chore: merge 3.0

2023-01-02 16:48:27 +08:00 · 2023-01-02 16:48:27 +08:00 · ea7c1338ff
parent 0fab1c8569 c684a7a334
commit ea7c1338ff
201 changed files with 13848 additions and 11403 deletions
--- a/cmake/cmake.define
+++ b/cmake/cmake.define
@ -1,6 +1,7 @@
 cmake_minimum_required(VERSION 3.0)

 set(CMAKE_VERBOSE_MAKEFILE OFF)
+set(TD_BUILD_TAOSA_INTERNAL FALSE)

 #set output directory
 SET(LIBRARY_OUTPUT_PATH ${PROJECT_BINARY_DIR}/build/lib)
--- a/cmake/cmake.install
+++ b/cmake/cmake.install
@ -21,7 +21,7 @@ IF (TD_LINUX)
 ELSEIF (TD_WINDOWS)
  SET(TD_MAKE_INSTALL_SH "${TD_SOURCE_DIR}/packaging/tools/make_install.bat")
  INSTALL(CODE "MESSAGE(\"make install script: ${TD_MAKE_INSTALL_SH}\")")
-  INSTALL(CODE "execute_process(COMMAND ${TD_MAKE_INSTALL_SH} :needAdmin ${TD_SOURCE_DIR} ${PROJECT_BINARY_DIR} Windows ${TD_VER_NUMBER})")
+  INSTALL(CODE "execute_process(COMMAND ${TD_MAKE_INSTALL_SH} :needAdmin ${TD_SOURCE_DIR} ${PROJECT_BINARY_DIR} Windows ${TD_VER_NUMBER} ${TD_BUILD_TAOSA_INTERNAL})")
 ELSEIF (TD_DARWIN)
  SET(TD_MAKE_INSTALL_SH "${TD_SOURCE_DIR}/packaging/tools/make_install.sh")
  INSTALL(CODE "MESSAGE(\"make install script: ${TD_MAKE_INSTALL_SH}\")")
--- a/cmake/taosadapter_CMakeLists.txt.in
+++ b/cmake/taosadapter_CMakeLists.txt.in
@ -2,7 +2,7 @@
 # taosadapter
 ExternalProject_Add(taosadapter
        GIT_REPOSITORY https://github.com/taosdata/taosadapter.git
-        GIT_TAG f0c1753
+        GIT_TAG a2e9920
        SOURCE_DIR "${TD_SOURCE_DIR}/tools/taosadapter"
        BINARY_DIR ""
        #BUILD_IN_SOURCE TRUE
--- a/docs/en/12-taos-sql/02-database.md
+++ b/docs/en/12-taos-sql/02-database.md
@ -30,6 +30,8 @@ database_option: {
  | WAL_LEVEL {1 | 2}
  | VGROUPS value
  | SINGLE_STABLE {0 | 1}
+  | TABLE_PREFIX value
+  | TABLE_SUFFIX value
  | WAL_RETENTION_PERIOD value
  | WAL_ROLL_PERIOD value
  | WAL_RETENTION_SIZE value
@ -67,6 +69,8 @@ database_option: {
 - SINGLE_STABLE: specifies whether the database can contain more than one supertable.
  - 0: The database can contain multiple supertables.
  - 1: The database can contain only one supertable.
+- TABLE_PREFIX：The prefix length in the table name that is ignored when distributing table to vnode based on table name.
+- TABLE_SUFFIX：The suffix length in the table name that is ignored when distributing table to vnode based on table name.
 - WAL_RETENTION_PERIOD: specifies the time after which WAL files are deleted. This parameter is used for data subscription. Enter a time in seconds. The default value of single copy is 0. A value of 0 indicates that each WAL file is deleted immediately after its contents are written to disk. -1: WAL files are never deleted. The default value of multiple copy is 4 days.
 - WAL_RETENTION_SIZE: specifies the size at which WAL files are deleted. This parameter is used for data subscription. Enter a size in KB. The default value of single copy is 0. A value of 0 indicates that each WAL file is deleted immediately after its contents are written to disk. -1: WAL files are never deleted. The default value of multiple copy is -1.
 - WAL_ROLL_PERIOD: specifies the time after which WAL files are rotated. After this period elapses, a new WAL file is created. The default value of single copy is 0. A value of 0 indicates that a new WAL file is created only after the previous WAL file was written to disk. The default values of multiple copy is 1 day.
--- a/docs/en/12-taos-sql/20-keywords.md
+++ b/docs/en/12-taos-sql/20-keywords.md
@ -17,6 +17,7 @@ The following list shows all reserved keywords:
 - ADD
 - AFTER
 - AGGREGATE
+- ALIVE
 - ALL
 - ALTER
 - ANALYZE
--- a/docs/en/12-taos-sql/24-show.md
+++ b/docs/en/12-taos-sql/24-show.md
@ -178,76 +178,138 @@ SHOW TABLE DISTRIBUTED table_name;

 Shows how table data is distributed.

-Examples： show table distributed d0\G;   Display the block distribution of table `d0` in detailed format.
+Examples： Below is an example of this command to display the block distribution of table `d0` in detailed format.

+```sql
+show table distributed d0\G;   
+```
+
+<details>
+ <summary> Show Example </summary>
+ <pre><code>
 *************************** 1.row ***************************
 _block_dist: Total_Blocks=[5] Total_Size=[93.65 Kb] Average_size=[18.73 Kb] Compression_Ratio=[23.98 %]

 Total_Blocks :  Table `d0` contains total 5 blocks
+
 Total_Size:  The total size of all the data blocks in table `d0` is 93.65 KB 
+
 Average_size:  The average size of each block is 18.73 KB
+
 Compression_Ratio: The data compression rate is 23.98%
 
 *************************** 2.row ***************************
 _block_dist: Total_Rows=[20000] Inmem_Rows=[0] MinRows=[3616] MaxRows=[4096] Average_Rows=[4000]

 Total_Rows: Table `d0` contains 20,000 rows
+
 Inmem_Rows： The rows still in memory, i.e. not committed in disk, is 0, i.e. none such rows
+
 MinRows：  The minimum number of rows in a block is 3,616 
+
 MaxRows： The maximum number of rows in a block is 4,096B
+
 Average_Rows： The average number of rows in a block is 4,000

 *************************** 3.row ***************************
 _block_dist: Total_Tables=[1] Total_Files=[2]

 Total_Tables:  The number of child tables, 1 in this example
+
 Total_Files：   The number of files storing the table's data, 2 in this example

 *************************** 4.row ***************************
+
 _block_dist: --------------------------------------------------------------------------------
+
 *************************** 5.row ***************************
+
 _block_dist: 0100 |
+
 *************************** 6.row ***************************
+
 _block_dist: 0299 |
+
 *************************** 7.row ***************************
+
 _block_dist: 0498 |
+
 *************************** 8.row ***************************
+
 _block_dist: 0697 |
+
 *************************** 9.row ***************************
+
 _block_dist: 0896 |
+
 *************************** 10.row ***************************
+
 _block_dist: 1095 |
+
 *************************** 11.row ***************************
+
 _block_dist: 1294 |
+
 *************************** 12.row ***************************
+
 _block_dist: 1493 |
+
 *************************** 13.row ***************************
+
 _block_dist: 1692 |
+
 *************************** 14.row ***************************
+
 _block_dist: 1891 |
+
 *************************** 15.row ***************************
+
 _block_dist: 2090 |
+
 *************************** 16.row ***************************
+
 _block_dist: 2289 |
+
 *************************** 17.row ***************************
+
 _block_dist: 2488 |
+
 *************************** 18.row ***************************
+
 _block_dist: 2687 |
+
 *************************** 19.row ***************************
+
 _block_dist: 2886 |
+
 *************************** 20.row ***************************
+
 _block_dist: 3085 |
+
 *************************** 21.row ***************************
+
 _block_dist: 3284 |
+
 *************************** 22.row ***************************
+
 _block_dist: 3483 |||||||||||||||||  1 (20.00%)
+
 *************************** 23.row ***************************
+
 _block_dist: 3682 |
+
 *************************** 24.row ***************************
+
 _block_dist: 3881 |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||  4 (80.00%)
+
 Query OK, 24 row(s) in set (0.002444s)

-  The above show the block distribution percentage according to the number of rows in each block. In the above example, `_block_dist: 3483 |||||||||||||||||  1 (20.00%)` means there is one block whose rows is between 3,483 and 3,681. `_block_dist: 3881 |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||  4 (80.00%)` means there are 4 blocks whose rows is between 3,881 and 4,096. The number of blocks whose rows fall in other range is zero.
+</code></pre>
+</details>
+
+  The above show the block distribution percentage according to the number of rows in each block. In the above example, we can get below information:
+  - `_block_dist: 3483 |||||||||||||||||  1 (20.00%)` means there is one block whose rows is between 3,483 and 3,681.
+  -  `_block_dist: 3881 |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||  4 (80.00%)` means there are 4 blocks whose rows is between 3,881 and 4,096.   -  The number of blocks whose rows fall in other range is zero.

 ## SHOW TAGS

--- a/docs/en/14-reference/04-taosadapter.md
+++ b/docs/en/14-reference/04-taosadapter.md
@ -21,6 +21,7 @@ taosAdapter provides the following features.
 - Seamless connection to collectd
 - Seamless connection to StatsD
 - Supports Prometheus remote_read and remote_write
+- Get table's VGroup ID

 ## taosAdapter architecture diagram

@ -178,6 +179,7 @@ See [example/config/taosadapter.toml](https://github.com/taosdata/taosadapter/bl
  node_export is an exporter for machine metrics. Please visit [https://github.com/prometheus/node_exporter](https://github.com/prometheus/node_exporter) for more information.
 - Support for Prometheus remote_read and remote_write
  remote_read and remote_write are interfaces for Prometheus data read and write from/to other data storage solution. Please visit [https://prometheus.io/blog/2019/10/10/remote-read-meets-streaming/#remote-apis](https://prometheus.io/blog/2019/10/10/remote-read-meets-streaming/#remote-apis) for more information.
+- Get table's VGroup ID. For more information about VGroup, please refer to [primary-logic-unit](/tdinternal/arch/#primary-logic-unit).

 ## Interfaces

@ -199,7 +201,7 @@ Support InfluxDB query parameters as follows.
 - `precision` The time precision used by TDengine
 - `u` TDengine user name
 - `p` TDengine password
- `ttl` The time to live of automatically created sub-table. This value cannot be updated. TDengine will use the ttl value of the frist data of sub-table to create sub-table. For more information, please refer [Create Table](/taos-sql/table/#create-table)
+- `ttl` The time to live of automatically created sub-table. This value cannot be updated. TDengine will use the ttl value of the first data of sub-table to create sub-table. For more information, please refer [Create Table](/taos-sql/table/#create-table)

 Note:      InfluxDB token authorization is not supported at present. Only Basic authorization and query parameter validation are supported.
 Example:   curl --request POST http://127.0.0.1:6041/influxdb/v1/write?db=test --user "root:taosdata" --data-binary "measurement,host=host1 field1=2i,field2=2.0 1577836800000000000"
@ -241,6 +243,10 @@ node_export is an exporter of hardware and OS metrics exposed by the \*NIX kerne

 <Prometheus />

+### Get table's VGroup ID
+
+You can call `http://<fqdn>:6041/rest/vgid?db=<db>&table=<table>` to get table's VGroup ID. For more information about VGroup, please refer to [primary-logic-unit](/tdinternal/arch/#primary-logic-unit).
+
 ## Memory usage optimization methods

 taosAdapter will monitor its memory usage during operation and adjust it with two thresholds. Valid values are integers between 1 to 100, and represent a percentage of the system's physical memory.
--- a/docs/en/28-releases/01-tdengine.md
+++ b/docs/en/28-releases/01-tdengine.md
@ -10,6 +10,10 @@ For TDengine 2.x installation packages by version, please visit [here](https://w

 import Release from "/components/ReleaseV3";

+## 3.0.2.2
+
+<Release type="tdengine" version="3.0.2.2" />
+
 ## 3.0.2.1

 <Release type="tdengine" version="3.0.2.1" />
--- a/docs/en/28-releases/02-tools.md
+++ b/docs/en/28-releases/02-tools.md
@ -10,6 +10,10 @@ For other historical version installers, please visit [here](https://www.taosdat

 import Release from "/components/ReleaseV3";

+## 2.4.0
+
+<Release type="tools" version="2.4.0" />
+
 ## 2.3.3

 <Release type="tools" version="2.3.3" />
--- a/docs/zh/12-taos-sql/02-database.md
+++ b/docs/zh/12-taos-sql/02-database.md
@ -30,6 +30,8 @@ database_option: {
  | WAL_LEVEL {1 | 2}
  | VGROUPS value
  | SINGLE_STABLE {0 | 1}
+  | TABLE_PREFIX value
+  | TABLE_SUFFIX value
  | WAL_RETENTION_PERIOD value
  | WAL_ROLL_PERIOD value
  | WAL_RETENTION_SIZE value
@ -67,6 +69,8 @@ database_option: {
 - SINGLE_STABLE：表示此数据库中是否只可以创建一个超级表，用于超级表列非常多的情况。
  - 0：表示可以创建多张超级表。
  - 1：表示只可以创建一张超级表。
+- TABLE_PREFIX：内部存储引擎根据表名分配存储该表数据的 VNODE 时要忽略的前缀的长度。
+- TABLE_SUFFIX：内部存储引擎根据表名分配存储该表数据的 VNODE 时要忽略的后缀的长度。
 - WAL_RETENTION_PERIOD：wal 文件的额外保留策略，用于数据订阅。wal 的保存时长，单位为 s。单副本默认为 0，即落盘后立即删除。-1 表示不删除。多副本默认为 4 天。
 - WAL_RETENTION_SIZE：wal 文件的额外保留策略，用于数据订阅。wal 的保存的最大上限，单位为 KB。单副本默认为 0，即落盘后立即删除。多副本默认为-1，表示不删除。
 - WAL_ROLL_PERIOD：wal 文件切换时长，单位为 s。当 wal 文件创建并写入后，经过该时间，会自动创建一个新的 wal 文件。单副本默认为 0，即仅在落盘时创建新文件。多副本默认为 1 天。
--- a/docs/zh/12-taos-sql/20-keywords.md
+++ b/docs/zh/12-taos-sql/20-keywords.md
@ -18,6 +18,7 @@ description: TDengine 保留关键字的详细列表
 - ADD
 - AFTER
 - AGGREGATE
+- ALIVE
 - ALL
 - ALTER
 - ANALYZE
--- a/docs/zh/12-taos-sql/24-show.md
+++ b/docs/zh/12-taos-sql/24-show.md
@ -183,6 +183,10 @@ SHOW TABLE DISTRIBUTED table_name;

 语句： show table distributed d0\G;   竖行显示表 d0 的 BLOCK 分布情况

+<details>
+ <summary>显示示例</summary>
+ <pre><code>
+
 *************************** 1.row ***************************

 _block_dist: Total_Blocks=[5] Total_Size=[93.65 Kb] Average_size=[18.73 Kb] Compression_Ratio=[23.98 %]
@ -244,6 +248,8 @@ _block_dist: 3881 ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||

 Query OK, 24 row(s) in set (0.002444s)

+</code></pre>
+ </details>

  上面是块中包含数据行数的块儿分布情况图，这里的 0100 0299 0498 … 表示的是每个块中包含的数据行数，上面的意思就是这个表的 5 个块，分布在 3483 ~3681 行的块有 1 个，占整个块的 20%，分布在 3881 ~ 4096（最大行数）的块数为 4 个，占整个块的 80%， 其它区域内分布块数为 0。

--- a/docs/zh/12-taos-sql/29-changes.md
+++ b/docs/zh/12-taos-sql/29-changes.md
@ -94,6 +94,7 @@ description: "TDengine 3.0 版本的语法变更说明"
 | 9 | SAMPLE | 增强 | 可以直接用于超级表了。没有PARTITION BY时，超级表的数据会被合并成一条时间线。
 | 10 | STATECOUNT | 增强 | 可以直接用于超级表了。没有PARTITION BY时，超级表的数据会被合并成一条时间线。
 | 11 | STATEDURATION | 增强 | 可以直接用于超级表了。没有PARTITION BY时，超级表的数据会被合并成一条时间线。
+| 12 | TIMETRUNCATE | 增强 | 增加ignore_timezone参数，可选是否使用，默认值为1.


 ## SCHEMALESS 变更
--- a/docs/zh/14-reference/04-taosadapter.md
+++ b/docs/zh/14-reference/04-taosadapter.md
@ -21,6 +21,7 @@ taosAdapter 提供以下功能：
 - 无缝连接到 collectd
 - 无缝连接到 StatsD
 - 支持 Prometheus remote_read 和 remote_write
+- 获取 table 所在的虚拟节点组（VGroup）的 VGroup ID

 ## taosAdapter 架构图

@ -178,6 +179,7 @@ AllowWebSockets
  node_export 是一个机器指标的导出器。请访问 [https://github.com/prometheus/node_exporter](https://github.com/prometheus/node_exporter) 了解更多信息。
 - 支持 Prometheus remote_read 和 remote_write
  remote_read 和 remote_write 是 Prometheus 数据读写分离的集群方案。请访问[https://prometheus.io/blog/2019/10/10/remote-read-meets-streaming/#remote-apis](https://prometheus.io/blog/2019/10/10/remote-read-meets-streaming/#remote-apis) 了解更多信息。
+- 获取 table 所在的虚拟节点组（VGroup）的 VGroup ID。关于虚拟节点组（VGroup）的更多信息，请访问[整体架构文档](/tdinternal/arch/#主要逻辑单元) 。

 ## 接口

@ -240,6 +242,10 @@ Prometheus 使用的由 \*NIX 内核暴露的硬件和操作系统指标的输

 <Prometheus />

+### 获取 table 的 VGroup ID
+
+可以访问 http 接口 `http://<fqdn>:6041/rest/vgid?db=<db>&table=<table>` 获取 table 的 VGroup ID。关于虚拟节点组（VGroup）的更多信息，请访问[整体架构文档](/tdinternal/arch/#主要逻辑单元) 。
+
 ## 内存使用优化方法

 taosAdapter 将监测自身运行过程中内存使用率并通过两个阈值进行调节。有效值范围为 -1 到 100 的整数，单位为系统物理内存的百分比。
@ -282,7 +288,7 @@ http 返回内容：

 ## taosAdapter 监控指标

-taosAdapter 采集 http 相关指标、cpu 百分比和内存百分比。
+taosAdapter 采集 http 相关指标、CPU 百分比和内存百分比。

 ### http 接口

@ -294,13 +300,13 @@ http://<fqdn>:6041/metrics

 ### 写入 TDengine

-taosAdapter 支持将 http 监控、cpu 百分比和内存百分比写入 TDengine。
+taosAdapter 支持将 http 监控、CPU 百分比和内存百分比写入 TDengine。

 有关配置参数

 | **配置项**                 | **描述**                                     | **默认值**  |
 |-------------------------|--------------------------------------------|----------|
-| monitor.collectDuration | cpu 和内存采集间隔                                | 3s       |
+| monitor.collectDuration | CPU 和内存采集间隔                                | 3s       |
 | monitor.identity        | 当前taosadapter 的标识符如果不设置将使用 'hostname:port' |          |
 | monitor.incgroup        | 是否是 cgroup 中运行(容器中运行设置为 true)              | false    |
 | monitor.writeToTD       | 是否写入到 TDengine                             | false    |
--- a/docs/zh/28-releases/01-tdengine.md
+++ b/docs/zh/28-releases/01-tdengine.md
@ -10,6 +10,10 @@ TDengine 2.x 各版本安装包请访问[这里](https://www.taosdata.com/all-do

 import Release from "/components/ReleaseV3";

+## 3.0.2.2
+
+<Release type="tdengine" version="3.0.2.2" />
+
 ## 3.0.2.1

 <Release type="tdengine" version="3.0.2.1" />
--- a/docs/zh/28-releases/02-tools.md
+++ b/docs/zh/28-releases/02-tools.md
@ -10,6 +10,10 @@ taosTools 各版本安装包下载链接如下：

 import Release from "/components/ReleaseV3";

+## 2.4.0
+
+<Release type="tools" version="2.4.0" />
+
 ## 2.3.3

 <Release type="tools" version="2.3.3" />
--- a/include/common/tcommon.h
+++ b/include/common/tcommon.h
@ -341,7 +341,7 @@ typedef struct SExprInfo {

 typedef struct {
  const char* key;
-  int32_t     keyLen;
+  size_t     keyLen;
  uint8_t     type;
  union {
    const char* value;
@ -350,7 +350,7 @@ typedef struct {
    double      d;
    float       f;
  };
-  int32_t length;
+  size_t length;
 } SSmlKv;

 #define QUERY_ASC_FORWARD_STEP  1
--- a/include/common/tdatablock.h
+++ b/include/common/tdatablock.h
@ -265,7 +265,7 @@ void blockDebugShowDataBlocks(const SArray* dataBlocks, const char* flag);
 // for debug
 char* dumpBlockData(SSDataBlock* pDataBlock, const char* flag, char** dumpBuf);

-int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SSDataBlock* pDataBlocks, STSchema* pTSchema, int32_t vgId,
+int32_t buildSubmitReqFromDataBlock(SSubmitReq2** pReq, const SSDataBlock* pDataBlocks, const STSchema* pTSchema, int64_t uid, int32_t vgId,
                                    tb_uid_t suid);

 char* buildCtbNameByGroupId(const char* stbName, uint64_t groupId);
--- a/include/common/tdataformat.h
+++ b/include/common/tdataformat.h
@ -44,18 +44,38 @@ typedef struct SColData SColData;
 #define HAS_VALUE ((uint8_t)0x4)

 // bitmap ================================
-const static uint8_t BIT2_MAP[4][4] = {{0b00000000, 0b00000001, 0b00000010, 0},
-                                       {0b00000000, 0b00000100, 0b00001000, 2},
-                                       {0b00000000, 0b00010000, 0b00100000, 4},
-                                       {0b00000000, 0b01000000, 0b10000000, 6}};
+const static uint8_t BIT1_MAP[8] = {0b11111110, 0b11111101, 0b11111011, 0b11110111,
+                                    0b11101111, 0b11011111, 0b10111111, 0b01111111};

-#define N1(n)             ((((uint8_t)1) << (n)) - 1)
-#define BIT1_SIZE(n)      ((((n)-1) >> 3) + 1)
-#define BIT2_SIZE(n)      ((((n)-1) >> 2) + 1)
-#define SET_BIT1(p, i, v) ((p)[(i) >> 3] = (p)[(i) >> 3] & N1((i)&7) | (((uint8_t)(v)) << ((i)&7)))
-#define GET_BIT1(p, i)    (((p)[(i) >> 3] >> ((i)&7)) & ((uint8_t)1))
-#define SET_BIT2(p, i, v) ((p)[(i) >> 2] = (p)[(i) >> 2] & N1(BIT2_MAP[(i)&3][3]) | BIT2_MAP[(i)&3][(v)])
-#define GET_BIT2(p, i)    (((p)[(i) >> 2] >> BIT2_MAP[(i)&3][3]) & ((uint8_t)3))
+const static uint8_t BIT2_MAP[4] = {0b11111100, 0b11110011, 0b11001111, 0b00111111};
+
+#define ONE               ((uint8_t)1)
+#define THREE             ((uint8_t)3)
+#define DIV_8(i)          ((i) >> 3)
+#define MOD_8(i)          ((i)&7)
+#define DIV_4(i)          ((i) >> 2)
+#define MOD_4(i)          ((i)&3)
+#define MOD_4_TIME_2(i)   (MOD_4(i) << 1)
+#define BIT1_SIZE(n)      (DIV_8((n)-1) + 1)
+#define BIT2_SIZE(n)      (DIV_4((n)-1) + 1)
+#define SET_BIT1(p, i, v) ((p)[DIV_8(i)] = (p)[DIV_8(i)] & BIT1_MAP[MOD_8(i)] | ((v) << MOD_8(i)))
+#define SET_BIT1_EX(p, i, v) \
+  do {                       \
+    if (MOD_8(i) == 0) {     \
+      (p)[DIV_8(i)] = 0;     \
+    }                        \
+    SET_BIT1(p, i, v);       \
+  } while (0)
+#define GET_BIT1(p, i)    (((p)[DIV_8(i)] >> MOD_8(i)) & ONE)
+#define SET_BIT2(p, i, v) ((p)[DIV_4(i)] = (p)[DIV_4(i)] & BIT2_MAP[MOD_4(i)] | ((v) << MOD_4_TIME_2(i)))
+#define SET_BIT2_EX(p, i, v) \
+  do {                       \
+    if (MOD_4(i) == 0) {     \
+      (p)[DIV_4(i)] = 0;     \
+    }                        \
+    SET_BIT2(p, i, v);       \
+  } while (0)
+#define GET_BIT2(p, i) (((p)[DIV_4(i)] >> MOD_4_TIME_2(i)) & THREE)

 // SBuffer ================================
 struct SBuffer {
@ -70,9 +90,6 @@ int32_t tBufferInit(SBuffer *pBuffer, int64_t size);
 int32_t tBufferPut(SBuffer *pBuffer, const void *pData, int64_t nData);
 int32_t tBufferReserve(SBuffer *pBuffer, int64_t nData, void **ppData);

-// STSchema ================================
-void tDestroyTSchema(STSchema *pTSchema);
-
 // SColVal ================================
 #define CV_FLAG_VALUE ((int8_t)0x0)
 #define CV_FLAG_NONE  ((int8_t)0x1)
@ -87,8 +104,12 @@ void tDestroyTSchema(STSchema *pTSchema);
 #define COL_VAL_IS_VALUE(CV) ((CV)->flag == CV_FLAG_VALUE)

 // SRow ================================
-int32_t tRowBuild(SArray *aColVal, STSchema *pTSchema, SBuffer *pBuffer);
+int32_t tRowBuild(SArray *aColVal, const STSchema *pTSchema, SRow **ppRow);
 void    tRowGet(SRow *pRow, STSchema *pTSchema, int32_t iCol, SColVal *pColVal);
+void    tRowDestroy(SRow *pRow);
+void    tRowSort(SArray *aRowP);
+int32_t tRowMerge(SArray *aRowP, STSchema *pTSchema, int8_t flag);
+int32_t tRowAppendToColData(SRow *pRow, STSchema *pTSchema, SColData *aColData, int32_t nColData);

 // SRowIter ================================
 int32_t  tRowIterOpen(SRow *pRow, STSchema *pTSchema, SRowIter **ppIter);
@ -110,15 +131,28 @@ void    debugPrintSTag(STag *pTag, const char *tag, int32_t ln);  // TODO: remov
 int32_t parseJsontoTagData(const char *json, SArray *pTagVals, STag **ppTag, void *pMsgBuf);

 // SColData ================================
+typedef void *(*xMallocFn)(void *, int32_t);
 void    tColDataDestroy(void *ph);
 void    tColDataInit(SColData *pColData, int16_t cid, int8_t type, int8_t smaOn);
 void    tColDataClear(SColData *pColData);
+void    tColDataDeepClear(SColData *pColData);
 int32_t tColDataAppendValue(SColData *pColData, SColVal *pColVal);
 void    tColDataGetValue(SColData *pColData, int32_t iVal, SColVal *pColVal);
 uint8_t tColDataGetBitValue(const SColData *pColData, int32_t iVal);
-int32_t tColDataCopy(SColData *pColDataSrc, SColData *pColDataDest);
+int32_t tColDataCopy(SColData *pColDataFrom, SColData *pColData, xMallocFn xMalloc, void *arg);
 extern void (*tColDataCalcSMA[])(SColData *pColData, int64_t *sum, int64_t *max, int64_t *min, int16_t *numOfNull);

+// for stmt bind
+int32_t tColDataAddValueByBind(SColData *pColData, TAOS_MULTI_BIND *pBind);
+void    tColDataSortMerge(SArray *colDataArr);
+
+//for raw block
+int32_t tColDataAddValueByDataBlock(SColData *pColData, int8_t type, int32_t bytes,
+                                    int32_t nRows, char* lengthOrbitmap, char *data);
+// for encode/decode
+int32_t tPutColData(uint8_t *pBuf, SColData *pColData);
+int32_t tGetColData(uint8_t *pBuf, SColData *pColData);
+
 // STRUCT ================================
 struct STColumn {
  col_id_t colId;
@ -225,23 +259,9 @@ struct STag {
    memcpy(varDataVal(x), (str), (_size));      \
  } while (0);

-// ----------------- SCHEMA BUILDER DEFINITION
-typedef struct {
-  int32_t      tCols;
-  int32_t      nCols;
-  schema_ver_t version;
-  uint16_t     flen;
-  int32_t      tlen;
-  STColumn    *columns;
-} STSchemaBuilder;
-
-int32_t   tdInitTSchemaBuilder(STSchemaBuilder *pBuilder, schema_ver_t version);
-void      tdDestroyTSchemaBuilder(STSchemaBuilder *pBuilder);
-void      tdResetTSchemaBuilder(STSchemaBuilder *pBuilder, schema_ver_t version);
-int32_t   tdAddColToSchema(STSchemaBuilder *pBuilder, int8_t type, int8_t flags, col_id_t colId, col_bytes_t bytes);
-STSchema *tdGetSchemaFromBuilder(STSchemaBuilder *pBuilder);
-
+// STSchema ================================
 STSchema *tBuildTSchema(SSchema *aSchema, int32_t numOfCols, int32_t version);
+void      tDestroyTSchema(STSchema *pTSchema);

 #endif

--- a/include/common/tglobal.h
+++ b/include/common/tglobal.h
@ -55,7 +55,7 @@ extern int32_t tsNumOfMnodeQueryThreads;
 extern int32_t tsNumOfMnodeFetchThreads;
 extern int32_t tsNumOfMnodeReadThreads;
 extern int32_t tsNumOfVnodeQueryThreads;
-extern int32_t tsNumOfVnodeStreamThreads;
+extern float   tsRatioOfVnodeStreamThreads;
 extern int32_t tsNumOfVnodeFetchThreads;
 extern int32_t tsNumOfVnodeRsmaThreads;
 extern int32_t tsNumOfQnodeQueryThreads;
--- a/include/common/tmsg.h
+++ b/include/common/tmsg.h
@ -482,8 +482,6 @@ static FORCE_INLINE int32_t tDecodeSSchemaWrapperEx(SDecoder* pDecoder, SSchemaW
  return 0;
 }

-STSchema* tdGetSTSChemaFromSSChema(SSchema* pSchema, int32_t nCols, int32_t sver);
-
 typedef struct {
  char     name[TSDB_TABLE_FNAME_LEN];
  int8_t   igExists;
@ -1734,6 +1732,8 @@ typedef struct {
  int32_t  execId;
 } STaskDropReq;

+int32_t tSerializeSTaskDropReq(void* buf, int32_t bufLen, STaskDropReq* pReq);
+int32_t tDeserializeSTaskDropReq(void* buf, int32_t bufLen, STaskDropReq* pReq);
 int32_t tSerializeSTaskDropReq(void* buf, int32_t bufLen, STaskDropReq* pReq);
 int32_t tDeserializeSTaskDropReq(void* buf, int32_t bufLen, STaskDropReq* pReq);

@ -1751,6 +1751,8 @@ typedef struct {
 #define STREAM_FILL_HISTORY_ON        1
 #define STREAM_FILL_HISTORY_OFF       0
 #define STREAM_DEFAULT_FILL_HISTORY   STREAM_FILL_HISTORY_OFF
+#define STREAM_CREATE_STABLE_TRUE     1
+#define STREAM_CREATE_STABLE_FALSE    0

 typedef struct {
  char    name[TSDB_STREAM_FNAME_LEN];
@ -1768,6 +1770,8 @@ typedef struct {
  SArray* pTags;  // array of SField
  // 3.0.20
  int64_t checkpointFreq;  // ms
+  // 3.0.2.3
+  int8_t createStb;
 } SCMCreateStreamReq;

 typedef struct {
@ -2081,10 +2085,15 @@ typedef struct SVCreateTbReq {
  };
 } SVCreateTbReq;

-int tEncodeSVCreateTbReq(SEncoder* pCoder, const SVCreateTbReq* pReq);
-int tDecodeSVCreateTbReq(SDecoder* pCoder, SVCreateTbReq* pReq);
+int  tEncodeSVCreateTbReq(SEncoder* pCoder, const SVCreateTbReq* pReq);
+int  tDecodeSVCreateTbReq(SDecoder* pCoder, SVCreateTbReq* pReq);
+void tDestroySVCreateTbReq(SVCreateTbReq* pReq, int32_t flags);

 static FORCE_INLINE void tdDestroySVCreateTbReq(SVCreateTbReq* req) {
+  if (NULL == req) {
+    return;
+  }
+
  taosMemoryFreeClear(req->name);
  taosMemoryFreeClear(req->comment);
  if (req->type == TSDB_CHILD_TABLE) {
@ -3232,6 +3241,57 @@ int32_t tSerializeSMqAskEpReq(void* buf, int32_t bufLen, SMqAskEpReq* pReq);
 int32_t tDeserializeSMqAskEpReq(void* buf, int32_t bufLen, SMqAskEpReq* pReq);
 int32_t tSerializeSMqHbReq(void* buf, int32_t bufLen, SMqHbReq* pReq);
 int32_t tDeserializeSMqHbReq(void* buf, int32_t bufLen, SMqHbReq* pReq);
+int32_t tSerializeSMqAskEpReq(void* buf, int32_t bufLen, SMqAskEpReq* pReq);
+int32_t tDeserializeSMqAskEpReq(void* buf, int32_t bufLen, SMqAskEpReq* pReq);
+int32_t tSerializeSMqHbReq(void* buf, int32_t bufLen, SMqHbReq* pReq);
+int32_t tDeserializeSMqHbReq(void* buf, int32_t bufLen, SMqHbReq* pReq);
+
+#define SUBMIT_REQ_AUTO_CREATE_TABLE  0x1
+#define SUBMIT_REQ_COLUMN_DATA_FORMAT 0x2
+
+typedef struct {
+  int32_t        flags;
+  SVCreateTbReq* pCreateTbReq;
+  int64_t        suid;
+  int64_t        uid;
+  int32_t        sver;
+  union {
+    SArray* aRowP;
+    SArray* aCol;
+  };
+} SSubmitTbData;
+
+typedef struct {
+  SArray* aSubmitTbData;  // SArray<SSubmitTbData>
+} SSubmitReq2;
+
+int32_t tEncodeSSubmitReq2(SEncoder* pCoder, const SSubmitReq2* pReq);
+int32_t tDecodeSSubmitReq2(SDecoder* pCoder, SSubmitReq2* pReq);
+void    tDestroySSubmitTbData(SSubmitTbData* pTbData, int32_t flag);
+void    tDestroySSubmitReq2(SSubmitReq2* pReq, int32_t flag);
+
+typedef struct {
+  int32_t affectedRows;
+  SArray* aCreateTbRsp;  // SArray<SVCreateTbRsp>
+} SSubmitRsp2;
+
+int32_t tEncodeSSubmitRsp2(SEncoder* pCoder, const SSubmitRsp2* pRsp);
+int32_t tDecodeSSubmitRsp2(SDecoder* pCoder, SSubmitRsp2* pRsp);
+void    tDestroySSubmitRsp2(SSubmitRsp2* pRsp, int32_t flag);
+
+#define TSDB_MSG_FLG_ENCODE 0x1
+#define TSDB_MSG_FLG_DECODE 0x2
+
+typedef struct {
+  union {
+    struct {
+      void*   msgStr;
+      int32_t msgLen;
+      int64_t ver;
+    };
+    void* pDataBlock;
+  };
+} SPackedData;

 #pragma pack(pop)

--- a/include/libs/executor/executor.h
+++ b/include/libs/executor/executor.h
@ -190,7 +190,9 @@ int32_t qStreamPrepareTsdbScan(qTaskInfo_t tinfo, uint64_t uid, int64_t ts);

 int32_t qStreamPrepareScan(qTaskInfo_t tinfo, STqOffsetVal* pOffset, int8_t subType);

-int32_t qStreamScanMemData(qTaskInfo_t tinfo, const SSubmitReq* pReq);
+// int32_t qStreamScanMemData(qTaskInfo_t tinfo, const SSubmitReq* pReq, int64_t ver);
+//
+int32_t qStreamSetScanMemData(qTaskInfo_t tinfo, SPackedData submit);

 int32_t qStreamExtractOffset(qTaskInfo_t tinfo, STqOffsetVal* pOffset);

@ -213,6 +215,7 @@ int32_t qStreamSourceRecoverStep1(qTaskInfo_t tinfo, int64_t ver);
 int32_t qStreamSourceRecoverStep2(qTaskInfo_t tinfo, int64_t ver);
 int32_t qStreamRecoverFinish(qTaskInfo_t tinfo);
 int32_t qStreamRestoreParam(qTaskInfo_t tinfo);
+bool    qStreamRecoverScanFinished(qTaskInfo_t tinfo);

 #ifdef __cplusplus
 }
--- a/include/libs/nodes/cmdnodes.h
+++ b/include/libs/nodes/cmdnodes.h
@ -295,7 +295,7 @@ typedef struct SShowTableTagsStmt {
  SNodeList* pTags;
 } SShowTableTagsStmt;

-typedef enum EIndexType { INDEX_TYPE_SMA = 1, INDEX_TYPE_FULLTEXT } EIndexType;
+typedef enum EIndexType { INDEX_TYPE_SMA = 1, INDEX_TYPE_FULLTEXT, INDEX_TYPE_NORMAL } EIndexType;

 typedef struct SIndexOptions {
  ENodeType  type;
@ -401,6 +401,7 @@ typedef struct SCreateStreamStmt {
  SNode*          pQuery;
  SNodeList*      pTags;
  SNode*          pSubtable;
+  SNodeList*      pCols;
 } SCreateStreamStmt;

 typedef struct SDropStreamStmt {
--- a/include/libs/nodes/querynodes.h
+++ b/include/libs/nodes/querynodes.h
@ -361,33 +361,33 @@ typedef struct SVgDataBlocks {
  void*       pData;  // SSubmitReq + SSubmitBlk + ...
 } SVgDataBlocks;

-typedef void (*FFreeDataBlockHash)(SHashObj*);
-typedef void (*FFreeDataBlockArray)(SArray*);
+typedef void (*FFreeTableBlockHash)(SHashObj*);
+typedef void (*FFreeVgourpBlockArray)(SArray*);

 typedef struct SVnodeModifyOpStmt {
-  ENodeType           nodeType;
-  ENodeType           sqlNodeType;
-  SArray*             pDataBlocks;  // data block for each vgroup, SArray<SVgDataBlocks*>.
-  uint32_t            insertType;   // insert data from [file|sql statement| bound statement]
-  const char*         pSql;         // current sql statement position
-  int32_t             totalRowsNum;
-  int32_t             totalTbNum;
-  SName               targetTableName;
-  SName               usingTableName;
-  const char*         pBoundCols;
-  struct STableMeta*  pTableMeta;
-  SHashObj*           pVgroupsHashObj;
-  SHashObj*           pTableBlockHashObj;
-  SHashObj*           pSubTableHashObj;
-  SHashObj*           pTableNameHashObj;
-  SHashObj*           pDbFNameHashObj;
-  SArray*             pVgDataBlocks;
-  SVCreateTbReq       createTblReq;
-  TdFilePtr           fp;
-  FFreeDataBlockHash  freeHashFunc;
-  FFreeDataBlockArray freeArrayFunc;
-  bool                usingTableProcessing;
-  bool                fileProcessing;
+  ENodeType             nodeType;
+  ENodeType             sqlNodeType;
+  SArray*               pDataBlocks;  // data block for each vgroup, SArray<SVgDataBlocks*>.
+  uint32_t              insertType;   // insert data from [file|sql statement| bound statement]
+  const char*           pSql;         // current sql statement position
+  int32_t               totalRowsNum;
+  int32_t               totalTbNum;
+  SName                 targetTableName;
+  SName                 usingTableName;
+  const char*           pBoundCols;
+  struct STableMeta*    pTableMeta;
+  SHashObj*             pVgroupsHashObj;
+  SHashObj*             pTableBlockHashObj;  // SHashObj<tuid, STableDataCxt*>
+  SHashObj*             pSubTableHashObj;
+  SHashObj*             pTableNameHashObj;
+  SHashObj*             pDbFNameHashObj;
+  SArray*               pVgDataBlocks;  // SArray<SVgroupDataCxt*>
+  SVCreateTbReq*        pCreateTblReq;
+  TdFilePtr             fp;
+  FFreeTableBlockHash   freeHashFunc;
+  FFreeVgourpBlockArray freeArrayFunc;
+  bool                  usingTableProcessing;
+  bool                  fileProcessing;
 } SVnodeModifyOpStmt;

 typedef struct SExplainOptions {
--- a/include/libs/parser/parser.h
+++ b/include/libs/parser/parser.h
@ -58,7 +58,6 @@ typedef struct SParseContext {
  bool             isSuperUser;
  bool             enableSysInfo;
  bool             async;
-  int8_t           schemalessType;
  const char*      svrVer;
  bool             nodeOffline;
  SArray*          pTableMetaPos;    // sql table pos => catalog data pos
@ -85,12 +84,12 @@ int32_t qSetSTableIdForRsma(SNode* pStmt, int64_t uid);
 void    qCleanupKeywordsTable();

 int32_t     qBuildStmtOutput(SQuery* pQuery, SHashObj* pVgHash, SHashObj* pBlockHash);
-int32_t     qResetStmtDataBlock(void* block, bool keepBuf);
-int32_t     qCloneStmtDataBlock(void** pDst, void* pSrc);
-void        qFreeStmtDataBlock(void* pDataBlock);
-int32_t     qRebuildStmtDataBlock(void** pDst, void* pSrc, uint64_t uid, int32_t vgId);
-void        qDestroyStmtDataBlock(void* pBlock);
-STableMeta* qGetTableMetaInDataBlock(void* pDataBlock);
+int32_t     qResetStmtDataBlock(STableDataCxt* block, bool keepBuf);
+int32_t     qCloneStmtDataBlock(STableDataCxt** pDst, STableDataCxt* pSrc, bool reset);
+int32_t     qRebuildStmtDataBlock(STableDataCxt** pDst, STableDataCxt* pSrc, uint64_t uid, uint64_t suid, int32_t vgId, bool rebuildCreateTb);
+void        qDestroyStmtDataBlock(STableDataCxt* pBlock);
+STableMeta* qGetTableMetaInDataBlock(STableDataCxt* pDataBlock);
+int32_t qCloneCurrentTbData(STableDataCxt*       pDataBlock, SSubmitTbData **pData);

 int32_t qStmtBindParams(SQuery* pQuery, TAOS_MULTI_BIND* pParams, int32_t colIdx);
 int32_t qStmtParseQuerySql(SParseContext* pCxt, SQuery* pQuery);
@ -105,11 +104,18 @@ void    destroyBoundColumnInfo(void* pBoundInfo);
 int32_t qCreateSName(SName* pName, const char* pTableName, int32_t acctId, char* dbName, char* msgBuf,
                     int32_t msgBufLen);

-void*   smlInitHandle(SQuery* pQuery);
-void    smlDestroyHandle(void* pHandle);
-int32_t smlBindData(void* handle, SArray* tags, SArray* colsSchema, SArray* cols, bool format, STableMeta* pTableMeta,
+void    qDestroyBoundColInfo(void* pInfo);
+
+SQuery* smlInitHandle();
+int32_t smlBuildRow(STableDataCxt* pTableCxt);
+int32_t smlBuildCol(STableDataCxt* pTableCxt, SSchema* schema, void *kv, int32_t index);
+STableDataCxt* smlInitTableDataCtx(SQuery* query, STableMeta* pTableMeta);
+
+int32_t smlBindData(SQuery* handle, bool dataFormat, SArray* tags, SArray* colsSchema, SArray* cols, STableMeta* pTableMeta,
                    char* tableName, const char* sTableName, int32_t sTableNameLen, int32_t ttl, char* msgBuf, int16_t msgBufLen);
-int32_t smlBuildOutput(void* handle, SHashObj* pVgHash);
+int32_t smlBuildOutput(SQuery* handle, SHashObj* pVgHash);
+
+int     rawBlockBindData(SQuery *query, STableMeta* pTableMeta, void* data, SVCreateTbReq* pCreateTb, TAOS_FIELD *fields, int numFields);

 int32_t rewriteToVnodeModifyOpStmt(SQuery* pQuery, SArray* pBufArray);
 SArray* serializeVgroupsCreateTableBatch(SHashObj* pVgroupHashmap);
--- a/include/libs/qcom/query.h
+++ b/include/libs/qcom/query.h
@ -163,6 +163,23 @@ typedef struct STargetInfo {
  int32_t     vgId;
 } STargetInfo;

+typedef struct SBoundColInfo {
+  int16_t* pColIndex;  // bound index => schema index
+  int32_t  numOfCols;
+  int32_t  numOfBound;
+} SBoundColInfo;
+
+typedef struct STableDataCxt {
+  STableMeta*    pMeta;
+  STSchema*      pSchema;
+  SBoundColInfo  boundColsInfo;
+  SArray*        pValues;
+  SSubmitTbData* pData;
+  TSKEY          lastTs;
+  bool           ordered;
+  bool           duplicateTs;
+} STableDataCxt;
+
 typedef int32_t (*__async_send_cb_fn_t)(void* param, SDataBuf* pMsg, int32_t code);
 typedef int32_t (*__async_exec_fn_t)(void* param);

@ -238,6 +255,7 @@ int32_t dataConverToStr(char* str, int type, void* buf, int32_t bufSize, int32_t
 char*   parseTagDatatoJson(void* p);
 int32_t cloneTableMeta(STableMeta* pSrc, STableMeta** pDst);
 int32_t cloneDbVgInfo(SDBVgInfo* pSrc, SDBVgInfo** pDst);
+int32_t cloneSVreateTbReq(SVCreateTbReq* pSrc, SVCreateTbReq** pDst);
 void    freeVgInfo(SDBVgInfo* vgInfo);

 extern int32_t (*queryBuildMsg[TDMT_MAX])(void* input, char** msg, int32_t msgSize, int32_t* msgLen,
@ -268,7 +286,7 @@ extern int32_t (*queryProcessMsgRsp[TDMT_MAX])(void* output, char* msg, int32_t
  ((_code) == TSDB_CODE_SYN_NOT_LEADER || (_code) == TSDB_CODE_SYN_RESTORING || (_code) == TSDB_CODE_SYN_INTERNAL_ERROR)
 #define SYNC_OTHER_LEADER_REDIRECT_ERROR(_code) ((_code) == TSDB_CODE_MNODE_NOT_FOUND)

-#define NO_RET_REDIRECT_ERROR(_code) ((_code) == TSDB_CODE_RPC_BROKEN_LINK || (_code) == TSDB_CODE_RPC_NETWORK_UNAVAIL)
+#define NO_RET_REDIRECT_ERROR(_code) ((_code) == TSDB_CODE_RPC_BROKEN_LINK || (_code) == TSDB_CODE_RPC_NETWORK_UNAVAIL || (_code) == TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED)

 #define NEED_REDIRECT_ERROR(_code)                                              \
  (NO_RET_REDIRECT_ERROR(_code) || SYNC_UNKNOWN_LEADER_REDIRECT_ERROR(_code) || \
--- a/include/libs/stream/tstream.h
+++ b/include/libs/stream/tstream.h
@ -103,6 +103,7 @@ typedef struct {
  int8_t type;
 } SStreamQueueItem;

+#if 0
 typedef struct {
  int8_t      type;
  int64_t     ver;
@ -116,6 +117,21 @@ typedef struct {
  SArray* dataRefs;  // SArray<int32_t*>
  SArray* reqs;      // SArray<SSubmitReq*>
 } SStreamMergedSubmit;
+#endif
+
+typedef struct {
+  int8_t      type;
+  int64_t     ver;
+  int32_t*    dataRef;
+  SPackedData submit;
+} SStreamDataSubmit2;
+
+typedef struct {
+  int8_t  type;
+  int64_t ver;
+  SArray* dataRefs;  // SArray<int32_t*>
+  SArray* submits;   // SArray<SPackedSubmit>
+} SStreamMergedSubmit2;

 typedef struct {
  int8_t type;
@ -219,11 +235,11 @@ static FORCE_INLINE void* streamQueueNextItem(SStreamQueue* queue) {
  }
 }

-SStreamDataSubmit* streamDataSubmitNew(SSubmitReq* pReq);
+SStreamDataSubmit2* streamDataSubmitNew(SPackedData submit);

-void streamDataSubmitRefDec(SStreamDataSubmit* pDataSubmit);
+void streamDataSubmitRefDec(SStreamDataSubmit2* pDataSubmit);

-SStreamDataSubmit* streamSubmitRefClone(SStreamDataSubmit* pSubmit);
+SStreamDataSubmit2* streamSubmitRefClone(SStreamDataSubmit2* pSubmit);

 typedef struct {
  char* qmsg;
@ -355,14 +371,15 @@ void         tFreeSStreamTask(SStreamTask* pTask);

 static FORCE_INLINE int32_t streamTaskInput(SStreamTask* pTask, SStreamQueueItem* pItem) {
  if (pItem->type == STREAM_INPUT__DATA_SUBMIT) {
-    SStreamDataSubmit* pSubmitClone = streamSubmitRefClone((SStreamDataSubmit*)pItem);
+    SStreamDataSubmit2* pSubmitClone = streamSubmitRefClone((SStreamDataSubmit2*)pItem);
    if (pSubmitClone == NULL) {
      qDebug("task %d %p submit enqueue failed since out of memory", pTask->taskId, pTask);
      terrno = TSDB_CODE_OUT_OF_MEMORY;
      atomic_store_8(&pTask->inputStatus, TASK_INPUT_STATUS__FAILED);
      return -1;
    }
-    qDebug("task %d %p submit enqueue %p %p %p", pTask->taskId, pTask, pItem, pSubmitClone, pSubmitClone->data);
+    qDebug("task %d %p submit enqueue %p %p %p %d %" PRId64, pTask->taskId, pTask, pItem, pSubmitClone,
+           pSubmitClone->submit.msgStr, pSubmitClone->submit.msgLen, pSubmitClone->submit.ver);
    taosWriteQitem(pTask->inputQueue->queue, pSubmitClone);
    // qStreamInput(pTask->exec.executor, pSubmitClone);
  } else if (pItem->type == STREAM_INPUT__DATA_BLOCK || pItem->type == STREAM_INPUT__DATA_RETRIEVE ||
@ -392,21 +409,6 @@ static FORCE_INLINE void streamTaskInputFail(SStreamTask* pTask) {
  atomic_store_8(&pTask->inputStatus, TASK_INPUT_STATUS__FAILED);
 }

-static FORCE_INLINE int32_t streamTaskOutput(SStreamTask* pTask, SStreamDataBlock* pBlock) {
-  if (pTask->outputType == TASK_OUTPUT__TABLE) {
-    pTask->tbSink.tbSinkFunc(pTask, pTask->tbSink.vnode, 0, pBlock->blocks);
-    taosArrayDestroyEx(pBlock->blocks, (FDelete)blockDataFreeRes);
-    taosFreeQitem(pBlock);
-  } else if (pTask->outputType == TASK_OUTPUT__SMA) {
-    pTask->smaSink.smaSink(pTask->smaSink.vnode, pTask->smaSink.smaId, pBlock->blocks);
-    taosArrayDestroyEx(pBlock->blocks, (FDelete)blockDataFreeRes);
-    taosFreeQitem(pBlock);
-  } else {
-    taosWriteQitem(pTask->outputQueue->queue, pBlock);
-  }
-  return 0;
-}
-
 typedef struct {
  SMsgHead head;
  int64_t  streamId;
@ -584,6 +586,7 @@ int32_t streamProcessRetrieveRsp(SStreamTask* pTask, SStreamRetrieveRsp* pRsp);

 int32_t streamTryExec(SStreamTask* pTask);
 int32_t streamSchedExec(SStreamTask* pTask);
+int32_t streamTaskOutput(SStreamTask* pTask, SStreamDataBlock* pBlock);

 int32_t streamScanExec(SStreamTask* pTask, int32_t batchSz);

--- a/include/os/osMemory.h
+++ b/include/os/osMemory.h
@ -22,12 +22,16 @@ extern "C" {

 // If the error is in a third-party library, place this header file under the third-party library header file.
 // When you want to use this feature, you should find or add the same function in the following sectio
+#if !defined(WINDOWS)
+
 #ifndef ALLOW_FORBID_FUNC
 #define malloc  MALLOC_FUNC_TAOS_FORBID
 #define calloc  CALLOC_FUNC_TAOS_FORBID
 #define realloc REALLOC_FUNC_TAOS_FORBID
 #define free    FREE_FUNC_TAOS_FORBID
-#endif
+#endif // ifndef ALLOW_FORBID_FUNC
+
+#endif // if !defined(WINDOWS)

 void   *taosMemoryMalloc(int64_t size);
 void   *taosMemoryCalloc(int64_t num, int64_t size);
--- a/include/os/osSysinfo.h
+++ b/include/os/osSysinfo.h
@ -70,6 +70,7 @@ typedef struct {

 SysNameInfo taosGetSysNameInfo();
 bool        taosCheckCurrentInDll();
+int         taosGetlocalhostname(char *hostname, size_t maxLen);

 #ifdef __cplusplus
 }
--- a/include/os/osSystem.h
+++ b/include/os/osSystem.h
@ -62,11 +62,38 @@ void    taosResetTerminalMode();
    taosMemoryFree(strings);                                               \
  }
 #else
+#include <windows.h>
+#include <dbghelp.h>
+
+#define STACKSIZE 64
 #define taosPrintTrace(flags, level, dflag)                                                                \
  {                                                                                                        \
-    taosPrintLog(flags, level, dflag,                                                                      \
-                 "backtrace not implemented on windows, so detailed stack information cannot be printed"); \
-    }
+    unsigned int   i;                                                                                                 \
+    void*          stack[STACKSIZE];                                                                                  \
+    unsigned short frames;                                                                                            \
+    SYMBOL_INFO*   symbol;                                                                                            \
+    HANDLE         process;                                                                                           \
+                                                                                                                      \
+    process = GetCurrentProcess();                                                                                    \
+                                                                                                                      \
+    SymInitialize(process, NULL, TRUE);                                                                               \
+                                                                                                                      \
+    frames = CaptureStackBackTrace(0, STACKSIZE, stack, NULL);                                                        \
+    symbol = (SYMBOL_INFO*)calloc(sizeof(SYMBOL_INFO) + 256 * sizeof(char), 1);                                       \
+    if (symbol != NULL) {                                                                                             \
+        symbol->MaxNameLen = 255;                                                                                     \
+        symbol->SizeOfStruct = sizeof(SYMBOL_INFO);                                                                   \
+                                                                                                                      \
+        if (frames > 0) {                                                                                             \
+          taosPrintLog(flags, level, dflag, "obtained %d stack frames", frames);                                      \
+          for (i = 0; i < frames; i++) {                                                                              \
+            SymFromAddr(process, (DWORD64)(stack[i]), 0, symbol);                                                     \
+            taosPrintLog(flags, level, dflag, "frame:%i: %s - 0x%0X", frames - i - 1, symbol->Name, symbol->Address); \
+          }                                                                                                           \
+        }                                                                                                             \
+        free(symbol);                                                                                                 \
+    }                                                                                                                 \
+  }
 #endif

 #ifdef __cplusplus
--- a/include/util/tRealloc.h
+++ b/include/util/tRealloc.h
@ -52,11 +52,13 @@ _exit:
  return code;
 }

-static FORCE_INLINE void tFree(uint8_t *pBuf) {
-  if (pBuf) {
-    taosMemoryFree(pBuf - sizeof(int64_t));
-  }
-}
+#define tFree(BUF)                                        \
+  do {                                                    \
+    if (BUF) {                                            \
+      taosMemoryFree((uint8_t *)(BUF) - sizeof(int64_t)); \
+      (BUF) = NULL;                                       \
+    }                                                     \
+  } while (0)

 #ifdef __cplusplus
 }
--- a/include/util/taoserror.h
+++ b/include/util/taoserror.h
@ -65,6 +65,8 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_RPC_PORT_EADDRINUSE           TAOS_DEF_ERROR_CODE(0, 0x0017) //
 #define TSDB_CODE_RPC_BROKEN_LINK               TAOS_DEF_ERROR_CODE(0, 0x0018) //
 #define TSDB_CODE_RPC_TIMEOUT                   TAOS_DEF_ERROR_CODE(0, 0x0019) //
+#define TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED    TAOS_DEF_ERROR_CODE(0, 0x0020) // "Vgroup could not be connected"
+#define TSDB_CODE_RPC_SOMENODE_BROKEN_LINK      TAOS_DEF_ERROR_CODE(0, 0x0021) // 

 //common & util
 #define TSDB_CODE_OPS_NOT_SUPPORT               TAOS_DEF_ERROR_CODE(0, 0x0100) //
@ -156,6 +158,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_TSC_QUERY_KILLED              TAOS_DEF_ERROR_CODE(0, 0X022D)
 #define TSDB_CODE_TSC_NO_EXEC_NODE              TAOS_DEF_ERROR_CODE(0, 0X022E)
 #define TSDB_CODE_TSC_NOT_STABLE_ERROR          TAOS_DEF_ERROR_CODE(0, 0X022F)
+#define TSDB_CODE_TSC_STMT_CACHE_ERROR          TAOS_DEF_ERROR_CODE(0, 0X0230)

 // mnode-common
 // #define TSDB_CODE_MND_MSG_NOT_PROCESSED      TAOS_DEF_ERROR_CODE(0, 0x0300) // 2.x
@ -517,6 +520,8 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_SYN_STANDBY_NOT_READY         TAOS_DEF_ERROR_CODE(0, 0x0912)
 #define TSDB_CODE_SYN_BATCH_ERROR               TAOS_DEF_ERROR_CODE(0, 0x0913)
 #define TSDB_CODE_SYN_RESTORING                 TAOS_DEF_ERROR_CODE(0, 0x0914)
+#define TSDB_CODE_SYN_INVALID_SNAPSHOT_MSG      TAOS_DEF_ERROR_CODE(0, 0x0915)  // internal
+#define TSDB_CODE_SYN_BUFFER_FULL               TAOS_DEF_ERROR_CODE(0, 0x0916)  //
 #define TSDB_CODE_SYN_INTERNAL_ERROR            TAOS_DEF_ERROR_CODE(0, 0x09FF)

 // tq
--- a/include/util/tarray.h
+++ b/include/util/tarray.h
@ -22,19 +22,6 @@
 extern "C" {
 #endif

-#if 0
-#define TARRAY(TYPE)             \
-  struct {                       \
-    int32_t      tarray_size_;   \
-    int32_t      tarray_neles_;  \
-    struct TYPE* td_array_data_; \
-  }
-
-#define TARRAY_SIZE(ARRAY)        (ARRAY)->tarray_size_
-#define TARRAY_NELES(ARRAY)       (ARRAY)->tarray_neles_
-#define TARRAY_ELE_AT(ARRAY, IDX) ((ARRAY)->td_array_data_ + idx)
-#endif
-
 #define TARRAY_MIN_SIZE               8
 #define TARRAY_GET_ELEM(array, index) ((void*)((char*)((array)->pData) + (index) * (array)->elemSize))
 #define TARRAY_ELEM_IDX(array, ele)   (POINTER_DISTANCE(ele, (array)->pData) / (array)->elemSize)
@ -46,6 +33,9 @@ typedef struct SArray {
  void*    pData;
 } SArray;

+#define TARRAY_SIZE(array) ((array)->size)
+#define TARRAY_DATA(array) ((array)->pData)
+
 /**
 *
 * @param size
@ -194,6 +184,13 @@ void taosArrayPopTailBatch(SArray* pArray, size_t cnt);
 */
 void taosArrayRemove(SArray* pArray, size_t index);

+/**
+ * remove batch entry from the given index
+ * @param pArray
+ * @param index
+ */
+void taosArrayRemoveBatch(SArray* pArray, size_t index, size_t num, FDelete fp);
+
 /**
 * copy the whole array from source to destination
 * @param pDst
--- a/include/util/tdef.h
+++ b/include/util/tdef.h
@ -189,12 +189,13 @@ typedef enum ELogicConditionType {
 #define TSDB_MAX_COLUMNS 4096
 #define TSDB_MIN_COLUMNS 2  // PRIMARY COLUMN(timestamp) + other columns

-#define TSDB_NODE_NAME_LEN  64
-#define TSDB_TABLE_NAME_LEN 193  // it is a null-terminated string
-#define TSDB_TOPIC_NAME_LEN 193  // it is a null-terminated string
-#define TSDB_CGROUP_LEN     193  // it is a null-terminated string
-#define TSDB_DB_NAME_LEN    65
-#define TSDB_DB_FNAME_LEN   (TSDB_ACCT_ID_LEN + TSDB_DB_NAME_LEN + TSDB_NAME_DELIMITER_LEN)
+#define TSDB_NODE_NAME_LEN   64
+#define TSDB_TABLE_NAME_LEN  193                                // it is a null-terminated string
+#define TSDB_TOPIC_NAME_LEN  193                                // it is a null-terminated string
+#define TSDB_CGROUP_LEN      193                                // it is a null-terminated string
+#define TSDB_USER_CGROUP_LEN (TSDB_USER_LEN + TSDB_CGROUP_LEN)  // it is a null-terminated string
+#define TSDB_DB_NAME_LEN     65
+#define TSDB_DB_FNAME_LEN    (TSDB_ACCT_ID_LEN + TSDB_DB_NAME_LEN + TSDB_NAME_DELIMITER_LEN)

 #define TSDB_FUNC_NAME_LEN       65
 #define TSDB_FUNC_COMMENT_LEN    1024 * 1024
@ -254,7 +255,7 @@ typedef enum ELogicConditionType {
 #define TSDB_EP_LEN             (TSDB_FQDN_LEN + 6)
 #define TSDB_IPv4ADDR_LEN       16
 #define TSDB_FILENAME_LEN       128
-#define TSDB_SHOW_SQL_LEN       1024
+#define TSDB_SHOW_SQL_LEN       2048
 #define TSDB_SLOW_QUERY_SQL_LEN 512
 #define TSDB_SHOW_SUBQUERY_LEN  1000

