Merge branch '3.0' into feature/qnode

CMakeLists.txt

@@ -35,7 +35,7 @@ endif(${BUILD_TEST})
 add_subdirectory(source)
 add_subdirectory(tools)
 add_subdirectory(tests)
-add_subdirectory(example)
+add_subdirectory(examples/c)
 
 # docs 
 add_subdirectory(docs)

cmake/cmake.define

@@ -46,7 +46,7 @@ ENDIF ()
 
 IF (TD_WINDOWS)
     MESSAGE("${Yellow} set compiler flag for Windows! ${ColourReset}")
-    SET(COMMON_FLAGS "/w /D_WIN32 /Zi")
+    SET(COMMON_FLAGS "/w /D_WIN32 /DWIN32 /Zi")
     SET(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} /MANIFEST:NO")
     # IF (MSVC AND (MSVC_VERSION GREATER_EQUAL 1900))
     #  SET(COMMON_FLAGS "${COMMON_FLAGS} /Wv:18")

docs-cn/12-taos-sql/07-function.md

@@ -1,6 +1,6 @@
 ---
-sidebar_label: SQL 函数
+sidebar_label: 函数
-title: SQL 函数
+title: 函数
 toc_max_heading_level: 4
 ---
 
@@ -20,7 +20,7 @@ toc_max_heading_level: 4
 
 **返回结果类型**：如果输入值为整数，输出值是 UBIGINT 类型。如果输入值是 FLOAT/DOUBLE 数据类型，输出值是 DOUBLE 数据类型。
 
-**适用数据类型**：不能应用在 timestamp、binary、nchar、bool 类型字段上；在超级表查询中使用时，不能应用在 tag 列
+**适用数据类型**：数值类型。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -38,7 +38,7 @@ toc_max_heading_level: 4
 
 **返回结果类型**：DOUBLE。如果输入值为 NULL，输出值也为 NULL
 
-**适用数据类型**：不能应用在 timestamp、binary、nchar、bool 类型字段上；在超级表查询中使用时，不能应用在 tag 列
+**适用数据类型**：数值类型。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -56,7 +56,7 @@ toc_max_heading_level: 4
 
 **返回结果类型**：DOUBLE。如果输入值为 NULL，输出值也为 NULL
 
-**适用数据类型**：不能应用在 timestamp、binary、nchar、bool 类型字段上；在超级表查询中使用时，不能应用在 tag 列
+**适用数据类型**：数值类型。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -75,7 +75,7 @@ toc_max_heading_level: 4
 
 **返回结果类型**：DOUBLE。如果输入值为 NULL，输出值也为 NULL
 
-**适用数据类型**：不能应用在 timestamp、binary、nchar、bool 类型字段上；在超级表查询中使用时，不能应用在 tag 列
+**适用数据类型**：数值类型。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -94,7 +94,7 @@ SELECT CEIL(field_name) FROM { tb_name | stb_name } [WHERE clause];
 
 **返回结果类型**：与指定列的原始数据类型一致。例如，如果指定列的原始数据类型为 Float，那么返回的数据类型也为 Float；如果指定列的原始数据类型为 Double，那么返回的数据类型也为 Double。
 
-**适用数据类型**：不能应用在 timestamp、binary、nchar、bool 类型字段上；在超级表查询中使用时，不能应用在 tag 列，无论 tag 列的类型是什么类型。
+**适用数据类型**：数值类型。
 
 **适用于**: 普通表、超级表。
 
@@ -115,7 +115,7 @@ SELECT CEIL(field_name) FROM { tb_name | stb_name } [WHERE clause];
 
 **返回结果类型**：DOUBLE。如果输入值为 NULL，输出值也为 NULL
 
-**适用数据类型**：不能应用在 timestamp、binary、nchar、bool 类型字段上；在超级表查询中使用时，不能应用在 tag 列
+**适用数据类型**：数值类型。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -142,7 +142,7 @@ SELECT FLOOR(field_name) FROM { tb_name | stb_name } [WHERE clause];
 
 **返回结果类型**：DOUBLE。如果输入值为 NULL，输出值也为 NULL
 
-**适用数据类型**：不能应用在 timestamp、binary、nchar、bool 类型字段上；在超级表查询中使用时，不能应用在 tag 列
+**适用数据类型**：数值类型。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -161,7 +161,7 @@ SELECT FLOOR(field_name) FROM { tb_name | stb_name } [WHERE clause];
 
 **返回结果类型**：DOUBLE。如果输入值为 NULL，输出值也为 NULL
 
-**适用数据类型**：不能应用在 timestamp、binary、nchar、bool 类型字段上；在超级表查询中使用时，不能应用在 tag 列
+**适用数据类型**：数值类型。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -190,7 +190,7 @@ SELECT ROUND(field_name) FROM { tb_name | stb_name } [WHERE clause];
 
 **返回结果类型**：DOUBLE。如果输入值为 NULL，输出值也为 NULL
 
-**适用数据类型**：不能应用在 timestamp、binary、nchar、bool 类型字段上；在超级表查询中使用时，不能应用在 tag 列
+**适用数据类型**：数值类型。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -208,7 +208,7 @@ SELECT ROUND(field_name) FROM { tb_name | stb_name } [WHERE clause];
 
 **返回结果类型**：DOUBLE。如果输入值为 NULL，输出值也为 NULL
 
-**适用数据类型**：不能应用在 timestamp、binary、nchar、bool 类型字段上；在超级表查询中使用时，不能应用在 tag 列
+**适用数据类型**：数值类型。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -226,7 +226,7 @@ SELECT ROUND(field_name) FROM { tb_name | stb_name } [WHERE clause];
 
 **返回结果类型**：DOUBLE。如果输入值为 NULL，输出值也为 NULL
 
-**适用数据类型**：不能应用在 timestamp、binary、nchar、bool 类型字段上；在超级表查询中使用时，不能应用在 tag 列
+**适用数据类型**：数值类型。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -248,7 +248,7 @@ SELECT ROUND(field_name) FROM { tb_name | stb_name } [WHERE clause];
 
 **返回结果类型**：INT。如果输入值为NULL，输出值为NULL。
 
-**适用数据类型**：输入参数是 BINARY 类型或者 NCHAR 类型的字符串或者列。不能应用在 TAG 列。
+**适用数据类型**：VARCHAR, NCHAR
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -262,9 +262,9 @@ SELECT ROUND(field_name) FROM { tb_name | stb_name } [WHERE clause];
 
 **功能说明**：字符串连接函数。
 
-**返回结果类型**：如果所有参数均为BINARY类型，则结果类型为BINARY。如果参数包含NCHAR类型，则结果类型为NCHAR。如果输入值为NULL，输出值为NULL。
+**返回结果类型**：如果所有参数均为 VARCHAR 类型，则结果类型为 VARCHAR。如果参数包含NCHAR类型，则结果类型为NCHAR。如果输入值为NULL，输出值为NULL。
 
-**适用数据类型**：BINARY, NCHAR。不能应用在 TAG 列。 该函数最小参数个数为2个，最大参数个数为8个。
+**适用数据类型**：VARCHAR, NCHAR。 该函数最小参数个数为2个，最大参数个数为8个。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -279,9 +279,9 @@ SELECT ROUND(field_name) FROM { tb_name | stb_name } [WHERE clause];
 
 **功能说明**：带分隔符的字符串连接函数。
 
-**返回结果类型**：如果所有参数均为BINARY类型，则结果类型为BINARY。如果参数包含NCHAR类型，则结果类型为NCHAR。如果输入值为NULL，输出值为NULL。如果separator值不为NULL，其他输入为NULL，输出为空串。
+**返回结果类型**：如果所有参数均为VARCHAR类型，则结果类型为VARCHAR。如果参数包含NCHAR类型，则结果类型为NCHAR。如果输入值为NULL，输出值为NULL。如果separator值不为NULL，其他输入为NULL，输出为空串。
 
-**适用数据类型**：BINARY, NCHAR。不能应用在 TAG 列。 该函数最小参数个数为3个，最大参数个数为9个。
+**适用数据类型**：VARCHAR, NCHAR。 该函数最小参数个数为3个，最大参数个数为9个。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -298,7 +298,7 @@ SELECT ROUND(field_name) FROM { tb_name | stb_name } [WHERE clause];
 
 **返回结果类型**：INT。
 
-**适用数据类型**：输入参数是 BINARY 类型或者 NCHAR 类型的字符串或者列。不能应用在 TAG 列。
+**适用数据类型**：输入参数是 VARCHAR 类型或者 NCHAR 类型的字符串或者列。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -315,7 +315,7 @@ SELECT ROUND(field_name) FROM { tb_name | stb_name } [WHERE clause];
 
 **返回结果类型**：同输入类型。如果输入值为NULL，输出值为NULL。
 
-**适用数据类型**：输入参数是 BINARY 类型或者 NCHAR 类型的字符串或者列。不能应用在 TAG 列。
+**适用数据类型**：输入参数是 VARCHAR 类型或者 NCHAR 类型的字符串或者列。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -332,7 +332,7 @@ SELECT ROUND(field_name) FROM { tb_name | stb_name } [WHERE clause];
 
 **返回结果类型**：同输入类型。如果输入值为NULL，输出值为NULL。
 
-**适用数据类型**：输入参数是 BINARY 类型或者 NCHAR 类型的字符串或者列。不能应用在 TAG 列。
+**适用数据类型**：输入参数是 VARCHAR 类型或者 NCHAR 类型的字符串或者列。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -349,7 +349,7 @@ SELECT ROUND(field_name) FROM { tb_name | stb_name } [WHERE clause];
 
 **返回结果类型**：同输入类型。如果输入值为NULL，输出值为NULL。
 
-**适用数据类型**：输入参数是 BINARY 类型或者 NCHAR 类型的字符串或者列。不能应用在 TAG 列。
+**适用数据类型**：输入参数是 VARCHAR 类型或者 NCHAR 类型的字符串或者列。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -366,7 +366,7 @@ SELECT ROUND(field_name) FROM { tb_name | stb_name } [WHERE clause];
 
 **返回结果类型**：同输入类型。如果输入值为NULL，输出值为NULL。
 
-**适用数据类型**：输入参数是 BINARY 类型或者 NCHAR 类型的字符串或者列。不能应用在 TAG 列。输入参数pos可以为正数，也可以为负数。如果pos是正数，表示开始位置从字符串开头正数计算。如果pos为负数，表示开始位置从字符串结尾倒数计算。如果输入参数len被忽略，返回的子串包含从pos开始的整个字串。
+**适用数据类型**：输入参数是 VARCHAR 类型或者 NCHAR 类型的字符串或者列。输入参数pos可以为正数，也可以为负数。如果pos是正数，表示开始位置从字符串开头正数计算。如果pos为负数，表示开始位置从字符串结尾倒数计算。如果输入参数len被忽略，返回的子串包含从pos开始的整个字串。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -383,7 +383,7 @@ SELECT ROUND(field_name) FROM { tb_name | stb_name } [WHERE clause];
 
 **返回结果类型**：同输入类型。如果输入值为NULL，输出值为NULL。
 
-**适用数据类型**：输入参数是 BINARY 类型或者 NCHAR 类型的字符串或者列。不能应用在 TAG 列。
+**适用数据类型**：输入参数是 VARCHAR 类型或者 NCHAR 类型的字符串或者列。
 
 **嵌套子查询支持**：适用于内层查询和外层查询。
 
@@ -400,7 +400,7 @@ SELECT ROUND(field_name) FROM { tb_name | stb_name } [WHERE clause];
   SELECT CAST(expression AS type_name) FROM { tb_name | stb_name } [WHERE clause]
 ```
 
-**功能说明**：数据类型转换函数，输入参数 expression 支持普通列、常量、标量函数及它们之间的四则运算，不支持 tag 列，只适用于 select 子句中。
+**功能说明**：数据类型转换函数，输入参数 expression 支持普通列、常量、标量函数及它们之间的四则运算，只适用于 select 子句中。
 
 **返回结果类型**：CAST 中指定的类型（type_name），可以是 BIGINT、BIGINT UNSIGNED、BINARY、VARCHAR、NCHAR和TIMESTAMP。
 
@@ -423,7 +423,7 @@ SELECT TO_ISO8601(ts_val | ts_col) FROM { tb_name | stb_name } [WHERE clause];
 
 **功能说明**：将 UNIX 时间戳转换成为 ISO8601 标准的日期时间格式，并附加客户端时区信息。
 
-**返回结果数据类型**：BINARY 类型。
+**返回结果数据类型**：VARCHAR 类型。
 
 **适用数据类型**：UNIX 时间戳常量或是 TIMESTAMP 类型的列
 
@@ -462,7 +462,7 @@ SELECT TO_UNIXTIMESTAMP(datetime_string | ts_col) FROM { tb_name | stb_name } [W
 
 **返回结果数据类型**：长整型 INT64。
 
-**应用字段**：字符串常量或是 BINARY/NCHAR 类型的列。
+**应用字段**：字符串常量或是 VARCHAR/NCHAR 类型的列。
 
 **适用于**：表、超级表。
 
@@ -549,7 +549,7 @@ SELECT TIMEZONE() FROM { tb_name | stb_name } [WHERE clause];
 
 **功能说明**：返回客户端当前时区信息。
 
-**返回结果数据类型**：BINARY 类型。
+**返回结果数据类型**：VARCHAR 类型。
 
 **应用字段**：无
 
@@ -668,7 +668,7 @@ SELECT LEASTSQUARES(field_name, start_val, step_val) FROM tb_name [WHERE clause]
 SELECT MODE(field_name) FROM tb_name [WHERE clause];
 ```
 
-**功能说明**：返回出现频率最高的值，若存在多个频率相同的最高值，输出空。不能匹配标签、时间戳输出。
+**功能说明**：返回出现频率最高的值，若存在多个频率相同的最高值，输出空。
 
 **返回数据类型**：同应用的字段。
 
@@ -808,6 +808,25 @@ SELECT BOTTOM(field_name, K) FROM { tb_name | stb_name } [WHERE clause];
 - 系统同时返回该记录关联的时间戳列；
 - 限制：BOTTOM 函数不支持 FILL 子句。
 
+### FIRST
+
+```
+SELECT FIRST(field_name) FROM { tb_name | stb_name } [WHERE clause];
+```
+
+**功能说明**：统计表/超级表中某列的值最先写入的非 NULL 值。
+
+**返回数据类型**：同应用的字段。
+
+**适用数据类型**：所有字段。
+
+**适用于**：表和超级表。
+
+**使用说明**:
+
+- 如果要返回各个列的首个（时间戳最小）非 NULL 值，可以使用 FIRST(\*)；
+- 如果结果集中的某列全部为 NULL 值，则该列的返回结果也是 NULL；
+- 如果结果集中所有列全部为 NULL 值，则不返回结果。
 
 ### INTERP
 
@@ -919,25 +938,6 @@ SELECT PERCENTILE(field_name, P) FROM { tb_name } [WHERE clause];
 
 **使用说明**：*P*值取值范围 0≤*P*≤100，为 0 的时候等同于 MIN，为 100 的时候等同于 MAX。
 
-### FIRST
-
-```
-SELECT FIRST(field_name) FROM { tb_name | stb_name } [WHERE clause];
-```
-
-**功能说明**：统计表/超级表中某列的值最先写入的非 NULL 值。
-
-**返回数据类型**：同应用的字段。
-
-**适用数据类型**：所有字段。
-
-**适用于**：表和超级表。
-
-**使用说明**:
-
-- 如果要返回各个列的首个（时间戳最小）非 NULL 值，可以使用 FIRST(\*)；
-- 如果结果集中的某列全部为 NULL 值，则该列的返回结果也是 NULL；
-- 如果结果集中所有列全部为 NULL 值，则不返回结果。
 
 ### TAIL
 
@@ -1005,7 +1005,7 @@ SELECT UNIQUE(field_name) FROM {tb_name | stb_name} [WHERE clause];
 
 **返回结果类型**： 输入列如果是整数类型返回值为长整型 （int64_t），浮点数返回值为双精度浮点数（Double）。无符号整数类型返回值为无符号长整型（uint64_t）。 返回结果中同时带有每行记录对应的时间戳。
 
-**适用数据类型**：不能应用在 timestamp、binary、nchar、bool 类型字段上；在超级表查询中使用时，不能应用在标签之上。
+**适用数据类型**：数值类型。
 
 **嵌套子查询支持**： 适用于内层查询和外层查询。
 
@@ -1076,7 +1076,7 @@ SELECT IRATE(field_name) FROM tb_name WHERE clause;
 
   **返回结果类型**： 返回双精度浮点数类型。
 
-  **适用数据类型**： 不能应用在 timestamp、binary、nchar、bool 类型上；在超级表查询中使用时，不能应用在标签之上。
+  **适用数据类型**： 数值类型。
 
   **嵌套子查询支持**： 适用于内层查询和外层查询。
 
@@ -1124,7 +1124,7 @@ SELECT STATECOUNT(field_name, oper, val) FROM { tb_name | stb_name } [WHERE clau
 
 **返回结果类型**：整形。
 
-**适用数据类型**：不能应用在 timestamp、binary、nchar、bool 类型字段上。
+**适用数据类型**：数值类型。
 
 **嵌套子查询支持**：不支持应用在子查询上。
 
@@ -1152,7 +1152,7 @@ SELECT stateDuration(field_name, oper, val, unit) FROM { tb_name | stb_name } [W
 
 **返回结果类型**：整形。
 
-**适用数据类型**：不能应用在 timestamp、binary、nchar、bool 类型字段上。
+**适用数据类型**：数值类型。
 
 **嵌套子查询支持**：不支持应用在子查询上。
 

docs-cn/12-taos-sql/13-operators.md

@@ -35,8 +35,8 @@ TDengine 支持 `UNION ALL` 和 `UNION` 操作符。UNION ALL 将查询返回的
 | --- | :---------------: | -------------------------------------------------------------------- | -------------------- |
 | 1   |         =         | 除 BLOB、MEDIUMBLOB 和 JSON 外的所有类型                             | 相等                 |
 | 2   |      <\>, !=      | 除 BLOB、MEDIUMBLOB 和 JSON 外的所有类型，且不可以为表的时间戳主键列 | 不相等               |
-| 3   |      \>, \<       | 除 BLOB、MEDIUMBLOB 和 JSON 外的所有类型                             | 大于，小于           |
+| 3   |      \>, <       | 除 BLOB、MEDIUMBLOB 和 JSON 外的所有类型                             | 大于，小于           |
-| 4   |     \>=, \<=      | 除 BLOB、MEDIUMBLOB 和 JSON 外的所有类型                             | 大于等于，小于等于   |
+| 4   |     \>=, <=      | 除 BLOB、MEDIUMBLOB 和 JSON 外的所有类型                             | 大于等于，小于等于   |
 | 5   |   IS [NOT] NULL   | 所有类型                                                             | 是否为空值           |
 | 6   | [NOT] BETWEEN AND | 除 BOOL、BLOB、MEDIUMBLOB 和 JSON 外的所有类型                       | 闭区间比较           |
 | 7   |        IN         | 除 BLOB、MEDIUMBLOB 和 JSON 外的所有类型，且不可以为表的时间戳主键列 | 与列表内的任意值相等 |

docs-cn/14-reference/03-connector/cpp.mdx

@@ -114,7 +114,6 @@ TDengine 客户端驱动的安装请参考 [安装指南](/reference/connector#
 <summary>订阅和消费</summary>
 
 ```c
-{{#include examples/c/subscribe.c}}
 ```
 
 </details>

docs-cn/20-third-party/01-grafana.mdx

@@ -18,21 +18,22 @@ TDengine 能够与开源数据可视化系统 [Grafana](https://www.grafana.com/
 
 ## 配置 Grafana
 
-TDengine 的 Grafana 插件托管在 GitHub，可从 <https://github.com/taosdata/grafanaplugin/releases/latest> 下载，当前最新版本为 3.1.4。
+使用 [`grafana-cli` 命令行工具](https://grafana.com/docs/grafana/latest/administration/cli/) 进行插件[安装](https://grafana.com/grafana/plugins/tdengine-datasource/?tab=installation)。
-
-推荐使用 [`grafana-cli` 命令行工具](https://grafana.com/docs/grafana/latest/administration/cli/) 进行插件安装。
 
 ```bash
-sudo -u grafana grafana-cli \
+grafana-cli plugins install tdengine-datasource
-  --pluginUrl https://github.com/taosdata/grafanaplugin/releases/download/v3.1.7/tdengine-datasource-3.1.7.zip \
+# with sudo
-  plugins install tdengine-datasource
+sudo -u grafana grafana-cli plugins install tdengine-datasource
 ```
 
-或者下载到本地并解压到 Grafana 插件目录。
+或者从 [GitHub](https://github.com/taosdata/grafanaplugin/releases/tag/latest) 或 [Grafana](https://grafana.com/grafana/plugins/tdengine-datasource/?tab=installation) 下载 .zip 文件到本地并解压到 Grafana 插件目录。命令行下载示例如下：
 
 ```bash
-GF_VERSION=3.1.7
+GF_VERSION=3.2.2
+# from GitHub
 wget https://github.com/taosdata/grafanaplugin/releases/download/v$GF_VERSION/tdengine-datasource-$GF_VERSION.zip
+# from Grafana
+wget -O tdengine-datasource-$GF_VERSION.zip https://grafana.com/api/plugins/tdengine-datasource/versions/$GF_VERSION/download
 ```
 
 以 CentOS 7.2 操作系统为例，将插件包解压到 /var/lib/grafana/plugins 目录下，重新启动 grafana 即可。
@@ -41,28 +42,17 @@ wget https://github.com/taosdata/grafanaplugin/releases/download/v$GF_VERSION/td
 sudo unzip tdengine-datasource-$GF_VERSION.zip -d /var/lib/grafana/plugins/
 ```
 
-:::note
+如果 Grafana 在 Docker 环境下运行，可以使用如下的环境变量设置自动安装 TDengine 数据源插件：
-3.1.6 和更早版本未签名，会在 Grafana 7.3+ / 8.x 版本签名检查时失败导致无法加载插件，需要在 grafana.ini 文件中修改配置如下：
-
-```ini
-[plugins]
-allow_loading_unsigned_plugins = tdengine-datasource
-```
-
-:::
-
-在 Docker 环境下，可以使用如下的环境变量设置自动安装并设置 TDengine 插件：
 