--- a/include/util/tencode.h
+++ b/include/util/tencode.h
@ -116,6 +116,7 @@ static int32_t tEncodeI64v(SEncoder* pCoder, int64_t val);
 static int32_t tEncodeFloat(SEncoder* pCoder, float val);
 static int32_t tEncodeDouble(SEncoder* pCoder, double val);
 static int32_t tEncodeBinary(SEncoder* pCoder, const uint8_t* val, uint32_t len);
+static int32_t tEncodeBinaryEx(SEncoder* pCoder, const uint8_t* val, uint32_t len);
 static int32_t tEncodeCStrWithLen(SEncoder* pCoder, const char* val, uint32_t len);
 static int32_t tEncodeCStr(SEncoder* pCoder, const char* val);

--- a/include/util/tworker.h
+++ b/include/util/tworker.h
@ -17,6 +17,7 @@
 #define _TD_UTIL_WORKER_H_

 #include "tqueue.h"
+#include "tarray.h"

 #ifdef __cplusplus
 extern "C" {
@ -26,10 +27,10 @@ typedef struct SQWorkerPool SQWorkerPool;
 typedef struct SWWorkerPool SWWorkerPool;

 typedef struct SQWorker {
-  int32_t       id;      // worker id
-  int64_t       pid;     // thread pid
-  TdThread      thread;  // thread id
-  SQWorkerPool *pool;
+  int32_t  id;      // worker id
+  int64_t  pid;     // thread pid
+  TdThread thread;  // thread id
+  void    *pool;
 } SQWorker;

 typedef struct SQWorkerPool {
@ -42,6 +43,14 @@ typedef struct SQWorkerPool {
  TdThreadMutex mutex;
 } SQWorkerPool;

+typedef struct SAutoQWorkerPool {
+  float         ratio;
+  STaosQset    *qset;
+  const char   *name;
+  SArray       *workers;
+  TdThreadMutex mutex;
+} SAutoQWorkerPool;
+
 typedef struct SWWorker {
  int32_t       id;      // worker id
  int64_t       pid;     // thread pid
@ -65,6 +74,11 @@ void        tQWorkerCleanup(SQWorkerPool *pool);
 STaosQueue *tQWorkerAllocQueue(SQWorkerPool *pool, void *ahandle, FItem fp);
 void        tQWorkerFreeQueue(SQWorkerPool *pool, STaosQueue *queue);

+int32_t     tAutoQWorkerInit(SAutoQWorkerPool *pool);
+void        tAutoQWorkerCleanup(SAutoQWorkerPool *pool);
+STaosQueue *tAutoQWorkerAllocQueue(SAutoQWorkerPool *pool, void *ahandle, FItem fp);
+void        tAutoQWorkerFreeQueue(SAutoQWorkerPool *pool, STaosQueue *queue);
+
 int32_t     tWWorkerInit(SWWorkerPool *pool);
 void        tWWorkerCleanup(SWWorkerPool *pool);
 STaosQueue *tWWorkerAllocQueue(SWWorkerPool *pool, void *ahandle, FItems fp);
--- a/packaging/tools/make_install.bat
+++ b/packaging/tools/make_install.bat
@ -14,6 +14,7 @@ set binary_dir=%3
 set binary_dir=%binary_dir:/=\\%
 set osType=%4
 set verNumber=%5
+set Enterprise=%6
 set target_dir=C:\\TDengine

 if not exist %target_dir% (
@ -57,7 +58,33 @@ if exist %binary_dir%\\build\\lib\\taosws.dll (
 if exist %binary_dir%\\build\\bin\\taosdump.exe (
    copy %binary_dir%\\build\\bin\\taosdump.exe %target_dir% > nul
 )
-
+if %Enterprise% == TRUE (
+    if exist %binary_dir%\\build\\bin\\taosx.exe (
+        copy %binary_dir%\\build\\bin\\taosx.exe %target_dir% > nul
+    )
+    if exist %binary_dir%\\build\\bin\\tmq_sim.exe (
+        copy %binary_dir%\\build\\bin\\tmq_sim.exe %target_dir% > nul
+    )
+    if exist %binary_dir%\\build\\bin\\tsim.exe (
+        copy %binary_dir%\\build\\bin\\tsim.exe %target_dir% > nul
+    )
+    if exist %binary_dir%\\build\\bin\\tmq_taosx_ci.exe (
+        copy %binary_dir%\\build\\bin\\tmq_taosx_ci.exe %target_dir% > nul
+    )
+    if exist %binary_dir%\\build\\bin\\tmq_demo.exe (
+        copy %binary_dir%\\build\\bin\\tmq_demo.exe %target_dir% > nul
+    )
+    if exist %binary_dir%\\build\\bin\\dumper.exe (
+        copy %binary_dir%\\build\\bin\\dumper.exe %target_dir% > nul
+    )
+    if exist %binary_dir%\\build\\bin\\runUdf.exe (
+        copy %binary_dir%\\build\\bin\\runUdf.exe %target_dir% > nul
+    )
+    if exist %binary_dir%\\build\\bin\\create_table.exe (
+        copy %binary_dir%\\build\\bin\\create_table.exe %target_dir% > nul
+    )
+)
+    
 copy %binary_dir%\\build\\bin\\taosd.exe %target_dir% > nul
 copy %binary_dir%\\build\\bin\\udfd.exe %target_dir% > nul

--- a/packaging/tools/makepkg.sh
+++ b/packaging/tools/makepkg.sh
@ -348,7 +348,8 @@ cd ${release_dir}
 #  install_dir has been distinguishes  cluster from  edege, so comments this code
 pkg_name=${install_dir}-${osType}-${cpuType}

-taostools_pkg_name=${taostools_install_dir}-${osType}-${cpuType}
+versionCompFirst=$(echo ${versionComp} | awk -F '.' '{print $1}')
+taostools_pkg_name=${taostools_install_dir}-${osType}-${cpuType}-comp${versionCompFirst}

 # if [ "$verMode" == "cluster" ]; then
 #   pkg_name=${install_dir}-${osType}-${cpuType}
--- a/source/client/inc/clientInt.h
+++ b/source/client/inc/clientInt.h
@ -149,7 +149,6 @@ typedef struct STscObj {
  int32_t       numOfReqs;  // number of sqlObj bound to this connection
  SAppInstInfo* pAppInfo;
  SHashObj*     pRequests;
-  int8_t        schemalessType;  // todo remove it, this attribute should be move to request
 } STscObj;

 typedef struct SResultColumn {
--- a/source/client/inc/clientLog.h
+++ b/source/client/inc/clientLog.h
@ -30,7 +30,7 @@ extern "C" {
 #define tscDebug(...)  do { if (cDebugFlag & DEBUG_DEBUG) { taosPrintLog("TSC ", DEBUG_DEBUG, cDebugFlag, __VA_ARGS__); }} while(0)
 #define tscTrace(...)  do { if (cDebugFlag & DEBUG_TRACE) { taosPrintLog("TSC ", DEBUG_TRACE, cDebugFlag, __VA_ARGS__); }} while(0)
 #define tscDebugL(...) do { if (cDebugFlag & DEBUG_DEBUG) { taosPrintLongString("TSC ", DEBUG_DEBUG, cDebugFlag, __VA_ARGS__); }} while(0)
-//#define tscPerf(...)   do { if (cDebugFlag & DEBUG_INFO)  { taosPrintLog("TSC ", DEBUG_INFO, cDebugFlag, __VA_ARGS__); }} while(0)
+#define tscPerf(...)   do { if (cDebugFlag & DEBUG_INFO)  { taosPrintLog("TSC ", 0, cDebugFlag, __VA_ARGS__); }} while(0)
 // clang-format on

 #ifdef __cplusplus
--- a/source/client/inc/clientSml.h
+++ b/source/client/inc/clientSml.h
@ -0,0 +1,238 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+#ifndef TDENGINE_CLIENTSML_H
+#define TDENGINE_CLIENTSML_H
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "catalog.h"
+#include "clientInt.h"
+#include "osThread.h"
+#include "query.h"
+#include "taos.h"
+#include "taoserror.h"
+#include "tcommon.h"
+#include "tdef.h"
+#include "tglobal.h"
+#include "tlog.h"
+#include "tmsg.h"
+#include "tname.h"
+#include "ttime.h"
+#include "ttypes.h"
+#include "cJSON.h"
+
+#if (defined(__GNUC__) && (__GNUC__ >= 3)) || (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 800)) || defined(__clang__)
+#  define expect(expr,value)    (__builtin_expect ((expr),(value)) )
+#else
+#  define expect(expr,value)    (expr)
+#endif
+
+#ifndef likely
+#define likely(expr)     expect((expr) != 0, 1)
+#endif
+#ifndef unlikely
+#define unlikely(expr)   expect((expr) != 0, 0)
+#endif
+
+#define SPACE ' '
+#define COMMA ','
+#define EQUAL '='
+#define QUOTE '"'
+#define SLASH '\\'
+
+#define JUMP_SPACE(sql, sqlEnd) \
+  while (sql < sqlEnd) {        \
+    if (unlikely(*sql == SPACE))          \
+      sql++;                    \
+    else                        \
+      break;                    \
+  }
+
+#define IS_INVALID_COL_LEN(len)   ((len) <= 0 || (len) >= TSDB_COL_NAME_LEN)
+#define IS_INVALID_TABLE_LEN(len) ((len) <= 0 || (len) >= TSDB_TABLE_NAME_LEN)
+
+#define TS        "_ts"
+#define TS_LEN    3
+#define VALUE     "_value"
+#define VALUE_LEN 6
+
+#define MAX_RETRY_TIMES 5
+typedef TSDB_SML_PROTOCOL_TYPE SMLProtocolType;
+
+typedef enum {
+  SCHEMA_ACTION_NULL,
+  SCHEMA_ACTION_CREATE_STABLE,
+  SCHEMA_ACTION_ADD_COLUMN,
+  SCHEMA_ACTION_ADD_TAG,
+  SCHEMA_ACTION_CHANGE_COLUMN_SIZE,
+  SCHEMA_ACTION_CHANGE_TAG_SIZE,
+} ESchemaAction;
+
+typedef struct {
+  const void  *key;
+  int32_t      keyLen;
+  void        *value;
+  bool         used;
+}Node;
+
+typedef struct NodeList{
+  Node             data;
+  struct NodeList* next;
+}NodeList;
+
+typedef struct {
+  char *measure;
+  char *tags;
+  char *cols;
+  char *timestamp;
+
+  int32_t measureLen;
+  int32_t measureTagsLen;
+  int32_t tagsLen;
+  int32_t colsLen;
+  int32_t timestampLen;
+
+  SArray *colArray;
+} SSmlLineInfo;
+
+typedef struct {
+  const char *sTableName;  // super table name
+  int32_t     sTableNameLen;
+  char        childTableName[TSDB_TABLE_NAME_LEN];
+  uint64_t    uid;
+  void       *key;        // for openTsdb
+
+  SArray *tags;
+
+  // elements are SHashObj<cols key string, SSmlKv*> for find by key quickly
+  SArray *cols;
+  STableDataCxt *tableDataCtx;
+} SSmlTableInfo;
+
+typedef struct {
+  SArray   *tags;     // save the origin order to create table
+  SHashObj *tagHash;  // elements are <key, index in tags>
+
+  SArray   *cols;
+  SHashObj *colHash;
+
+  STableMeta *tableMeta;
+} SSmlSTableMeta;
+
+typedef struct {
+  int32_t len;
+  char   *buf;
+} SSmlMsgBuf;
+
+typedef struct {
+  int32_t code;
+  int32_t lineNum;
+
+  int32_t numOfSTables;
+  int32_t numOfCTables;
+  int32_t numOfCreateSTables;
+  int32_t numOfAlterColSTables;
+  int32_t numOfAlterTagSTables;
+
+  int64_t parseTime;
+  int64_t schemaTime;
+  int64_t insertBindTime;
+  int64_t insertRpcTime;
+  int64_t endTime;
+} SSmlCostInfo;
+
+typedef struct {
+  int64_t id;
+
+  SMLProtocolType protocol;
+  int8_t          precision;
+  bool            reRun;
+  bool            dataFormat;  // true means that the name and order of keys in each line are the same(only for influx protocol)
+  bool            isRawLine;
+  int32_t         ttl;
+
+  NodeList *childTables;
+  NodeList *superTables;
+  SHashObj *pVgHash;
+
+  STscObj     *taos;
+  SCatalog    *pCatalog;
+  SRequestObj *pRequest;
+  SQuery      *pQuery;
+
+  SSmlCostInfo cost;
+  int32_t      lineNum;
+  SSmlMsgBuf   msgBuf;
+
+//  cJSON       *root;  // for parse json
+  int8_t             offset[4];
+  SSmlLineInfo      *lines; // element is SSmlLineInfo
+
+  //
+  SArray      *preLineTagKV;
+  SArray      *preLineColKV;
+
+  SSmlLineInfo preLine;
+  STableMeta  *currSTableMeta;
+  STableDataCxt *currTableDataCtx;
+  bool         needModifySchema;
+} SSmlHandle;
+
+#define IS_SAME_CHILD_TABLE (elements->measureTagsLen == info->preLine.measureTagsLen \
+&& memcmp(elements->measure, info->preLine.measure, elements->measureTagsLen) == 0)
+
+#define IS_SAME_SUPER_TABLE (elements->measureLen == info->preLine.measureLen \
+&& memcmp(elements->measure, info->preLine.measure, elements->measureLen) == 0)
+
+#define IS_SAME_KEY (preKV->keyLen == kv.keyLen && memcmp(preKV->key, kv.key, kv.keyLen) == 0)
+
+extern int64_t smlFactorNS[3];
+extern int64_t smlFactorS[3];
+
+typedef int32_t (*_equal_fn_sml)(const void *, const void *);
+
+SSmlHandle   *smlBuildSmlInfo(TAOS *taos);
+void          smlDestroyInfo(SSmlHandle *info);
+void          smlJsonParseObjFirst(char **start, SSmlLineInfo *element, int8_t *offset);
+void          smlJsonParseObj(char **start, SSmlLineInfo *element, int8_t *offset);
+SArray       *smlJsonParseTags(char *start, char *end);
+bool          smlParseNumberOld(SSmlKv *kvVal, SSmlMsgBuf *msg);
+void*         nodeListGet(NodeList* list, const void *key, int32_t len, _equal_fn_sml fn);
+int           nodeListSet(NodeList** list, const void *key, int32_t len, void* value, _equal_fn_sml fn);
+int           nodeListSize(NodeList* list);
+bool          smlDoubleToInt64OverFlow(double num);
+int32_t       smlBuildInvalidDataMsg(SSmlMsgBuf *pBuf, const char *msg1, const char *msg2);
+bool          smlParseNumber(SSmlKv *kvVal, SSmlMsgBuf *msg);
+int64_t       smlGetTimeValue(const char *value, int32_t len, uint8_t fromPrecision, uint8_t toPrecision);
+int8_t        smlGetTsTypeByLen(int32_t len);
+SSmlTableInfo*    smlBuildTableInfo(int numRows, const char* measure, int32_t measureLen);
+SSmlSTableMeta*   smlBuildSTableMeta(bool isDataFormat);
+int32_t           smlSetCTableName(SSmlTableInfo *oneTable);
+STableMeta*       smlGetMeta(SSmlHandle *info, const void* measure, int32_t measureLen);
+int32_t           is_same_child_table_telnet(const void *a, const void *b);
+int64_t           smlParseOpenTsdbTime(SSmlHandle *info, const char *data, int32_t len);
+int32_t           smlClearForRerun(SSmlHandle *info);
+int32_t           smlParseValue(SSmlKv *pVal, SSmlMsgBuf *msg);
+
+int32_t smlParseInfluxString(SSmlHandle *info, char *sql, char *sqlEnd, SSmlLineInfo *elements);
+int32_t smlParseTelnetString(SSmlHandle *info, char *sql, char *sqlEnd, SSmlLineInfo *elements);
+int32_t smlParseJSON(SSmlHandle *info, char *payload);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // TDENGINE_CLIENTSML_H
--- a/source/client/inc/clientStmt.h
+++ b/source/client/inc/clientStmt.h
@ -21,8 +21,6 @@ extern "C" {
 #endif
 #include "catalog.h"

-typedef void STableDataBlocks;
-
 typedef enum {
  STMT_TYPE_INSERT = 1,
  STMT_TYPE_MULTI_INSERT,
@ -43,8 +41,8 @@ typedef enum {
 } STMT_STATUS;

 typedef struct SStmtTableCache {
-  STableDataBlocks *pDataBlock;
-  void             *boundTags;
+  STableDataCxt *pDataCtx;
+  void          *boundTags;
 } SStmtTableCache;

 typedef struct SStmtQueryResInfo {
@ -71,10 +69,11 @@ typedef struct SStmtBindInfo {
 } SStmtBindInfo;

 typedef struct SStmtExecInfo {
-  int32_t      affectedRows;
-  SRequestObj *pRequest;
-  SHashObj    *pBlockHash;
-  bool         autoCreateTbl;
+  int32_t        affectedRows;
+  SRequestObj   *pRequest;
+  SHashObj      *pBlockHash;
+  STableDataCxt *pCurrBlock;
+  SSubmitTbData *pCurrTbData;
 } SStmtExecInfo;

 typedef struct SStmtSQLInfo {
--- a/source/client/src/clientEnv.c
+++ b/source/client/src/clientEnv.c
@ -76,13 +76,19 @@ static void deregisterRequest(SRequestObj *pRequest) {
           "current:%d, app current:%d",
           pRequest->self, pTscObj->id, pRequest->requestId, duration / 1000.0, num, currentInst);

+  tscPerf("insert duration %" PRId64 "us: syntax:%" PRId64 "us, ctg:%" PRId64 "us, semantic:%" PRId64
+          "us, exec:%" PRId64 "us, stmtType:%d",
+          duration, pRequest->metric.syntaxEnd - pRequest->metric.syntaxStart,
+          pRequest->metric.ctgEnd - pRequest->metric.ctgStart, pRequest->metric.semanticEnd - pRequest->metric.ctgEnd,
+          pRequest->metric.execEnd - pRequest->metric.semanticEnd, pRequest->stmtType);
+
  if (QUERY_NODE_VNODE_MODIFY_STMT == pRequest->stmtType) {
-    //    tscPerf("insert duration %" PRId64 "us: syntax:%" PRId64 "us, ctg:%" PRId64 "us, semantic:%" PRId64
-    //            "us, exec:%" PRId64 "us",
-    //            duration, pRequest->metric.syntaxEnd - pRequest->metric.syntaxStart,
-    //            pRequest->metric.ctgEnd - pRequest->metric.ctgStart, pRequest->metric.semanticEnd -
-    //            pRequest->metric.ctgEnd, pRequest->metric.execEnd - pRequest->metric.semanticEnd);
-    atomic_add_fetch_64((int64_t *)&pActivity->insertElapsedTime, duration);
+    //        tscPerf("insert duration %" PRId64 "us: syntax:%" PRId64 "us, ctg:%" PRId64 "us, semantic:%" PRId64
+    //                "us, exec:%" PRId64 "us",
+    //                duration, pRequest->metric.syntaxEnd - pRequest->metric.syntaxStart,
+    //                pRequest->metric.ctgEnd - pRequest->metric.ctgStart, pRequest->metric.semanticEnd -
+    //                pRequest->metric.ctgEnd, pRequest->metric.execEnd - pRequest->metric.semanticEnd);
+    //    atomic_add_fetch_64((int64_t *)&pActivity->insertElapsedTime, duration);
  } else if (QUERY_NODE_SELECT_STMT == pRequest->stmtType) {
    //    tscPerf("select duration %" PRId64 "us: syntax:%" PRId64 "us, ctg:%" PRId64 "us, semantic:%" PRId64
    //            "us, planner:%" PRId64 "us, exec:%" PRId64 "us, reqId:0x%" PRIx64,
@ -264,7 +270,6 @@ void *createTscObj(const char *user, const char *auth, const char *db, int32_t c

  taosThreadMutexInit(&pObj->mutex, NULL);
  pObj->id = taosAddRef(clientConnRefPool, pObj);
-  pObj->schemalessType = 1;

  atomic_add_fetch_64(&pObj->pAppInfo->numOfConns, 1);

--- a/source/client/src/clientImpl.c
+++ b/source/client/src/clientImpl.c
@ -239,7 +239,6 @@ int32_t parseSql(SRequestObj* pRequest, bool topicQuery, SQuery** pQuery, SStmtC
                       .pTransporter = pTscObj->pAppInfo->pTransporter,
                       .pStmtCb = pStmtCb,
                       .pUser = pTscObj->user,
-                       .schemalessType = pTscObj->schemalessType,
                       .isSuperUser = (0 == strcmp(pTscObj->user, TSDB_DEFAULT_USER)),
                       .enableSysInfo = pTscObj->sysInfo,
                       .svrVer = pTscObj->sVer,
@ -741,47 +740,21 @@ int32_t scheduleQuery(SRequestObj* pRequest, SQueryPlan* pDag, SArray* pNodeList
 }

 int32_t handleSubmitExecRes(SRequestObj* pRequest, void* res, SCatalog* pCatalog, SEpSet* epset) {
-  int32_t     code = 0;
-  SArray*     pArray = NULL;
-  SSubmitRsp* pRsp = (SSubmitRsp*)res;
-  if (pRsp->nBlocks <= 0) {
-    taosMemoryFreeClear(pRsp->pBlocks);
+  SArray*      pArray = NULL;
+  SSubmitRsp2* pRsp = (SSubmitRsp2*)res;
+  if (NULL == pRsp->aCreateTbRsp) {
    return TSDB_CODE_SUCCESS;
  }

-  pArray = taosArrayInit(pRsp->nBlocks, sizeof(STbSVersion));
-  if (NULL == pArray) {
-    terrno = TSDB_CODE_OUT_OF_MEMORY;
-    return TSDB_CODE_OUT_OF_MEMORY;
+  int32_t tbNum = taosArrayGetSize(pRsp->aCreateTbRsp);
+  for (int32_t i = 0; i < tbNum; ++i) {
+    SVCreateTbRsp* pTbRsp = (SVCreateTbRsp*)taosArrayGet(pRsp->aCreateTbRsp, i);
+    if (pTbRsp->pMeta) {
+      handleCreateTbExecRes(pTbRsp->pMeta, pCatalog);
+    }
  }

-  for (int32_t i = 0; i < pRsp->nBlocks; ++i) {
-    SSubmitBlkRsp* blk = pRsp->pBlocks + i;
-    if (blk->pMeta) {
-      handleCreateTbExecRes(blk->pMeta, pCatalog);
-      tFreeSTableMetaRsp(blk->pMeta);
-      taosMemoryFreeClear(blk->pMeta);
-    }
-
-    if (NULL == blk->tblFName || 0 == blk->tblFName[0]) {
-      continue;
-    }
-
-    STbSVersion tbSver = {.tbFName = blk->tblFName, .sver = blk->sver};
-    taosArrayPush(pArray, &tbSver);
-  }
-
-  SRequestConnInfo conn = {.pTrans = pRequest->pTscObj->pAppInfo->pTransporter,
-                           .requestId = pRequest->requestId,
-                           .requestObjRefId = pRequest->self,
-                           .mgmtEps = *epset};
-
-  code = catalogChkTbMetaVersion(pCatalog, &conn, pArray);
-
-_return:
-
-  taosArrayDestroy(pArray);
-  return code;
+  return TSDB_CODE_SUCCESS;
 }

 int32_t handleQueryExecRes(SRequestObj* pRequest, void* res, SCatalog* pCatalog, SEpSet* epset) {
@ -1430,6 +1403,21 @@ void processMsgFromServer(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet) {
    memcpy((void*)tEpSet, (void*)pEpSet, sizeof(SEpSet));
  }

+  // pMsg is response msg
+  if (pMsg->msgType == TDMT_MND_CONNECT + 1) {
+    // restore origin code
+    if (pMsg->code == TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED) {
+      pMsg->code = TSDB_CODE_RPC_NETWORK_UNAVAIL;
+    } else if (pMsg->code == TSDB_CODE_RPC_SOMENODE_BROKEN_LINK) {
+      pMsg->code = TSDB_CODE_RPC_BROKEN_LINK;
+    }
+  } else {
+    // uniform to one error code: TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED
+    if (pMsg->code == TSDB_CODE_RPC_SOMENODE_BROKEN_LINK) {
+      pMsg->code = TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED;
+    }
+  }
+
  AsyncArg* arg = taosMemoryCalloc(1, sizeof(AsyncArg));
  arg->msg = *pMsg;
  arg->pEpset = tEpSet;
--- a/source/client/src/clientMain.c
+++ b/source/client/src/clientMain.c
@ -866,7 +866,6 @@ int32_t createParseContext(const SRequestObj *pRequest, SParseContext **pCxt) {
                           .pTransporter = pTscObj->pAppInfo->pTransporter,
                           .pStmtCb = NULL,
                           .pUser = pTscObj->user,
-                           .schemalessType = pTscObj->schemalessType,
                           .isSuperUser = (0 == strcmp(pTscObj->user, TSDB_DEFAULT_USER)),
                           .enableSysInfo = pTscObj->sysInfo,
                           .async = true,
--- a/source/client/src/clientRawBlockWrite.c
+++ b/source/client/src/clientRawBlockWrite.c
@ -1240,22 +1240,12 @@ static int32_t taosAlterTable(TAOS* taos, void* meta, int32_t metaLen) {
  return code;
 }

-typedef struct {
-  SVgroupInfo vg;
-  void*       data;
-} VgData;
-
-static void destroyVgHash(void* data) {
-  VgData* vgData = (VgData*)data;
-  taosMemoryFreeClear(vgData->data);
-}
-
 int taos_write_raw_block_with_fields(TAOS* taos, int rows, char* pData, const char* tbname, TAOS_FIELD* fields,
                                     int numFields) {
  int32_t     code = TSDB_CODE_SUCCESS;
  STableMeta* pTableMeta = NULL;
  SQuery*     pQuery = NULL;
-  SSubmitReq* subReq = NULL;
+  SHashObj*   pVgHash = NULL;

  SRequestObj* pRequest = (SRequestObj*)createRequest(*(int64_t*)taos, TSDB_SQL_INSERT, 0);
  if (!pRequest) {
@ -1300,149 +1290,25 @@ int taos_write_raw_block_with_fields(TAOS* taos, int rows, char* pData, const ch
    uError("WriteRaw:catalogGetTableMeta failed. table name: %s", tbname);
    goto end;
  }
-  uint64_t suid = (TSDB_NORMAL_TABLE == pTableMeta->tableType ? 0 : pTableMeta->suid);
-  uint64_t uid = pTableMeta->uid;
-  int32_t  numOfCols = pTableMeta->tableInfo.numOfColumns;
-
-  uint16_t fLen = 0;
-  int32_t  rowSize = 0;
-  int16_t  nVar = 0;
-  for (int i = 0; i < pTableMeta->tableInfo.numOfColumns; i++) {
-    SSchema* schema = pTableMeta->schema + i;
-    fLen += TYPE_BYTES[schema->type];
-    rowSize += schema->bytes;
-    if (IS_VAR_DATA_TYPE(schema->type)) {
-      nVar++;
-    }
-  }
-
-  fLen -= sizeof(TSKEY);
-
-  int32_t extendedRowSize = rowSize + TD_ROW_HEAD_LEN - sizeof(TSKEY) + nVar * sizeof(VarDataOffsetT) +
-                            (int32_t)TD_BITMAP_BYTES(numOfCols - 1);
-  int32_t schemaLen = 0;
-  int32_t submitLen = sizeof(SSubmitBlk) + schemaLen + rows * extendedRowSize;
-
-  int32_t totalLen = sizeof(SSubmitReq) + submitLen;
-  subReq = taosMemoryCalloc(1, totalLen);
-  SSubmitBlk* blk = POINTER_SHIFT(subReq, sizeof(SSubmitReq));
-  void*       blkSchema = POINTER_SHIFT(blk, sizeof(SSubmitBlk));
-  STSRow*     rowData = POINTER_SHIFT(blkSchema, schemaLen);
-
-  SRowBuilder rb = {0};
-  tdSRowInit(&rb, pTableMeta->sversion);
-  tdSRowSetTpInfo(&rb, numOfCols, fLen);
-  int32_t dataLen = 0;
-
-  // | version | total length | total rows | total columns | flag seg| block group id | column schema | each column
-  // length |
-  char*    pStart = pData + getVersion1BlockMetaSize(pData, numFields);
-  int32_t* colLength = (int32_t*)pStart;
-  pStart += sizeof(int32_t) * numFields;
-
-  SResultColumn* pCol = taosMemoryCalloc(numFields, sizeof(SResultColumn));
-
-  for (int32_t i = 0; i < numFields; ++i) {
-    if (IS_VAR_DATA_TYPE(fields[i].type)) {
-      pCol[i].offset = (int32_t*)pStart;
-      pStart += rows * sizeof(int32_t);
-    } else {
-      pCol[i].nullbitmap = pStart;
-      pStart += BitmapLen(rows);
-    }
-
-    pCol[i].pData = pStart;
-    pStart += colLength[i];
-  }
-
-  SHashObj* schemaHash = taosHashInit(16, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), false, HASH_NO_LOCK);
-  for (int i = 0; i < numFields; i++) {
-    TAOS_FIELD* schema = &fields[i];
-    taosHashPut(schemaHash, schema->name, strlen(schema->name), &i, sizeof(int32_t));
-  }
-
-  for (int32_t j = 0; j < rows; j++) {
-    tdSRowResetBuf(&rb, rowData);
-    int32_t offset = 0;
-    for (int32_t k = 0; k < numOfCols; k++) {
-      const SSchema* pColumn = &pTableMeta->schema[k];
-      int32_t*       index = taosHashGet(schemaHash, pColumn->name, strlen(pColumn->name));
-      if (!index) {  // add none
-        tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NONE, NULL, false, offset, k);
-      } else {
-        if (IS_VAR_DATA_TYPE(pColumn->type)) {
-          if (pCol[*index].offset[j] != -1) {
-            char* data = pCol[*index].pData + pCol[*index].offset[j];
-            tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NORM, data, true, offset, k);
-          } else {
-            tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NULL, NULL, false, offset, k);
-          }
-        } else {
-          if (!colDataIsNull_f(pCol[*index].nullbitmap, j)) {
-            char* data = pCol[*index].pData + pColumn->bytes * j;
-            tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NORM, data, true, offset, k);
-          } else {
-            tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NULL, NULL, false, offset, k);
-          }
-        }
-      }
-
-      if (pColumn->colId != PRIMARYKEY_TIMESTAMP_COL_ID) {
-        offset += TYPE_BYTES[pColumn->type];
-      }
-    }
-    tdSRowEnd(&rb);
-    int32_t rowLen = TD_ROW_LEN(rowData);
-    rowData = POINTER_SHIFT(rowData, rowLen);
-    dataLen += rowLen;
-  }
-
-  taosHashCleanup(schemaHash);
-  taosMemoryFree(pCol);
-
-  blk->uid = htobe64(uid);
-  blk->suid = htobe64(suid);
-  blk->sversion = htonl(pTableMeta->sversion);
-  blk->schemaLen = htonl(schemaLen);
-  blk->numOfRows = htonl(rows);
-  blk->dataLen = htonl(dataLen);
-  subReq->length = sizeof(SSubmitReq) + sizeof(SSubmitBlk) + schemaLen + dataLen;
-  subReq->numOfBlocks = 1;
-
-  pQuery = (SQuery*)nodesMakeNode(QUERY_NODE_QUERY);
-  if (NULL == pQuery) {
-    uError("create SQuery error");
+  pQuery = smlInitHandle();
+  if (pQuery == NULL) {
    code = TSDB_CODE_OUT_OF_MEMORY;
    goto end;
  }
-  pQuery->execMode = QUERY_EXEC_MODE_SCHEDULE;
-  pQuery->haveResultSet = false;
-  pQuery->msgType = TDMT_VND_SUBMIT;
-  pQuery->pRoot = (SNode*)nodesMakeNode(QUERY_NODE_VNODE_MODIFY_STMT);
-  if (NULL == pQuery->pRoot) {
-    uError("create pQuery->pRoot error");
-    code = TSDB_CODE_OUT_OF_MEMORY;
-    goto end;
-  }
-  SVnodeModifyOpStmt* nodeStmt = (SVnodeModifyOpStmt*)(pQuery->pRoot);
-  nodeStmt->pDataBlocks = taosArrayInit(1, POINTER_BYTES);
+  pVgHash = taosHashInit(16, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_NO_LOCK);
+  taosHashPut(pVgHash, (const char*)&vgData.vgId, sizeof(vgData.vgId), (char*)&vgData, sizeof(vgData));

-  SVgDataBlocks* dst = taosMemoryCalloc(1, sizeof(SVgDataBlocks));
-  if (NULL == dst) {
-    code = TSDB_CODE_OUT_OF_MEMORY;
+  code = rawBlockBindData(pQuery, pTableMeta, pData, NULL, fields, numFields);
+  if (code != TSDB_CODE_SUCCESS) {
+    uError("WriteRaw:rawBlockBindData failed");
    goto end;
  }
-  dst->vg = vgData;
-  dst->numOfTables = subReq->numOfBlocks;
-  dst->size = subReq->length;
-  dst->pData = (char*)subReq;
-  subReq->header.vgId = htonl(dst->vg.vgId);
-  subReq->version = htonl(1);
-  subReq->header.contLen = htonl(subReq->length);
-  subReq->length = htonl(subReq->length);
-  subReq->numOfBlocks = htonl(subReq->numOfBlocks);
-  subReq = NULL;  // no need free
-  taosArrayPush(nodeStmt->pDataBlocks, &dst);
+
+  code = smlBuildOutput(pQuery, pVgHash);
+  if (code != TSDB_CODE_SUCCESS) {
+    uError("smlBuildOutput failed");
+    return code;
+  }

  launchQueryImpl(pRequest, pQuery, true, NULL);
  code = pRequest->code;
@ -1450,7 +1316,8 @@ int taos_write_raw_block_with_fields(TAOS* taos, int rows, char* pData, const ch
 end:
  taosMemoryFreeClear(pTableMeta);
  qDestroyQuery(pQuery);
-  taosMemoryFree(subReq);
+  destroyRequest(pRequest);
+  taosHashCleanup(pVgHash);
  return code;
 }

@ -1458,7 +1325,7 @@ int taos_write_raw_block(TAOS* taos, int rows, char* pData, const char* tbname)
  int32_t     code = TSDB_CODE_SUCCESS;
  STableMeta* pTableMeta = NULL;
  SQuery*     pQuery = NULL;
-  SSubmitReq* subReq = NULL;
+  SHashObj*   pVgHash = NULL;

  SRequestObj* pRequest = (SRequestObj*)createRequest(*(int64_t*)taos, TSDB_SQL_INSERT, 0);
  if (!pRequest) {
@ -1503,138 +1370,25 @@ int taos_write_raw_block(TAOS* taos, int rows, char* pData, const char* tbname)
    uError("WriteRaw:catalogGetTableMeta failed. table name: %s", tbname);
    goto end;
  }
-  uint64_t suid = (TSDB_NORMAL_TABLE == pTableMeta->tableType ? 0 : pTableMeta->suid);
-  uint64_t uid = pTableMeta->uid;
-  int32_t  numOfCols = pTableMeta->tableInfo.numOfColumns;
-
-  uint16_t fLen = 0;
-  int32_t  rowSize = 0;
-  int16_t  nVar = 0;
-  for (int i = 0; i < numOfCols; i++) {
-    SSchema* schema = pTableMeta->schema + i;
-    fLen += TYPE_BYTES[schema->type];
-    rowSize += schema->bytes;
-    if (IS_VAR_DATA_TYPE(schema->type)) {
-      nVar++;
-    }
-  }
-  fLen -= sizeof(TSKEY);
-
-  int32_t extendedRowSize = rowSize + TD_ROW_HEAD_LEN - sizeof(TSKEY) + nVar * sizeof(VarDataOffsetT) +
-                            (int32_t)TD_BITMAP_BYTES(numOfCols - 1);
-  int32_t schemaLen = 0;
-  int32_t submitLen = sizeof(SSubmitBlk) + schemaLen + rows * extendedRowSize;
-
-  int32_t totalLen = sizeof(SSubmitReq) + submitLen;
-  subReq = taosMemoryCalloc(1, totalLen);
-  SSubmitBlk* blk = POINTER_SHIFT(subReq, sizeof(SSubmitReq));
-  void*       blkSchema = POINTER_SHIFT(blk, sizeof(SSubmitBlk));
-  STSRow*     rowData = POINTER_SHIFT(blkSchema, schemaLen);
-
-  SRowBuilder rb = {0};
-  tdSRowInit(&rb, pTableMeta->sversion);
-  tdSRowSetTpInfo(&rb, numOfCols, fLen);
-  int32_t dataLen = 0;
-
-  // | version | total length | total rows | total columns | flag seg| block group id | column schema | each column
-  // length |
-  char*    pStart = pData + getVersion1BlockMetaSize(pData, numOfCols);
-  int32_t* colLength = (int32_t*)pStart;
-  pStart += sizeof(int32_t) * numOfCols;
-
-  SResultColumn* pCol = taosMemoryCalloc(numOfCols, sizeof(SResultColumn));
-
-  for (int32_t i = 0; i < numOfCols; ++i) {
-    if (IS_VAR_DATA_TYPE(pTableMeta->schema[i].type)) {
-      pCol[i].offset = (int32_t*)pStart;
-      pStart += rows * sizeof(int32_t);
-    } else {
-      pCol[i].nullbitmap = pStart;
-      pStart += BitmapLen(rows);
-    }
-
-    pCol[i].pData = pStart;
-    pStart += colLength[i];
-  }
-
-  for (int32_t j = 0; j < rows; j++) {
-    tdSRowResetBuf(&rb, rowData);
-    int32_t offset = 0;
-    for (int32_t k = 0; k < numOfCols; k++) {
-      const SSchema* pColumn = &pTableMeta->schema[k];
-
-      if (IS_VAR_DATA_TYPE(pColumn->type)) {
-        if (pCol[k].offset[j] != -1) {
-          char* data = pCol[k].pData + pCol[k].offset[j];
-          tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NORM, data, true, offset, k);
-        } else {
-          tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NULL, NULL, false, offset, k);
-        }
-      } else {
-        if (!colDataIsNull_f(pCol[k].nullbitmap, j)) {
-          char* data = pCol[k].pData + pColumn->bytes * j;
-          tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NORM, data, true, offset, k);
-        } else {
-          tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NULL, NULL, false, offset, k);
-        }
-      }
-
-      if (pColumn->colId != PRIMARYKEY_TIMESTAMP_COL_ID) {
-        offset += TYPE_BYTES[pColumn->type];
-      }
-    }
-    tdSRowEnd(&rb);
-    int32_t rowLen = TD_ROW_LEN(rowData);
-    rowData = POINTER_SHIFT(rowData, rowLen);
-    dataLen += rowLen;
-  }
-
-  taosMemoryFree(pCol);
-
-  blk->uid = htobe64(uid);
-  blk->suid = htobe64(suid);
-  blk->sversion = htonl(pTableMeta->sversion);
-  blk->schemaLen = htonl(schemaLen);
-  blk->numOfRows = htonl(rows);
-  blk->dataLen = htonl(dataLen);
-  subReq->length = sizeof(SSubmitReq) + sizeof(SSubmitBlk) + schemaLen + dataLen;
-  subReq->numOfBlocks = 1;
-
-  pQuery = (SQuery*)nodesMakeNode(QUERY_NODE_QUERY);
-  if (NULL == pQuery) {
-    uError("create SQuery error");
-    code = TSDB_CODE_OUT_OF_MEMORY;
-    taosMemoryFree(subReq);
-    goto end;
-  }
-  pQuery->execMode = QUERY_EXEC_MODE_SCHEDULE;
-  pQuery->haveResultSet = false;
-  pQuery->msgType = TDMT_VND_SUBMIT;
-  pQuery->pRoot = (SNode*)nodesMakeNode(QUERY_NODE_VNODE_MODIFY_STMT);
-  if (NULL == pQuery->pRoot) {
-    uError("create pQuery->pRoot error");
+  pQuery = smlInitHandle();
+  if (pQuery == NULL) {
    code = TSDB_CODE_OUT_OF_MEMORY;
    goto end;
  }
-  SVnodeModifyOpStmt* nodeStmt = (SVnodeModifyOpStmt*)(pQuery->pRoot);
-  nodeStmt->pDataBlocks = taosArrayInit(1, POINTER_BYTES);
+  pVgHash = taosHashInit(16, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_NO_LOCK);
+  taosHashPut(pVgHash, (const char*)&vgData.vgId, sizeof(vgData.vgId), (char*)&vgData, sizeof(vgData));

-  SVgDataBlocks* dst = taosMemoryCalloc(1, sizeof(SVgDataBlocks));
-  if (NULL == dst) {
-    code = TSDB_CODE_OUT_OF_MEMORY;
+  code = rawBlockBindData(pQuery, pTableMeta, pData, NULL, NULL, 0);
+  if (code != TSDB_CODE_SUCCESS) {
+    uError("WriteRaw:rawBlockBindData failed");
    goto end;
  }
-  dst->vg = vgData;
-  dst->numOfTables = subReq->numOfBlocks;
-  dst->size = subReq->length;
-  dst->pData = (char*)subReq;
-  subReq->header.vgId = htonl(dst->vg.vgId);
-  subReq->version = htonl(1);
-  subReq->header.contLen = htonl(subReq->length);
-  subReq->length = htonl(subReq->length);
-  subReq->numOfBlocks = htonl(subReq->numOfBlocks);
-  subReq = NULL;  // no need free
-  taosArrayPush(nodeStmt->pDataBlocks, &dst);
+
+  code = smlBuildOutput(pQuery, pVgHash);
+  if (code != TSDB_CODE_SUCCESS) {
+    uError("smlBuildOutput failed");
+    return code;
+  }

  launchQueryImpl(pRequest, pQuery, true, NULL);
  code = pRequest->code;
@ -1642,7 +1396,8 @@ int taos_write_raw_block(TAOS* taos, int rows, char* pData, const char* tbname)
  end:
  taosMemoryFreeClear(pTableMeta);
  qDestroyQuery(pQuery);
-  taosMemoryFree(subReq);
+  destroyRequest(pRequest);
+  taosHashCleanup(pVgHash);
  return code;
 }

@ -1679,8 +1434,6 @@ static int32_t tmqWriteRawDataImpl(TAOS* taos, void* data, int32_t dataLen) {
    goto end;
  }

-  pVgHash = taosHashInit(16, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_NO_LOCK);
-  taosHashSetFreeFp(pVgHash, destroyVgHash);
  struct SCatalog* pCatalog = NULL;
  code = catalogGetHandle(pRequest->pTscObj->pAppInfo->clusterId, &pCatalog);
  if (code != TSDB_CODE_SUCCESS) {
@ -1694,6 +1447,12 @@ static int32_t tmqWriteRawDataImpl(TAOS* taos, void* data, int32_t dataLen) {
  conn.requestObjRefId = pRequest->self;
  conn.mgmtEps = getEpSet_s(&pRequest->pTscObj->pAppInfo->mgmtEp);

+  pQuery = smlInitHandle();
+  if (pQuery == NULL) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+    goto end;
+  }
+  pVgHash = taosHashInit(16, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_NO_LOCK);
  uDebug("raw data block num:%d\n", rspObj.rsp.blockNum);
  while (++rspObj.resIter < rspObj.rsp.blockNum) {
    SRetrieveTableRsp* pRetrieve = (SRetrieveTableRsp*)taosArrayGetP(rspObj.rsp.blockData, rspObj.resIter);
@ -1701,14 +1460,6 @@ static int32_t tmqWriteRawDataImpl(TAOS* taos, void* data, int32_t dataLen) {
      uError("WriteRaw:no schema, iter:%d", rspObj.resIter);
      goto end;
    }
-    SSchemaWrapper* pSW = (SSchemaWrapper*)taosArrayGetP(rspObj.rsp.blockSchema, rspObj.resIter);
-    setResSchemaInfo(&rspObj.resInfo, pSW->pSchema, pSW->nCols);
-
-    code = setQueryResultFromRsp(&rspObj.resInfo, pRetrieve, false, false);
-    if (code != TSDB_CODE_SUCCESS) {
-      uError("WriteRaw: setQueryResultFromRsp error");
-      goto end;
-    }

    const char* tbName = (const char*)taosArrayGetP(rspObj.rsp.blockTbName, rspObj.resIter);
    if (!tbName) {
@ -1722,13 +1473,6 @@ static int32_t tmqWriteRawDataImpl(TAOS* taos, void* data, int32_t dataLen) {
    strcpy(pName.dbname, pRequest->pDb);
    strcpy(pName.tname, tbName);

-    VgData vgData = {0};
-    code = catalogGetTableHashVgroup(pCatalog, &conn, &pName, &(vgData.vg));
-    if (code != TSDB_CODE_SUCCESS) {
-      uError("WriteRaw:catalogGetTableHashVgroup failed. table name: %s", tbName);
-      goto end;
-    }
-
    code = catalogGetTableMeta(pCatalog, &conn, &pName, &pTableMeta);
    if (code == TSDB_CODE_PAR_TABLE_NOT_EXIST) {
      uError("WriteRaw:catalogGetTableMeta table not exist. table name: %s", tbName);
@ -1740,164 +1484,29 @@ static int32_t tmqWriteRawDataImpl(TAOS* taos, void* data, int32_t dataLen) {
      goto end;
    }

-    uint16_t fLen = 0;
-    int32_t  rowSize = 0;
-    int16_t  nVar = 0;
-    for (int i = 0; i < pTableMeta->tableInfo.numOfColumns; i++) {
-      SSchema* schema = &pTableMeta->schema[i];
-      fLen += TYPE_BYTES[schema->type];
-      rowSize += schema->bytes;
-      if (IS_VAR_DATA_TYPE(schema->type)) {
-        nVar++;
-      }
-    }
-    fLen -= sizeof(TSKEY);
-
-    int32_t rows = rspObj.resInfo.numOfRows;
-    int32_t extendedRowSize = rowSize + TD_ROW_HEAD_LEN - sizeof(TSKEY) + nVar * sizeof(VarDataOffsetT) +
-                              (int32_t)TD_BITMAP_BYTES(pTableMeta->tableInfo.numOfColumns - 1);
-    int32_t schemaLen = 0;
-    int32_t submitLen = sizeof(SSubmitBlk) + schemaLen + rows * extendedRowSize;
-
-    SSubmitReq* subReq = NULL;
-    SSubmitBlk* blk = NULL;
-    void*       hData = taosHashGet(pVgHash, &vgData.vg.vgId, sizeof(vgData.vg.vgId));
-    if (hData) {
-      vgData = *(VgData*)hData;
-
-      int32_t totalLen = ((SSubmitReq*)(vgData.data))->length + submitLen;
-      void*   tmp = taosMemoryRealloc(vgData.data, totalLen);
-      if (tmp == NULL) {
-        code = TSDB_CODE_OUT_OF_MEMORY;
-        goto end;
-      }
-      vgData.data = tmp;
-      ((VgData*)hData)->data = tmp;
-      subReq = (SSubmitReq*)(vgData.data);
-      blk = POINTER_SHIFT(vgData.data, subReq->length);
-    } else {
-      int32_t totalLen = sizeof(SSubmitReq) + submitLen;
-      void*   tmp = taosMemoryCalloc(1, totalLen);
-      if (tmp == NULL) {
-        code = TSDB_CODE_OUT_OF_MEMORY;
-        goto end;
-      }
-      vgData.data = tmp;
-      taosHashPut(pVgHash, (const char*)&vgData.vg.vgId, sizeof(vgData.vg.vgId), (char*)&vgData, sizeof(vgData));
-      subReq = (SSubmitReq*)(vgData.data);
-      subReq->length = sizeof(SSubmitReq);
-      subReq->numOfBlocks = 0;
-
-      blk = POINTER_SHIFT(vgData.data, sizeof(SSubmitReq));
-    }
-
-    // pSW->pSchema should be same as pTableMeta->schema
-    //    ASSERT(pSW->nCols == pTableMeta->tableInfo.numOfColumns);
-    uint64_t suid = (TSDB_NORMAL_TABLE == pTableMeta->tableType ? 0 : pTableMeta->suid);
-    uint64_t uid = pTableMeta->uid;
-    int16_t  sver = pTableMeta->sversion;
-
-    void*   blkSchema = POINTER_SHIFT(blk, sizeof(SSubmitBlk));
-    STSRow* rowData = POINTER_SHIFT(blkSchema, schemaLen);
-
-    SRowBuilder rb = {0};
-    tdSRowInit(&rb, sver);
-    tdSRowSetTpInfo(&rb, pTableMeta->tableInfo.numOfColumns, fLen);
-    int32_t totalLen = 0;
-
-    SHashObj* schemaHash = taosHashInit(16, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), false, HASH_NO_LOCK);
-    for (int i = 0; i < pSW->nCols; i++) {
-      SSchema* schema = &pSW->pSchema[i];
-      taosHashPut(schemaHash, schema->name, strlen(schema->name), &i, sizeof(int32_t));
-    }
-
-    for (int32_t j = 0; j < rows; j++) {
-      tdSRowResetBuf(&rb, rowData);
-
-      doSetOneRowPtr(&rspObj.resInfo);
-      rspObj.resInfo.current += 1;
-
-      int32_t offset = 0;
-      for (int32_t k = 0; k < pTableMeta->tableInfo.numOfColumns; k++) {
-        const SSchema* pColumn = &pTableMeta->schema[k];
-        int32_t*       index = taosHashGet(schemaHash, pColumn->name, strlen(pColumn->name));
-        if (!index) {
-          tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NONE, NULL, false, offset, k);
-        } else {
-          char* colData = rspObj.resInfo.row[*index];
-          if (!colData) {
-            tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NULL, NULL, false, offset, k);
-          } else {
-            if (IS_VAR_DATA_TYPE(pColumn->type)) {
-              colData -= VARSTR_HEADER_SIZE;
-            }
-            tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NORM, colData, true, offset, k);
-          }
-        }
-        if (pColumn->colId != PRIMARYKEY_TIMESTAMP_COL_ID) {
-          offset += TYPE_BYTES[pColumn->type];
-        }
-      }
-      tdSRowEnd(&rb);
-      int32_t rowLen = TD_ROW_LEN(rowData);
-      rowData = POINTER_SHIFT(rowData, rowLen);
-      totalLen += rowLen;
-    }
-
-    taosHashCleanup(schemaHash);
-    blk->uid = htobe64(uid);
-    blk->suid = htobe64(suid);
-    blk->sversion = htonl(sver);
-    blk->schemaLen = htonl(schemaLen);
-    blk->numOfRows = htonl(rows);
-    blk->dataLen = htonl(totalLen);
-    subReq->length += sizeof(SSubmitBlk) + schemaLen + totalLen;
-    subReq->numOfBlocks++;
-    taosMemoryFreeClear(pTableMeta);
-    rspObj.resInfo.pRspMsg = NULL;
-    doFreeReqResultInfo(&rspObj.resInfo);
-  }
-
-  pQuery = (SQuery*)nodesMakeNode(QUERY_NODE_QUERY);
-  if (NULL == pQuery) {
-    uError("create SQuery error");
-    code = TSDB_CODE_OUT_OF_MEMORY;
-    goto end;
-  }
-  pQuery->execMode = QUERY_EXEC_MODE_SCHEDULE;
-  pQuery->haveResultSet = false;
-  pQuery->msgType = TDMT_VND_SUBMIT;
-  pQuery->pRoot = (SNode*)nodesMakeNode(QUERY_NODE_VNODE_MODIFY_STMT);
-  if (NULL == pQuery->pRoot) {
-    uError("create pQuery->pRoot error");
-    code = TSDB_CODE_OUT_OF_MEMORY;
-    goto end;
-  }
-  SVnodeModifyOpStmt* nodeStmt = (SVnodeModifyOpStmt*)(pQuery->pRoot);
-
-  int32_t numOfVg = taosHashGetSize(pVgHash);
-  nodeStmt->pDataBlocks = taosArrayInit(numOfVg, POINTER_BYTES);
-
-  VgData* vData = (VgData*)taosHashIterate(pVgHash, NULL);
-  while (vData) {
-    SVgDataBlocks* dst = taosMemoryCalloc(1, sizeof(SVgDataBlocks));
-    if (NULL == dst) {
-      code = TSDB_CODE_OUT_OF_MEMORY;
+    SVgroupInfo vg;
+    code = catalogGetTableHashVgroup(pCatalog, &conn, &pName, &vg);
+    if (code != TSDB_CODE_SUCCESS) {
+      uError("WriteRaw:catalogGetTableHashVgroup failed. table name: %s", tbName);
      goto end;
    }
-    dst->vg = vData->vg;
-    SSubmitReq* subReq = (SSubmitReq*)(vData->data);
-    dst->numOfTables = subReq->numOfBlocks;
-    dst->size = subReq->length;
-    dst->pData = (char*)subReq;
-    vData->data = NULL;  // no need free
-    subReq->header.vgId = htonl(dst->vg.vgId);
-    subReq->version = htonl(1);
-    subReq->header.contLen = htonl(subReq->length);
-    subReq->length = htonl(subReq->length);
-    subReq->numOfBlocks = htonl(subReq->numOfBlocks);
-    taosArrayPush(nodeStmt->pDataBlocks, &dst);
-    vData = (VgData*)taosHashIterate(pVgHash, vData);
+
+    void* hData = taosHashGet(pVgHash, &vg.vgId, sizeof(vg.vgId));
+    if (hData == NULL) {
+      taosHashPut(pVgHash, (const char*)&vg.vgId, sizeof(vg.vgId), (char*)&vg, sizeof(vg));
+    }
+
+    code = rawBlockBindData(pQuery, pTableMeta, pRetrieve->data, NULL, NULL, 0);
+    if (code != TSDB_CODE_SUCCESS) {
+      uError("WriteRaw:rawBlockBindData failed");
+      goto end;
+    }
+  }
+
+  code = smlBuildOutput(pQuery, pVgHash);
+  if (code != TSDB_CODE_SUCCESS) {
+    uError("smlBuildOutput failed");
+    return code;
  }

  launchQueryImpl(pRequest, pQuery, true, NULL);
@ -1905,8 +1514,6 @@ static int32_t tmqWriteRawDataImpl(TAOS* taos, void* data, int32_t dataLen) {

  end:
  tDeleteSMqDataRsp(&rspObj.rsp);
-  rspObj.resInfo.pRspMsg = NULL;
-  doFreeReqResultInfo(&rspObj.resInfo);
  tDecoderClear(&decoder);
  qDestroyQuery(pQuery);
  destroyRequest(pRequest);
@ -1948,8 +1555,6 @@ static int32_t tmqWriteRawMetaDataImpl(TAOS* taos, void* data, int32_t dataLen)
    goto end;
  }

-  pVgHash = taosHashInit(16, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_NO_LOCK);
-  taosHashSetFreeFp(pVgHash, destroyVgHash);
  struct SCatalog* pCatalog = NULL;
  code = catalogGetHandle(pRequest->pTscObj->pAppInfo->clusterId, &pCatalog);
  if (code != TSDB_CODE_SUCCESS) {
@ -1963,6 +1568,13 @@ static int32_t tmqWriteRawMetaDataImpl(TAOS* taos, void* data, int32_t dataLen)
  conn.requestObjRefId = pRequest->self;
  conn.mgmtEps = getEpSet_s(&pRequest->pTscObj->pAppInfo->mgmtEp);

+  pQuery = smlInitHandle();
+  if (pQuery == NULL) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+    goto end;
+  }
+  pVgHash = taosHashInit(16, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_NO_LOCK);
+
  uDebug("raw data block num:%d\n", rspObj.rsp.blockNum);
  while (++rspObj.resIter < rspObj.rsp.blockNum) {
    SRetrieveTableRsp* pRetrieve = (SRetrieveTableRsp*)taosArrayGetP(rspObj.rsp.blockData, rspObj.resIter);
@ -1970,14 +1582,6 @@ static int32_t tmqWriteRawMetaDataImpl(TAOS* taos, void* data, int32_t dataLen)
      uError("WriteRaw:no schema, iter:%d", rspObj.resIter);
      goto end;
    }
-    SSchemaWrapper* pSW = (SSchemaWrapper*)taosArrayGetP(rspObj.rsp.blockSchema, rspObj.resIter);
-    setResSchemaInfo(&rspObj.resInfo, pSW->pSchema, pSW->nCols);
-
-    code = setQueryResultFromRsp(&rspObj.resInfo, pRetrieve, false, false);
-    if (code != TSDB_CODE_SUCCESS) {
-      uError("WriteRaw: setQueryResultFromRsp error");
-      goto end;
-    }

    const char* tbName = (const char*)taosArrayGetP(rspObj.rsp.blockTbName, rspObj.resIter);
    if (!tbName) {
@ -1991,44 +1595,28 @@ static int32_t tmqWriteRawMetaDataImpl(TAOS* taos, void* data, int32_t dataLen)
    strcpy(pName.dbname, pRequest->pDb);
    strcpy(pName.tname, tbName);

-    VgData vgData = {0};
-    code = catalogGetTableHashVgroup(pCatalog, &conn, &pName, &(vgData.vg));
-    if (code != TSDB_CODE_SUCCESS) {
-      uError("WriteRaw:catalogGetTableHashVgroup failed. table name: %s", tbName);
-      goto end;
-    }
-
    // find schema data info
-    int32_t schemaLen = 0;
-    void*   schemaData = NULL;
+    SVCreateTbReq pCreateReq = {0};
+
    for (int j = 0; j < rspObj.rsp.createTableNum; j++) {
      void**   dataTmp = taosArrayGet(rspObj.rsp.createTableReq, j);
      int32_t* lenTmp = taosArrayGet(rspObj.rsp.createTableLen, j);

-      SDecoder      decoderTmp = {0};
-      SVCreateTbReq pCreateReq = {0};
-
+      SDecoder decoderTmp = {0};
      tDecoderInit(&decoderTmp, *dataTmp, *lenTmp);
+      memset(&pCreateReq, 0, sizeof(SVCreateTbReq));
      if (tDecodeSVCreateTbReq(&decoderTmp, &pCreateReq) < 0) {
        tDecoderClear(&decoderTmp);
-        taosMemoryFreeClear(pCreateReq.comment);
-        taosArrayDestroy(pCreateReq.ctb.tagName);
        goto end;
      }

      ASSERT(pCreateReq.type == TSDB_CHILD_TABLE);
      if (strcmp(tbName, pCreateReq.name) == 0) {
-        schemaLen = *lenTmp;
-        schemaData = *dataTmp;
        strcpy(pName.tname, pCreateReq.ctb.stbName);
        tDecoderClear(&decoderTmp);
-        taosMemoryFreeClear(pCreateReq.comment);
-        taosArrayDestroy(pCreateReq.ctb.tagName);
        break;
      }
      tDecoderClear(&decoderTmp);
-      taosMemoryFreeClear(pCreateReq.comment);
-      taosArrayDestroy(pCreateReq.ctb.tagName);
    }

    code = catalogGetTableMeta(pCatalog, &conn, &pName, &pTableMeta);
@ -2042,167 +1630,23 @@ static int32_t tmqWriteRawMetaDataImpl(TAOS* taos, void* data, int32_t dataLen)
      goto end;
    }

-    uint16_t fLen = 0;
-    int32_t  rowSize = 0;
-    int16_t  nVar = 0;
-    for (int i = 0; i < pTableMeta->tableInfo.numOfColumns; i++) {
-      SSchema* schema = &pTableMeta->schema[i];
-      fLen += TYPE_BYTES[schema->type];
-      rowSize += schema->bytes;
-      if (IS_VAR_DATA_TYPE(schema->type)) {
-        nVar++;
-      }
-    }
-    fLen -= sizeof(TSKEY);
-
-    int32_t rows = rspObj.resInfo.numOfRows;
-    int32_t extendedRowSize = rowSize + TD_ROW_HEAD_LEN - sizeof(TSKEY) + nVar * sizeof(VarDataOffsetT) +
-                              (int32_t)TD_BITMAP_BYTES(pTableMeta->tableInfo.numOfColumns - 1);
-
-    int32_t submitLen = sizeof(SSubmitBlk) + schemaLen + rows * extendedRowSize;
-
-    SSubmitReq* subReq = NULL;
-    SSubmitBlk* blk = NULL;
-    void*       hData = taosHashGet(pVgHash, &vgData.vg.vgId, sizeof(vgData.vg.vgId));
-    if (hData) {
-      vgData = *(VgData*)hData;
-
-      int32_t totalLen = ((SSubmitReq*)(vgData.data))->length + submitLen;
-      void*   tmp = taosMemoryRealloc(vgData.data, totalLen);
-      if (tmp == NULL) {
-        code = TSDB_CODE_OUT_OF_MEMORY;
-        goto end;
-      }
-      vgData.data = tmp;
-      ((VgData*)hData)->data = tmp;
-      subReq = (SSubmitReq*)(vgData.data);
-      blk = POINTER_SHIFT(vgData.data, subReq->length);
-    } else {
-      int32_t totalLen = sizeof(SSubmitReq) + submitLen;
-      void*   tmp = taosMemoryCalloc(1, totalLen);
-      if (tmp == NULL) {
-        code = TSDB_CODE_OUT_OF_MEMORY;
-        goto end;
-      }
-      vgData.data = tmp;
-      taosHashPut(pVgHash, (const char*)&vgData.vg.vgId, sizeof(vgData.vg.vgId), (char*)&vgData, sizeof(vgData));
-      subReq = (SSubmitReq*)(vgData.data);
-      subReq->length = sizeof(SSubmitReq);
-      subReq->numOfBlocks = 0;
-
-      blk = POINTER_SHIFT(vgData.data, sizeof(SSubmitReq));
-    }
-
-    // pSW->pSchema should be same as pTableMeta->schema
-    //    ASSERT(pSW->nCols == pTableMeta->tableInfo.numOfColumns);
-    uint64_t suid = (TSDB_NORMAL_TABLE == pTableMeta->tableType ? 0 : pTableMeta->suid);
-    uint64_t uid = pTableMeta->uid;
-    int16_t  sver = pTableMeta->sversion;
-
-    void* blkSchema = POINTER_SHIFT(blk, sizeof(SSubmitBlk));
-    if (schemaData) {
-      memcpy(blkSchema, schemaData, schemaLen);
-    }
-    STSRow* rowData = POINTER_SHIFT(blkSchema, schemaLen);
-
-    SRowBuilder rb = {0};
-    tdSRowInit(&rb, sver);
-    tdSRowSetTpInfo(&rb, pTableMeta->tableInfo.numOfColumns, fLen);
-    int32_t totalLen = 0;
-
-    SHashObj* schemaHash = taosHashInit(16, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), false, HASH_NO_LOCK);
-    for (int i = 0; i < pSW->nCols; i++) {
-      SSchema* schema = &pSW->pSchema[i];
-      taosHashPut(schemaHash, schema->name, strlen(schema->name), &i, sizeof(int32_t));
-    }
-
-    for (int32_t j = 0; j < rows; j++) {
-      tdSRowResetBuf(&rb, rowData);
-
-      doSetOneRowPtr(&rspObj.resInfo);
-      rspObj.resInfo.current += 1;
-
-      int32_t offset = 0;
-      for (int32_t k = 0; k < pTableMeta->tableInfo.numOfColumns; k++) {
-        const SSchema* pColumn = &pTableMeta->schema[k];
-        int32_t*       index = taosHashGet(schemaHash, pColumn->name, strlen(pColumn->name));
-        if (!index) {
-          tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NONE, NULL, false, offset, k);
-        } else {
-          char* colData = rspObj.resInfo.row[*index];
-          if (!colData) {
-            tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NULL, NULL, false, offset, k);
-          } else {
-            if (IS_VAR_DATA_TYPE(pColumn->type)) {
-              colData -= VARSTR_HEADER_SIZE;
-            }
-            tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NORM, colData, true, offset, k);
-          }
-        }
-        if (pColumn->colId != PRIMARYKEY_TIMESTAMP_COL_ID) {
-          offset += TYPE_BYTES[pColumn->type];
-        }
-      }
-      tdSRowEnd(&rb);
-      int32_t rowLen = TD_ROW_LEN(rowData);
-      rowData = POINTER_SHIFT(rowData, rowLen);
-      totalLen += rowLen;
-    }
-
-    taosHashCleanup(schemaHash);
-    blk->uid = htobe64(uid);
-    blk->suid = htobe64(suid);
-    blk->sversion = htonl(sver);
-    blk->schemaLen = htonl(schemaLen);
-    blk->numOfRows = htonl(rows);
-    blk->dataLen = htonl(totalLen);
-    subReq->length += sizeof(SSubmitBlk) + schemaLen + totalLen;
-    subReq->numOfBlocks++;
-    taosMemoryFreeClear(pTableMeta);
-    rspObj.resInfo.pRspMsg = NULL;
-    doFreeReqResultInfo(&rspObj.resInfo);
-  }
-
-  pQuery = (SQuery*)nodesMakeNode(QUERY_NODE_QUERY);
-  if (NULL == pQuery) {
-    uError("create SQuery error");
-    code = TSDB_CODE_OUT_OF_MEMORY;
-    goto end;
-  }
-  pQuery->execMode = QUERY_EXEC_MODE_SCHEDULE;
-  pQuery->haveResultSet = false;
-  pQuery->msgType = TDMT_VND_SUBMIT;
-  pQuery->pRoot = (SNode*)nodesMakeNode(QUERY_NODE_VNODE_MODIFY_STMT);
-  if (NULL == pQuery->pRoot) {
-    uError("create pQuery->pRoot error");
-    code = TSDB_CODE_OUT_OF_MEMORY;
-    goto end;
-  }
-  SVnodeModifyOpStmt* nodeStmt = (SVnodeModifyOpStmt*)(pQuery->pRoot);
-
-  int32_t numOfVg = taosHashGetSize(pVgHash);
-  nodeStmt->pDataBlocks = taosArrayInit(numOfVg, POINTER_BYTES);
-
-  VgData* vData = (VgData*)taosHashIterate(pVgHash, NULL);
-  while (vData) {
-    SVgDataBlocks* dst = taosMemoryCalloc(1, sizeof(SVgDataBlocks));
-    if (NULL == dst) {
-      code = TSDB_CODE_OUT_OF_MEMORY;
+    SVgroupInfo vg;
+    code = catalogGetTableHashVgroup(pCatalog, &conn, &pName, &vg);
+    if (code != TSDB_CODE_SUCCESS) {
+      uError("WriteRaw:catalogGetTableHashVgroup failed. table name: %s", tbName);
+      goto end;
+    }
+
+    void* hData = taosHashGet(pVgHash, &vg.vgId, sizeof(vg.vgId));
+    if (hData == NULL) {
+      taosHashPut(pVgHash, (const char*)&vg.vgId, sizeof(vg.vgId), (char*)&vg, sizeof(vg));
+    }
+
+    code = rawBlockBindData(pQuery, pTableMeta, pRetrieve->data, &pCreateReq, NULL, 0);
+    if (code != TSDB_CODE_SUCCESS) {
+      uError("WriteRaw:rawBlockBindData failed");
      goto end;
    }
-    dst->vg = vData->vg;
-    SSubmitReq* subReq = (SSubmitReq*)(vData->data);
-    dst->numOfTables = subReq->numOfBlocks;
-    dst->size = subReq->length;
-    dst->pData = (char*)subReq;
-    vData->data = NULL;  // no need free
-    subReq->header.vgId = htonl(dst->vg.vgId);
-    subReq->version = htonl(1);
-    subReq->header.contLen = htonl(subReq->length);
-    subReq->length = htonl(subReq->length);
-    subReq->numOfBlocks = htonl(subReq->numOfBlocks);
-    taosArrayPush(nodeStmt->pDataBlocks, &dst);
-    vData = (VgData*)taosHashIterate(pVgHash, vData);
  }

  launchQueryImpl(pRequest, pQuery, true, NULL);
@ -2210,8 +1654,6 @@ static int32_t tmqWriteRawMetaDataImpl(TAOS* taos, void* data, int32_t dataLen)

 end:
  tDeleteSTaosxRsp(&rspObj.rsp);
-  rspObj.resInfo.pRspMsg = NULL;
-  doFreeReqResultInfo(&rspObj.resInfo);
  tDecoderClear(&decoder);
  qDestroyQuery(pQuery);
  destroyRequest(pRequest);
--- a/source/client/src/clientSml.c
+++ b/source/client/src/clientSml.c
--- a/source/client/src/clientSmlJson.c
+++ b/source/client/src/clientSmlJson.c
@ -0,0 +1,501 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <ctype.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include "clientSml.h"
+
+#define JUMP_JSON_SPACE(start) \
+while(*(start)){\
+  if(unlikely(*(start) > 32))\
+    break;\
+  else\
+    (start)++;\
+  }
+
+SArray *smlJsonParseTags(char *start, char *end){
+  SArray *tags = taosArrayInit(4, sizeof(SSmlKv));
+  while(start < end){
+    SSmlKv kv = {0};
+    kv.type = TSDB_DATA_TYPE_NCHAR;
+    bool isInQuote = false;
+    while(start < end){
+      if(unlikely(!isInQuote && *start == '"')){
+        start++;
+        kv.key = start;
+        isInQuote = true;
+        continue;
+      }
+      if(unlikely(isInQuote && *start == '"')){
+        kv.keyLen = start - kv.key;
+        start++;
+        break;
+      }
+      start++;
+    }
+    bool hasColon = false;
+    while(start < end){
+      if(unlikely(!hasColon && *start == ':')){
+        start++;
+        hasColon = true;
+        continue;
+      }
+      if(unlikely(hasColon && kv.value == NULL && (*start > 32 && *start != '"'))){
+        kv.value = start;
+        start++;
+        continue;
+      }
+
+      if(unlikely(hasColon && kv.value != NULL && (*start == '"' || *start == ',' || *start == '}'))){
+        kv.length = start - kv.value;
+        taosArrayPush(tags, &kv);
+        start++;
+        break;
+      }
+      start++;
+    }
+  }
+  return tags;
+}
+
+static int32_t smlParseTagsFromJSON(SSmlHandle *info, SSmlLineInfo *elements) {
+  int32_t ret = TSDB_CODE_SUCCESS;
+
+  if(is_same_child_table_telnet(elements, &info->preLine) == 0){
+    return TSDB_CODE_SUCCESS;
+  }
+
+  bool isSameMeasure = IS_SAME_SUPER_TABLE;
+
+  int     cnt = 0;
+  SArray *preLineKV = info->preLineTagKV;
+  bool    isSuperKVInit = true;
+  SArray *superKV = NULL;
+  if(info->dataFormat){
+    if(unlikely(!isSameMeasure)){
+      SSmlSTableMeta *sMeta = (SSmlSTableMeta *)nodeListGet(info->superTables, elements->measure, elements->measureLen, NULL);
+
+      if(unlikely(sMeta == NULL)){
+        sMeta = smlBuildSTableMeta(info->dataFormat);
+        STableMeta * pTableMeta = smlGetMeta(info, elements->measure, elements->measureLen);
+        sMeta->tableMeta = pTableMeta;
+        if(pTableMeta == NULL){
+          info->dataFormat = false;
+          info->reRun      = true;
+          return TSDB_CODE_SUCCESS;
+        }
+        nodeListSet(&info->superTables, elements->measure, elements->measureLen, sMeta, NULL);
+      }
+      info->currSTableMeta = sMeta->tableMeta;
+      superKV = sMeta->tags;
+
+      if(unlikely(taosArrayGetSize(superKV) == 0)){
+        isSuperKVInit = false;
+      }
+      taosArraySetSize(preLineKV, 0);
+    }
+  }else{
+    taosArraySetSize(preLineKV, 0);
+  }
+
+  SArray *tags = smlJsonParseTags(elements->tags, elements->tags + elements->tagsLen);
+  int32_t tagNum = taosArrayGetSize(tags);
+  for (int32_t i = 0; i < tagNum; ++i) {
+    SSmlKv kv = *(SSmlKv*)taosArrayGet(tags, i);
+
+    if(info->dataFormat){
+      if(unlikely(cnt + 1 > info->currSTableMeta->tableInfo.numOfTags)){
+        info->dataFormat = false;
+        info->reRun      = true;
+        taosArrayDestroy(tags);
+        return TSDB_CODE_SUCCESS;
+      }
+
+      if(isSameMeasure){
+        if(unlikely(cnt >= taosArrayGetSize(preLineKV))) {
+          info->dataFormat = false;
+          info->reRun      = true;
+          taosArrayDestroy(tags);
+          return TSDB_CODE_SUCCESS;
+        }
+        SSmlKv *preKV = (SSmlKv *)taosArrayGet(preLineKV, cnt);
+        if(unlikely(kv.length > preKV->length)){
+          preKV->length = kv.length;
+          SSmlSTableMeta *tableMeta = (SSmlSTableMeta *)nodeListGet(info->superTables, elements->measure, elements->measureLen, NULL);
+          ASSERT(tableMeta != NULL);
+
+          SSmlKv *oldKV = (SSmlKv *)taosArrayGet(tableMeta->tags, cnt);
+          oldKV->length = kv.length;
+          info->needModifySchema = true;
+        }
+        if(unlikely(!IS_SAME_KEY)){
+          info->dataFormat = false;
+          info->reRun      = true;
+          taosArrayDestroy(tags);
+          return TSDB_CODE_SUCCESS;
+        }
+      }else{
+        if(isSuperKVInit){
+          if(unlikely(cnt >= taosArrayGetSize(superKV))) {
+            info->dataFormat = false;
+            info->reRun      = true;
+            taosArrayDestroy(tags);
+            return TSDB_CODE_SUCCESS;
+          }
+          SSmlKv *preKV = (SSmlKv *)taosArrayGet(superKV, cnt);
+          if(unlikely(kv.length > preKV->length)) {
+            preKV->length = kv.length;
+          }else{
+            kv.length = preKV->length;
+          }
+          info->needModifySchema = true;
+
+          if(unlikely(!IS_SAME_KEY)){
+            info->dataFormat = false;
+            info->reRun      = true;
+            taosArrayDestroy(tags);
+            return TSDB_CODE_SUCCESS;
+          }
+        }else{
+          taosArrayPush(superKV, &kv);
+        }
+        taosArrayPush(preLineKV, &kv);
+      }
+    }else{
+      taosArrayPush(preLineKV, &kv);
+    }
+    cnt++;
+  }
+  taosArrayDestroy(tags);
+
+  SSmlTableInfo *tinfo = (SSmlTableInfo *)nodeListGet(info->childTables, elements, POINTER_BYTES, is_same_child_table_telnet);
+  if (unlikely(tinfo == NULL)) {
+    tinfo = smlBuildTableInfo(1, elements->measure, elements->measureLen);
+    if (unlikely(!tinfo)) {
+      return TSDB_CODE_OUT_OF_MEMORY;
+    }
+    tinfo->tags = taosArrayDup(preLineKV, NULL);
+
+    smlSetCTableName(tinfo);
+    if (info->dataFormat) {
+      info->currSTableMeta->uid = tinfo->uid;
+      tinfo->tableDataCtx = smlInitTableDataCtx(info->pQuery, info->currSTableMeta);
+      if (tinfo->tableDataCtx == NULL) {
+        smlBuildInvalidDataMsg(&info->msgBuf, "smlInitTableDataCtx error", NULL);
+        return TSDB_CODE_SML_INVALID_DATA;
+      }
+    }
+
+    SSmlLineInfo *key = (SSmlLineInfo *)taosMemoryMalloc(sizeof(SSmlLineInfo));
+    *key = *elements;
+    tinfo->key = key;
+    nodeListSet(&info->childTables, key, POINTER_BYTES, tinfo, is_same_child_table_telnet);
+  }
+  if (info->dataFormat) info->currTableDataCtx = tinfo->tableDataCtx;
+
+  return ret;
+}
+
+static char* smlJsonGetObj(char *payload){
+  int   leftBracketCnt = 0;
+  while(*payload) {
+    if (unlikely(*payload == '{')) {
+      leftBracketCnt++;
+      payload++;
+      continue;
+    }
+    if (unlikely(*payload == '}')) {
+      leftBracketCnt--;
+      payload++;
+      if (leftBracketCnt == 0) {
+        return payload;
+      } else if (leftBracketCnt < 0) {
+        return NULL;
+      }
+      continue;
+    }
+    payload++;
+  }
+  return NULL;
+}
+
+void smlJsonParseObjFirst(char **start, SSmlLineInfo *element, int8_t *offset){
+  int index = 0;
+  while(*(*start)){
+    if((*start)[0] != '"'){
+      (*start)++;
+      continue;
+    }
+
+    if(unlikely(index >= 4)) {
+      uError("index >= 4, %s", *start)
+      break;
+    }
+    char *sTmp = *start;
+    if((*start)[1] == 'm' && (*start)[2] == 'e' && (*start)[3] == 't'
+       && (*start)[4] == 'r' &&  (*start)[5] == 'i' && (*start)[6] == 'c' && (*start)[7] == '"'){
+
+      (*start) += 8;
+      bool isInQuote = false;
+      while(*(*start)){
+        if(unlikely(!isInQuote && *(*start) == '"')){
+          (*start)++;
+          offset[index++] = *start - sTmp;
+          element->measure = (*start);
+          isInQuote = true;
+          continue;
+        }
+        if(unlikely(isInQuote && *(*start) == '"')){
+          element->measureLen = (*start) - element->measure;
+          break;
+        }
+        (*start)++;
+      }
+    }else if((*start)[1] == 't' && (*start)[2] == 'i' && (*start)[3] == 'm'
+             && (*start)[4] == 'e' &&  (*start)[5] == 's' && (*start)[6] == 't'
+             && (*start)[7] == 'a' &&  (*start)[8] == 'm' && (*start)[9] == 'p' && (*start)[10] == '"'){
+
+      (*start) += 11;
+      bool hasColon = false;
+      while(*(*start)){
+        if(unlikely(!hasColon && *(*start) == ':')){
+          (*start)++;
+          JUMP_JSON_SPACE((*start))
+          offset[index++] = *start - sTmp;
+          element->timestamp = (*start);
+          hasColon = true;
+          continue;
+        }
+        if(unlikely(hasColon && (*(*start) == ',' || *(*start) == '}' || (*(*start)) <= 32))){
+          element->timestampLen = (*start) - element->timestamp;
+          break;
+        }
+        (*start)++;
+      }
+    }else if((*start)[1] == 'v' && (*start)[2] == 'a' && (*start)[3] == 'l'
+             && (*start)[4] == 'u' &&  (*start)[5] == 'e' && (*start)[6] == '"'){
+
+      (*start) += 7;
+
+      bool hasColon = false;
+      while(*(*start)){
+        if(unlikely(!hasColon && *(*start) == ':')){
+          (*start)++;
+          JUMP_JSON_SPACE((*start))
+          offset[index++] = *start - sTmp;
+          element->cols = (*start);
+          hasColon = true;
+          continue;
+        }
+        if(unlikely(hasColon && (*(*start) == ',' || *(*start) == '}' || (*(*start)) <= 32))){
+          element->colsLen = (*start) - element->cols;
+          break;
+        }
+        (*start)++;
+      }
+    }else if((*start)[1] == 't' && (*start)[2] == 'a' && (*start)[3] == 'g'
+             && (*start)[4] == 's' && (*start)[5] == '"'){
+      (*start) += 6;
+
+      while(*(*start)){
+        if(unlikely(*(*start) == ':')){
+          (*start)++;
+          JUMP_JSON_SPACE((*start))
+          offset[index++] = *start - sTmp;
+          element->tags = (*start);
+          char* tmp = smlJsonGetObj((*start));
+          if(tmp){
+            element->tagsLen = tmp - (*start);
+            *start = tmp;
+          }
+          break;
+        }
+        (*start)++;
+      }
+    }
+    if(*(*start) == '}'){
+      (*start)++;
+      break;
+    }
+    (*start)++;
+  }
+}
+
+void smlJsonParseObj(char **start, SSmlLineInfo *element, int8_t *offset){
+  int index = 0;
+  while(*(*start)){
+    if((*start)[0] != '"'){
+      (*start)++;
+      continue;
+    }
+
+    if(unlikely(index >= 4)) {
+      uError("index >= 4, %s", *start)
+      break;
+    }
+    if((*start)[1] == 'm'){
+      (*start) += offset[index++];
+      element->measure = *start;
+      while(*(*start)){
+        if(unlikely(*(*start) == '"')){
+          element->measureLen = (*start) - element->measure;
+          break;
+        }
+        (*start)++;
+      }
+    }else if((*start)[1] == 't' && (*start)[2] == 'i'){
+      (*start) += offset[index++];
+      element->timestamp = *start;
+      while(*(*start)){
+        if(unlikely(*(*start) == ',' || *(*start) == '}' || (*(*start)) <= 32)){
+          element->timestampLen = (*start) - element->timestamp;
+          break;
+        }
+        (*start)++;
+      }
+    }else if((*start)[1] == 'v'){
+      (*start) += offset[index++];
+      element->cols = *start;
+      while(*(*start)){
+        if(unlikely( *(*start) == ',' || *(*start) == '}' || (*(*start)) <= 32)){
+          element->colsLen = (*start) - element->cols;
+          break;
+        }
+        (*start)++;
+      }
+    }else if((*start)[1] == 't' && (*start)[2] == 'a'){
+      (*start) += offset[index++];
+      element->tags = (*start);
+      char* tmp = smlJsonGetObj((*start));
+      if(tmp){
+        element->tagsLen = tmp - (*start);
+        *start = tmp;
+      }
+      break;
+    }
+    if(*(*start) == '}'){
+      (*start)++;
+      break;
+    }
+    (*start)++;
+  }
+}
+
+static int32_t smlParseJSONString(SSmlHandle *info, char **start, SSmlLineInfo *elements) {
+  int32_t ret = TSDB_CODE_SUCCESS;
+
+  if(info->offset[0] == 0){
+    smlJsonParseObjFirst(start, elements, info->offset);
+  }else{
+    smlJsonParseObj(start, elements, info->offset);
+  }
+  if(**start == '\0') return TSDB_CODE_SUCCESS;
+
+  SSmlKv kv = {.key = VALUE, .keyLen = VALUE_LEN, .value = elements->cols, .length = (size_t)elements->colsLen};
+  if (smlParseNumber(&kv, &info->msgBuf)) {
+    kv.length = (int16_t)tDataTypes[kv.type].bytes;
+  }else{
+    return TSDB_CODE_TSC_INVALID_VALUE;
+  }
+
+  // Parse tags
+  ret = smlParseTagsFromJSON(info, elements);
+  if (unlikely(ret)) {
+    uError("OTD:0x%" PRIx64 " Unable to parse tags from JSON payload", info->id);
+    return ret;
+  }
+
+  if(unlikely(info->reRun)){
+    return TSDB_CODE_SUCCESS;
+  }
+
+  // Parse timestamp
+  // notice!!! put ts back to tag to ensure get meta->precision
+  int64_t ts = smlParseOpenTsdbTime(info, elements->timestamp, elements->timestampLen);
+  if (unlikely(ts < 0)) {
+    uError("OTD:0x%" PRIx64 " Unable to parse timestamp from JSON payload", info->id);
+    return TSDB_CODE_INVALID_TIMESTAMP;
+  }
+  SSmlKv kvTs = { .key = TS, .keyLen = TS_LEN, .type = TSDB_DATA_TYPE_TIMESTAMP, .i = ts, .length = (size_t)tDataTypes[TSDB_DATA_TYPE_TIMESTAMP].bytes};
+
+  if(info->dataFormat){
+    ret = smlBuildCol(info->currTableDataCtx, info->currSTableMeta->schema, &kvTs, 0);
+    if(ret == TSDB_CODE_SUCCESS){
+      ret = smlBuildCol(info->currTableDataCtx, info->currSTableMeta->schema, &kv, 1);
+    }
+    if(ret == TSDB_CODE_SUCCESS){
+      ret = smlBuildRow(info->currTableDataCtx);
+    }
+    if (unlikely(ret != TSDB_CODE_SUCCESS)) {
+      smlBuildInvalidDataMsg(&info->msgBuf, "smlBuildCol error", NULL);
+      return ret;
+    }
+  }else{
+    if(elements->colArray == NULL){
+      elements->colArray = taosArrayInit(16, sizeof(SSmlKv));
+    }
+    taosArrayPush(elements->colArray, &kvTs);
+    taosArrayPush(elements->colArray, &kv);
+  }
+  info->preLine = *elements;
+
+  return TSDB_CODE_SUCCESS;
+}
+
+int32_t smlParseJSON(SSmlHandle *info, char *payload) {
+  int32_t payloadNum = 1 << 15;
+  int32_t ret = TSDB_CODE_SUCCESS;
+
+  int cnt = 0;
+  char *dataPointStart = payload;
+  while (1) {
+    if(info->dataFormat) {
+      SSmlLineInfo element = {0};
+      ret = smlParseJSONString(info, &dataPointStart, &element);
+    }else{
+      if(cnt >= payloadNum){
+        payloadNum = payloadNum << 1;
+        void* tmp = taosMemoryRealloc(info->lines, payloadNum * sizeof(SSmlLineInfo));
+        if(tmp != NULL){
+          info->lines = (SSmlLineInfo*)tmp;
+        }
+      }
+      ret = smlParseJSONString(info, &dataPointStart, info->lines + cnt);
+    }
+    if (unlikely(ret != TSDB_CODE_SUCCESS)) {
+      uError("SML:0x%" PRIx64 " Invalid JSON Payload", info->id);
+      return ret;
+    }
+
+    if(*dataPointStart == '\0') break;
+
+    if(unlikely(info->reRun)){
+      cnt = 0;
+      dataPointStart = payload;
+      info->lineNum = payloadNum;
+      ret = smlClearForRerun(info);
+      if(ret != TSDB_CODE_SUCCESS){
+        return ret;
+      }
+      continue;
+    }
+    cnt++;
+  }
+  info->lineNum = cnt;
+
+  return TSDB_CODE_SUCCESS;
+}
--- a/source/client/src/clientSmlLine.c
+++ b/source/client/src/clientSmlLine.c
@ -0,0 +1,700 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <ctype.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "clientSml.h"
+
+// comma ,
+//#define IS_SLASH_COMMA(sql) (*(sql) == COMMA && *((sql)-1) == SLASH)
+#define IS_COMMA(sql)       (*(sql) == COMMA && *((sql)-1) != SLASH)
+// space
+//#define IS_SLASH_SPACE(sql) (*(sql) == SPACE && *((sql)-1) == SLASH)
+#define IS_SPACE(sql)       (*(sql) == SPACE && *((sql)-1) != SLASH)
+// equal =
+//#define IS_SLASH_EQUAL(sql) (*(sql) == EQUAL && *((sql)-1) == SLASH)
+#define IS_EQUAL(sql)       (*(sql) == EQUAL && *((sql)-1) != SLASH)
+// quote "
+//#define IS_SLASH_QUOTE(sql) (*(sql) == QUOTE && *((sql)-1) == SLASH)
+#define IS_QUOTE(sql)       (*(sql) == QUOTE && *((sql)-1) != SLASH)
+// SLASH
+//#define IS_SLASH_SLASH(sql) (*(sql) == SLASH && *((sql)-1) == SLASH)
+
+#define IS_SLASH_LETTER(sql) \
+  (*((sql)-1) == SLASH && (*(sql) == COMMA || *(sql) == SPACE || *(sql) == EQUAL || *(sql) == QUOTE || *(sql) == SLASH))                          \
+//  (IS_SLASH_COMMA(sql) || IS_SLASH_SPACE(sql) || IS_SLASH_EQUAL(sql) || IS_SLASH_QUOTE(sql) || IS_SLASH_SLASH(sql))
+
+#define MOVE_FORWARD_ONE(sql, len) (memmove((void *)((sql)-1), (sql), len))
+
+#define PROCESS_SLASH(key, keyLen)           \
+  for (int i = 1; i < keyLen; ++i) {         \
+    if (IS_SLASH_LETTER(key + i)) {          \
+      MOVE_FORWARD_ONE(key + i, keyLen - i); \
+      i--;                                   \
+      keyLen--;                              \
+    }                                        \
+  }
+
+#define BINARY_ADD_LEN 2  // "binary"   2 means " "
+#define NCHAR_ADD_LEN  3  // L"nchar"   3 means L" "
+
+uint8_t smlPrecisionConvert[7] = {TSDB_TIME_PRECISION_NANO, TSDB_TIME_PRECISION_HOURS, TSDB_TIME_PRECISION_MINUTES,
+                                  TSDB_TIME_PRECISION_SECONDS, TSDB_TIME_PRECISION_MILLI, TSDB_TIME_PRECISION_MICRO,
+                                  TSDB_TIME_PRECISION_NANO};
+
+static bool smlParseBool(SSmlKv *kvVal) {
+  const char *pVal = kvVal->value;
+  int32_t     len = kvVal->length;
+  if ((len == 1) && (pVal[0] == 't' || pVal[0] == 'T')) {
+    kvVal->i = TSDB_TRUE;
+    return true;
+  }
+
+  if ((len == 1) && (pVal[0] == 'f' || pVal[0] == 'F')) {
+    kvVal->i = TSDB_FALSE;
+    return true;
+  }
+
+  if ((len == 4) && !strncasecmp(pVal, "true", len)) {
+    kvVal->i = TSDB_TRUE;
+    return true;
+  }
+  if ((len == 5) && !strncasecmp(pVal, "false", len)) {
+    kvVal->i = TSDB_FALSE;
+    return true;
+  }
+  return false;
+}
+
+static bool smlIsBinary(const char *pVal, uint16_t len) {
+  // binary: "abc"
+  if (len < 2) {
+    return false;
+  }
+  if (pVal[0] == '"' && pVal[len - 1] == '"') {
+    return true;
+  }
+  return false;
+}
+
+static bool smlIsNchar(const char *pVal, uint16_t len) {
+  // nchar: L"abc"
+  if (len < 3) {
+    return false;
+  }
+  if (pVal[1] == '"' && pVal[len - 1] == '"' && (pVal[0] == 'l' || pVal[0] == 'L')) {
+    return true;
+  }
+  return false;
+}
+
+static int64_t smlParseInfluxTime(SSmlHandle *info, const char *data, int32_t len) {
+  uint8_t toPrecision = info->currSTableMeta ? info->currSTableMeta->tableInfo.precision : TSDB_TIME_PRECISION_NANO;
+
+  if(unlikely(len == 0 || (len == 1 && data[0] == '0'))){
+    return taosGetTimestampNs()/smlFactorNS[toPrecision];
+  }
+
+  uint8_t fromPrecision = smlPrecisionConvert[info->precision];
+
+  int64_t ts = smlGetTimeValue(data, len, fromPrecision, toPrecision);
+  if (unlikely(ts == -1)) {
+    smlBuildInvalidDataMsg(&info->msgBuf, "invalid timestamp", data);
+    return -1;
+  }
+  return ts;
+}
+
+int32_t smlParseValue(SSmlKv *pVal, SSmlMsgBuf *msg) {
+  if (pVal->value[0] == '"'){  // binary
+    if (pVal->length >= 2 && pVal->value[pVal->length - 1] == '"') {
+      pVal->type = TSDB_DATA_TYPE_BINARY;
+      pVal->length -= BINARY_ADD_LEN;
+      if (pVal->length > TSDB_MAX_BINARY_LEN - VARSTR_HEADER_SIZE) {
+        return TSDB_CODE_PAR_INVALID_VAR_COLUMN_LEN;
+      }
+      pVal->value += (BINARY_ADD_LEN - 1);
+      return TSDB_CODE_SUCCESS;
+    }
+    return TSDB_CODE_TSC_INVALID_VALUE;
+  }
+
+  if(pVal->value[0] == 'l' || pVal->value[0] == 'L'){  // nchar
+    if (pVal->value[1] == '"' && pVal->value[pVal->length - 1] == '"' && pVal->length >= 3){
+      pVal->type = TSDB_DATA_TYPE_NCHAR;
+      pVal->length -= NCHAR_ADD_LEN;
+      if (pVal->length > (TSDB_MAX_NCHAR_LEN - VARSTR_HEADER_SIZE) / TSDB_NCHAR_SIZE) {
+        return TSDB_CODE_PAR_INVALID_VAR_COLUMN_LEN;
+      }
+      pVal->value += (NCHAR_ADD_LEN - 1);
+      return TSDB_CODE_SUCCESS;
+    }
+    return TSDB_CODE_TSC_INVALID_VALUE;
+  }
+
+  if (pVal->value[0] == 't' || pVal->value[0] == 'T'){
+    if(pVal->length == 1 || (pVal->length == 4 && (pVal->value[1] == 'r' || pVal->value[1] == 'R')
+                              && (pVal->value[2] == 'u' || pVal->value[2] == 'U')
+                              && (pVal->value[3] == 'e' || pVal->value[3] == 'E'))){
+      pVal->i = TSDB_TRUE;
+      pVal->type = TSDB_DATA_TYPE_BOOL;
+      pVal->length = (int16_t)tDataTypes[pVal->type].bytes;
+      return TSDB_CODE_SUCCESS;
+    }
+    return TSDB_CODE_TSC_INVALID_VALUE;
+  }
+
+  if (pVal->value[0] == 'f' || pVal->value[0] == 'F'){
+    if(pVal->length == 1 || (pVal->length == 5 && (pVal->value[1] == 'a' || pVal->value[1] == 'A')
+                             && (pVal->value[2] == 'l' || pVal->value[2] == 'L')
+                             && (pVal->value[3] == 's' || pVal->value[3] == 'S')
+                             && (pVal->value[4] == 'e' || pVal->value[4] == 'E'))){
+      pVal->i = TSDB_FALSE;
+      pVal->type = TSDB_DATA_TYPE_BOOL;
+      pVal->length = (int16_t)tDataTypes[pVal->type].bytes;
+      return TSDB_CODE_SUCCESS;
+    }
+    return TSDB_CODE_TSC_INVALID_VALUE;
+  }
+
+  // number
+  if (smlParseNumber(pVal, msg)) {
+    pVal->length = (int16_t)tDataTypes[pVal->type].bytes;
+    return TSDB_CODE_SUCCESS;
+  }
+
+  return TSDB_CODE_TSC_INVALID_VALUE;
+}
+
+static int32_t smlParseTagKv(SSmlHandle *info, char **sql, char *sqlEnd,
+                          SSmlLineInfo* currElement, bool isSameMeasure, bool isSameCTable){
+  if(isSameCTable){
+    return TSDB_CODE_SUCCESS;
+  }
+
+  int     cnt = 0;
+  SArray *preLineKV = info->preLineTagKV;
+  bool    isSuperKVInit = true;
+  SArray *superKV = NULL;
+  if(info->dataFormat){
+    if(unlikely(!isSameMeasure)){
+      SSmlSTableMeta *sMeta = (SSmlSTableMeta *)nodeListGet(info->superTables, currElement->measure, currElement->measureLen, NULL);
+
+      if(unlikely(sMeta == NULL)){
+        sMeta = smlBuildSTableMeta(info->dataFormat);
+        STableMeta * pTableMeta = smlGetMeta(info, currElement->measure, currElement->measureLen);
+        sMeta->tableMeta = pTableMeta;
+        if(pTableMeta == NULL){
+          info->dataFormat = false;
+          info->reRun      = true;
+          return TSDB_CODE_SUCCESS;
+        }
+        nodeListSet(&info->superTables, currElement->measure, currElement->measureLen, sMeta, NULL);
+      }
+      info->currSTableMeta = sMeta->tableMeta;
+      superKV = sMeta->tags;
+
+      if(unlikely(taosArrayGetSize(superKV) == 0)){
+        isSuperKVInit = false;
+      }
+      taosArraySetSize(preLineKV, 0);
+    }
+  }else{
+    taosArraySetSize(preLineKV, 0);
+  }
+
+
+  while (*sql < sqlEnd) {
+    if (unlikely(IS_SPACE(*sql))) {
+      break;
+    }
+
+    bool hasSlash = false;
+    // parse key
+    const char *key = *sql;
+    size_t     keyLen = 0;
+    while (*sql < sqlEnd) {
+      if (unlikely(IS_COMMA(*sql))) {
+        smlBuildInvalidDataMsg(&info->msgBuf, "invalid data", *sql);
+        return TSDB_CODE_SML_INVALID_DATA;
+      }
+      if (unlikely(IS_EQUAL(*sql))) {
+        keyLen = *sql - key;
+        (*sql)++;
+        break;
+      }
+      if(!hasSlash){
+        hasSlash = (*(*sql) == SLASH);
+      }
+      (*sql)++;
+    }
+    if(unlikely(hasSlash)) {
+      PROCESS_SLASH(key, keyLen)
+    }
+
+    if (unlikely(IS_INVALID_COL_LEN(keyLen))) {
+      smlBuildInvalidDataMsg(&info->msgBuf, "invalid key or key is too long than 64", key);
+      return TSDB_CODE_TSC_INVALID_COLUMN_LENGTH;
+    }
+
+    // parse value
+    const char *value = *sql;
+    size_t     valueLen = 0;
+    hasSlash = false;
+    while (*sql < sqlEnd) {
+      // parse value
+      if (unlikely(IS_SPACE(*sql) || IS_COMMA(*sql))) {
+        break;
+      }else if (unlikely(IS_EQUAL(*sql))) {
+        smlBuildInvalidDataMsg(&info->msgBuf, "invalid data", *sql);
+        return TSDB_CODE_SML_INVALID_DATA;
+      }
+
+      if(!hasSlash){
+        hasSlash = (*(*sql) == SLASH);
+      }
+
+      (*sql)++;
+    }
+    valueLen = *sql - value;
+
+    if (unlikely(valueLen == 0)) {
+      smlBuildInvalidDataMsg(&info->msgBuf, "invalid value", value);
+      return TSDB_CODE_SML_INVALID_DATA;
+    }
+
+    if(unlikely(hasSlash)) {
+      PROCESS_SLASH(value, valueLen)
+    }
+
+    if (unlikely(valueLen > (TSDB_MAX_NCHAR_LEN - VARSTR_HEADER_SIZE) / TSDB_NCHAR_SIZE)) {
+      return TSDB_CODE_PAR_INVALID_VAR_COLUMN_LEN;
+    }
+
+    SSmlKv kv = {.key = key, .keyLen = keyLen, .type = TSDB_DATA_TYPE_NCHAR, .value = value, .length = valueLen};
+    if(info->dataFormat){
+      if(unlikely(cnt + 1 > info->currSTableMeta->tableInfo.numOfTags)){
+        info->dataFormat = false;
+        info->reRun      = true;
+        return TSDB_CODE_SUCCESS;
+      }
+
+      if(isSameMeasure){
+        if(unlikely(cnt >= taosArrayGetSize(preLineKV))) {
+          info->dataFormat = false;
+          info->reRun      = true;
+          return TSDB_CODE_SUCCESS;
+        }
+        SSmlKv *preKV = (SSmlKv *)taosArrayGet(preLineKV, cnt);
+        if(unlikely(kv.length > preKV->length)){
+          preKV->length = kv.length;
+          SSmlSTableMeta *tableMeta = (SSmlSTableMeta *)nodeListGet(info->superTables, currElement->measure, currElement->measureLen, NULL);
+          ASSERT(tableMeta != NULL);
+
+          SSmlKv *oldKV = (SSmlKv *)taosArrayGet(tableMeta->tags, cnt);
+          oldKV->length = kv.length;
+          info->needModifySchema = true;
+        }
+        if(unlikely(!IS_SAME_KEY)){
+          info->dataFormat = false;
+          info->reRun      = true;
+          return TSDB_CODE_SUCCESS;
+        }
+      }else{
+        if(isSuperKVInit){
+          if(unlikely(cnt >= taosArrayGetSize(superKV))) {
+            info->dataFormat = false;
+            info->reRun      = true;
+            return TSDB_CODE_SUCCESS;
+          }
+          SSmlKv *preKV = (SSmlKv *)taosArrayGet(superKV, cnt);
+          if(unlikely(kv.length > preKV->length)) {
+            preKV->length = kv.length;
+          }else{
+            kv.length = preKV->length;
+          }
+          info->needModifySchema = true;
+
+          if(unlikely(!IS_SAME_KEY)){
+            info->dataFormat = false;
+            info->reRun      = true;
+            return TSDB_CODE_SUCCESS;
+          }
+        }else{
+          taosArrayPush(superKV, &kv);
+        }
+        taosArrayPush(preLineKV, &kv);
+      }
+    }else{
+      taosArrayPush(preLineKV, &kv);
+    }
+
+    cnt++;
+    if(IS_SPACE(*sql)){
+      break;
+    }
+    (*sql)++;
+  }
+
+  void* oneTable = nodeListGet(info->childTables, currElement->measure, currElement->measureTagsLen, NULL);
+  if ((oneTable != NULL)) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+  SSmlTableInfo *tinfo = smlBuildTableInfo(1, currElement->measure, currElement->measureLen);
+  if (unlikely(!tinfo)) {
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+  tinfo->tags = taosArrayDup(preLineKV, NULL);
+
+  smlSetCTableName(tinfo);
+  if(info->dataFormat) {
+    info->currSTableMeta->uid = tinfo->uid;
+    tinfo->tableDataCtx = smlInitTableDataCtx(info->pQuery, info->currSTableMeta);
+    if(tinfo->tableDataCtx == NULL){
+      smlBuildInvalidDataMsg(&info->msgBuf, "smlInitTableDataCtx error", NULL);
+      return TSDB_CODE_SML_INVALID_DATA;
+    }
+  }
+
+  nodeListSet(&info->childTables, currElement->measure, currElement->measureTagsLen, tinfo, NULL);
+
+  return TSDB_CODE_SUCCESS;
+}
+
+static int32_t smlParseColKv(SSmlHandle *info, char **sql, char *sqlEnd,
+                          SSmlLineInfo* currElement, bool isSameMeasure, bool isSameCTable){
+  int     cnt = 0;
+  SArray *preLineKV = info->preLineColKV;
+  bool    isSuperKVInit = true;
+  SArray *superKV = NULL;
+  if(info->dataFormat){
+    if(unlikely(!isSameCTable)){
+      SSmlTableInfo *oneTable = (SSmlTableInfo *)nodeListGet(info->childTables, currElement->measure, currElement->measureTagsLen, NULL);
+      if (unlikely(oneTable == NULL)) {
+        smlBuildInvalidDataMsg(&info->msgBuf, "child table should inside", currElement->measure);
+        return TSDB_CODE_SML_INVALID_DATA;
+      }
+      info->currTableDataCtx = oneTable->tableDataCtx;
+    }
+
+    if(unlikely(!isSameMeasure)){
+      SSmlSTableMeta *sMeta = (SSmlSTableMeta *)nodeListGet(info->superTables, currElement->measure, currElement->measureLen, NULL);
+
+      if(unlikely(sMeta == NULL)){
+        sMeta = smlBuildSTableMeta(info->dataFormat);
+        STableMeta * pTableMeta = smlGetMeta(info, currElement->measure, currElement->measureLen);
+        sMeta->tableMeta = pTableMeta;
+        if(pTableMeta == NULL){
+          info->dataFormat = false;
+          info->reRun      = true;
+          return TSDB_CODE_SUCCESS;
+        }
+        nodeListSet(&info->superTables, currElement->measure, currElement->measureLen, sMeta, NULL);
+      }
+      info->currSTableMeta = sMeta->tableMeta;
+      superKV = sMeta->cols;
+      if(unlikely(taosArrayGetSize(superKV) == 0)){
+        isSuperKVInit = false;
+      }
+      taosArraySetSize(preLineKV, 0);
+    }
+  }
+
+  while (*sql < sqlEnd) {
+    if (unlikely(IS_SPACE(*sql))) {
+      break;
+    }
+
+    bool hasSlash = false;
+    // parse key
+    const char *key = *sql;
+    size_t     keyLen = 0;
+    while (*sql < sqlEnd) {
+      if (unlikely(IS_COMMA(*sql))) {
+        smlBuildInvalidDataMsg(&info->msgBuf, "invalid data", *sql);
+        return TSDB_CODE_SML_INVALID_DATA;
+      }
+      if (unlikely(IS_EQUAL(*sql))) {
+        keyLen = *sql - key;
+        (*sql)++;
+        break;
+      }
+      if(!hasSlash){
+        hasSlash = (*(*sql) == SLASH);
+      }
+      (*sql)++;
+    }
+    if(unlikely(hasSlash)) {
+      PROCESS_SLASH(key, keyLen)
+    }
+
+    if (unlikely(IS_INVALID_COL_LEN(keyLen))) {
+      smlBuildInvalidDataMsg(&info->msgBuf, "invalid key or key is too long than 64", key);
+      return TSDB_CODE_TSC_INVALID_COLUMN_LENGTH;
+    }
+
+    // parse value
+    const char *value = *sql;
+    size_t     valueLen = 0;
+    hasSlash              = false;
+    bool        isInQuote = false;
+    while (*sql < sqlEnd) {
+      // parse value
+      if (unlikely(IS_QUOTE(*sql))) {
+        isInQuote = !isInQuote;
+        (*sql)++;
+        continue;
+      }
+      if (!isInQuote){
+        if (unlikely(IS_SPACE(*sql) || IS_COMMA(*sql))) {
+          break;
+        } else if (unlikely(IS_EQUAL(*sql))) {
+          smlBuildInvalidDataMsg(&info->msgBuf, "invalid data", *sql);
+          return TSDB_CODE_SML_INVALID_DATA;
+        }
+      }
+      if(!hasSlash){
+        hasSlash = (*(*sql) == SLASH);
+      }
+
+      (*sql)++;
+    }
+    valueLen = *sql - value;
+
+    if (unlikely(isInQuote)) {
+      smlBuildInvalidDataMsg(&info->msgBuf, "only one quote", value);
+      return TSDB_CODE_SML_INVALID_DATA;
+    }
+    if (unlikely(valueLen == 0)) {
+      smlBuildInvalidDataMsg(&info->msgBuf, "invalid value", value);
+      return TSDB_CODE_SML_INVALID_DATA;
+    }
+    if(unlikely(hasSlash)) {
+      PROCESS_SLASH(value, valueLen)
+    }
+
+    SSmlKv kv = {.key = key, .keyLen = keyLen, .value = value, .length = valueLen};
+    int32_t ret = smlParseValue(&kv, &info->msgBuf);
+    if (ret != TSDB_CODE_SUCCESS) {
+      smlBuildInvalidDataMsg(&info->msgBuf, "smlParseValue error", value);
+      return ret;
+    }
+
+    if(info->dataFormat){
+      //cnt begin 0, add ts so + 2
+      if(unlikely(cnt + 2 > info->currSTableMeta->tableInfo.numOfColumns)){
+        info->dataFormat = false;
+        info->reRun      = true;
+        return TSDB_CODE_SUCCESS;
+      }
+      // bind data
+      ret = smlBuildCol(info->currTableDataCtx, info->currSTableMeta->schema, &kv, cnt + 1);
+      if (unlikely(ret != TSDB_CODE_SUCCESS)) {
+        uError("smlBuildCol error, retry");
+        info->dataFormat = false;
+        info->reRun      = true;
+        return TSDB_CODE_SUCCESS;
+      }
+
+      if(isSameMeasure){
+        if(cnt >= taosArrayGetSize(preLineKV)) {
+          info->dataFormat = false;
+          info->reRun      = true;
+          return TSDB_CODE_SUCCESS;
+        }
+        SSmlKv *preKV = (SSmlKv *)taosArrayGet(preLineKV, cnt);
+        if(kv.type != preKV->type){
+          info->dataFormat = false;
+          info->reRun      = true;
+          return TSDB_CODE_SUCCESS;
+        }
+
+        if(unlikely(IS_VAR_DATA_TYPE(kv.type) && kv.length > preKV->length)){
+          preKV->length = kv.length;
+          SSmlSTableMeta *tableMeta = (SSmlSTableMeta *)nodeListGet(info->superTables, currElement->measure, currElement->measureLen, NULL);
+          ASSERT(tableMeta != NULL);
+
+          SSmlKv *oldKV = (SSmlKv *)taosArrayGet(tableMeta->cols, cnt);
+          oldKV->length = kv.length;
+          info->needModifySchema = true;
+        }
+        if(unlikely(!IS_SAME_KEY)){
+          info->dataFormat = false;
+          info->reRun      = true;
+          return TSDB_CODE_SUCCESS;
+        }
+      }else{
+        if(isSuperKVInit){
+          if(unlikely(cnt >= taosArrayGetSize(superKV))) {
+            info->dataFormat = false;
+            info->reRun      = true;
+            return TSDB_CODE_SUCCESS;
+          }
+          SSmlKv *preKV = (SSmlKv *)taosArrayGet(superKV, cnt);
+          if(unlikely(kv.type != preKV->type)){
+            info->dataFormat = false;
+            info->reRun      = true;
+            return TSDB_CODE_SUCCESS;
+          }
+
+          if(IS_VAR_DATA_TYPE(kv.type)){
+            if(kv.length > preKV->length) {
+              preKV->length = kv.length;
+            }else{
+              kv.length = preKV->length;
+            }
+            info->needModifySchema = true;
+          }
+          if(unlikely(!IS_SAME_KEY)){
+            info->dataFormat = false;
+            info->reRun      = true;
+            return TSDB_CODE_SUCCESS;
+          }
+        }else{
+          taosArrayPush(superKV, &kv);
+        }
+        taosArrayPush(preLineKV, &kv);
+      }
+    }else{
+      if(currElement->colArray == NULL){
+        currElement->colArray = taosArrayInit(16, sizeof(SSmlKv));
+        taosArraySetSize(currElement->colArray, 1);
+      }
+      taosArrayPush(currElement->colArray, &kv);   //reserve for timestamp
+    }
+
+    cnt++;
+    if(IS_SPACE(*sql)){
+      break;
+    }
+    (*sql)++;
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
+int32_t smlParseInfluxString(SSmlHandle *info, char *sql, char *sqlEnd, SSmlLineInfo *elements) {
+  if (!sql) return TSDB_CODE_SML_INVALID_DATA;
+  JUMP_SPACE(sql, sqlEnd)
+  if (unlikely(*sql == COMMA)) return TSDB_CODE_SML_INVALID_DATA;
+  elements->measure = sql;
+
+  // parse measure
+  while (sql < sqlEnd) {
+    if (unlikely((sql != elements->measure) && IS_SLASH_LETTER(sql))) {
+      MOVE_FORWARD_ONE(sql, sqlEnd - sql);
+      sqlEnd--;
+      continue;
+    }
+    if (unlikely(IS_COMMA(sql))) {
+      break;
+    }
+
+    if (unlikely(IS_SPACE(sql))) {
+      break;
+    }
+    sql++;
+  }
+  elements->measureLen = sql - elements->measure;
+  if (unlikely(IS_INVALID_TABLE_LEN(elements->measureLen))) {
+    smlBuildInvalidDataMsg(&info->msgBuf, "measure is empty or too large than 192", NULL);
+    return TSDB_CODE_TSC_INVALID_TABLE_ID_LENGTH;
+  }
+
+  // to get measureTagsLen before
+  const char* tmp = sql;
+  while (tmp < sqlEnd){
+    if (unlikely(IS_SPACE(tmp))) {
+      break;
+    }
+    tmp++;
+  }
+  elements->measureTagsLen = tmp - elements->measure;
+
+  bool isSameCTable = false;
+  bool isSameMeasure = false;
+  if(IS_SAME_CHILD_TABLE){
+    isSameCTable = true;
+    isSameMeasure = true;
+  }else if(info->dataFormat) {
+    isSameMeasure = IS_SAME_SUPER_TABLE;
+  }
+  // parse tag
+  if (*sql == COMMA) sql++;
+  elements->tags = sql;
+
+  int ret = smlParseTagKv(info, &sql, sqlEnd, elements, isSameMeasure, isSameCTable);
+  if(unlikely(ret != TSDB_CODE_SUCCESS)){
+    return ret;
+  }
+  if(unlikely(info->reRun)){
+    return TSDB_CODE_SUCCESS;
+  }
+
+  sql = elements->measure + elements->measureTagsLen;
+  elements->tagsLen = sql - elements->tags;
+
+  // parse cols
+  JUMP_SPACE(sql, sqlEnd)
+  elements->cols = sql;
+
+  ret = smlParseColKv(info, &sql, sqlEnd, elements, isSameMeasure, isSameCTable);
+  if(unlikely(ret != TSDB_CODE_SUCCESS)){
+    return ret;
+  }
+
+  if(unlikely(info->reRun)){
+    return TSDB_CODE_SUCCESS;
+  }
+
+  elements->colsLen = sql - elements->cols;
+  if (unlikely(elements->colsLen == 0)) {
+    smlBuildInvalidDataMsg(&info->msgBuf, "cols is empty", NULL);
+    return TSDB_CODE_SML_INVALID_DATA;
+  }
+
+  // parse timestamp
+  JUMP_SPACE(sql, sqlEnd)
+  elements->timestamp = sql;
+  while (sql < sqlEnd) {
+    if (unlikely(isspace(*sql))) {
+      break;
+    }
+    sql++;
+  }
+  elements->timestampLen = sql - elements->timestamp;
+
+  int64_t ts = smlParseInfluxTime(info, elements->timestamp, elements->timestampLen);
+  if (unlikely(ts <= 0)) {
+    uError("SML:0x%" PRIx64 " smlParseTS error:%" PRId64, info->id, ts);
+    return TSDB_CODE_INVALID_TIMESTAMP;
+  }
+  // add ts to
+  SSmlKv kv = { .key = TS, .keyLen = TS_LEN, .type = TSDB_DATA_TYPE_TIMESTAMP, .i = ts, .length = (size_t)tDataTypes[TSDB_DATA_TYPE_TIMESTAMP].bytes};
+  if(info->dataFormat){
+    smlBuildCol(info->currTableDataCtx, info->currSTableMeta->schema, &kv, 0);
+    smlBuildRow(info->currTableDataCtx);
+  }else{
+    taosArraySet(elements->colArray, 0, &kv);
+  }
+  info->preLine = *elements;
+
+  return ret;
+}
+
--- a/source/client/src/clientSmlTelnet.c
+++ b/source/client/src/clientSmlTelnet.c
@ -0,0 +1,333 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <ctype.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "clientSml.h"
+
+int32_t is_same_child_table_telnet(const void *a, const void *b){
+  SSmlLineInfo *t1 = (SSmlLineInfo *)a;
+  SSmlLineInfo *t2 = (SSmlLineInfo *)b;
+  return (((t1->measureLen == t2->measureLen) && memcmp(t1->measure, t2->measure, t1->measureLen) == 0)
+          && ((t1->tagsLen == t2->tagsLen) && memcmp(t1->tags, t2->tags, t1->tagsLen) == 0)) ? 0 : 1;
+}
+
+int64_t smlParseOpenTsdbTime(SSmlHandle *info, const char *data, int32_t len) {
+  uint8_t toPrecision = info->currSTableMeta ? info->currSTableMeta->tableInfo.precision : TSDB_TIME_PRECISION_NANO;
+
+  if (unlikely(!data)) {
+    smlBuildInvalidDataMsg(&info->msgBuf, "timestamp can not be null", NULL);
+    return -1;
+  }
+  if (unlikely(len == 1 && data[0] == '0')) {
+    return taosGetTimestampNs()/smlFactorNS[toPrecision];
+  }
+  int8_t fromPrecision = smlGetTsTypeByLen(len);
+  if (unlikely(fromPrecision == -1)) {
+    smlBuildInvalidDataMsg(&info->msgBuf,
+                           "timestamp precision can only be seconds(10 digits) or milli seconds(13 digits)", data);
+    return -1;
+  }
+  int64_t ts = smlGetTimeValue(data, len, fromPrecision, toPrecision);
+  if (unlikely(ts == -1)) {
+    smlBuildInvalidDataMsg(&info->msgBuf, "invalid timestamp", data);
+    return -1;
+  }
+  return ts;
+}
+
+
+static void smlParseTelnetElement(char **sql, char *sqlEnd, char **data, int32_t *len) {
+  while (*sql < sqlEnd) {
+    if (unlikely((**sql != SPACE && !(*data)))) {
+      *data = *sql;
+    } else if (unlikely(**sql == SPACE && *data)) {
+      *len = *sql - *data;
+      break;
+    }
+    (*sql)++;
+  }
+}
+
+static int32_t smlParseTelnetTags(SSmlHandle *info, char *data, char *sqlEnd, SSmlLineInfo *elements, SSmlMsgBuf *msg) {
+  if(is_same_child_table_telnet(elements, &info->preLine) == 0){
+    return TSDB_CODE_SUCCESS;
+  }
+
+  bool isSameMeasure = IS_SAME_SUPER_TABLE;
+
+  int     cnt = 0;
+  SArray *preLineKV = info->preLineTagKV;
+  bool    isSuperKVInit = true;
+  SArray *superKV = NULL;
+  if(info->dataFormat){
+    if(!isSameMeasure){
+      SSmlSTableMeta *sMeta = (SSmlSTableMeta *)nodeListGet(info->superTables, elements->measure, elements->measureLen, NULL);
+
+      if(unlikely(sMeta == NULL)){
+        sMeta = smlBuildSTableMeta(info->dataFormat);
+        STableMeta * pTableMeta = smlGetMeta(info, elements->measure, elements->measureLen);
+        sMeta->tableMeta = pTableMeta;
+        if(pTableMeta == NULL){
+          info->dataFormat = false;
+          info->reRun      = true;
+          return TSDB_CODE_SUCCESS;
+        }
+        nodeListSet(&info->superTables, elements->measure, elements->measureLen, sMeta, NULL);
+      }
+      info->currSTableMeta = sMeta->tableMeta;
+      superKV = sMeta->tags;
+
+      if(unlikely(taosArrayGetSize(superKV) == 0)){
+        isSuperKVInit = false;
+      }
+      taosArraySetSize(preLineKV, 0);
+    }
+  }else{
+    taosArraySetSize(preLineKV, 0);
+  }
+
+  const char *sql = data;
+  while (sql < sqlEnd) {
+    JUMP_SPACE(sql, sqlEnd)
+    if (unlikely(*sql == '\0')) break;
+
+    const char *key = sql;
+    size_t     keyLen = 0;
+
+    // parse key
+    while (sql < sqlEnd) {
+      if (unlikely(*sql == SPACE)) {
+        smlBuildInvalidDataMsg(msg, "invalid data", sql);
+        return TSDB_CODE_SML_INVALID_DATA;
+      }
+      if (unlikely(*sql == EQUAL)) {
+        keyLen = sql - key;
+        sql++;
+        break;
+      }
+      sql++;
+    }
+
+    if (unlikely(IS_INVALID_COL_LEN(keyLen))) {
+      smlBuildInvalidDataMsg(msg, "invalid key or key is too long than 64", key);
+      return TSDB_CODE_TSC_INVALID_COLUMN_LENGTH;
+    }
+//    if (smlCheckDuplicateKey(key, keyLen, dumplicateKey)) {
+//      smlBuildInvalidDataMsg(msg, "dumplicate key", key);
+//      return TSDB_CODE_TSC_DUP_NAMES;
+//    }
+
+    // parse value
+    const char *value = sql;
+    size_t     valueLen = 0;
+    while (sql < sqlEnd) {
+      // parse value
+      if (unlikely(*sql == SPACE)) {
+        break;
+      }
+      if (unlikely(*sql == EQUAL)) {
+        smlBuildInvalidDataMsg(msg, "invalid data", sql);
+        return TSDB_CODE_SML_INVALID_DATA;
+      }
+      sql++;
+    }
+    valueLen = sql - value;
+
+    if (unlikely(valueLen == 0)) {
+      smlBuildInvalidDataMsg(msg, "invalid value", value);
+      return TSDB_CODE_TSC_INVALID_VALUE;
+    }
+
+    if (unlikely(valueLen > (TSDB_MAX_NCHAR_LEN - VARSTR_HEADER_SIZE) / TSDB_NCHAR_SIZE)) {
+      return TSDB_CODE_PAR_INVALID_VAR_COLUMN_LEN;
+    }
+
+    SSmlKv kv = {.key = key, .keyLen = keyLen, .type = TSDB_DATA_TYPE_NCHAR, .value = value, .length = valueLen};
+
+    if(info->dataFormat){
+      if(unlikely(cnt + 1 > info->currSTableMeta->tableInfo.numOfTags)){
+        info->dataFormat = false;
+        info->reRun      = true;
+        return TSDB_CODE_SUCCESS;
+      }
+
+      if(isSameMeasure){
+        if(unlikely(cnt >= taosArrayGetSize(preLineKV))) {
+          info->dataFormat = false;
+          info->reRun      = true;
+          return TSDB_CODE_SUCCESS;
+        }
+        SSmlKv *preKV = (SSmlKv *)taosArrayGet(preLineKV, cnt);
+        if(unlikely(kv.length > preKV->length)){
+          preKV->length = kv.length;
+          SSmlSTableMeta *tableMeta = (SSmlSTableMeta *)nodeListGet(info->superTables, elements->measure, elements->measureLen, NULL);
+          ASSERT(tableMeta != NULL);
+
+          SSmlKv *oldKV = (SSmlKv *)taosArrayGet(tableMeta->tags, cnt);
+          oldKV->length = kv.length;
+          info->needModifySchema = true;
+        }
+        if(unlikely(!IS_SAME_KEY)){
+          info->dataFormat = false;
+          info->reRun      = true;
+          return TSDB_CODE_SUCCESS;
+        }
+      }else{
+        if(isSuperKVInit){
+          if(unlikely(cnt >= taosArrayGetSize(superKV))) {
+            info->dataFormat = false;
+            info->reRun      = true;
+            return TSDB_CODE_SUCCESS;
+          }
+          SSmlKv *preKV = (SSmlKv *)taosArrayGet(superKV, cnt);
+          if(unlikely(kv.length > preKV->length)) {
+            preKV->length = kv.length;
+          }else{
+            kv.length = preKV->length;
+          }
+          info->needModifySchema = true;
+
+          if(unlikely(!IS_SAME_KEY)){
+            info->dataFormat = false;
+            info->reRun      = true;
+            return TSDB_CODE_SUCCESS;
+          }
+        }else{
+          taosArrayPush(superKV, &kv);
+        }
+        taosArrayPush(preLineKV, &kv);
+      }
+    }else{
+      taosArrayPush(preLineKV, &kv);
+    }
+    cnt++;
+  }
+  SSmlTableInfo *tinfo = (SSmlTableInfo *)nodeListGet(info->childTables, elements, POINTER_BYTES, is_same_child_table_telnet);
+  if (unlikely(tinfo == NULL)) {
+    tinfo = smlBuildTableInfo(1, elements->measure, elements->measureLen);
+    if (!tinfo) {
+      return TSDB_CODE_OUT_OF_MEMORY;
+    }
+    tinfo->tags = taosArrayDup(preLineKV, NULL);
+
+    smlSetCTableName(tinfo);
+    if (info->dataFormat) {
+      info->currSTableMeta->uid = tinfo->uid;
+      tinfo->tableDataCtx = smlInitTableDataCtx(info->pQuery, info->currSTableMeta);
+      if (tinfo->tableDataCtx == NULL) {
+        smlBuildInvalidDataMsg(&info->msgBuf, "smlInitTableDataCtx error", NULL);
+        return TSDB_CODE_SML_INVALID_DATA;
+      }
+    }
+
+    SSmlLineInfo *key = (SSmlLineInfo *)taosMemoryMalloc(sizeof(SSmlLineInfo));
+    *key = *elements;
+    tinfo->key = key;
+    nodeListSet(&info->childTables, key, POINTER_BYTES, tinfo, is_same_child_table_telnet);
+  }
+  if (info->dataFormat) info->currTableDataCtx = tinfo->tableDataCtx;
+
+  return TSDB_CODE_SUCCESS;
+}
+
+// format: <metric> <timestamp> <value> <tagk_1>=<tagv_1>[ <tagk_n>=<tagv_n>]
+int32_t smlParseTelnetString(SSmlHandle *info, char *sql, char *sqlEnd, SSmlLineInfo *elements) {
+  if (!sql) return TSDB_CODE_SML_INVALID_DATA;
+
+  // parse metric
+  smlParseTelnetElement(&sql, sqlEnd, &elements->measure, &elements->measureLen);
+  if (unlikely((!(elements->measure) || IS_INVALID_TABLE_LEN(elements->measureLen)))) {
+    smlBuildInvalidDataMsg(&info->msgBuf, "invalid data", sql);
+    return TSDB_CODE_TSC_INVALID_TABLE_ID_LENGTH;
+  }
+
+  // parse timestamp
+  smlParseTelnetElement(&sql, sqlEnd, &elements->timestamp, &elements->timestampLen);
+  if (unlikely(!elements->timestamp || elements->timestampLen == 0)) {
+    smlBuildInvalidDataMsg(&info->msgBuf, "invalid timestamp", sql);
+    return TSDB_CODE_SML_INVALID_DATA;
+  }
+
+  bool needConverTime = false;  // get TS before parse tag(get meta), so need conver time
+  if(info->dataFormat && info->currSTableMeta == NULL){
+    needConverTime = true;
+  }
+  int64_t ts = smlParseOpenTsdbTime(info, elements->timestamp, elements->timestampLen);
+  if (unlikely(ts < 0)) {
+    smlBuildInvalidDataMsg(&info->msgBuf, "invalid timestamp", sql);
+    return TSDB_CODE_INVALID_TIMESTAMP;
+  }
+  SSmlKv kvTs = { .key = TS, .keyLen = TS_LEN, .type = TSDB_DATA_TYPE_TIMESTAMP, .i = ts, .length = (size_t)tDataTypes[TSDB_DATA_TYPE_TIMESTAMP].bytes};
+
+  // parse value
+  smlParseTelnetElement(&sql, sqlEnd, &elements->cols, &elements->colsLen);
+  if (unlikely(!elements->cols || elements->colsLen == 0)) {
+    smlBuildInvalidDataMsg(&info->msgBuf, "invalid value", sql);
+    return TSDB_CODE_TSC_INVALID_VALUE;
+  }
+
+  SSmlKv kv = {.key = VALUE, .keyLen = VALUE_LEN, .value = elements->cols, .length = (size_t)elements->colsLen};
+  if (smlParseValue(&kv, &info->msgBuf) != TSDB_CODE_SUCCESS) {
+    return TSDB_CODE_TSC_INVALID_VALUE;
+  }
+
+  JUMP_SPACE(sql, sqlEnd)
+
+  elements->tags = sql;
+  elements->tagsLen = sqlEnd - sql;
+  if (unlikely(!elements->tags || elements->tagsLen == 0)) {
+    smlBuildInvalidDataMsg(&info->msgBuf, "invalid value", sql);
+    return TSDB_CODE_TSC_INVALID_VALUE;
+  }
+
+  int ret = smlParseTelnetTags(info, sql, sqlEnd, elements, &info->msgBuf);
+  if (unlikely(ret != TSDB_CODE_SUCCESS)) {
+    return ret;
+  }
+
+  if(unlikely(info->reRun)){
+    return TSDB_CODE_SUCCESS;
+  }
+
+  if(info->dataFormat){
+    if(needConverTime) {
+      kvTs.i = convertTimePrecision(kvTs.i, TSDB_TIME_PRECISION_NANO, info->currSTableMeta->tableInfo.precision);
+    }
+    ret = smlBuildCol(info->currTableDataCtx, info->currSTableMeta->schema, &kvTs, 0);
+    if(ret == TSDB_CODE_SUCCESS){
+      ret = smlBuildCol(info->currTableDataCtx, info->currSTableMeta->schema, &kv, 1);
+    }
+    if(ret == TSDB_CODE_SUCCESS){
+      ret = smlBuildRow(info->currTableDataCtx);
+    }
+    if (unlikely(ret != TSDB_CODE_SUCCESS)) {
+      smlBuildInvalidDataMsg(&info->msgBuf, "smlBuildCol error", NULL);
+      return ret;
+    }
+  }else{
+    if(elements->colArray == NULL){
+      elements->colArray = taosArrayInit(16, sizeof(SSmlKv));
+    }
+    taosArrayPush(elements->colArray, &kvTs);
+    taosArrayPush(elements->colArray, &kv);
+  }
+  info->preLine = *elements;
+
+  return TSDB_CODE_SUCCESS;
+}
--- a/source/client/src/clientStmt.c
+++ b/source/client/src/clientStmt.c
@ -152,9 +152,10 @@ int32_t stmtRestoreQueryFields(STscStmt* pStmt) {
  return TSDB_CODE_SUCCESS;
 }