 ```bash
-GF_INSTALL_PLUGINS=https://github.com/taosdata/grafanaplugin/releases/download/v3.1.4/tdengine-datasource-3.1.4.zip;tdengine-datasource
+GF_INSTALL_PLUGINS=tdengine-datasource
-GF_PLUGINS_ALLOW_LOADING_UNSIGNED_PLUGINS=tdengine-datasource
 ```
 
 ## 使用 Grafana
 
 ### 配置数据源
 
-用户可以直接通过 http://localhost:3000 的网址，登录 Grafana 服务器（用户名/密码：admin/admin），通过左侧 `Configuration -> Data Sources` 可以添加数据源，如下图所示：
+用户可以直接通过 <http://localhost:3000> 的网址，登录 Grafana 服务器（用户名/密码：admin/admin），通过左侧 `Configuration -> Data Sources` 可以添加数据源，如下图所示：
 
 ![TDengine Database Grafana plugin add data source](./add_datasource1.webp)
 
@@ -74,7 +64,7 @@ GF_PLUGINS_ALLOW_LOADING_UNSIGNED_PLUGINS=tdengine-datasource
 
 ![TDengine Database Grafana plugin add data source](./add_datasource3.webp)
 
-- Host： TDengine 集群中提供 REST 服务 （在 2.4 之前由 taosd 提供， 从 2.4 开始由 taosAdapter 提供）的组件所在服务器的 IP 地址与 TDengine REST 服务的端口号(6041)，默认 http://localhost:6041。
+- Host： TDengine 集群中提供 REST 服务 （在 2.4 之前由 taosd 提供， 从 2.4 开始由 taosAdapter 提供）的组件所在服务器的 IP 地址与 TDengine REST 服务的端口号(6041)，默认 <http://localhost:6041>。
 - User：TDengine 用户名。
 - Password：TDengine 用户密码。
 

docs-en/12-taos-sql/13-operators.md

@@ -35,8 +35,8 @@ TDengine provides 2 set operators: `UNION ALL` and `UNION`. `UNION ALL` combines
 | --- | :---------------: | ------------------------------------------------------------------- | ----------------------------------------------- |
 | 1   |         =         | Except for BLOB, MEDIUMBLOB and JSON                                | Equal                                           |
 | 2   |      <\>, !=      | Except for BLOB, MEDIUMBLOB, JSON and primary key of timestamp type | Not equal                                       |
-| 3   |      \>, \<       | Except for BLOB, MEDIUMBLOB and JSON                                | Greater than, less than                         |
+| 3   |      \>, <       | Except for BLOB, MEDIUMBLOB and JSON                                | Greater than, less than                         |
-| 4   |     \>=, \<=      | Except for BLOB, MEDIUMBLOB and JSON                                | Greater than or equal to, less than or equal to |
+| 4   |     \>=, <=      | Except for BLOB, MEDIUMBLOB and JSON                                | Greater than or equal to, less than or equal to |
 | 5   |   IS [NOT] NULL   | Any types                                                           | Is NULL or NOT                                  |
 | 6   | [NOT] BETWEEN AND | Except for BLOB, MEDIUMBLOB and JSON                                | In a value range or not                         |
 | 7   |        IN         | Except for BLOB, MEDIUMBLOB, JSON and primary key of timestamp type | In a list of values or not                      |

docs-en/14-reference/03-connector/cpp.mdx

@@ -114,7 +114,6 @@ This section shows sample code for standard access methods to TDengine clusters
 <summary>Subscribe and consume</summary>
 
 ```c
-{{#include examples/c/subscribe.c}}
 ```
 
 </details>

docs-en/20-third-party/01-grafana.mdx

@@ -9,7 +9,8 @@ You can learn more about using the TDengine plugin on [GitHub](https://github.co
 
 ## Prerequisites
 
-In order for Grafana to add the TDengine data source successfully, the following preparation is required:
+In order for Grafana to add the TDengine data source successfully, the following preparations are required:
+
 1. The TDengine cluster is deployed and functioning properly
 2. taosAdapter is installed and running properly. Please refer to the taosAdapter manual for details.
 
@@ -19,41 +20,34 @@ TDengine currently supports Grafana versions 7.0 and above. Users can go to the
 
 ## Configuring Grafana
 
-You can download The Grafana plugin for TDengine from <https://github.com/taosdata/grafanaplugin/releases/latest>. The current latest version is 3.1.4.
+Follow the installation steps in [Grafana](https://grafana.com/grafana/plugins/tdengine-datasource/?tab=installation) with the [``grafana-cli`` command-line tool](https://grafana.com/docs/grafana/latest/administration/cli/) for plugin installation.
-
-Recommend using the [``grafana-cli`` command-line tool](https://grafana.com/docs/grafana/latest/administration/cli/) for plugin installation.
 
 ```bash
-sudo -u grafana grafana-cli \
+grafana-cli plugins install tdengine-datasource
-  --pluginUrl https://github.com/taosdata/grafanaplugin/releases/download/v3.1.4/tdengine-datasource-3.1.4.zip \
+# with sudo
-  plugins install tdengine-datasource
+sudo -u grafana grafana-cli plugins install tdengine-datasource
 ```
 
-Or download it locally and extract it to the Grafana plugin directory.
+Alternatively, you can manually download the .zip file from [GitHub](https://github.com/taosdata/grafanaplugin/releases/tag/latest) or [Grafana](https://grafana.com/grafana/plugins/tdengine-datasource/?tab=installation) and unpack it into your grafana plugins directory.
 
 ```bash
-GF_VERSION=3.1.4
+GF_VERSION=3.2.2
+# from GitHub
 wget https://github.com/taosdata/grafanaplugin/releases/download/v$GF_VERSION/tdengine-datasource-$GF_VERSION.zip
+# from Grafana
+wget -O tdengine-datasource-$GF_VERSION.zip https://grafana.com/api/plugins/tdengine-datasource/versions/$GF_VERSION/download
 ```
 
-In CentOS 7.2 for example, extract the plugin package to /var/lib/grafana/plugins directory, and restart grafana.
+Take CentOS 7.2 for example, extract the plugin package to /var/lib/grafana/plugins directory, and restart grafana.
 
 ```bash
 sudo unzip tdengine-datasource-$GF_VERSION.zip -d /var/lib/grafana/plugins/
 ```
 
-Grafana versions 7.3+ / 8.x do signature checks on plugins, so you also need to add the following line to the grafana.ini file to use the plugin correctly.
+If Grafana is running in a Docker environment, the TDengine plugin can be automatically installed and set up using the following environment variable settings:
-
-```ini
-[plugins]
-allow_loading_unsigned_plugins = tdengine-datasource
-```
-
-The TDengine plugin can be automatically installed and set up using the following environment variable settings in a Docker environment.
 
 ```bash
-GF_INSTALL_PLUGINS=https://github.com/taosdata/grafanaplugin/releases/download/v3.1.4/tdengine-datasource-3.1.4.zip;tdengine- datasource
+GF_INSTALL_PLUGINS=tdengine-datasource
-GF_PLUGINS_ALLOW_LOADING_UNSIGNED_PLUGINS=tdengine-datasource
 ```
 
 ## Using Grafana

example/CMakeLists.txt

@@ -1,49 +0,0 @@
-add_executable(tmq "")
-add_executable(tstream "")
-add_executable(demoapi "")
-
-target_sources(tmq
-    PRIVATE
-    "src/tmq.c"
-)
-
-target_sources(tstream
-    PRIVATE
-    "src/tstream.c"
-)
-
-target_sources(demoapi
-    PRIVATE
-    "src/demoapi.c"
-)
-
-target_link_libraries(tmq
-    taos_static
-)
-
-target_link_libraries(tstream
-    taos_static
-)
-
-target_link_libraries(demoapi
-    taos_static
-)
-
-target_include_directories(tmq
-    PUBLIC "${TD_SOURCE_DIR}/include/os"
-    PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
-)
-
-target_include_directories(tstream
-    PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
-)
-
-target_include_directories(demoapi
-    PUBLIC "${TD_SOURCE_DIR}/include/client"
-    PUBLIC "${TD_SOURCE_DIR}/include/os"
-    PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
-)
-
-SET_TARGET_PROPERTIES(tmq PROPERTIES OUTPUT_NAME tmq)
-SET_TARGET_PROPERTIES(tstream PROPERTIES OUTPUT_NAME tstream)
-SET_TARGET_PROPERTIES(demoapi PROPERTIES OUTPUT_NAME demoapi)

examples/c/CMakeLists.txt

@@ -3,20 +3,70 @@ PROJECT(TDengine)
 IF (TD_LINUX)
   INCLUDE_DIRECTORIES(. ${TD_SOURCE_DIR}/src/inc ${TD_SOURCE_DIR}/src/client/inc  ${TD_SOURCE_DIR}/inc)
   AUX_SOURCE_DIRECTORY(. SRC)
-  ADD_EXECUTABLE(demo apitest.c)
+  #  ADD_EXECUTABLE(demo apitest.c)
-  TARGET_LINK_LIBRARIES(demo taos_static trpc tutil pthread )
+  #TARGET_LINK_LIBRARIES(demo taos_static trpc tutil pthread )
-  ADD_EXECUTABLE(sml schemaless.c)
+  #ADD_EXECUTABLE(sml schemaless.c)
-  TARGET_LINK_LIBRARIES(sml taos_static trpc tutil pthread )
+  #TARGET_LINK_LIBRARIES(sml taos_static trpc tutil pthread )
-  ADD_EXECUTABLE(subscribe subscribe.c)
+  #ADD_EXECUTABLE(subscribe subscribe.c)
-  TARGET_LINK_LIBRARIES(subscribe taos_static trpc tutil pthread )
+  #TARGET_LINK_LIBRARIES(subscribe taos_static trpc tutil pthread )
-  ADD_EXECUTABLE(epoll epoll.c)
+  #ADD_EXECUTABLE(epoll epoll.c)
-  TARGET_LINK_LIBRARIES(epoll taos_static trpc tutil pthread lua)
+  #TARGET_LINK_LIBRARIES(epoll taos_static trpc tutil pthread lua)
+
+  add_executable(tmq "")
+  add_executable(stream_demo "")
+  add_executable(demoapi "")
+
+  target_sources(tmq
+      PRIVATE
+      "tmq.c"
+      )
+
+  target_sources(stream_demo
+      PRIVATE
+      "stream_demo.c"
+      )
+
+  target_sources(demoapi
+      PRIVATE
+      "demoapi.c"
+      )
+
+  target_link_libraries(tmq
+      taos_static
+      )
+
+  target_link_libraries(stream_demo
+      taos_static
+      )
+
+  target_link_libraries(demoapi
+      taos_static
+      )
+
+  target_include_directories(tmq
+      PUBLIC "${TD_SOURCE_DIR}/include/os"
+      PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
+      )
+
+  target_include_directories(stream_demo
+      PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
+      )
+
+  target_include_directories(demoapi
+      PUBLIC "${TD_SOURCE_DIR}/include/client"
+      PUBLIC "${TD_SOURCE_DIR}/include/os"
+      PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
+      )
+
+  SET_TARGET_PROPERTIES(tmq PROPERTIES OUTPUT_NAME tmq)
+  SET_TARGET_PROPERTIES(stream_demo PROPERTIES OUTPUT_NAME stream_demo)
+  SET_TARGET_PROPERTIES(demoapi PROPERTIES OUTPUT_NAME demoapi)
 ENDIF ()
 IF (TD_DARWIN)
   INCLUDE_DIRECTORIES(. ${TD_SOURCE_DIR}/src/inc ${TD_SOURCE_DIR}/src/client/inc  ${TD_SOURCE_DIR}/inc)
   AUX_SOURCE_DIRECTORY(. SRC)
-  ADD_EXECUTABLE(demo demo.c)
+  #ADD_EXECUTABLE(demo demo.c)
-  TARGET_LINK_LIBRARIES(demo taos_static trpc tutil pthread lua)
+  #TARGET_LINK_LIBRARIES(demo taos_static trpc tutil pthread lua)
-  ADD_EXECUTABLE(epoll epoll.c)
+  #ADD_EXECUTABLE(epoll epoll.c)
-  TARGET_LINK_LIBRARIES(epoll taos_static trpc tutil pthread lua)
+  #TARGET_LINK_LIBRARIES(epoll taos_static trpc tutil pthread lua)
 ENDIF ()

examples/c/subscribe.c

@@ -1,263 +0,0 @@
-// sample code for TDengine subscribe/consume API
-// to compile: gcc -o subscribe subscribe.c -ltaos
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-#include <unistd.h>
-#include "../../../include/client/taos.h"  // include TDengine header file
-
-int nTotalRows;
-
-void print_result(TAOS_RES* res, int blockFetch) {
-  TAOS_ROW    row = NULL;
-  int         num_fields = taos_num_fields(res);
-  TAOS_FIELD* fields = taos_fetch_fields(res);
-  int         nRows = 0;
-  
-  if (blockFetch) {
-    nRows = taos_fetch_block(res, &row);
-    //for (int i = 0; i < nRows; i++) {
-    //  taos_print_row(buf, row + i, fields, num_fields);
-    //  puts(buf);
-    //}
-  } else {
-    while ((row = taos_fetch_row(res))) {
-      char buf[4096] = {0};
-      taos_print_row(buf, row, fields, num_fields);
-      puts(buf);
-      nRows++;
-    }
-  }
-
-  nTotalRows += nRows;
-  printf("%d rows consumed.\n", nRows);
-}
-
-
-void subscribe_callback(TAOS_SUB* tsub, TAOS_RES *res, void* param, int code) {
-  print_result(res, *(int*)param);
-}
-
-
-void check_row_count(int line, TAOS_RES* res, int expected) {
-  int actual = 0;
-  TAOS_ROW    row;
-  while ((row = taos_fetch_row(res))) {
-    actual++;
-  }
-  if (actual != expected) {
-    printf("line %d: row count mismatch, expected: %d, actual: %d\n", line, expected, actual);
-  } else {
-    printf("line %d: %d rows consumed as expected\n", line, actual);
-  }
-}
-
-
-void do_query(TAOS* taos, const char* sql) {
-  TAOS_RES* res = taos_query(taos, sql);
-  taos_free_result(res);
-}
-
-
-void run_test(TAOS* taos) {
-  do_query(taos, "drop database if exists test;");
-  
-  usleep(100000);
-  do_query(taos, "create database test;");
-  usleep(100000);
-  do_query(taos, "use test;");
-
-  usleep(100000);
-  do_query(taos, "create table meters(ts timestamp, a int) tags(area int);");
-
-  do_query(taos, "create table t0 using meters tags(0);");
-  do_query(taos, "create table t1 using meters tags(1);");
-  do_query(taos, "create table t2 using meters tags(2);");
-  do_query(taos, "create table t3 using meters tags(3);");
-  do_query(taos, "create table t4 using meters tags(4);");
-  do_query(taos, "create table t5 using meters tags(5);");
-  do_query(taos, "create table t6 using meters tags(6);");
-  do_query(taos, "create table t7 using meters tags(7);");
-  do_query(taos, "create table t8 using meters tags(8);");
-  do_query(taos, "create table t9 using meters tags(9);");
-
-  do_query(taos, "insert into t0 values('2020-01-01 00:00:00.000', 0);");
-  do_query(taos, "insert into t0 values('2020-01-01 00:01:00.000', 0);");
-  do_query(taos, "insert into t0 values('2020-01-01 00:02:00.000', 0);");
-  do_query(taos, "insert into t1 values('2020-01-01 00:00:00.000', 0);");
-  do_query(taos, "insert into t1 values('2020-01-01 00:01:00.000', 0);");
-  do_query(taos, "insert into t1 values('2020-01-01 00:02:00.000', 0);");
-  do_query(taos, "insert into t1 values('2020-01-01 00:03:00.000', 0);");
-  do_query(taos, "insert into t2 values('2020-01-01 00:00:00.000', 0);");
-  do_query(taos, "insert into t2 values('2020-01-01 00:01:00.000', 0);");
-  do_query(taos, "insert into t2 values('2020-01-01 00:01:01.000', 0);");
-  do_query(taos, "insert into t2 values('2020-01-01 00:01:02.000', 0);");
-  do_query(taos, "insert into t3 values('2020-01-01 00:01:02.000', 0);");
-  do_query(taos, "insert into t4 values('2020-01-01 00:01:02.000', 0);");
-  do_query(taos, "insert into t5 values('2020-01-01 00:01:02.000', 0);");
-  do_query(taos, "insert into t6 values('2020-01-01 00:01:02.000', 0);");
-  do_query(taos, "insert into t7 values('2020-01-01 00:01:02.000', 0);");
-  do_query(taos, "insert into t8 values('2020-01-01 00:01:02.000', 0);");
-  do_query(taos, "insert into t9 values('2020-01-01 00:01:02.000', 0);");
-
-  // super tables subscription
-  usleep(1000000);
-
-  TAOS_SUB* tsub = taos_subscribe(taos, 0, "test", "select * from meters;", NULL, NULL, 0);
-  TAOS_RES* res = taos_consume(tsub);
-  check_row_count(__LINE__, res, 18);
-
-  res = taos_consume(tsub);
-  check_row_count(__LINE__, res, 0);
-
-  do_query(taos, "insert into t0 values('2020-01-01 00:02:00.001', 0);");
-  do_query(taos, "insert into t8 values('2020-01-01 00:01:03.000', 0);");
-  res = taos_consume(tsub);
-  check_row_count(__LINE__, res, 2);
-
-  do_query(taos, "insert into t2 values('2020-01-01 00:01:02.001', 0);");
-  do_query(taos, "insert into t1 values('2020-01-01 00:03:00.001', 0);");
-  res = taos_consume(tsub);
-  check_row_count(__LINE__, res, 2);
-
-  do_query(taos, "insert into t1 values('2020-01-01 00:03:00.002', 0);");
-  res = taos_consume(tsub);
-  check_row_count(__LINE__, res, 1);
-
-  // keep progress information and restart subscription
-  taos_unsubscribe(tsub, 1);
-  do_query(taos, "insert into t0 values('2020-01-01 00:04:00.000', 0);");
-  tsub = taos_subscribe(taos, 1, "test", "select * from meters;", NULL, NULL, 0);
-  res = taos_consume(tsub);
-  check_row_count(__LINE__, res, 24);
-
-  // keep progress information and continue previous subscription
-  taos_unsubscribe(tsub, 1);
-  tsub = taos_subscribe(taos, 0, "test", "select * from meters;", NULL, NULL, 0);
-  res = taos_consume(tsub);
-  check_row_count(__LINE__, res, 0);
-
-  // don't keep progress information and continue previous subscription
-  taos_unsubscribe(tsub, 0);
-  tsub = taos_subscribe(taos, 0, "test", "select * from meters;", NULL, NULL, 0);
-  res = taos_consume(tsub);
-  check_row_count(__LINE__, res, 24);
-
-  // single meter subscription
-
-  taos_unsubscribe(tsub, 0);
-  tsub = taos_subscribe(taos, 0, "test", "select * from t0;", NULL, NULL, 0);
-  res = taos_consume(tsub);
-  check_row_count(__LINE__, res, 5);
-
-  res = taos_consume(tsub);
-  check_row_count(__LINE__, res, 0);
-
-  do_query(taos, "insert into t0 values('2020-01-01 00:04:00.001', 0);");
-  res = taos_consume(tsub);
-  check_row_count(__LINE__, res, 1);
-
-  taos_unsubscribe(tsub, 0);
-}
-
-
-int main(int argc, char *argv[]) {
-  const char* host = "127.0.0.1";
-  const char* user = "root";
-  const char* passwd = "taosdata";
-  const char* sql = "select * from meters;";
-  const char* topic = "test-multiple";
-  int async = 1, restart = 0, keep = 1, test = 0, blockFetch = 0;
-
-  for (int i = 1; i < argc; i++) {
-    if (strncmp(argv[i], "-h=", 3) == 0) {
-      host = argv[i] + 3;
-      continue;
-    }
-    if (strncmp(argv[i], "-u=", 3) == 0) {
-      user = argv[i] + 3;
-      continue;
-    }
-    if (strncmp(argv[i], "-p=", 3) == 0) {
-      passwd = argv[i] + 3;
-      continue;
-    }
-    if (strcmp(argv[i], "-sync") == 0) {
-      async = 0;
-      continue;
-    }
-    if (strcmp(argv[i], "-restart") == 0) {
-      restart = 1;
-      continue;
-    }
-    if (strcmp(argv[i], "-single") == 0) {
-      sql = "select * from t0;";
-      topic = "test-single";
-      continue;
-    }
-    if (strcmp(argv[i], "-nokeep") == 0) {
-      keep = 0;
-      continue;
-    }
-    if (strncmp(argv[i], "-sql=", 5) == 0) {
-      sql = argv[i] + 5;
-      topic = "test-custom";
-      continue;
-    }
-    if (strcmp(argv[i], "-test") == 0) {
-      test = 1;
-      continue;
-    }
-    if (strcmp(argv[i], "-block-fetch") == 0) {
-      blockFetch = 1;
-      continue;
-    }
-  }
-
-  TAOS* taos = taos_connect(host, user, passwd, "", 0);
-  if (taos == NULL) {
-    printf("failed to connect to db, reason:%s\n", taos_errstr(taos));
-    exit(1);
-  }
-
-  if (test) {
-    run_test(taos);
-    taos_close(taos);
-    exit(0);
-  }
-
-  taos_select_db(taos, "test");
-  TAOS_SUB* tsub = NULL;
-  if (async) {
-    // create an asynchronized subscription, the callback function will be called every 1s
-    tsub = taos_subscribe(taos, restart, topic, sql, subscribe_callback, &blockFetch, 1000);
-  } else {
-    // create an synchronized subscription, need to call 'taos_consume' manually
-    tsub = taos_subscribe(taos, restart, topic, sql, NULL, NULL, 0);
-  }
-
-  if (tsub == NULL) {
-    printf("failed to create subscription.\n");
-    exit(0);
-  } 
-
-  if (async) {
-    getchar();
-  } else while(1) {
-    TAOS_RES* res = taos_consume(tsub);
-    if (res == NULL) {
-      printf("failed to consume data.");
-      break;
-    } else {
-      print_result(res, blockFetch);
-      getchar();
-    }
-  }
-
-  printf("total rows consumed: %d\n", nTotalRows);
-  taos_unsubscribe(tsub, keep);
-  taos_close(taos);
-
-  return 0;
-}

examples/c/tmq.c

@@ -24,6 +24,7 @@ static void msg_process(TAOS_RES* msg) {
   char buf[1024];
   /*memset(buf, 0, 1024);*/
   printf("topic: %s\n", tmq_get_topic_name(msg));
+  printf("db: %s\n", tmq_get_db_name(msg));
   printf("vg: %d\n", tmq_get_vgroup_id(msg));
   while (1) {
     TAOS_ROW row = taos_fetch_row(msg);
@@ -165,7 +166,6 @@ tmq_t* build_consumer() {
   tmq_conf_set(conf, "group.id", "tg2");
   tmq_conf_set(conf, "td.connect.user", "root");
   tmq_conf_set(conf, "td.connect.pass", "taosdata");
-  /*tmq_conf_set(conf, "td.connect.db", "abc1");*/
   tmq_conf_set(conf, "msg.with.table.name", "true");
   tmq_conf_set(conf, "enable.auto.commit", "false");
   tmq_conf_set_auto_commit_cb(conf, tmq_commit_cb_print, NULL);
@@ -191,21 +191,18 @@ void basic_consume_loop(tmq_t* tmq, tmq_list_t* topics) {
     return;
   }
   int32_t cnt = 0;
-  /*clock_t startTime = clock();*/
   while (running) {
     TAOS_RES* tmqmessage = tmq_consumer_poll(tmq, 0);
     if (tmqmessage) {
       cnt++;
       msg_process(tmqmessage);
-      if (cnt >= 2) break;
+      /*if (cnt >= 2) break;*/
       /*printf("get data\n");*/
       taos_free_result(tmqmessage);
       /*} else {*/
       /*break;*/
     }
   }
-  /*clock_t endTime = clock();*/
-  /*printf("log cnt: %d %f s\n", cnt, (double)(endTime - startTime) / CLOCKS_PER_SEC);*/
 
   err = tmq_consumer_close(tmq);
   if (err)

include/client/taos.h

@@ -144,8 +144,8 @@ DLL_EXPORT int        taos_stmt_set_tbname_tags(TAOS_STMT *stmt, const char *nam
 DLL_EXPORT int        taos_stmt_set_tbname(TAOS_STMT *stmt, const char *name);
 DLL_EXPORT int        taos_stmt_set_tags(TAOS_STMT *stmt, TAOS_MULTI_BIND *tags);
 DLL_EXPORT int        taos_stmt_set_sub_tbname(TAOS_STMT *stmt, const char *name);
-DLL_EXPORT int        taos_stmt_get_tag_fields(TAOS_STMT *stmt, int* fieldNum, TAOS_FIELD_E** fields);
+DLL_EXPORT int        taos_stmt_get_tag_fields(TAOS_STMT *stmt, int *fieldNum, TAOS_FIELD_E **fields);
-DLL_EXPORT int        taos_stmt_get_col_fields(TAOS_STMT *stmt, int* fieldNum, TAOS_FIELD_E** fields);
+DLL_EXPORT int        taos_stmt_get_col_fields(TAOS_STMT *stmt, int *fieldNum, TAOS_FIELD_E **fields);
 
 DLL_EXPORT int       taos_stmt_is_insert(TAOS_STMT *stmt, int *insert);
 DLL_EXPORT int       taos_stmt_num_params(TAOS_STMT *stmt, int *nums);
@@ -269,6 +269,7 @@ DLL_EXPORT void           tmq_conf_set_auto_commit_cb(tmq_conf_t *conf, tmq_comm
 /* -------------------------TMQ MSG HANDLE INTERFACE---------------------- */
 
 DLL_EXPORT const char *tmq_get_topic_name(TAOS_RES *res);
+DLL_EXPORT const char *tmq_get_db_name(TAOS_RES *res);
 DLL_EXPORT int32_t     tmq_get_vgroup_id(TAOS_RES *res);
 DLL_EXPORT const char *tmq_get_table_name(TAOS_RES *res);
 

include/common/taosdef.h

@@ -97,6 +97,7 @@ extern char *qtypeStr[];
 
 #undef TD_DEBUG_PRINT_ROW
 #undef TD_DEBUG_PRINT_TSDB_LOAD_DCOLS
+#undef TD_DEBUG_PRINT_TAG
 
 #ifdef __cplusplus
 }

include/common/tdataformat.h

@@ -18,6 +18,7 @@
 
 #include "os.h"
 #include "talgo.h"
+#include "tarray.h"
 #include "tencode.h"
 #include "ttypes.h"
 #include "tutil.h"
@@ -29,6 +30,7 @@ extern "C" {
 typedef struct SSchema       SSchema;
 typedef struct STColumn      STColumn;
 typedef struct STSchema      STSchema;
+typedef struct SValue        SValue;
 typedef struct SColVal       SColVal;
 typedef struct STSRow2       STSRow2;
 typedef struct STSRowBuilder STSRowBuilder;
@@ -39,32 +41,37 @@ typedef struct STag          STag;
 int32_t tTSchemaCreate(int32_t sver, SSchema *pSchema, int32_t nCols, STSchema **ppTSchema);
 void    tTSchemaDestroy(STSchema *pTSchema);
 
-// SColVal
-#define ColValNONE               ((SColVal){.type = COL_VAL_NONE, .nData = 0, .pData = NULL})
-#define ColValNULL               ((SColVal){.type = COL_VAL_NULL, .nData = 0, .pData = NULL})
-#define ColValDATA(nData, pData) ((SColVal){.type = COL_VAL_DATA, .nData = (nData), .pData = (pData)})
-
 // STSRow2
+#define COL_VAL_NONE(CID)     ((SColVal){.cid = (CID), .isNone = 1})
+#define COL_VAL_NULL(CID)     ((SColVal){.cid = (CID), .isNull = 1})
+#define COL_VAL_VALUE(CID, V) ((SColVal){.cid = (CID), .value = (V)})
+
+int32_t tTSRowClone(const STSRow2 *pRow, STSRow2 **ppRow);
+void    tTSRowFree(STSRow2 *pRow);
+void    tTSRowGet(STSRow2 *pRow, STSchema *pTSchema, int32_t iCol, SColVal *pColVal);
+int32_t tTSRowToArray(STSRow2 *pRow, STSchema *pTSchema, SArray **ppArray);
 int32_t tPutTSRow(uint8_t *p, STSRow2 *pRow);
 int32_t tGetTSRow(uint8_t *p, STSRow2 *pRow);
-int32_t tTSRowDup(const STSRow2 *pRow, STSRow2 **ppRow);
-void    tTSRowFree(STSRow2 *pRow);
-int32_t tTSRowGet(const STSRow2 *pRow, STSchema *pTSchema, int32_t iCol, SColVal *pColVal);
 
 // STSRowBuilder
+#if 0
 int32_t tTSRowBuilderInit(STSRowBuilder *pBuilder, int32_t sver, int32_t nCols, SSchema *pSchema);
 void    tTSRowBuilderClear(STSRowBuilder *pBuilder);
 void    tTSRowBuilderReset(STSRowBuilder *pBuilder);
 int32_t tTSRowBuilderPut(STSRowBuilder *pBuilder, int32_t cid, uint8_t *pData, uint32_t nData);
 int32_t tTSRowBuilderGetRow(STSRowBuilder *pBuilder, const STSRow2 **ppRow);
+#endif
 
 // STag
-int32_t tTagNew(STagVal *pTagVals, int16_t nTag, STag **ppTag);
+int32_t tTagNew(SArray *pArray, int32_t version, int8_t isJson, STag **ppTag);
 void    tTagFree(STag *pTag);
-int32_t tTagSet(STag *pTag, SSchema *pSchema, int32_t nCols, int iCol, uint8_t *pData, uint32_t nData, STag **ppTag);
+bool    tTagGet(const STag *pTag, STagVal *pTagVal);
-void    tTagGet(STag *pTag, int16_t cid, int8_t type, uint8_t **ppData, uint32_t *nData);
+char*   tTagValToData(const STagVal *pTagVal, bool isJson);
 int32_t tEncodeTag(SEncoder *pEncoder, const STag *pTag);
 int32_t tDecodeTag(SDecoder *pDecoder, STag **ppTag);
+int32_t tTagToValArray(const STag *pTag, SArray **ppArray);
+void    debugPrintSTag(STag *pTag, const char *tag, int32_t ln); // TODO: remove
+void    debugCheckTags(STag *pTag); // TODO: remove
 
 // STRUCT =================
 struct STColumn {
@@ -87,7 +94,9 @@ struct STSchema {
 #define TSROW_HAS_NONE ((uint8_t)0x1)
 #define TSROW_HAS_NULL ((uint8_t)0x2U)
 #define TSROW_HAS_VAL  ((uint8_t)0x4U)
-#define TSROW_KV_ROW   ((uint8_t)0x10U)
+#define TSROW_KV_SMALL ((uint8_t)0x10U)
+#define TSROW_KV_MID   ((uint8_t)0x20U)
+#define TSROW_KV_BIG   ((uint8_t)0x40U)
 struct STSRow2 {