-int32_t stmtUpdateBindInfo(TAOS_STMT* stmt, STableMeta* pTableMeta, void* tags, SName* tbName, const char* sTableName, bool autoCreateTbl) {
+int32_t stmtUpdateBindInfo(TAOS_STMT* stmt, STableMeta* pTableMeta, void* tags, SName* tbName, const char* sTableName,
+                           bool autoCreateTbl) {
  STscStmt* pStmt = (STscStmt*)stmt;
-  char           tbFName[TSDB_TABLE_FNAME_LEN];
+  char      tbFName[TSDB_TABLE_FNAME_LEN];
  tNameExtractFullName(tbName, tbFName);

  memcpy(&pStmt->bInfo.sname, tbName, sizeof(*tbName));
@ -171,12 +172,11 @@ int32_t stmtUpdateBindInfo(TAOS_STMT* stmt, STableMeta* pTableMeta, void* tags,
  return TSDB_CODE_SUCCESS;
 }

-int32_t stmtUpdateExecInfo(TAOS_STMT* stmt, SHashObj* pVgHash, SHashObj* pBlockHash, bool autoCreateTbl) {
+int32_t stmtUpdateExecInfo(TAOS_STMT* stmt, SHashObj* pVgHash, SHashObj* pBlockHash) {
  STscStmt* pStmt = (STscStmt*)stmt;

  pStmt->sql.pVgHash = pVgHash;
  pStmt->exec.pBlockHash = pBlockHash;
-  pStmt->exec.autoCreateTbl = autoCreateTbl;

  return TSDB_CODE_SUCCESS;
 }
@ -186,7 +186,7 @@ int32_t stmtUpdateInfo(TAOS_STMT* stmt, STableMeta* pTableMeta, void* tags, SNam
  STscStmt* pStmt = (STscStmt*)stmt;

  STMT_ERR_RET(stmtUpdateBindInfo(stmt, pTableMeta, tags, tbName, sTableName, autoCreateTbl));
-  STMT_ERR_RET(stmtUpdateExecInfo(stmt, pVgHash, pBlockHash, autoCreateTbl));
+  STMT_ERR_RET(stmtUpdateExecInfo(stmt, pVgHash, pBlockHash));

  pStmt->sql.autoCreateTbl = autoCreateTbl;

@ -214,16 +214,16 @@ int32_t stmtCacheBlock(STscStmt* pStmt) {
    return TSDB_CODE_SUCCESS;
  }

-  STableDataBlocks** pSrc = taosHashGet(pStmt->exec.pBlockHash, pStmt->bInfo.tbFName, strlen(pStmt->bInfo.tbFName));
+  STableDataCxt** pSrc = taosHashGet(pStmt->exec.pBlockHash, pStmt->bInfo.tbFName, strlen(pStmt->bInfo.tbFName));
  if (!pSrc) {
    return TSDB_CODE_OUT_OF_MEMORY;
  }
-  STableDataBlocks* pDst = NULL;
+  STableDataCxt* pDst = NULL;

-  STMT_ERR_RET(qCloneStmtDataBlock(&pDst, *pSrc));
+  STMT_ERR_RET(qCloneStmtDataBlock(&pDst, *pSrc, true));

  SStmtTableCache cache = {
-      .pDataBlock = pDst,
+      .pDataCtx = pDst,
      .boundTags = pStmt->bInfo.boundTags,
  };

@ -241,6 +241,8 @@ int32_t stmtCacheBlock(STscStmt* pStmt) {
 }

 int32_t stmtParseSql(STscStmt* pStmt) {
+  pStmt->exec.pCurrBlock = NULL;
+
  SStmtCallback stmtCb = {
      .pStmt = pStmt,
      .getTbNameFn = stmtGetTbName,
@ -273,7 +275,7 @@ int32_t stmtCleanBindInfo(STscStmt* pStmt) {
  pStmt->bInfo.tbName[0] = 0;
  pStmt->bInfo.tbFName[0] = 0;
  if (!pStmt->bInfo.tagsCached) {
-    destroyBoundColumnInfo(pStmt->bInfo.boundTags);
+    qDestroyBoundColInfo(pStmt->bInfo.boundTags);
    taosMemoryFreeClear(pStmt->bInfo.boundTags);
  }
  memset(pStmt->bInfo.stbFName, 0, TSDB_TABLE_FNAME_LEN);
@ -289,29 +291,25 @@ int32_t stmtCleanExecInfo(STscStmt* pStmt, bool keepTable, bool deepClean) {
  size_t keyLen = 0;
  void*  pIter = taosHashIterate(pStmt->exec.pBlockHash, NULL);
  while (pIter) {
-    STableDataBlocks* pBlocks = *(STableDataBlocks**)pIter;
-    char*             key = taosHashGetKey(pIter, &keyLen);
-    STableMeta*       pMeta = qGetTableMetaInDataBlock(pBlocks);
+    STableDataCxt* pBlocks = *(STableDataCxt**)pIter;
+    char*          key = taosHashGetKey(pIter, &keyLen);
+    STableMeta*    pMeta = qGetTableMetaInDataBlock(pBlocks);

-    if (keepTable && (strlen(pStmt->bInfo.tbFName) == keyLen) && strncmp(pStmt->bInfo.tbFName, key, keyLen) == 0) {
-      STMT_ERR_RET(qResetStmtDataBlock(pBlocks, true));
+    if (keepTable && pBlocks == pStmt->exec.pCurrBlock) {
+      ASSERT(NULL == pBlocks->pData);
+      TSWAP(pBlocks->pData, pStmt->exec.pCurrTbData);
+      STMT_ERR_RET(qResetStmtDataBlock(pBlocks, false));

      pIter = taosHashIterate(pStmt->exec.pBlockHash, pIter);
      continue;
    }

-    if (STMT_TYPE_MULTI_INSERT == pStmt->sql.type) {
-      qFreeStmtDataBlock(pBlocks);
-    } else {
-      qDestroyStmtDataBlock(pBlocks);
-    }
+    qDestroyStmtDataBlock(pBlocks);
    taosHashRemove(pStmt->exec.pBlockHash, key, keyLen);

    pIter = taosHashIterate(pStmt->exec.pBlockHash, pIter);
  }

-  pStmt->exec.autoCreateTbl = false;
-
  if (keepTable) {
    return TSDB_CODE_SUCCESS;
  }
@ -319,6 +317,9 @@ int32_t stmtCleanExecInfo(STscStmt* pStmt, bool keepTable, bool deepClean) {
  taosHashCleanup(pStmt->exec.pBlockHash);
  pStmt->exec.pBlockHash = NULL;

+  tDestroySSubmitTbData(pStmt->exec.pCurrTbData, TSDB_MSG_FLG_ENCODE);
+  taosMemoryFreeClear(pStmt->exec.pCurrTbData);
+
  STMT_ERR_RET(stmtCleanBindInfo(pStmt));

  return TSDB_CODE_SUCCESS;
@ -337,8 +338,8 @@ int32_t stmtCleanSQLInfo(STscStmt* pStmt) {
  while (pIter) {
    SStmtTableCache* pCache = (SStmtTableCache*)pIter;

-    qDestroyStmtDataBlock(pCache->pDataBlock);
-    destroyBoundColumnInfo(pCache->boundTags);
+    qDestroyStmtDataBlock(pCache->pDataCtx);
+    qDestroyBoundColInfo(pCache->boundTags);
    taosMemoryFreeClear(pCache->boundTags);

    pIter = taosHashIterate(pStmt->sql.pTableCache, pIter);
@ -354,7 +355,8 @@ int32_t stmtCleanSQLInfo(STscStmt* pStmt) {
  return TSDB_CODE_SUCCESS;
 }

-int32_t stmtRebuildDataBlock(STscStmt* pStmt, STableDataBlocks* pDataBlock, STableDataBlocks** newBlock, uint64_t uid) {
+int32_t stmtRebuildDataBlock(STscStmt* pStmt, STableDataCxt* pDataBlock, STableDataCxt** newBlock, uint64_t uid,
+                             uint64_t suid) {
  SEpSet           ep = getEpSet_s(&pStmt->taos->pAppInfo->mgmtEp);
  SVgroupInfo      vgInfo = {0};
  SRequestConnInfo conn = {.pTrans = pStmt->taos->pAppInfo->pTransporter,
@ -366,7 +368,9 @@ int32_t stmtRebuildDataBlock(STscStmt* pStmt, STableDataBlocks* pDataBlock, STab
  STMT_ERR_RET(
      taosHashPut(pStmt->sql.pVgHash, (const char*)&vgInfo.vgId, sizeof(vgInfo.vgId), (char*)&vgInfo, sizeof(vgInfo)));

-  STMT_ERR_RET(qRebuildStmtDataBlock(newBlock, pDataBlock, uid, vgInfo.vgId));
+  STMT_ERR_RET(qRebuildStmtDataBlock(newBlock, pDataBlock, uid, suid, vgInfo.vgId, pStmt->sql.autoCreateTbl));
+
+  STMT_DLOG("tableDataCxt rebuilt, uid:%" PRId64 ", vgId:%d", uid, vgInfo.vgId);

  return TSDB_CODE_SUCCESS;
 }
@ -375,12 +379,13 @@ int32_t stmtGetFromCache(STscStmt* pStmt) {
  pStmt->bInfo.needParse = true;
  pStmt->bInfo.inExecCache = false;

-  STableDataBlocks* pBlockInExec =
-      taosHashGet(pStmt->exec.pBlockHash, pStmt->bInfo.tbFName, strlen(pStmt->bInfo.tbFName));
-  if (pBlockInExec) {
+  STableDataCxt** pCxtInExec = taosHashGet(pStmt->exec.pBlockHash, pStmt->bInfo.tbFName, strlen(pStmt->bInfo.tbFName));
+  if (pCxtInExec) {
    pStmt->bInfo.needParse = false;
    pStmt->bInfo.inExecCache = true;

+    pStmt->exec.pCurrBlock = *pCxtInExec;
+
    if (pStmt->sql.autoCreateTbl) {
      tscDebug("reuse stmt block for tb %s in execBlock", pStmt->bInfo.tbFName);
      return TSDB_CODE_SUCCESS;
@ -407,18 +412,18 @@ int32_t stmtGetFromCache(STscStmt* pStmt) {
    SStmtTableCache* pCache = taosHashGet(pStmt->sql.pTableCache, &pStmt->bInfo.tbSuid, sizeof(pStmt->bInfo.tbSuid));
    if (pCache) {
      pStmt->bInfo.needParse = false;
-      pStmt->exec.autoCreateTbl = true;
-
      pStmt->bInfo.tbUid = 0;

-      STableDataBlocks* pNewBlock = NULL;
-      STMT_ERR_RET(stmtRebuildDataBlock(pStmt, pCache->pDataBlock, &pNewBlock, 0));
+      STableDataCxt* pNewBlock = NULL;
+      STMT_ERR_RET(stmtRebuildDataBlock(pStmt, pCache->pDataCtx, &pNewBlock, 0, pStmt->bInfo.tbSuid));

      if (taosHashPut(pStmt->exec.pBlockHash, pStmt->bInfo.tbFName, strlen(pStmt->bInfo.tbFName), &pNewBlock,
                      POINTER_BYTES)) {
        STMT_ERR_RET(TSDB_CODE_OUT_OF_MEMORY);
      }

+      pStmt->exec.pCurrBlock = pNewBlock;
+
      tscDebug("reuse stmt block for tb %s in sqlBlock, suid:0x%" PRIx64, pStmt->bInfo.tbFName, pStmt->bInfo.tbSuid);

      return TSDB_CODE_SUCCESS;
@ -489,14 +494,16 @@ int32_t stmtGetFromCache(STscStmt* pStmt) {
    pStmt->bInfo.boundTags = pCache->boundTags;
    pStmt->bInfo.tagsCached = true;

-    STableDataBlocks* pNewBlock = NULL;
-    STMT_ERR_RET(stmtRebuildDataBlock(pStmt, pCache->pDataBlock, &pNewBlock, uid));
+    STableDataCxt* pNewBlock = NULL;
+    STMT_ERR_RET(stmtRebuildDataBlock(pStmt, pCache->pDataCtx, &pNewBlock, uid, suid));

    if (taosHashPut(pStmt->exec.pBlockHash, pStmt->bInfo.tbFName, strlen(pStmt->bInfo.tbFName), &pNewBlock,
                    POINTER_BYTES)) {
      STMT_ERR_RET(TSDB_CODE_OUT_OF_MEMORY);
    }

+    pStmt->exec.pCurrBlock = pNewBlock;
+
    tscDebug("tb %s in sqlBlock list, set to current", pStmt->bInfo.tbFName);

    return TSDB_CODE_SUCCESS;
@ -614,8 +621,8 @@ int stmtSetTbTags(TAOS_STMT* stmt, TAOS_MULTI_BIND* tags) {
    return TSDB_CODE_SUCCESS;
  }

-  STableDataBlocks** pDataBlock =
-      (STableDataBlocks**)taosHashGet(pStmt->exec.pBlockHash, pStmt->bInfo.tbFName, strlen(pStmt->bInfo.tbFName));
+  STableDataCxt** pDataBlock =
+      (STableDataCxt**)taosHashGet(pStmt->exec.pBlockHash, pStmt->bInfo.tbFName, strlen(pStmt->bInfo.tbFName));
  if (NULL == pDataBlock) {
    tscError("table %s not found in exec blockHash", pStmt->bInfo.tbFName);
    STMT_ERR_RET(TSDB_CODE_APP_ERROR);
@ -626,8 +633,6 @@ int stmtSetTbTags(TAOS_STMT* stmt, TAOS_MULTI_BIND* tags) {
                                  pStmt->bInfo.sname.tname, tags, pStmt->exec.pRequest->msgBuf,
                                  pStmt->exec.pRequest->msgBufLen));

-  pStmt->exec.autoCreateTbl = true;
-
  return TSDB_CODE_SUCCESS;
 }

@ -637,8 +642,8 @@ int stmtFetchTagFields(STscStmt* pStmt, int32_t* fieldNum, TAOS_FIELD_E** fields
    STMT_ERR_RET(TSDB_CODE_TSC_STMT_API_ERROR);
  }

-  STableDataBlocks** pDataBlock =
-      (STableDataBlocks**)taosHashGet(pStmt->exec.pBlockHash, pStmt->bInfo.tbFName, strlen(pStmt->bInfo.tbFName));
+  STableDataCxt** pDataBlock =
+      (STableDataCxt**)taosHashGet(pStmt->exec.pBlockHash, pStmt->bInfo.tbFName, strlen(pStmt->bInfo.tbFName));
  if (NULL == pDataBlock) {
    tscError("table %s not found in exec blockHash", pStmt->bInfo.tbFName);
    STMT_ERR_RET(TSDB_CODE_APP_ERROR);
@ -655,8 +660,8 @@ int stmtFetchColFields(STscStmt* pStmt, int32_t* fieldNum, TAOS_FIELD_E** fields
    STMT_ERR_RET(TSDB_CODE_TSC_STMT_API_ERROR);
  }

-  STableDataBlocks** pDataBlock =
-      (STableDataBlocks**)taosHashGet(pStmt->exec.pBlockHash, pStmt->bInfo.tbFName, strlen(pStmt->bInfo.tbFName));
+  STableDataCxt** pDataBlock =
+      (STableDataCxt**)taosHashGet(pStmt->exec.pBlockHash, pStmt->bInfo.tbFName, strlen(pStmt->bInfo.tbFName));
  if (NULL == pDataBlock) {
    tscError("table %s not found in exec blockHash", pStmt->bInfo.tbFName);
    STMT_ERR_RET(TSDB_CODE_APP_ERROR);
@ -729,11 +734,18 @@ int stmtBindBatch(TAOS_STMT* stmt, TAOS_MULTI_BIND* bind, int32_t colIdx) {
    return TSDB_CODE_SUCCESS;
  }

-  STableDataBlocks** pDataBlock =
-      (STableDataBlocks**)taosHashGet(pStmt->exec.pBlockHash, pStmt->bInfo.tbFName, strlen(pStmt->bInfo.tbFName));
-  if (NULL == pDataBlock) {
-    tscError("table %s not found in exec blockHash", pStmt->bInfo.tbFName);
-    STMT_ERR_RET(TSDB_CODE_APP_ERROR);
+  STableDataCxt** pDataBlock = NULL;
+
+  if (pStmt->exec.pCurrBlock) {
+    pDataBlock = &pStmt->exec.pCurrBlock;
+  } else {
+    pDataBlock =
+        (STableDataCxt**)taosHashGet(pStmt->exec.pBlockHash, pStmt->bInfo.tbFName, strlen(pStmt->bInfo.tbFName));
+    if (NULL == pDataBlock) {
+      tscError("table %s not found in exec blockHash", pStmt->bInfo.tbFName);
+      STMT_ERR_RET(TSDB_CODE_TSC_STMT_CACHE_ERROR);
+    }
+    pStmt->exec.pCurrBlock = *pDataBlock;
  }

  if (colIdx < 0) {
@ -779,10 +791,10 @@ int stmtUpdateTableUid(STscStmt* pStmt, SSubmitRsp* pRsp) {
  int32_t code = 0;
  int32_t finalCode = 0;
  size_t  keyLen = 0;
-  STableDataBlocks** pIter = taosHashIterate(pStmt->exec.pBlockHash, NULL);
+  void*   pIter = taosHashIterate(pStmt->exec.pBlockHash, NULL);
  while (pIter) {
-    STableDataBlocks* pBlock = *pIter;
-    char*             key = taosHashGetKey(pIter, &keyLen);
+    STableDataCxt* pBlock = *(STableDataCxt**)pIter;
+    char*          key = taosHashGetKey(pIter, &keyLen);

    STableMeta* pMeta = qGetTableMetaInDataBlock(pBlock);
    if (pMeta->uid) {
@ -848,7 +860,7 @@ int stmtUpdateTableUid(STscStmt* pStmt, SSubmitRsp* pRsp) {

      pMeta->uid = pTableMeta->uid;
      pStmt->bInfo.tbUid = pTableMeta->uid;
-      taosMemoryFree(pTableMeta);      
+      taosMemoryFree(pTableMeta);
    }

    pIter = taosHashIterate(pStmt->exec.pBlockHash, pIter);
@ -861,7 +873,6 @@ int stmtExec(TAOS_STMT* stmt) {
  STscStmt*   pStmt = (STscStmt*)stmt;
  int32_t     code = 0;
  SSubmitRsp* pRsp = NULL;
-  bool        autoCreateTbl = pStmt->exec.autoCreateTbl;

  STMT_DLOG_E("start to exec");

@ -870,8 +881,13 @@ int stmtExec(TAOS_STMT* stmt) {
  if (STMT_TYPE_QUERY == pStmt->sql.type) {
    launchQueryImpl(pStmt->exec.pRequest, pStmt->sql.pQuery, true, NULL);
  } else {
+    tDestroySSubmitTbData(pStmt->exec.pCurrTbData, TSDB_MSG_FLG_ENCODE);
+    taosMemoryFreeClear(pStmt->exec.pCurrTbData);
+
+    STMT_ERR_RET(qCloneCurrentTbData(pStmt->exec.pCurrBlock, &pStmt->exec.pCurrTbData));
+
    STMT_ERR_RET(qBuildStmtOutput(pStmt->sql.pQuery, pStmt->sql.pVgHash, pStmt->exec.pBlockHash));
-    launchQueryImpl(pStmt->exec.pRequest, pStmt->sql.pQuery, true, (autoCreateTbl ? (void**)&pRsp : NULL));
+    launchQueryImpl(pStmt->exec.pRequest, pStmt->sql.pQuery, true, NULL);
  }

  if (pStmt->exec.pRequest->code && NEED_CLIENT_HANDLE_ERROR(pStmt->exec.pRequest->code)) {
@ -894,15 +910,6 @@ _return:

  stmtCleanExecInfo(pStmt, (code ? false : true), false);

-  if (TSDB_CODE_SUCCESS == code && autoCreateTbl) {
-    if (NULL == pRsp) {
-      tscError("no submit resp got for auto create table");
-      code = TSDB_CODE_APP_ERROR;
-    } else {
-      code = stmtUpdateTableUid(pStmt, pRsp);
-    }
-  }
-
  tFreeSSubmitRsp(pRsp);

  ++pStmt->sql.runTimes;
--- a/source/client/test/smlTest.cpp
+++ b/source/client/test/smlTest.cpp
--- a/source/common/src/tdatablock.c
+++ b/source/common/src/tdatablock.c
@ -19,7 +19,7 @@
 #include "tlog.h"
 #include "tname.h"

-#define MALLOC_ALIGN_BYTES  256
+#define MALLOC_ALIGN_BYTES 32

 int32_t colDataGetLength(const SColumnInfoData* pColumnInfoData, int32_t numOfRows) {
  ASSERT(pColumnInfoData != NULL);
@ -38,7 +38,8 @@ int32_t colDataGetFullLength(const SColumnInfoData* pColumnInfoData, int32_t num
  if (IS_VAR_DATA_TYPE(pColumnInfoData->info.type)) {
    return pColumnInfoData->varmeta.length + sizeof(int32_t) * numOfRows;
  } else {
-    return ((pColumnInfoData->info.type == TSDB_DATA_TYPE_NULL) ? 0 : pColumnInfoData->info.bytes * numOfRows) + BitmapLen(numOfRows);
+    return ((pColumnInfoData->info.type == TSDB_DATA_TYPE_NULL) ? 0 : pColumnInfoData->info.bytes * numOfRows) +
+           BitmapLen(numOfRows);
  }
 }

@ -279,7 +280,7 @@ int32_t colDataMergeCol(SColumnInfoData* pColumnInfoData, int32_t numOfRow1, int
      pColumnInfoData->varmeta.allocLen = len + oldLen;
    }

-    if (pColumnInfoData->pData && pSource->pData) { // TD-20382
+    if (pColumnInfoData->pData && pSource->pData) {  // TD-20382
      memcpy(pColumnInfoData->pData + oldLen, pSource->pData, len);
    }
    pColumnInfoData->varmeta.length = len + oldLen;
@ -359,7 +360,7 @@ size_t blockDataGetNumOfRows(const SSDataBlock* pBlock) { return pBlock->info.ro

 int32_t blockDataUpdateTsWindow(SSDataBlock* pDataBlock, int32_t tsColumnIndex) {
  if (pDataBlock->info.rows > 0) {
-//    ASSERT(pDataBlock->info.dataLoad == 1);
+    //    ASSERT(pDataBlock->info.dataLoad == 1);
  }

  if (pDataBlock == NULL || pDataBlock->info.rows <= 0 || pDataBlock->info.dataLoad == 0) {
@ -1167,8 +1168,11 @@ void blockDataEmpty(SSDataBlock* pDataBlock) {
 }

 // todo temporarily disable it
-static int32_t doEnsureCapacity(SColumnInfoData* pColumn, const SDataBlockInfo* pBlockInfo, uint32_t numOfRows, bool clearPayload) {
+
+static int32_t doEnsureCapacity(SColumnInfoData* pColumn, const SDataBlockInfo* pBlockInfo, uint32_t numOfRows,
+                                bool clearPayload) {
  ASSERT(numOfRows > 0);
+
  if (numOfRows <= pBlockInfo->capacity) {
    return TSDB_CODE_SUCCESS;
  }
@ -1225,7 +1229,7 @@ static int32_t doEnsureCapacity(SColumnInfoData* pColumn, const SDataBlockInfo*
  return TSDB_CODE_SUCCESS;
 }

-void  colInfoDataCleanup(SColumnInfoData* pColumn, uint32_t numOfRows) {
+void colInfoDataCleanup(SColumnInfoData* pColumn, uint32_t numOfRows) {
  pColumn->hasNull = false;

  if (IS_VAR_DATA_TYPE(pColumn->info.type)) {
@ -1953,7 +1957,8 @@ char* dumpBlockData(SSDataBlock* pDataBlock, const char* flag, char** pDataBuf)
                  "===stream===%s|block type %d|child id %d|group id:%" PRIu64 "|uid:%" PRId64
                  "|rows:%d|version:%" PRIu64 "|cal start:%" PRIu64 "|cal end:%" PRIu64 "\n",
                  flag, (int32_t)pDataBlock->info.type, pDataBlock->info.childId, pDataBlock->info.id.groupId,
-                  pDataBlock->info.id.uid, pDataBlock->info.rows, pDataBlock->info.version, pDataBlock->info.calWin.skey, pDataBlock->info.calWin.ekey);
+                  pDataBlock->info.id.uid, pDataBlock->info.rows, pDataBlock->info.version,
+                  pDataBlock->info.calWin.skey, pDataBlock->info.calWin.ekey);
  if (len >= size - 1) return dumpBuf;

  for (int32_t j = 0; j < rows; j++) {
@ -2055,6 +2060,7 @@ char* dumpBlockData(SSDataBlock* pDataBlock, const char* flag, char** pDataBuf)
 * @param suid
 *
 */
+#if 0
 int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SSDataBlock* pDataBlock, STSchema* pTSchema, int32_t vgId,
                                    tb_uid_t suid) {
  int32_t bufSize = sizeof(SSubmitReq);
@ -2220,32 +2226,186 @@ int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SSDataBlock* pDataB

  return TSDB_CODE_SUCCESS;
 }
+#endif
+
+int32_t buildSubmitReqFromDataBlock(SSubmitReq2** ppReq, const SSDataBlock* pDataBlock, const STSchema* pTSchema,
+                                    int64_t uid, int32_t vgId, tb_uid_t suid) {
+  SSubmitReq2* pReq = *ppReq;
+  SArray*      pVals = NULL;
+  int32_t      numOfBlks = 0;
+  int32_t      sz = 1;
+
+  terrno = TSDB_CODE_SUCCESS;
+
+  if (NULL == pReq) {
+    if (!(pReq = taosMemoryMalloc(sizeof(SSubmitReq2)))) {
+      terrno = TSDB_CODE_OUT_OF_MEMORY;
+      goto _end;
+    }
+
+    if (!(pReq->aSubmitTbData = taosArrayInit(1, sizeof(SSubmitTbData)))) {
+      goto _end;
+    }
+  }
+
+  for (int32_t i = 0; i < sz; ++i) {
+    int32_t colNum = taosArrayGetSize(pDataBlock->pDataBlock);
+    int32_t rows = pDataBlock->info.rows;
+
+    if (colNum <= 1) {  // invalid if only with TS col
+      continue;
+    }
+
+    // the rsma result should has the same column number with schema.
+    ASSERT(colNum == pTSchema->numOfCols);
+
+    SSubmitTbData tbData = {0};
+
+    if (!(tbData.aRowP = taosArrayInit(rows, sizeof(SRow*)))) {
+      goto _end;
+    }
+    tbData.suid = suid;
+    tbData.uid = uid;
+    tbData.sver = pTSchema->version;
+
+    if (!pVals && !(pVals = taosArrayInit(colNum, sizeof(SColVal)))) {
+      taosArrayDestroy(tbData.aRowP);
+      goto _end;
+    }
+
+    for (int32_t j = 0; j < rows; ++j) {  // iterate by row
+
+      taosArrayClear(pVals);
+
+      bool    isStartKey = false;
+      int32_t offset = 0;
+      for (int32_t k = 0; k < colNum; ++k) {  // iterate by column
+        SColumnInfoData* pColInfoData = taosArrayGet(pDataBlock->pDataBlock, k);
+        const STColumn*  pCol = &pTSchema->columns[k];
+        void*            var = POINTER_SHIFT(pColInfoData->pData, j * pColInfoData->info.bytes);
+
+        switch (pColInfoData->info.type) {
+          case TSDB_DATA_TYPE_TIMESTAMP:
+            ASSERT(pColInfoData->info.type == pCol->type);
+            if (!isStartKey) {
+              isStartKey = true;
+              ASSERT(PRIMARYKEY_TIMESTAMP_COL_ID == pCol->colId);
+              SColVal cv = COL_VAL_VALUE(pCol->colId, pCol->type, (SValue){.val = *(TSKEY*)var});
+              taosArrayPush(pVals, &cv);
+            } else if (colDataIsNull_s(pColInfoData, j)) {
+              SColVal cv = COL_VAL_NULL(pCol->colId, pCol->type);
+              taosArrayPush(pVals, &cv);
+            } else {
+              SColVal cv = COL_VAL_VALUE(pCol->colId, pCol->type, (SValue){.val = *(int64_t*)var});
+              taosArrayPush(pVals, &cv);
+            }
+            break;
+          case TSDB_DATA_TYPE_NCHAR:
+          case TSDB_DATA_TYPE_VARCHAR: {  // TSDB_DATA_TYPE_BINARY
+            ASSERT(pColInfoData->info.type == pCol->type);
+            if (colDataIsNull_s(pColInfoData, j)) {
+              SColVal cv = COL_VAL_NULL(pCol->colId, pCol->type);
+              taosArrayPush(pVals, &cv);
+            } else {
+              void*   data = colDataGetVarData(pColInfoData, j);
+              SValue  sv = (SValue){.nData = varDataLen(data), .pData = varDataVal(data)};  // address copy, no value
+              SColVal cv = COL_VAL_VALUE(pCol->colId, pCol->type, sv);
+              taosArrayPush(pVals, &cv);
+            }
+            break;
+          }
+          case TSDB_DATA_TYPE_VARBINARY:
+          case TSDB_DATA_TYPE_DECIMAL:
+          case TSDB_DATA_TYPE_BLOB:
+          case TSDB_DATA_TYPE_JSON:
+          case TSDB_DATA_TYPE_MEDIUMBLOB:
+            uError("the column type %" PRIi16 " is defined but not implemented yet", pColInfoData->info.type);
+            ASSERT(0);
+            break;
+          default:
+            if (pColInfoData->info.type < TSDB_DATA_TYPE_MAX && pColInfoData->info.type > TSDB_DATA_TYPE_NULL) {
+              if (colDataIsNull_s(pColInfoData, j)) {
+                SColVal cv = COL_VAL_NULL(pCol->colId, pCol->type);  // should use pCol->type
+                taosArrayPush(pVals, &cv);
+              } else {
+                SValue sv;
+                if (pCol->type == pColInfoData->info.type) {
+                  memcpy(&sv.val, var, tDataTypes[pCol->type].bytes);
+                } else {
+                  /**
+                   *  1. sum/avg would convert to int64_t/uint64_t/double during aggregation
+                   *  2. below conversion may lead to overflow or loss, the app should select the right data type.
+                   */
+                  char tv[8] = {0};
+                  if (pColInfoData->info.type == TSDB_DATA_TYPE_FLOAT) {
+                    float v = 0;
+                    GET_TYPED_DATA(v, float, pColInfoData->info.type, var);
+                    SET_TYPED_DATA(&tv, pCol->type, v);
+                  } else if (pColInfoData->info.type == TSDB_DATA_TYPE_DOUBLE) {
+                    double v = 0;
+                    GET_TYPED_DATA(v, double, pColInfoData->info.type, var);
+                    SET_TYPED_DATA(&tv, pCol->type, v);
+                  } else if (IS_SIGNED_NUMERIC_TYPE(pColInfoData->info.type)) {
+                    int64_t v = 0;
+                    GET_TYPED_DATA(v, int64_t, pColInfoData->info.type, var);
+                    SET_TYPED_DATA(&tv, pCol->type, v);
+                  } else {
+                    uint64_t v = 0;
+                    GET_TYPED_DATA(v, uint64_t, pColInfoData->info.type, var);
+                    SET_TYPED_DATA(&tv, pCol->type, v);
+                  }
+                  memcpy(&sv.val, tv, tDataTypes[pCol->type].bytes);
+                }
+                SColVal cv = COL_VAL_VALUE(pCol->colId, pColInfoData->info.type, sv);
+                taosArrayPush(pVals, &cv);
+              }
+            } else {
+              uError("the column type %" PRIi16 " is undefined\n", pColInfoData->info.type);
+              ASSERT(0);
+            }
+            break;
+        }
+      }
+      SRow* pRow = NULL;
+      if ((terrno = tRowBuild(pVals, pTSchema, &pRow)) < 0) {
+        tDestroySSubmitTbData(&tbData, TSDB_MSG_FLG_ENCODE);
+        goto _end;
+      }
+      ASSERT(pRow);
+      taosArrayPush(tbData.aRowP, &pRow);
+    }
+
+    taosArrayPush(pReq->aSubmitTbData, &tbData);
+  }
+_end:
+  taosArrayDestroy(pVals);
+  if (terrno != 0) {
+    *ppReq = NULL;
+    if (pReq) tDestroySSubmitReq2(pReq, TSDB_MSG_FLG_ENCODE);
+    return TSDB_CODE_FAILED;
+  }
+  *ppReq = pReq;
+  return TSDB_CODE_SUCCESS;
+}

 char* buildCtbNameByGroupId(const char* stbFullName, uint64_t groupId) {
  ASSERT(stbFullName[0] != 0);
-  SArray* tags = taosArrayInit(0, sizeof(void*));
+  SArray* tags = taosArrayInit(0, sizeof(SSmlKv));
  if (tags == NULL) {
    return NULL;
  }

-  SSmlKv* pTag = taosMemoryCalloc(1, sizeof(SSmlKv));
-  if (pTag == NULL) {
-    taosArrayDestroy(tags);
-    return NULL;
-  }
-
  void* cname = taosMemoryCalloc(1, TSDB_TABLE_NAME_LEN + 1);
  if (cname == NULL) {
    taosArrayDestroy(tags);
-    taosMemoryFree(pTag);
    return NULL;
  }

-  pTag->key = "group_id";
-  pTag->keyLen = strlen(pTag->key);
-  pTag->type = TSDB_DATA_TYPE_UBIGINT;
-  pTag->u = groupId;
-  pTag->length = sizeof(uint64_t);
+  SSmlKv pTag = {.key = "group_id",
+                 .keyLen = sizeof("group_id") - 1,
+                 .type = TSDB_DATA_TYPE_UBIGINT,
+                 .u = groupId,
+                 .length = sizeof(uint64_t)};
  taosArrayPush(tags, &pTag);

  RandTableName rname = {
@ -2257,7 +2417,6 @@ char* buildCtbNameByGroupId(const char* stbFullName, uint64_t groupId) {

  buildChildTableName(&rname);

-  taosMemoryFree(pTag);
  taosArrayDestroy(tags);

  ASSERT(rname.ctbShortName && rname.ctbShortName[0]);
--- a/source/common/src/tdataformat.c
+++ b/source/common/src/tdataformat.c
--- a/source/common/src/tglobal.c
+++ b/source/common/src/tglobal.c
@ -47,7 +47,7 @@ int32_t tsNumOfMnodeQueryThreads = 4;
 int32_t tsNumOfMnodeFetchThreads = 1;
 int32_t tsNumOfMnodeReadThreads = 1;
 int32_t tsNumOfVnodeQueryThreads = 4;
-int32_t tsNumOfVnodeStreamThreads = 2;
+float   tsRatioOfVnodeStreamThreads = 1.0;
 int32_t tsNumOfVnodeFetchThreads = 4;
 int32_t tsNumOfVnodeRsmaThreads = 2;
 int32_t tsNumOfQnodeQueryThreads = 4;
@ -392,9 +392,7 @@ static int32_t taosAddServerCfg(SConfig *pCfg) {
  tsNumOfVnodeQueryThreads = TMAX(tsNumOfVnodeQueryThreads, 4);
  if (cfgAddInt32(pCfg, "numOfVnodeQueryThreads", tsNumOfVnodeQueryThreads, 4, 1024, 0) != 0) return -1;

-  tsNumOfVnodeStreamThreads = tsNumOfCores / 4;
-  tsNumOfVnodeStreamThreads = TMAX(tsNumOfVnodeStreamThreads, 4);
-  if (cfgAddInt32(pCfg, "numOfVnodeStreamThreads", tsNumOfVnodeStreamThreads, 4, 1024, 0) != 0) return -1;
+  if (cfgAddFloat(pCfg, "ratioOfVnodeStreamThreads", tsRatioOfVnodeStreamThreads, 0.01, 100, 0) != 0) return -1;

  tsNumOfVnodeFetchThreads = tsNumOfCores / 4;
  tsNumOfVnodeFetchThreads = TMAX(tsNumOfVnodeFetchThreads, 4);
@ -513,11 +511,9 @@ static int32_t taosUpdateServerCfg(SConfig *pCfg) {
    pItem->stype = stype;
  }

-  pItem = cfgGetItem(tsCfg, "numOfVnodeStreamThreads");
+  pItem = cfgGetItem(tsCfg, "ratioOfVnodeStreamThreads");
  if (pItem != NULL && pItem->stype == CFG_STYPE_DEFAULT) {
-    tsNumOfVnodeStreamThreads = numOfCores / 4;
-    tsNumOfVnodeStreamThreads = TMAX(tsNumOfVnodeStreamThreads, 4);
-    pItem->i32 = tsNumOfVnodeStreamThreads;
+    pItem->fval = tsRatioOfVnodeStreamThreads;
    pItem->stype = stype;
  }

@ -710,7 +706,7 @@ static int32_t taosSetServerCfg(SConfig *pCfg) {
  tsNumOfCommitThreads = cfgGetItem(pCfg, "numOfCommitThreads")->i32;
  tsNumOfMnodeReadThreads = cfgGetItem(pCfg, "numOfMnodeReadThreads")->i32;
  tsNumOfVnodeQueryThreads = cfgGetItem(pCfg, "numOfVnodeQueryThreads")->i32;
-  tsNumOfVnodeStreamThreads = cfgGetItem(pCfg, "numOfVnodeStreamThreads")->i32;
+  tsRatioOfVnodeStreamThreads = cfgGetItem(pCfg, "ratioOfVnodeStreamThreads")->fval;
  tsNumOfVnodeFetchThreads = cfgGetItem(pCfg, "numOfVnodeFetchThreads")->i32;
  tsNumOfVnodeRsmaThreads = cfgGetItem(pCfg, "numOfVnodeRsmaThreads")->i32;
  tsNumOfQnodeQueryThreads = cfgGetItem(pCfg, "numOfQnodeQueryThreads")->i32;
@ -1316,12 +1312,17 @@ void taosSetDebugFlag(int32_t *pFlagPtr, const char *flagName, int32_t flagVal,
  if (pItem != NULL && (rewrite || pItem->i32 == 0)) {
    pItem->i32 = flagVal;
  }
-  *pFlagPtr = flagVal;
+  if (pFlagPtr != NULL) {
+    *pFlagPtr = flagVal;
+  }
 }

 void taosSetAllDebugFlag(int32_t flag, bool rewrite) {
  if (flag <= 0) return;

+  taosSetDebugFlag(NULL, "debugFlag", flag, rewrite);
+  taosSetDebugFlag(NULL, "simDebugFlag", flag, rewrite);
+  taosSetDebugFlag(NULL, "tmrDebugFlag", flag, rewrite);
  taosSetDebugFlag(&uDebugFlag, "uDebugFlag", flag, rewrite);
  taosSetDebugFlag(&rpcDebugFlag, "rpcDebugFlag", flag, rewrite);
  taosSetDebugFlag(&jniDebugFlag, "jniDebugFlag", flag, rewrite);
@ -1340,6 +1341,5 @@ void taosSetAllDebugFlag(int32_t flag, bool rewrite) {
  taosSetDebugFlag(&idxDebugFlag, "idxDebugFlag", flag, rewrite);
  taosSetDebugFlag(&tdbDebugFlag, "tdbDebugFlag", flag, rewrite);
  taosSetDebugFlag(&metaDebugFlag, "metaDebugFlag", flag, rewrite);
-  taosSetDebugFlag(&metaDebugFlag, "tmrDebugFlag", flag, rewrite);
  uInfo("all debug flag are set to %d", flag);
 }
--- a/source/common/src/tmsg.c
+++ b/source/common/src/tmsg.c
@ -5424,6 +5424,7 @@ int32_t tSerializeSCMCreateStreamReq(void *buf, int32_t bufLen, const SCMCreateS
    if (tEncodeI32(&encoder, pField->bytes) < 0) return -1;
    if (tEncodeCStr(&encoder, pField->name) < 0) return -1;
  }
+  if (tEncodeI8(&encoder, pReq->createStb) < 0) return -1;

  tEndEncode(&encoder);

@ -5484,6 +5485,7 @@ int32_t tDeserializeSCMCreateStreamReq(void *buf, int32_t bufLen, SCMCreateStrea
      }
    }
  }
+  if (tDecodeI8(&decoder, &pReq->createStb) < 0) return -1;

  tEndDecode(&decoder);

@ -5627,30 +5629,6 @@ int tDecodeSVCreateStbReq(SDecoder *pCoder, SVCreateStbReq *pReq) {
  return 0;
 }

-STSchema *tdGetSTSChemaFromSSChema(SSchema *pSchema, int32_t nCols, int32_t sver) {
-  STSchemaBuilder schemaBuilder = {0};
-  if (tdInitTSchemaBuilder(&schemaBuilder, sver) < 0) {
-    return NULL;
-  }
-
-  for (int i = 0; i < nCols; i++) {
-    SSchema *schema = pSchema + i;
-    if (tdAddColToSchema(&schemaBuilder, schema->type, schema->flags, schema->colId, schema->bytes) < 0) {
-      tdDestroyTSchemaBuilder(&schemaBuilder);
-      return NULL;
-    }
-  }
-
-  STSchema *pNSchema = tdGetSchemaFromBuilder(&schemaBuilder);
-  if (pNSchema == NULL) {
-    tdDestroyTSchemaBuilder(&schemaBuilder);
-    return NULL;
-  }
-
-  tdDestroyTSchemaBuilder(&schemaBuilder);
-  return pNSchema;
-}
-
 int tEncodeSVCreateTbReq(SEncoder *pCoder, const SVCreateTbReq *pReq) {
  if (tStartEncode(pCoder) < 0) return -1;

@ -5728,6 +5706,25 @@ int tDecodeSVCreateTbReq(SDecoder *pCoder, SVCreateTbReq *pReq) {
  return 0;
 }

+void tDestroySVCreateTbReq(SVCreateTbReq *pReq, int32_t flags) {
+  if (pReq == NULL) return;
+
+  if (flags & TSDB_MSG_FLG_ENCODE) {
+    // TODO
+  } else if (flags & TSDB_MSG_FLG_DECODE) {
+    if (pReq->comment) {
+      pReq->comment = NULL;
+      taosMemoryFree(pReq->comment);
+    }
+
+    if (pReq->type == TSDB_CHILD_TABLE) {
+      if (pReq->ctb.tagName) taosArrayDestroy(pReq->ctb.tagName);
+    } else if (pReq->type == TSDB_NORMAL_TABLE) {
+      if (pReq->ntb.schemaRow.pSchema) taosMemoryFree(pReq->ntb.schemaRow.pSchema);
+    }
+  }
+}
+
 int tEncodeSVCreateTbBatchReq(SEncoder *pCoder, const SVCreateTbBatchReq *pReq) {
  int32_t nReq = taosArrayGetSize(pReq->pArray);

@ -6719,3 +6716,328 @@ int32_t tDecodeSBatchDeleteReq(SDecoder *pDecoder, SBatchDeleteReq *pReq) {
  }
  return 0;
 }
+
+static int32_t tEncodeSSubmitTbData(SEncoder *pCoder, const SSubmitTbData *pSubmitTbData) {
+  if (tStartEncode(pCoder) < 0) return -1;
+
+  if (tEncodeI32v(pCoder, pSubmitTbData->flags) < 0) return -1;
+
+  // auto create table
+  if (pSubmitTbData->flags & SUBMIT_REQ_AUTO_CREATE_TABLE) {
+    ASSERT(pSubmitTbData->pCreateTbReq);
+    if (tEncodeSVCreateTbReq(pCoder, pSubmitTbData->pCreateTbReq) < 0) return -1;
+  }
+
+  // submit data
+  if (tEncodeI64(pCoder, pSubmitTbData->suid) < 0) return -1;
+  if (tEncodeI64(pCoder, pSubmitTbData->uid) < 0) return -1;
+  if (tEncodeI32v(pCoder, pSubmitTbData->sver) < 0) return -1;
+
+  if (pSubmitTbData->flags & SUBMIT_REQ_COLUMN_DATA_FORMAT) {
+    uint64_t  nColData = TARRAY_SIZE(pSubmitTbData->aCol);
+    SColData *aColData = (SColData *)TARRAY_DATA(pSubmitTbData->aCol);
+
+    if (tEncodeU64v(pCoder, nColData) < 0) return -1;
+
+    for (uint64_t i = 0; i < nColData; i++) {
+      pCoder->pos += tPutColData(pCoder->data ? pCoder->data + pCoder->pos : NULL, &aColData[i]);
+    }
+  } else {
+    if (tEncodeU64v(pCoder, TARRAY_SIZE(pSubmitTbData->aRowP)) < 0) return -1;
+
+    SRow **rows = (SRow **)TARRAY_DATA(pSubmitTbData->aRowP);
+    for (int32_t iRow = 0; iRow < TARRAY_SIZE(pSubmitTbData->aRowP); ++iRow) {
+      if (pCoder->data) memcpy(pCoder->data + pCoder->pos, rows[iRow], rows[iRow]->len);
+      pCoder->pos += rows[iRow]->len;
+    }
+  }
+
+  tEndEncode(pCoder);
+  return 0;
+}
+
+static int32_t tDecodeSSubmitTbData(SDecoder *pCoder, SSubmitTbData *pSubmitTbData) {
+  int32_t code = 0;
+
+  if (tStartDecode(pCoder) < 0) {
+    code = TSDB_CODE_INVALID_MSG;
+    goto _exit;
+  }
+
+  if (tDecodeI32v(pCoder, &pSubmitTbData->flags) < 0) return -1;
+
+  if (pSubmitTbData->flags & SUBMIT_REQ_AUTO_CREATE_TABLE) {
+    pSubmitTbData->pCreateTbReq = taosMemoryCalloc(1, sizeof(SVCreateTbReq));
+    if (pSubmitTbData->pCreateTbReq == NULL) {
+      code = TSDB_CODE_OUT_OF_MEMORY;
+      goto _exit;
+    }
+
+    if (tDecodeSVCreateTbReq(pCoder, pSubmitTbData->pCreateTbReq) < 0) {
+      code = TSDB_CODE_INVALID_MSG;
+      goto _exit;
+    }
+  }
+
+  // submit data
+  if (tDecodeI64(pCoder, &pSubmitTbData->suid) < 0) {
+    code = TSDB_CODE_INVALID_MSG;
+    goto _exit;
+  }
+  if (tDecodeI64(pCoder, &pSubmitTbData->uid) < 0) {
+    code = TSDB_CODE_INVALID_MSG;
+    goto _exit;
+  }
+  if (tDecodeI32v(pCoder, &pSubmitTbData->sver) < 0) {
+    code = TSDB_CODE_INVALID_MSG;
+    goto _exit;
+  }
+
+  if (pSubmitTbData->flags & SUBMIT_REQ_COLUMN_DATA_FORMAT) {
+    uint64_t nColData;
+
+    if (tDecodeU64v(pCoder, &nColData) < 0) {
+      code = TSDB_CODE_INVALID_MSG;
+      goto _exit;
+    }
+
+    pSubmitTbData->aCol = taosArrayInit(nColData, sizeof(SColData));
+    if (pSubmitTbData->aCol == NULL) {
+      code = TSDB_CODE_OUT_OF_MEMORY;
+      goto _exit;
+    }
+
+    for (int32_t i = 0; i < nColData; ++i) {
+      pCoder->pos += tGetColData(pCoder->data + pCoder->pos, taosArrayReserve(pSubmitTbData->aCol, 1));
+    }
+  } else {
+    uint64_t nRow;
+    if (tDecodeU64v(pCoder, &nRow) < 0) {
+      code = TSDB_CODE_INVALID_MSG;
+      goto _exit;
+    }
+
+    pSubmitTbData->aRowP = taosArrayInit(nRow, sizeof(SRow *));
+    if (pSubmitTbData->aRowP == NULL) {
+      code = TSDB_CODE_OUT_OF_MEMORY;
+      goto _exit;
+    }
+
+    for (int32_t iRow = 0; iRow < nRow; ++iRow) {
+      SRow **ppRow = taosArrayReserve(pSubmitTbData->aRowP, 1);
+
+      *ppRow = (SRow *)(pCoder->data + pCoder->pos);
+      pCoder->pos += (*ppRow)->len;
+    }
+  }
+
+  tEndDecode(pCoder);
+
+_exit:
+  if (code) {
+    // TODO: clear
+  }
+  return 0;
+}
+
+int32_t tEncodeSSubmitReq2(SEncoder *pCoder, const SSubmitReq2 *pReq) {
+  if (tStartEncode(pCoder) < 0) return -1;
+
+  if (tEncodeU64v(pCoder, taosArrayGetSize(pReq->aSubmitTbData)) < 0) return -1;
+  for (uint64_t i = 0; i < taosArrayGetSize(pReq->aSubmitTbData); i++) {
+    if (tEncodeSSubmitTbData(pCoder, taosArrayGet(pReq->aSubmitTbData, i)) < 0) return -1;
+  }
+
+  tEndEncode(pCoder);
+  return 0;
+}
+
+int32_t tDecodeSSubmitReq2(SDecoder *pCoder, SSubmitReq2 *pReq) {
+  int32_t code = 0;
+
+  memset(pReq, 0, sizeof(*pReq));
+
+  // decode
+  if (tStartDecode(pCoder) < 0) {
+    code = TSDB_CODE_INVALID_MSG;
+    goto _exit;
+  }
+
+  uint64_t nSubmitTbData;
+  if (tDecodeU64v(pCoder, &nSubmitTbData) < 0) {
+    code = TSDB_CODE_INVALID_MSG;
+    goto _exit;
+  }
+
+  pReq->aSubmitTbData = taosArrayInit(nSubmitTbData, sizeof(SSubmitTbData));
+  if (pReq->aSubmitTbData == NULL) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+    goto _exit;
+  }
+
+  for (uint64_t i = 0; i < nSubmitTbData; i++) {
+    if (tDecodeSSubmitTbData(pCoder, taosArrayReserve(pReq->aSubmitTbData, 1)) < 0) {
+      code = TSDB_CODE_INVALID_MSG;
+      goto _exit;
+    }
+  }
+
+  tEndDecode(pCoder);
+
+_exit:
+  if (code) {
+    if (pReq->aSubmitTbData) {
+      // todo
+      taosArrayDestroy(pReq->aSubmitTbData);
+      pReq->aSubmitTbData = NULL;
+    }
+  }
+  return code;
+}
+
+void tDestroySSubmitTbData(SSubmitTbData *pTbData, int32_t flag) {
+  if (NULL == pTbData) {
+    return;
+  }
+
+  if (flag == TSDB_MSG_FLG_ENCODE) {
+    if (pTbData->pCreateTbReq) {
+      tdDestroySVCreateTbReq(pTbData->pCreateTbReq);
+      taosMemoryFree(pTbData->pCreateTbReq);
+    }
+
+    if (pTbData->flags & SUBMIT_REQ_COLUMN_DATA_FORMAT) {
+      int32_t   nColData = TARRAY_SIZE(pTbData->aCol);
+      SColData *aColData = (SColData *)TARRAY_DATA(pTbData->aCol);
+
+      for (int32_t i = 0; i < nColData; ++i) {
+        tColDataDestroy(&aColData[i]);
+      }
+      taosArrayDestroy(pTbData->aCol);
+    } else {
+      int32_t nRow = TARRAY_SIZE(pTbData->aRowP);
+      SRow  **rows = (SRow **)TARRAY_DATA(pTbData->aRowP);
+
+      for (int32_t i = 0; i < nRow; ++i) {
+        tRowDestroy(rows[i]);
+      }
+      taosArrayDestroy(pTbData->aRowP);
+    }
+  } else if (flag == TSDB_MSG_FLG_DECODE) {
+    if (pTbData->pCreateTbReq) {
+      tDestroySVCreateTbReq(pTbData->pCreateTbReq, TSDB_MSG_FLG_DECODE);
+      taosMemoryFree(pTbData->pCreateTbReq);
+    }
+
+    if (pTbData->flags & SUBMIT_REQ_COLUMN_DATA_FORMAT) {
+      taosArrayDestroy(pTbData->aCol);
+    } else {
+      taosArrayDestroy(pTbData->aRowP);
+    }
+  }
+}
+
+void tDestroySSubmitReq2(SSubmitReq2 *pReq, int32_t flag) {
+  if (pReq->aSubmitTbData == NULL) return;
+
+  int32_t        nSubmitTbData = TARRAY_SIZE(pReq->aSubmitTbData);
+  SSubmitTbData *aSubmitTbData = (SSubmitTbData *)TARRAY_DATA(pReq->aSubmitTbData);
+
+  for (int32_t i = 0; i < nSubmitTbData; i++) {
+    tDestroySSubmitTbData(&aSubmitTbData[i], flag);
+  }
+  taosArrayDestroy(pReq->aSubmitTbData);
+}
+
+int32_t tEncodeSSubmitRsp2(SEncoder *pCoder, const SSubmitRsp2 *pRsp) {
+  if (tStartEncode(pCoder) < 0) return -1;
+
+  if (tEncodeI32v(pCoder, pRsp->affectedRows) < 0) return -1;
+
+  if (tEncodeU64v(pCoder, taosArrayGetSize(pRsp->aCreateTbRsp)) < 0) return -1;
+  for (int32_t i = 0; i < taosArrayGetSize(pRsp->aCreateTbRsp); ++i) {
+    if (tEncodeSVCreateTbRsp(pCoder, taosArrayGet(pRsp->aCreateTbRsp, i)) < 0) return -1;
+  }
+
+  tEndEncode(pCoder);
+  return 0;
+}
+
+int32_t tDecodeSSubmitRsp2(SDecoder *pCoder, SSubmitRsp2 *pRsp) {
+  int32_t code = 0;
+
+  memset(pRsp, 0, sizeof(SSubmitRsp2));
+
+  // decode
+  if (tStartDecode(pCoder) < 0) {
+    code = TSDB_CODE_INVALID_MSG;
+    goto _exit;
+  }
+
+  if (tDecodeI32v(pCoder, &pRsp->affectedRows) < 0) {
+    code = TSDB_CODE_INVALID_MSG;
+    goto _exit;
+  }
+
+  uint64_t nCreateTbRsp;
+  if (tDecodeU64v(pCoder, &nCreateTbRsp) < 0) {
+    code = TSDB_CODE_INVALID_MSG;
+    goto _exit;
+  }
+
+  if (nCreateTbRsp) {
+    pRsp->aCreateTbRsp = taosArrayInit(nCreateTbRsp, sizeof(SVCreateTbRsp));
+    if (pRsp->aCreateTbRsp == NULL) {
+      code = TSDB_CODE_OUT_OF_MEMORY;
+      goto _exit;
+    }
+
+    for (int32_t i = 0; i < nCreateTbRsp; ++i) {
+      SVCreateTbRsp *pCreateTbRsp = taosArrayReserve(pRsp->aCreateTbRsp, 1);
+      if (tDecodeSVCreateTbRsp(pCoder, pCreateTbRsp) < 0) {
+        code = TSDB_CODE_INVALID_MSG;
+        goto _exit;
+      }
+    }
+  }
+
+  tEndDecode(pCoder);
+
+_exit:
+  if (code) {
+    if (pRsp->aCreateTbRsp) {
+      taosArrayDestroyEx(pRsp->aCreateTbRsp, NULL /* todo */);
+    }
+  }
+  return code;
+}
+
+void tDestroySSubmitRsp2(SSubmitRsp2 *pRsp, int32_t flag) {
+  if (NULL == pRsp) {
+    return;
+  }
+
+  if (flag & TSDB_MSG_FLG_ENCODE) {
+    if (pRsp->aCreateTbRsp) {
+      int32_t        nCreateTbRsp = TARRAY_SIZE(pRsp->aCreateTbRsp);
+      SVCreateTbRsp *aCreateTbRsp = TARRAY_DATA(pRsp->aCreateTbRsp);
+      for (int32_t i = 0; i < nCreateTbRsp; ++i) {
+        if (aCreateTbRsp[i].pMeta) {
+          taosMemoryFree(aCreateTbRsp[i].pMeta);
+        }
+      }
+      taosArrayDestroy(pRsp->aCreateTbRsp);
+    }
+  } else if (flag & TSDB_MSG_FLG_DECODE) {
+    if (pRsp->aCreateTbRsp) {
+      int32_t        nCreateTbRsp = TARRAY_SIZE(pRsp->aCreateTbRsp);
+      SVCreateTbRsp *aCreateTbRsp = TARRAY_DATA(pRsp->aCreateTbRsp);
+      for (int32_t i = 0; i < nCreateTbRsp; ++i) {
+        if (aCreateTbRsp[i].pMeta) {
+          taosMemoryFree(aCreateTbRsp[i].pMeta);
+        }
+      }
+      taosArrayDestroy(pRsp->aCreateTbRsp);
+    }
+  }
+}
--- a/source/common/src/tname.c
+++ b/source/common/src/tname.c
@ -298,8 +298,8 @@ int32_t tNameFromString(SName* dst, const char* str, uint32_t type) {
 }

 static int compareKv(const void* p1, const void* p2) {
-  SSmlKv* kv1 = *(SSmlKv**)p1;
-  SSmlKv* kv2 = *(SSmlKv**)p2;
+  SSmlKv* kv1 = (SSmlKv*)p1;
+  SSmlKv* kv2 = (SSmlKv*)p2;
  int32_t kvLen1 = kv1->keyLen;
  int32_t kvLen2 = kv2->keyLen;
  int32_t res = strncasecmp(kv1->key, kv2->key, TMIN(kvLen1, kvLen2));
@ -320,7 +320,7 @@ void buildChildTableName(RandTableName* rName) {
  taosArraySort(rName->tags, compareKv);
  for (int j = 0; j < taosArrayGetSize(rName->tags); ++j) {
    taosStringBuilderAppendChar(&sb, ',');
-    SSmlKv* tagKv = taosArrayGetP(rName->tags, j);
+    SSmlKv* tagKv = taosArrayGet(rName->tags, j);
    taosStringBuilderAppendStringLen(&sb, tagKv->key, tagKv->keyLen);
    taosStringBuilderAppendChar(&sb, '=');
    if (IS_VAR_DATA_TYPE(tagKv->type)) {
--- a/source/common/test/dataformatTest.cpp
+++ b/source/common/test/dataformatTest.cpp
@ -117,7 +117,7 @@ STSchema *genSTSchema(int16_t nCols) {
  }

  STSchema *pResult = NULL;
-  pResult = tdGetSTSChemaFromSSChema(pSchema, nCols, 1);
+  pResult = tBuildTSchema(pSchema, nCols, 1);

  taosMemoryFree(pSchema);
  return pResult;
--- a/source/dnode/mgmt/exe/dmMain.c
+++ b/source/dnode/mgmt/exe/dmMain.c
@ -268,6 +268,11 @@ int mainWindows(int argc, char **argv) {

  if (dmInit() != 0) {
    dError("failed to init dnode since %s", terrstr());
+
+    taosCleanupCfg();
+    taosCloseLog();
+    taosCleanupArgs();
+    taosConvDestroy();
    return -1;
  }