   TSKEY    ts;
   uint8_t  flags;
@@ -110,20 +119,60 @@ struct STSRowBuilder {
   STSRow2   row;
 };
 
-typedef enum { COL_VAL_NONE = 0, COL_VAL_NULL = 1, COL_VAL_DATA = 2 } EColValT;
+struct SValue {
-struct SColVal {
+  union {
-  EColValT type;
+    int8_t   i8;   // TSDB_DATA_TYPE_BOOL||TSDB_DATA_TYPE_TINYINT
-  uint32_t nData;
+    uint8_t  u8;   // TSDB_DATA_TYPE_UTINYINT
-  uint8_t *pData;
+    int16_t  i16;  // TSDB_DATA_TYPE_SMALLINT
+    uint16_t u16;  // TSDB_DATA_TYPE_USMALLINT
+    int32_t  i32;  // TSDB_DATA_TYPE_INT
+    uint32_t u32;  // TSDB_DATA_TYPE_UINT
+    int64_t  i64;  // TSDB_DATA_TYPE_BIGINT
+    uint64_t u64;  // TSDB_DATA_TYPE_UBIGINT
+    TSKEY    ts;   // TSDB_DATA_TYPE_TIMESTAMP
+    float    f;    // TSDB_DATA_TYPE_FLOAT
+    double   d;    // TSDB_DATA_TYPE_DOUBLE
+    struct {
+      uint32_t nData;
+      uint8_t *pData;
+    };
+  };
 };
 
-struct STagVal {
+struct SColVal {
-  int16_t  cid;
+  int16_t cid;
-  int8_t   type;
+  int8_t  isNone;
-  uint32_t nData;
+  int8_t  isNull;
-  uint8_t *pData;
+  SValue  value;
 };
 
+#pragma pack(push, 1)
+struct STagVal {
+  union {
+    int16_t cid;
+    char   *pKey;
+  };
+  int8_t type;
+  union {
+    int64_t  i64;
+    struct {
+      uint32_t nData;
+      uint8_t *pData;
+    };
+  };
+};
+
+#define TD_TAG_JSON  ((int8_t)0x40)   // distinguish JSON string and JSON value with the highest bit
+#define TD_TAG_LARGE ((int8_t)0x20)
+struct STag {
+  int8_t  flags;
+  int16_t len;
+  int16_t nTag;
+  int32_t ver;
+  int8_t  idx[];
+};
+#pragma pack(pop)
+
 #if 1  //================================================================================================================================================
 // Imported since 3.0 and use bitmap to demonstrate None/Null/Norm, while use Null/Norm below 3.0 without of bitmap.
 #define TD_SUPPORT_BITMAP
@@ -366,109 +415,6 @@ SDataCols *tdFreeDataCols(SDataCols *pCols);
 int32_t    tdMergeDataCols(SDataCols *target, SDataCols *source, int32_t rowsToMerge, int32_t *pOffset, bool update,
                            TDRowVerT maxVer);
 
-// ----------------- K-V data row structure
-/* |<-------------------------------------- len -------------------------------------------->|
- * |<----- header  ----->|<--------------------------- body -------------------------------->|
- * +----------+----------+---------------------------------+---------------------------------+
- * | uint16_t |  int16_t |                                 |                                 |
- * +----------+----------+---------------------------------+---------------------------------+
- * |    len   |   ncols  |           cols index            |             data part           |
- * +----------+----------+---------------------------------+---------------------------------+
- */
-typedef void *SKVRow;
-
-typedef struct {
-  int16_t  colId;
-  uint16_t offset;
-} SColIdx;
-
-#define TD_KV_ROW_HEAD_SIZE (sizeof(uint16_t) + sizeof(int16_t))
-
-#define kvRowLen(r)            (*(uint16_t *)(r))
-#define kvRowNCols(r)          (*(int16_t *)POINTER_SHIFT(r, sizeof(uint16_t)))
-#define kvRowSetLen(r, len)    kvRowLen(r) = (len)
-#define kvRowSetNCols(r, n)    kvRowNCols(r) = (n)
-#define kvRowColIdx(r)         (SColIdx *)POINTER_SHIFT(r, TD_KV_ROW_HEAD_SIZE)
-#define kvRowValues(r)         POINTER_SHIFT(r, TD_KV_ROW_HEAD_SIZE + sizeof(SColIdx) * kvRowNCols(r))
-#define kvRowCpy(dst, r)       memcpy((dst), (r), kvRowLen(r))
-#define kvRowColVal(r, colIdx) POINTER_SHIFT(kvRowValues(r), (colIdx)->offset)
-#define kvRowColIdxAt(r, i)    (kvRowColIdx(r) + (i))
-#define kvRowFree(r)           taosMemoryFreeClear(r)
-#define kvRowEnd(r)            POINTER_SHIFT(r, kvRowLen(r))
-#define kvRowValLen(r)         (kvRowLen(r) - TD_KV_ROW_HEAD_SIZE - sizeof(SColIdx) * kvRowNCols(r))
-#define kvRowTKey(r)           (*(TKEY *)(kvRowValues(r)))
-#define kvRowKey(r)            tdGetKey(kvRowTKey(r))
-#define kvRowKeys(r)           POINTER_SHIFT(r, *(uint16_t *)POINTER_SHIFT(r, TD_KV_ROW_HEAD_SIZE + sizeof(int16_t)))
-#define kvRowDeleted(r)        TKEY_IS_DELETED(kvRowTKey(r))
-
-SKVRow  tdKVRowDup(SKVRow row);
-int32_t tdSetKVRowDataOfCol(SKVRow *orow, int16_t colId, int8_t type, void *value);
-int32_t tdEncodeKVRow(void **buf, SKVRow row);
-void   *tdDecodeKVRow(void *buf, SKVRow *row);
-void    tdSortKVRowByColIdx(SKVRow row);
-
-static FORCE_INLINE int32_t comparTagId(const void *key1, const void *key2) {
-  if (*(int16_t *)key1 > ((SColIdx *)key2)->colId) {
-    return 1;
-  } else if (*(int16_t *)key1 < ((SColIdx *)key2)->colId) {
-    return -1;
-  } else {
-    return 0;
-  }
-}
-
-static FORCE_INLINE void *tdGetKVRowValOfCol(const SKVRow row, int16_t colId) {
-  void *ret = taosbsearch(&colId, kvRowColIdx(row), kvRowNCols(row), sizeof(SColIdx), comparTagId, TD_EQ);
-  if (ret == NULL) return NULL;
-  return kvRowColVal(row, (SColIdx *)ret);
-}
-
-static FORCE_INLINE void *tdGetKVRowIdxOfCol(SKVRow row, int16_t colId) {
-  return taosbsearch(&colId, kvRowColIdx(row), kvRowNCols(row), sizeof(SColIdx), comparTagId, TD_EQ);
-}
-
-// ----------------- K-V data row builder
-typedef struct {
-  int16_t  tCols;
-  int16_t  nCols;
-  SColIdx *pColIdx;
-  uint16_t alloc;
-  uint16_t size;
-  void    *buf;
-} SKVRowBuilder;
-
-int32_t tdInitKVRowBuilder(SKVRowBuilder *pBuilder);
-void    tdDestroyKVRowBuilder(SKVRowBuilder *pBuilder);
-void    tdResetKVRowBuilder(SKVRowBuilder *pBuilder);
-SKVRow  tdGetKVRowFromBuilder(SKVRowBuilder *pBuilder);
-
-static FORCE_INLINE int32_t tdAddColToKVRow(SKVRowBuilder *pBuilder, col_id_t colId, const void *value, int32_t tlen) {
-  if (pBuilder->nCols >= pBuilder->tCols) {
-    pBuilder->tCols *= 2;
-    SColIdx *pColIdx = (SColIdx *)taosMemoryRealloc((void *)(pBuilder->pColIdx), sizeof(SColIdx) * pBuilder->tCols);
-    if (pColIdx == NULL) return -1;
-    pBuilder->pColIdx = pColIdx;
-  }
-
-  pBuilder->pColIdx[pBuilder->nCols].colId = colId;
-  pBuilder->pColIdx[pBuilder->nCols].offset = pBuilder->size;
-
-  pBuilder->nCols++;
-
-  if (tlen > pBuilder->alloc - pBuilder->size) {
-    while (tlen > pBuilder->alloc - pBuilder->size) {
-      pBuilder->alloc *= 2;
-    }
-    void *buf = taosMemoryRealloc(pBuilder->buf, pBuilder->alloc);
-    if (buf == NULL) return -1;
-    pBuilder->buf = buf;
-  }
-
-  memcpy(POINTER_SHIFT(pBuilder->buf, pBuilder->size), value, tlen);
-  pBuilder->size += tlen;
-
-  return 0;
-}
 #endif
 
 #ifdef __cplusplus
@@ -476,3 +422,4 @@ static FORCE_INLINE int32_t tdAddColToKVRow(SKVRowBuilder *pBuilder, col_id_t co
 #endif
 
 #endif /*_TD_COMMON_DATA_FORMAT_H_*/
+

include/common/tmsg.h

@@ -244,7 +244,7 @@ typedef struct {
   const void* pMsg;
 } SSubmitMsgIter;
 
-int32_t tInitSubmitMsgIter(SSubmitReq* pMsg, SSubmitMsgIter* pIter);
+int32_t tInitSubmitMsgIter(const SSubmitReq* pMsg, SSubmitMsgIter* pIter);
 int32_t tGetSubmitMsgNext(SSubmitMsgIter* pIter, SSubmitBlk** pPBlock);
 int32_t tInitSubmitBlkIter(SSubmitMsgIter* pMsgIter, SSubmitBlk* pBlock, SSubmitBlkIter* pIter);
 STSRow* tGetSubmitBlkNext(SSubmitBlkIter* pIter);
@@ -287,7 +287,7 @@ typedef struct SSchema {
   char     name[TSDB_COL_NAME_LEN];
 } SSchema;
 
-#define COL_IS_SET(FLG)  ((FLG) & (COL_SET_VAL | COL_SET_NULL) != 0)
+#define COL_IS_SET(FLG)  (((FLG) & (COL_SET_VAL | COL_SET_NULL)) != 0)
 #define COL_CLR_SET(FLG) ((FLG) &= (~(COL_SET_VAL | COL_SET_NULL)))
 
 #define IS_BSMA_ON(s) (((s)->flags & 0x01) == COL_SMA_ON)
@@ -945,7 +945,6 @@ typedef struct {
   int64_t timeInFetchQueue;
 } SQnodeLoad;
 
-
 typedef struct {
   int32_t     sver;      // software version
   int64_t     dnodeVer;  // dnode table version in sdb
@@ -1002,7 +1001,6 @@ typedef struct {
 
 typedef struct {
   int32_t  vgId;
-  int32_t  dnodeId;
   char     db[TSDB_DB_FNAME_LEN];
   int64_t  dbUid;
   int32_t  vgVersion;
@@ -1025,16 +1023,14 @@ typedef struct {
   int8_t   compression;
   int8_t   strict;
   int8_t   cacheLastRow;
+  int8_t   isTsma;
+  int8_t   standby;
   int8_t   replica;
   int8_t   selfIndex;
   SReplica replicas[TSDB_MAX_REPLICA];
   int32_t  numOfRetensions;
   SArray*  pRetensions;  // SRetention
-
+  void*    pTsma;
-  // for tsma
-  int8_t isTsma;
-  void*  pTsma;
-
 } SCreateVnodeReq;
 
 int32_t tSerializeSCreateVnodeReq(void* buf, int32_t bufLen, SCreateVnodeReq* pReq);
@@ -1072,8 +1068,8 @@ typedef struct {
   int8_t   walLevel;
   int8_t   strict;
   int8_t   cacheLastRow;
-  int8_t   replica;
   int8_t   selfIndex;
+  int8_t   replica;
   SReplica replicas[TSDB_MAX_REPLICA];
 } SAlterVnodeReq;
 
@@ -1786,6 +1782,15 @@ typedef struct SVCreateTbReq {
 int tEncodeSVCreateTbReq(SEncoder* pCoder, const SVCreateTbReq* pReq);
 int tDecodeSVCreateTbReq(SDecoder* pCoder, SVCreateTbReq* pReq);
 
+static FORCE_INLINE void tdDestroySVCreateTbReq(SVCreateTbReq* req) {
+  taosMemoryFreeClear(req->name);
+  if (req->type == TSDB_CHILD_TABLE) {
+    taosMemoryFreeClear(req->ctb.pTag);
+  } else if (req->type == TSDB_NORMAL_TABLE) {
+    taosMemoryFreeClear(req->ntb.schemaRow.pSchema);
+  }
+}
+
 typedef struct {
   int32_t nReqs;
   union {
@@ -1977,7 +1982,7 @@ typedef struct {
   int8_t   killConnection;
   int8_t   align[3];
   SEpSet   epSet;
-  SArray  *pQnodeList;
+  SArray*  pQnodeList;
 } SQueryHbRspBasic;
 
 typedef struct {
@@ -2207,10 +2212,8 @@ typedef struct {
   int64_t newConsumerId;
   char    subKey[TSDB_SUBSCRIBE_KEY_LEN];
   int8_t  subType;
-  // int8_t  withTbName;
+  char*   qmsg;
-  // int8_t  withSchema;
+  int64_t suid;
-  // int8_t  withTag;
-  char* qmsg;
 } SMqRebVgReq;
 
 static FORCE_INLINE int32_t tEncodeSMqRebVgReq(void** buf, const SMqRebVgReq* pReq) {
@@ -2221,11 +2224,10 @@ static FORCE_INLINE int32_t tEncodeSMqRebVgReq(void** buf, const SMqRebVgReq* pR
   tlen += taosEncodeFixedI64(buf, pReq->newConsumerId);
   tlen += taosEncodeString(buf, pReq->subKey);
   tlen += taosEncodeFixedI8(buf, pReq->subType);
-  // tlen += taosEncodeFixedI8(buf, pReq->withTbName);
-  // tlen += taosEncodeFixedI8(buf, pReq->withSchema);
-  // tlen += taosEncodeFixedI8(buf, pReq->withTag);
   if (pReq->subType == TOPIC_SUB_TYPE__COLUMN) {
     tlen += taosEncodeString(buf, pReq->qmsg);
+  } else if (pReq->subType == TOPIC_SUB_TYPE__TABLE) {
+    tlen += taosEncodeFixedI64(buf, pReq->suid);
   }
   return tlen;
 }
@@ -2237,11 +2239,10 @@ static FORCE_INLINE void* tDecodeSMqRebVgReq(const void* buf, SMqRebVgReq* pReq)
   buf = taosDecodeFixedI64(buf, &pReq->newConsumerId);
   buf = taosDecodeStringTo(buf, pReq->subKey);
   buf = taosDecodeFixedI8(buf, &pReq->subType);
-  // buf = taosDecodeFixedI8(buf, &pReq->withTbName);
-  // buf = taosDecodeFixedI8(buf, &pReq->withSchema);
-  // buf = taosDecodeFixedI8(buf, &pReq->withTag);
   if (pReq->subType == TOPIC_SUB_TYPE__COLUMN) {
     buf = taosDecodeString(buf, &pReq->qmsg);
+  } else if (pReq->subType == TOPIC_SUB_TYPE__TABLE) {
+    buf = taosDecodeFixedI64(buf, &pReq->suid);
   }
   return (void*)buf;
 }
@@ -2305,6 +2306,7 @@ typedef struct {
   int8_t   intervalUnit;  // MACRO: TIME_UNIT_XXX
   int8_t   slidingUnit;   // MACRO: TIME_UNIT_XXX
   int8_t   timezoneInt;   // sma data expired if timezone changes.
+  int32_t  dstVgId;
   char     indexName[TSDB_INDEX_NAME_LEN];
   int32_t  exprLen;
   int32_t  tagsFilterLen;
@@ -2474,7 +2476,7 @@ static FORCE_INLINE void* tDecodeSMqSubVgEp(void* buf, SMqSubVgEp* pVgEp) {
 
 typedef struct {
   char           topic[TSDB_TOPIC_FNAME_LEN];
-  int8_t         isSchemaAdaptive;
+  char           db[TSDB_DB_FNAME_LEN];
   SArray*        vgs;  // SArray<SMqSubVgEp>
   SSchemaWrapper schema;
 } SMqSubTopicEp;
@@ -2482,7 +2484,7 @@ typedef struct {
 static FORCE_INLINE int32_t tEncodeSMqSubTopicEp(void** buf, const SMqSubTopicEp* pTopicEp) {
   int32_t tlen = 0;
   tlen += taosEncodeString(buf, pTopicEp->topic);
-  tlen += taosEncodeFixedI8(buf, pTopicEp->isSchemaAdaptive);
+  tlen += taosEncodeString(buf, pTopicEp->db);
   int32_t sz = taosArrayGetSize(pTopicEp->vgs);
   tlen += taosEncodeFixedI32(buf, sz);
   for (int32_t i = 0; i < sz; i++) {
@@ -2495,7 +2497,7 @@ static FORCE_INLINE int32_t tEncodeSMqSubTopicEp(void** buf, const SMqSubTopicEp
 
 static FORCE_INLINE void* tDecodeSMqSubTopicEp(void* buf, SMqSubTopicEp* pTopicEp) {
   buf = taosDecodeStringTo(buf, pTopicEp->topic);
-  buf = taosDecodeFixedI8(buf, &pTopicEp->isSchemaAdaptive);
+  buf = taosDecodeStringTo(buf, pTopicEp->db);
   int32_t sz;
   buf = taosDecodeFixedI32(buf, &sz);
   pTopicEp->vgs = taosArrayInit(sz, sizeof(SMqSubVgEp));
@@ -2573,6 +2575,12 @@ static FORCE_INLINE void* tDecodeSMqDataBlkRsp(const void* buf, SMqDataBlkRsp* p
     buf = taosDecodeFixedI8(buf, &pRsp->withTbName);
     buf = taosDecodeFixedI8(buf, &pRsp->withSchema);
     buf = taosDecodeFixedI8(buf, &pRsp->withTag);
+    if (pRsp->withTbName) {
+      pRsp->blockTbName = taosArrayInit(pRsp->blockNum, sizeof(void*));
+    }
+    if (pRsp->withSchema) {
+      pRsp->blockSchema = taosArrayInit(pRsp->blockNum, sizeof(void*));
+    }
 
     for (int32_t i = 0; i < pRsp->blockNum; i++) {
       int32_t bLen = 0;
@@ -2582,20 +2590,14 @@ static FORCE_INLINE void* tDecodeSMqDataBlkRsp(const void* buf, SMqDataBlkRsp* p
       taosArrayPush(pRsp->blockDataLen, &bLen);
       taosArrayPush(pRsp->blockData, &data);
       if (pRsp->withSchema) {
-        pRsp->blockSchema = taosArrayInit(pRsp->blockNum, sizeof(void*));
         SSchemaWrapper* pSW = (SSchemaWrapper*)taosMemoryMalloc(sizeof(SSchemaWrapper));
         buf = taosDecodeSSchemaWrapper(buf, pSW);
         taosArrayPush(pRsp->blockSchema, &pSW);
-      } else {
-        pRsp->blockSchema = NULL;
       }
       if (pRsp->withTbName) {
-        pRsp->blockTbName = taosArrayInit(pRsp->blockNum, sizeof(void*));
         char* name = NULL;
         buf = taosDecodeString(buf, &name);
         taosArrayPush(pRsp->blockTbName, &name);
-      } else {
-        pRsp->blockTbName = NULL;
       }
     }
   }
@@ -2662,6 +2664,23 @@ typedef struct {
 int32_t tEncodeSVSubmitReq(SEncoder* pCoder, const SVSubmitReq* pReq);
 int32_t tDecodeSVSubmitReq(SDecoder* pCoder, SVSubmitReq* pReq);
 
+// TDMT_VND_DELETE
+typedef struct {
+  TSKEY sKey;
+  TSKEY eKey;
+
+  // super table
+  char* stbName;
+
+  // child/normal
+  char* tbName;
+} SVDeleteReq;
+
+typedef struct {
+  int32_t code;
+  // TODO
+} SVDeleteRsp;
+
 #pragma pack(pop)
 
 #ifdef __cplusplus

include/common/tmsgdef.h

@@ -221,9 +221,11 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_VND_SYNC_APPLY_MSG, "vnode-sync-apply-msg", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_SYNC_CONFIG_CHANGE, "vnode-sync-config-change", NULL, NULL)
   
-  TD_DEF_MSG_TYPE(TDMT_VND_SYNC_VNODE, "vnode-sync-vnode", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_VND_ALTER_CONFIG, "vnode-alter-config", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_ALTER_VNODE, "vnode-alter-vnode", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_VND_ALTER_REPLICA, "vnode-alter-replica", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_COMPACT_VNODE, "vnode-compact-vnode", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_VND_COMPACT, "vnode-compact", NULL, NULL)
+
+  TD_DEF_MSG_TYPE(TDMT_VND_DELETE, "vnode-delete-data", SVDeleteReq, SVDeleteRsp)
 
   // Requests handled by QNODE
   TD_NEW_MSG_SEG(TDMT_QND_MSG)

include/libs/function/functionMgt.h

@@ -156,6 +156,9 @@ bool fmIsDynamicScanOptimizedFunc(int32_t funcId);
 bool fmIsMultiResFunc(int32_t funcId);
 bool fmIsRepeatScanFunc(int32_t funcId);
 bool fmIsUserDefinedFunc(int32_t funcId);
+bool fmIsDistExecFunc(int32_t funcId);
+
+int32_t fmGetDistMethod(const SFunctionNode* pFunc, SFunctionNode** pPartialFunc, SFunctionNode** pMergeFunc);
 
 typedef enum EFuncDataRequired {
   FUNC_DATA_REQUIRED_DATA_LOAD = 1,

include/libs/nodes/nodes.h

@@ -189,6 +189,7 @@ typedef enum ENodeType {
   QUERY_NODE_LOGIC_PLAN_PROJECT,
   QUERY_NODE_LOGIC_PLAN_VNODE_MODIF,
   QUERY_NODE_LOGIC_PLAN_EXCHANGE,
+  QUERY_NODE_LOGIC_PLAN_MERGE,
   QUERY_NODE_LOGIC_PLAN_WINDOW,
   QUERY_NODE_LOGIC_PLAN_FILL,
   QUERY_NODE_LOGIC_PLAN_SORT,
@@ -206,6 +207,7 @@ typedef enum ENodeType {
   QUERY_NODE_PHYSICAL_PLAN_JOIN,
   QUERY_NODE_PHYSICAL_PLAN_AGG,
   QUERY_NODE_PHYSICAL_PLAN_EXCHANGE,
+  QUERY_NODE_PHYSICAL_PLAN_MERGE,
   QUERY_NODE_PHYSICAL_PLAN_SORT,
   QUERY_NODE_PHYSICAL_PLAN_INTERVAL,
   QUERY_NODE_PHYSICAL_PLAN_STREAM_INTERVAL,

include/libs/nodes/plannodes.h

@@ -95,9 +95,15 @@ typedef struct SVnodeModifLogicNode {
 typedef struct SExchangeLogicNode {
   SLogicNode node;
   int32_t    srcGroupId;
-  uint8_t    precision;
 } SExchangeLogicNode;
 
+typedef struct SMergeLogicNode {
+  SLogicNode node;
+  SNodeList* pMergeKeys;
+  int32_t    numOfChannels;
+  int32_t    srcGroupId;
+} SMergeLogicNode;
+
 typedef enum EWindowType { WINDOW_TYPE_INTERVAL = 1, WINDOW_TYPE_SESSION, WINDOW_TYPE_STATE } EWindowType;
 
 typedef struct SWindowLogicNode {
@@ -268,6 +274,13 @@ typedef struct SExchangePhysiNode {
   SNodeList* pSrcEndPoints;  // element is SDownstreamSource, scheduler fill by calling qSetSuplanExecutionNode
 } SExchangePhysiNode;
 
+typedef struct SMergePhysiNode {
+  SPhysiNode node;
+  SNodeList* pMergeKeys;
+  int32_t    numOfChannels;
+  int32_t    srcGroupId;
+} SMergePhysiNode;
+
 typedef struct SWinodwPhysiNode {
   SPhysiNode node;
   SNodeList* pExprs;  // these are expression list of parameter expression of function

include/libs/stream/tstream.h

@@ -61,11 +61,8 @@ enum {
 };
 
 typedef struct {
-  int8_t type;
+  int8_t      type;
-
+  int64_t     ver;
-  int32_t sourceVg;
-  int64_t sourceVer;
-
   int32_t*    dataRef;
   SSubmitReq* data;
 } SStreamDataSubmit;
@@ -83,6 +80,37 @@ typedef struct {
   int8_t type;
 } SStreamCheckpoint;
 
+typedef struct {
+  STaosQueue* queue;
+  STaosQall*  qall;
+  void*       qItem;
+  int8_t      failed;
+} SStreamQ;
+
+static FORCE_INLINE void* streamQCurItem(SStreamQ* queue) {
+  //
+  return queue->qItem;
+}
+
+static FORCE_INLINE void* streamQNextItem(SStreamQ* queue) {
+  int8_t failed = atomic_load_8(&queue->failed);
+  if (failed) {
+    ASSERT(queue->qItem != NULL);
+    return streamQCurItem(queue);
+  } else {
+    taosGetQitem(queue->qall, &queue->qItem);
+    if (queue->qItem == NULL) {
+      taosReadAllQitems(queue->queue, queue->qall);
+      taosGetQitem(queue->qall, &queue->qItem);
+    }
+    return streamQCurItem(queue);
+  }
+}
+
+static FORCE_INLINE void streamQSetFail(SStreamQ* queue) { atomic_store_8(&queue->failed, 1); }
+
+static FORCE_INLINE void streamQSetSuccess(SStreamQ* queue) { atomic_store_8(&queue->failed, 0); }
+
 static FORCE_INLINE SStreamDataSubmit* streamDataSubmitNew(SSubmitReq* pReq) {
   SStreamDataSubmit* pDataSubmit = (SStreamDataSubmit*)taosAllocateQitem(sizeof(SStreamDataSubmit), DEF_QITEM);
   if (pDataSubmit == NULL) return NULL;
@@ -111,6 +139,8 @@ static FORCE_INLINE void streamDataSubmitRefDec(SStreamDataSubmit* pDataSubmit)
   }
 }
 
+SStreamDataSubmit* streamSubmitRefClone(SStreamDataSubmit* pSubmit);
+
 int32_t streamDataBlockEncode(void** buf, const SStreamDataBlock* pOutput);
 void*   streamDataBlockDecode(const void* buf, SStreamDataBlock* pInput);
 
@@ -209,8 +239,6 @@ struct SStreamTask {
   int32_t nodeId;
   SEpSet  epSet;
 
-  // source preprocess
-
   // exec
   STaskExec exec;
 
@@ -318,8 +346,6 @@ int32_t streamDequeueOutput(SStreamTask* pTask, void** output);
 
 int32_t streamTaskRun(SStreamTask* pTask);
 
-int32_t streamTaskHandleInput(SStreamTask* pTask, void* data);
-
 int32_t streamTaskProcessRunReq(SStreamTask* pTask, SMsgCb* pMsgCb);
 int32_t streamProcessDispatchReq(SStreamTask* pTask, SMsgCb* pMsgCb, SStreamDispatchReq* pReq, SRpcMsg* pMsg);
 int32_t streamProcessDispatchRsp(SStreamTask* pTask, SMsgCb* pMsgCb, SStreamDispatchRsp* pRsp);

include/os/osDir.h

@@ -33,8 +33,19 @@ extern "C" {
 
 #ifdef WINDOWS
 #define TD_TMP_DIR_PATH "C:\\Windows\\Temp\\"
+#define TD_CFG_DIR_PATH "C:\\TDengine\\cfg\\"
+#define TD_DATA_DIR_PATH "C:\\TDengine\\data\\"
+#define TD_LOG_DIR_PATH "C:\\TDengine\\log\\"
+#elif defined(_TD_DARWIN_64)
+#define TD_TMP_DIR_PATH "/tmp/taosd/"
+#define TD_CFG_DIR_PATH "/usr/local/etc/taos/"
+#define TD_DATA_DIR_PATH "/usr/local/var/lib/taos/"
+#define TD_LOG_DIR_PATH "/usr/local/var/log/taos/"
 #else
 #define TD_TMP_DIR_PATH "/tmp/"
+#define TD_CFG_DIR_PATH "/etc/taos/"
+#define TD_DATA_DIR_PATH "/var/lib/taos/"
+#define TD_LOG_DIR_PATH "/var/log/taos/"
 #endif
 
 typedef struct TdDir *TdDirPtr;

include/util/taoserror.h

@@ -70,6 +70,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_NEED_RETRY                    TAOS_DEF_ERROR_CODE(0, 0x0028)
 #define TSDB_CODE_OUT_OF_RPC_MEMORY_QUEUE       TAOS_DEF_ERROR_CODE(0, 0x0029)
 #define TSDB_CODE_INVALID_TIMESTAMP             TAOS_DEF_ERROR_CODE(0, 0x0030)
+#define TSDB_CODE_MSG_DECODE_ERROR              TAOS_DEF_ERROR_CODE(0, 0x0031)
 
 #define TSDB_CODE_REF_NO_MEMORY                 TAOS_DEF_ERROR_CODE(0, 0x0040)
 #define TSDB_CODE_REF_FULL                      TAOS_DEF_ERROR_CODE(0, 0x0041)
@@ -84,6 +85,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_RPC_NETWORK_UNAVAIL           TAOS_DEF_ERROR_CODE(0, 0x0102)
 #define TSDB_CODE_RPC_FQDN_ERROR                TAOS_DEF_ERROR_CODE(0, 0x0103)
 #define TSDB_CODE_RPC_PORT_EADDRINUSE           TAOS_DEF_ERROR_CODE(0, 0x0104)
+#define TSDB_CODE_RPC_INDIRECT_NETWORK_UNAVAIL  TAOS_DEF_ERROR_CODE(0, 0x0105)
 
 //client
 #define TSDB_CODE_TSC_INVALID_OPERATION         TAOS_DEF_ERROR_CODE(0, 0x0200)
@@ -183,8 +185,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_MND_BNODE_NOT_EXIST           TAOS_DEF_ERROR_CODE(0, 0x0357)
 #define TSDB_CODE_MND_TOO_FEW_MNODES            TAOS_DEF_ERROR_CODE(0, 0x0358)
 #define TSDB_CODE_MND_TOO_MANY_MNODES           TAOS_DEF_ERROR_CODE(0, 0x0359)
-#define TSDB_CODE_MND_MNODE_DEPLOYED            TAOS_DEF_ERROR_CODE(0, 0x035A)
+#define TSDB_CODE_MND_CANT_DROP_MASTER          TAOS_DEF_ERROR_CODE(0, 0x035A)
-#define TSDB_CODE_MND_CANT_DROP_MASTER          TAOS_DEF_ERROR_CODE(0, 0x035B)
 
 // mnode-acct
 #define TSDB_CODE_MND_ACCT_ALREADY_EXIST        TAOS_DEF_ERROR_CODE(0, 0x0360)

include/util/tdef.h

@@ -209,7 +209,7 @@ typedef enum ELogicConditionType {
 #define TSDB_INDEX_FNAME_LEN     (TSDB_DB_FNAME_LEN + TSDB_INDEX_NAME_LEN + TSDB_NAME_DELIMITER_LEN)
 #define TSDB_TYPE_STR_MAX_LEN    32
 #define TSDB_TABLE_FNAME_LEN     (TSDB_DB_FNAME_LEN + TSDB_TABLE_NAME_LEN + TSDB_NAME_DELIMITER_LEN)
-#define TSDB_TOPIC_FNAME_LEN     TSDB_TABLE_FNAME_LEN
+#define TSDB_TOPIC_FNAME_LEN     (TSDB_ACCT_ID_LEN + TSDB_TABLE_NAME_LEN + TSDB_NAME_DELIMITER_LEN)
 #define TSDB_STREAM_FNAME_LEN    TSDB_TABLE_FNAME_LEN
 #define TSDB_SUBSCRIBE_KEY_LEN   (TSDB_CGROUP_LEN + TSDB_TOPIC_FNAME_LEN + 2)
 #define TSDB_PARTITION_KEY_LEN   (TSDB_SUBSCRIBE_KEY_LEN + 20)

include/util/tencode.h

@@ -534,6 +534,26 @@ static FORCE_INLINE int32_t tPutI64(uint8_t* p, int64_t v) {
   return sizeof(int64_t);
 }
 
+static FORCE_INLINE int32_t tPutFloat(uint8_t* p, float f) {
+  union {
+    uint32_t ui;
+    float    f;
+  } v;
+  v.f = f;
+
+  return tPutU32(p, v.ui);
+}
+
+static FORCE_INLINE int32_t tPutDouble(uint8_t* p, double d) {
+  union {
+    uint64_t ui;
+    double   d;
+  } v;
+  v.d = d;
+
+  return tPutU64(p, v.ui);
+}
+
 static FORCE_INLINE int32_t tPutU16v(uint8_t* p, uint16_t v) { tPutV(p, v); }
 
 static FORCE_INLINE int32_t tPutI16v(uint8_t* p, int16_t v) { return tPutU16v(p, ZIGZAGE(int16_t, v)); }
@@ -623,6 +643,34 @@ static FORCE_INLINE int32_t tGetI64v(uint8_t* p, int64_t* v) {
   return n;
 }
 
+static FORCE_INLINE int32_t tGetFloat(uint8_t* p, float* f) {
+  int32_t n = 0;
+
+  union {
+    uint32_t ui;
+    float    f;
+  } v;
+
+  n = tGetU32(p, &v.ui);
+
+  *f = v.f;
+  return n;
+}
+
+static FORCE_INLINE int32_t tGetDouble(uint8_t* p, double* d) {
+  int32_t n = 0;
+
+  union {
+    uint64_t ui;
+    double   d;
+  } v;
+
+  n = tGetU64(p, &v.ui);
+
+  *d = v.d;
+  return n;
+}
+
 // =====================
 static FORCE_INLINE int32_t tPutBinary(uint8_t* p, uint8_t* pData, uint32_t nData) {
   int n = 0;
@@ -646,6 +694,11 @@ static FORCE_INLINE int32_t tGetBinary(uint8_t* p, uint8_t** ppData, uint32_t* n
   return n;
 }
 
+static FORCE_INLINE int32_t tPutCStr(uint8_t* p, char* pData) {
+  return tPutBinary(p, (uint8_t*)pData, strlen(pData) + 1);
+}
+static FORCE_INLINE int32_t tGetCStr(uint8_t* p, char** ppData) { return tGetBinary(p, (uint8_t**)ppData, NULL); }
+
 #ifdef __cplusplus
 }
 #endif

include/util/ttimer.h

@@ -31,16 +31,16 @@ extern int32_t taosTmrThreads;
 
 void *taosTmrInit(int32_t maxTmr, int32_t resoultion, int32_t longest, const char *label);
 
+void taosTmrCleanUp(void *handle);
+
 tmr_h taosTmrStart(TAOS_TMR_CALLBACK fp, int32_t mseconds, void *param, void *handle);
 
 bool taosTmrStop(tmr_h tmrId);
 
-bool taosTmrStopA(tmr_h *timerId);
+bool taosTmrStopA(tmr_h *tmrId);
 
 bool taosTmrReset(TAOS_TMR_CALLBACK fp, int32_t mseconds, void *param, void *handle, tmr_h *pTmrId);
 
-void taosTmrCleanUp(void *handle);
-
 #ifdef __cplusplus
 }
 #endif

source/client/inc/clientInt.h

@@ -192,6 +192,7 @@ typedef struct SRequestSendRecvBody {
 typedef struct {
   int8_t         resType;
   char           topic[TSDB_TOPIC_FNAME_LEN];
+  char           db[TSDB_DB_FNAME_LEN];
   int32_t        vgId;
   SSchemaWrapper schema;
   int32_t        resIter;
@@ -220,7 +221,8 @@ typedef struct SRequestObj {
 void*   doFetchRows(SRequestObj* pRequest, bool setupOneRowPtr, bool convertUcs4);
 void    doSetOneRowPtr(SReqResultInfo* pResultInfo);
 void    setResPrecision(SReqResultInfo* pResInfo, int32_t precision);
-int32_t setQueryResultFromRsp(SReqResultInfo* pResultInfo, const SRetrieveTableRsp* pRsp, bool convertUcs4);
+int32_t setQueryResultFromRsp(SReqResultInfo* pResultInfo, const SRetrieveTableRsp* pRsp, bool convertUcs4,
+                              bool freeAfterUse);
 void    setResSchemaInfo(SReqResultInfo* pResInfo, const SSchema* pSchema, int32_t numOfCols);
 void    doFreeReqResultInfo(SReqResultInfo* pResInfo);
 
@@ -242,7 +244,7 @@ static FORCE_INLINE SReqResultInfo* tmqGetNextResInfo(TAOS_RES* res, bool conver
       taosMemoryFreeClear(msg->resInfo.length);
       taosMemoryFreeClear(msg->resInfo.convertBuf);
     }
-    setQueryResultFromRsp(&msg->resInfo, pRetrieve, convertUcs4);
+    setQueryResultFromRsp(&msg->resInfo, pRetrieve, convertUcs4, false);
     return &msg->resInfo;
   }
   return NULL;
@@ -320,7 +322,7 @@ SRequestObj* launchQueryImpl(SRequestObj* pRequest, SQuery* pQuery, int32_t code
 int32_t      getQueryPlan(SRequestObj* pRequest, SQuery* pQuery, SArray** pNodeList);
 int32_t      scheduleQuery(SRequestObj* pRequest, SQueryPlan* pDag, SArray* pNodeList);
 int32_t      refreshMeta(STscObj* pTscObj, SRequestObj* pRequest);
-int32_t updateQnodeList(SAppInstInfo*pInfo, SArray* pNodeList);
+int32_t      updateQnodeList(SAppInstInfo* pInfo, SArray* pNodeList);
 
 #ifdef __cplusplus
 }

source/client/src/clientImpl.c

@@ -117,7 +117,7 @@ TAOS* taos_connect_internal(const char* ip, const char* user, const char* pass,
   SAppInstInfo* p = NULL;
   if (pInst == NULL) {
     p = taosMemoryCalloc(1, sizeof(struct SAppInstInfo));
-    p->mgmtEp = epSet;    
+    p->mgmtEp = epSet;
     taosThreadMutexInit(&p->qnodeMutex, NULL);
     p->pTransporter = openTransporter(user, secretEncrypt, tsNumOfCores);
     p->pAppHbMgr = appHbMgrInit(p, key);
@@ -203,7 +203,7 @@ int32_t execLocalCmd(SRequestObj* pRequest, SQuery* pQuery) {
   SRetrieveTableRsp* pRsp = NULL;
   int32_t            code = qExecCommand(pQuery->pRoot, &pRsp);
   if (TSDB_CODE_SUCCESS == code && NULL != pRsp) {
-    code = setQueryResultFromRsp(&pRequest->body.resInfo, pRsp, false);
+    code = setQueryResultFromRsp(&pRequest->body.resInfo, pRsp, false, false);
   }
   return code;
 }
@@ -230,23 +230,23 @@ int32_t execDdlQuery(SRequestObj* pRequest, SQuery* pQuery) {
 }
 
 int compareQueryNodeLoad(const void* elem1, const void* elem2) {
-  SQueryNodeLoad *node1 = (SQueryNodeLoad *)elem1;
+  SQueryNodeLoad* node1 = (SQueryNodeLoad*)elem1;
-  SQueryNodeLoad *node2 = (SQueryNodeLoad *)elem2;
+  SQueryNodeLoad* node2 = (SQueryNodeLoad*)elem2;
 
   if (node1->load < node2->load) {
     return -1;
   }
-  
+
   return node1->load > node2->load;
 }
 
-int32_t updateQnodeList(SAppInstInfo*pInfo, SArray* pNodeList) {
+int32_t updateQnodeList(SAppInstInfo* pInfo, SArray* pNodeList) {
   taosThreadMutexLock(&pInfo->qnodeMutex);
   if (pInfo->pQnodeList) {
     taosArrayDestroy(pInfo->pQnodeList);
     pInfo->pQnodeList = NULL;
   }
-  
+
   if (pNodeList) {
     pInfo->pQnodeList = taosArrayDup(pNodeList);
     taosArraySort(pInfo->pQnodeList, compareQueryNodeLoad);
@@ -257,9 +257,9 @@ int32_t updateQnodeList(SAppInstInfo*pInfo, SArray* pNodeList) {
 }
 
 int32_t getQnodeList(SRequestObj* pRequest, SArray** pNodeList) {
-  SAppInstInfo*pInfo = pRequest->pTscObj->pAppInfo;
+  SAppInstInfo* pInfo = pRequest->pTscObj->pAppInfo;
-  int32_t code = 0;
+  int32_t       code = 0;
-  
+
   taosThreadMutexLock(&pInfo->qnodeMutex);
   if (pInfo->pQnodeList) {
     *pNodeList = taosArrayDup(pInfo->pQnodeList);
@@ -267,14 +267,14 @@ int32_t getQnodeList(SRequestObj* pRequest, SArray** pNodeList) {
   taosThreadMutexUnlock(&pInfo->qnodeMutex);
 
   if (NULL == *pNodeList) {
-    SEpSet       mgmtEpSet = getEpSet_s(&pRequest->pTscObj->pAppInfo->mgmtEp);
+    SEpSet    mgmtEpSet = getEpSet_s(&pRequest->pTscObj->pAppInfo->mgmtEp);
-    SCatalog*    pCatalog = NULL;
+    SCatalog* pCatalog = NULL;
     code = catalogGetHandle(pRequest->pTscObj->pAppInfo->clusterId, &pCatalog);
     if (TSDB_CODE_SUCCESS == code) {
       *pNodeList = taosArrayInit(5, sizeof(SQueryNodeLoad));
       code = catalogGetQnodeList(pCatalog, pRequest->pTscObj->pAppInfo->pTransporter, &mgmtEpSet, *pNodeList);
     }
-    
+
     if (TSDB_CODE_SUCCESS == code && *pNodeList) {
       code = updateQnodeList(pInfo, *pNodeList);
     }
@@ -342,7 +342,7 @@ void setResPrecision(SReqResultInfo* pResInfo, int32_t precision) {
 
 int32_t scheduleAsyncQuery(SRequestObj* pRequest, SQueryPlan* pDag, SArray* pNodeList) {
   void* pTransporter = pRequest->pTscObj->pAppInfo->pTransporter;
-  
+
   tsem_init(&schdRspSem, 0, 0);
 
   SQueryResult res = {.code = 0, .numOfRows = 0};
@@ -362,7 +362,7 @@ int32_t scheduleAsyncQuery(SRequestObj* pRequest, SQueryPlan* pDag, SArray* pNod
       return pRequest->code;
     } else {
       tsem_wait(&schdRspSem);
-      
+
       if (res.code) {
         code = res.code;
       } else {
@@ -384,7 +384,6 @@ int32_t scheduleAsyncQuery(SRequestObj* pRequest, SQueryPlan* pDag, SArray* pNod
   return pRequest->code;
 }
 
-
 int32_t scheduleQuery(SRequestObj* pRequest, SQueryPlan* pDag, SArray* pNodeList) {
   void* pTransporter = pRequest->pTscObj->pAppInfo->pTransporter;
 
@@ -807,7 +806,7 @@ void updateTargetEpSet(SMsgSendInfo* pSendInfo, STscObj* pTscObj, SRpcMsg* pMsg,
   if (NULL == pEpSet) {
     return;
   }
-  
+
   switch (pSendInfo->target.type) {
     case TARGET_TYPE_MNODE:
       if (NULL == pTscObj) {
@@ -815,7 +814,7 @@ void updateTargetEpSet(SMsgSendInfo* pSendInfo, STscObj* pTscObj, SRpcMsg* pMsg,
         return;
       }
 
-      updateEpSet_s(&pTscObj->pAppInfo->mgmtEp, pEpSet);   
+      updateEpSet_s(&pTscObj->pAppInfo->mgmtEp, pEpSet);
       break;
     case TARGET_TYPE_VNODE: {
       if (NULL == pTscObj) {
@@ -824,12 +823,13 @@ void updateTargetEpSet(SMsgSendInfo* pSendInfo, STscObj* pTscObj, SRpcMsg* pMsg,
       }
 
       SCatalog* pCatalog = NULL;
-      int32_t code = catalogGetHandle(pTscObj->pAppInfo->clusterId, &pCatalog);
+      int32_t   code = catalogGetHandle(pTscObj->pAppInfo->clusterId, &pCatalog);
       if (code != TSDB_CODE_SUCCESS) {
-        tscError("fail to get catalog handle, clusterId:%" PRIx64 ", error %s", pTscObj->pAppInfo->clusterId, tstrerror(code));
+        tscError("fail to get catalog handle, clusterId:%" PRIx64 ", error %s", pTscObj->pAppInfo->clusterId,
+                 tstrerror(code));
         return;
       }
-    
+
       catalogUpdateVgEpSet(pCatalog, pSendInfo->target.dbFName, pSendInfo->target.vgId, pEpSet);
       break;
     }
@@ -839,12 +839,11 @@ void updateTargetEpSet(SMsgSendInfo* pSendInfo, STscObj* pTscObj, SRpcMsg* pMsg,
   }
 }
 
-
 void processMsgFromServer(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet) {
   SMsgSendInfo* pSendInfo = (SMsgSendInfo*)pMsg->info.ahandle;
   assert(pMsg->info.ahandle != NULL);
   SRequestObj* pRequest = NULL;
-  STscObj* pTscObj = NULL;
+  STscObj*     pTscObj = NULL;
 
   if (pSendInfo->requestObjRefId != 0) {
     SRequestObj* pRequest = (SRequestObj*)taosAcquireRef(clientReqRefPool, pSendInfo->requestObjRefId);
@@ -971,7 +970,8 @@ void* doAsyncFetchRows(SRequestObj* pRequest, bool setupOneRowPtr, bool convertU
       return NULL;
     }
 
-    pRequest->code = setQueryResultFromRsp(&pRequest->body.resInfo, (SRetrieveTableRsp*)pResInfo->pData, convertUcs4);
+    pRequest->code =
+        setQueryResultFromRsp(&pRequest->body.resInfo, (SRetrieveTableRsp*)pResInfo->pData, convertUcs4, true);
     if (pRequest->code != TSDB_CODE_SUCCESS) {
       pResultInfo->numOfRows = 0;
       return NULL;
@@ -993,9 +993,8 @@ void* doAsyncFetchRows(SRequestObj* pRequest, bool setupOneRowPtr, bool convertU
   return pResultInfo->row;
 }
 
-
 void* doFetchRows(SRequestObj* pRequest, bool setupOneRowPtr, bool convertUcs4) {
-  //return doAsyncFetchRows(pRequest, setupOneRowPtr, convertUcs4);
+  // return doAsyncFetchRows(pRequest, setupOneRowPtr, convertUcs4);
   assert(pRequest != NULL);
 
   SReqResultInfo* pResultInfo = &pRequest->body.resInfo;
@@ -1013,7 +1012,8 @@ void* doFetchRows(SRequestObj* pRequest, bool setupOneRowPtr, bool convertUcs4)
       return NULL;
     }
 
-    pRequest->code = setQueryResultFromRsp(&pRequest->body.resInfo, (SRetrieveTableRsp*)pResInfo->pData, convertUcs4);
+    pRequest->code =
+        setQueryResultFromRsp(&pRequest->body.resInfo, (SRetrieveTableRsp*)pResInfo->pData, convertUcs4, true);
     if (pRequest->code != TSDB_CODE_SUCCESS) {
       pResultInfo->numOfRows = 0;
       return NULL;
@@ -1057,27 +1057,20 @@ static char* parseTagDatatoJson(void* p) {
     goto end;
   }
 
-  int16_t nCols = kvRowNCols(p);
+  SArray* pTagVals = NULL;
+  if (tTagToValArray((const STag*)p, &pTagVals) != 0) {
+    goto end;
+  }
+
+  int16_t nCols = taosArrayGetSize(pTagVals);
   char    tagJsonKey[256] = {0};
   for (int j = 0; j < nCols; ++j) {
-    SColIdx* pColIdx = kvRowColIdxAt(p, j);
+    STagVal* pTagVal = (STagVal*)taosArrayGet(pTagVals, j);
-    char*    val = (char*)(kvRowColVal(p, pColIdx));
-    if (j == 0) {
-      if (*val == TSDB_DATA_TYPE_NULL) {
-        string = taosMemoryCalloc(1, 8);
-        sprintf(string, "%s", TSDB_DATA_NULL_STR_L);
-        goto end;
-      }
-      continue;
-    }
-
     // json key  encode by binary
     memset(tagJsonKey, 0, sizeof(tagJsonKey));
-    memcpy(tagJsonKey, varDataVal(val), varDataLen(val));
+    memcpy(tagJsonKey, pTagVal->pKey, strlen(pTagVal->pKey));
     // json value
-    val += varDataTLen(val);
+    char type = pTagVal->type;
-    char* realData = POINTER_SHIFT(val, CHAR_BYTES);
-    char  type = *val;
     if (type == TSDB_DATA_TYPE_NULL) {
       cJSON* value = cJSON_CreateNull();
       if (value == NULL) {
@@ -1086,11 +1079,12 @@ static char* parseTagDatatoJson(void* p) {
       cJSON_AddItemToObject(json, tagJsonKey, value);
     } else if (type == TSDB_DATA_TYPE_NCHAR) {
       cJSON* value = NULL;
-      if (varDataLen(realData) > 0) {
+      if (pTagVal->nData > 0) {
-        char*   tagJsonValue = taosMemoryCalloc(varDataLen(realData), 1);
+        char*   tagJsonValue = taosMemoryCalloc(pTagVal->nData, 1);
-        int32_t length = taosUcs4ToMbs((TdUcs4*)varDataVal(realData), varDataLen(realData), tagJsonValue);
+        int32_t length = taosUcs4ToMbs((TdUcs4*)pTagVal->pData, pTagVal->nData, tagJsonValue);
         if (length < 0) {
-          tscError("charset:%s to %s. val:%s convert json value failed.", DEFAULT_UNICODE_ENCODEC, tsCharset, val);
+          tscError("charset:%s to %s. val:%s convert json value failed.", DEFAULT_UNICODE_ENCODEC, tsCharset,
+                   pTagVal->pData);
           taosMemoryFree(tagJsonValue);
           goto end;
         }
@@ -1099,7 +1093,7 @@ static char* parseTagDatatoJson(void* p) {
         if (value == NULL) {
           goto end;
         }
-      } else if (varDataLen(realData) == 0) {
+      } else if (pTagVal->nData == 0) {
         value = cJSON_CreateString("");
       } else {
         ASSERT(0);
@@ -1107,22 +1101,14 @@ static char* parseTagDatatoJson(void* p) {
 
       cJSON_AddItemToObject(json, tagJsonKey, value);
     } else if (type == TSDB_DATA_TYPE_DOUBLE) {
-      double jsonVd = *(double*)(realData);
+      double jsonVd = *(double*)(&pTagVal->i64);
       cJSON* value = cJSON_CreateNumber(jsonVd);
       if (value == NULL) {
         goto end;
       }
       cJSON_AddItemToObject(json, tagJsonKey, value);
-      //    }else if(type == TSDB_DATA_TYPE_BIGINT){
-      //      int64_t jsonVd = *(int64_t*)(realData);
-      //      cJSON* value = cJSON_CreateNumber((double)jsonVd);
-      //      if (value == NULL)
-      //      {
-      //        goto end;
-      //      }
-      //      cJSON_AddItemToObject(json, tagJsonKey, value);
     } else if (type == TSDB_DATA_TYPE_BOOL) {
-      char   jsonVd = *(char*)(realData);
+      char   jsonVd = *(char*)(&pTagVal->i64);
       cJSON* value = cJSON_CreateBool(jsonVd);
       if (value == NULL) {
         goto end;
@@ -1187,7 +1173,7 @@ static int32_t doConvertUCS4(SReqResultInfo* pResultInfo, int32_t numOfRows, int
           if (jsonInnerType == TSDB_DATA_TYPE_NULL) {
             sprintf(varDataVal(dst), "%s", TSDB_DATA_NULL_STR_L);