--- a/source/dnode/mgmt/mgmt_vnode/inc/vmInt.h
+++ b/source/dnode/mgmt/mgmt_vnode/inc/vmInt.h
@ -26,20 +26,20 @@ extern "C" {
 #endif

 typedef struct SVnodeMgmt {
-  SDnodeData    *pData;
-  SMsgCb         msgCb;
-  const char    *path;
-  const char    *name;
-  SQWorkerPool   queryPool;
-  SQWorkerPool   streamPool;
-  SWWorkerPool   fetchPool;
-  SSingleWorker  mgmtWorker;
-  SHashObj      *hash;
-  TdThreadRwlock lock;
-  SVnodesStat    state;
-  STfs          *pTfs;
-  TdThread       thread;
-  bool           stop;
+  SDnodeData      *pData;
+  SMsgCb           msgCb;
+  const char      *path;
+  const char      *name;
+  SQWorkerPool     queryPool;
+  SAutoQWorkerPool streamPool;
+  SWWorkerPool     fetchPool;
+  SSingleWorker    mgmtWorker;
+  SHashObj        *hash;
+  TdThreadRwlock   lock;
+  SVnodesStat      state;
+  STfs            *pTfs;
+  TdThread         thread;
+  bool             stop;
 } SVnodeMgmt;

 typedef struct {
--- a/source/dnode/mgmt/mgmt_vnode/src/vmWorker.c
+++ b/source/dnode/mgmt/mgmt_vnode/src/vmWorker.c
@ -140,6 +140,12 @@ static void vmProcessSyncQueue(SQueueInfo *pInfo, STaosQall *qall, int32_t numOf

 static int32_t vmPutMsgToQueue(SVnodeMgmt *pMgmt, SRpcMsg *pMsg, EQueueType qtype) {
  const STraceId *trace = &pMsg->info.traceId;
+  if (pMsg->contLen < sizeof(SMsgHead)) {
+    dGError("invalid rpc msg with no msg head at pCont. pMsg:%p, type:%s, contLen:%d", pMsg, TMSG_INFO(pMsg->msgType),
+            pMsg->contLen);
+    return -1;
+  }
+
  SMsgHead       *pHead = pMsg->pCont;
  int32_t         code = 0;

@ -233,6 +239,14 @@ int32_t vmPutMsgToMgmtQueue(SVnodeMgmt *pMgmt, SRpcMsg *pMsg) {
 }

 int32_t vmPutRpcMsgToQueue(SVnodeMgmt *pMgmt, EQueueType qtype, SRpcMsg *pRpc) {
+  if (pRpc->contLen < sizeof(SMsgHead)) {
+    dError("invalid rpc msg with no msg head at pCont. pRpc:%p, type:%s, len:%d", pRpc, TMSG_INFO(pRpc->msgType),
+           pRpc->contLen);
+    rpcFreeCont(pRpc->pCont);
+    pRpc->pCont = NULL;
+    return -1;
+  }
+
  SRpcMsg *pMsg = taosAllocateQitem(sizeof(SRpcMsg), RPC_QITEM, pRpc->contLen);
  if (pMsg == NULL) {
    rpcFreeCont(pRpc->pCont);
@ -304,7 +318,7 @@ int32_t vmAllocQueue(SVnodeMgmt *pMgmt, SVnodeObj *pVnode) {
  (void)tMultiWorkerInit(&pVnode->pApplyW, &acfg);

  pVnode->pQueryQ = tQWorkerAllocQueue(&pMgmt->queryPool, pVnode, (FItem)vmProcessQueryQueue);
-  pVnode->pStreamQ = tQWorkerAllocQueue(&pMgmt->streamPool, pVnode, (FItem)vmProcessStreamQueue);
+  pVnode->pStreamQ = tAutoQWorkerAllocQueue(&pMgmt->streamPool, pVnode, (FItem)vmProcessStreamQueue);
  pVnode->pFetchQ = tWWorkerAllocQueue(&pMgmt->fetchPool, pVnode, (FItems)vmProcessFetchQueue);

  if (pVnode->pWriteW.queue == NULL || pVnode->pSyncW.queue == NULL || pVnode->pSyncCtrlW.queue == NULL ||
@ -330,7 +344,7 @@ int32_t vmAllocQueue(SVnodeMgmt *pMgmt, SVnodeObj *pVnode) {

 void vmFreeQueue(SVnodeMgmt *pMgmt, SVnodeObj *pVnode) {
  tQWorkerFreeQueue(&pMgmt->queryPool, pVnode->pQueryQ);
-  tQWorkerFreeQueue(&pMgmt->streamPool, pVnode->pStreamQ);
+  tAutoQWorkerFreeQueue(&pMgmt->streamPool, pVnode->pStreamQ);
  tWWorkerFreeQueue(&pMgmt->fetchPool, pVnode->pFetchQ);
  pVnode->pQueryQ = NULL;
  pVnode->pStreamQ = NULL;
@ -345,11 +359,10 @@ int32_t vmStartWorker(SVnodeMgmt *pMgmt) {
  pQPool->max = tsNumOfVnodeQueryThreads;
  if (tQWorkerInit(pQPool) != 0) return -1;

-  SQWorkerPool *pStreamPool = &pMgmt->streamPool;
+  SAutoQWorkerPool *pStreamPool = &pMgmt->streamPool;
  pStreamPool->name = "vnode-stream";
-  pStreamPool->min = tsNumOfVnodeStreamThreads;
-  pStreamPool->max = tsNumOfVnodeStreamThreads;
-  if (tQWorkerInit(pStreamPool) != 0) return -1;
+  pStreamPool->ratio = tsRatioOfVnodeStreamThreads;
+  if (tAutoQWorkerInit(pStreamPool) != 0) return -1;

  SWWorkerPool *pFPool = &pMgmt->fetchPool;
  pFPool->name = "vnode-fetch";
@ -371,7 +384,7 @@ int32_t vmStartWorker(SVnodeMgmt *pMgmt) {

 void vmStopWorker(SVnodeMgmt *pMgmt) {
  tQWorkerCleanup(&pMgmt->queryPool);
-  tQWorkerCleanup(&pMgmt->streamPool);
+  tAutoQWorkerCleanup(&pMgmt->streamPool);
  tWWorkerCleanup(&pMgmt->fetchPool);
  dDebug("vnode workers are closed");
 }
--- a/source/dnode/mgmt/node_mgmt/src/dmTransport.c
+++ b/source/dnode/mgmt/node_mgmt/src/dmTransport.c
@ -111,12 +111,12 @@ static void dmProcessRpcMsg(SDnode *pDnode, SRpcMsg *pRpc, SEpSet *pEpSet) {
    dGError("msg:%p, type:%s pCont is NULL", pRpc, TMSG_INFO(pRpc->msgType));
    terrno = TSDB_CODE_INVALID_MSG_LEN;
    goto _OVER;
-  } /* else if ((pRpc->code == TSDB_CODE_RPC_NETWORK_UNAVAIL || pRpc->code == TSDB_CODE_RPC_BROKEN_LINK) &&
-              (!IsReq(pRpc)) && (pRpc->pCont == NULL)) {
-     dGError("msg:%p, type:%s pCont is NULL, err: %s", pRpc, TMSG_INFO(pRpc->msgType), tstrerror(pRpc->code));
-     terrno = pRpc->code;
-     goto _OVER;
-   }*/
+  } else if ((pRpc->code == TSDB_CODE_RPC_NETWORK_UNAVAIL || pRpc->code == TSDB_CODE_RPC_BROKEN_LINK) &&
+             (!IsReq(pRpc)) && (pRpc->pCont == NULL)) {
+    dGError("msg:%p, type:%s pCont is NULL, err: %s", pRpc, TMSG_INFO(pRpc->msgType), tstrerror(pRpc->code));
+    terrno = pRpc->code;
+    goto _OVER;
+  }

  if (pHandle->defaultNtype == NODE_END) {
    dGError("msg:%p, type:%s not processed since no handle", pRpc, TMSG_INFO(pRpc->msgType));
@ -248,8 +248,9 @@ static inline void dmReleaseHandle(SRpcHandleInfo *pHandle, int8_t type) { rpcRe

 static bool rpcRfp(int32_t code, tmsg_t msgType) {
  if (code == TSDB_CODE_RPC_NETWORK_UNAVAIL || code == TSDB_CODE_RPC_BROKEN_LINK || code == TSDB_CODE_MNODE_NOT_FOUND ||
-      code == TSDB_CODE_SYN_NOT_LEADER || code == TSDB_CODE_SYN_RESTORING || code == TSDB_CODE_VND_STOPPED ||
-      code == TSDB_CODE_APP_IS_STARTING || code == TSDB_CODE_APP_IS_STOPPING) {
+      code == TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED || code == TSDB_CODE_SYN_NOT_LEADER ||
+      code == TSDB_CODE_SYN_RESTORING || code == TSDB_CODE_VND_STOPPED || code == TSDB_CODE_APP_IS_STARTING ||
+      code == TSDB_CODE_APP_IS_STOPPING) {
    if (msgType == TDMT_SCH_QUERY || msgType == TDMT_SCH_MERGE_QUERY || msgType == TDMT_SCH_FETCH ||
        msgType == TDMT_SCH_MERGE_FETCH) {
      return false;
--- a/source/dnode/mnode/impl/inc/mndDef.h
+++ b/source/dnode/mnode/impl/inc/mndDef.h
@ -193,6 +193,7 @@ typedef struct {
  int64_t    lastAccessTime;
  int32_t    accessTimes;
  int32_t    numOfVnodes;
+  int32_t    numOfOtherNodes;
  int32_t    numOfSupportVnodes;
  float      numOfCores;
  int64_t    memTotal;
--- a/source/dnode/mnode/impl/src/mndConsumer.c
+++ b/source/dnode/mnode/impl/src/mndConsumer.c
@ -558,6 +558,10 @@ static int32_t mndProcessSubscribeReq(SRpcMsg *pMsg) {
      goto SUBSCRIBE_OVER;
    }

+    if (mndCheckTopicPrivilege(pMnode, pMsg->info.conn.user, MND_OPER_SUBSCRIBE, pTopic) != 0) {
+      goto SUBSCRIBE_OVER;
+    }
+
    mndReleaseTopic(pMnode, pTopic);
  }

--- a/source/dnode/mnode/impl/src/mndDnode.c
+++ b/source/dnode/mnode/impl/src/mndDnode.c
@ -397,8 +397,6 @@ static int32_t mndProcessStatusReq(SRpcMsg *pReq) {
  bool    reboot = (pDnode->rebootTime != statusReq.rebootTime);
  bool    needCheck = !online || dnodeChanged || reboot;

-  pDnode->accessTimes++;
-  pDnode->lastAccessTime = curMs;
  const STraceId *trace = &pReq->info.traceId;
  mGTrace("dnode:%d, status received, accessTimes:%d check:%d online:%d reboot:%d changed:%d statusSeq:%d", pDnode->id,
          pDnode->accessTimes, needCheck, online, reboot, dnodeChanged, statusReq.statusSeq);
@ -534,6 +532,8 @@ static int32_t mndProcessStatusReq(SRpcMsg *pReq) {
    pReq->info.rsp = pHead;
  }

+  pDnode->accessTimes++;
+  pDnode->lastAccessTime = curMs;
  code = 0;

 _OVER:
--- a/source/dnode/mnode/impl/src/mndStream.c
+++ b/source/dnode/mnode/impl/src/mndStream.c
@ -637,7 +637,7 @@ static int32_t mndProcessCreateStreamReq(SRpcMsg *pReq) {
  if (mndTrancCheckConflict(pMnode, pTrans) != 0) goto _OVER;

  // create stb for stream
-  if (mndCreateStbForStream(pMnode, pTrans, &streamObj, pReq->info.conn.user) < 0) {
+  if (createStreamReq.createStb == STREAM_CREATE_STABLE_TRUE && mndCreateStbForStream(pMnode, pTrans, &streamObj, pReq->info.conn.user) < 0) {
    mError("trans:%d, failed to create stb for stream %s since %s", pTrans->id, createStreamReq.name, terrstr());
    mndTransDrop(pTrans);
    goto _OVER;
--- a/source/dnode/mnode/impl/src/mndTopic.c
+++ b/source/dnode/mnode/impl/src/mndTopic.c
@ -696,14 +696,9 @@ static int32_t mndProcessDropTopicReq(SRpcMsg *pReq) {
    sdbRelease(pSdb, pConsumer);
  }

-#if 0
-  if (pTopic->refConsumerCnt != 0) {
-    mndReleaseTopic(pMnode, pTopic);
-    terrno = TSDB_CODE_MND_TOPIC_SUBSCRIBED;
-    mError("topic:%s, failed to drop since %s", dropReq.name, terrstr());
+  if (mndCheckDbPrivilegeByName(pMnode, pReq->info.conn.user, MND_OPER_READ_DB, pTopic->db) != 0) {
    return -1;
  }
-#endif

  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_DB_INSIDE, pReq, "drop-topic");
  if (pTrans == NULL) {
--- a/source/dnode/mnode/impl/src/mndTrans.c
+++ b/source/dnode/mnode/impl/src/mndTrans.c
@ -957,7 +957,7 @@ static void mndTransSendRpcRsp(SMnode *pMnode, STrans *pTrans) {
  for (int32_t i = 0; i < size; ++i) {
    SRpcHandleInfo *pInfo = taosArrayGet(pTrans->pRpcArray, i);
    if (pInfo->handle != NULL) {
-      if (code == TSDB_CODE_RPC_NETWORK_UNAVAIL) {
+      if (code == TSDB_CODE_RPC_NETWORK_UNAVAIL || code == TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED) {
        code = TSDB_CODE_MND_TRANS_NETWORK_UNAVAILL;
      }
      if (code == TSDB_CODE_SYN_TIMEOUT) {
--- a/source/dnode/mnode/impl/src/mndVgroup.c
+++ b/source/dnode/mnode/impl/src/mndVgroup.c
@ -425,6 +425,7 @@ void *mndBuildDropVnodeReq(SMnode *pMnode, SDnodeObj *pDnode, SDbObj *pDb, SVgOb
 static bool mndResetDnodesArrayFp(SMnode *pMnode, void *pObj, void *p1, void *p2, void *p3) {
  SDnodeObj *pDnode = pObj;
  pDnode->numOfVnodes = 0;
+  pDnode->numOfOtherNodes = 0;
  return true;
 }

@ -447,7 +448,7 @@ static bool mndBuildDnodesArrayFp(SMnode *pMnode, void *pObj, void *p1, void *p2
        pDnode->numOfVnodes, pDnode->numOfSupportVnodes, isMnode, online, pDnode->memAvail, pDnode->memUsed);

  if (isMnode) {
-    pDnode->numOfVnodes++;
+    pDnode->numOfOtherNodes++;
  }

  if (online && pDnode->numOfSupportVnodes > 0) {
@ -468,14 +469,25 @@ SArray *mndBuildDnodesArray(SMnode *pMnode, int32_t exceptDnodeId) {

  sdbTraverse(pSdb, SDB_DNODE, mndResetDnodesArrayFp, NULL, NULL, NULL);
  sdbTraverse(pSdb, SDB_DNODE, mndBuildDnodesArrayFp, pArray, &exceptDnodeId, NULL);
+
+  mDebug("build %d dnodes array", (int32_t)taosArrayGetSize(pArray));
+  for (int32_t i = 0; i < (int32_t)taosArrayGetSize(pArray); ++i) {
+    SDnodeObj *pDnode = taosArrayGet(pArray, i);
+    mDebug("dnode:%d, vnodes:%d others:%d", pDnode->id, pDnode->numOfVnodes, pDnode->numOfOtherNodes);
+  }
  return pArray;
 }

 static int32_t mndCompareDnodeId(int32_t *dnode1Id, int32_t *dnode2Id) { return *dnode1Id >= *dnode2Id ? 1 : 0; }

+static float mndGetDnodeScore(SDnodeObj *pDnode, int32_t additionDnodes, float ratio) {
+  float totalDnodes = pDnode->numOfVnodes + (float)pDnode->numOfOtherNodes * ratio + additionDnodes;
+  return totalDnodes / pDnode->numOfSupportVnodes;
+}
+
 static int32_t mndCompareDnodeVnodes(SDnodeObj *pDnode1, SDnodeObj *pDnode2) {
-  float d1Score = (float)pDnode1->numOfVnodes / pDnode1->numOfSupportVnodes;
-  float d2Score = (float)pDnode2->numOfVnodes / pDnode2->numOfSupportVnodes;
+  float d1Score = mndGetDnodeScore(pDnode1, 0, 0.9);
+  float d2Score = mndGetDnodeScore(pDnode2, 0, 0.9);
  return d1Score >= d2Score ? 1 : 0;
 }

@ -494,7 +506,12 @@ static int32_t mndGetAvailableDnode(SMnode *pMnode, SDbObj *pDb, SVgObj *pVgroup
  int32_t allocedVnodes = 0;
  void   *pIter = NULL;

+  mDebug("start to sort %d dnodes", (int32_t)taosArrayGetSize(pArray));
  taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);
+  for (int32_t i = 0; i < (int32_t)taosArrayGetSize(pArray); ++i) {
+    SDnodeObj *pDnode = taosArrayGet(pArray, i);
+    mDebug("dnode:%d, score:%f", pDnode->id, mndGetDnodeScore(pDnode, 0, 0.9));
+  }

  int32_t size = taosArrayGetSize(pArray);
  if (size < pVgroup->replica) {
@ -875,7 +892,7 @@ static int32_t mndAddVnodeToVgroup(SMnode *pMnode, STrans *pTrans, SVgObj *pVgro
  taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);
  for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {
    SDnodeObj *pDnode = taosArrayGet(pArray, i);
-    mInfo("dnode:%d, equivalent vnodes:%d", pDnode->id, pDnode->numOfVnodes);
+    mInfo("dnode:%d, equivalent vnodes:%d others:%d", pDnode->id, pDnode->numOfVnodes, pDnode->numOfOtherNodes);
  }

  SVnodeGid *pVgid = &pVgroup->vnodeGid[pVgroup->replica];
@ -935,7 +952,7 @@ static int32_t mndRemoveVnodeFromVgroup(SMnode *pMnode, STrans *pTrans, SVgObj *
  taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);
  for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {
    SDnodeObj *pDnode = taosArrayGet(pArray, i);
-    mInfo("dnode:%d, equivalent vnodes:%d", pDnode->id, pDnode->numOfVnodes);
+    mInfo("dnode:%d, equivalent vnodes:%d others:%d", pDnode->id, pDnode->numOfVnodes, pDnode->numOfOtherNodes);
  }

  int32_t code = -1;
@ -1970,16 +1987,16 @@ static int32_t mndBalanceVgroup(SMnode *pMnode, SRpcMsg *pReq, SArray *pArray) {
    taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);
    for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {
      SDnodeObj *pDnode = taosArrayGet(pArray, i);
-      mInfo("dnode:%d, equivalent vnodes:%d support:%d, score:%f", pDnode->id, pDnode->numOfVnodes,
-            pDnode->numOfSupportVnodes, (float)pDnode->numOfVnodes / pDnode->numOfSupportVnodes);
+      mInfo("dnode:%d, equivalent vnodes:%d others:%d support:%d, score:%f", pDnode->id, pDnode->numOfVnodes,
+            pDnode->numOfSupportVnodes, pDnode->numOfOtherNodes, mndGetDnodeScore(pDnode, 0, 1));
    }

    SDnodeObj *pSrc = taosArrayGet(pArray, taosArrayGetSize(pArray) - 1);
    SDnodeObj *pDst = taosArrayGet(pArray, 0);

-    float srcScore = (float)(pSrc->numOfVnodes - 1) / pSrc->numOfSupportVnodes;
-    float dstScore = (float)(pDst->numOfVnodes + 1) / pDst->numOfSupportVnodes;
-    mInfo("trans:%d, after balance, src dnode:%d score:%f, dst dnode:%d score:%f", pTrans->id, pSrc->id, srcScore,
+    float srcScore = mndGetDnodeScore(pSrc, -1, 1);
+    float dstScore = mndGetDnodeScore(pDst, 1, 1);
+    mInfo("trans:%d, after balance, src dnode:%d score:%f, dst dnode:%d score:%f", pTrans->id, pSrc->id, dstScore,
          pDst->id, dstScore);

    if (srcScore > dstScore - 0.000001) {
--- a/source/dnode/vnode/inc/vnode.h
+++ b/source/dnode/vnode/inc/vnode.h
@ -222,11 +222,19 @@ typedef struct SSnapContext {
 } SSnapContext;

 typedef struct STqReader {
-  int64_t           ver;
-  const SSubmitReq *pMsg;
-  SSubmitBlk       *pBlock;
-  SSubmitMsgIter    msgIter;
-  SSubmitBlkIter    blkIter;
+  // const SSubmitReq *pMsg;
+  //   SSubmitBlk       *pBlock;
+  //   SSubmitMsgIter    msgIter;
+  //   SSubmitBlkIter    blkIter;
+
+  int64_t     ver;
+  SPackedData msg2;
+
+  int8_t      setMsg;
+  SSubmitReq2 submit;
+  int32_t     nextBlk;
+
+  int64_t lastBlkUid;

  SWalReader *pWalReader;

@ -251,11 +259,14 @@ int32_t tqReaderRemoveTbUidList(STqReader *pReader, const SArray *tbUidList);
 int32_t tqSeekVer(STqReader *pReader, int64_t ver);
 int32_t tqNextBlock(STqReader *pReader, SFetchRet *ret);

-int32_t tqReaderSetDataMsg(STqReader *pReader, const SSubmitReq *pMsg, int64_t ver);
-bool    tqNextDataBlock(STqReader *pReader);
-bool    tqNextDataBlockFilterOut(STqReader *pReader, SHashObj *filterOutUids);
-int32_t tqRetrieveDataBlock(SSDataBlock *pBlock, STqReader *pReader);
-int32_t tqRetrieveTaosxBlock(STqReader *pReader, SArray *blocks, SArray *schemas);
+int32_t tqReaderSetSubmitReq2(STqReader *pReader, void *msgStr, int32_t msgLen, int64_t ver);
+// int32_t tqReaderSetDataMsg(STqReader *pReader, const SSubmitReq *pMsg, int64_t ver);
+bool    tqNextDataBlock2(STqReader *pReader);
+bool    tqNextDataBlockFilterOut2(STqReader *pReader, SHashObj *filterOutUids);
+int32_t tqRetrieveDataBlock2(SSDataBlock *pBlock, STqReader *pReader);
+int32_t tqRetrieveTaosxBlock2(STqReader *pReader, SArray *blocks, SArray *schemas);
+// int32_t tqRetrieveDataBlock(SSDataBlock *pBlock, STqReader *pReader);
+// int32_t tqRetrieveTaosxBlock(STqReader *pReader, SArray *blocks, SArray *schemas);

 int32_t vnodeEnqueueStreamMsg(SVnode *pVnode, SRpcMsg *pMsg);

--- a/source/dnode/vnode/src/inc/tq.h
+++ b/source/dnode/vnode/src/inc/tq.h
@ -154,7 +154,8 @@ int32_t tqScanData(STQ* pTq, const STqHandle* pHandle, SMqDataRsp* pRsp, STqOffs
 int64_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, SWalCkHead** pHeadWithCkSum);

 // tqExec
-int32_t tqTaosxScanLog(STQ* pTq, STqHandle* pHandle, SSubmitReq* pReq, STaosxRsp* pRsp);
+int32_t tqTaosxScanLog(STQ* pTq, STqHandle* pHandle, SPackedData submit, STaosxRsp* pRsp);
+// int32_t tqTaosxScanLog(STQ* pTq, STqHandle* pHandle, SSubmitReq* pReq, STaosxRsp* pRsp);
 int32_t tqAddBlockDataToRsp(const SSDataBlock* pBlock, SMqDataRsp* pRsp, int32_t numOfCols, int8_t precision);
 int32_t tqSendDataRsp(STQ* pTq, const SRpcMsg* pMsg, const SMqPollReq* pReq, const SMqDataRsp* pRsp);
 int32_t tqPushDataRsp(STQ* pTq, STqPushEntry* pPushEntry);
@ -182,7 +183,8 @@ int32_t         tqOffsetCommitFile(STqOffsetStore* pStore);

 // tqSink
 // void tqSinkToTableMerge(SStreamTask* pTask, void* vnode, int64_t ver, void* data);
-void tqSinkToTablePipeline(SStreamTask* pTask, void* vnode, int64_t ver, void* data);
+// void tqSinkToTablePipeline(SStreamTask* pTask, void* vnode, int64_t ver, void* data);
+void tqSinkToTablePipeline2(SStreamTask* pTask, void* vnode, int64_t ver, void* data);

 // tqOffset
 char*   tqOffsetBuildFName(const char* path, int32_t fVer);
--- a/source/dnode/vnode/src/inc/tsdb.h
+++ b/source/dnode/vnode/src/inc/tsdb.h
@ -70,6 +70,9 @@ typedef struct SDiskData        SDiskData;
 typedef struct SDiskDataBuilder SDiskDataBuilder;
 typedef struct SBlkInfo         SBlkInfo;

+#define TSDBROW_ROW_FMT ((int8_t)0x0)
+#define TSDBROW_COL_FMT ((int8_t)0x1)
+
 #define TSDB_FILE_DLMT     ((uint32_t)0xF00AFA0F)
 #define TSDB_MAX_SUBBLOCKS 8
 #define TSDB_FHDR_SIZE     512
@ -102,26 +105,29 @@ static FORCE_INLINE int64_t tsdbLogicToFileSize(int64_t lSize, int32_t szPage) {

 // tsdbUtil.c ==============================================================================================
 // TSDBROW
-#define TSDBROW_TS(ROW)                       (((ROW)->type == 0) ? (ROW)->pTSRow->ts : (ROW)->pBlockData->aTSKEY[(ROW)->iRow])
-#define TSDBROW_VERSION(ROW)                  (((ROW)->type == 0) ? (ROW)->version : (ROW)->pBlockData->aVersion[(ROW)->iRow])
-#define TSDBROW_SVERSION(ROW)                 TD_ROW_SVER((ROW)->pTSRow)
-#define TSDBROW_KEY(ROW)                      ((TSDBKEY){.version = TSDBROW_VERSION(ROW), .ts = TSDBROW_TS(ROW)})
-#define tsdbRowFromTSRow(VERSION, TSROW)      ((TSDBROW){.type = 0, .version = (VERSION), .pTSRow = (TSROW)})
-#define tsdbRowFromBlockData(BLOCKDATA, IROW) ((TSDBROW){.type = 1, .pBlockData = (BLOCKDATA), .iRow = (IROW)})
-void tsdbRowGetColVal(TSDBROW *pRow, STSchema *pTSchema, int32_t iCol, SColVal *pColVal);
-// int32_t tPutTSDBRow(uint8_t *p, TSDBROW *pRow);
+#define TSDBROW_TS(ROW) (((ROW)->type == TSDBROW_ROW_FMT) ? (ROW)->pTSRow->ts : (ROW)->pBlockData->aTSKEY[(ROW)->iRow])
+#define TSDBROW_VERSION(ROW) \
+  (((ROW)->type == TSDBROW_ROW_FMT) ? (ROW)->version : (ROW)->pBlockData->aVersion[(ROW)->iRow])
+#define TSDBROW_SVERSION(ROW)            ((ROW)->type == TSDBROW_ROW_FMT ? (ROW)->pTSRow->sver : -1)
+#define TSDBROW_KEY(ROW)                 ((TSDBKEY){.version = TSDBROW_VERSION(ROW), .ts = TSDBROW_TS(ROW)})
+#define tsdbRowFromTSRow(VERSION, TSROW) ((TSDBROW){.type = TSDBROW_ROW_FMT, .version = (VERSION), .pTSRow = (TSROW)})
+#define tsdbRowFromBlockData(BLOCKDATA, IROW) \
+  ((TSDBROW){.type = TSDBROW_COL_FMT, .pBlockData = (BLOCKDATA), .iRow = (IROW)})
+
+void    tsdbRowGetColVal(TSDBROW *pRow, STSchema *pTSchema, int32_t iCol, SColVal *pColVal);
 int32_t tsdbRowCmprFn(const void *p1, const void *p2);
 // STSDBRowIter
-void     tsdbRowIterInit(STSDBRowIter *pIter, TSDBROW *pRow, STSchema *pTSchema);
+int32_t  tsdbRowIterOpen(STSDBRowIter *pIter, TSDBROW *pRow, STSchema *pTSchema);
+void     tsdbRowClose(STSDBRowIter *pIter);
 SColVal *tsdbRowIterNext(STSDBRowIter *pIter);
 // SRowMerger
-int32_t tRowMergerInit2(SRowMerger *pMerger, STSchema *pResTSchema, TSDBROW *pRow, STSchema *pTSchema);
-int32_t tRowMergerAdd(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema);
+int32_t tsdbRowMergerInit2(SRowMerger *pMerger, STSchema *pResTSchema, TSDBROW *pRow, STSchema *pTSchema);
+int32_t tsdbRowMergerAdd(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema);

-int32_t tRowMergerInit(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema);
-void    tRowMergerClear(SRowMerger *pMerger);
-int32_t tRowMerge(SRowMerger *pMerger, TSDBROW *pRow);
-int32_t tRowMergerGetRow(SRowMerger *pMerger, STSRow **ppRow);
+int32_t tsdbRowMergerInit(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema);
+void    tsdbRowMergerClear(SRowMerger *pMerger);
+int32_t tsdbRowMerge(SRowMerger *pMerger, TSDBROW *pRow);
+int32_t tsdbRowMergerGetRow(SRowMerger *pMerger, SRow **ppRow);
 // TABLEID
 int32_t tTABLEIDCmprFn(const void *p1, const void *p2);
 // TSDBKEY
@ -146,24 +152,22 @@ int32_t tGetBlockIdx(uint8_t *p, void *ph);
 int32_t tCmprBlockIdx(void const *lhs, void const *rhs);
 int32_t tCmprBlockL(void const *lhs, void const *rhs);
 // SBlockData
-#define tBlockDataFirstRow(PBLOCKDATA) tsdbRowFromBlockData(PBLOCKDATA, 0)
-#define tBlockDataLastRow(PBLOCKDATA)  tsdbRowFromBlockData(PBLOCKDATA, (PBLOCKDATA)->nRow - 1)
-#define tBlockDataFirstKey(PBLOCKDATA) TSDBROW_KEY(&tBlockDataFirstRow(PBLOCKDATA))
-#define tBlockDataLastKey(PBLOCKDATA)  TSDBROW_KEY(&tBlockDataLastRow(PBLOCKDATA))
+#define tBlockDataFirstRow(PBLOCKDATA)             tsdbRowFromBlockData(PBLOCKDATA, 0)
+#define tBlockDataLastRow(PBLOCKDATA)              tsdbRowFromBlockData(PBLOCKDATA, (PBLOCKDATA)->nRow - 1)
+#define tBlockDataFirstKey(PBLOCKDATA)             TSDBROW_KEY(&tBlockDataFirstRow(PBLOCKDATA))
+#define tBlockDataLastKey(PBLOCKDATA)              TSDBROW_KEY(&tBlockDataLastRow(PBLOCKDATA))
+#define tBlockDataGetColDataByIdx(PBLOCKDATA, IDX) (&(PBLOCKDATA)->aColData[IDX])

-int32_t   tBlockDataCreate(SBlockData *pBlockData);
-void      tBlockDataDestroy(SBlockData *pBlockData, int8_t deepClear);
-int32_t   tBlockDataInit(SBlockData *pBlockData, TABLEID *pId, STSchema *pTSchema, int16_t *aCid, int32_t nCid);
-void      tBlockDataReset(SBlockData *pBlockData);
-int32_t   tBlockDataAppendRow(SBlockData *pBlockData, TSDBROW *pRow, STSchema *pTSchema, int64_t uid);
-void      tBlockDataClear(SBlockData *pBlockData);
-SColData *tBlockDataGetColDataByIdx(SBlockData *pBlockData, int32_t idx);
-void      tBlockDataGetColData(SBlockData *pBlockData, int16_t cid, SColData **ppColData);
-int32_t   tBlockDataMerge(SBlockData *pBlockData1, SBlockData *pBlockData2, SBlockData *pBlockData);
-int32_t   tBlockDataAddColData(SBlockData *pBlockData, SColData **ppColData);
-int32_t   tCmprBlockData(SBlockData *pBlockData, int8_t cmprAlg, uint8_t **ppOut, int32_t *szOut, uint8_t *aBuf[],
-                         int32_t aBufN[]);
-int32_t   tDecmprBlockData(uint8_t *pIn, int32_t szIn, SBlockData *pBlockData, uint8_t *aBuf[]);
+int32_t tBlockDataCreate(SBlockData *pBlockData);
+void    tBlockDataDestroy(SBlockData *pBlockData);
+int32_t tBlockDataInit(SBlockData *pBlockData, TABLEID *pId, STSchema *pTSchema, int16_t *aCid, int32_t nCid);
+void    tBlockDataReset(SBlockData *pBlockData);
+int32_t tBlockDataAppendRow(SBlockData *pBlockData, TSDBROW *pRow, STSchema *pTSchema, int64_t uid);
+void    tBlockDataClear(SBlockData *pBlockData);
+void    tBlockDataGetColData(SBlockData *pBlockData, int16_t cid, SColData **ppColData);
+int32_t tCmprBlockData(SBlockData *pBlockData, int8_t cmprAlg, uint8_t **ppOut, int32_t *szOut, uint8_t *aBuf[],
+                       int32_t aBufN[]);
+int32_t tDecmprBlockData(uint8_t *pIn, int32_t szIn, SBlockData *pBlockData, uint8_t *aBuf[]);
 // SDiskDataHdr
 int32_t tPutDiskDataHdr(uint8_t *p, const SDiskDataHdr *pHdr);
 int32_t tGetDiskDataHdr(uint8_t *p, void *ph);
@ -335,10 +339,13 @@ struct SVersionRange {
 typedef struct SMemSkipListNode SMemSkipListNode;
 struct SMemSkipListNode {
  int8_t            level;
+  int8_t            flag;  // TSDBROW_ROW_FMT for row format, TSDBROW_COL_FMT for col format
+  int32_t           iRow;
  int64_t           version;
-  STSRow           *pTSRow;
+  void             *pData;
  SMemSkipListNode *forwards[0];
 };
+
 typedef struct SMemSkipList {
  int64_t           size;
  uint32_t          seed;
@ -376,11 +383,11 @@ struct SMemTable {
 };

 struct TSDBROW {
-  int8_t type;  // 0 for row from tsRow, 1 for row from block data
+  int8_t type;  // TSDBROW_ROW_FMT for row from tsRow, TSDBROW_COL_FMT for row from block data
  union {
    struct {
      int64_t version;
-      STSRow *pTSRow;
+      SRow   *pTSRow;
    };
    struct {
      SBlockData *pBlockData;
@ -466,14 +473,14 @@ struct SSttBlk {
 // (SBlockData){.suid = suid, .uid = 0}: block data for N child tables int .last file
 // (SBlockData){.suid = 0, .uid = uid}: block data for 1 normal table int .last/.data file
 struct SBlockData {
-  int64_t  suid;      // 0 means normal table block data, otherwise child table block data
-  int64_t  uid;       // 0 means block data in .last file, otherwise in .data file
-  int32_t  nRow;      // number of rows
-  int64_t *aUid;      // uids of each row, only exist in block data in .last file (uid == 0)
-  int64_t *aVersion;  // versions of each row
-  TSKEY   *aTSKEY;    // timestamp of each row
-  int32_t  nColData;
-  SArray  *aColData;  // SArray<SColData>
+  int64_t   suid;      // 0 means normal table block data, otherwise child table block data
+  int64_t   uid;       // 0 means block data in .last file, otherwise in .data file
+  int32_t   nRow;      // number of rows
+  int64_t  *aUid;      // uids of each row, only exist in block data in .last file (uid == 0)
+  int64_t  *aVersion;  // versions of each row
+  TSKEY    *aTSKEY;    // timestamp of each row
+  int32_t   nColData;
+  SColData *aColData;
 };

 struct TABLEID {
@ -565,10 +572,14 @@ struct SDFileSet {
 };

 struct STSDBRowIter {
-  TSDBROW  *pRow;
-  STSchema *pTSchema;
-  SColVal   colVal;
-  int32_t   i;
+  TSDBROW *pRow;
+  union {
+    SRowIter *pIter;
+    struct {
+      int32_t iColData;
+      SColVal cv;
+    };
+  };
 };
 struct SRowMerger {
  STSchema *pTSchema;
@ -739,8 +750,8 @@ typedef struct {

 int32_t tsdbOpenCache(STsdb *pTsdb);
 void    tsdbCloseCache(STsdb *pTsdb);
-int32_t tsdbCacheInsertLast(SLRUCache *pCache, tb_uid_t uid, STSRow *row, STsdb *pTsdb);
-int32_t tsdbCacheInsertLastrow(SLRUCache *pCache, STsdb *pTsdb, tb_uid_t uid, STSRow *row, bool dup);
+int32_t tsdbCacheInsertLast(SLRUCache *pCache, tb_uid_t uid, TSDBROW *row, STsdb *pTsdb);
+int32_t tsdbCacheInsertLastrow(SLRUCache *pCache, STsdb *pTsdb, tb_uid_t uid, TSDBROW *row, bool dup);
 int32_t tsdbCacheGetLastH(SLRUCache *pCache, tb_uid_t uid, SCacheRowsReader *pr, LRUHandle **h);
 int32_t tsdbCacheGetLastrowH(SLRUCache *pCache, tb_uid_t uid, SCacheRowsReader *pr, LRUHandle **h);
 int32_t tsdbCacheRelease(SLRUCache *pCache, LRUHandle *h);
@ -752,6 +763,8 @@ int32_t tsdbCacheDelete(SLRUCache *pCache, tb_uid_t uid, TSKEY eKey);
 void   tsdbCacheSetCapacity(SVnode *pVnode, size_t capacity);
 size_t tsdbCacheGetCapacity(SVnode *pVnode);

+// int32_t tsdbCacheLastArray2Row(SArray *pLastArray, STSRow **ppRow, STSchema *pSchema);
+
 // ========== inline functions ==========
 static FORCE_INLINE int32_t tsdbKeyCmprFn(const void *p1, const void *p2) {
  TSDBKEY *pKey1 = (TSDBKEY *)p1;
@ -793,8 +806,13 @@ static FORCE_INLINE TSDBROW *tsdbTbDataIterGet(STbDataIter *pIter) {
  }

  pIter->pRow = &pIter->row;
-  pIter->pRow->version = pIter->pNode->version;
-  pIter->pRow->pTSRow = pIter->pNode->pTSRow;
+  if (pIter->pNode->flag == TSDBROW_ROW_FMT) {
+    pIter->row = tsdbRowFromTSRow(pIter->pNode->version, pIter->pNode->pData);
+  } else if (pIter->pNode->flag == TSDBROW_COL_FMT) {
+    pIter->row = tsdbRowFromBlockData(pIter->pNode->pData, pIter->pNode->iRow);
+  } else {
+    ASSERT(0);
+  }

  return pIter->pRow;
 }
--- a/source/dnode/vnode/src/inc/vnodeInt.h
+++ b/source/dnode/vnode/src/inc/vnodeInt.h
@ -90,6 +90,7 @@ typedef struct SCommitInfo        SCommitInfo;
 #define VND_INFO_FNAME "vnode.json"

 // vnd.h
+
 void* vnodeBufPoolMalloc(SVBufPool* pPool, int size);
 void* vnodeBufPoolMallocAligned(SVBufPool* pPool, int size);
 void  vnodeBufPoolFree(SVBufPool* pPool, void* p);
@ -162,10 +163,9 @@ int32_t tsdbCommit(STsdb* pTsdb, SCommitInfo* pInfo);
 int32_t tsdbFinishCommit(STsdb* pTsdb);
 int32_t tsdbRollbackCommit(STsdb* pTsdb);
 int32_t tsdbDoRetention(STsdb* pTsdb, int64_t now);
-int     tsdbScanAndConvertSubmitMsg(STsdb* pTsdb, SSubmitReq* pMsg);
-int     tsdbInsertData(STsdb* pTsdb, int64_t version, SSubmitReq* pMsg, SSubmitRsp* pRsp);
-int32_t tsdbInsertTableData(STsdb* pTsdb, int64_t version, SSubmitMsgIter* pMsgIter, SSubmitBlk* pBlock,
-                            SSubmitBlkRsp* pRsp);
+int     tsdbScanAndConvertSubmitMsg(STsdb* pTsdb, SSubmitReq2* pMsg);
+int     tsdbInsertData(STsdb* pTsdb, int64_t version, SSubmitReq2* pMsg, SSubmitRsp2* pRsp);
+int32_t tsdbInsertTableData(STsdb* pTsdb, int64_t version, SSubmitTbData* pSubmitTbData, int32_t* affectedRows);
 int32_t tsdbDeleteTableData(STsdb* pTsdb, int64_t version, tb_uid_t suid, tb_uid_t uid, TSKEY sKey, TSKEY eKey);
 int32_t tsdbSetKeepCfg(STsdb* pTsdb, STsdbCfg* pCfg);

@ -190,7 +190,7 @@ int32_t tqProcessTaskDeployReq(STQ* pTq, int64_t version, char* msg, int32_t msg
 int32_t tqProcessTaskDropReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
 int32_t tqProcessStreamTaskCheckReq(STQ* pTq, SRpcMsg* pMsg);
 int32_t tqProcessStreamTaskCheckRsp(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
-int32_t tqProcessSubmitReq(STQ* pTq, SSubmitReq* data, int64_t ver);
+int32_t tqProcessSubmitReq(STQ* pTq, SPackedData submit);
 int32_t tqProcessDelReq(STQ* pTq, void* pReq, int32_t len, int64_t ver);
 int32_t tqProcessTaskRunReq(STQ* pTq, SRpcMsg* pMsg);
 int32_t tqProcessTaskDispatchReq(STQ* pTq, SRpcMsg* pMsg, bool exec);
@ -203,9 +203,9 @@ int32_t tqProcessTaskRecoverFinishReq(STQ* pTq, SRpcMsg* pMsg);
 int32_t tqProcessTaskRecoverFinishRsp(STQ* pTq, SRpcMsg* pMsg);
 int32_t tqCheckLogInWal(STQ* pTq, int64_t version);

-SSubmitReq* tqBlockToSubmit(SVnode* pVnode, const SArray* pBlocks, const STSchema* pSchema,
-                            SSchemaWrapper* pTagSchemaWrapper, bool createTb, int64_t suid, const char* stbFullName,
-                            SBatchDeleteReq* pDeleteReq);
+int32_t tqBlockToSubmit(SVnode* pVnode, const SArray* pBlocks, const STSchema* pSchema,
+                        SSchemaWrapper* pTagSchemaWrapper, bool createTb, int64_t suid, const char* stbFullName,
+                        SBatchDeleteReq* pDeleteReq, void** ppData, int32_t* pLen);

 // sma
 int32_t smaInit();
@ -223,7 +223,7 @@ int32_t tdProcessTSmaCreate(SSma* pSma, int64_t version, const char* msg);
 int32_t tdProcessTSmaInsert(SSma* pSma, int64_t indexUid, const char* msg);

 int32_t tdProcessRSmaCreate(SSma* pSma, SVCreateStbReq* pReq);
-int32_t tdProcessRSmaSubmit(SSma* pSma, void* pMsg, int32_t inputType);
+int32_t tdProcessRSmaSubmit(SSma* pSma, int64_t version, void* pReq, void* pMsg, int32_t len, int32_t inputType);
 int32_t tdProcessRSmaDrop(SSma* pSma, SVDropStbReq* pReq);
 int32_t tdFetchTbUidList(SSma* pSma, STbUidStore** ppStore, tb_uid_t suid, tb_uid_t uid);
 int32_t tdUpdateTbUidList(SSma* pSma, STbUidStore* pUidStore, bool isAdd);
--- a/source/dnode/vnode/src/meta/metaCache.c
+++ b/source/dnode/vnode/src/meta/metaCache.c
@ -57,7 +57,6 @@ struct SMetaCache {
    TdThreadMutex lock;
    SHashObj*  pTableEntry;
    SLRUCache* pUidResCache;
-    uint64_t   keyBuf[3];
  } sTagFilterResCache;
 };

@ -211,7 +210,7 @@ _exit:
 int32_t metaCacheUpsert(SMeta* pMeta, SMetaInfo* pInfo) {
  int32_t code = 0;

-  // ASSERT(metaIsWLocked(pMeta));
+  // meta is wlocked for calling this func.

  // search
  SMetaCache*       pCache = pMeta->pCache;
@ -222,7 +221,10 @@ int32_t metaCacheUpsert(SMeta* pMeta, SMetaInfo* pInfo) {
  }

  if (*ppEntry) {  // update
-    ASSERT(pInfo->suid == (*ppEntry)->info.suid);
+    if (pInfo->suid != (*ppEntry)->info.suid) {
+      metaError("meta/cache: suid should be same as the one in cache.");
+      return TSDB_CODE_FAILED;
+    }
    if (pInfo->version > (*ppEntry)->info.version) {
      (*ppEntry)->info.version = pInfo->version;
      (*ppEntry)->info.skmVer = pInfo->skmVer;
@ -341,7 +343,7 @@ _exit:
 int32_t metaStatsCacheUpsert(SMeta* pMeta, SMetaStbStats* pInfo) {
  int32_t code = 0;

-  // ASSERT(metaIsWLocked(pMeta));
+  // meta is wlocked for calling this func.

  // search
  SMetaCache*          pCache = pMeta->pCache;
@ -429,20 +431,20 @@ int32_t metaGetCachedTableUidList(SMeta* pMeta, tb_uid_t suid, const uint8_t* pK
                                  bool* acquireRes) {
  // generate the composed key for LRU cache
  SLRUCache*     pCache = pMeta->pCache->sTagFilterResCache.pUidResCache;
-  uint64_t*      pBuf = pMeta->pCache->sTagFilterResCache.keyBuf;
  SHashObj*      pTableMap = pMeta->pCache->sTagFilterResCache.pTableEntry;
  TdThreadMutex* pLock = &pMeta->pCache->sTagFilterResCache.lock;

+  uint64_t buf[3] = {0};
  uint32_t times = 0;

  *acquireRes = 0;
-  pBuf[0] = suid;
-  memcpy(&pBuf[1], pKey, keyLen);
+  buf[0] = suid;
+  memcpy(&buf[1], pKey, keyLen);

  taosThreadMutexLock(pLock);

  int32_t    len = keyLen + sizeof(uint64_t);
-  LRUHandle* pHandle = taosLRUCacheLookup(pCache, pBuf, len);
+  LRUHandle* pHandle = taosLRUCacheLookup(pCache, buf, len);
  if (pHandle == NULL) {
    taosThreadMutexUnlock(pLock);
    return TSDB_CODE_SUCCESS;
@ -450,7 +452,11 @@ int32_t metaGetCachedTableUidList(SMeta* pMeta, tb_uid_t suid, const uint8_t* pK

  // do some book mark work after acquiring the filter result from cache
  STagFilterResEntry** pEntry = taosHashGet(pTableMap, &suid, sizeof(uint64_t));
-  ASSERT(pEntry != NULL);
+  if (NULL == pEntry) {
+    metaError("meta/cache: pEntry should not be NULL.");
+    return TSDB_CODE_FAILED;
+  }
+
  *acquireRes = 1;

  const char* p = taosLRUCacheValue(pCache, pHandle);
@ -476,10 +482,10 @@ int32_t metaGetCachedTableUidList(SMeta* pMeta, tb_uid_t suid, const uint8_t* pK

    SListNode* pNode = NULL;
    while ((pNode = tdListNext(&iter)) != NULL) {
-      memcpy(&pBuf[1], pNode->data, keyLen);
+      memcpy(&buf[1], pNode->data, keyLen);

      // check whether it is existed in LRU cache, and remove it from linked list if not.
-      LRUHandle* pRes = taosLRUCacheLookup(pCache, pBuf, len);
+      LRUHandle* pRes = taosLRUCacheLookup(pCache, buf, len);
      if (pRes == NULL) {  // remove the item in the linked list
        taosArrayPush(pInvalidRes, &pNode);
      } else {
@ -495,7 +501,7 @@ int32_t metaGetCachedTableUidList(SMeta* pMeta, tb_uid_t suid, const uint8_t* pK
      taosMemoryFree(*p1);
    }

-    atomic_store_32(&(*pEntry)->qTimes, 0); // reset the query times
+    atomic_store_32(&(*pEntry)->qTimes, 0);  // reset the query times
    taosArrayDestroy(pInvalidRes);

    taosThreadMutexUnlock(pLock);
@ -547,14 +553,17 @@ int32_t metaUidFilterCachePut(SMeta* pMeta, uint64_t suid, const void* pKey, int
    tdListAppend(&(*pEntry)->list, pKey);
  }

-  uint64_t* pBuf = pMeta->pCache->sTagFilterResCache.keyBuf;
-  pBuf[0] = suid;
+  uint64_t buf[3] = {0};
+  buf[0] = suid;

-  memcpy(&pBuf[1], pKey, keyLen);
-  ASSERT(sizeof(uint64_t) + keyLen == 24);
+  memcpy(&buf[1], pKey, keyLen);
+  if (sizeof(uint64_t) + keyLen != 24) {
+    metaError("meta/cache: incorrect keyLen:%" PRId32 " length.", keyLen);
+    return TSDB_CODE_FAILED;
+  }

  // add to cache.
-  taosLRUCacheInsert(pCache, pBuf, sizeof(uint64_t) + keyLen, pPayload, payloadLen, freePayload, NULL,
+  taosLRUCacheInsert(pCache, buf, sizeof(uint64_t) + keyLen, pPayload, payloadLen, freePayload, NULL,
                     TAOS_LRU_PRIORITY_LOW);

  taosThreadMutexUnlock(pLock);
--- a/source/dnode/vnode/src/meta/metaEntry.c
+++ b/source/dnode/vnode/src/meta/metaEntry.c
@ -51,7 +51,9 @@ int metaEncodeEntry(SEncoder *pCoder, const SMetaEntry *pME) {
  } else if (pME->type == TSDB_TSMA_TABLE) {
    if (tEncodeTSma(pCoder, pME->smaEntry.tsma) < 0) return -1;
  } else {
-    ASSERT(0);
+    metaError("meta/entry: invalide table type: %" PRId8 " encode failed.", pME->type);
+
+    return -1;
  }

  tEndEncode(pCoder);
@ -99,7 +101,9 @@ int metaDecodeEntry(SDecoder *pCoder, SMetaEntry *pME) {
    }
    if (tDecodeTSma(pCoder, pME->smaEntry.tsma, true) < 0) return -1;
  } else {
-    ASSERT(0);
+    metaError("meta/entry: invalide table type: %" PRId8 " decode failed.", pME->type);
+
+    return -1;
  }

  tEndDecode(pCoder);
--- a/source/dnode/vnode/src/meta/metaOpen.c
+++ b/source/dnode/vnode/src/meta/metaOpen.c
@ -358,7 +358,10 @@ static int tagIdxKeyCmpr(const void *pKey1, int kLen1, const void *pKey2, int kL
    return -1;
  }

-  ASSERT(pTagIdxKey1->type == pTagIdxKey2->type);
+  if (pTagIdxKey1->type != pTagIdxKey2->type) {
+    metaError("meta/open: incorrect tag idx type.");
+    return TSDB_CODE_FAILED;
+  }

  // check NULL, NULL is always the smallest
  if (pTagIdxKey1->isNull && !pTagIdxKey2->isNull) {
--- a/source/dnode/vnode/src/meta/metaQuery.c
+++ b/source/dnode/vnode/src/meta/metaQuery.c
@ -611,23 +611,14 @@ tb_uid_t metaStbCursorNext(SMStbCursor *pStbCur) {
 }

 STSchema *metaGetTbTSchema(SMeta *pMeta, tb_uid_t uid, int32_t sver, int lock) {
-  // SMetaReader     mr = {0};
  STSchema       *pTSchema = NULL;
  SSchemaWrapper *pSW = NULL;
-  STSchemaBuilder sb = {0};
-  SSchema        *pSchema;
+  SSchema        *pSchema = NULL;

  pSW = metaGetTableSchema(pMeta, uid, sver, lock);
  if (!pSW) return NULL;

-  tdInitTSchemaBuilder(&sb, pSW->version);
-  for (int i = 0; i < pSW->nCols; i++) {
-    pSchema = pSW->pSchema + i;
-    tdAddColToSchema(&sb, pSchema->type, pSchema->flags, pSchema->colId, pSchema->bytes);
-  }
-  pTSchema = tdGetSchemaFromBuilder(&sb);
-
-  tdDestroyTSchemaBuilder(&sb);
+  pTSchema = tBuildTSchema(pSW->pSchema, pSW->nCols, pSW->version);

  taosMemoryFree(pSW->pSchema);
  taosMemoryFree(pSW);
@ -661,7 +652,11 @@ int32_t metaGetTbTSchemaEx(SMeta *pMeta, tb_uid_t suid, tb_uid_t uid, int32_t sv
        goto _exit;
      }

-      ASSERT(c);
+      if (c == 0) {
+        metaError("meta/query: incorrect c: %" PRId32 ".", c);
+        code = TSDB_CODE_FAILED;
+        goto _exit;
+      }

      if (c < 0) {
        tdbTbcMoveToPrev(pSkmDbC);
@ -685,7 +680,11 @@ int32_t metaGetTbTSchemaEx(SMeta *pMeta, tb_uid_t suid, tb_uid_t uid, int32_t sv
    }
  }

-  ASSERT(sver > 0);
+  if (sver <= 0) {
+    metaError("meta/query: incorrect sver: %" PRId32 ".", sver);
+    code = TSDB_CODE_FAILED;
+    goto _exit;
+  }

  skmDbKey.uid = suid ? suid : uid;
  skmDbKey.sver = sver;
@ -708,21 +707,11 @@ int32_t metaGetTbTSchemaEx(SMeta *pMeta, tb_uid_t suid, tb_uid_t uid, int32_t sv
  tdbFree(pData);

  // convert
-  STSchemaBuilder sb = {0};
-
-  tdInitTSchemaBuilder(&sb, pSchemaWrapper->version);
-  for (int i = 0; i < pSchemaWrapper->nCols; i++) {
-    SSchema *pSchema = pSchemaWrapper->pSchema + i;
-    tdAddColToSchema(&sb, pSchema->type, pSchema->flags, pSchema->colId, pSchema->bytes);
-  }
-
-  STSchema *pTSchema = tdGetSchemaFromBuilder(&sb);
+  STSchema *pTSchema = tBuildTSchema(pSchemaWrapper->pSchema, pSchemaWrapper->nCols, pSchemaWrapper->version);
  if (pTSchema == NULL) {
    code = TSDB_CODE_OUT_OF_MEMORY;
  }

-  tdDestroyTSchemaBuilder(&sb);
-
  *ppTSchema = pTSchema;
  taosMemoryFree(pSchemaWrapper->pSchema);

--- a/source/dnode/vnode/src/meta/metaSnapshot.c
+++ b/source/dnode/vnode/src/meta/metaSnapshot.c
@ -100,7 +100,10 @@ int32_t metaSnapRead(SMetaSnapReader* pReader, uint8_t** ppData) {
    break;
  }

-  ASSERT(pData && nData);
+  if (!pData || !nData) {
+    metaError("meta/snap: invalide nData: %" PRId32 " meta snap read failed.", nData);
+    goto _exit;
+  }

  *ppData = taosMemoryMalloc(sizeof(SSnapDataHdr) + nData);
  if (*ppData == NULL) {
@ -113,8 +116,8 @@ int32_t metaSnapRead(SMetaSnapReader* pReader, uint8_t** ppData) {
  pHdr->size = nData;
  memcpy(pHdr->data, pData, nData);

-  metaInfo("vgId:%d, vnode snapshot meta read data, version:%" PRId64 " uid:%" PRId64 " nData:%d",
-           TD_VID(pReader->pMeta->pVnode), key.version, key.uid, nData);
+  metaDebug("vgId:%d, vnode snapshot meta read data, version:%" PRId64 " uid:%" PRId64 " blockLen:%d",
+            TD_VID(pReader->pMeta->pVnode), key.version, key.uid, nData);

 _exit:
  return code;
@ -356,7 +359,11 @@ int32_t buildSnapContext(SMeta* pMeta, int64_t snapVersion, int64_t suid, int8_t
  for (int i = 0; i < taosArrayGetSize(ctx->idList); i++) {
    int64_t* uid = taosArrayGet(ctx->idList, i);
    SIdInfo* idData = (SIdInfo*)taosHashGet(ctx->idVersion, uid, sizeof(int64_t));
-    ASSERT(idData);
+    if (!idData) {
+      metaError("meta/snap: null idData");
+      return TSDB_CODE_FAILED;
+    }
+
    idData->index = i;
    metaDebug("tmqsnap init idVersion uid:%" PRIi64 " version:%" PRIi64 " index:%d", *uid, idData->version,
              idData->index);
@ -473,7 +480,10 @@ int32_t getMetafromSnapShot(SSnapContext* ctx, void** pBuf, int32_t* contLen, in
    int64_t* uidTmp = taosArrayGet(ctx->idList, ctx->index);
    ctx->index++;
    SIdInfo* idInfo = (SIdInfo*)taosHashGet(ctx->idVersion, uidTmp, sizeof(tb_uid_t));
-    ASSERT(idInfo);
+    if (!idInfo) {
+      metaError("meta/snap: null idInfo");
+      return TSDB_CODE_FAILED;
+    }

    *uid = *uidTmp;
    ret = MoveToPosition(ctx, idInfo->version, *uidTmp);
@ -507,7 +517,11 @@ int32_t getMetafromSnapShot(SSnapContext* ctx, void** pBuf, int32_t* contLen, in
             (ctx->subType == TOPIC_SUB_TYPE__TABLE && me.type == TSDB_CHILD_TABLE && me.ctbEntry.suid == ctx->suid)) {
    STableInfoForChildTable* data =
        (STableInfoForChildTable*)taosHashGet(ctx->suidInfo, &me.ctbEntry.suid, sizeof(tb_uid_t));
-    ASSERT(data);
+    if (!data) {
+      metaError("meta/snap: null data");
+      return TSDB_CODE_FAILED;
+    }
+
    SVCreateTbReq req = {0};

    req.type = TSDB_CHILD_TABLE;
@ -528,7 +542,8 @@ int32_t getMetafromSnapShot(SSnapContext* ctx, void** pBuf, int32_t* contLen, in
    } else {
      SArray* pTagVals = NULL;
      if (tTagToValArray((const STag*)p, &pTagVals) != 0) {
-        ASSERT(0);
+        metaError("meta/snap: tag to val array failed.");
+        return TSDB_CODE_FAILED;
      }
      int16_t nCols = taosArrayGetSize(pTagVals);
      for (int j = 0; j < nCols; ++j) {
@ -572,7 +587,8 @@ int32_t getMetafromSnapShot(SSnapContext* ctx, void** pBuf, int32_t* contLen, in
    ret = buildNormalChildTableInfo(&req, pBuf, contLen);
    *type = TDMT_VND_CREATE_TABLE;
  } else {
-    ASSERT(0);
+    metaError("meta/snap: invalid topic sub type: %" PRId8 " get meta from snap failed.", ctx->subType);
+    ret = -1;
  }
  tDecoderClear(&dc);

@ -593,7 +609,10 @@ SMetaTableInfo getUidfromSnapShot(SSnapContext* ctx) {
    int64_t* uidTmp = taosArrayGet(ctx->idList, ctx->index);
    ctx->index++;
    SIdInfo* idInfo = (SIdInfo*)taosHashGet(ctx->idVersion, uidTmp, sizeof(tb_uid_t));
-    ASSERT(idInfo);
+    if (!idInfo) {
+      metaError("meta/snap: null idInfo");
+      return result;
+    }

    int32_t ret = MoveToPosition(ctx, idInfo->version, *uidTmp);
    if (ret != 0) {
--- a/source/dnode/vnode/src/meta/metaTable.c
+++ b/source/dnode/vnode/src/meta/metaTable.c
@ -46,7 +46,7 @@ static void metaGetEntryInfo(const SMetaEntry *pEntry, SMetaInfo *pInfo) {
    pInfo->suid = 0;
    pInfo->skmVer = pEntry->ntbEntry.schemaRow.version;
  } else {
-    ASSERT(0);
+    metaError("meta/table: invalide table type: %" PRId8 " get entry info failed.", pEntry->type);
  }
 }

@ -342,10 +342,18 @@ int metaAlterSTable(SMeta *pMeta, int64_t version, SVCreateStbReq *pReq) {

  tdbTbcOpen(pMeta->pTbDb, &pTbDbc, NULL);
  ret = tdbTbcMoveTo(pTbDbc, &((STbDbKey){.uid = pReq->suid, .version = oversion}), sizeof(STbDbKey), &c);
-  ASSERT(ret == 0 && c == 0);
+  if (!(ret == 0 && c == 0)) {
+    metaError("meta/table: invalide ret: %" PRId32 " or c: %" PRId32 "alter stb failed.", ret, c);
+    return -1;
+  }

  ret = tdbTbcGet(pTbDbc, NULL, NULL, &pData, &nData);
-  ASSERT(ret == 0);
+  if (ret < 0) {
+    tdbTbcClose(pTbDbc);
+
+    terrno = TSDB_CODE_TDB_STB_NOT_EXIST;
+    return -1;
+  }

  oStbEntry.pBuf = taosMemoryMalloc(nData);
  memcpy(oStbEntry.pBuf, pData, nData);
@ -558,7 +566,8 @@ static void metaBuildTtlIdxKey(STtlIdxKey *ttlKey, const SMetaEntry *pME) {
    ctime = pME->ntbEntry.ctime;
    ttlDays = pME->ntbEntry.ttlDays;
  } else {
-    ASSERT(0);
+    metaError("meta/table: invalide table type: %" PRId8 " build ttl idx key failed.", pME->type);
+    return;
  }

  if (ttlDays <= 0) return;
@ -773,7 +782,10 @@ static int metaAlterTableColumn(SMeta *pMeta, int64_t version, SVAlterTbReq *pAl

  tdbTbcOpen(pMeta->pUidIdx, &pUidIdxc, NULL);
  tdbTbcMoveTo(pUidIdxc, &uid, sizeof(uid), &c);
-  ASSERT(c == 0);
+  if (c != 0) {
+    metaError("meta/table: invalide c: %" PRId32 " alt tb column failed.", c);
+    return -1;
+  }

  tdbTbcGet(pUidIdxc, NULL, NULL, &pData, &nData);
  oversion = ((SUidIdxVal *)pData)[0].version;
@ -783,7 +795,11 @@ static int metaAlterTableColumn(SMeta *pMeta, int64_t version, SVAlterTbReq *pAl

  tdbTbcOpen(pMeta->pTbDb, &pTbDbc, NULL);
  tdbTbcMoveTo(pTbDbc, &((STbDbKey){.uid = uid, .version = oversion}), sizeof(STbDbKey), &c);
-  ASSERT(c == 0);
+  if (c != 0) {
+    metaError("meta/table: invalide c: %" PRId32 " alt tb column failed.", c);
+    return -1;
+  }
+
  tdbTbcGet(pTbDbc, NULL, NULL, &pData, &nData);