             varDataSetLen(dst, strlen(varDataVal(dst)));
-          } else if (jsonInnerType == TSDB_DATA_TYPE_JSON) {
+          } else if (jsonInnerType == TD_TAG_JSON) {
             char* jsonString = parseTagDatatoJson(jsonInnerData);
             STR_TO_VARSTR(dst, jsonString);
             taosMemoryFree(jsonString);
@@ -1206,10 +1192,6 @@ static int32_t doConvertUCS4(SReqResultInfo* pResultInfo, int32_t numOfRows, int
             double jsonVd = *(double*)(jsonInnerData);
             sprintf(varDataVal(dst), "%.9lf", jsonVd);
             varDataSetLen(dst, strlen(varDataVal(dst)));
-          } else if (jsonInnerType == TSDB_DATA_TYPE_BIGINT) {
-            int64_t jsonVd = *(int64_t*)(jsonInnerData);
-            sprintf(varDataVal(dst), "%" PRId64, jsonVd);
-            varDataSetLen(dst, strlen(varDataVal(dst)));
           } else if (jsonInnerType == TSDB_DATA_TYPE_BOOL) {
             sprintf(varDataVal(dst), "%s", (*((char*)jsonInnerData) == 1) ? "true" : "false");
             varDataSetLen(dst, strlen(varDataVal(dst)));
@@ -1320,11 +1302,12 @@ void resetConnectDB(STscObj* pTscObj) {
   taosThreadMutexUnlock(&pTscObj->mutex);
 }
 
-int32_t setQueryResultFromRsp(SReqResultInfo* pResultInfo, const SRetrieveTableRsp* pRsp, bool convertUcs4) {
+int32_t setQueryResultFromRsp(SReqResultInfo* pResultInfo, const SRetrieveTableRsp* pRsp, bool convertUcs4,
+                              bool freeAfterUse) {
   assert(pResultInfo != NULL && pRsp != NULL);
 
-  taosMemoryFreeClear(pResultInfo->pRspMsg);
+  if (freeAfterUse) taosMemoryFreeClear(pResultInfo->pRspMsg);
-  
+
   pResultInfo->pRspMsg = (const char*)pRsp;
   pResultInfo->pData = (void*)pRsp->data;
   pResultInfo->numOfRows = htonl(pRsp->numOfRows);

source/client/src/tmq.c

@@ -143,6 +143,7 @@ typedef struct {
 typedef struct {
   // subscribe info
   char* topicName;
+  char  db[TSDB_DB_FNAME_LEN];
 
   SArray* vgs;  // SArray<SMqClientVg>
 
@@ -1039,6 +1040,7 @@ bool tmqUpdateEp(tmq_t* tmq, int32_t epoch, SMqAskEpRsp* pRsp) {
     topic.schema = pTopicEp->schema;
     taosHashClear(pHash);
     topic.topicName = strdup(pTopicEp->topic);
+    tstrncpy(topic.db, pTopicEp->db, TSDB_DB_FNAME_LEN);
 
     tscDebug("consumer %ld update topic: %s", tmq->consumerId, topic.topicName);
     int32_t topicNumCur = taosArrayGetSize(tmq->clientTopics);
@@ -1283,7 +1285,8 @@ SMqPollReq* tmqBuildConsumeReqImpl(tmq_t* tmq, int64_t timeout, SMqClientTopic*
 SMqRspObj* tmqBuildRspFromWrapper(SMqPollRspWrapper* pWrapper) {
   SMqRspObj* pRspObj = taosMemoryCalloc(1, sizeof(SMqRspObj));
   pRspObj->resType = RES_TYPE__TMQ;
-  strncpy(pRspObj->topic, pWrapper->topicHandle->topicName, TSDB_TOPIC_FNAME_LEN);
+  tstrncpy(pRspObj->topic, pWrapper->topicHandle->topicName, TSDB_TOPIC_FNAME_LEN);
+  tstrncpy(pRspObj->db, pWrapper->topicHandle->db, TSDB_DB_FNAME_LEN);
   pRspObj->vgId = pWrapper->vgHandle->vgId;
   pRspObj->resIter = -1;
   memcpy(&pRspObj->rsp, &pWrapper->msg, sizeof(SMqDataBlkRsp));
@@ -1506,6 +1509,15 @@ const char* tmq_get_topic_name(TAOS_RES* res) {
   }
 }
 
+const char* tmq_get_db_name(TAOS_RES* res) {
+  if (TD_RES_TMQ(res)) {
+    SMqRspObj* pRspObj = (SMqRspObj*)res;
+    return strchr(pRspObj->db, '.') + 1;
+  } else {
+    return NULL;
+  }
+}
+
 int32_t tmq_get_vgroup_id(TAOS_RES* res) {
   if (TD_RES_TMQ(res)) {
     SMqRspObj* pRspObj = (SMqRspObj*)res;

source/common/src/systable.c

@@ -36,6 +36,7 @@ static const SSysDbTableSchema mnodesSchema[] = {
     {.name = "id", .bytes = 4, .type = TSDB_DATA_TYPE_INT},
     {.name = "endpoint", .bytes = TSDB_EP_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
     {.name = "role", .bytes = 12 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
+    {.name = "status", .bytes = 9 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
     {.name = "create_time", .bytes = 8, .type = TSDB_DATA_TYPE_TIMESTAMP},
 };
 

source/common/src/tdatablock.c

@@ -116,22 +116,23 @@ int32_t colDataAppend(SColumnInfoData* pColumnInfoData, uint32_t currentRow, con
 
   int32_t type = pColumnInfoData->info.type;
   if (IS_VAR_DATA_TYPE(type)) {
-    int32_t dataLen = varDataTLen(pData);
+    int32_t dataLen = 0;
     if (type == TSDB_DATA_TYPE_JSON) {
       if (*pData == TSDB_DATA_TYPE_NULL) {
-        dataLen = 0;
-      } else if (*pData == TSDB_DATA_TYPE_NCHAR) {
-        dataLen = varDataTLen(pData + CHAR_BYTES);
-      } else if (*pData == TSDB_DATA_TYPE_DOUBLE) {
-        dataLen = DOUBLE_BYTES;
-      } else if (*pData == TSDB_DATA_TYPE_BOOL) {
         dataLen = CHAR_BYTES;
-      } else if (*pData == TSDB_DATA_TYPE_JSON) {
+      } else if (*pData == TSDB_DATA_TYPE_NCHAR) {
-        dataLen = kvRowLen(pData + CHAR_BYTES);
+        dataLen = varDataTLen(pData + CHAR_BYTES) + CHAR_BYTES;
+      } else if (*pData == TSDB_DATA_TYPE_DOUBLE) {
+        dataLen = DOUBLE_BYTES + CHAR_BYTES;
+      } else if (*pData == TSDB_DATA_TYPE_BOOL) {
+        dataLen = CHAR_BYTES + CHAR_BYTES;
+      } else if (*pData == TD_TAG_JSON) {   // json string
+        dataLen = ((STag*)(pData))->len;
       } else {
         ASSERT(0);
       }
-      dataLen += CHAR_BYTES;
+    }else {
+      dataLen = varDataTLen(pData);
     }
 
     SVarColAttr* pAttr = &pColumnInfoData->varmeta;
@@ -1634,6 +1635,11 @@ int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SArray* pDataBlocks
 SSubmitReq* tdBlockToSubmit(const SArray* pBlocks, const STSchema* pTSchema, bool createTb, int64_t suid,
                             const char* stbFullName, int32_t vgId) {
   SSubmitReq* ret = NULL;
+  SArray*     tagArray = taosArrayInit(1, sizeof(STagVal));
+  if(!tagArray) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    return NULL;
+  }
 
   // cal size
   int32_t cap = sizeof(SSubmitReq);
@@ -1655,18 +1661,33 @@ SSubmitReq* tdBlockToSubmit(const SArray* pBlocks, const STSchema* pTSchema, boo
       createTbReq.type = TSDB_CHILD_TABLE;
       createTbReq.ctb.suid = suid;
 
-      SKVRowBuilder kvRowBuilder = {0};
+
-      if (tdInitKVRowBuilder(&kvRowBuilder) < 0) {
+
-        ASSERT(0);
+      STagVal tagVal = {.cid = 1,
+                        .type = TSDB_DATA_TYPE_UBIGINT,
+                        .pData = (uint8_t*)&pDataBlock->info.groupId,
+                        .nData = sizeof(uint64_t)};
+      STag*   pTag = NULL;
+      taosArrayClear(tagArray);
+      taosArrayPush(tagArray, &tagVal);
+      tTagNew(tagArray, 1, false, &pTag);
+      if (!pTag) {
+        tdDestroySVCreateTbReq(&createTbReq);
+        taosArrayDestroy(tagArray);
+        return NULL;
       }
-      tdAddColToKVRow(&kvRowBuilder, 1, &pDataBlock->info.groupId, sizeof(uint64_t));
+      createTbReq.ctb.pTag = (uint8_t*)pTag;
-      createTbReq.ctb.pTag = tdGetKVRowFromBuilder(&kvRowBuilder);
-      tdDestroyKVRowBuilder(&kvRowBuilder);
 
       int32_t code;
       tEncodeSize(tEncodeSVCreateTbReq, &createTbReq, schemaLen, code);
-      if (code < 0) return NULL;
+
-      taosMemoryFree(cname);
+      tdDestroySVCreateTbReq(&createTbReq);
+
+      if (code < 0) {
+        tdDestroySVCreateTbReq(&createTbReq);
+        taosArrayDestroy(tagArray);
+        return NULL;
+      }
     }
 
     cap += sizeof(SSubmitBlk) + schemaLen + rows * maxLen;
@@ -1709,22 +1730,42 @@ SSubmitReq* tdBlockToSubmit(const SArray* pBlocks, const STSchema* pTSchema, boo
       createTbReq.type = TSDB_CHILD_TABLE;
       createTbReq.ctb.suid = suid;
 
-      SKVRowBuilder kvRowBuilder = {0};
+      STagVal tagVal = {.cid = 1,
-      if (tdInitKVRowBuilder(&kvRowBuilder) < 0) {
+                        .type = TSDB_DATA_TYPE_UBIGINT,
-        ASSERT(0);
+                        .pData = (uint8_t*)&pDataBlock->info.groupId,
+                        .nData = sizeof(uint64_t)};
+      taosArrayClear(tagArray);
+      taosArrayPush(tagArray, &tagVal);
+      STag* pTag = NULL;
+      tTagNew(tagArray, 1, false, &pTag);
+      if (!pTag) {
+        tdDestroySVCreateTbReq(&createTbReq);
+        taosArrayDestroy(tagArray);
+        taosMemoryFreeClear(ret);
+        return NULL;
       }
-      tdAddColToKVRow(&kvRowBuilder, 1, &pDataBlock->info.groupId, sizeof(uint64_t));
+      createTbReq.ctb.pTag = (uint8_t*)pTag;
-      createTbReq.ctb.pTag = tdGetKVRowFromBuilder(&kvRowBuilder);
-      tdDestroyKVRowBuilder(&kvRowBuilder);
 
       int32_t code;
       tEncodeSize(tEncodeSVCreateTbReq, &createTbReq, schemaLen, code);
-      if (code < 0) return NULL;
+      if (code < 0) {
+        tdDestroySVCreateTbReq(&createTbReq);
+        taosArrayDestroy(tagArray);
+        taosMemoryFreeClear(ret);
+        return NULL;
+      }
 
       SEncoder encoder = {0};
       tEncoderInit(&encoder, blockData, schemaLen);
-      if (tEncodeSVCreateTbReq(&encoder, &createTbReq) < 0) return NULL;
+      code = tEncodeSVCreateTbReq(&encoder, &createTbReq);
       tEncoderClear(&encoder);
+      tdDestroySVCreateTbReq(&createTbReq);
+
+      if (code < 0) {
+        taosArrayDestroy(tagArray);
+        taosMemoryFreeClear(ret);
+        return NULL;
+      }
     }
     blkHead->schemaLen = htonl(schemaLen);
 
@@ -1759,5 +1800,6 @@ SSubmitReq* tdBlockToSubmit(const SArray* pBlocks, const STSchema* pTSchema, boo
   }
 
   ret->length = htonl(ret->length);
+        taosArrayDestroy(tagArray);
   return ret;
 }

source/common/src/tmsg.c

@@ -28,7 +28,7 @@
 #undef TD_MSG_SEG_CODE_
 #include "tmsgdef.h"
 
-int32_t tInitSubmitMsgIter(SSubmitReq *pMsg, SSubmitMsgIter *pIter) {
+int32_t tInitSubmitMsgIter(const SSubmitReq *pMsg, SSubmitMsgIter *pIter) {
   if (pMsg == NULL) {
     terrno = TSDB_CODE_TDB_SUBMIT_MSG_MSSED_UP;
     return -1;
@@ -165,7 +165,6 @@ int32_t tDecodeSQueryNodeLoad(SDecoder *pDecoder, SQueryNodeLoad *pLoad) {
   return 0;
 }
 
-
 int32_t taosEncodeSEpSet(void **buf, const SEpSet *pEp) {
   int32_t tlen = 0;
   tlen += taosEncodeFixedI8(buf, pEp->inUse);
@@ -2935,7 +2934,6 @@ int32_t tSerializeSCreateVnodeReq(void *buf, int32_t bufLen, SCreateVnodeReq *pR
 
   if (tStartEncode(&encoder) < 0) return -1;
   if (tEncodeI32(&encoder, pReq->vgId) < 0) return -1;
-  if (tEncodeI32(&encoder, pReq->dnodeId) < 0) return -1;
   if (tEncodeCStr(&encoder, pReq->db) < 0) return -1;
   if (tEncodeI64(&encoder, pReq->dbUid) < 0) return -1;
   if (tEncodeI32(&encoder, pReq->vgVersion) < 0) return -1;
@@ -2958,6 +2956,7 @@ int32_t tSerializeSCreateVnodeReq(void *buf, int32_t bufLen, SCreateVnodeReq *pR
   if (tEncodeI8(&encoder, pReq->compression) < 0) return -1;
   if (tEncodeI8(&encoder, pReq->strict) < 0) return -1;
   if (tEncodeI8(&encoder, pReq->cacheLastRow) < 0) return -1;
+  if (tEncodeI8(&encoder, pReq->standby) < 0) return -1;
   if (tEncodeI8(&encoder, pReq->replica) < 0) return -1;
   if (tEncodeI8(&encoder, pReq->selfIndex) < 0) return -1;
   for (int32_t i = 0; i < TSDB_MAX_REPLICA; ++i) {
@@ -2992,7 +2991,6 @@ int32_t tDeserializeSCreateVnodeReq(void *buf, int32_t bufLen, SCreateVnodeReq *
 
   if (tStartDecode(&decoder) < 0) return -1;
   if (tDecodeI32(&decoder, &pReq->vgId) < 0) return -1;
-  if (tDecodeI32(&decoder, &pReq->dnodeId) < 0) return -1;
   if (tDecodeCStrTo(&decoder, pReq->db) < 0) return -1;
   if (tDecodeI64(&decoder, &pReq->dbUid) < 0) return -1;
   if (tDecodeI32(&decoder, &pReq->vgVersion) < 0) return -1;
@@ -3015,6 +3013,7 @@ int32_t tDeserializeSCreateVnodeReq(void *buf, int32_t bufLen, SCreateVnodeReq *
   if (tDecodeI8(&decoder, &pReq->compression) < 0) return -1;
   if (tDecodeI8(&decoder, &pReq->strict) < 0) return -1;
   if (tDecodeI8(&decoder, &pReq->cacheLastRow) < 0) return -1;
+  if (tDecodeI8(&decoder, &pReq->standby) < 0) return -1;
   if (tDecodeI8(&decoder, &pReq->replica) < 0) return -1;
   if (tDecodeI8(&decoder, &pReq->selfIndex) < 0) return -1;
   for (int32_t i = 0; i < TSDB_MAX_REPLICA; ++i) {
@@ -3054,7 +3053,7 @@ int32_t tDeserializeSCreateVnodeReq(void *buf, int32_t bufLen, SCreateVnodeReq *
 int32_t tFreeSCreateVnodeReq(SCreateVnodeReq *pReq) {
   taosArrayDestroy(pReq->pRetensions);
   pReq->pRetensions = NULL;
-  if(pReq->isTsma) {
+  if (pReq->isTsma) {
     taosMemoryFreeClear(pReq->pTsma);
   }
   return 0;
@@ -3135,8 +3134,8 @@ int32_t tSerializeSAlterVnodeReq(void *buf, int32_t bufLen, SAlterVnodeReq *pReq
   if (tEncodeI8(&encoder, pReq->walLevel) < 0) return -1;
   if (tEncodeI8(&encoder, pReq->strict) < 0) return -1;
   if (tEncodeI8(&encoder, pReq->cacheLastRow) < 0) return -1;
-  if (tEncodeI8(&encoder, pReq->replica) < 0) return -1;
   if (tEncodeI8(&encoder, pReq->selfIndex) < 0) return -1;
+  if (tEncodeI8(&encoder, pReq->replica) < 0) return -1;
   for (int32_t i = 0; i < TSDB_MAX_REPLICA; ++i) {
     SReplica *pReplica = &pReq->replicas[i];
     if (tEncodeSReplica(&encoder, pReplica) < 0) return -1;
@@ -3166,8 +3165,8 @@ int32_t tDeserializeSAlterVnodeReq(void *buf, int32_t bufLen, SAlterVnodeReq *pR
   if (tDecodeI8(&decoder, &pReq->walLevel) < 0) return -1;
   if (tDecodeI8(&decoder, &pReq->strict) < 0) return -1;
   if (tDecodeI8(&decoder, &pReq->cacheLastRow) < 0) return -1;
-  if (tDecodeI8(&decoder, &pReq->replica) < 0) return -1;
   if (tDecodeI8(&decoder, &pReq->selfIndex) < 0) return -1;
+  if (tDecodeI8(&decoder, &pReq->replica) < 0) return -1;
   for (int32_t i = 0; i < TSDB_MAX_REPLICA; ++i) {
     SReplica *pReplica = &pReq->replicas[i];
     if (tDecodeSReplica(&decoder, pReplica) < 0) return -1;
@@ -3655,6 +3654,7 @@ int32_t tEncodeTSma(SEncoder *pCoder, const STSma *pSma) {
   if (tEncodeI8(pCoder, pSma->intervalUnit) < 0) return -1;
   if (tEncodeI8(pCoder, pSma->slidingUnit) < 0) return -1;
   if (tEncodeI8(pCoder, pSma->timezoneInt) < 0) return -1;
+  if (tEncodeI32(pCoder, pSma->dstVgId) < 0) return -1;
   if (tEncodeCStr(pCoder, pSma->indexName) < 0) return -1;
   if (tEncodeI32(pCoder, pSma->exprLen) < 0) return -1;
   if (tEncodeI32(pCoder, pSma->tagsFilterLen) < 0) return -1;
@@ -3677,6 +3677,7 @@ int32_t tDecodeTSma(SDecoder *pCoder, STSma *pSma) {
   if (tDecodeI8(pCoder, &pSma->version) < 0) return -1;
   if (tDecodeI8(pCoder, &pSma->intervalUnit) < 0) return -1;
   if (tDecodeI8(pCoder, &pSma->slidingUnit) < 0) return -1;
+  if (tDecodeI32(pCoder, &pSma->dstVgId) < 0) return -1;
   if (tDecodeI8(pCoder, &pSma->timezoneInt) < 0) return -1;
   if (tDecodeCStrTo(pCoder, pSma->indexName) < 0) return -1;
   if (tDecodeI32(pCoder, &pSma->exprLen) < 0) return -1;
@@ -3910,7 +3911,7 @@ int tEncodeSVCreateTbReq(SEncoder *pCoder, const SVCreateTbReq *pReq) {
 
   if (pReq->type == TSDB_CHILD_TABLE) {
     if (tEncodeI64(pCoder, pReq->ctb.suid) < 0) return -1;
-    if (tEncodeBinary(pCoder, pReq->ctb.pTag, kvRowLen(pReq->ctb.pTag)) < 0) return -1;
+    if (tEncodeTag(pCoder, (const STag *)pReq->ctb.pTag) < 0) return -1;
   } else if (pReq->type == TSDB_NORMAL_TABLE) {
     if (tEncodeSSchemaWrapper(pCoder, &pReq->ntb.schemaRow) < 0) return -1;
   } else {
@@ -3922,8 +3923,6 @@ int tEncodeSVCreateTbReq(SEncoder *pCoder, const SVCreateTbReq *pReq) {
 }
 
 int tDecodeSVCreateTbReq(SDecoder *pCoder, SVCreateTbReq *pReq) {
-  uint32_t len;
-
   if (tStartDecode(pCoder) < 0) return -1;
 
   if (tDecodeI32v(pCoder, &pReq->flags) < 0) return -1;
@@ -3935,7 +3934,7 @@ int tDecodeSVCreateTbReq(SDecoder *pCoder, SVCreateTbReq *pReq) {
 
   if (pReq->type == TSDB_CHILD_TABLE) {
     if (tDecodeI64(pCoder, &pReq->ctb.suid) < 0) return -1;
-    if (tDecodeBinary(pCoder, &pReq->ctb.pTag, &len) < 0) return -1;
+    if (tDecodeTag(pCoder, (STag **)&pReq->ctb.pTag) < 0) return -1;
   } else if (pReq->type == TSDB_NORMAL_TABLE) {
     if (tDecodeSSchemaWrapper(pCoder, &pReq->ntb.schemaRow) < 0) return -1;
   } else {

source/dnode/mgmt/mgmt_mnode/src/mmHandle.c

@@ -219,9 +219,9 @@ SArray *mmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_VND_DROP_TASK, mmPutNodeMsgToQueryQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_QUERY_HEARTBEAT, mmPutNodeMsgToQueryQueue, 1) == NULL) goto _OVER;
 
-  if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_VNODE_RSP, mmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_CONFIG_RSP, mmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_SYNC_VNODE_RSP, mmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_REPLICA_RSP, mmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_COMPACT_VNODE_RSP, mmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_COMPACT_RSP, mmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
 
   if (dmSetMgmtHandle(pArray, TDMT_VND_SYNC_TIMEOUT, mmPutNodeMsgToSyncQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_SYNC_PING, mmPutNodeMsgToSyncQueue, 1) == NULL) goto _OVER;

source/dnode/mgmt/mgmt_vnode/src/vmHandle.c

@@ -150,20 +150,26 @@ static void vmGenerateVnodeCfg(SCreateVnodeReq *pCreate, SVnodeCfg *pCfg) {
   pCfg->tsdbCfg.minRows = pCreate->minRows;
   pCfg->tsdbCfg.maxRows = pCreate->maxRows;
   for (size_t i = 0; i < taosArrayGetSize(pCreate->pRetensions); ++i) {
-    memcpy(&pCfg->tsdbCfg.retentions[i], taosArrayGet(pCreate->pRetensions, i), sizeof(SRetention));
+    SRetention *pRetention = &pCfg->tsdbCfg.retentions[i];
+    memcpy(pRetention, taosArrayGet(pCreate->pRetensions, i), sizeof(SRetention));
+    if (i == 0) {
+      if ((pRetention->freq > 0 && pRetention->keep > 0)) pCfg->isRsma = 1;
+    }
   }
+
   pCfg->walCfg.vgId = pCreate->vgId;
   pCfg->hashBegin = pCreate->hashBegin;
   pCfg->hashEnd = pCreate->hashEnd;
   pCfg->hashMethod = pCreate->hashMethod;
 
+  pCfg->standby = pCfg->standby;
   pCfg->syncCfg.myIndex = pCreate->selfIndex;
   pCfg->syncCfg.replicaNum = pCreate->replica;
   memset(&pCfg->syncCfg.nodeInfo, 0, sizeof(pCfg->syncCfg.nodeInfo));
   for (int i = 0; i < pCreate->replica; ++i) {
-    pCfg->syncCfg.nodeInfo[i].nodePort = pCreate->replicas[i].port;
+    SNodeInfo *pNode = &pCfg->syncCfg.nodeInfo[i];
-    snprintf(pCfg->syncCfg.nodeInfo[i].nodeFqdn, sizeof(pCfg->syncCfg.nodeInfo[i].nodeFqdn), "%s",
+    pNode->nodePort = pCreate->replicas[i].port;
-             pCreate->replicas[i].fqdn);
+    tstrncpy(pNode->nodeFqdn, pCreate->replicas[i].fqdn, sizeof(pNode->nodeFqdn));
   }
 }
 
@@ -176,6 +182,8 @@ static void vmGenerateWrapperCfg(SVnodeMgmt *pMgmt, SCreateVnodeReq *pCreate, SW
 
 int32_t vmProcessCreateVnodeReq(SVnodeMgmt *pMgmt, SRpcMsg *pMsg) {
   SCreateVnodeReq createReq = {0};
+  SVnodeCfg       vnodeCfg = {0};
+  SWrapperCfg     wrapperCfg = {0};
   int32_t         code = -1;
   char            path[TSDB_FILENAME_LEN] = {0};
 
@@ -184,12 +192,9 @@ int32_t vmProcessCreateVnodeReq(SVnodeMgmt *pMgmt, SRpcMsg *pMsg) {
     return -1;
   }
 
-  dDebug("vgId:%d, create vnode req is received, tsma:%d", createReq.vgId, createReq.isTsma);
+  dDebug("vgId:%d, create vnode req is received, tsma:%d standby:%d", createReq.vgId, createReq.isTsma,
-
+         createReq.standby);
-  SVnodeCfg vnodeCfg = {0};
   vmGenerateVnodeCfg(&createReq, &vnodeCfg);
-
-  SWrapperCfg wrapperCfg = {0};
   vmGenerateWrapperCfg(pMgmt, &createReq, &wrapperCfg);
 