  // get table entry
@ -792,7 +808,11 @@ static int metaAlterTableColumn(SMeta *pMeta, int64_t version, SVAlterTbReq *pAl
  memcpy(entry.pBuf, pData, nData);
  tDecoderInit(&dc, entry.pBuf, nData);
  ret = metaDecodeEntry(&dc, &entry);
-  ASSERT(ret == 0);
+  if (ret != 0) {
+    tDecoderClear(&dc);
+    metaError("meta/table: invalide ret: %" PRId32 " alt tb column failed.", ret);
+    return -1;
+  }

  if (entry.type != TSDB_NORMAL_TABLE) {
    terrno = TSDB_CODE_VND_INVALID_TABLE_ACTION;
@ -812,7 +832,11 @@ static int metaAlterTableColumn(SMeta *pMeta, int64_t version, SVAlterTbReq *pAl
    if (iCol >= pSchema->nCols) break;
    pColumn = &pSchema->pSchema[iCol];

-    ASSERT(pAlterTbReq->colName);
+    if (NULL == pAlterTbReq->colName) {
+      metaError("meta/table: null pAlterTbReq->colName");
+      return -1;
+    }
+
    if (strcmp(pColumn->name, pAlterTbReq->colName) == 0) break;
    iCol++;
  }
@ -964,7 +988,10 @@ static int metaUpdateTableTagVal(SMeta *pMeta, int64_t version, SVAlterTbReq *pA

  tdbTbcOpen(pMeta->pUidIdx, &pUidIdxc, NULL);
  tdbTbcMoveTo(pUidIdxc, &uid, sizeof(uid), &c);
-  ASSERT(c == 0);
+  if (c != 0) {
+    metaError("meta/table: invalide c: %" PRId32 " update tb tag val failed.", c);
+    return -1;
+  }

  tdbTbcGet(pUidIdxc, NULL, NULL, &pData, &nData);
  oversion = ((SUidIdxVal *)pData)[0].version;
@ -977,7 +1004,11 @@ static int metaUpdateTableTagVal(SMeta *pMeta, int64_t version, SVAlterTbReq *pA
  /* get ctbEntry */
  tdbTbcOpen(pMeta->pTbDb, &pTbDbc, NULL);
  tdbTbcMoveTo(pTbDbc, &((STbDbKey){.uid = uid, .version = oversion}), sizeof(STbDbKey), &c);
-  ASSERT(c == 0);
+  if (c != 0) {
+    metaError("meta/table: invalide c: %" PRId32 " update tb tag val failed.", c);
+    return -1;
+  }
+
  tdbTbcGet(pTbDbc, NULL, NULL, &pData, &nData);

  ctbEntry.pBuf = taosMemoryMalloc(nData);
@ -1075,7 +1106,11 @@ static int metaUpdateTableTagVal(SMeta *pMeta, int64_t version, SVAlterTbReq *pA
    metaUpdateTagIdx(pMeta, &ctbEntry);
  }

-  ASSERT(ctbEntry.ctbEntry.pTags);
+  if (NULL == ctbEntry.ctbEntry.pTags) {
+    metaError("meta/table: null tags, update tag val failed.");
+    goto _err;
+  }
+
  SCtbIdxKey ctbIdxKey = {.suid = ctbEntry.ctbEntry.suid, .uid = uid};
  tdbTbUpsert(pMeta->pCtbIdx, &ctbIdxKey, sizeof(ctbIdxKey), ctbEntry.ctbEntry.pTags,
              ((STag *)(ctbEntry.ctbEntry.pTags))->len, pMeta->txn);
@ -1130,7 +1165,10 @@ static int metaUpdateTableOptions(SMeta *pMeta, int64_t version, SVAlterTbReq *p

  tdbTbcOpen(pMeta->pUidIdx, &pUidIdxc, NULL);
  tdbTbcMoveTo(pUidIdxc, &uid, sizeof(uid), &c);
-  ASSERT(c == 0);
+  if (c != 0) {
+    metaError("meta/table: invalide c: %" PRId32 " update tb options failed.", c);
+    return -1;
+  }

  tdbTbcGet(pUidIdxc, NULL, NULL, &pData, &nData);
  oversion = ((SUidIdxVal *)pData)[0].version;
@ -1140,7 +1178,11 @@ static int metaUpdateTableOptions(SMeta *pMeta, int64_t version, SVAlterTbReq *p

  tdbTbcOpen(pMeta->pTbDb, &pTbDbc, NULL);
  tdbTbcMoveTo(pTbDbc, &((STbDbKey){.uid = uid, .version = oversion}), sizeof(STbDbKey), &c);
-  ASSERT(c == 0);
+  if (c != 0) {
+    metaError("meta/table: invalide c: %" PRId32 " update tb options failed.", c);
+    return -1;
+  }
+
  tdbTbcGet(pTbDbc, NULL, NULL, &pData, &nData);

  // get table entry
@ -1149,7 +1191,11 @@ static int metaUpdateTableOptions(SMeta *pMeta, int64_t version, SVAlterTbReq *p
  memcpy(entry.pBuf, pData, nData);
  tDecoderInit(&dc, entry.pBuf, nData);
  ret = metaDecodeEntry(&dc, &entry);
-  ASSERT(ret == 0);
+  if (ret != 0) {
+    tDecoderClear(&dc);
+    metaError("meta/table: invalide ret: %" PRId32 " alt tb options failed.", ret);
+    return -1;
+  }

  entry.version = version;
  metaWLock(pMeta);
@ -1408,7 +1454,8 @@ static int metaSaveToSkmDb(SMeta *pMeta, const SMetaEntry *pME) {
  } else if (pME->type == TSDB_NORMAL_TABLE) {
    pSW = &pME->ntbEntry.schemaRow;
  } else {
-    ASSERT(0);
+    metaError("meta/table: invalide table type: %" PRId8 " save skm db failed.", pME->type);
+    return TSDB_CODE_FAILED;
  }

  skmDbKey.uid = pME->uid;
--- a/source/dnode/vnode/src/sma/smaRollup.c
+++ b/source/dnode/vnode/src/sma/smaRollup.c
@ -563,8 +563,8 @@ static int32_t tdProcessSubmitReq(STsdb *pTsdb, int64_t version, void *pReq) {
    return TSDB_CODE_FAILED;
  }

-  SSubmitReq *pSubmitReq = (SSubmitReq *)pReq;
-  // TODO: spin lock for race condition
+  SSubmitReq2 *pSubmitReq = (SSubmitReq2 *)pReq;
+  // spin lock for race condition during insert data
  if (tsdbInsertData(pTsdb, version, pSubmitReq, NULL) < 0) {
    return TSDB_CODE_FAILED;
  }
@ -572,29 +572,19 @@ static int32_t tdProcessSubmitReq(STsdb *pTsdb, int64_t version, void *pReq) {
  return TSDB_CODE_SUCCESS;
 }

-static int32_t tdFetchSubmitReqSuids(SSubmitReq *pMsg, STbUidStore *pStore) {
-  SSubmitMsgIter msgIter = {0};
-  SSubmitBlk    *pBlock = NULL;
-  SSubmitBlkIter blkIter = {0};
-  STSRow        *row = NULL;
+static int32_t tdFetchSubmitReqSuids(SSubmitReq2 *pMsg, STbUidStore *pStore) {
+  SArray *pSubmitTbData = pMsg ? pMsg->aSubmitTbData : NULL;
+  int32_t size = taosArrayGetSize(pSubmitTbData);

  terrno = TSDB_CODE_SUCCESS;

-  if (tInitSubmitMsgIter(pMsg, &msgIter) < 0) {
-    return -1;
-  }
-  while (true) {
-    if (tGetSubmitMsgNext(&msgIter, &pBlock) < 0) {
+  for (int32_t i = 0; i < size; ++i) {
+    SSubmitTbData *pData = TARRAY_GET_ELEM(pSubmitTbData, i);
+    if ((terrno = tdUidStorePut(pStore, pData->suid, NULL)) < 0) {
      return -1;
    }
-
-    if (!pBlock) break;
-    tdUidStorePut(pStore, msgIter.suid, NULL);
  }

-  if (terrno != TSDB_CODE_SUCCESS) {
-    return -1;
-  }
  return 0;
 }

@ -673,11 +663,11 @@ static int32_t tdRSmaExecAndSubmitResult(SSma *pSma, qTaskInfo_t taskInfo, SRSma
      smaDebug("result block, uid:%" PRIu64 ", groupid:%" PRIu64 ", rows:%d", output->info.id.uid,
               output->info.id.groupId, output->info.rows);

-      STsdb      *sinkTsdb = (pItem->level == TSDB_RETENTION_L1 ? pSma->pRSmaTsdb[0] : pSma->pRSmaTsdb[1]);
-      SSubmitReq *pReq = NULL;
+      STsdb       *sinkTsdb = (pItem->level == TSDB_RETENTION_L1 ? pSma->pRSmaTsdb[0] : pSma->pRSmaTsdb[1]);
+      SSubmitReq2 *pReq = NULL;

      // TODO: the schema update should be handled later(TD-17965)
-      if (buildSubmitReqFromDataBlock(&pReq, output, pTSchema, SMA_VID(pSma), suid) < 0) {
+      if (buildSubmitReqFromDataBlock(&pReq, output, pTSchema, output->info.id.groupId, SMA_VID(pSma), suid) < 0) {
        smaError("vgId:%d, build submit req for rsma table suid:%" PRIu64 ", uid:%" PRIu64 ", level %" PRIi8
                 " failed since %s",
                 SMA_VID(pSma), suid, output->info.id.groupId, pItem->level, terrstr());
@ -685,19 +675,21 @@ static int32_t tdRSmaExecAndSubmitResult(SSma *pSma, qTaskInfo_t taskInfo, SRSma
      }

      if (pReq && tdProcessSubmitReq(sinkTsdb, output->info.version, pReq) < 0) {
-        taosMemoryFreeClear(pReq);
+        tDestroySSubmitReq2(pReq, TSDB_MSG_FLG_ENCODE);
+        taosMemoryFree(pReq);
        smaError("vgId:%d, process submit req for rsma suid:%" PRIu64 ", uid:%" PRIu64 " level %" PRIi8
                 " failed since %s",
                 SMA_VID(pSma), suid, output->info.id.groupId, pItem->level, terrstr());
        goto _err;
      }

-      smaDebug("vgId:%d, process submit req for rsma suid:%" PRIu64 ",uid:%" PRIu64 ", level %" PRIi8 " ver %" PRIi64
-               " len %" PRIu32,
-               SMA_VID(pSma), suid, output->info.id.groupId, pItem->level, output->info.version,
-               htonl(pReq->header.contLen));
+      smaDebug("vgId:%d, process submit req for rsma suid:%" PRIu64 ",uid:%" PRIu64 ", level %" PRIi8 " ver %" PRIi64,
+               SMA_VID(pSma), suid, output->info.id.groupId, pItem->level, output->info.version);

-      taosMemoryFreeClear(pReq);
+      if (pReq) {
+        tDestroySSubmitReq2(pReq, TSDB_MSG_FLG_ENCODE);
+        taosMemoryFree(pReq);
+      }
    }
  }

@ -715,22 +707,29 @@ _err:
 * @brief Copy msg to rsmaQueueBuffer for batch process
 *
 * @param pSma
+ * @param version
 * @param pMsg
+ * @param len
 * @param inputType
 * @param pInfo
 * @param suid
 * @return int32_t
 */
-static int32_t tdExecuteRSmaImplAsync(SSma *pSma, const void *pMsg, int32_t inputType, SRSmaInfo *pInfo,
-                                      tb_uid_t suid) {
-  const SSubmitReq *pReq = (const SSubmitReq *)pMsg;
+static int32_t tdExecuteRSmaImplAsync(SSma *pSma, int64_t version, const void *pMsg, int32_t len, int32_t inputType,
+                                      SRSmaInfo *pInfo, tb_uid_t suid) {
+  int32_t size = sizeof(int32_t) + sizeof(int64_t) + len;
+  void   *qItem = taosAllocateQitem(size, DEF_QITEM, 0);

-  void *qItem = taosAllocateQitem(pReq->header.contLen, DEF_QITEM, 0);
  if (!qItem) {
    return TSDB_CODE_FAILED;
  }

-  memcpy(qItem, pMsg, pReq->header.contLen);
+  void *pItem = qItem;
+
+  *(int32_t *)pItem = len;
+  pItem = POINTER_SHIFT(pItem, sizeof(int32_t));
+  *(int64_t *)pItem = version;
+  memcpy(POINTER_SHIFT(pItem, sizeof(int64_t)), pMsg, len);

  taosWriteQitem(pInfo->queue, qItem);

@ -997,12 +996,14 @@ static FORCE_INLINE void tdReleaseRSmaInfo(SSma *pSma, SRSmaInfo *pInfo) {
 * @brief async mode
 *
 * @param pSma
+ * @param version
 * @param pMsg
 * @param inputType
 * @param suid
 * @return int32_t
 */
-static int32_t tdExecuteRSmaAsync(SSma *pSma, const void *pMsg, int32_t inputType, tb_uid_t suid) {
+static int32_t tdExecuteRSmaAsync(SSma *pSma, int64_t version, const void *pMsg, int32_t len, int32_t inputType,
+                                  tb_uid_t suid) {
  SRSmaInfo *pRSmaInfo = tdAcquireRSmaInfoBySuid(pSma, suid);
  if (!pRSmaInfo) {
    smaDebug("vgId:%d, execute rsma, no rsma info for suid:%" PRIu64, SMA_VID(pSma), suid);
@ -1010,7 +1011,7 @@ static int32_t tdExecuteRSmaAsync(SSma *pSma, const void *pMsg, int32_t inputTyp
  }

  if (inputType == STREAM_INPUT__DATA_SUBMIT) {
-    if (tdExecuteRSmaImplAsync(pSma, pMsg, inputType, pRSmaInfo, suid) < 0) {
+    if (tdExecuteRSmaImplAsync(pSma, version, pMsg, len, inputType, pRSmaInfo, suid) < 0) {
      tdReleaseRSmaInfo(pSma, pRSmaInfo);
      return TSDB_CODE_FAILED;
    }
@ -1036,7 +1037,7 @@ static int32_t tdExecuteRSmaAsync(SSma *pSma, const void *pMsg, int32_t inputTyp
  return TSDB_CODE_SUCCESS;
 }

-int32_t tdProcessRSmaSubmit(SSma *pSma, void *pMsg, int32_t inputType) {
+int32_t tdProcessRSmaSubmit(SSma *pSma, int64_t version, void *pReq, void *pMsg, int32_t len, int32_t inputType) {
  SSmaEnv *pEnv = SMA_RSMA_ENV(pSma);
  if (!pEnv) {
    // only applicable when rsma env exists
@ -1046,19 +1047,22 @@ int32_t tdProcessRSmaSubmit(SSma *pSma, void *pMsg, int32_t inputType) {
  STbUidStore uidStore = {0};

  if (inputType == STREAM_INPUT__DATA_SUBMIT) {
-    if (tdFetchSubmitReqSuids(pMsg, &uidStore) < 0) {
+    if (tdFetchSubmitReqSuids(pReq, &uidStore) < 0) {
+      smaError("vgId:%d, failed to process rsma submit fetch suid since: %s", SMA_VID(pSma), terrstr());
      goto _err;
    }

    if (uidStore.suid != 0) {
-      if (tdExecuteRSmaAsync(pSma, pMsg, inputType, uidStore.suid) < 0) {
+      if (tdExecuteRSmaAsync(pSma, version, pMsg, len, inputType, uidStore.suid) < 0) {
+        smaError("vgId:%d, failed to process rsma submit exec 1 since: %s", SMA_VID(pSma), terrstr());
        goto _err;
      }

      void *pIter = NULL;
      while ((pIter = taosHashIterate(uidStore.uidHash, pIter))) {
        tb_uid_t *pTbSuid = (tb_uid_t *)taosHashGetKey(pIter, NULL);
-        if (tdExecuteRSmaAsync(pSma, pMsg, inputType, *pTbSuid) < 0) {
+        if (tdExecuteRSmaAsync(pSma, version, pMsg, len, inputType, *pTbSuid) < 0) {
+          smaError("vgId:%d, failed to process rsma submit exec 2 since: %s", SMA_VID(pSma), terrstr());
          goto _err;
        }
      }
@ -1068,7 +1072,6 @@ int32_t tdProcessRSmaSubmit(SSma *pSma, void *pMsg, int32_t inputType) {
  return TSDB_CODE_SUCCESS;
 _err:
  tdUidStoreDestory(&uidStore);
-  smaError("vgId:%d, failed to process rsma submit since: %s", SMA_VID(pSma), terrstr());
  return TSDB_CODE_FAILED;
 }

@ -1428,7 +1431,8 @@ _end:

 static void tdFreeRSmaSubmitItems(SArray *pItems) {
  for (int32_t i = 0; i < taosArrayGetSize(pItems); ++i) {
-    taosFreeQitem(*(void **)taosArrayGet(pItems, i));
+    SPackedData *packData = taosArrayGet(pItems, i);
+    taosFreeQitem(POINTER_SHIFT(packData->msgStr, -sizeof(int32_t) - sizeof(int64_t)));
  }
  taosArrayClear(pItems);
 }
@ -1497,8 +1501,12 @@ static int32_t tdRSmaBatchExec(SSma *pSma, SRSmaInfo *pInfo, STaosQall *qall, SA
    void *msg = NULL;
    taosGetQitem(qall, (void **)&msg);
    if (msg) {
-      if (!taosArrayPush(pSubmitArr, &msg)) {
-        terrno = TSDB_CODE_OUT_OF_MEMORY;
+      SPackedData packData = {.msgLen = *(int32_t *)msg,
+                              .ver = *(int64_t *)POINTER_SHIFT(msg, sizeof(int32_t)),
+                              .msgStr = POINTER_SHIFT(msg, sizeof(int32_t) + sizeof(int64_t))};
+
+      if (!taosArrayPush(pSubmitArr, &packData)) {
+        tdFreeRSmaSubmitItems(pSubmitArr);
        goto _err;
      }
    } else {
--- a/source/dnode/vnode/src/sma/smaTimeRange.c
+++ b/source/dnode/vnode/src/sma/smaTimeRange.c
@ -217,19 +217,18 @@ static int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char
  }

  SBatchDeleteReq deleteReq = {0};
-  SSubmitReq     *pSubmitReq =
-      tqBlockToSubmit(pSma->pVnode, (const SArray *)msg, pTsmaStat->pTSchema, &pTsmaStat->pTSma->schemaTag, true,
-                      pTsmaStat->pTSma->dstTbUid, pTsmaStat->pTSma->dstTbName, &deleteReq);
-  // TODO deleteReq
-  taosArrayDestroy(deleteReq.deleteReqs);
-  
+  void           *pSubmitReq = NULL;
+  int32_t         contLen = 0;

-  if (!pSubmitReq) {
-    smaError("vgId:%d, failed to gen submit blk while tsma insert for smaIndex %" PRIi64 " since %s", SMA_VID(pSma),
+  if (tqBlockToSubmit(pSma->pVnode, (const SArray *)msg, pTsmaStat->pTSchema, &pTsmaStat->pTSma->schemaTag, true,
+                      pTsmaStat->pTSma->dstTbUid, pTsmaStat->pTSma->dstTbName, &deleteReq, &pSubmitReq, &contLen) < 0) {
+    smaError("vgId:%d, failed to gen submit msg while tsma insert for smaIndex %" PRIi64 " since %s", SMA_VID(pSma),
             indexUid, tstrerror(terrno));
    goto _err;
  }
-  
+
+  // TODO deleteReq
+  taosArrayDestroy(deleteReq.deleteReqs);
 #if 0
  if (!strncasecmp("td.tsma.rst.tb", pTsmaStat->pTSma->dstTbName, 14)) {
    terrno = TSDB_CODE_APP_ERROR;
@ -242,7 +241,7 @@ static int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char
  SRpcMsg submitReqMsg = {
      .msgType = TDMT_VND_SUBMIT,
      .pCont = pSubmitReq,
-      .contLen = ntohl(pSubmitReq->length),
+      .contLen = ntohl(contLen),
  };

  if (tmsgPutToQueue(&pSma->pVnode->msgCb, WRITE_QUEUE, &submitReqMsg) < 0) {
--- a/source/dnode/vnode/src/tq/tq.c
+++ b/source/dnode/vnode/src/tq/tq.c
@ -677,9 +677,12 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg) {
              req.epoch, TD_VID(pTq->pVnode), fetchVer, pHead->msgType);

      if (pHead->msgType == TDMT_VND_SUBMIT) {
-        SSubmitReq* pCont = (SSubmitReq*)&pHead->body;
-
-        if (tqTaosxScanLog(pTq, pHandle, pCont, &taosxRsp) < 0) {
+        SPackedData submit = {
+            .msgStr = POINTER_SHIFT(pHead->body, sizeof(SMsgHead)),
+            .msgLen = pHead->bodyLen - sizeof(SMsgHead),
+            .ver = pHead->version,
+        };
+        if (tqTaosxScanLog(pTq, pHandle, submit, &taosxRsp) < 0) {
        }
        if (taosxRsp.blockNum > 0 /* threshold */) {
          tqOffsetResetToLog(&taosxRsp.rspOffset, fetchVer);
@ -731,6 +734,7 @@ int32_t tqProcessDeleteSubReq(STQ* pTq, int64_t version, char* msg, int32_t msgL

  STqHandle* pHandle = taosHashGet(pTq->pHandle, pReq->subKey, strlen(pReq->subKey));
  if (pHandle) {
+    // walCloseRef(pHandle->pWalReader->pWal, pHandle->pRef->refId);
    if (pHandle->pRef) {
      walCloseRef(pTq->pVnode->pWal, pHandle->pRef->refId);
    }
@ -954,14 +958,14 @@ int32_t tqExpandTask(STQ* pTq, SStreamTask* pTask, int64_t ver) {
    pTask->smaSink.smaSink = smaHandleRes;
  } else if (pTask->outputType == TASK_OUTPUT__TABLE) {
    pTask->tbSink.vnode = pTq->pVnode;
-    pTask->tbSink.tbSinkFunc = tqSinkToTablePipeline;
+    pTask->tbSink.tbSinkFunc = tqSinkToTablePipeline2;

    /*A(pTask->tbSink.pSchemaWrapper);*/
    /*A(pTask->tbSink.pSchemaWrapper->pSchema);*/

    pTask->tbSink.pTSchema =
-        tdGetSTSChemaFromSSChema(pTask->tbSink.pSchemaWrapper->pSchema, pTask->tbSink.pSchemaWrapper->nCols, 1);
-    /*A(pTask->tbSink.pTSchema);*/
+        tBuildTSchema(pTask->tbSink.pSchemaWrapper->pSchema, pTask->tbSink.pSchemaWrapper->nCols, 1);
+    ASSERT(pTask->tbSink.pTSchema);
  }

  streamSetupTrigger(pTask);
@ -1010,6 +1014,7 @@ int32_t tqProcessStreamTaskCheckReq(STQ* pTq, SRpcMsg* pMsg) {
    tqError("unable to encode rsp %d", __LINE__);
    return -1;
  }
+
  void* buf = rpcMallocCont(sizeof(SMsgHead) + len);
  ((SMsgHead*)buf)->vgId = htonl(req.upstreamNodeId);

@ -1334,12 +1339,12 @@ int32_t tqProcessDelReq(STQ* pTq, void* pReq, int32_t len, int64_t ver) {
  return 0;
 }

-int32_t tqProcessSubmitReq(STQ* pTq, SSubmitReq* pReq, int64_t ver) {
-  void*              pIter = NULL;
-  bool               failed = false;
-  SStreamDataSubmit* pSubmit = NULL;
+int32_t tqProcessSubmitReq(STQ* pTq, SPackedData submit) {
+  void*               pIter = NULL;
+  bool                failed = false;
+  SStreamDataSubmit2* pSubmit = NULL;

-  pSubmit = streamDataSubmitNew(pReq);
+  pSubmit = streamDataSubmitNew(submit);
  if (pSubmit == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    tqError("failed to create data submit for stream since out of memory");
@ -1356,7 +1361,7 @@ int32_t tqProcessSubmitReq(STQ* pTq, SSubmitReq* pReq, int64_t ver) {
      continue;
    }

-    tqDebug("data submit enqueue stream task: %d, ver: %" PRId64, pTask->taskId, ver);
+    tqDebug("data submit enqueue stream task: %d, ver: %" PRId64, pTask->taskId, submit.ver);

    if (!failed) {
      if (streamTaskInput(pTask, (SStreamQueueItem*)pSubmit) < 0) {
--- a/source/dnode/vnode/src/tq/tqExec.c
+++ b/source/dnode/vnode/src/tq/tqExec.c
@ -113,14 +113,8 @@ int32_t tqScanData(STQ* pTq, const STqHandle* pHandle, SMqDataRsp* pRsp, STqOffs
    return -1;
  }

-  if (pRsp->withTbName) {
-    if (pRsp->rspOffset.type == TMQ_OFFSET__LOG) {
-      int64_t uid = pExec->pExecReader->msgIter.uid;
-      tqAddTbNameToRsp(pTq, uid, pRsp, 1);
-    } else {
-      pRsp->withTbName = false;
-    }
-  }
+  ASSERT(pRsp->withTbName == false);
+  ASSERT(pRsp->withSchema == false);

  return 0;
 }
@ -157,9 +151,8 @@ int32_t tqScanTaosx(STQ* pTq, const STqHandle* pHandle, STaosxRsp* pRsp, SMqMeta

    if (pDataBlock != NULL) {
      if (pRsp->withTbName) {
-        int64_t uid = 0;
        if (pOffset->type == TMQ_OFFSET__LOG) {
-          uid = pExec->pExecReader->msgIter.uid;
+          int64_t uid = pExec->pExecReader->lastBlkUid;
          if (tqAddTbNameToRsp(pTq, uid, (SMqDataRsp*)pRsp, 1) < 0) {
            continue;
          }
@ -229,7 +222,7 @@ int32_t tqScanTaosx(STQ* pTq, const STqHandle* pHandle, STaosxRsp* pRsp, SMqMeta
  return 0;
 }

-int32_t tqTaosxScanLog(STQ* pTq, STqHandle* pHandle, SSubmitReq* pReq, STaosxRsp* pRsp) {
+int32_t tqTaosxScanLog(STQ* pTq, STqHandle* pHandle, SPackedData submit, STaosxRsp* pRsp) {
  STqExecHandle* pExec = &pHandle->execHandle;
  /*A(pExec->subType != TOPIC_SUB_TYPE__COLUMN);*/

@ -238,8 +231,9 @@ int32_t tqTaosxScanLog(STQ* pTq, STqHandle* pHandle, SSubmitReq* pReq, STaosxRsp

  if (pExec->subType == TOPIC_SUB_TYPE__TABLE) {
    STqReader* pReader = pExec->pExecReader;
-    tqReaderSetDataMsg(pReader, pReq, 0);
-    while (tqNextDataBlock(pReader)) {
+    /*tqReaderSetDataMsg(pReader, pReq, 0);*/
+    tqReaderSetSubmitReq2(pReader, submit.msgStr, submit.msgLen, submit.ver);
+    while (tqNextDataBlock2(pReader)) {
      /*SSDataBlock block = {0};*/
      /*if (tqRetrieveDataBlock(&block, pReader) < 0) {*/
      /*if (terrno == TSDB_CODE_TQ_TABLE_SCHEMA_NOT_FOUND) continue;*/
@ -247,11 +241,12 @@ int32_t tqTaosxScanLog(STQ* pTq, STqHandle* pHandle, SSubmitReq* pReq, STaosxRsp

      taosArrayClear(pBlocks);
      taosArrayClear(pSchemas);
-      if (tqRetrieveTaosxBlock(pReader, pBlocks, pSchemas) < 0) {
+      if (tqRetrieveTaosxBlock2(pReader, pBlocks, pSchemas) < 0) {
        if (terrno == TSDB_CODE_TQ_TABLE_SCHEMA_NOT_FOUND) continue;
      }
      if (pRsp->withTbName) {
-        int64_t uid = pExec->pExecReader->msgIter.uid;
+        /*int64_t uid = pExec->pExecReader->msgIter.uid;*/
+        int64_t uid = pExec->pExecReader->lastBlkUid;
        if (tqAddTbNameToRsp(pTq, uid, (SMqDataRsp*)pRsp, taosArrayGetSize(pBlocks)) < 0) {
          taosArrayDestroyEx(pBlocks, (FDelete)blockDataFreeRes);
          taosArrayDestroyP(pSchemas, (FDelete)tDeleteSSchemaWrapper);
@ -261,7 +256,9 @@ int32_t tqTaosxScanLog(STQ* pTq, STqHandle* pHandle, SSubmitReq* pReq, STaosxRsp
        }
      }
      if (pHandle->fetchMeta) {
+#if 0
        SSubmitBlk* pBlk = pReader->pBlock;
+        int64_t uid = pExec->pExecReader->lastBlkUid;
        int32_t     schemaLen = htonl(pBlk->schemaLen);
        if (schemaLen > 0) {
          if (pRsp->createTableNum == 0) {
@ -274,6 +271,7 @@ int32_t tqTaosxScanLog(STQ* pTq, STqHandle* pHandle, SSubmitReq* pReq, STaosxRsp
          taosArrayPush(pRsp->createTableReq, &createReq);
          pRsp->createTableNum++;
        }
+#endif
      }
      for (int32_t i = 0; i < taosArrayGetSize(pBlocks); i++) {
        SSDataBlock* pBlock = taosArrayGet(pBlocks, i);
@ -287,19 +285,20 @@ int32_t tqTaosxScanLog(STQ* pTq, STqHandle* pHandle, SSubmitReq* pReq, STaosxRsp
    }
  } else if (pExec->subType == TOPIC_SUB_TYPE__DB) {
    STqReader* pReader = pExec->pExecReader;
-    tqReaderSetDataMsg(pReader, pReq, 0);
-    while (tqNextDataBlockFilterOut(pReader, pExec->execDb.pFilterOutTbUid)) {
+    /*tqReaderSetDataMsg(pReader, pReq, 0);*/
+    tqReaderSetSubmitReq2(pReader, submit.msgStr, submit.msgLen, submit.ver);
+    while (tqNextDataBlockFilterOut2(pReader, pExec->execDb.pFilterOutTbUid)) {
      /*SSDataBlock block = {0};*/
      /*if (tqRetrieveDataBlock(&block, pReader) < 0) {*/
      /*if (terrno == TSDB_CODE_TQ_TABLE_SCHEMA_NOT_FOUND) continue;*/
      /*}*/
      taosArrayClear(pBlocks);
      taosArrayClear(pSchemas);
-      if (tqRetrieveTaosxBlock(pReader, pBlocks, pSchemas) < 0) {
+      if (tqRetrieveTaosxBlock2(pReader, pBlocks, pSchemas) < 0) {
        if (terrno == TSDB_CODE_TQ_TABLE_SCHEMA_NOT_FOUND) continue;
      }
      if (pRsp->withTbName) {
-        int64_t uid = pExec->pExecReader->msgIter.uid;
+        int64_t uid = pExec->pExecReader->lastBlkUid;
        if (tqAddTbNameToRsp(pTq, uid, (SMqDataRsp*)pRsp, taosArrayGetSize(pBlocks)) < 0) {
          taosArrayDestroyEx(pBlocks, (FDelete)blockDataFreeRes);
          taosArrayDestroyP(pSchemas, (FDelete)tDeleteSSchemaWrapper);
@ -309,6 +308,7 @@ int32_t tqTaosxScanLog(STQ* pTq, STqHandle* pHandle, SSubmitReq* pReq, STaosxRsp
        }
      }
      if (pHandle->fetchMeta) {
+#if 0
        SSubmitBlk* pBlk = pReader->pBlock;
        int32_t     schemaLen = htonl(pBlk->schemaLen);
        if (schemaLen > 0) {
@ -322,6 +322,7 @@ int32_t tqTaosxScanLog(STQ* pTq, STqHandle* pHandle, SSubmitReq* pReq, STaosxRsp
          taosArrayPush(pRsp->createTableReq, &createReq);
          pRsp->createTableNum++;
        }
+#endif
      }
      /*tqAddBlockDataToRsp(&block, (SMqDataRsp*)pRsp, taosArrayGetSize(block.pDataBlock),*/
      /*pTq->pVnode->config.tsdbCfg.precision);*/
--- a/source/dnode/vnode/src/tq/tqPush.c
+++ b/source/dnode/vnode/src/tq/tqPush.c
@ -207,7 +207,11 @@ int32_t tqPushMsgNew(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_
 #endif

 int tqPushMsg(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_t ver) {
-  tqDebug("vgId:%d, tq push msg ver %" PRId64 ", type: %s", pTq->pVnode->config.vgId, ver, TMSG_INFO(msgType));
+  void*   pReq = POINTER_SHIFT(msg, sizeof(SMsgHead));
+  int32_t len = msgLen - sizeof(SMsgHead);
+
+  tqDebug("vgId:%d, tq push msg ver %" PRId64 ", type: %s, p head %p, p body %p, len %d", pTq->pVnode->config.vgId, ver,
+          TMSG_INFO(msgType), msg, pReq, len);

  if (msgType == TDMT_VND_SUBMIT) {
    // lock push mgr to avoid potential msg lost
@ -216,7 +220,7 @@ int tqPushMsg(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_t ver)
    if (taosHashGetSize(pTq->pPushMgr) != 0) {
      SArray* cachedKeys = taosArrayInit(0, sizeof(void*));
      SArray* cachedKeyLens = taosArrayInit(0, sizeof(size_t));
-      void*   data = taosMemoryMalloc(msgLen);
+      void*   data = taosMemoryMalloc(len);
      if (data == NULL) {
        terrno = TSDB_CODE_OUT_OF_MEMORY;
        tqError("failed to copy data for stream since out of memory");
@ -224,9 +228,7 @@ int tqPushMsg(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_t ver)
        taosArrayDestroy(cachedKeyLens);
        return -1;
      }
-      memcpy(data, msg, msgLen);
-      SSubmitReq* pReq = (SSubmitReq*)data;
-      pReq->version = ver;
+      memcpy(data, pReq, len);

      void* pIter = NULL;
      while (1) {
@ -250,7 +252,12 @@ int tqPushMsg(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_t ver)
        SMqDataRsp* pRsp = &pPushEntry->dataRsp;

        // prepare scan mem data
-        qStreamScanMemData(task, pReq);
+        SPackedData submit = {
+            .msgStr = data,
+            .msgLen = len,
+            .ver = ver,
+        };
+        qStreamSetScanMemData(task, submit);

        // exec
        while (1) {
@ -304,17 +311,22 @@ int tqPushMsg(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_t ver)
  if (vnodeIsRoleLeader(pTq->pVnode)) {
    if (taosHashGetSize(pTq->pStreamMeta->pTasks) == 0) return 0;
    if (msgType == TDMT_VND_SUBMIT) {
-      void* data = taosMemoryMalloc(msgLen);
+      void* data = taosMemoryMalloc(len);
      if (data == NULL) {
        terrno = TSDB_CODE_OUT_OF_MEMORY;
        tqError("failed to copy data for stream since out of memory");
        return -1;
      }
-      memcpy(data, msg, msgLen);
-      SSubmitReq* pReq = (SSubmitReq*)data;
-      pReq->version = ver;
+      memcpy(data, pReq, len);
+      SPackedData submit = {
+          .msgStr = data,
+          .msgLen = len,
+          .ver = ver,
+      };

-      tqProcessSubmitReq(pTq, data, ver);
+      tqDebug("tq copy write msg %p %d %" PRId64 " from %p", data, len, ver, pReq);
+
+      tqProcessSubmitReq(pTq, submit);
    }
    if (msgType == TDMT_VND_DELETE) {
      tqProcessDelReq(pTq, POINTER_SHIFT(msg, sizeof(SMsgHead)), msgLen - sizeof(SMsgHead), ver);
--- a/source/dnode/vnode/src/tq/tqRead.c
+++ b/source/dnode/vnode/src/tq/tqRead.c
@ -247,7 +247,7 @@ END:
 }

 STqReader* tqOpenReader(SVnode* pVnode) {
-  STqReader* pReader = taosMemoryMalloc(sizeof(STqReader));
+  STqReader* pReader = taosMemoryCalloc(1, sizeof(STqReader));
  if (pReader == NULL) {
    return NULL;
  }
@ -259,7 +259,7 @@ STqReader* tqOpenReader(SVnode* pVnode) {
  }

  pReader->pVnodeMeta = pVnode->pMeta;
-  pReader->pMsg = NULL;
+  /*pReader->pMsg = NULL;*/
  pReader->ver = -1;
  pReader->pColIdList = NULL;
  pReader->cachedSchemaVer = 0;
@ -298,7 +298,7 @@ int32_t tqSeekVer(STqReader* pReader, int64_t ver) {
 }

 int32_t tqNextBlock(STqReader* pReader, SFetchRet* ret) {
-  bool fromProcessedMsg = pReader->pMsg != NULL;
+  bool fromProcessedMsg = pReader->msg2.msgStr != NULL;

  while (1) {
    if (!fromProcessedMsg) {
@ -311,7 +311,9 @@ int32_t tqNextBlock(STqReader* pReader, SFetchRet* ret) {
        tqDebug("return offset %" PRId64 ", no more valid", ret->offset.version);
        return -1;
      }
-      void* body = pReader->pWalReader->pHead->head.body;
+      void*   body = POINTER_SHIFT(pReader->pWalReader->pHead->head.body, sizeof(SMsgHead));
+      int32_t bodyLen = pReader->pWalReader->pHead->head.bodyLen - sizeof(SMsgHead);
+      int64_t ver = pReader->pWalReader->pHead->head.version;
 #if 0
      if (pReader->pWalReader->pHead->head.msgType != TDMT_VND_SUBMIT) {
        // TODO do filter
@ -320,16 +322,17 @@ int32_t tqNextBlock(STqReader* pReader, SFetchRet* ret) {
        return 0;
      } else {
 #endif
-      tqReaderSetDataMsg(pReader, body, pReader->pWalReader->pHead->head.version);
+      tqReaderSetSubmitReq2(pReader, body, bodyLen, ver);
+      /*tqReaderSetDataMsg(pReader, body, pReader->pWalReader->pHead->head.version);*/
 #if 0
      }
 #endif
    }

-    while (tqNextDataBlock(pReader)) {
+    while (tqNextDataBlock2(pReader)) {
      // TODO mem free
      memset(&ret->data, 0, sizeof(SSDataBlock));
-      int32_t code = tqRetrieveDataBlock(&ret->data, pReader);
+      int32_t code = tqRetrieveDataBlock2(&ret->data, pReader);
      if (code != 0 || ret->data.info.rows == 0) {
        continue;
      }
@ -348,6 +351,7 @@ int32_t tqNextBlock(STqReader* pReader, SFetchRet* ret) {
  }
 }

+#if 0
 int32_t tqReaderSetDataMsg(STqReader* pReader, const SSubmitReq* pMsg, int64_t ver) {
  pReader->pMsg = pMsg;

@ -362,7 +366,33 @@ int32_t tqReaderSetDataMsg(STqReader* pReader, const SSubmitReq* pMsg, int64_t v
  memset(&pReader->blkIter, 0, sizeof(SSubmitBlkIter));
  return 0;
 }
+#endif

+int32_t tqReaderSetSubmitReq2(STqReader* pReader, void* msgStr, int32_t msgLen, int64_t ver) {
+  ASSERT(pReader->msg2.msgStr == NULL);
+  ASSERT(msgStr);
+  ASSERT(msgLen);
+  ASSERT(ver >= 0);
+  pReader->msg2.msgStr = msgStr;
+  pReader->msg2.msgLen = msgLen;
+  pReader->msg2.ver = ver;
+  pReader->ver = ver;
+
+  tqDebug("tq reader set msg %p %d", msgStr, msgLen);
+
+  if (pReader->setMsg == 0) {
+    SDecoder decoder;
+    tDecoderInit(&decoder, pReader->msg2.msgStr, pReader->msg2.msgLen);
+    if (tDecodeSSubmitReq2(&decoder, &pReader->submit) < 0) {
+      ASSERT(0);
+    }
+    tDecoderClear(&decoder);
+    pReader->setMsg = 1;
+  }
+  return 0;
+}
+
+#if 0
 bool tqNextDataBlock(STqReader* pReader) {
  if (pReader->pMsg == NULL) return false;
  while (1) {
@ -386,6 +416,59 @@ bool tqNextDataBlock(STqReader* pReader) {
  }
  return false;
 }
+#endif
+
+bool tqNextDataBlock2(STqReader* pReader) {
+  if (pReader->msg2.msgStr == NULL) return false;
+  ASSERT(pReader->setMsg == 1);
+
+  tqDebug("tq reader next data block %p, %d %" PRId64 " %d", pReader->msg2.msgStr, pReader->msg2.msgLen,
+          pReader->msg2.ver, pReader->nextBlk);
+
+  int32_t blockSz = taosArrayGetSize(pReader->submit.aSubmitTbData);
+  while (pReader->nextBlk < blockSz) {
+    SSubmitTbData* pSubmitTbData = taosArrayGet(pReader->submit.aSubmitTbData, pReader->nextBlk);
+    ASSERT(pSubmitTbData->uid);
+
+    if (pReader->tbIdHash == NULL) return true;
+
+    void* ret = taosHashGet(pReader->tbIdHash, &pSubmitTbData->uid, sizeof(int64_t));
+    if (ret != NULL) {
+      return true;
+    }
+    pReader->nextBlk++;
+  }
+
+  tDestroySSubmitReq2(&pReader->submit, TSDB_MSG_FLG_DECODE);
+  pReader->setMsg = 0;
+  pReader->nextBlk = 0;
+  pReader->msg2.msgStr = NULL;
+
+  return false;
+}
+
+bool tqNextDataBlockFilterOut2(STqReader* pReader, SHashObj* filterOutUids) {
+  if (pReader->msg2.msgStr == NULL) return false;
+  ASSERT(pReader->setMsg == 1);
+
+  int32_t blockSz = taosArrayGetSize(pReader->submit.aSubmitTbData);
+  while (pReader->nextBlk < blockSz) {
+    SSubmitTbData* pSubmitTbData = taosArrayGet(pReader->submit.aSubmitTbData, pReader->nextBlk);
+    if (pReader->tbIdHash == NULL) return true;
+
+    void* ret = taosHashGet(pReader->tbIdHash, &pSubmitTbData->uid, sizeof(int64_t));
+    if (ret == NULL) {
+      return true;
+    }
+  }
+
+  tDestroySSubmitReq2(&pReader->submit, TSDB_MSG_FLG_DECODE);
+  pReader->setMsg = 0;
+  pReader->nextBlk = 0;
+  pReader->msg2.msgStr = NULL;
+
+  return false;
+}

 int32_t tqMaskBlock(SSchemaWrapper* pDst, SSDataBlock* pBlock, const SSchemaWrapper* pSrc, char* mask) {
  int32_t code;
@ -416,6 +499,7 @@ int32_t tqMaskBlock(SSchemaWrapper* pDst, SSDataBlock* pBlock, const SSchemaWrap
  return 0;
 }

+#if 0
 bool tqNextDataBlockFilterOut(STqReader* pHandle, SHashObj* filterOutUids) {
  while (1) {
    if (tGetSubmitMsgNext(&pHandle->msgIter, &pHandle->pBlock) < 0) {
@ -431,6 +515,253 @@ bool tqNextDataBlockFilterOut(STqReader* pHandle, SHashObj* filterOutUids) {
  return false;
 }

+int32_t tqScanSubmitSplit(SArray* pBlocks, SArray* schemas, STqReader* pReader) {
+  //
+  int32_t sversion = htonl(pReader->pBlock->sversion);
+  if (pReader->cachedSchemaSuid == 0 || pReader->cachedSchemaVer != sversion ||
+      pReader->cachedSchemaSuid != pReader->msgIter.suid) {
+    taosMemoryFree(pReader->pSchema);
+    pReader->pSchema = metaGetTbTSchema(pReader->pVnodeMeta, pReader->msgIter.uid, sversion, 1);
+    if (pReader->pSchema == NULL) {
+      tqWarn("vgId:%d, cannot found tsschema for table: uid:%" PRId64 " (suid:%" PRId64
+             "), version %d, possibly dropped table",
+             pReader->pWalReader->pWal->cfg.vgId, pReader->msgIter.uid, pReader->msgIter.suid, sversion);
+      pReader->cachedSchemaSuid = 0;
+      terrno = TSDB_CODE_TQ_TABLE_SCHEMA_NOT_FOUND;
+      return -1;
+    }
+
+    tDeleteSSchemaWrapper(pReader->pSchemaWrapper);
+    pReader->pSchemaWrapper = metaGetTableSchema(pReader->pVnodeMeta, pReader->msgIter.uid, sversion, 1);
+    if (pReader->pSchemaWrapper == NULL) {
+      tqWarn("vgId:%d, cannot found schema wrapper for table: suid:%" PRId64 ", version %d, possibly dropped table",
+             pReader->pWalReader->pWal->cfg.vgId, pReader->msgIter.uid, pReader->cachedSchemaVer);
+      pReader->cachedSchemaSuid = 0;
+      terrno = TSDB_CODE_TQ_TABLE_SCHEMA_NOT_FOUND;
+      return -1;
+    }
+
+    STSchema*       pTschema = pReader->pSchema;
+    SSchemaWrapper* pSchemaWrapper = pReader->pSchemaWrapper;
+
+    int32_t colNumNeed = taosArrayGetSize(pReader->pColIdList);
+  }
+  return 0;
+}
+#endif
+
+int32_t tqRetrieveDataBlock2(SSDataBlock* pBlock, STqReader* pReader) {
+  int32_t blockSz = taosArrayGetSize(pReader->submit.aSubmitTbData);
+  ASSERT(pReader->nextBlk < blockSz);
+
+  tqDebug("tq reader retrieve data block %p, %d", pReader->msg2.msgStr, pReader->nextBlk);
+
+  SSubmitTbData* pSubmitTbData = taosArrayGet(pReader->submit.aSubmitTbData, pReader->nextBlk);
+  pReader->nextBlk++;
+
+  int32_t sversion = pSubmitTbData->sver;
+  int64_t suid = pSubmitTbData->suid;
+  int64_t uid = pSubmitTbData->uid;
+  pReader->lastBlkUid = uid;
+
+  pBlock->info.id.uid = uid;
+  pBlock->info.version = pReader->msg2.ver;
+
+  if (pReader->cachedSchemaSuid == 0 || pReader->cachedSchemaVer != sversion || pReader->cachedSchemaSuid != suid) {
+    taosMemoryFree(pReader->pSchema);
+    pReader->pSchema = metaGetTbTSchema(pReader->pVnodeMeta, uid, sversion, 1);
+    if (pReader->pSchema == NULL) {
+      tqWarn("vgId:%d, cannot found tsschema for table: uid:%" PRId64 " (suid:%" PRId64
+             "), version %d, possibly dropped table",
+             pReader->pWalReader->pWal->cfg.vgId, uid, suid, sversion);
+      pReader->cachedSchemaSuid = 0;
+      terrno = TSDB_CODE_TQ_TABLE_SCHEMA_NOT_FOUND;
+      return -1;
+    }
+
+    tDeleteSSchemaWrapper(pReader->pSchemaWrapper);
+    pReader->pSchemaWrapper = metaGetTableSchema(pReader->pVnodeMeta, uid, sversion, 1);
+    if (pReader->pSchemaWrapper == NULL) {
+      tqWarn("vgId:%d, cannot found schema wrapper for table: suid:%" PRId64 ", version %d, possibly dropped table",
+             pReader->pWalReader->pWal->cfg.vgId, uid, pReader->cachedSchemaVer);
+      pReader->cachedSchemaSuid = 0;
+      terrno = TSDB_CODE_TQ_TABLE_SCHEMA_NOT_FOUND;
+      return -1;
+    }
+
+    STSchema*       pTschema = pReader->pSchema;
+    SSchemaWrapper* pSchemaWrapper = pReader->pSchemaWrapper;
+
+    int32_t colNumNeed = taosArrayGetSize(pReader->pColIdList);
+
+    if (colNumNeed == 0) {
+      int32_t colMeta = 0;
+      while (colMeta < pSchemaWrapper->nCols) {
+        SSchema*        pColSchema = &pSchemaWrapper->pSchema[colMeta];
+        SColumnInfoData colInfo = createColumnInfoData(pColSchema->type, pColSchema->bytes, pColSchema->colId);
+        int32_t         code = blockDataAppendColInfo(pBlock, &colInfo);
+        if (code != TSDB_CODE_SUCCESS) {
+          goto FAIL;
+        }
+        colMeta++;
+      }
+    } else {
+      if (colNumNeed > pSchemaWrapper->nCols) {
+        colNumNeed = pSchemaWrapper->nCols;
+      }
+
+      int32_t colMeta = 0;
+      int32_t colNeed = 0;
+      while (colMeta < pSchemaWrapper->nCols && colNeed < colNumNeed) {
+        SSchema* pColSchema = &pSchemaWrapper->pSchema[colMeta];
+        col_id_t colIdSchema = pColSchema->colId;
+        col_id_t colIdNeed = *(col_id_t*)taosArrayGet(pReader->pColIdList, colNeed);
+        if (colIdSchema < colIdNeed) {
+          colMeta++;
+        } else if (colIdSchema > colIdNeed) {
+          colNeed++;
+        } else {
+          SColumnInfoData colInfo = createColumnInfoData(pColSchema->type, pColSchema->bytes, pColSchema->colId);
+          int32_t         code = blockDataAppendColInfo(pBlock, &colInfo);
+          if (code != TSDB_CODE_SUCCESS) {
+            goto FAIL;
+          }
+          colMeta++;
+          colNeed++;
+        }
+      }
+    }
+
+    int32_t numOfRows = 0;
+
+    if (pSubmitTbData->flags & SUBMIT_REQ_COLUMN_DATA_FORMAT) {
+      SArray*   pCols = pSubmitTbData->aCol;
+      SColData* pCol = taosArrayGet(pCols, 0);
+      numOfRows = pCol->nVal;
+    } else {
+      SArray* pRows = pSubmitTbData->aRowP;
+      numOfRows = taosArrayGetSize(pRows);
+    }
+
+    if (blockDataEnsureCapacity(pBlock, numOfRows) < 0) {
+      terrno = TSDB_CODE_OUT_OF_MEMORY;
+      goto FAIL;
+    }
+    pBlock->info.rows = numOfRows;
+
+    int32_t colActual = blockDataGetNumOfCols(pBlock);
+
+    // convert and scan one block
+    if (pSubmitTbData->flags & SUBMIT_REQ_COLUMN_DATA_FORMAT) {
+      SArray* pCols = pSubmitTbData->aCol;
+      int32_t numOfCols = taosArrayGetSize(pCols);
+      int32_t targetIdx = 0;
+      int32_t sourceIdx = 0;
+      while (targetIdx < colActual) {
+        ASSERT(sourceIdx < numOfCols);
+
+        SColData*        pCol = taosArrayGet(pCols, sourceIdx);
+        SColumnInfoData* pColData = taosArrayGet(pBlock->pDataBlock, targetIdx);
+        SColVal          colVal;
+
+        ASSERT(pCol->nVal == numOfRows);
+
+        if (pCol->cid < pColData->info.colId) {
+          sourceIdx++;
+        } else if (pCol->cid == pColData->info.colId) {
+          for (int32_t i = 0; i < pCol->nVal; i++) {
+            tColDataGetValue(pCol, sourceIdx, &colVal);
+#if 0
+            void* val = NULL;
+            if (IS_STR_DATA_TYPE(colVal.type)) {
+              val = colVal.value.pData;
+            } else {
+              val = &colVal.value.val;
+            }
+            if (colDataAppend(pColData, i, val, !COL_VAL_IS_VALUE(&colVal)) < 0) {
+              goto FAIL;
+            }
+#endif
+            if (IS_STR_DATA_TYPE(colVal.type)) {
+              if (colVal.value.pData != NULL) {
+                char val[65535 + 2];
+                memcpy(varDataVal(val), colVal.value.pData, colVal.value.nData);
+                varDataSetLen(val, colVal.value.nData);
+                if (colDataAppend(pColData, i, val, !COL_VAL_IS_VALUE(&colVal)) < 0) {
+                  goto FAIL;
+                }
+              } else {
+                colDataAppendNULL(pColData, i);
+              }
+            } else {
+              if (colDataAppend(pColData, i, (void*)&colVal.value.val, !COL_VAL_IS_VALUE(&colVal)) < 0) {
+                goto FAIL;
+              }
+            }
+          }
+          sourceIdx++;
+          targetIdx++;
+        } else {
+          ASSERT(0);
+        }
+      }
+    } else {
+      SArray* pRows = pSubmitTbData->aRowP;
+
+      for (int32_t i = 0; i < numOfRows; i++) {
+        SRow*   pRow = taosArrayGetP(pRows, i);
+        int32_t targetIdx = 0;
+        int32_t sourceIdx = 0;
+
+        for (int32_t j = 0; j < colActual; j++) {
+          SColumnInfoData* pColData = taosArrayGet(pBlock->pDataBlock, j);
+          while (1) {
+            ASSERT(sourceIdx < pTschema->numOfCols);
+
+            SColVal colVal;
+            tRowGet(pRow, pTschema, sourceIdx, &colVal);
+            if (colVal.cid < pColData->info.colId) {
+              sourceIdx++;
+              continue;
+            } else if (colVal.cid == pColData->info.colId) {
+              if (IS_STR_DATA_TYPE(colVal.type)) {
+                if (colVal.value.pData != NULL) {
+                  char val[65535 + 2];
+                  memcpy(varDataVal(val), colVal.value.pData, colVal.value.nData);
+                  varDataSetLen(val, colVal.value.nData);
+                  if (colDataAppend(pColData, i, val, !COL_VAL_IS_VALUE(&colVal)) < 0) {
+                    goto FAIL;
+                  }
+                } else {
+                  colDataAppendNULL(pColData, i);
+                }
+                /*val = colVal.value.pData;*/
+              } else {
+                if (colDataAppend(pColData, i, (void*)&colVal.value.val, !COL_VAL_IS_VALUE(&colVal)) < 0) {
+                  goto FAIL;
+                }
+              }
+
+              sourceIdx++;
+              targetIdx++;
+              break;
+            } else {
+              ASSERT(0);
+            }
+          }
+        }
+      }
+    }
+  }
+
+  return 0;
+
+FAIL:
+  blockDataFreeRes(pBlock);
+  return -1;
+}
+
+#if 0
 int32_t tqRetrieveDataBlock(SSDataBlock* pBlock, STqReader* pReader) {
  // TODO: cache multiple schema
  int32_t sversion = htonl(pReader->pBlock->sversion);
@ -673,6 +1004,18 @@ FAIL:
  taosMemoryFree(assigned);
  return -1;
 }
+#endif
+
+int32_t tqRetrieveTaosxBlock2(STqReader* pReader, SArray* blocks, SArray* schemas) {
+  SSDataBlock block = {0};
+  if (tqRetrieveDataBlock2(&block, pReader) == 0) {
+    taosArrayPush(blocks, &block);
+    SSchemaWrapper* pSW = tCloneSSchemaWrapper(pReader->pSchemaWrapper);
+    taosArrayPush(schemas, &pSW);
+    return 0;
+  }
+  return -1;
+}

 void tqReaderSetColIdList(STqReader* pReader, SArray* pColIdList) { pReader->pColIdList = pColIdList; }

--- a/source/dnode/vnode/src/tq/tqSink.c
+++ b/source/dnode/vnode/src/tq/tqSink.c
@ -71,6 +71,7 @@ int32_t tqBuildDeleteReq(SVnode* pVnode, const char* stbFullName, const SSDataBl
  return 0;
 }

+#if 0
 SSubmitReq* tqBlockToSubmit(SVnode* pVnode, const SArray* pBlocks, const STSchema* pTSchema,
                            SSchemaWrapper* pTagSchemaWrapper, bool createTb, int64_t suid, const char* stbFullName,
                            SBatchDeleteReq* pDeleteReq) {
@ -251,8 +252,6 @@ SSubmitReq* tqBlockToSubmit(SVnode* pVnode, const SArray* pBlocks, const STSchem

    int32_t rows = pDataBlock->info.rows;

-    tqDebug("tq sink, convert block1 %d, rows: %d", i, rows);
-
    int32_t dataLen = 0;
    int32_t schemaLen = 0;
    void*   blkSchema = POINTER_SHIFT(blkHead, sizeof(SSubmitBlk));
@ -298,6 +297,201 @@ SSubmitReq* tqBlockToSubmit(SVnode* pVnode, const SArray* pBlocks, const STSchem

  return ret;
 }
+#endif
+
+int32_t tqBlockToSubmit(SVnode* pVnode, const SArray* pBlocks, const STSchema* pTSchema,
+                        SSchemaWrapper* pTagSchemaWrapper, bool createTb, int64_t suid, const char* stbFullName,
+                        SBatchDeleteReq* pDeleteReq, void** ppData, int32_t* pLen) {
+  void*        pBuf = NULL;
+  int32_t      len = 0;
+  SSubmitReq2* pReq = NULL;
+  SArray*      tagArray = NULL;
+  SArray*      createTbArray = NULL;
+  SArray*      pVals = NULL;
+
+  int32_t sz = taosArrayGetSize(pBlocks);
+
+  if (!(tagArray = taosArrayInit(1, sizeof(STagVal)))) {
+    goto _end;
+  }
+
+  if (!(createTbArray = taosArrayInit(sz, POINTER_BYTES))) {
+    goto _end;
+  }
+
+  if (!(pReq = taosMemoryCalloc(1, sizeof(SSubmitReq2)))) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    goto _end;
+  }
+
+  if (!(pReq->aSubmitTbData = taosArrayInit(1, sizeof(SSubmitTbData)))) {
+    goto _end;
+  }
+
+  // create table req
+  if (createTb) {
+    for (int32_t i = 0; i < sz; ++i) {
+      SSDataBlock*   pDataBlock = taosArrayGet(pBlocks, i);
+      SVCreateTbReq* pCreateTbReq = NULL;
+      if (pDataBlock->info.type == STREAM_DELETE_RESULT) {
+        taosArrayPush(createTbArray, &pCreateTbReq);
+        continue;
+      }
+
+      if (!(pCreateTbReq = taosMemoryCalloc(1, sizeof(SVCreateStbReq)))) {
+        goto _end;
+      };
+
+      // don't move to the end of loop as to destroy in the end of func when error occur
+      taosArrayPush(createTbArray, &pCreateTbReq);
+
+      // set const
+      pCreateTbReq->flags = 0;
+      pCreateTbReq->type = TSDB_CHILD_TABLE;
+      pCreateTbReq->ctb.suid = suid;
+
+      // set super table name
+      SName name = {0};
+      tNameFromString(&name, stbFullName, T_NAME_ACCT | T_NAME_DB | T_NAME_TABLE);
+      pCreateTbReq->ctb.stbName = strdup((char*)tNameGetTableName(&name));  // strdup(stbFullName);
+
+      // set tag content
+      taosArrayClear(tagArray);
+      STagVal tagVal = {
+          .cid = taosArrayGetSize(pDataBlock->pDataBlock) + 1,
+          .type = TSDB_DATA_TYPE_UBIGINT,
+          .i64 = (int64_t)pDataBlock->info.id.groupId,
+      };
+      taosArrayPush(tagArray, &tagVal);
+      pCreateTbReq->ctb.tagNum = taosArrayGetSize(tagArray);
+
+      STag* pTag = NULL;
+      tTagNew(tagArray, 1, false, &pTag);
+      if (pTag == NULL) {
+        terrno = TSDB_CODE_OUT_OF_MEMORY;
+        goto _end;
+      }
+      pCreateTbReq->ctb.pTag = (uint8_t*)pTag;
+
+      // set tag name
+      SArray* tagName = taosArrayInit(1, TSDB_COL_NAME_LEN);
+      char    tagNameStr[TSDB_COL_NAME_LEN] = {0};
+      strcpy(tagNameStr, "group_id");
+      taosArrayPush(tagName, tagNameStr);
+      pCreateTbReq->ctb.tagName = tagName;
+
+      // set table name
+      if (pDataBlock->info.parTbName[0]) {
+        pCreateTbReq->name = strdup(pDataBlock->info.parTbName);
+      } else {
+        pCreateTbReq->name = buildCtbNameByGroupId(stbFullName, pDataBlock->info.id.groupId);
+      }
+    }
+  }
+
+  // SSubmitTbData req
+  for (int32_t i = 0; i < sz; ++i) {
+    SSDataBlock* pDataBlock = taosArrayGet(pBlocks, i);
+    if (pDataBlock->info.type == STREAM_DELETE_RESULT) {
+      pDeleteReq->suid = suid;
+      pDeleteReq->deleteReqs = taosArrayInit(0, sizeof(SSingleDeleteReq));
+      tqBuildDeleteReq(pVnode, stbFullName, pDataBlock, pDeleteReq);
+      continue;
+    }
+
+    int32_t rows = pDataBlock->info.rows;
+
+    SSubmitTbData* pTbData = (SSubmitTbData*)taosMemoryCalloc(1, sizeof(SSubmitTbData));
+    if (!pTbData) {
+      terrno = TSDB_CODE_OUT_OF_MEMORY;
+      goto _end;
+    }
+
+    if (!(pTbData->aRowP = taosArrayInit(rows, sizeof(SRow*)))) {
+      taosMemoryFree(pTbData);
+      goto _end;
+    }
+    pTbData->suid = suid;
+    pTbData->uid = 0;  // uid is assigned by vnode
+    pTbData->sver = pTSchema->version;
+
+    if (createTb) {
+      pTbData->pCreateTbReq = taosArrayGetP(createTbArray, i);
+      if (pTbData->pCreateTbReq) pTbData->flags = SUBMIT_REQ_AUTO_CREATE_TABLE;
+    }
+
+    if (!pVals && !(pVals = taosArrayInit(pTSchema->numOfCols, sizeof(SColVal)))) {
+      taosArrayDestroy(pTbData->aRowP);
+      taosMemoryFree(pTbData);
+      goto _end;
+    }
+
+    for (int32_t j = 0; j < rows; j++) {
+      taosArrayClear(pVals);
+      for (int32_t k = 0; k < pTSchema->numOfCols; k++) {
+        const STColumn*  pCol = &pTSchema->columns[k];
+        SColumnInfoData* pColData = taosArrayGet(pDataBlock->pDataBlock, k);
+        if (colDataIsNull_s(pColData, j)) {
+          SColVal cv = COL_VAL_NULL(pCol->colId, pCol->type);
+          taosArrayPush(pVals, &cv);
+        } else {
+          void* data = colDataGetData(pColData, j);
+          if (IS_STR_DATA_TYPE(pCol->type)) {
+            SValue  sv = (SValue){.nData = varDataLen(data), .pData = varDataVal(data)};  // address copy, no value
+            SColVal cv = COL_VAL_VALUE(pCol->colId, pCol->type, sv);
+            taosArrayPush(pVals, &cv);
+          } else {
+            SValue sv;
+            memcpy(&sv.val, data, tDataTypes[pCol->type].bytes);
+            SColVal cv = COL_VAL_VALUE(pCol->colId, pCol->type, sv);
+            taosArrayPush(pVals, &cv);
+          }
+        }
+      }
+      SRow* pRow = NULL;
+      if ((terrno = tRowBuild(pVals, (STSchema*)pTSchema, &pRow)) < 0) {
+        tDestroySSubmitTbData(pTbData, TSDB_MSG_FLG_ENCODE);
+        goto _end;
+      }
+      ASSERT(pRow);
+      taosArrayPush(pTbData->aRowP, &pRow);
+    }
+
+    taosArrayPush(pReq->aSubmitTbData, pTbData);
+  }
+
+  // encode
+  tEncodeSize(tEncodeSSubmitReq2, pReq, len, terrno);
+  if (TSDB_CODE_SUCCESS == terrno) {
+    SEncoder encoder;
+    len += sizeof(SMsgHead);
+    pBuf = rpcMallocCont(len);
+    if (NULL == pBuf) {
+      goto _end;
+    }
+    ((SMsgHead*)pBuf)->vgId = htonl(TD_VID(pVnode));
+    ((SMsgHead*)pBuf)->contLen = htonl(len);
+    tEncoderInit(&encoder, POINTER_SHIFT(pBuf, sizeof(SMsgHead)), len - sizeof(SMsgHead));
+    if (tEncodeSSubmitReq2(&encoder, pReq) < 0) {
+      terrno = TSDB_CODE_OUT_OF_MEMORY;
+      tqError("failed to encode submit req since %s", terrstr());
+    }
+    tEncoderClear(&encoder);
+  }
+_end:
+  taosArrayDestroy(tagArray);
+  taosArrayDestroy(pVals);
+  tDestroySSubmitReq2(pReq, TSDB_MSG_FLG_ENCODE);
+
+  if (terrno != 0) {
+    rpcFreeCont(pBuf);
+    taosArrayDestroy(pDeleteReq->deleteReqs);
+    return TSDB_CODE_FAILED;
+  }
+  if (ppData) *ppData = pBuf;
+  if (pLen) *pLen = len;
+  return TSDB_CODE_SUCCESS;
+}

 void tqSinkToTablePipeline(SStreamTask* pTask, void* vnode, int64_t ver, void* data) {
  const SArray*   pBlocks = (const SArray*)data;
@ -492,7 +686,7 @@ void tqSinkToTablePipeline(SStreamTask* pTask, void* vnode, int64_t ver, void* d
      blkHead->uid = 0;
      blkHead->schemaLen = 0;

-      tqDebug("tq sink, convert block2 %d, rows: %d", i, rows);
+      tqDebug("tq sink pipe1, convert block2 %d, rows: %d", i, rows);

      int32_t dataLen = 0;
      void*   blkSchema = POINTER_SHIFT(blkHead, sizeof(SSubmitBlk));
@ -521,7 +715,7 @@ void tqSinkToTablePipeline(SStreamTask* pTask, void* vnode, int64_t ver, void* d
          } else {
            void* colData = colDataGetData(pColData, j);
            if (k == 0) {
-              tqDebug("tq sink, row %d ts %" PRId64, j, *(int64_t*)colData);
+              tqDebug("tq sink pipe1, row %d ts %" PRId64, j, *(int64_t*)colData);
            }
            tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NORM, colData, true, pColumn->offset, k);
          }
@ -550,6 +744,247 @@ void tqSinkToTablePipeline(SStreamTask* pTask, void* vnode, int64_t ver, void* d
  taosArrayDestroy(tagArray);
 }

+void tqSinkToTablePipeline2(SStreamTask* pTask, void* vnode, int64_t ver, void* data) {
+  const SArray* pBlocks = (const SArray*)data;
+  SVnode*       pVnode = (SVnode*)vnode;
+  int64_t       suid = pTask->tbSink.stbUid;
+  char*         stbFullName = pTask->tbSink.stbFullName;
+  STSchema*     pTSchema = pTask->tbSink.pTSchema;
+  /*SSchemaWrapper* pSchemaWrapper = pTask->tbSink.pSchemaWrapper;*/
+
+  int32_t blockSz = taosArrayGetSize(pBlocks);
+
+  tqDebug("vgId:%d, task %d write into table, block num: %d", TD_VID(pVnode), pTask->taskId, blockSz);
+
+  void*   pBuf = NULL;
+  SArray* tagArray = NULL;
+  SArray* pVals = NULL;
+
+  if (!(tagArray = taosArrayInit(1, sizeof(STagVal)))) {
+    goto _end;
+  }
+
+  for (int32_t i = 0; i < blockSz; i++) {
+    SSDataBlock* pDataBlock = taosArrayGet(pBlocks, i);
+    if (pDataBlock->info.type == STREAM_DELETE_RESULT) {
+      SBatchDeleteReq deleteReq = {0};
+      deleteReq.deleteReqs = taosArrayInit(0, sizeof(SSingleDeleteReq));
+      deleteReq.suid = suid;
+      tqBuildDeleteReq(pVnode, stbFullName, pDataBlock, &deleteReq);
+      if (taosArrayGetSize(deleteReq.deleteReqs) == 0) {
+        taosArrayDestroy(deleteReq.deleteReqs);
+        continue;
+      }
+
+      int32_t len;
+      int32_t code;
+      tEncodeSize(tEncodeSBatchDeleteReq, &deleteReq, len, code);
+      if (code < 0) {
+        //
+        ASSERT(0);
+      }
+      SEncoder encoder;
+      void*    serializedDeleteReq = rpcMallocCont(len + sizeof(SMsgHead));
+      void*    abuf = POINTER_SHIFT(serializedDeleteReq, sizeof(SMsgHead));
+      tEncoderInit(&encoder, abuf, len);
+      tEncodeSBatchDeleteReq(&encoder, &deleteReq);
+      tEncoderClear(&encoder);
+      taosArrayDestroy(deleteReq.deleteReqs);
+
+      ((SMsgHead*)serializedDeleteReq)->vgId = pVnode->config.vgId;
+
+      SRpcMsg msg = {
+          .msgType = TDMT_VND_BATCH_DEL,
+          .pCont = serializedDeleteReq,
+          .contLen = len + sizeof(SMsgHead),
+      };
+      if (tmsgPutToQueue(&pVnode->msgCb, WRITE_QUEUE, &msg) != 0) {
+        tqDebug("failed to put delete req into write-queue since %s", terrstr());
+      }
+    } else {
+      SSubmitTbData tbData = {0};
+      int32_t       rows = pDataBlock->info.rows;
+      tqDebug("tq sink pipe2, convert block1 %d, rows: %d", i, rows);
+
+      if (!(tbData.aRowP = taosArrayInit(rows, sizeof(SRow*)))) {
+        goto _end;
+      }
+
+      tbData.suid = suid;
+      tbData.uid = 0;  // uid is assigned by vnode
+      tbData.sver = pTSchema->version;
+
+      char* ctbName = NULL;
+      if (pDataBlock->info.parTbName[0]) {
+        ctbName = strdup(pDataBlock->info.parTbName);
+      } else {
+        ctbName = buildCtbNameByGroupId(stbFullName, pDataBlock->info.id.groupId);
+      }
+
+      SMetaReader mr = {0};
+      metaReaderInit(&mr, pVnode->pMeta, 0);
+      if (metaGetTableEntryByName(&mr, ctbName) < 0) {
+        metaReaderClear(&mr);
+        tqDebug("vgId:%d, stream write into %s, table auto created", TD_VID(pVnode), ctbName);
+
+        SVCreateTbReq* pCreateTbReq = NULL;
+
+        if (!(pCreateTbReq = taosMemoryCalloc(1, sizeof(SVCreateStbReq)))) {
+          goto _end;
+        };
+
+        // set const
+        pCreateTbReq->flags = 0;
+        pCreateTbReq->type = TSDB_CHILD_TABLE;
+        pCreateTbReq->ctb.suid = suid;
+
+        // set super table name
+        SName name = {0};
+        tNameFromString(&name, stbFullName, T_NAME_ACCT | T_NAME_DB | T_NAME_TABLE);
+        pCreateTbReq->ctb.stbName = strdup((char*)tNameGetTableName(&name));  // strdup(stbFullName);
+
+        // set tag content
+        taosArrayClear(tagArray);
+        STagVal tagVal = {
+            .cid = taosArrayGetSize(pDataBlock->pDataBlock) + 1,
+            .type = TSDB_DATA_TYPE_UBIGINT,
+            .i64 = (int64_t)pDataBlock->info.id.groupId,
+        };
+        taosArrayPush(tagArray, &tagVal);
+        pCreateTbReq->ctb.tagNum = taosArrayGetSize(tagArray);
+
+        STag* pTag = NULL;
+        tTagNew(tagArray, 1, false, &pTag);
+        if (pTag == NULL) {
+          terrno = TSDB_CODE_OUT_OF_MEMORY;
+          goto _end;
+        }
+        pCreateTbReq->ctb.pTag = (uint8_t*)pTag;
+
+        // set tag name
+        SArray* tagName = taosArrayInit(1, TSDB_COL_NAME_LEN);
+        char    tagNameStr[TSDB_COL_NAME_LEN] = {0};
+        strcpy(tagNameStr, "group_id");
+        taosArrayPush(tagName, tagNameStr);
+        pCreateTbReq->ctb.tagName = tagName;
+
+        // set table name
+        pCreateTbReq->name = ctbName;
+        ctbName = NULL;
+
+        tbData.pCreateTbReq = pCreateTbReq;
+        tbData.flags = SUBMIT_REQ_AUTO_CREATE_TABLE;
+      } else {
+        if (mr.me.type != TSDB_CHILD_TABLE) {
+          tqError("vgId:%d, failed to write into %s, since table type incorrect, type %d", TD_VID(pVnode), ctbName,
+                  mr.me.type);
+          metaReaderClear(&mr);
+          taosMemoryFree(ctbName);
+          continue;
+        }
+
+        if (mr.me.ctbEntry.suid != suid) {
+          tqError("vgId:%d, failed to write into %s, since suid mismatch, expect suid: %" PRId64
+                  ", actual suid %" PRId64 "",
+                  TD_VID(pVnode), ctbName, suid, mr.me.ctbEntry.suid);
+          metaReaderClear(&mr);
+          taosMemoryFree(ctbName);
+        }
+
+        tbData.uid = mr.me.uid;
+        metaReaderClear(&mr);
+        taosMemoryFreeClear(ctbName);
+      }
+
+      // rows
+      if (!pVals && !(pVals = taosArrayInit(pTSchema->numOfCols, sizeof(SColVal)))) {
+        taosArrayDestroy(tbData.aRowP);
+        goto _end;
+      }
+
+      for (int32_t j = 0; j < rows; j++) {
+        taosArrayClear(pVals);
+        for (int32_t k = 0; k < pTSchema->numOfCols; k++) {
+          const STColumn*  pCol = &pTSchema->columns[k];
+          SColumnInfoData* pColData = taosArrayGet(pDataBlock->pDataBlock, k);
+          if (k == 0) {
+            void* colData = colDataGetData(pColData, j);
+            tqDebug("tq sink pipe2, row %d, col %d ts %" PRId64, j, k, *(int64_t*)colData);
+          }
+          if (colDataIsNull_s(pColData, j)) {
+            SColVal cv = COL_VAL_NULL(pCol->colId, pCol->type);
+            taosArrayPush(pVals, &cv);
+          } else {
+            void* colData = colDataGetData(pColData, j);
+            if (IS_STR_DATA_TYPE(pCol->type)) {
+              SValue sv =
+                  (SValue){.nData = varDataLen(colData), .pData = varDataVal(colData)};  // address copy, no value
+              SColVal cv = COL_VAL_VALUE(pCol->colId, pCol->type, sv);
+              taosArrayPush(pVals, &cv);
+            } else {
+              SValue sv;
+              memcpy(&sv.val, colData, tDataTypes[pCol->type].bytes);
+              SColVal cv = COL_VAL_VALUE(pCol->colId, pCol->type, sv);
+              taosArrayPush(pVals, &cv);
+            }
+          }
+        }
+        SRow* pRow = NULL;
+        if ((terrno = tRowBuild(pVals, (STSchema*)pTSchema, &pRow)) < 0) {
+          tDestroySSubmitTbData(&tbData, TSDB_MSG_FLG_ENCODE);
+          goto _end;
+        }
+        ASSERT(pRow);
+        taosArrayPush(tbData.aRowP, &pRow);
+      }
+
+      SSubmitReq2 submitReq = {0};
+      if (!(submitReq.aSubmitTbData = taosArrayInit(1, sizeof(SSubmitTbData)))) {
+        goto _end;
+      }
+
+      taosArrayPush(submitReq.aSubmitTbData, &tbData);
+
+      // encode
+      int32_t len;
+      int32_t code;
+      tEncodeSize(tEncodeSSubmitReq2, &submitReq, len, code);
+      SEncoder encoder;
+      len += sizeof(SMsgHead);
+      pBuf = rpcMallocCont(len);
+      if (NULL == pBuf) {
+        goto _end;
+      }
+      ((SMsgHead*)pBuf)->vgId = TD_VID(pVnode);
+      ((SMsgHead*)pBuf)->contLen = htonl(len);
+      tEncoderInit(&encoder, POINTER_SHIFT(pBuf, sizeof(SMsgHead)), len - sizeof(SMsgHead));
+      if (tEncodeSSubmitReq2(&encoder, &submitReq) < 0) {
+        terrno = TSDB_CODE_OUT_OF_MEMORY;
+        tqError("failed to encode submit req since %s", terrstr());
+        tEncoderClear(&encoder);
+        rpcFreeCont(pBuf);
+        continue;
+      }
+      tEncoderClear(&encoder);
+      tDestroySSubmitReq2(&submitReq, TSDB_MSG_FLG_ENCODE);
+
+      SRpcMsg msg = {
+          .msgType = TDMT_VND_SUBMIT,
+          .pCont = pBuf,
+          .contLen = len,
+      };
+
+      if (tmsgPutToQueue(&pVnode->msgCb, WRITE_QUEUE, &msg) != 0) {
+        tqDebug("failed to put into write-queue since %s", terrstr());
+      }
+    }
+  }
+_end:
+  taosArrayDestroy(tagArray);
+  taosArrayDestroy(pVals);
+  // TODO: change
+}
+
 #if 0
 void tqSinkToTableMerge(SStreamTask* pTask, void* vnode, int64_t ver, void* data) {
  const SArray*   pRes = (const SArray*)data;
--- a/source/dnode/vnode/src/tsdb/tsdbCache.c
+++ b/source/dnode/vnode/src/tsdb/tsdbCache.c
@ -91,12 +91,10 @@ int32_t tsdbCacheDeleteLastrow(SLRUCache *pCache, tb_uid_t uid, TSKEY eKey) {
      }
    }

+    taosLRUCacheRelease(pCache, h, invalidate);
    if (invalidate) {
-      taosLRUCacheRelease(pCache, h, true);
-    } else {
-      taosLRUCacheRelease(pCache, h, false);
+      taosLRUCacheErase(pCache, key, keyLen);
    }
-    // void taosLRUCacheErase(SLRUCache * cache, const void *key, size_t keyLen);
  }

  return code;
@ -124,12 +122,10 @@ int32_t tsdbCacheDeleteLast(SLRUCache *pCache, tb_uid_t uid, TSKEY eKey) {
      }
    }

+    taosLRUCacheRelease(pCache, h, invalidate);
    if (invalidate) {
-      taosLRUCacheRelease(pCache, h, true);
-    } else {
-      taosLRUCacheRelease(pCache, h, false);
+      taosLRUCacheErase(pCache, key, keyLen);
    }
-    // void taosLRUCacheErase(SLRUCache * cache, const void *key, size_t keyLen);
  }

  return code;
@ -190,7 +186,7 @@ int32_t tsdbCacheDelete(SLRUCache *pCache, tb_uid_t uid, TSKEY eKey) {
  return code;
 }

-int32_t tsdbCacheInsertLastrow(SLRUCache *pCache, STsdb *pTsdb, tb_uid_t uid, STSRow *row, bool dup) {
+int32_t tsdbCacheInsertLastrow(SLRUCache *pCache, STsdb *pTsdb, tb_uid_t uid, TSDBROW *row, bool dup) {
  int32_t code = 0;
  STSRow *cacheRow = NULL;
  char    key[32] = {0};
@ -201,13 +197,51 @@ int32_t tsdbCacheInsertLastrow(SLRUCache *pCache, STsdb *pTsdb, tb_uid_t uid, ST
  LRUHandle *h = taosLRUCacheLookup(pCache, key, keyLen);
  if (h) {
    STSchema *pTSchema = metaGetTbTSchema(pTsdb->pVnode->pMeta, uid, -1, 1);
-    TSKEY     keyTs = row->ts;
+    TSKEY     keyTs = TSDBROW_TS(row);
    bool      invalidate = false;

    SArray *pLast = (SArray *)taosLRUCacheValue(pCache, h);
    int16_t nCol = taosArrayGetSize(pLast);
    int16_t iCol = 0;

+    if (nCol <= 0) {
+      nCol = pTSchema->numOfCols;
+
+      STColumn *pTColumn = &pTSchema->columns[0];
+      SColVal   tColVal = COL_VAL_VALUE(pTColumn->colId, pTColumn->type, (SValue){.val = keyTs});
+      if (taosArrayPush(pLast, &(SLastCol){.ts = keyTs, .colVal = tColVal}) == NULL) {
+        terrno = TSDB_CODE_OUT_OF_MEMORY;
+        code = TSDB_CODE_OUT_OF_MEMORY;
+        goto _invalidate;
+      }
+
+      for (iCol = 1; iCol < nCol; ++iCol) {
+        SColVal colVal = {0};
+        tsdbRowGetColVal(row, pTSchema, iCol, &colVal);
+
+        SLastCol lastCol = {.ts = keyTs, .colVal = colVal};
+        if (IS_VAR_DATA_TYPE(colVal.type) && colVal.value.nData > 0) {
+          SLastCol *pLastCol = (SLastCol *)taosArrayGet(pLast, iCol);
+          taosMemoryFree(pLastCol->colVal.value.pData);
+
+          lastCol.colVal.value.pData = taosMemoryMalloc(colVal.value.nData);
+          if (lastCol.colVal.value.pData == NULL) {
+            terrno = TSDB_CODE_OUT_OF_MEMORY;
+            code = TSDB_CODE_OUT_OF_MEMORY;
+            goto _invalidate;
+          }
+          memcpy(lastCol.colVal.value.pData, colVal.value.pData, colVal.value.nData);
+        }
+
+        if (taosArrayPush(pLast, &lastCol) == NULL) {
+          code = TSDB_CODE_OUT_OF_MEMORY;
+          goto _invalidate;
+        }
+      }
+
+      goto _invalidate;
+    }
+
    SLastCol *tTsVal = (SLastCol *)taosArrayGet(pLast, iCol);
    if (keyTs > tTsVal->ts) {
      STColumn *pTColumn = &pTSchema->columns[0];
@ -222,7 +256,7 @@ int32_t tsdbCacheInsertLastrow(SLRUCache *pCache, STsdb *pTsdb, tb_uid_t uid, ST
        SColVal *tColVal = &tTsVal1->colVal;

        SColVal colVal = {0};
-        tTSRowGetVal(row, pTSchema, iCol, &colVal);
+        tsdbRowGetColVal(row, pTSchema, iCol, &colVal);
        if (!COL_VAL_IS_NONE(&colVal)) {
          if (keyTs == tTsVal1->ts && !COL_VAL_IS_NONE(tColVal)) {
            invalidate = true;
@ -261,7 +295,7 @@ int32_t tsdbCacheInsertLastrow(SLRUCache *pCache, STsdb *pTsdb, tb_uid_t uid, ST
  return code;
 }

-int32_t tsdbCacheInsertLast(SLRUCache *pCache, tb_uid_t uid, STSRow *row, STsdb *pTsdb) {
+int32_t tsdbCacheInsertLast(SLRUCache *pCache, tb_uid_t uid, TSDBROW *row, STsdb *pTsdb) {
  int32_t code = 0;
  STSRow *cacheRow = NULL;
  char    key[32] = {0};
@ -272,13 +306,51 @@ int32_t tsdbCacheInsertLast(SLRUCache *pCache, tb_uid_t uid, STSRow *row, STsdb
  LRUHandle *h = taosLRUCacheLookup(pCache, key, keyLen);
  if (h) {
    STSchema *pTSchema = metaGetTbTSchema(pTsdb->pVnode->pMeta, uid, -1, 1);
-    TSKEY     keyTs = row->ts;
+    TSKEY     keyTs = TSDBROW_TS(row);
    bool      invalidate = false;

    SArray *pLast = (SArray *)taosLRUCacheValue(pCache, h);
    int16_t nCol = taosArrayGetSize(pLast);
    int16_t iCol = 0;

+    if (nCol <= 0) {
+      nCol = pTSchema->numOfCols;
+
+      STColumn *pTColumn = &pTSchema->columns[0];
+      SColVal   tColVal = COL_VAL_VALUE(pTColumn->colId, pTColumn->type, (SValue){.val = keyTs});
+      if (taosArrayPush(pLast, &(SLastCol){.ts = keyTs, .colVal = tColVal}) == NULL) {
+        terrno = TSDB_CODE_OUT_OF_MEMORY;
+        code = TSDB_CODE_OUT_OF_MEMORY;
+        goto _invalidate;
+      }
+
+      for (iCol = 1; iCol < nCol; ++iCol) {
+        SColVal colVal = {0};
+        tsdbRowGetColVal(row, pTSchema, iCol, &colVal);
+
+        SLastCol lastCol = {.ts = keyTs, .colVal = colVal};
+        if (IS_VAR_DATA_TYPE(colVal.type) && colVal.value.nData > 0) {
+          SLastCol *pLastCol = (SLastCol *)taosArrayGet(pLast, iCol);
+          taosMemoryFree(pLastCol->colVal.value.pData);
+
+          lastCol.colVal.value.pData = taosMemoryMalloc(colVal.value.nData);
+          if (lastCol.colVal.value.pData == NULL) {
+            terrno = TSDB_CODE_OUT_OF_MEMORY;
+            code = TSDB_CODE_OUT_OF_MEMORY;
+            goto _invalidate;
+          }
+          memcpy(lastCol.colVal.value.pData, colVal.value.pData, colVal.value.nData);
+        }
+
+        if (taosArrayPush(pLast, &lastCol) == NULL) {
+          code = TSDB_CODE_OUT_OF_MEMORY;
+          goto _invalidate;
+        }
+      }
+
+      goto _invalidate;
+    }
+
    SLastCol *tTsVal = (SLastCol *)taosArrayGet(pLast, iCol);
    if (keyTs > tTsVal->ts) {
      STColumn *pTColumn = &pTSchema->columns[0];
@ -293,7 +365,7 @@ int32_t tsdbCacheInsertLast(SLRUCache *pCache, tb_uid_t uid, STSRow *row, STsdb
        SColVal *tColVal = &tTsVal1->colVal;

        SColVal colVal = {0};
-        tTSRowGetVal(row, pTSchema, iCol, &colVal);
+        tsdbRowGetColVal(row, pTSchema, iCol, &colVal);
        if (COL_VAL_IS_VALUE(&colVal)) {
          if (keyTs == tTsVal1->ts && COL_VAL_IS_VALUE(tColVal)) {
            invalidate = true;
@ -617,7 +689,7 @@ static int32_t getNextRowFromFS(void *iter, TSDBROW **ppRow) {
      } else {
        // tBlockDataDestroy(&state->blockData, 1);
        if (state->pBlockData) {
-          tBlockDataDestroy(state->pBlockData, 1);
+          tBlockDataDestroy(state->pBlockData);
          state->pBlockData = NULL;
        }

@ -685,7 +757,8 @@ static int32_t getNextRowFromFS(void *iter, TSDBROW **ppRow) {
        tBlockDataReset(state->pBlockData);

        tMapDataGetItemByIdx(&state->blockMap, state->iBlock, &block, tGetDataBlk);
-        /* code = tsdbReadBlockData(state->pDataFReader, &state->blockIdx, &block, &state->blockData, NULL, NULL); */
+        /* code = tsdbReadBlockData(state->pDataFReader, &state->blockIdx, &block, &state->blockData, NULL, NULL);
+         */
        tBlockDataReset(state->pBlockData);
        TABLEID tid = {.suid = state->suid, .uid = state->uid};
        code = tBlockDataInit(state->pBlockData, &tid, state->pTSchema, NULL, 0);
@ -739,7 +812,7 @@ _err:
    state->aBlockIdx = NULL;
  }
  if (state->pBlockData) {
-    tBlockDataDestroy(state->pBlockData, 1);
+    tBlockDataDestroy(state->pBlockData);
    state->pBlockData = NULL;
  }

@ -766,7 +839,7 @@ int32_t clearNextRowFromFS(void *iter) {
  }
  if (state->pBlockData) {
    // tBlockDataDestroy(&state->blockData, 1);
-    tBlockDataDestroy(state->pBlockData, 1);
+    tBlockDataDestroy(state->pBlockData);
    state->pBlockData = NULL;
  }

@ -1219,12 +1292,12 @@ static int32_t mergeLastRow(tb_uid_t uid, STsdb *pTsdb, bool *dup, SArray **ppCo

  // build the result ts row here
  *dup = false;
-  if (taosArrayGetSize(pColArray) != nCol) {
-    *ppColArray = NULL;
-    taosArrayDestroy(pColArray);
-  } else {
-    *ppColArray = pColArray;
-  }
+  // if (taosArrayGetSize(pColArray) != nCol) {
+  //*ppColArray = NULL;
+  // taosArrayDestroy(pColArray);
+  //} else {
+  *ppColArray = pColArray;
+  //}

  nextRowIterClose(&iter);
  // taosMemoryFreeClear(pTSchema);
@ -1337,12 +1410,12 @@ static int32_t mergeLast(tb_uid_t uid, STsdb *pTsdb, SArray **ppLastArray, SCach
    }
  } while (setNoneCol);

-  if (taosArrayGetSize(pColArray) <= 0) {
-    *ppLastArray = NULL;
-    taosArrayDestroy(pColArray);
-  } else {
-    *ppLastArray = pColArray;
-  }
+  // if (taosArrayGetSize(pColArray) <= 0) {
+  //*ppLastArray = NULL;
+  // taosArrayDestroy(pColArray);
+  //} else {
+  *ppLastArray = pColArray;
+  //}

  nextRowIterClose(&iter);
  // taosMemoryFreeClear(pTSchema);
@ -1373,8 +1446,8 @@ int32_t tsdbCacheGetLastrowH(SLRUCache *pCache, tb_uid_t uid, SCacheRowsReader *
      SArray *pArray = NULL;
      bool    dup = false;  // which is always false for now
      code = mergeLastRow(uid, pTsdb, &dup, &pArray, pr);
-      // if table's empty or error, return code of -1
-      if (code < 0 || pArray == NULL) {
+      // if table's empty or error, set handle NULL and return
+      if (code < 0 /* || pArray == NULL*/) {
        if (!dup && pArray) {
          taosArrayDestroy(pArray);
        }
@ -1392,13 +1465,8 @@ int32_t tsdbCacheGetLastrowH(SLRUCache *pCache, tb_uid_t uid, SCacheRowsReader *
      if (status != TAOS_LRU_STATUS_OK) {
        code = -1;
      }
-
-      // taosThreadMutexUnlock(&pTsdb->lruMutex);
-
-      // h = taosLRUCacheLookup(pCache, key, keyLen);
-    }  // else {
+    }
    taosThreadMutexUnlock(&pTsdb->lruMutex);
-    //}
  }

  *handle = h;
@ -1422,8 +1490,8 @@ int32_t tsdbCacheGetLastH(SLRUCache *pCache, tb_uid_t uid, SCacheRowsReader *pr,
    if (!h) {
      SArray *pLastArray = NULL;
      code = mergeLast(uid, pTsdb, &pLastArray, pr);
-      // if table's empty or error, return code of -1
-      if (code < 0 || pLastArray == NULL) {
+      // if table's empty or error, set handle NULL and return
+      if (code < 0 /* || pLastArray == NULL*/) {
        taosThreadMutexUnlock(&pTsdb->lruMutex);

        *handle = NULL;
@ -1437,13 +1505,8 @@ int32_t tsdbCacheGetLastH(SLRUCache *pCache, tb_uid_t uid, SCacheRowsReader *pr,
      if (status != TAOS_LRU_STATUS_OK) {
        code = -1;
      }
-
-      // taosThreadMutexUnlock(&pTsdb->lruMutex);
-
-      // h = taosLRUCacheLookup(pCache, key, keyLen);
-    }  // else {
+    }
    taosThreadMutexUnlock(&pTsdb->lruMutex);
-    //}
  }

  *handle = h;
--- a/source/dnode/vnode/src/tsdb/tsdbCacheRead.c
+++ b/source/dnode/vnode/src/tsdb/tsdbCacheRead.c
@ -258,6 +258,10 @@ int32_t tsdbRetrieveCacheRows(void* pReader, SSDataBlock* pResBlock, const int32
      if (h == NULL) {
        continue;
      }
+      if (taosArrayGetSize(pRow) <= 0) {
+        tsdbCacheRelease(lruCache, h);
+        continue;
+      }

      {
        for (int32_t k = 0; k < pr->numOfCols; ++k) {
@ -327,6 +331,10 @@ int32_t tsdbRetrieveCacheRows(void* pReader, SSDataBlock* pResBlock, const int32
      if (h == NULL) {
        continue;
      }
+      if (taosArrayGetSize(pRow) <= 0) {
+        tsdbCacheRelease(lruCache, h);
+        continue;
+      }

      saveOneRow(pRow, pResBlock, pr, slotIds, pRes, pr->idstr);
      // TODO reset the pRes
--- a/source/dnode/vnode/src/tsdb/tsdbCommit.c
+++ b/source/dnode/vnode/src/tsdb/tsdbCommit.c
@ -913,24 +913,24 @@ static void tsdbCommitDataEnd(SCommitter *pCommitter) {
  // reader
  taosArrayDestroy(pCommitter->dReader.aBlockIdx);
  tMapDataClear(&pCommitter->dReader.mBlock);
-  tBlockDataDestroy(&pCommitter->dReader.bData, 1);
+  tBlockDataDestroy(&pCommitter->dReader.bData);

  // merger
  for (int32_t iStt = 0; iStt < TSDB_MAX_STT_TRIGGER; iStt++) {
    SDataIter *pIter = &pCommitter->aDataIter[iStt];
    taosArrayDestroy(pIter->aSttBlk);
-    tBlockDataDestroy(&pIter->bData, 1);
+    tBlockDataDestroy(&pIter->bData);
  }

  // writer
  taosArrayDestroy(pCommitter->dWriter.aBlockIdx);
  taosArrayDestroy(pCommitter->dWriter.aSttBlk);
  tMapDataClear(&pCommitter->dWriter.mBlock);
-  tBlockDataDestroy(&pCommitter->dWriter.bData, 1);
+  tBlockDataDestroy(&pCommitter->dWriter.bData);
 #if USE_STREAM_COMPRESSION
  tDiskDataBuilderDestroy(pCommitter->dWriter.pBuilder);
 #else
-  tBlockDataDestroy(&pCommitter->dWriter.bDatal, 1);
+  tBlockDataDestroy(&pCommitter->dWriter.bDatal);
 #endif
  tDestroyTSchema(pCommitter->skmTable.pTSchema);
  tDestroyTSchema(pCommitter->skmRow.pTSchema);
@ -1183,7 +1183,6 @@ static int32_t tsdbCommitAheadBlock(SCommitter *pCommitter, SDataBlk *pDataBlk)

  tBlockDataClear(pBlockData);
  while (pRowInfo) {
-    ASSERT(pRowInfo->row.type == 0);
    code = tsdbCommitterUpdateRowSchema(pCommitter, id.suid, id.uid, TSDBROW_SVERSION(&pRowInfo->row));
    TSDB_CHECK_CODE(code, lino, _exit);

@ -1251,7 +1250,6 @@ static int32_t tsdbCommitMergeBlock(SCommitter *pCommitter, SDataBlk *pDataBlk)
        pRow = NULL;
      }
    } else if (c > 0) {
-      ASSERT(pRowInfo->row.type == 0);
      code = tsdbCommitterUpdateRowSchema(pCommitter, id.suid, id.uid, TSDBROW_SVERSION(&pRowInfo->row));
      TSDB_CHECK_CODE(code, lino, _exit);

@ -1493,7 +1491,7 @@ static int32_t tsdbCommitTableData(SCommitter *pCommitter, TABLEID id) {

    while (pRowInfo) {
      STSchema *pTSchema = NULL;
-      if (pRowInfo->row.type == 0) {
+      if (pRowInfo->row.type == TSDBROW_ROW_FMT) {
        code = tsdbCommitterUpdateRowSchema(pCommitter, id.suid, id.uid, TSDBROW_SVERSION(&pRowInfo->row));
        TSDB_CHECK_CODE(code, lino, _exit);
        pTSchema = pCommitter->skmRow.pTSchema;
@ -1536,7 +1534,7 @@ static int32_t tsdbCommitTableData(SCommitter *pCommitter, TABLEID id) {

    while (pRowInfo) {
      STSchema *pTSchema = NULL;
-      if (pRowInfo->row.type == 0) {
+      if (pRowInfo->row.type == TSDBROW_ROW_FMT) {
        code = tsdbCommitterUpdateRowSchema(pCommitter, id.suid, id.uid, TSDBROW_SVERSION(&pRowInfo->row));
        TSDB_CHECK_CODE(code, lino, _exit);
        pTSchema = pCommitter->skmRow.pTSchema;
--- a/source/dnode/vnode/src/tsdb/tsdbDiskData.c
+++ b/source/dnode/vnode/src/tsdb/tsdbDiskData.c
@ -596,7 +596,7 @@ int32_t tDiskDataAddRow(SDiskDataBuilder *pBuilder, TSDBROW *pRow, STSchema *pTS
  if (pBuilder->bi.maxKey < kRow.ts) pBuilder->bi.maxKey = kRow.ts;

  STSDBRowIter iter = {0};
-  tsdbRowIterInit(&iter, pRow, pTSchema);
+  tsdbRowIterOpen(&iter, pRow, pTSchema);

  SColVal *pColVal = tsdbRowIterNext(&iter);
  for (int32_t iBuilder = 0; iBuilder < pBuilder->nBuilder; iBuilder++) {
--- a/source/dnode/vnode/src/tsdb/tsdbMemTable.c
+++ b/source/dnode/vnode/src/tsdb/tsdbMemTable.c
@ -28,8 +28,10 @@

 static void    tbDataMovePosTo(STbData *pTbData, SMemSkipListNode **pos, TSDBKEY *pKey, int32_t flags);
 static int32_t tsdbGetOrCreateTbData(SMemTable *pMemTable, tb_uid_t suid, tb_uid_t uid, STbData **ppTbData);
-static int32_t tsdbInsertTableDataImpl(SMemTable *pMemTable, STbData *pTbData, int64_t version,
-                                       SSubmitMsgIter *pMsgIter, SSubmitBlk *pBlock, SSubmitBlkRsp *pRsp);
+static int32_t tsdbInsertRowDataToTable(SMemTable *pMemTable, STbData *pTbData, int64_t version,
+                                        SSubmitTbData *pSubmitTbData, int32_t *affectedRows);
+static int32_t tsdbInsertColDataToTable(SMemTable *pMemTable, STbData *pTbData, int64_t version,
+                                        SSubmitTbData *pSubmitTbData, int32_t *affectedRows);

 int32_t tsdbMemTableCreate(STsdb *pTsdb, SMemTable **ppMemTable) {
  int32_t    code = 0;
@ -95,14 +97,14 @@ STbData *tsdbGetTbDataFromMemTable(SMemTable *pMemTable, tb_uid_t suid, tb_uid_t
  return pTbData;
 }

-int32_t tsdbInsertTableData(STsdb *pTsdb, int64_t version, SSubmitMsgIter *pMsgIter, SSubmitBlk *pBlock,
-                            SSubmitBlkRsp *pRsp) {
+int32_t tsdbInsertTableData(STsdb *pTsdb, int64_t version, SSubmitTbData *pSubmitTbData, int32_t *affectedRows) {
  int32_t    code = 0;
  SMemTable *pMemTable = pTsdb->mem;
  STbData   *pTbData = NULL;
-  tb_uid_t   suid = pMsgIter->suid;
-  tb_uid_t   uid = pMsgIter->uid;
+  tb_uid_t   suid = pSubmitTbData->suid;
+  tb_uid_t   uid = pSubmitTbData->uid;

+#if 0
  SMetaInfo info;
  code = metaGetInfo(pTsdb->pVnode->pMeta, uid, &info, NULL);
  if (code) {
@ -116,14 +118,15 @@ int32_t tsdbInsertTableData(STsdb *pTsdb, int64_t version, SSubmitMsgIter *pMsgI
  if (info.suid) {
    metaGetInfo(pTsdb->pVnode->pMeta, info.suid, &info, NULL);
  }
-  if (pMsgIter->sversion != info.skmVer) {
+  if (pSubmitTbData->sver != info.skmVer) {
    tsdbError("vgId:%d, req sver:%d, skmVer:%d suid:%" PRId64 " uid:%" PRId64, TD_VID(pTsdb->pVnode),
-              pMsgIter->sversion, info.skmVer, suid, uid);
+              pSubmitTbData->sver, info.skmVer, suid, uid);
    code = TSDB_CODE_TDB_INVALID_TABLE_SCHEMA_VER;
    goto _err;
  }

-  pRsp->sver = info.skmVer;
+  if (pRsp) pRsp->sver = info.skmVer;
+#endif

  // create/get STbData to op
  code = tsdbGetOrCreateTbData(pMemTable, suid, uid, &pTbData);
@ -132,10 +135,12 @@ int32_t tsdbInsertTableData(STsdb *pTsdb, int64_t version, SSubmitMsgIter *pMsgI
  }

  // do insert impl
-  code = tsdbInsertTableDataImpl(pMemTable, pTbData, version, pMsgIter, pBlock, pRsp);
-  if (code) {
-    goto _err;
+  if (pSubmitTbData->flags & SUBMIT_REQ_COLUMN_DATA_FORMAT) {
+    code = tsdbInsertColDataToTable(pMemTable, pTbData, version, pSubmitTbData, affectedRows);
+  } else {
+    code = tsdbInsertRowDataToTable(pMemTable, pTbData, version, pSubmitTbData, affectedRows);
  }
+  if (code) goto _err;

  return code;

@ -242,7 +247,6 @@ void tsdbTbDataIterOpen(STbData *pTbData, TSDBKEY *pFrom, int8_t backward, STbDa
  pIter->pTbData = pTbData;
  pIter->backward = backward;
  pIter->pRow = NULL;
-  pIter->row.type = 0;
  if (pFrom == NULL) {
    // create from head or tail
    if (backward) {
@ -410,8 +414,13 @@ static void tbDataMovePosTo(STbData *pTbData, SMemSkipListNode **pos, TSDBKEY *p
      for (int8_t iLevel = pTbData->sl.level - 1; iLevel >= 0; iLevel--) {
        pn = SL_NODE_BACKWARD(px, iLevel);
        while (pn != pTbData->sl.pHead) {
-          tKey.version = pn->version;
-          tKey.ts = pn->pTSRow->ts;
+          if (pn->flag == TSDBROW_ROW_FMT) {
+            tKey.version = pn->version;
+            tKey.ts = ((SRow *)pn->pData)->ts;
+          } else if (pn->flag == TSDBROW_COL_FMT) {
+            tKey.version = ((SBlockData *)pn->pData)->aVersion[pn->iRow];
+            tKey.ts = ((SBlockData *)pn->pData)->aTSKEY[pn->iRow];
+          }

          int32_t c = tsdbKeyCmprFn(&tKey, pKey);
          if (c <= 0) {
@ -440,8 +449,13 @@ static void tbDataMovePosTo(STbData *pTbData, SMemSkipListNode **pos, TSDBKEY *p
      for (int8_t iLevel = pTbData->sl.level - 1; iLevel >= 0; iLevel--) {
        pn = SL_NODE_FORWARD(px, iLevel);
        while (pn != pTbData->sl.pTail) {
-          tKey.version = pn->version;
-          tKey.ts = pn->pTSRow->ts;
+          if (pn->flag == TSDBROW_ROW_FMT) {
+            tKey.version = pn->version;
+            tKey.ts = ((SRow *)pn->pData)->ts;
+          } else if (pn->flag == TSDBROW_COL_FMT) {
+            tKey.version = ((SBlockData *)pn->pData)->aVersion[pn->iRow];
+            tKey.ts = ((SBlockData *)pn->pData)->aTSKEY[pn->iRow];
+          }

          int32_t c = tsdbKeyCmprFn(&tKey, pKey);
          if (c >= 0) {
@ -468,29 +482,39 @@ static FORCE_INLINE int8_t tsdbMemSkipListRandLevel(SMemSkipList *pSl) {

  return level;
 }
-static int32_t tbDataDoPut(SMemTable *pMemTable, STbData *pTbData, SMemSkipListNode **pos, int64_t version,
-                           STSRow *pRow, int8_t forward) {
+static int32_t tbDataDoPut(SMemTable *pMemTable, STbData *pTbData, SMemSkipListNode **pos, TSDBROW *pRow,
+                           int8_t forward) {
  int32_t           code = 0;
  int8_t            level;
  SMemSkipListNode *pNode;
  SVBufPool        *pPool = pMemTable->pTsdb->pVnode->inUse;

+  ASSERT(pPool != NULL);
+
  // node
  level = tsdbMemSkipListRandLevel(&pTbData->sl);
-  ASSERT(pPool != NULL);
  pNode = (SMemSkipListNode *)vnodeBufPoolMalloc(pPool, SL_NODE_SIZE(level));
  if (pNode == NULL) {
    code = TSDB_CODE_OUT_OF_MEMORY;
    goto _exit;
  }
  pNode->level = level;
-  pNode->version = version;
-  pNode->pTSRow = vnodeBufPoolMalloc(pPool, pRow->len);
-  if (NULL == pNode->pTSRow) {
-    code = TSDB_CODE_OUT_OF_MEMORY;
-    goto _exit;
+  pNode->flag = pRow->type;
+
+  if (pRow->type == TSDBROW_ROW_FMT) {
+    pNode->version = pRow->version;
+    pNode->pData = vnodeBufPoolMalloc(pPool, pRow->pTSRow->len);
+    if (NULL == pNode->pData) {
+      code = TSDB_CODE_OUT_OF_MEMORY;
+      goto _exit;
+    }
+    memcpy(pNode->pData, pRow->pTSRow, pRow->pTSRow->len);
+  } else if (pRow->type == TSDBROW_COL_FMT) {
+    pNode->iRow = pRow->iRow;
+    pNode->pData = pRow->pBlockData;
+  } else {
+    ASSERT(0);
  }
-  memcpy(pNode->pTSRow, pRow, pRow->len);

  for (int8_t iLevel = level - 1; iLevel >= 0; iLevel--) {
    SMemSkipListNode *pn = pos[iLevel];
@ -537,69 +561,166 @@ _exit:
  return code;
 }

-static int32_t tsdbInsertTableDataImpl(SMemTable *pMemTable, STbData *pTbData, int64_t version,
-                                       SSubmitMsgIter *pMsgIter, SSubmitBlk *pBlock, SSubmitBlkRsp *pRsp) {
-  int32_t           code = 0;
-  SSubmitBlkIter    blkIter = {0};
-  TSDBKEY           key = {.version = version};
-  SMemSkipListNode *pos[SL_MAX_LEVEL];
-  TSDBROW           row = tsdbRowFromTSRow(version, NULL);
-  int32_t           nRow = 0;
-  STSRow           *pLastRow = NULL;
+static int32_t tsdbInsertColDataToTable(SMemTable *pMemTable, STbData *pTbData, int64_t version,
+                                        SSubmitTbData *pSubmitTbData, int32_t *affectedRows) {
+  int32_t code = 0;

-  tInitSubmitBlkIter(pMsgIter, pBlock, &blkIter);
+  SVBufPool *pPool = pMemTable->pTsdb->pVnode->inUse;
+  int32_t    nColData = TARRAY_SIZE(pSubmitTbData->aCol);
+  SColData  *aColData = (SColData *)TARRAY_DATA(pSubmitTbData->aCol);

-  // backward put first data
-  row.pTSRow = tGetSubmitBlkNext(&blkIter);
-  if (row.pTSRow == NULL) return code;
+  ASSERT(aColData[0].cid == PRIMARYKEY_TIMESTAMP_COL_ID);
+  ASSERT(aColData[0].type == TSDB_DATA_TYPE_TIMESTAMP);
+  ASSERT(aColData[0].flag == HAS_VALUE);

-  key.ts = row.pTSRow->ts;
-  nRow++;
-  tbDataMovePosTo(pTbData, pos, &key, SL_MOVE_BACKWARD);
-  code = tbDataDoPut(pMemTable, pTbData, pos, version, row.pTSRow, 0);
-  if (code) {
-    goto _err;
+  // copy and construct block data
+  SBlockData *pBlockData = vnodeBufPoolMalloc(pPool, sizeof(*pBlockData));
+  if (pBlockData == NULL) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+    goto _exit;
  }

+  pBlockData->suid = pTbData->suid;
+  pBlockData->uid = pTbData->uid;
+  pBlockData->nRow = aColData[0].nVal;
+  pBlockData->aUid = NULL;
+  pBlockData->aVersion = vnodeBufPoolMalloc(pPool, aColData[0].nData);
+  if (pBlockData->aVersion == NULL) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+    goto _exit;
+  }
+  for (int32_t i = 0; i < pBlockData->nRow; i++) {  // todo: here can be optimized
+    pBlockData->aVersion[i] = version;
+  }
+
+  pBlockData->aTSKEY = vnodeBufPoolMalloc(pPool, aColData[0].nData);
+  if (pBlockData->aTSKEY == NULL) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+    goto _exit;
+  }
+  memcpy(pBlockData->aTSKEY, aColData[0].pData, aColData[0].nData);
+
+  pBlockData->nColData = nColData - 1;
+  pBlockData->aColData = vnodeBufPoolMalloc(pPool, sizeof(SColData) * pBlockData->nColData);
+  if (pBlockData->aColData == NULL) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+  }
+  for (int32_t iColData = 0; iColData < pBlockData->nColData; ++iColData) {
+    code = tColDataCopy(&aColData[iColData + 1], &pBlockData->aColData[iColData], (xMallocFn)vnodeBufPoolMalloc, pPool);
+    if (code) goto _exit;
+  }
+
+  // loop to add each row to the skiplist
+  SMemSkipListNode *pos[SL_MAX_LEVEL];
+  TSDBROW           tRow = tsdbRowFromBlockData(pBlockData, 0);
+  TSDBKEY           key = {.version = version, .ts = pBlockData->aTSKEY[0]};
+  TSDBROW           lRow;  // last row
+
+  // first row
+  tbDataMovePosTo(pTbData, pos, &key, SL_MOVE_BACKWARD);
+  if ((code = tbDataDoPut(pMemTable, pTbData, pos, &tRow, 0))) goto _exit;
  pTbData->minKey = TMIN(pTbData->minKey, key.ts);
+  lRow = tRow;

-  pLastRow = row.pTSRow;
-
-  // forward put rest data
-  row.pTSRow = tGetSubmitBlkNext(&blkIter);
-  if (row.pTSRow) {
+  // remain row
+  ++tRow.iRow;
+  if (tRow.iRow < pBlockData->nRow) {
    for (int8_t iLevel = pos[0]->level; iLevel < pTbData->sl.maxLevel; iLevel++) {
      pos[iLevel] = SL_NODE_BACKWARD(pos[iLevel], iLevel);
    }
-    do {
-      key.ts = row.pTSRow->ts;
-      nRow++;
+
+    while (tRow.iRow < pBlockData->nRow) {
+      key.ts = pBlockData->aTSKEY[tRow.iRow];
+
      if (SL_NODE_FORWARD(pos[0], 0) != pTbData->sl.pTail) {
        tbDataMovePosTo(pTbData, pos, &key, SL_MOVE_FROM_POS);
      }
-      code = tbDataDoPut(pMemTable, pTbData, pos, version, row.pTSRow, 1);
-      if (code) {
-        goto _err;
-      }

-      pLastRow = row.pTSRow;
+      if ((code = tbDataDoPut(pMemTable, pTbData, pos, &tRow, 1))) goto _exit;
+      lRow = tRow;

-      row.pTSRow = tGetSubmitBlkNext(&blkIter);
-    } while (row.pTSRow);
+      ++tRow.iRow;
+    }
  }

  if (key.ts >= pTbData->maxKey) {
-    if (key.ts > pTbData->maxKey) {
-      pTbData->maxKey = key.ts;
-    }
+    pTbData->maxKey = key.ts;

-    if (TSDB_CACHE_LAST_ROW(pMemTable->pTsdb->pVnode->config) && pLastRow != NULL) {
-      tsdbCacheInsertLastrow(pMemTable->pTsdb->lruCache, pMemTable->pTsdb, pTbData->uid, pLastRow, true);
+    if (TSDB_CACHE_LAST_ROW(pMemTable->pTsdb->pVnode->config)) {
+      tsdbCacheInsertLastrow(pMemTable->pTsdb->lruCache, pMemTable->pTsdb, pTbData->uid, &lRow, true);
    }
  }

  if (TSDB_CACHE_LAST(pMemTable->pTsdb->pVnode->config)) {
-    tsdbCacheInsertLast(pMemTable->pTsdb->lruCache, pTbData->uid, pLastRow, pMemTable->pTsdb);
+    tsdbCacheInsertLast(pMemTable->pTsdb->lruCache, pTbData->uid, &lRow, pMemTable->pTsdb);
+  }
+
+  // SMemTable
+  pMemTable->minKey = TMIN(pMemTable->minKey, pTbData->minKey);
+  pMemTable->maxKey = TMAX(pMemTable->maxKey, pTbData->maxKey);
+  pMemTable->nRow += pBlockData->nRow;
+
+  if (affectedRows) *affectedRows = pBlockData->nRow;
+
+_exit:
+  return code;
+}
+
+static int32_t tsdbInsertRowDataToTable(SMemTable *pMemTable, STbData *pTbData, int64_t version,
+                                        SSubmitTbData *pSubmitTbData, int32_t *affectedRows) {
+  int32_t code = 0;
+
+  int32_t           nRow = TARRAY_SIZE(pSubmitTbData->aRowP);
+  SRow            **aRow = (SRow **)TARRAY_DATA(pSubmitTbData->aRowP);
+  TSDBKEY           key = {.version = version};
+  SMemSkipListNode *pos[SL_MAX_LEVEL];
+  TSDBROW           tRow = {.type = TSDBROW_ROW_FMT, .version = version};
+  int32_t           iRow = 0;
+  TSDBROW           lRow;
+
+  // backward put first data
+  tRow.pTSRow = aRow[iRow++];
+  key.ts = tRow.pTSRow->ts;
+  tbDataMovePosTo(pTbData, pos, &key, SL_MOVE_BACKWARD);
+  code = tbDataDoPut(pMemTable, pTbData, pos, &tRow, 0);
+  if (code) goto _exit;
+  lRow = tRow;
+
+  pTbData->minKey = TMIN(pTbData->minKey, key.ts);
+
+  // forward put rest data
+  if (iRow < nRow) {
+    for (int8_t iLevel = pos[0]->level; iLevel < pTbData->sl.maxLevel; iLevel++) {
+      pos[iLevel] = SL_NODE_BACKWARD(pos[iLevel], iLevel);
+    }
+
+    while (iRow < nRow) {
+      tRow.pTSRow = aRow[iRow];
+      key.ts = tRow.pTSRow->ts;
+
+      if (SL_NODE_FORWARD(pos[0], 0) != pTbData->sl.pTail) {
+        tbDataMovePosTo(pTbData, pos, &key, SL_MOVE_FROM_POS);
+      }
+
+      code = tbDataDoPut(pMemTable, pTbData, pos, &tRow, 1);
+      if (code) goto _exit;
+
+      lRow = tRow;
+
+      iRow++;
+    }
+  }
+
+  if (key.ts >= pTbData->maxKey) {
+    pTbData->maxKey = key.ts;
+
+    if (TSDB_CACHE_LAST_ROW(pMemTable->pTsdb->pVnode->config)) {
+      tsdbCacheInsertLastrow(pMemTable->pTsdb->lruCache, pMemTable->pTsdb, pTbData->uid, &lRow, true);
+    }
+  }
+
+  if (TSDB_CACHE_LAST(pMemTable->pTsdb->pVnode->config)) {
+    tsdbCacheInsertLast(pMemTable->pTsdb->lruCache, pTbData->uid, &lRow, pMemTable->pTsdb);
  }

  // SMemTable
@ -607,12 +728,9 @@ static int32_t tsdbInsertTableDataImpl(SMemTable *pMemTable, STbData *pTbData, i
  pMemTable->maxKey = TMAX(pMemTable->maxKey, pTbData->maxKey);
  pMemTable->nRow += nRow;

-  pRsp->numOfRows = nRow;
-  pRsp->affectedRows = nRow;
+  if (affectedRows) *affectedRows = nRow;

-  return code;
-
-_err:
+_exit:
  return code;
 }

--- a/source/dnode/vnode/src/tsdb/tsdbMergeTree.c
+++ b/source/dnode/vnode/src/tsdb/tsdbMergeTree.c
@ -89,8 +89,8 @@ void *destroyLastBlockLoadInfo(SSttBlockLoadInfo *pLoadInfo) {
    pLoadInfo[i].blockIndex[0] = -1;
    pLoadInfo[i].blockIndex[1] = -1;

-    tBlockDataDestroy(&pLoadInfo[i].blockData[0], true);
-    tBlockDataDestroy(&pLoadInfo[i].blockData[1], true);
+    tBlockDataDestroy(&pLoadInfo[i].blockData[0]);
+    tBlockDataDestroy(&pLoadInfo[i].blockData[1]);

    taosArrayDestroy(pLoadInfo[i].aSttBlk);
  }
@ -567,7 +567,6 @@ int32_t tMergeTreeOpen(SMergeTree *pMTree, int8_t backward, SDataFReader *pFRead

  pMTree->pLoadInfo = pBlockLoadInfo;
  pMTree->destroyLoadInfo = destroyLoadInfo;
-  ASSERT(pMTree->pLoadInfo != NULL);

  for (int32_t i = 0; i < pFReader->pSet->nSttF; ++i) {  // open all last file
    struct SLDataIter *pIter = NULL;
--- a/source/dnode/vnode/src/tsdb/tsdbRead.c
+++ b/source/dnode/vnode/src/tsdb/tsdbRead.c
--- a/source/dnode/vnode/src/tsdb/tsdbSnapshot.c
+++ b/source/dnode/vnode/src/tsdb/tsdbSnapshot.c
@ -538,7 +538,7 @@ _exit:
    if (pReader) {
      taosArrayDestroy(pReader->aDelData);
      taosArrayDestroy(pReader->aDelIdx);
-      tBlockDataDestroy(&pReader->bData, 1);
+      tBlockDataDestroy(&pReader->bData);
      tsdbFSDestroy(&pReader->fs);
      taosMemoryFree(pReader);
    }
@ -565,10 +565,10 @@ int32_t tsdbSnapReaderClose(STsdbSnapReader** ppReader) {
      taosArrayDestroy(pIter->aSttBlk);
    }

-    tBlockDataDestroy(&pIter->bData, 1);
+    tBlockDataDestroy(&pIter->bData);
  }

-  tBlockDataDestroy(&pReader->bData, 1);
+  tBlockDataDestroy(&pReader->bData);
  tDestroyTSchema(pReader->skmTable.pTSchema);

  // del
@ -1359,13 +1359,13 @@ _exit:
      if (pWriter->aDelIdxW) taosArrayDestroy(pWriter->aDelIdxW);
      if (pWriter->aDelData) taosArrayDestroy(pWriter->aDelData);
      if (pWriter->aDelIdxR) taosArrayDestroy(pWriter->aDelIdxR);
-      tBlockDataDestroy(&pWriter->dWriter.sData, 1);
-      tBlockDataDestroy(&pWriter->dWriter.bData, 1);
+      tBlockDataDestroy(&pWriter->dWriter.sData);
+      tBlockDataDestroy(&pWriter->dWriter.bData);
      if (pWriter->dWriter.aSttBlk) taosArrayDestroy(pWriter->dWriter.aSttBlk);
      if (pWriter->dWriter.aBlockIdx) taosArrayDestroy(pWriter->dWriter.aBlockIdx);
-      tBlockDataDestroy(&pWriter->dReader.bData, 1);
+      tBlockDataDestroy(&pWriter->dReader.bData);
      if (pWriter->dReader.aBlockIdx) taosArrayDestroy(pWriter->dReader.aBlockIdx);
-      tBlockDataDestroy(&pWriter->bData, 1);
+      tBlockDataDestroy(&pWriter->bData);
      tsdbFSDestroy(&pWriter->fs);
      taosMemoryFree(pWriter);
    }
@ -1425,18 +1425,18 @@ int32_t tsdbSnapWriterClose(STsdbSnapWriter** ppWriter, int8_t rollback) {
  // SNAP_DATA_TSDB

  // Writer
-  tBlockDataDestroy(&pWriter->dWriter.sData, 1);
-  tBlockDataDestroy(&pWriter->dWriter.bData, 1);
+  tBlockDataDestroy(&pWriter->dWriter.sData);
+  tBlockDataDestroy(&pWriter->dWriter.bData);
  taosArrayDestroy(pWriter->dWriter.aSttBlk);
  tMapDataClear(&pWriter->dWriter.mDataBlk);
  taosArrayDestroy(pWriter->dWriter.aBlockIdx);

  // Reader
-  tBlockDataDestroy(&pWriter->dReader.bData, 1);
+  tBlockDataDestroy(&pWriter->dReader.bData);
  tMapDataClear(&pWriter->dReader.mDataBlk);
  taosArrayDestroy(pWriter->dReader.aBlockIdx);

-  tBlockDataDestroy(&pWriter->bData, 1);
+  tBlockDataDestroy(&pWriter->bData);
  tDestroyTSchema(pWriter->skmTable.pTSchema);

  for (int32_t iBuf = 0; iBuf < sizeof(pWriter->aBuf) / sizeof(uint8_t*); iBuf++) {
--- a/source/dnode/vnode/src/tsdb/tsdbUtil.c
+++ b/source/dnode/vnode/src/tsdb/tsdbUtil.c
@ -23,7 +23,7 @@ void tMapDataReset(SMapData *pMapData) {
 }

 void tMapDataClear(SMapData *pMapData) {
-  tFree((uint8_t *)pMapData->aOffset);
+  tFree(pMapData->aOffset);
  tFree(pMapData->pData);
  pMapData->pData = NULL;
  pMapData->aOffset = NULL;
@ -572,9 +572,9 @@ void tsdbRowGetColVal(TSDBROW *pRow, STSchema *pTSchema, int32_t iCol, SColVal *

  ASSERT(iCol > 0);

-  if (pRow->type == 0) {
-    tTSRowGetVal(pRow->pTSRow, pTSchema, iCol, pColVal);
-  } else if (pRow->type == 1) {
+  if (pRow->type == TSDBROW_ROW_FMT) {
+    tRowGet(pRow->pTSRow, pTSchema, iCol, pColVal);
+  } else if (pRow->type == TSDBROW_COL_FMT) {
    SColData *pColData;

    tBlockDataGetColData(pRow->pBlockData, pTColumn->colId, &pColData);
@ -589,60 +589,61 @@ void tsdbRowGetColVal(TSDBROW *pRow, STSchema *pTSchema, int32_t iCol, SColVal *
  }
 }

-// int32_t tPutTSDBRow(uint8_t *p, TSDBROW *pRow) {
-//   int32_t n = 0;
-
-//   n += tPutI64(p, pRow->version);
-//   if (p) memcpy(p + n, pRow->pTSRow, pRow->pTSRow->len);
-//   n += pRow->pTSRow->len;
-
-//   return n;
-// }
-
 int32_t tsdbRowCmprFn(const void *p1, const void *p2) {
  return tsdbKeyCmprFn(&TSDBROW_KEY((TSDBROW *)p1), &TSDBROW_KEY((TSDBROW *)p2));
 }

 // STSDBRowIter ======================================================
-void tsdbRowIterInit(STSDBRowIter *pIter, TSDBROW *pRow, STSchema *pTSchema) {
+int32_t tsdbRowIterOpen(STSDBRowIter *pIter, TSDBROW *pRow, STSchema *pTSchema) {
+  int32_t code = 0;
+
  pIter->pRow = pRow;
-  if (pRow->type == 0) {
-    ASSERT(pTSchema);
-    pIter->pTSchema = pTSchema;
-    pIter->i = 1;
-  } else if (pRow->type == 1) {
-    pIter->pTSchema = NULL;
-    pIter->i = 0;
+  if (pRow->type == TSDBROW_ROW_FMT) {
+    code = tRowIterOpen(pRow->pTSRow, pTSchema, &pIter->pIter);
+    if (code) goto _exit;
+  } else if (pRow->type == TSDBROW_COL_FMT) {
+    pIter->iColData = 0;
  } else {
    ASSERT(0);
  }
+
+_exit:
+  return code;
+}
+
+void tsdbRowClose(STSDBRowIter *pIter) {
+  if (pIter->pRow->type == TSDBROW_ROW_FMT) {
+    tRowIterClose(&pIter->pIter);
+  }
 }