   SVnodeObj *pVnode = vmAcquireVnode(pMgmt, createReq.vgId);
@@ -218,9 +223,20 @@ int32_t vmProcessCreateVnodeReq(SVnodeMgmt *pMgmt, SRpcMsg *pMsg) {
   code = vmOpenVnode(pMgmt, &wrapperCfg, pImpl);
   if (code != 0) {
     dError("vgId:%d, failed to open vnode since %s", createReq.vgId, terrstr());
+    code = terrno;
     goto _OVER;
   }
 
+  if (createReq.isTsma) {
+    SMsgHead *smaMsg = createReq.pTsma;
+    uint32_t  contLen = (uint32_t)(htonl(smaMsg->contLen) - sizeof(SMsgHead));
+    if (vnodeProcessCreateTSma(pImpl, POINTER_SHIFT(smaMsg, sizeof(SMsgHead)), contLen) < 0) {
+      dError("vgId:%d, failed to create tsma since %s", createReq.vgId, terrstr());
+      code = terrno;
+      goto _OVER;
+    };
+  }
+
   code = vnodeStart(pImpl);
   if (code != 0) {
     dError("vgId:%d, failed to start sync since %s", createReq.vgId, terrstr());
@@ -228,7 +244,10 @@ int32_t vmProcessCreateVnodeReq(SVnodeMgmt *pMgmt, SRpcMsg *pMsg) {
   }
 
   code = vmWriteVnodeListToFile(pMgmt);
-  if (code != 0) goto _OVER;
+  if (code != 0) {
+    code = terrno;
+    goto _OVER;
+  }
 
 _OVER:
   if (code != 0) {
@@ -314,8 +333,9 @@ SArray *vmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_VND_TASK_DISPATCH, vmPutNodeMsgToFetchQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_TASK_RECOVER, vmPutNodeMsgToFetchQueue, 0) == NULL) goto _OVER;
 
-  if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_VNODE, vmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_REPLICA, vmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_COMPACT_VNODE, vmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_CONFIG, vmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_COMPACT, vmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_DND_CREATE_VNODE, vmPutNodeMsgToMgmtQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_DND_DROP_VNODE, vmPutNodeMsgToMgmtQueue, 0) == NULL) goto _OVER;
 

source/dnode/mnode/impl/inc/mndDef.h

@@ -168,7 +168,7 @@ typedef struct {
   int64_t    createdTime;
   int64_t    updateTime;
   SDnodeObj* pDnode;
-  SQnodeLoad load;  
+  SQnodeLoad load;
 } SQnodeObj;
 
 typedef struct {
@@ -403,26 +403,22 @@ int32_t tEncodeSMqOffsetObj(void** buf, const SMqOffsetObj* pOffset);
 void*   tDecodeSMqOffsetObj(void* buf, SMqOffsetObj* pOffset);
 
 typedef struct {
-  char    name[TSDB_TOPIC_FNAME_LEN];
+  char           name[TSDB_TOPIC_FNAME_LEN];
-  char    db[TSDB_DB_FNAME_LEN];
+  char           db[TSDB_DB_FNAME_LEN];
-  int64_t createTime;
+  int64_t        createTime;
-  int64_t updateTime;
+  int64_t        updateTime;
-  int64_t uid;
+  int64_t        uid;
-  int64_t dbUid;
+  int64_t        dbUid;
-  int32_t version;
+  int32_t        version;
-  int8_t  subType;  // column, db or stable
+  int8_t         subType;  // column, db or stable
-  // int8_t         withTbName;
-  // int8_t         withSchema;
-  // int8_t         withTag;
   SRWLatch       lock;
-  int32_t        consumerCnt;
   int32_t        sqlLen;
   int32_t        astLen;
   char*          sql;
   char*          ast;
   char*          physicalPlan;
   SSchemaWrapper schema;
-  // int32_t        refConsumerCnt;
+  int64_t        stbUid;
 } SMqTopicObj;
 
 typedef struct {
@@ -476,14 +472,12 @@ int32_t        tEncodeSMqConsumerEp(void** buf, const SMqConsumerEp* pEp);
 void*          tDecodeSMqConsumerEp(const void* buf, SMqConsumerEp* pEp);
 
 typedef struct {
-  char     key[TSDB_SUBSCRIBE_KEY_LEN];
+  char      key[TSDB_SUBSCRIBE_KEY_LEN];
-  SRWLatch lock;
+  SRWLatch  lock;
-  int64_t  dbUid;
+  int64_t   dbUid;
-  int32_t  vgNum;
+  int32_t   vgNum;
-  int8_t   subType;
+  int8_t    subType;
-  // int8_t    withTbName;
+  int64_t   stbUid;
-  // int8_t    withSchema;
-  // int8_t    withTag;
   SHashObj* consumerHash;   // consumerId -> SMqConsumerEp
   SArray*   unassignedVgs;  // SArray<SMqVgEp*>
 } SMqSubscribeObj;
@@ -535,7 +529,7 @@ typedef struct {
 } SMqRebOutputObj;
 
 typedef struct {
-  char           name[TSDB_TOPIC_FNAME_LEN];
+  char           name[TSDB_STREAM_FNAME_LEN];
   char           sourceDb[TSDB_DB_FNAME_LEN];
   char           targetDb[TSDB_DB_FNAME_LEN];
   char           targetSTbName[TSDB_TABLE_FNAME_LEN];

source/dnode/mnode/impl/inc/mndTrans.h

@@ -61,7 +61,7 @@ int32_t mndTransAppendRedoAction(STrans *pTrans, STransAction *pAction);
 int32_t mndTransAppendUndoAction(STrans *pTrans, STransAction *pAction);
 void    mndTransSetRpcRsp(STrans *pTrans, void *pCont, int32_t contLen);
 void    mndTransSetCb(STrans *pTrans, ETrnFunc startFunc, ETrnFunc stopFunc, void *param, int32_t paramLen);
-void    mndTransSetDbInfo(STrans *pTrans, SDbObj *pDb);
+void    mndTransSetDbName(STrans *pTrans, const char *dbname);
 void    mndTransSetSerial(STrans *pTrans);
 
 int32_t mndTransPrepare(SMnode *pMnode, STrans *pTrans);

source/dnode/mnode/impl/inc/mndVgroup.h

@@ -36,7 +36,7 @@ SArray *mndBuildDnodesArray(SMnode *pMnode);
 int32_t mndAddVnodeToVgroup(SMnode *pMnode, SVgObj *pVgroup, SArray *pArray);
 int32_t mndRemoveVnodeFromVgroup(SMnode *pMnode, SVgObj *pVgroup, SArray *pArray, SVnodeGid *del1, SVnodeGid *del2);
 
-void *mndBuildCreateVnodeReq(SMnode *pMnode, SDnodeObj *pDnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen);
+void *mndBuildCreateVnodeReq(SMnode *pMnode, SDnodeObj *pDnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen, bool standby);
 void *mndBuildDropVnodeReq(SMnode *pMnode, SDnodeObj *pDnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen);
 void *mndBuildAlterVnodeReq(SMnode *pMnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen);
 

source/dnode/mnode/impl/src/mndConsumer.c

@@ -306,6 +306,7 @@ static int32_t mndProcessAskEpReq(SRpcMsg *pMsg) {
       SMqTopicObj *pTopic = mndAcquireTopic(pMnode, topic);
       ASSERT(pTopic);
       taosRLockLatch(&pTopic->lock);
+      tstrncpy(topicEp.db, pTopic->db, TSDB_DB_FNAME_LEN);
       topicEp.schema.nCols = pTopic->schema.nCols;
       if (topicEp.schema.nCols) {
         topicEp.schema.pSchema = taosMemoryCalloc(topicEp.schema.nCols, sizeof(SSchema));

source/dnode/mnode/impl/src/mndDb.c

@@ -261,7 +261,7 @@ void mndReleaseDb(SMnode *pMnode, SDbObj *pDb) {
   sdbRelease(pSdb, pDb);
 }
 
-static int32_t mndAddCreateVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, SVnodeGid *pVgid) {
+static int32_t mndAddCreateVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, SVnodeGid *pVgid, bool standby) {
   STransAction action = {0};
 
   SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);
@@ -270,7 +270,7 @@ static int32_t mndAddCreateVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *p
   mndReleaseDnode(pMnode, pDnode);
 
   int32_t contLen = 0;
-  void   *pReq = mndBuildCreateVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen);
+  void   *pReq = mndBuildCreateVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen, standby);
   if (pReq == NULL) return -1;
 
   action.pCont = pReq;
@@ -286,7 +286,7 @@ static int32_t mndAddCreateVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *p
   return 0;
 }
 
-static int32_t mndAddAlterVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup) {
+static int32_t mndAddAlterVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, tmsg_t msgType) {
   STransAction action = {0};
   action.epSet = mndGetVgroupEpset(pMnode, pVgroup);
 
@@ -296,7 +296,7 @@ static int32_t mndAddAlterVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pD
 
   action.pCont = pReq;
   action.contLen = contLen;
-  action.msgType = TDMT_VND_ALTER_VNODE;
+  action.msgType = msgType;
 
   if (mndTransAppendRedoAction(pTrans, &action) != 0) {
     taosMemoryFree(pReq);
@@ -388,7 +388,7 @@ static int32_t mndCheckDbCfg(SMnode *pMnode, SDbCfg *pCfg) {
   }
 
   terrno = 0;
-  return TSDB_CODE_SUCCESS;
+  return terrno;
 }
 
 static void mndSetDefaultDbCfg(SDbCfg *pCfg) {
@@ -467,7 +467,7 @@ static int32_t mndSetCreateDbRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj
 
     for (int32_t vn = 0; vn < pVgroup->replica; ++vn) {
       SVnodeGid *pVgid = pVgroup->vnodeGid + vn;
-      if (mndAddCreateVnodeAction(pMnode, pTrans, pDb, pVgroup, pVgid) != 0) {
+      if (mndAddCreateVnodeAction(pMnode, pTrans, pDb, pVgroup, pVgid, false) != 0) {
         return -1;
       }
     }
@@ -550,7 +550,7 @@ static int32_t mndCreateDb(SMnode *pMnode, SRpcMsg *pReq, SCreateDbReq *pCreate,
 
   mDebug("trans:%d, used to create db:%s", pTrans->id, pCreate->db);
 
-  mndTransSetDbInfo(pTrans, &dbObj);
+  mndTransSetDbName(pTrans, dbObj.name);
   if (mndSetCreateDbRedoLogs(pMnode, pTrans, &dbObj, pVgroups) != 0) goto _OVER;
   if (mndSetCreateDbUndoLogs(pMnode, pTrans, &dbObj, pVgroups) != 0) goto _OVER;
   if (mndSetCreateDbCommitLogs(pMnode, pTrans, &dbObj, pVgroups) != 0) goto _OVER;
@@ -688,29 +688,37 @@ static int32_t mndSetDbCfgFromAlterDbReq(SDbObj *pDb, SAlterDbReq *pAlter) {
 static int32_t mndSetAlterDbRedoLogs(SMnode *pMnode, STrans *pTrans, SDbObj *pOld, SDbObj *pNew) {
   SSdbRaw *pRedoRaw = mndDbActionEncode(pOld);
   if (pRedoRaw == NULL) return -1;
-  if (mndTransAppendRedolog(pTrans, pRedoRaw) != 0) return -1;
+  if (mndTransAppendRedolog(pTrans, pRedoRaw) != 0) {
-  if (sdbSetRawStatus(pRedoRaw, SDB_STATUS_READY) != 0) return -1;
+    sdbFreeRaw(pRedoRaw);
+    return -1;
+  }
 
+  sdbSetRawStatus(pRedoRaw, SDB_STATUS_READY);
   return 0;
 }
 
 static int32_t mndSetAlterDbCommitLogs(SMnode *pMnode, STrans *pTrans, SDbObj *pOld, SDbObj *pNew) {
   SSdbRaw *pCommitRaw = mndDbActionEncode(pNew);
   if (pCommitRaw == NULL) return -1;
-  if (mndTransAppendCommitlog(pTrans, pCommitRaw) != 0) return -1;
+  if (mndTransAppendCommitlog(pTrans, pCommitRaw) != 0) {
-  if (sdbSetRawStatus(pCommitRaw, SDB_STATUS_READY) != 0) return -1;
+    sdbFreeRaw(pCommitRaw);
+    return -1;
+  }
 
+  sdbSetRawStatus(pCommitRaw, SDB_STATUS_READY);
   return 0;
 }
 
 static int32_t mndBuildAlterVgroupAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, SArray *pArray) {
   if (pVgroup->replica <= 0 || pVgroup->replica == pDb->cfg.replications) {
-    if (mndAddAlterVnodeAction(pMnode, pTrans, pDb, pVgroup) != 0) {
+    if (mndAddAlterVnodeAction(pMnode, pTrans, pDb, pVgroup, TDMT_VND_ALTER_CONFIG) != 0) {
       return -1;
     }
   } else {
     SVgObj newVgroup = {0};
     memcpy(&newVgroup, pVgroup, sizeof(SVgObj));
+    mndTransSetSerial(pTrans);
+
     if (newVgroup.replica < pDb->cfg.replications) {
       mInfo("db:%s, vgId:%d, will add 2 vnodes, vn:0 dnode:%d", pVgroup->dbName, pVgroup->vgId,
             pVgroup->vnodeGid[0].dnodeId);
@@ -720,9 +728,9 @@ static int32_t mndBuildAlterVgroupAction(SMnode *pMnode, STrans *pTrans, SDbObj
         return -1;
       }
       newVgroup.replica = pDb->cfg.replications;
-      if (mndAddAlterVnodeAction(pMnode, pTrans, pDb, &newVgroup) != 0) return -1;
+      if (mndAddCreateVnodeAction(pMnode, pTrans, pDb, &newVgroup, &newVgroup.vnodeGid[1], true) != 0) return -1;
-      if (mndAddCreateVnodeAction(pMnode, pTrans, pDb, &newVgroup, &newVgroup.vnodeGid[1]) != 0) return -1;
+      if (mndAddCreateVnodeAction(pMnode, pTrans, pDb, &newVgroup, &newVgroup.vnodeGid[2], true) != 0) return -1;
-      if (mndAddCreateVnodeAction(pMnode, pTrans, pDb, &newVgroup, &newVgroup.vnodeGid[2]) != 0) return -1;
+      if (mndAddAlterVnodeAction(pMnode, pTrans, pDb, &newVgroup, TDMT_VND_ALTER_REPLICA) != 0) return -1;
     } else {
       mInfo("db:%s, vgId:%d, will remove 2 vnodes", pVgroup->dbName, pVgroup->vgId);
 
@@ -733,15 +741,18 @@ static int32_t mndBuildAlterVgroupAction(SMnode *pMnode, STrans *pTrans, SDbObj
         return -1;
       }
       newVgroup.replica = pDb->cfg.replications;
-      if (mndAddAlterVnodeAction(pMnode, pTrans, pDb, &newVgroup) != 0) return -1;
+      if (mndAddAlterVnodeAction(pMnode, pTrans, pDb, &newVgroup, TDMT_VND_ALTER_REPLICA) != 0) return -1;
       if (mndAddDropVnodeAction(pMnode, pTrans, pDb, &newVgroup, &del1, true) != 0) return -1;
       if (mndAddDropVnodeAction(pMnode, pTrans, pDb, &newVgroup, &del2, true) != 0) return -1;
     }
 
     SSdbRaw *pVgRaw = mndVgroupActionEncode(&newVgroup);
     if (pVgRaw == NULL) return -1;
-    if (mndTransAppendCommitlog(pTrans, pVgRaw) != 0) return -1;
+    if (mndTransAppendCommitlog(pTrans, pVgRaw) != 0) {
-    if (sdbSetRawStatus(pVgRaw, SDB_STATUS_READY) != 0) return -1;
+      sdbFreeRaw(pVgRaw);
+      return -1;
+    }
+    sdbSetRawStatus(pVgRaw, SDB_STATUS_READY);
   }
 
   return 0;
@@ -774,18 +785,16 @@ static int32_t mndSetAlterDbRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *
 }
 
 static int32_t mndAlterDb(SMnode *pMnode, SRpcMsg *pReq, SDbObj *pOld, SDbObj *pNew) {
-  int32_t code = -1;
   STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_DB, pReq);
-  if (pTrans == NULL) goto _OVER;
+  if (pTrans == NULL) return -1;
-
   mDebug("trans:%d, used to alter db:%s", pTrans->id, pOld->name);
 
-  mndTransSetDbInfo(pTrans, pOld);
+  int32_t code = -1;
+  mndTransSetDbName(pTrans, pOld->name);
   if (mndSetAlterDbRedoLogs(pMnode, pTrans, pOld, pNew) != 0) goto _OVER;
   if (mndSetAlterDbCommitLogs(pMnode, pTrans, pOld, pNew) != 0) goto _OVER;
   if (mndSetAlterDbRedoActions(pMnode, pTrans, pOld, pNew) != 0) goto _OVER;
   if (mndTransPrepare(pMnode, pTrans) != 0) goto _OVER;
-
   code = 0;
 
 _OVER:
@@ -1040,7 +1049,7 @@ static int32_t mndDropDb(SMnode *pMnode, SRpcMsg *pReq, SDbObj *pDb) {
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to drop db:%s", pTrans->id, pDb->name);
-  mndTransSetDbInfo(pTrans, pDb);
+  mndTransSetDbName(pTrans, pDb->name);
 
   if (mndSetDropDbRedoLogs(pMnode, pTrans, pDb) != 0) goto _OVER;
   if (mndSetDropDbCommitLogs(pMnode, pTrans, pDb) != 0) goto _OVER;

source/dnode/mnode/impl/src/mndDef.c

@@ -395,10 +395,8 @@ SMqSubscribeObj *tCloneSubscribeObj(const SMqSubscribeObj *pSub) {
   taosInitRWLatch(&pSubNew->lock);
 
   pSubNew->dbUid = pSub->dbUid;
+  pSubNew->stbUid = pSub->stbUid;
   pSubNew->subType = pSub->subType;
-  /*pSubNew->withTbName = pSub->withTbName;*/
-  /*pSubNew->withSchema = pSub->withSchema;*/
-  /*pSubNew->withTag = pSub->withTag;*/
 
   pSubNew->vgNum = pSub->vgNum;
   pSubNew->consumerHash = taosHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), false, HASH_NO_LOCK);
@@ -431,9 +429,7 @@ int32_t tEncodeSubscribeObj(void **buf, const SMqSubscribeObj *pSub) {
   tlen += taosEncodeFixedI64(buf, pSub->dbUid);
   tlen += taosEncodeFixedI32(buf, pSub->vgNum);
   tlen += taosEncodeFixedI8(buf, pSub->subType);
-  /*tlen += taosEncodeFixedI8(buf, pSub->withTbName);*/
+  tlen += taosEncodeFixedI64(buf, pSub->stbUid);
-  /*tlen += taosEncodeFixedI8(buf, pSub->withSchema);*/
-  /*tlen += taosEncodeFixedI8(buf, pSub->withTag);*/
 
   void   *pIter = NULL;
   int32_t sz = taosHashGetSize(pSub->consumerHash);
@@ -458,9 +454,7 @@ void *tDecodeSubscribeObj(const void *buf, SMqSubscribeObj *pSub) {
   buf = taosDecodeFixedI64(buf, &pSub->dbUid);
   buf = taosDecodeFixedI32(buf, &pSub->vgNum);
   buf = taosDecodeFixedI8(buf, &pSub->subType);
-  /*buf = taosDecodeFixedI8(buf, &pSub->withTbName);*/
+  buf = taosDecodeFixedI64(buf, &pSub->stbUid);
-  /*buf = taosDecodeFixedI8(buf, &pSub->withSchema);*/
-  /*buf = taosDecodeFixedI8(buf, &pSub->withTag);*/
 
   int32_t sz;
   buf = taosDecodeFixedI32(buf, &sz);

source/dnode/mnode/impl/src/mndDnode.c

@@ -613,7 +613,7 @@ static int32_t mndProcessDropDnodeReq(SRpcMsg *pReq) {
 
   pMObj = mndAcquireMnode(pMnode, dropReq.dnodeId);
   if (pMObj != NULL) {
-    terrno = TSDB_CODE_MND_MNODE_DEPLOYED;
+    terrno = TSDB_CODE_MND_MNODE_NOT_EXIST;
     goto DROP_DNODE_OVER;
   }
 

source/dnode/mnode/impl/src/mndMnode.c

@@ -392,11 +392,6 @@ static int32_t mndProcessCreateMnodeReq(SRpcMsg *pReq) {
 
   mDebug("mnode:%d, start to create", createReq.dnodeId);
 
-  if (sdbGetSize(pMnode->pSdb, SDB_MNODE) >= 3) {
-    terrno = TSDB_CODE_MND_TOO_MANY_MNODES;
-    goto _OVER;
-  }
-
   pObj = mndAcquireMnode(pMnode, createReq.dnodeId);
   if (pObj != NULL) {
     terrno = TSDB_CODE_MND_MNODE_ALREADY_EXIST;
@@ -405,12 +400,22 @@ static int32_t mndProcessCreateMnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
+  if (sdbGetSize(pMnode->pSdb, SDB_MNODE) >= 3) {
+    terrno = TSDB_CODE_MND_TOO_MANY_MNODES;
+    goto _OVER;
+  }
+
   pDnode = mndAcquireDnode(pMnode, createReq.dnodeId);
   if (pDnode == NULL) {
     terrno = TSDB_CODE_MND_DNODE_NOT_EXIST;
     goto _OVER;
   }
 
+  if (!mndIsDnodeOnline(pMnode, pDnode, taosGetTimestampMs())) {
+    terrno = TSDB_CODE_NODE_OFFLINE;
+    goto _OVER;
+  }
+
   pUser = mndAcquireUser(pMnode, pReq->conn.user);
   if (pUser == NULL) {
     terrno = TSDB_CODE_MND_NO_USER_FROM_CONN;
@@ -632,11 +637,12 @@ static int32_t mndRetrieveMnodes(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pB
   int32_t    numOfRows = 0;
   int32_t    cols = 0;
   SMnodeObj *pObj = NULL;
+  ESdbStatus objStatus;
   char      *pWrite;
   int64_t    curMs = taosGetTimestampMs();
 
   while (numOfRows < rows) {
-    pShow->pIter = sdbFetch(pSdb, SDB_MNODE, pShow->pIter, (void **)&pObj);
+    pShow->pIter = sdbFetchAll(pSdb, SDB_MNODE, pShow->pIter, (void **)&pObj, &objStatus);
     if (pShow->pIter == NULL) break;
 
     cols = 0;
@@ -649,23 +655,26 @@ static int32_t mndRetrieveMnodes(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pB
     pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
     colDataAppend(pColInfo, numOfRows, b1, false);
 
-    bool        online = mndIsDnodeOnline(pMnode, pObj->pDnode, curMs);
+    const char *roles = "OFFLINE";
-    const char *roles = NULL;
     if (pObj->id == pMnode->selfDnodeId) {
       roles = syncStr(TAOS_SYNC_STATE_LEADER);
-    } else {
-      if (!online) {
-        roles = "OFFLINE";
-      } else {
-        roles = syncStr(pObj->state);
-      }
     }
-    char *b2 = taosMemoryCalloc(1, 12 + VARSTR_HEADER_SIZE);
+    if (pObj->pDnode && mndIsDnodeOnline(pMnode, pObj->pDnode, curMs)) {
+      roles = syncStr(pObj->state);
+    }
+    char b2[12 + VARSTR_HEADER_SIZE] = {0};
     STR_WITH_MAXSIZE_TO_VARSTR(b2, roles, pShow->pMeta->pSchemas[cols].bytes);
-
     pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
     colDataAppend(pColInfo, numOfRows, (const char *)b2, false);
 
+    const char *status = "READY";
+    if (objStatus == SDB_STATUS_CREATING) status = "CREATING";
+    if (objStatus == SDB_STATUS_DROPPING) status = "DROPPING";
+    char b3[9 + VARSTR_HEADER_SIZE] = {0};
+    STR_WITH_MAXSIZE_TO_VARSTR(b3, status, pShow->pMeta->pSchemas[cols].bytes);
+    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+    colDataAppend(pColInfo, numOfRows, (const char *)b3, false);
+
     pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
     colDataAppend(pColInfo, numOfRows, (const char *)&pObj->createdTime, false);
 

source/dnode/mnode/impl/src/mndSma.c

@@ -426,7 +426,7 @@ static int32_t mndSetCreateSmaVgroupRedoActions(SMnode *pMnode, STrans *pTrans,
   pVgroup->pTsma = pSmaReq;
 