 SColVal *tsdbRowIterNext(STSDBRowIter *pIter) {
-  if (pIter->pRow->type == 0) {
-    if (pIter->i < pIter->pTSchema->numOfCols) {
-      tTSRowGetVal(pIter->pRow->pTSRow, pIter->pTSchema, pIter->i, &pIter->colVal);
-      pIter->i++;
+  if (pIter->pRow->type == TSDBROW_ROW_FMT) {
+    return tRowIterNext(pIter->pIter);
+  } else if (pIter->pRow->type == TSDBROW_COL_FMT) {
+    if (pIter->iColData == 0) {
+      pIter->cv = COL_VAL_VALUE(PRIMARYKEY_TIMESTAMP_COL_ID, TSDB_DATA_TYPE_TIMESTAMP,
+                                (SValue){.val = pIter->pRow->pBlockData->aTSKEY[pIter->pRow->iRow]});
+      ++pIter->iColData;
+      return &pIter->cv;
+    }

-      return &pIter->colVal;
+    if (pIter->iColData < pIter->pRow->pBlockData->nColData) {
+      tColDataGetValue(&pIter->pRow->pBlockData->aColData[pIter->iColData], pIter->pRow->iRow, &pIter->cv);
+      ++pIter->iColData;
+      return &pIter->cv;
+    } else {
+      return NULL;
    }
  } else {
-    if (pIter->i < pIter->pRow->pBlockData->nColData) {
-      SColData *pColData = tBlockDataGetColDataByIdx(pIter->pRow->pBlockData, pIter->i);
-
-      tColDataGetValue(pColData, pIter->pRow->iRow, &pIter->colVal);
-      pIter->i++;
-
-      return &pIter->colVal;
-    }
+    ASSERT(0);
+    return NULL;   // suppress error report by compiler
  }
-
-  return NULL;
 }

 // SRowMerger ======================================================

-int32_t tRowMergerInit2(SRowMerger *pMerger, STSchema *pResTSchema, TSDBROW *pRow, STSchema *pTSchema) {
+int32_t tsdbRowMergerInit2(SRowMerger *pMerger, STSchema *pResTSchema, TSDBROW *pRow, STSchema *pTSchema) {
  int32_t   code = 0;
  TSDBKEY   key = TSDBROW_KEY(pRow);
  SColVal  *pColVal = &(SColVal){0};
@ -697,7 +698,7 @@ _exit:
  return code;
 }

-int32_t tRowMergerAdd(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema) {
+int32_t tsdbRowMergerAdd(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema) {
  int32_t   code = 0;
  TSDBKEY   key = TSDBROW_KEY(pRow);
  SColVal  *pColVal = &(SColVal){0};
@ -736,7 +737,7 @@ int32_t tRowMergerAdd(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema) {
  return code;
 }

-int32_t tRowMergerInit(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema) {
+int32_t tsdbRowMergerInit(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema) {
  int32_t   code = 0;
  TSDBKEY   key = TSDBROW_KEY(pRow);
  SColVal  *pColVal = &(SColVal){0};
@ -775,9 +776,9 @@ _exit:
  return code;
 }

-void tRowMergerClear(SRowMerger *pMerger) { taosArrayDestroy(pMerger->pArray); }
+void tsdbRowMergerClear(SRowMerger *pMerger) { taosArrayDestroy(pMerger->pArray); }

-int32_t tRowMerge(SRowMerger *pMerger, TSDBROW *pRow) {
+int32_t tsdbRowMerge(SRowMerger *pMerger, TSDBROW *pRow) {
  int32_t  code = 0;
  TSDBKEY  key = TSDBROW_KEY(pRow);
  SColVal *pColVal = &(SColVal){0};
@ -807,12 +808,8 @@ _exit:
  return code;
 }

-int32_t tRowMergerGetRow(SRowMerger *pMerger, STSRow **ppRow) {
-  int32_t code = 0;
-
-  code = tdSTSRowNew(pMerger->pArray, pMerger->pTSchema, ppRow);
-
-  return code;
+int32_t tsdbRowMergerGetRow(SRowMerger *pMerger, SRow **ppRow) {
+  return tRowBuild(pMerger->pArray, pMerger->pTSchema, ppRow);
 }

 // delete skyline ======================================================
@ -931,27 +928,49 @@ int32_t tBlockDataCreate(SBlockData *pBlockData) {
  pBlockData->aVersion = NULL;
  pBlockData->aTSKEY = NULL;
  pBlockData->nColData = 0;
-  pBlockData->aColData = taosArrayInit(0, sizeof(SColData));
-  if (pBlockData->aColData == NULL) {
-    code = TSDB_CODE_OUT_OF_MEMORY;
-    goto _exit;
-  }
+  pBlockData->aColData = NULL;

 _exit:
  return code;
 }

-void tBlockDataDestroy(SBlockData *pBlockData, int8_t deepClear) {
-  tFree((uint8_t *)pBlockData->aUid);
-  tFree((uint8_t *)pBlockData->aVersion);
-  tFree((uint8_t *)pBlockData->aTSKEY);
-  taosArrayDestroyEx(pBlockData->aColData, deepClear ? tColDataDestroy : NULL);
-  pBlockData->aUid = NULL;
-  pBlockData->aVersion = NULL;
-  pBlockData->aTSKEY = NULL;
-  pBlockData->aColData = NULL;
+void tBlockDataDestroy(SBlockData *pBlockData) {
+  tFree(pBlockData->aUid);
+  tFree(pBlockData->aVersion);
+  tFree(pBlockData->aTSKEY);
+
+  for (int32_t i = 0; i < pBlockData->nColData; i++) {
+    tColDataDestroy(&pBlockData->aColData[i]);
+  }
+
+  if (pBlockData->aColData) {
+    taosMemoryFree(pBlockData->aColData);
+    pBlockData->aColData = NULL;
+  }
 }

+static int32_t tBlockDataAdjustColData(SBlockData *pBlockData, int32_t nColData) {
+  int32_t code = 0;
+
+  if (pBlockData->nColData > nColData) {
+    for (int32_t i = nColData; i < pBlockData->nColData; i++) {
+      tColDataDestroy(&pBlockData->aColData[i]);
+    }
+  } else if (pBlockData->nColData < nColData) {
+    SColData *aColData = taosMemoryRealloc(pBlockData->aColData, sizeof(SBlockData) * nColData);
+    if (aColData == NULL) {
+      code = TSDB_CODE_OUT_OF_MEMORY;
+      goto _exit;
+    }
+
+    pBlockData->aColData = aColData;
+    memset(&pBlockData->aColData[pBlockData->nColData], 0, sizeof(SBlockData) * (nColData - pBlockData->nColData));
+  }
+  pBlockData->nColData = nColData;
+
+_exit:
+  return code;
+}
 int32_t tBlockDataInit(SBlockData *pBlockData, TABLEID *pId, STSchema *pTSchema, int16_t *aCid, int32_t nCid) {
  int32_t code = 0;

@ -961,37 +980,35 @@ int32_t tBlockDataInit(SBlockData *pBlockData, TABLEID *pId, STSchema *pTSchema,
  pBlockData->uid = pId->uid;
  pBlockData->nRow = 0;

-  pBlockData->nColData = 0;
  if (aCid) {
+    code = tBlockDataAdjustColData(pBlockData, nCid);
+    if (code) goto _exit;
+
    int32_t   iColumn = 1;
    STColumn *pTColumn = &pTSchema->columns[iColumn];
    for (int32_t iCid = 0; iCid < nCid; iCid++) {
-      while (pTColumn && pTColumn->colId < aCid[iCid]) {
+      ASSERT(pTColumn);
+      while (pTColumn->colId < aCid[iCid]) {
        iColumn++;
-        pTColumn = (iColumn < pTSchema->numOfCols) ? &pTSchema->columns[iColumn] : NULL;
+        ASSERT(iColumn < pTSchema->numOfCols);
+        pTColumn = &pTSchema->columns[iColumn];
      }

-      if (pTColumn == NULL) {
-        break;
-      } else if (pTColumn->colId == aCid[iCid]) {
-        SColData *pColData;
-        code = tBlockDataAddColData(pBlockData, &pColData);
-        if (code) goto _exit;
-        tColDataInit(pColData, pTColumn->colId, pTColumn->type, (pTColumn->flags & COL_SMA_ON) ? 1 : 0);
+      ASSERT(pTColumn->colId == aCid[iCid]);
+      tColDataInit(&pBlockData->aColData[iCid], pTColumn->colId, pTColumn->type,
+                   (pTColumn->flags & COL_SMA_ON) ? 1 : 0);

-        iColumn++;
-        pTColumn = (iColumn < pTSchema->numOfCols) ? &pTSchema->columns[iColumn] : NULL;
-      }
+      iColumn++;
+      pTColumn = (iColumn < pTSchema->numOfCols) ? &pTSchema->columns[iColumn] : NULL;
    }
  } else {
-    for (int32_t iColumn = 1; iColumn < pTSchema->numOfCols; iColumn++) {
-      STColumn *pTColumn = &pTSchema->columns[iColumn];
+    code = tBlockDataAdjustColData(pBlockData, pTSchema->numOfCols - 1);
+    if (code) goto _exit;

-      SColData *pColData;
-      code = tBlockDataAddColData(pBlockData, &pColData);
-      if (code) goto _exit;
-
-      tColDataInit(pColData, pTColumn->colId, pTColumn->type, (pTColumn->flags & COL_SMA_ON) ? 1 : 0);
+    for (int32_t iColData = 0; iColData < pBlockData->nColData; iColData++) {
+      STColumn *pTColumn = &pTSchema->columns[iColData + 1];
+      tColDataInit(&pBlockData->aColData[iColData], pTColumn->colId, pTColumn->type,
+                   (pTColumn->flags & COL_SMA_ON) ? 1 : 0);
    }
  }

@ -1002,8 +1019,6 @@ _exit:
 void tBlockDataReset(SBlockData *pBlockData) {
  pBlockData->suid = 0;
  pBlockData->uid = 0;
-  pBlockData->nRow = 0;
-  pBlockData->nColData = 0;
 }

 void tBlockDataClear(SBlockData *pBlockData) {
@ -1011,49 +1026,22 @@ void tBlockDataClear(SBlockData *pBlockData) {

  pBlockData->nRow = 0;
  for (int32_t iColData = 0; iColData < pBlockData->nColData; iColData++) {
-    SColData *pColData = tBlockDataGetColDataByIdx(pBlockData, iColData);
-    tColDataClear(pColData);
+    tColDataClear(tBlockDataGetColDataByIdx(pBlockData, iColData));
  }
 }

-int32_t tBlockDataAddColData(SBlockData *pBlockData, SColData **ppColData) {
-  int32_t   code = 0;
-  SColData *pColData = NULL;
-
-  if (pBlockData->nColData >= taosArrayGetSize(pBlockData->aColData)) {
-    if (taosArrayPush(pBlockData->aColData, &((SColData){0})) == NULL) {
-      code = TSDB_CODE_OUT_OF_MEMORY;
-      goto _err;
-    }
-  }
-  pColData = (SColData *)taosArrayGet(pBlockData->aColData, pBlockData->nColData);
-
-  pBlockData->nColData++;
-
-  *ppColData = pColData;
-  return code;
-
-_err:
-  *ppColData = NULL;
-  return code;
-}
-
 static int32_t tBlockDataAppendBlockRow(SBlockData *pBlockData, SBlockData *pBlockDataFrom, int32_t iRow) {
  int32_t code = 0;

  SColVal   cv = {0};
  int32_t   iColDataFrom = 0;
-  SColData *pColDataFrom =
-      (iColDataFrom < pBlockDataFrom->nColData) ? &((SColData *)pBlockDataFrom->aColData->pData)[iColDataFrom] : NULL;
+  SColData *pColDataFrom = (iColDataFrom < pBlockDataFrom->nColData) ? &pBlockDataFrom->aColData[iColDataFrom] : NULL;

  for (int32_t iColDataTo = 0; iColDataTo < pBlockData->nColData; iColDataTo++) {
-    SColData *pColDataTo = &((SColData *)pBlockData->aColData->pData)[iColDataTo];
+    SColData *pColDataTo = &pBlockData->aColData[iColDataTo];

    while (pColDataFrom && pColDataFrom->cid < pColDataTo->cid) {
-      iColDataFrom++;
-      pColDataFrom = (iColDataFrom < pBlockDataFrom->nColData)
-                         ? &((SColData *)pBlockDataFrom->aColData->pData)[iColDataFrom]
-                         : NULL;
+      pColDataFrom = (++iColDataFrom < pBlockDataFrom->nColData) ? &pBlockDataFrom->aColData[iColDataFrom] : NULL;
    }

    if (pColDataFrom == NULL || pColDataFrom->cid > pColDataTo->cid) {
@ -1065,157 +1053,7 @@ static int32_t tBlockDataAppendBlockRow(SBlockData *pBlockData, SBlockData *pBlo
      code = tColDataAppendValue(pColDataTo, &cv);
      if (code) goto _exit;

-      iColDataFrom++;
-      pColDataFrom = (iColDataFrom < pBlockDataFrom->nColData)
-                         ? &((SColData *)pBlockDataFrom->aColData->pData)[iColDataFrom]
-                         : NULL;
-    }
-  }
-
-_exit:
-  return code;
-}
-
-static int32_t tBlockDataAppendTPRow(SBlockData *pBlockData, STSRow *pRow, STSchema *pTSchema) {
-  int32_t code = 0;
-
-  int32_t   iTColumn = 1;
-  STColumn *pTColumn = (iTColumn < pTSchema->numOfCols) ? &pTSchema->columns[iTColumn] : NULL;
-  void     *pBitmap = pRow->statis ? tdGetBitmapAddrTp(pRow, pTSchema->flen) : NULL;
-
-  for (int32_t iColData = 0; iColData < pBlockData->nColData; iColData++) {
-    SColData *pColData = &((SColData *)pBlockData->aColData->pData)[iColData];
-
-    while (pTColumn && pTColumn->colId < pColData->cid) {
-      iTColumn++;
-      pTColumn = (iTColumn < pTSchema->numOfCols) ? &pTSchema->columns[iTColumn] : NULL;
-    }
-
-    if (pTColumn == NULL || pTColumn->colId > pColData->cid) {
-      code = tColDataAppendValue(pColData, &COL_VAL_NONE(pColData->cid, pColData->type));
-      if (code) goto _exit;
-    } else {
-      ASSERT(pTColumn->type == pColData->type);
-
-      SColVal cv = {.cid = pTColumn->colId, .type = pTColumn->type};
-
-      if (pRow->statis) {
-        TDRowValT vt = TD_VTYPE_MAX;
-        tdGetBitmapValTypeII(pBitmap, iTColumn - 1, &vt);
-
-        if (vt == TD_VTYPE_NORM) {
-          cv.flag = CV_FLAG_VALUE;
-
-          if (IS_VAR_DATA_TYPE(pTColumn->type)) {
-            void *pData = (char *)pRow + *(int32_t *)(pRow->data + pTColumn->offset);
-            cv.value.nData = varDataLen(pData);
-            cv.value.pData = varDataVal(pData);
-          } else {
-            memcpy(&cv.value.val, pRow->data + pTColumn->offset, pTColumn->bytes);
-          }
-
-          code = tColDataAppendValue(pColData, &cv);
-          if (code) goto _exit;
-        } else if (vt == TD_VTYPE_NONE) {
-          code = tColDataAppendValue(pColData, &COL_VAL_NONE(pColData->cid, pColData->type));
-          if (code) goto _exit;
-        } else if (vt == TD_VTYPE_NULL) {
-          code = tColDataAppendValue(pColData, &COL_VAL_NULL(pColData->cid, pColData->type));
-          if (code) goto _exit;
-        } else {
-          ASSERT(0);
-        }
-      } else {
-        cv.flag = CV_FLAG_VALUE;
-
-        if (IS_VAR_DATA_TYPE(pTColumn->type)) {
-          void *pData = (char *)pRow + *(int32_t *)(pRow->data + pTColumn->offset);
-          cv.value.nData = varDataLen(pData);
-          cv.value.pData = varDataVal(pData);
-        } else {
-          memcpy(&cv.value.val, pRow->data + pTColumn->offset, pTColumn->bytes);
-        }
-
-        code = tColDataAppendValue(pColData, &cv);
-        if (code) goto _exit;
-      }
-
-      iTColumn++;
-      pTColumn = (iTColumn < pTSchema->numOfCols) ? &pTSchema->columns[iTColumn] : NULL;
-    }
-  }
-
-_exit:
-  return code;
-}
-
-static int32_t tBlockDataAppendKVRow(SBlockData *pBlockData, STSRow *pRow, STSchema *pTSchema) {
-  int32_t code = 0;
-
-  col_id_t  kvIter = 0;
-  col_id_t  nKvCols = tdRowGetNCols(pRow) - 1;
-  void     *pColIdx = TD_ROW_COL_IDX(pRow);
-  void     *pBitmap = tdGetBitmapAddrKv(pRow, tdRowGetNCols(pRow));
-  int32_t   iTColumn = 1;
-  STColumn *pTColumn = (iTColumn < pTSchema->numOfCols) ? &pTSchema->columns[iTColumn] : NULL;
-
-  for (int32_t iColData = 0; iColData < pBlockData->nColData; iColData++) {
-    SColData *pColData = &((SColData *)pBlockData->aColData->pData)[iColData];
-
-    while (pTColumn && pTColumn->colId < pColData->cid) {
-      iTColumn++;
-      pTColumn = (iTColumn < pTSchema->numOfCols) ? &pTSchema->columns[iTColumn] : NULL;
-    }
-
-    if (pTColumn == NULL || pTColumn->colId > pColData->cid) {
-      code = tColDataAppendValue(pColData, &COL_VAL_NONE(pColData->cid, pColData->type));
-      if (code) goto _exit;
-    } else {
-      ASSERT(pTColumn->type == pColData->type);
-
-      SColVal    cv = {.cid = pTColumn->colId, .type = pTColumn->type};
-      TDRowValT  vt = TD_VTYPE_NONE;  // default is NONE
-      SKvRowIdx *pKvIdx = NULL;
-
-      while (kvIter < nKvCols) {
-        pKvIdx = (SKvRowIdx *)POINTER_SHIFT(pColIdx, kvIter * sizeof(SKvRowIdx));
-        if (pKvIdx->colId == pTColumn->colId) {
-          tdGetBitmapValTypeII(pBitmap, kvIter, &vt);
-          ++kvIter;
-          break;
-        } else if (pKvIdx->colId > pTColumn->colId) {
-          vt = TD_VTYPE_NONE;
-          break;
-        } else {
-          ++kvIter;
-        }
-      }
-
-      if (vt == TD_VTYPE_NORM) {
-        cv.flag = CV_FLAG_VALUE;
-
-        void *pData = POINTER_SHIFT(pRow, pKvIdx->offset);
-        if (IS_VAR_DATA_TYPE(pTColumn->type)) {
-          cv.value.nData = varDataLen(pData);
-          cv.value.pData = varDataVal(pData);
-        } else {
-          memcpy(&cv.value.val, pData, pTColumn->bytes);
-        }
-
-        code = tColDataAppendValue(pColData, &cv);
-        if (code) goto _exit;
-      } else if (vt == TD_VTYPE_NONE) {
-        code = tColDataAppendValue(pColData, &COL_VAL_NONE(pColData->cid, pColData->type));
-        if (code) goto _exit;
-      } else if (vt == TD_VTYPE_NULL) {
-        code = tColDataAppendValue(pColData, &COL_VAL_NULL(pColData->cid, pColData->type));
-        if (code) goto _exit;
-      } else {
-        ASSERT(0);
-      }
-
-      iTColumn++;
-      pTColumn = (iTColumn < pTSchema->numOfCols) ? &pTSchema->columns[iTColumn] : NULL;
+      pColDataFrom = (++iColDataFrom < pBlockDataFrom->nColData) ? &pBlockDataFrom->aColData[iColDataFrom] : NULL;
    }
  }

@ -1245,19 +1083,14 @@ int32_t tBlockDataAppendRow(SBlockData *pBlockData, TSDBROW *pRow, STSchema *pTS
  pBlockData->aTSKEY[pBlockData->nRow] = TSDBROW_TS(pRow);

  SColVal cv = {0};
-  if (pRow->type == 0) {
-    if (TD_IS_TP_ROW(pRow->pTSRow)) {
-      code = tBlockDataAppendTPRow(pBlockData, pRow->pTSRow, pTSchema);
-      if (code) goto _err;
-    } else if (TD_IS_KV_ROW(pRow->pTSRow)) {
-      code = tBlockDataAppendKVRow(pBlockData, pRow->pTSRow, pTSchema);
-      if (code) goto _err;
-    } else {
-      ASSERT(0);
-    }
-  } else {
+  if (pRow->type == TSDBROW_ROW_FMT) {
+    code = tRowAppendToColData(pRow->pTSRow, pTSchema, pBlockData->aColData, pBlockData->nColData);
+    if (code) goto _err;
+  } else if (pRow->type == TSDBROW_COL_FMT) {
    code = tBlockDataAppendBlockRow(pBlockData, pRow->pBlockData, pRow->iRow);
    if (code) goto _err;
+  } else {
+    ASSERT(0);
  }
  pBlockData->nRow++;

@ -1267,133 +1100,13 @@ _err:
  return code;
 }

-int32_t tBlockDataCorrectSchema(SBlockData *pBlockData, SBlockData *pBlockDataFrom) {
-  int32_t code = 0;
-
-  int32_t iColData = 0;
-  for (int32_t iColDataFrom = 0; iColDataFrom < pBlockDataFrom->nColData; iColDataFrom++) {
-    SColData *pColDataFrom = tBlockDataGetColDataByIdx(pBlockDataFrom, iColDataFrom);
-
-    while (true) {
-      SColData *pColData;
-      if (iColData < pBlockData->nColData) {
-        pColData = tBlockDataGetColDataByIdx(pBlockData, iColData);
-      } else {
-        pColData = NULL;
-      }
-
-      if (pColData == NULL || pColData->cid > pColDataFrom->cid) {
-        code = tBlockDataAddColData(pBlockData, &pColData);
-        if (code) goto _exit;
-
-        tColDataInit(pColData, pColDataFrom->cid, pColDataFrom->type, pColDataFrom->smaOn);
-        for (int32_t iRow = 0; iRow < pBlockData->nRow; iRow++) {
-          code = tColDataAppendValue(pColData, &COL_VAL_NONE(pColData->cid, pColData->type));
-          if (code) goto _exit;
-        }
-
-        iColData++;
-        break;
-      } else if (pColData->cid == pColDataFrom->cid) {
-        iColData++;
-        break;
-      } else {
-        iColData++;
-      }
-    }
-  }
-
-_exit:
-  return code;
-}
-
-int32_t tBlockDataMerge(SBlockData *pBlockData1, SBlockData *pBlockData2, SBlockData *pBlockData) {
-  int32_t code = 0;
-
-  ASSERT(pBlockData->suid == pBlockData1->suid);
-  ASSERT(pBlockData->uid == pBlockData1->uid);
-  ASSERT(pBlockData1->nRow > 0);
-  ASSERT(pBlockData2->nRow > 0);
-
-  tBlockDataClear(pBlockData);
-
-  TSDBROW  row1 = tsdbRowFromBlockData(pBlockData1, 0);
-  TSDBROW  row2 = tsdbRowFromBlockData(pBlockData2, 0);
-  TSDBROW *pRow1 = &row1;
-  TSDBROW *pRow2 = &row2;
-
-  while (pRow1 && pRow2) {
-    int32_t c = tsdbRowCmprFn(pRow1, pRow2);
-
-    if (c < 0) {
-      code = tBlockDataAppendRow(pBlockData, pRow1, NULL,
-                                 pBlockData1->uid ? pBlockData1->uid : pBlockData1->aUid[pRow1->iRow]);
-      if (code) goto _exit;
-
-      pRow1->iRow++;
-      if (pRow1->iRow < pBlockData1->nRow) {
-        *pRow1 = tsdbRowFromBlockData(pBlockData1, pRow1->iRow);
-      } else {
-        pRow1 = NULL;
-      }
-    } else if (c > 0) {
-      code = tBlockDataAppendRow(pBlockData, pRow2, NULL,
-                                 pBlockData2->uid ? pBlockData2->uid : pBlockData2->aUid[pRow2->iRow]);
-      if (code) goto _exit;
-
-      pRow2->iRow++;
-      if (pRow2->iRow < pBlockData2->nRow) {
-        *pRow2 = tsdbRowFromBlockData(pBlockData2, pRow2->iRow);
-      } else {
-        pRow2 = NULL;
-      }
-    } else {
-      ASSERT(0);
-    }
-  }
-
-  while (pRow1) {
-    code = tBlockDataAppendRow(pBlockData, pRow1, NULL,
-                               pBlockData1->uid ? pBlockData1->uid : pBlockData1->aUid[pRow1->iRow]);
-    if (code) goto _exit;
-
-    pRow1->iRow++;
-    if (pRow1->iRow < pBlockData1->nRow) {
-      *pRow1 = tsdbRowFromBlockData(pBlockData1, pRow1->iRow);
-    } else {
-      pRow1 = NULL;
-    }
-  }
-
-  while (pRow2) {
-    code = tBlockDataAppendRow(pBlockData, pRow2, NULL,
-                               pBlockData2->uid ? pBlockData2->uid : pBlockData2->aUid[pRow2->iRow]);
-    if (code) goto _exit;
-
-    pRow2->iRow++;
-    if (pRow2->iRow < pBlockData2->nRow) {
-      *pRow2 = tsdbRowFromBlockData(pBlockData2, pRow2->iRow);
-    } else {
-      pRow2 = NULL;
-    }
-  }
-
-_exit:
-  return code;
-}
-
-SColData *tBlockDataGetColDataByIdx(SBlockData *pBlockData, int32_t idx) {
-  ASSERT(idx >= 0 && idx < pBlockData->nColData);
-  return (SColData *)taosArrayGet(pBlockData->aColData, idx);
-}
-
 void tBlockDataGetColData(SBlockData *pBlockData, int16_t cid, SColData **ppColData) {
  ASSERT(cid != PRIMARYKEY_TIMESTAMP_COL_ID);
  int32_t lidx = 0;
  int32_t ridx = pBlockData->nColData - 1;

  while (lidx <= ridx) {
-    int32_t   midx = (lidx + ridx) / 2;
+    int32_t   midx = (lidx + ridx) >> 1;
    SColData *pColData = tBlockDataGetColDataByIdx(pBlockData, midx);
    int32_t   c = (pColData->cid == cid) ? 0 : ((pColData->cid > cid) ? 1 : -1);

@ -1540,15 +1253,25 @@ int32_t tDecmprBlockData(uint8_t *pIn, int32_t szIn, SBlockData *pBlockData, uin
  // loop to decode each column data
  if (hdr.szBlkCol == 0) goto _exit;

+  int32_t nColData = 0;
  int32_t nt = 0;
  while (nt < hdr.szBlkCol) {
    SBlockCol blockCol = {0};
    nt += tGetBlockCol(pIn + n + nt, &blockCol);
-    ASSERT(nt <= hdr.szBlkCol);
+    ++nColData;
+  }
+  ASSERT(nt == hdr.szBlkCol);

-    SColData *pColData;
-    code = tBlockDataAddColData(pBlockData, &pColData);
-    if (code) goto _exit;
+  code = tBlockDataAdjustColData(pBlockData, nColData);
+  if (code) goto _exit;
+
+  nt = 0;
+  int32_t iColData = 0;
+  while (nt < hdr.szBlkCol) {
+    SBlockCol blockCol = {0};
+    nt += tGetBlockCol(pIn + n + nt, &blockCol);
+
+    SColData *pColData = &pBlockData->aColData[iColData++];

    tColDataInit(pColData, blockCol.cid, blockCol.type, blockCol.smaOn);
    if (blockCol.flag == HAS_NULL) {
--- a/source/dnode/vnode/src/tsdb/tsdbWrite.c
+++ b/source/dnode/vnode/src/tsdb/tsdbWrite.c
@ -26,14 +26,17 @@ static int64_t tsMaxKeyByPrecision[] = {31556995200000L, 31556995200000000L, 921

 // static int tsdbScanAndConvertSubmitMsg(STsdb *pTsdb, SSubmitReq *pMsg);

-int tsdbInsertData(STsdb *pTsdb, int64_t version, SSubmitReq *pMsg, SSubmitRsp *pRsp) {
-  SSubmitMsgIter msgIter = {0};
-  SSubmitBlk    *pBlock = NULL;
-  int32_t        affectedrows = 0;
-  int32_t        numOfRows = 0;
+int tsdbInsertData(STsdb *pTsdb, int64_t version, SSubmitReq2 *pMsg, SSubmitRsp2 *pRsp) {
+  int32_t arrSize = 0;
+  int32_t affectedrows = 0;
+  int32_t numOfRows = 0;

  ASSERT(pTsdb->mem != NULL);

+  if (pMsg) {
+    arrSize = taosArrayGetSize(pMsg->aSubmitTbData);
+  }
+
  // scan and convert
  if (tsdbScanAndConvertSubmitMsg(pTsdb, pMsg) < 0) {
    if (terrno != TSDB_CODE_TDB_TABLE_RECONFIGURE) {
@ -43,18 +46,10 @@ int tsdbInsertData(STsdb *pTsdb, int64_t version, SSubmitReq *pMsg, SSubmitRsp *
  }

  // loop to insert
-  if (tInitSubmitMsgIter(pMsg, &msgIter) < 0) {
-    return -1;
-  }
-  while (true) {
-    SSubmitBlkRsp r = {0};
-    tGetSubmitMsgNext(&msgIter, &pBlock);
-    if (pBlock == NULL) break;
-    if ((terrno = tsdbInsertTableData(pTsdb, version, &msgIter, pBlock, &r)) < 0) {
+  for (int32_t i = 0; i < arrSize; ++i) {
+    if ((terrno = tsdbInsertTableData(pTsdb, version, taosArrayGet(pMsg->aSubmitTbData, i), &affectedrows)) < 0) {
      return -1;
    }
-
-    numOfRows += msgIter.numOfRows;
  }

  if (pRsp != NULL) {
@ -82,9 +77,8 @@ static FORCE_INLINE int tsdbCheckRowRange(STsdb *pTsdb, STable *pTable, STSRow *
 }
 #endif

-static FORCE_INLINE int tsdbCheckRowRange(STsdb *pTsdb, tb_uid_t uid, STSRow *row, TSKEY minKey, TSKEY maxKey,
+static FORCE_INLINE int tsdbCheckRowRange(STsdb *pTsdb, tb_uid_t uid, TSKEY rowKey, TSKEY minKey, TSKEY maxKey,
                                          TSKEY now) {
-  TSKEY rowKey = TD_ROW_KEY(row);
  if (rowKey < minKey || rowKey > maxKey) {
    tsdbError("vgId:%d, table uid %" PRIu64 " timestamp is out of range! now %" PRId64 " minKey %" PRId64
              " maxKey %" PRId64 " row key %" PRId64,
@ -96,6 +90,7 @@ static FORCE_INLINE int tsdbCheckRowRange(STsdb *pTsdb, tb_uid_t uid, STSRow *ro
  return 0;
 }

+#if 0
 int tsdbScanAndConvertSubmitMsg(STsdb *pTsdb, SSubmitReq *pMsg) {
  ASSERT(pMsg != NULL);
  // STsdbMeta *    pMeta = pTsdb->tsdbMeta;
@ -163,6 +158,46 @@ int tsdbScanAndConvertSubmitMsg(STsdb *pTsdb, SSubmitReq *pMsg) {
    }
  }

+  if (terrno != TSDB_CODE_SUCCESS) return -1;
+  return 0;
+}
+#endif
+
+int tsdbScanAndConvertSubmitMsg(STsdb *pTsdb, SSubmitReq2 *pMsg) {
+  ASSERT(pMsg != NULL);
+  STsdbKeepCfg *pCfg = &pTsdb->keepCfg;
+  TSKEY         now = taosGetTimestamp(pCfg->precision);
+  TSKEY         minKey = now - tsTickPerMin[pCfg->precision] * pCfg->keep2;
+  TSKEY         maxKey = tsMaxKeyByPrecision[pCfg->precision];
+  int32_t       size = taosArrayGetSize(pMsg->aSubmitTbData);
+
+  terrno = TSDB_CODE_SUCCESS;
+
+  for (int32_t i = 0; i < size; ++i) {
+    SSubmitTbData *pData = TARRAY_GET_ELEM(pMsg->aSubmitTbData, i);
+    if (pData->flags & SUBMIT_REQ_COLUMN_DATA_FORMAT) {
+      uint64_t  nColData = TARRAY_SIZE(pData->aCol);
+      SColData *aColData = (SColData *)TARRAY_DATA(pData->aCol);
+      if (nColData > 0) {
+        int32_t nRows = aColData[0].nVal;
+        TSKEY  *aKey = (TSKEY *)aColData[0].pData;
+        for (int32_t r = 0; r < nRows; ++r) {
+          if (tsdbCheckRowRange(pTsdb, pData->uid, aKey[r], minKey, maxKey, now) < 0) {
+            return -1;
+          }
+        }
+      }
+    } else {
+      int32_t nRows = taosArrayGetSize(pData->aRowP);
+      for (int32_t r = 0; r < nRows; ++r) {
+        SRow *pRow = (SRow *)taosArrayGetP(pData->aRowP, r);
+        if (tsdbCheckRowRange(pTsdb, pData->uid, pRow->ts, minKey, maxKey, now) < 0) {
+          return -1;
+        }
+      }
+    }
+  }
+
  if (terrno != TSDB_CODE_SUCCESS) return -1;
  return 0;
 }
--- a/source/dnode/vnode/src/vnd/vnodeBufPool.c
+++ b/source/dnode/vnode/src/vnd/vnodeBufPool.c
@ -123,6 +123,7 @@ void vnodeBufPoolReset(SVBufPool *pPool) {
  pPool->ptr = pPool->node.data;
 }

+
 void *vnodeBufPoolMallocAligned(SVBufPool *pPool, int size) {
  SVBufPoolNode *pNode;
  void          *p = NULL;
--- a/source/dnode/vnode/src/vnd/vnodeCommit.c
+++ b/source/dnode/vnode/src/vnd/vnodeCommit.c
@ -184,18 +184,51 @@ _err:
  return -1;
 }

-static int32_t vnodePrepareCommit(SVnode *pVnode) {
+static int32_t vnodePrepareCommit(SVnode *pVnode, SCommitInfo *pInfo) {
  int32_t code = 0;
+  int32_t lino = 0;
+  char    dir[TSDB_FILENAME_LEN] = {0};
+
  tsem_wait(&pVnode->canCommit);

+  pVnode->state.commitTerm = pVnode->state.applyTerm;
+
+  pInfo->info.config = pVnode->config;
+  pInfo->info.state.committed = pVnode->state.applied;
+  pInfo->info.state.commitTerm = pVnode->state.applyTerm;
+  pInfo->info.state.commitID = pVnode->state.commitID;
+  pInfo->pVnode = pVnode;
+  pInfo->txn = metaGetTxn(pVnode->pMeta);
+
+  // save info
+  if (pVnode->pTfs) {
+    snprintf(dir, TSDB_FILENAME_LEN, "%s%s%s", tfsGetPrimaryPath(pVnode->pTfs), TD_DIRSEP, pVnode->path);
+  } else {
+    snprintf(dir, TSDB_FILENAME_LEN, "%s", pVnode->path);
+  }
+  if (vnodeSaveInfo(dir, &pInfo->info) < 0) {
+    code = terrno;
+    TSDB_CHECK_CODE(code, lino, _exit);
+  }
+
  tsdbPrepareCommit(pVnode->pTsdb);
+
  metaPrepareAsyncCommit(pVnode->pMeta);
+
  code = smaPrepareAsyncCommit(pVnode->pSma);
  if (code) goto _exit;

-_exit:
  vnodeBufPoolUnRef(pVnode->inUse);
  pVnode->inUse = NULL;
+
+_exit:
+  if (code) {
+    vError("vgId:%d, %s failed at line %d since %s, commit id:%" PRId64, TD_VID(pVnode), __func__, lino,
+           tstrerror(code), pVnode->state.commitID);
+  } else {
+    vDebug("vgId:%d, %s done", TD_VID(pVnode), __func__);
+  }
+
  return code;
 }

@ -217,33 +250,32 @@ _exit:
 int vnodeAsyncCommit(SVnode *pVnode) {
  int32_t code = 0;

-  // prepare to commit
-  code = vnodePrepareCommit(pVnode);
-  if (code) goto _exit;
-
-  // schedule the task
-  pVnode->state.commitTerm = pVnode->state.applyTerm;
-
  SCommitInfo *pInfo = (SCommitInfo *)taosMemoryCalloc(1, sizeof(*pInfo));
  if (NULL == pInfo) {
    code = TSDB_CODE_OUT_OF_MEMORY;
    goto _exit;
  }
-  pInfo->info.config = pVnode->config;
-  pInfo->info.state.committed = pVnode->state.applied;
-  pInfo->info.state.commitTerm = pVnode->state.applyTerm;
-  pInfo->info.state.commitID = pVnode->state.commitID;
-  pInfo->pVnode = pVnode;
-  pInfo->txn = metaGetTxn(pVnode->pMeta);
+
+  // prepare to commit
+  code = vnodePrepareCommit(pVnode, pInfo);
+  if (TSDB_CODE_SUCCESS != code) {
+    goto _exit;
+  }
+
+  // schedule the task
  code = vnodeScheduleTask(vnodeCommitTask, pInfo);

 _exit:
  if (code) {
+    if (NULL != pInfo) {
+      taosMemoryFree(pInfo);
+    }
    tsem_post(&pVnode->canCommit);
    vError("vgId:%d, %s failed since %s, commit id:%" PRId64, TD_VID(pVnode), __func__, tstrerror(code),
           pVnode->state.commitID);
  } else {
-    vDebug("vgId:%d, %s done", TD_VID(pVnode), __func__);
+    vInfo("vgId:%d, vnode async commit done, commitId:%" PRId64 " term:%" PRId64 " applied:%" PRId64, TD_VID(pVnode),
+          pVnode->state.commitID, pVnode->state.applyTerm, pVnode->state.applied);
  }
  return code;
 }
@ -262,7 +294,7 @@ static int vnodeCommitImpl(SCommitInfo *pInfo) {
  char    dir[TSDB_FILENAME_LEN] = {0};
  SVnode *pVnode = pInfo->pVnode;

-  vInfo("vgId:%d, start to commit, commit ID:%" PRId64 " version:%" PRId64 " term: %" PRId64, TD_VID(pVnode),
+  vInfo("vgId:%d, start to commit, commitId:%" PRId64 " version:%" PRId64 " term: %" PRId64, TD_VID(pVnode),
        pVnode->state.commitID, pVnode->state.applied, pVnode->state.applyTerm);

  // persist wal before starting
@ -271,16 +303,11 @@ static int vnodeCommitImpl(SCommitInfo *pInfo) {
    return -1;
  }

-  // save info
  if (pVnode->pTfs) {
    snprintf(dir, TSDB_FILENAME_LEN, "%s%s%s", tfsGetPrimaryPath(pVnode->pTfs), TD_DIRSEP, pVnode->path);
  } else {
    snprintf(dir, TSDB_FILENAME_LEN, "%s", pVnode->path);
  }
-  if (vnodeSaveInfo(dir, &pInfo->info) < 0) {
-    code = terrno;
-    TSDB_CHECK_CODE(code, lino, _exit);
-  }

  // walBeginSnapshot(pVnode->pWal, pVnode->state.applied);
  syncBeginSnapshot(pVnode->sync, pVnode->state.applied);
--- a/source/dnode/vnode/src/vnd/vnodeSnapshot.c
+++ b/source/dnode/vnode/src/vnd/vnodeSnapshot.c
@ -257,8 +257,8 @@ _exit:
    pReader->index++;
    *nData = sizeof(SSnapDataHdr) + pHdr->size;
    pHdr->index = pReader->index;
-    vInfo("vgId:%d, vnode snapshot read data,index:%" PRId64 " type:%d nData:%d ", TD_VID(pReader->pVnode),
-          pReader->index, pHdr->type, *nData);
+    vDebug("vgId:%d, vnode snapshot read data, index:%" PRId64 " type:%d blockLen:%d ", TD_VID(pReader->pVnode),
+           pReader->index, pHdr->type, *nData);
  } else {
    vInfo("vgId:%d, vnode snapshot read data end, index:%" PRId64, TD_VID(pReader->pVnode), pReader->index);
  }
@ -423,8 +423,8 @@ int32_t vnodeSnapWrite(SVSnapWriter *pWriter, uint8_t *pData, uint32_t nData) {
  ASSERT(pHdr->index == pWriter->index + 1);
  pWriter->index = pHdr->index;

-  vInfo("vgId:%d, vnode snapshot write data, index:%" PRId64 " type:%d nData:%d", TD_VID(pVnode), pHdr->index,
-        pHdr->type, nData);
+  vDebug("vgId:%d, vnode snapshot write data, index:%" PRId64 " type:%d blockLen:%d", TD_VID(pVnode), pHdr->index,
+         pHdr->type, nData);

  switch (pHdr->type) {
    case SNAP_DATA_CFG: {
--- a/source/dnode/vnode/src/vnd/vnodeSvr.c
+++ b/source/dnode/vnode/src/vnd/vnodeSvr.c
@ -77,52 +77,72 @@ int32_t vnodePreProcessWriteMsg(SVnode *pVnode, SRpcMsg *pMsg) {
      tDecoderClear(&dc);
    } break;
    case TDMT_VND_SUBMIT: {
-      SSubmitMsgIter msgIter = {0};
-      SSubmitReq    *pSubmitReq = (SSubmitReq *)pMsg->pCont;
-      SSubmitBlk    *pBlock = NULL;
-      int64_t        ctime = taosGetTimestampMs();
-      tb_uid_t       uid;
+      int64_t ctime = taosGetTimestampMs();

-      if (tInitSubmitMsgIter(pSubmitReq, &msgIter) < 0) {
-        code = terrno;
+      tDecoderInit(&dc, (uint8_t *)pMsg->pCont + sizeof(SMsgHead), pMsg->contLen - sizeof(SMsgHead));
+      tStartDecode(&dc);
+
+      uint64_t nSubmitTbData;
+      if (tDecodeU64v(&dc, &nSubmitTbData) < 0) {
+        code = TSDB_CODE_INVALID_MSG;
        goto _err;
      }

-      for (;;) {
-        tGetSubmitMsgNext(&msgIter, &pBlock);
-        if (pBlock == NULL) break;
+      for (int32_t i = 0; i < nSubmitTbData; i++) {
+        if (tStartDecode(&dc) < 0) {
+          code = TSDB_CODE_INVALID_MSG;
+          goto _err;
+        }

-        if (msgIter.schemaLen > 0) {
-          char *name = NULL;
+        int32_t flags;
+        if (tDecodeI32v(&dc, &flags) < 0) {
+          code = TSDB_CODE_INVALID_MSG;
+          goto _err;
+        }

-          tDecoderInit(&dc, pBlock->data, msgIter.schemaLen);
+        if (flags & SUBMIT_REQ_AUTO_CREATE_TABLE) {
+          // SVCreateTbReq
          if (tStartDecode(&dc) < 0) {
            code = TSDB_CODE_INVALID_MSG;
-            return code;
+            goto _err;
          }

          if (tDecodeI32v(&dc, NULL) < 0) {
            code = TSDB_CODE_INVALID_MSG;
-            return code;
-          }
-          if (tDecodeCStr(&dc, &name) < 0) {
-            code = TSDB_CODE_INVALID_MSG;
-            return code;
+            goto _err;
          }

-          uid = metaGetTableEntryUidByName(pVnode->pMeta, name);
+          char *name = NULL;
+          if (tDecodeCStr(&dc, &name) < 0) {
+            code = TSDB_CODE_INVALID_MSG;
+            goto _err;
+          }
+
+          int64_t uid = metaGetTableEntryUidByName(pVnode->pMeta, name);
          if (uid == 0) {
            uid = tGenIdPI64();
          }
+
          *(int64_t *)(dc.data + dc.pos) = uid;
          *(int64_t *)(dc.data + dc.pos + 8) = ctime;
-          pBlock->uid = htobe64(uid);

          tEndDecode(&dc);
-          tDecoderClear(&dc);
+
+          // SSubmitTbData
+          int64_t suid;
+          if (tDecodeI64(&dc, &suid) < 0) {
+            code = TSDB_CODE_INVALID_MSG;
+            goto _err;
+          }
+
+          *(int64_t *)(dc.data + dc.pos) = uid;
        }
+
+        tEndDecode(&dc);
      }

+      tEndDecode(&dc);
+      tDecoderClear(&dc);
    } break;
    case TDMT_VND_DELETE: {
      int32_t     size;
@ -197,7 +217,7 @@ int32_t vnodeProcessWriteMsg(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRp

  if (!syncUtilUserCommit(pMsg->msgType)) goto _exit;

-  if (pMsg->msgType == TDMT_VND_STREAM_RECOVER_BLOCKING_STAGE) {
+  if (pMsg->msgType == TDMT_VND_STREAM_RECOVER_BLOCKING_STAGE || pMsg->msgType == TDMT_STREAM_TASK_CHECK_RSP) {
    if (tqCheckLogInWal(pVnode->pTq, version)) return 0;
  }

@ -236,7 +256,7 @@ int32_t vnodeProcessWriteMsg(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRp
      break;
    /* TSDB */
    case TDMT_VND_SUBMIT:
-      if (vnodeProcessSubmitReq(pVnode, version, pMsg->pCont, pMsg->contLen, pRsp) < 0) goto _err;
+      if (vnodeProcessSubmitReq(pVnode, version, pReq, len, pRsp) < 0) goto _err;
      break;
    case TDMT_VND_DELETE:
      if (vnodeProcessDeleteReq(pVnode, version, pReq, len, pRsp) < 0) goto _err;
@ -855,57 +875,198 @@ static int32_t vnodeDebugPrintSingleSubmitMsg(SMeta *pMeta, SSubmitBlk *pBlock,
  return TSDB_CODE_SUCCESS;
 }

-static int32_t vnodeDebugPrintSubmitMsg(SVnode *pVnode, SSubmitReq *pMsg, const char *tags) {
-  ASSERT(pMsg != NULL);
-  SSubmitMsgIter msgIter = {0};
-  SMeta         *pMeta = pVnode->pMeta;
-  SSubmitBlk    *pBlock = NULL;
+static int32_t vnodeProcessSubmitReq(SVnode *pVnode, int64_t version, void *pReq, int32_t len, SRpcMsg *pRsp) {
+#if 1
+  int32_t code = 0;
+  terrno = 0;

-  if (tInitSubmitMsgIter(pMsg, &msgIter) < 0) return -1;
-  while (true) {
-    if (tGetSubmitMsgNext(&msgIter, &pBlock) < 0) return -1;
-    if (pBlock == NULL) break;
+  SSubmitReq2 *pSubmitReq = &(SSubmitReq2){0};
+  SSubmitRsp2 *pSubmitRsp = &(SSubmitRsp2){0};
+  SArray      *newTbUids = NULL;
+  int32_t      ret;
+  SEncoder     ec = {0};

-    vnodeDebugPrintSingleSubmitMsg(pMeta, pBlock, &msgIter, tags);
+  pRsp->code = TSDB_CODE_SUCCESS;
+
+  // decode
+  SDecoder dc = {0};
+  tDecoderInit(&dc, pReq, len);
+  if (tDecodeSSubmitReq2(&dc, pSubmitReq) < 0) {
+    code = TSDB_CODE_INVALID_MSG;
+    goto _exit;
+  }
+  tDecoderClear(&dc);
+
+  // check
+  code = tsdbScanAndConvertSubmitMsg(pVnode->pTsdb, pSubmitReq);
+  if (code) {
+    goto _exit;
  }

-  return 0;
-}
+  for (int32_t i = 0; i < TARRAY_SIZE(pSubmitReq->aSubmitTbData); ++i) {
+    SSubmitTbData *pSubmitTbData = taosArrayGet(pSubmitReq->aSubmitTbData, i);

-static int32_t vnodeProcessSubmitReq(SVnode *pVnode, int64_t version, void *pReq, int32_t len, SRpcMsg *pRsp) {
-  SSubmitReq    *pSubmitReq = (SSubmitReq *)pReq;
-  SSubmitRsp     submitRsp = {0};
-  SSubmitMsgIter msgIter = {0};
-  SSubmitBlk    *pBlock = NULL;
-  SVCreateTbReq  createTbReq = {0};
-  SDecoder       decoder = {0};
-  int32_t        nRows = 0;
-  int32_t        tsize, ret;
-  SEncoder       encoder = {0};
-  SArray        *newTbUids = NULL;
-  SVStatis       statis = {0};
-  bool           tbCreated = false;
+    if (pSubmitTbData->pCreateTbReq) {
+      pSubmitTbData->uid = pSubmitTbData->pCreateTbReq->uid;
+    } else {
+      SMetaInfo info = {0};
+
+      code = metaGetInfo(pVnode->pMeta, pSubmitTbData->uid, &info, NULL);
+      if (code) {
+        code = TSDB_CODE_TDB_TABLE_NOT_EXIST;
+        vWarn("vgId:%d, table uid:%" PRId64 " not exists", TD_VID(pVnode), pSubmitTbData->uid);
+        goto _exit;
+      }
+
+      if (info.suid != pSubmitTbData->suid) {
+        code = TSDB_CODE_INVALID_MSG;
+        goto _exit;
+      }
+
+      if (info.suid) {
+        metaGetInfo(pVnode->pMeta, info.suid, &info, NULL);
+      }
+
+      if (pSubmitTbData->sver != info.skmVer) {
+        code = TSDB_CODE_TDB_INVALID_TABLE_SCHEMA_VER;
+        goto _exit;
+      }
+    }
+
+    if (pSubmitTbData->flags & SUBMIT_REQ_COLUMN_DATA_FORMAT) {
+      int32_t   nColData = TARRAY_SIZE(pSubmitTbData->aCol);
+      SColData *aColData = (SColData *)TARRAY_DATA(pSubmitTbData->aCol);
+
+      if (nColData <= 0) {
+        code = TSDB_CODE_INVALID_MSG;
+        goto _exit;
+      }
+
+      if (aColData[0].cid != PRIMARYKEY_TIMESTAMP_COL_ID || aColData[0].type != TSDB_DATA_TYPE_TIMESTAMP ||
+          aColData[0].nVal <= 0) {
+        code = TSDB_CODE_INVALID_MSG;
+        goto _exit;
+      }
+
+      for (int32_t i = 1; i < nColData; i++) {
+        if (aColData[i].nVal != aColData[0].nVal) {
+          code = TSDB_CODE_INVALID_MSG;
+          goto _exit;
+        }
+      }
+    }
+  }
+
+  vDebug("vgId:%d, submit block size %d", TD_VID(pVnode), (int32_t)taosArrayGetSize(pSubmitReq->aSubmitTbData));
+
+  // loop to handle
+  for (int32_t i = 0; i < TARRAY_SIZE(pSubmitReq->aSubmitTbData); ++i) {
+    SSubmitTbData *pSubmitTbData = taosArrayGet(pSubmitReq->aSubmitTbData, i);
+
+    // create table
+    if (pSubmitTbData->pCreateTbReq) {
+      // check (TODO: move check to create table)
+      code = grantCheck(TSDB_GRANT_TIMESERIES);
+      if (code) goto _exit;
+
+      code = grantCheck(TSDB_GRANT_TABLE);
+      if (code) goto _exit;
+
+      // alloc if need
+      if (pSubmitRsp->aCreateTbRsp == NULL &&
+          (pSubmitRsp->aCreateTbRsp = taosArrayInit(TARRAY_SIZE(pSubmitReq->aSubmitTbData), sizeof(SVCreateTbRsp))) ==
+              NULL) {
+        code = TSDB_CODE_OUT_OF_MEMORY;
+        goto _exit;
+      }
+
+      SVCreateTbRsp *pCreateTbRsp = taosArrayReserve(pSubmitRsp->aCreateTbRsp, 1);
+
+      // create table
+      if (metaCreateTable(pVnode->pMeta, version, pSubmitTbData->pCreateTbReq, &pCreateTbRsp->pMeta) ==
+          0) {  // create table success
+
+        if (newTbUids == NULL &&
+            (newTbUids = taosArrayInit(TARRAY_SIZE(pSubmitReq->aSubmitTbData), sizeof(int64_t))) == NULL) {
+          code = TSDB_CODE_OUT_OF_MEMORY;
+          goto _exit;
+        }
+
+        taosArrayPush(newTbUids, &pSubmitTbData->uid);
+
+        if (pCreateTbRsp->pMeta) {
+          vnodeUpdateMetaRsp(pVnode, pCreateTbRsp->pMeta);
+        }
+      } else {  // create table failed
+        if (terrno != TSDB_CODE_TDB_TABLE_ALREADY_EXIST) {
+          code = terrno;
+          goto _exit;
+        }
+      }
+    }
+
+    // insert data
+    int32_t affectedRows;
+    code = tsdbInsertTableData(pVnode->pTsdb, version, pSubmitTbData, &affectedRows);
+    if (code) goto _exit;
+
+    pSubmitRsp->affectedRows += affectedRows;
+  }
+
+  // update table uid list
+  if (taosArrayGetSize(newTbUids) > 0) {
+    vDebug("vgId:%d, add %d table into query table list in handling submit", TD_VID(pVnode),
+           (int32_t)taosArrayGetSize(newTbUids));
+    tqUpdateTbUidList(pVnode->pTq, newTbUids, true);
+  }
+
+_exit:
+  // message
+  pRsp->code = code;
+  tEncodeSize(tEncodeSSubmitRsp2, pSubmitRsp, pRsp->contLen, ret);
+  pRsp->pCont = rpcMallocCont(pRsp->contLen);
+  tEncoderInit(&ec, pRsp->pCont, pRsp->contLen);
+  tEncodeSSubmitRsp2(&ec, pSubmitRsp);
+  tEncoderClear(&ec);
+
+  // update statistics
+  atomic_add_fetch_64(&pVnode->statis.nInsert, pSubmitRsp->affectedRows);
+  atomic_add_fetch_64(&pVnode->statis.nInsertSuccess, pSubmitRsp->affectedRows);
+  atomic_add_fetch_64(&pVnode->statis.nBatchInsert, 1);
+  if (code == 0) {
+    atomic_add_fetch_64(&pVnode->statis.nBatchInsertSuccess, 1);
+    tdProcessRSmaSubmit(pVnode->pSma, version, pSubmitReq, pReq, len, STREAM_INPUT__DATA_SUBMIT);
+  }
+
+  // clear
+  taosArrayDestroy(newTbUids);
+  tDestroySSubmitReq2(pSubmitReq, TSDB_MSG_FLG_DECODE);
+  tDestroySSubmitRsp2(pSubmitRsp, TSDB_MSG_FLG_ENCODE);
+
+  if (code) terrno = code;
+
+  return code;
+
+#else
+  SSubmitReq *pSubmitReq = (SSubmitReq *)pReq;
+  SSubmitRsp  submitRsp = {0};
+  int32_t     nRows = 0;
+  int32_t     tsize, ret;
+  SEncoder    encoder = {0};
+  SArray     *newTbUids = NULL;
+  SVStatis    statis = {0};
+  bool        tbCreated = false;
  terrno = TSDB_CODE_SUCCESS;

  pRsp->code = 0;
  pSubmitReq->version = version;
  statis.nBatchInsert = 1;

-#ifdef TD_DEBUG_PRINT_ROW
-  vnodeDebugPrintSubmitMsg(pVnode, pReq, __func__);
-#endif
-
  if (tsdbScanAndConvertSubmitMsg(pVnode->pTsdb, pSubmitReq) < 0) {
    pRsp->code = terrno;
    goto _exit;
  }

-  // handle the request
-  if (tInitSubmitMsgIter(pSubmitReq, &msgIter) < 0) {
-    pRsp->code = TSDB_CODE_INVALID_MSG;
-    goto _exit;
-  }
-
  submitRsp.pArray = taosArrayInit(msgIter.numOfBlocks, sizeof(SSubmitBlkRsp));
  newTbUids = taosArrayInit(msgIter.numOfBlocks, sizeof(int64_t));
  if (!submitRsp.pArray || !newTbUids) {
@ -922,42 +1083,42 @@ static int32_t vnodeProcessSubmitReq(SVnode *pVnode, int64_t version, void *pReq

    // create table for auto create table mode
    if (msgIter.schemaLen > 0) {
-      tDecoderInit(&decoder, pBlock->data, msgIter.schemaLen);
-      if (tDecodeSVCreateTbReq(&decoder, &createTbReq) < 0) {
-        pRsp->code = TSDB_CODE_INVALID_MSG;
-        tDecoderClear(&decoder);
-        taosArrayDestroy(createTbReq.ctb.tagName);
-        goto _exit;
-      }
+      // tDecoderInit(&decoder, pBlock->data, msgIter.schemaLen);
+      // if (tDecodeSVCreateTbReq(&decoder, &createTbReq) < 0) {
+      //   pRsp->code = TSDB_CODE_INVALID_MSG;
+      //   tDecoderClear(&decoder);
+      //   taosArrayDestroy(createTbReq.ctb.tagName);
+      //   goto _exit;
+      // }

-      if ((terrno = grantCheck(TSDB_GRANT_TIMESERIES)) < 0) {
-        pRsp->code = terrno;
-        tDecoderClear(&decoder);
-        taosArrayDestroy(createTbReq.ctb.tagName);
-        goto _exit;
-      }
+      // if ((terrno = grantCheck(TSDB_GRANT_TIMESERIES)) < 0) {
+      //   pRsp->code = terrno;
+      //   tDecoderClear(&decoder);
+      //   taosArrayDestroy(createTbReq.ctb.tagName);
+      //   goto _exit;
+      // }

-      if ((terrno = grantCheck(TSDB_GRANT_TABLE)) < 0) {
-        pRsp->code = terrno;
-        tDecoderClear(&decoder);
-        taosArrayDestroy(createTbReq.ctb.tagName);
-        goto _exit;
-      }
+      // if ((terrno = grantCheck(TSDB_GRANT_TABLE)) < 0) {
+      //   pRsp->code = terrno;
+      //   tDecoderClear(&decoder);
+      //   taosArrayDestroy(createTbReq.ctb.tagName);
+      //   goto _exit;
+      // }

      if (metaCreateTable(pVnode->pMeta, version, &createTbReq, &submitBlkRsp.pMeta) < 0) {
-        if (terrno != TSDB_CODE_TDB_TABLE_ALREADY_EXIST) {
-          submitBlkRsp.code = terrno;
-          pRsp->code = terrno;
-          tDecoderClear(&decoder);
-          taosArrayDestroy(createTbReq.ctb.tagName);
-          goto _exit;
-        }
+        // if (terrno != TSDB_CODE_TDB_TABLE_ALREADY_EXIST) {
+        //   submitBlkRsp.code = terrno;
+        //   pRsp->code = terrno;
+        //   tDecoderClear(&decoder);
+        //   taosArrayDestroy(createTbReq.ctb.tagName);
+        //   goto _exit;
+        // }
      } else {
        if (NULL != submitBlkRsp.pMeta) {
          vnodeUpdateMetaRsp(pVnode, submitBlkRsp.pMeta);
        }

-        taosArrayPush(newTbUids, &createTbReq.uid);
+        // taosArrayPush(newTbUids, &createTbReq.uid);

        submitBlkRsp.uid = createTbReq.uid;
        submitBlkRsp.tblFName = taosMemoryMalloc(strlen(pVnode->config.dbname) + strlen(createTbReq.name) + 2);
@ -965,18 +1126,15 @@ static int32_t vnodeProcessSubmitReq(SVnode *pVnode, int64_t version, void *pReq
        tbCreated = true;
      }

-      msgIter.uid = createTbReq.uid;
-      if (createTbReq.type == TSDB_CHILD_TABLE) {
-        msgIter.suid = createTbReq.ctb.suid;
-      } else {
-        msgIter.suid = 0;
-      }
+      // msgIter.uid = createTbReq.uid;
+      // if (createTbReq.type == TSDB_CHILD_TABLE) {
+      //   msgIter.suid = createTbReq.ctb.suid;
+      // } else {
+      //   msgIter.suid = 0;
+      // }

-#ifdef TD_DEBUG_PRINT_ROW
-      vnodeDebugPrintSingleSubmitMsg(pVnode->pMeta, pBlock, &msgIter, "real uid");
-#endif
-      tDecoderClear(&decoder);
-      taosArrayDestroy(createTbReq.ctb.tagName);
+      // tDecoderClear(&decoder);
+      // taosArrayDestroy(createTbReq.ctb.tagName);
    }

    if (tsdbInsertTableData(pVnode->pTsdb, version, &msgIter, pBlock, &submitBlkRsp) < 0) {
@ -990,21 +1148,21 @@ static int32_t vnodeProcessSubmitReq(SVnode *pVnode, int64_t version, void *pReq
    }
  }

-  if (taosArrayGetSize(newTbUids) > 0) {
-    vDebug("vgId:%d, add %d table into query table list in handling submit", TD_VID(pVnode),
-           (int32_t)taosArrayGetSize(newTbUids));
-  }
+  // if (taosArrayGetSize(newTbUids) > 0) {
+  //   vDebug("vgId:%d, add %d table into query table list in handling submit", TD_VID(pVnode),
+  //          (int32_t)taosArrayGetSize(newTbUids));
+  // }

-  tqUpdateTbUidList(pVnode->pTq, newTbUids, true);
+  // tqUpdateTbUidList(pVnode->pTq, newTbUids, true);

 _exit:
  taosArrayDestroy(newTbUids);
-  tEncodeSize(tEncodeSSubmitRsp, &submitRsp, tsize, ret);
-  pRsp->pCont = rpcMallocCont(tsize);
-  pRsp->contLen = tsize;
-  tEncoderInit(&encoder, pRsp->pCont, tsize);
-  tEncodeSSubmitRsp(&encoder, &submitRsp);
-  tEncoderClear(&encoder);
+  // tEncodeSize(tEncodeSSubmitRsp, &submitRsp, tsize, ret);
+  // pRsp->pCont = rpcMallocCont(tsize);
+  // pRsp->contLen = tsize;
+  // tEncoderInit(&encoder, pRsp->pCont, tsize);
+  // tEncodeSSubmitRsp(&encoder, &submitRsp);
+  // tEncoderClear(&encoder);

  taosArrayDestroyEx(submitRsp.pArray, tFreeSSubmitBlkRsp);

@ -1025,6 +1183,8 @@ _exit:

  vDebug("vgId:%d, submit success, index:%" PRId64, pVnode->config.vgId, version);
  return 0;
+#endif
+  return 0;
 }

 static int32_t vnodeProcessCreateTSmaReq(SVnode *pVnode, int64_t version, void *pReq, int32_t len, SRpcMsg *pRsp) {
--- a/Show More
+++ b/Show More