   int32_t contLen = 0;
-  void   *pReq = mndBuildCreateVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen);
+  void   *pReq = mndBuildCreateVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen, false);
   taosMemoryFreeClear(pSmaReq);
   if (pReq == NULL) return -1;
 
@@ -512,7 +512,7 @@ static int32_t mndCreateSma(SMnode *pMnode, SRpcMsg *pReq, SMCreateSmaReq *pCrea
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to create sma:%s", pTrans->id, pCreate->name);
-  mndTransSetDbInfo(pTrans, pDb);
+  mndTransSetDbName(pTrans, pDb->name);
   mndTransSetSerial(pTrans);
 
   if (mndSetCreateSmaRedoLogs(pMnode, pTrans, &smaObj) != 0) goto _OVER;
@@ -757,7 +757,7 @@ static int32_t mndDropSma(SMnode *pMnode, SRpcMsg *pReq, SDbObj *pDb, SSmaObj *p
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to drop sma:%s", pTrans->id, pSma->name);
-  mndTransSetDbInfo(pTrans, pDb);
+  mndTransSetDbName(pTrans, pDb->name);
 
   if (mndSetDropSmaRedoLogs(pMnode, pTrans, pSma) != 0) goto _OVER;
   if (mndSetDropSmaVgroupRedoLogs(pMnode, pTrans, pVgroup) != 0) goto _OVER;

source/dnode/mnode/impl/src/mndStb.c

@@ -754,7 +754,7 @@ _OVER:
 }
 
 int32_t mndAddStbToTrans(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SStbObj *pStb) {
-  mndTransSetDbInfo(pTrans, pDb);
+  mndTransSetDbName(pTrans, pDb->name);
   if (mndSetCreateStbRedoLogs(pMnode, pTrans, pDb, pStb) != 0) return -1;
   if (mndSetCreateStbUndoLogs(pMnode, pTrans, pDb, pStb) != 0) return -1;
   if (mndSetCreateStbCommitLogs(pMnode, pTrans, pDb, pStb) != 0) return -1;
@@ -1375,7 +1375,7 @@ static int32_t mndAlterStb(SMnode *pMnode, SRpcMsg *pReq, const SMAlterStbReq *p
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to alter stb:%s", pTrans->id, pAlter->name);
-  mndTransSetDbInfo(pTrans, pDb);
+  mndTransSetDbName(pTrans, pDb->name);
 
   if (needRsp) {
     void* pCont = NULL;
@@ -1528,7 +1528,7 @@ static int32_t mndDropStb(SMnode *pMnode, SRpcMsg *pReq, SDbObj *pDb, SStbObj *p
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to drop stb:%s", pTrans->id, pStb->name);
-  mndTransSetDbInfo(pTrans, pDb);
+  mndTransSetDbName(pTrans, pDb->name);
 
   if (mndSetDropStbRedoLogs(pMnode, pTrans, pStb) != 0) goto _OVER;
   if (mndSetDropStbCommitLogs(pMnode, pTrans, pStb) != 0) goto _OVER;

source/dnode/mnode/impl/src/mndSubscribe.c

@@ -93,10 +93,8 @@ static SMqSubscribeObj *mndCreateSub(SMnode *pMnode, const SMqTopicObj *pTopic,
     return NULL;
   }
   pSub->dbUid = pTopic->dbUid;
+  pSub->stbUid = pTopic->stbUid;
   pSub->subType = pTopic->subType;
-  /*pSub->withTbName = pTopic->withTbName;*/
-  /*pSub->withSchema = pTopic->withSchema;*/
-  /*pSub->withTag = pTopic->withTag;*/
 
   ASSERT(pSub->unassignedVgs->size == 0);
   ASSERT(taosHashGetSize(pSub->consumerHash) == 0);
@@ -121,9 +119,7 @@ static int32_t mndBuildSubChangeReq(void **pBuf, int32_t *pLen, const SMqSubscri
   req.vgId = pRebVg->pVgEp->vgId;
   req.qmsg = pRebVg->pVgEp->qmsg;
   req.subType = pSub->subType;
-  /*req.withTbName = pSub->withTbName;*/
+  req.suid = pSub->stbUid;
-  /*req.withSchema = pSub->withSchema;*/
-  /*req.withTag = pSub->withTag;*/
   strncpy(req.subKey, pSub->key, TSDB_SUBSCRIBE_KEY_LEN);
 
   int32_t tlen = sizeof(SMsgHead) + tEncodeSMqRebVgReq(NULL, &req);

source/dnode/mnode/impl/src/mndTopic.c

@@ -96,11 +96,8 @@ SSdbRaw *mndTopicActionEncode(SMqTopicObj *pTopic) {
   SDB_SET_INT64(pRaw, dataPos, pTopic->dbUid, TOPIC_ENCODE_OVER);
   SDB_SET_INT32(pRaw, dataPos, pTopic->version, TOPIC_ENCODE_OVER);
   SDB_SET_INT8(pRaw, dataPos, pTopic->subType, TOPIC_ENCODE_OVER);
-  /*SDB_SET_INT8(pRaw, dataPos, pTopic->withTbName, TOPIC_ENCODE_OVER);*/
-  /*SDB_SET_INT8(pRaw, dataPos, pTopic->withSchema, TOPIC_ENCODE_OVER);*/
-  /*SDB_SET_INT8(pRaw, dataPos, pTopic->withTag, TOPIC_ENCODE_OVER);*/
 
-  SDB_SET_INT32(pRaw, dataPos, pTopic->consumerCnt, TOPIC_ENCODE_OVER);
+  SDB_SET_INT64(pRaw, dataPos, pTopic->stbUid, TOPIC_ENCODE_OVER);
   SDB_SET_INT32(pRaw, dataPos, pTopic->sqlLen, TOPIC_ENCODE_OVER);
   SDB_SET_BINARY(pRaw, dataPos, pTopic->sql, pTopic->sqlLen, TOPIC_ENCODE_OVER);
   SDB_SET_INT32(pRaw, dataPos, pTopic->astLen, TOPIC_ENCODE_OVER);
@@ -122,8 +119,6 @@ SSdbRaw *mndTopicActionEncode(SMqTopicObj *pTopic) {
     SDB_SET_BINARY(pRaw, dataPos, swBuf, schemaLen, TOPIC_ENCODE_OVER);
   }
 
-  /*SDB_SET_INT32(pRaw, dataPos, pTopic->refConsumerCnt, TOPIC_ENCODE_OVER);*/
-
   SDB_SET_RESERVE(pRaw, dataPos, MND_TOPIC_RESERVE_SIZE, TOPIC_ENCODE_OVER);
   SDB_SET_DATALEN(pRaw, dataPos, TOPIC_ENCODE_OVER);
 
@@ -168,12 +163,8 @@ SSdbRow *mndTopicActionDecode(SSdbRaw *pRaw) {
   SDB_GET_INT64(pRaw, dataPos, &pTopic->dbUid, TOPIC_DECODE_OVER);
   SDB_GET_INT32(pRaw, dataPos, &pTopic->version, TOPIC_DECODE_OVER);
   SDB_GET_INT8(pRaw, dataPos, &pTopic->subType, TOPIC_DECODE_OVER);
-  /*SDB_GET_INT8(pRaw, dataPos, &pTopic->withTbName, TOPIC_DECODE_OVER);*/
-  /*SDB_GET_INT8(pRaw, dataPos, &pTopic->withSchema, TOPIC_DECODE_OVER);*/
-  /*SDB_GET_INT8(pRaw, dataPos, &pTopic->withTag, TOPIC_DECODE_OVER);*/
-
-  SDB_GET_INT32(pRaw, dataPos, &pTopic->consumerCnt, TOPIC_DECODE_OVER);
 
+  SDB_GET_INT64(pRaw, dataPos, &pTopic->stbUid, TOPIC_DECODE_OVER);
   SDB_GET_INT32(pRaw, dataPos, &pTopic->sqlLen, TOPIC_DECODE_OVER);
   pTopic->sql = taosMemoryCalloc(pTopic->sqlLen, sizeof(char));
   if (pTopic->sql == NULL) {
@@ -222,8 +213,6 @@ SSdbRow *mndTopicActionDecode(SSdbRaw *pRaw) {
     pTopic->schema.pSchema = NULL;
   }
 
-  /*SDB_GET_INT32(pRaw, dataPos, &pTopic->refConsumerCnt, TOPIC_DECODE_OVER);*/
-
   SDB_GET_RESERVE(pRaw, dataPos, MND_TOPIC_RESERVE_SIZE, TOPIC_DECODE_OVER);
 
   terrno = TSDB_CODE_SUCCESS;
@@ -254,8 +243,6 @@ static int32_t mndTopicActionUpdate(SSdb *pSdb, SMqTopicObj *pOldTopic, SMqTopic
   atomic_exchange_64(&pOldTopic->updateTime, pNewTopic->updateTime);
   atomic_exchange_32(&pOldTopic->version, pNewTopic->version);
 
-  /*atomic_store_32(&pOldTopic->refConsumerCnt, pNewTopic->refConsumerCnt);*/
-
   /*taosWLockLatch(&pOldTopic->lock);*/
 
   // TODO handle update
@@ -278,18 +265,6 @@ void mndReleaseTopic(SMnode *pMnode, SMqTopicObj *pTopic) {
   sdbRelease(pSdb, pTopic);
 }
 
-#if 0
-static SDbObj *mndAcquireDbByTopic(SMnode *pMnode, char *topicName) {
-  SName name = {0};
-  tNameFromString(&name, topicName, T_NAME_ACCT | T_NAME_DB | T_NAME_TABLE);
-
-  char db[TSDB_TOPIC_FNAME_LEN] = {0};
-  tNameGetFullDbName(&name, db);
-
-  return mndAcquireDb(pMnode, db);
-}
-#endif
-
 static SDDropTopicReq *mndBuildDropTopicMsg(SMnode *pMnode, SVgObj *pVgroup, SMqTopicObj *pTopic) {
   int32_t contLen = sizeof(SDDropTopicReq);
 
@@ -341,8 +316,6 @@ static int32_t mndCreateTopic(SMnode *pMnode, SRpcMsg *pReq, SCMCreateTopicReq *
   if (pCreate->subType == TOPIC_SUB_TYPE__COLUMN) {
     topicObj.ast = strdup(pCreate->ast);
     topicObj.astLen = strlen(pCreate->ast) + 1;
-    /*topicObj.withTbName = pCreate->withTbName;*/
-    /*topicObj.withSchema = pCreate->withSchema;*/
 
     SNode *pAst = NULL;
     if (nodesStringToNode(pCreate->ast, &pAst) != 0) {
@@ -375,13 +348,16 @@ static int32_t mndCreateTopic(SMnode *pMnode, SRpcMsg *pReq, SCMCreateTopicReq *
       taosMemoryFree(topicObj.sql);
       return -1;
     }
-    /*} else if (pCreate->subType == TOPIC_SUB_TYPE__DB) {*/
+  } else if (pCreate->subType == TOPIC_SUB_TYPE__TABLE) {
-    /*topicObj.ast = NULL;*/
+    SStbObj *pStb = mndAcquireStb(pMnode, pCreate->subStbName);
-    /*topicObj.astLen = 0;*/
+    topicObj.stbUid = pStb->uid;
-    /*topicObj.physicalPlan = NULL;*/
-    /*topicObj.withTbName = 1;*/
-    /*topicObj.withSchema = 1;*/
   }
+  /*} else if (pCreate->subType == TOPIC_SUB_TYPE__DB) {*/
+  /*topicObj.ast = NULL;*/
+  /*topicObj.astLen = 0;*/
+  /*topicObj.physicalPlan = NULL;*/
+  /*topicObj.withTbName = 1;*/
+  /*topicObj.withSchema = 1;*/
 
   STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) {

source/dnode/mnode/impl/src/mndTrans.c

@@ -344,7 +344,7 @@ static SSdbRow *mndTransActionDecode(SSdbRaw *pRaw) {
       SDB_GET_INT32(pRaw, dataPos, &dataLen, _OVER)
       action.pRaw = taosMemoryMalloc(dataLen);
       if (action.pRaw == NULL) goto _OVER;
-      mTrace("raw:%p, is created", pData);
+      mTrace("raw:%p, is created", action.pRaw);
       SDB_GET_BINARY(pRaw, dataPos, (void *)action.pRaw, dataLen, _OVER);
       if (taosArrayPush(pTrans->commitActions, &action) == NULL) goto _OVER;
       action.pRaw = NULL;
@@ -619,9 +619,7 @@ void mndTransSetCb(STrans *pTrans, ETrnFunc startFunc, ETrnFunc stopFunc, void *
   pTrans->paramLen = paramLen;
 }
 
-void mndTransSetDbInfo(STrans *pTrans, SDbObj *pDb) {
+void mndTransSetDbName(STrans *pTrans, const char *dbname) { memcpy(pTrans->dbname, dbname, TSDB_DB_FNAME_LEN); }
-  memcpy(pTrans->dbname, pDb->name, TSDB_DB_FNAME_LEN);
-}
 
 void mndTransSetSerial(STrans *pTrans) { pTrans->exec = TRN_EXEC_SERIAL; }
 
@@ -753,22 +751,30 @@ static void mndTransSendRpcRsp(SMnode *pMnode, STrans *pTrans) {
       sendRsp = true;
     }
   } else {
-    if (pTrans->stage == TRN_STAGE_REDO_ACTION && pTrans->failedTimes > 6) {
+    if (pTrans->stage == TRN_STAGE_REDO_ACTION && pTrans->failedTimes > 3) {
       if (code == 0) code = TSDB_CODE_MND_TRANS_UNKNOW_ERROR;
       sendRsp = true;
     }
   }
 
   if (sendRsp && pTrans->rpcInfo.handle != NULL) {
-    void *rpcCont = rpcMallocCont(pTrans->rpcRspLen);
-    if (rpcCont != NULL) {
-      memcpy(rpcCont, pTrans->rpcRsp, pTrans->rpcRspLen);
-    }
-    taosMemoryFree(pTrans->rpcRsp);
-
     mDebug("trans:%d, send rsp, code:0x%x stage:%s app:%p", pTrans->id, code, mndTransStr(pTrans->stage),
            pTrans->rpcInfo.ahandle);
-    SRpcMsg rspMsg = {.code = code, .pCont = rpcCont, .contLen = pTrans->rpcRspLen, .info = pTrans->rpcInfo};
+    if (code == TSDB_CODE_RPC_NETWORK_UNAVAIL) {
+      code = TSDB_CODE_RPC_INDIRECT_NETWORK_UNAVAIL;
+    }
+    SRpcMsg rspMsg = {.code = code, .info = pTrans->rpcInfo};
+
+    if (pTrans->rpcRspLen != 0) {
+      void *rpcCont = rpcMallocCont(pTrans->rpcRspLen);
+      if (rpcCont != NULL) {
+        memcpy(rpcCont, pTrans->rpcRsp, pTrans->rpcRspLen);
+        rspMsg.pCont = rpcCont;
+        rspMsg.contLen = pTrans->rpcRspLen;
+      }
+      taosMemoryFree(pTrans->rpcRsp);
+    }
+
     tmsgSendRsp(&rspMsg);
     pTrans->rpcInfo.handle = NULL;
     pTrans->rpcRsp = NULL;
@@ -1000,6 +1006,9 @@ static int32_t mndTransExecuteRedoActionsSerial(SMnode *pMnode, STrans *pTrans)
         if (pAction->msgReceived) {
           if (pAction->errCode != 0 && pAction->errCode != pAction->acceptableCode) {
             code = pAction->errCode;
+            pAction->msgSent = 0;
+            pAction->msgReceived = 0;
+            mDebug("trans:%d, %s:%d execute status is reset", pTrans->id, mndTransStr(pAction->stage), action);
           }
         } else {
           code = TSDB_CODE_ACTION_IN_PROGRESS;
@@ -1025,18 +1034,23 @@ static int32_t mndTransExecuteRedoActionsSerial(SMnode *pMnode, STrans *pTrans)
     }
 
     if (code == 0) {
+      pTrans->code = 0;
       pTrans->redoActionPos++;
       mDebug("trans:%d, %s:%d is executed and need sync to other mnodes", pTrans->id, mndTransStr(pAction->stage),
              pAction->id);
       code = mndTransSync(pMnode, pTrans);
       if (code != 0) {
-        mError("trans:%d, failed to sync redoActionPos since %s", pTrans->id, terrstr());
+        pTrans->code = terrno;
+        mError("trans:%d, %s:%d is executed and failed to sync to other mnodes since %s", pTrans->id,
+               mndTransStr(pAction->stage), pAction->id, terrstr());
         break;
       }
     } else if (code == TSDB_CODE_ACTION_IN_PROGRESS) {
       mDebug("trans:%d, %s:%d is in progress and wait it finish", pTrans->id, mndTransStr(pAction->stage), pAction->id);
       break;
     } else {
+      terrno = code;
+      pTrans->code = code;
       mError("trans:%d, %s:%d failed to execute since %s", pTrans->id, mndTransStr(pAction->stage), pAction->id,
              terrstr());
       break;
@@ -1239,19 +1253,8 @@ int32_t mndKillTrans(SMnode *pMnode, STrans *pTrans) {
     return -1;
   }
 
-  int32_t size = taosArrayGetSize(pArray);
+  for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {
-
-  for (int32_t i = 0; i < size; ++i) {
     STransAction *pAction = taosArrayGet(pArray, i);
-    if (pAction == NULL) continue;
-
-    if (pAction->msgReceived == 0) {
-      mInfo("trans:%d, %s:%d set processed for kill msg received", pTrans->id, mndTransStr(pAction->stage), i);
-      pAction->msgSent = 1;
-      pAction->msgReceived = 1;
-      pAction->errCode = 0;
-    }
-
     if (pAction->errCode != 0) {
       mInfo("trans:%d, %s:%d set processed for kill msg received, errCode from %s to success", pTrans->id,
             mndTransStr(pAction->stage), i, tstrerror(pAction->errCode));
@@ -1290,9 +1293,7 @@ static int32_t mndProcessKillTransReq(SRpcMsg *pReq) {
 
   pTrans = mndAcquireTrans(pMnode, killReq.transId);
   if (pTrans == NULL) {
-    terrno = TSDB_CODE_MND_TRANS_NOT_EXIST;
+    goto _OVER;
-    mError("trans:%d, failed to kill since %s", killReq.transId, terrstr());
-    return -1;
   }
 
   code = mndKillTrans(pMnode, pTrans);
@@ -1300,9 +1301,9 @@ static int32_t mndProcessKillTransReq(SRpcMsg *pReq) {
 _OVER:
   if (code != 0) {
     mError("trans:%d, failed to kill since %s", killReq.transId, terrstr());
-    return -1;
   }
 
+  mndReleaseUser(pMnode, pUser);
   mndReleaseTrans(pMnode, pTrans);
   return code;
 }

source/dnode/mnode/impl/src/mndVgroup.c

@@ -51,9 +51,10 @@ int32_t mndInitVgroup(SMnode *pMnode) {
   };
 
   mndSetMsgHandle(pMnode, TDMT_DND_CREATE_VNODE_RSP, mndProcessCreateVnodeRsp);
-  mndSetMsgHandle(pMnode, TDMT_VND_ALTER_VNODE_RSP, mndProcessAlterVnodeRsp);
+  mndSetMsgHandle(pMnode, TDMT_VND_ALTER_REPLICA_RSP, mndProcessAlterVnodeRsp);
+  mndSetMsgHandle(pMnode, TDMT_VND_ALTER_CONFIG_RSP, mndProcessAlterVnodeRsp);
   mndSetMsgHandle(pMnode, TDMT_DND_DROP_VNODE_RSP, mndProcessDropVnodeRsp);
-  mndSetMsgHandle(pMnode, TDMT_VND_COMPACT_VNODE_RSP, mndProcessCompactVnodeRsp);
+  mndSetMsgHandle(pMnode, TDMT_VND_COMPACT_RSP, mndProcessCompactVnodeRsp);
 
   mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_VGROUP, mndRetrieveVgroups);
   mndAddShowFreeIterHandle(pMnode, TSDB_MGMT_TABLE_VGROUP, mndCancelGetNextVgroup);
@@ -188,10 +189,10 @@ void mndReleaseVgroup(SMnode *pMnode, SVgObj *pVgroup) {
   sdbRelease(pSdb, pVgroup);
 }
 
-void *mndBuildCreateVnodeReq(SMnode *pMnode, SDnodeObj *pDnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen) {
+void *mndBuildCreateVnodeReq(SMnode *pMnode, SDnodeObj *pDnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen,
+                             bool standby) {
   SCreateVnodeReq createReq = {0};
   createReq.vgId = pVgroup->vgId;
-  createReq.dnodeId = pDnode->id;
   memcpy(createReq.db, pDb->name, TSDB_DB_FNAME_LEN);
   createReq.dbUid = pDb->uid;
   createReq.vgVersion = pVgroup->version;
@@ -218,6 +219,7 @@ void *mndBuildCreateVnodeReq(SMnode *pMnode, SDnodeObj *pDnode, SDbObj *pDb, SVg
   createReq.hashMethod = pDb->cfg.hashMethod;
   createReq.numOfRetensions = pDb->cfg.numOfRetensions;
   createReq.pRetensions = pDb->cfg.pRetensions;
+  createReq.standby = standby;
   createReq.isTsma = pVgroup->isTsma;
   createReq.pTsma = pVgroup->pTsma;
 
@@ -276,7 +278,6 @@ void *mndBuildAlterVnodeReq(SMnode *pMnode, SDbObj *pDb, SVgObj *pVgroup, int32_
   alterReq.strict = pDb->cfg.strict;
   alterReq.cacheLastRow = pDb->cfg.cacheLastRow;
   alterReq.replica = pVgroup->replica;
-  alterReq.selfIndex = -1;
 
   for (int32_t v = 0; v < pVgroup->replica; ++v) {
     SReplica  *pReplica = &alterReq.replicas[v];
@@ -292,13 +293,6 @@ void *mndBuildAlterVnodeReq(SMnode *pMnode, SDbObj *pDb, SVgObj *pVgroup, int32_
     mndReleaseDnode(pMnode, pVgidDnode);
   }
 
-#if 0
-  if (alterReq.selfIndex == -1) {
-    terrno = TSDB_CODE_MND_APP_ERROR;
-    return NULL;
-  }
-#endif
-
   int32_t contLen = tSerializeSAlterVnodeReq(NULL, 0, &alterReq);
   if (contLen < 0) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
@@ -538,7 +532,7 @@ int32_t mndAddVnodeToVgroup(SMnode *pMnode, SVgObj *pVgroup, SArray *pArray) {
 
     SVnodeGid *pVgid = &pVgroup->vnodeGid[maxPos];
     pVgid->dnodeId = pDnode->id;
-    pVgid->role = TAOS_SYNC_STATE_FOLLOWER;
+    pVgid->role = TAOS_SYNC_STATE_ERROR;
     pDnode->numOfVnodes++;
 
     mInfo("db:%s, vgId:%d, vnode_index:%d dnode:%d is added", pVgroup->dbName, pVgroup->vgId, maxPos, pVgid->dnodeId);
@@ -551,14 +545,13 @@ int32_t mndAddVnodeToVgroup(SMnode *pMnode, SVgObj *pVgroup, SArray *pArray) {
 }
 
 int32_t mndRemoveVnodeFromVgroup(SMnode *pMnode, SVgObj *pVgroup, SArray *pArray, SVnodeGid *del1, SVnodeGid *del2) {
-  int32_t removedNum = 0;
-
   taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);
   for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {
     SDnodeObj *pDnode = taosArrayGet(pArray, i);
     mDebug("dnode:%d, equivalent vnodes:%d", pDnode->id, pDnode->numOfVnodes);
   }
 
+  int32_t removedNum = 0;
   for (int32_t d = taosArrayGetSize(pArray) - 1; d >= 0; --d) {
     SDnodeObj *pDnode = taosArrayGet(pArray, d);
 
@@ -664,6 +657,7 @@ static int32_t mndRetrieveVgroups(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *p
   int32_t numOfRows = 0;
   SVgObj *pVgroup = NULL;
   int32_t cols = 0;
+  int64_t curMs = taosGetTimestampMs();
 
   SDbObj *pDb = NULL;
   if (strlen(pShow->db) > 0) {
@@ -703,12 +697,15 @@ static int32_t mndRetrieveVgroups(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *p
       if (i < pVgroup->replica) {
         colDataAppend(pColInfo, numOfRows, (const char *)&pVgroup->vnodeGid[i].dnodeId, false);
 
+        bool       online = false;
+        SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgroup->vnodeGid[i].dnodeId);
+        if (pDnode != NULL) {
+          online = mndIsDnodeOnline(pMnode, pDnode, curMs);
+          mndReleaseDnode(pMnode, pDnode);
+        }
+
         char        buf1[20] = {0};
-        SDnodeObj *pDnodeObj = mndAcquireDnode(pMnode, pVgroup->vnodeGid[i].dnodeId);
+        const char *role = online ? syncStr(pVgroup->vnodeGid[i].role) : "OFFLINE";
-        ASSERT(pDnodeObj != NULL);
-        bool isOffLine = !mndIsDnodeOnline(pMnode, pDnodeObj, taosGetTimestampMs());
-        const char *role = isOffLine ? "OFFLINE" : syncStr(pVgroup->vnodeGid[i].role);
-        
         STR_WITH_MAXSIZE_TO_VARSTR(buf1, role, pShow->pMeta->pSchemas[cols].bytes);
 
         pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);

source/dnode/mnode/impl/test/trans/trans2.cpp

@@ -128,7 +128,7 @@ class MndTestTrans2 : public ::testing::Test {
     mndTransSetCb(pTrans, TRANS_START_FUNC_TEST, TRANS_STOP_FUNC_TEST, param, strlen(param) + 1);
 
     if (pDb != NULL) {
-      mndTransSetDbInfo(pTrans, pDb);
+      mndTransSetDbName(pTrans, pDb->name);
     }
 
     int32_t code = mndTransPrepare(pMnode, pTrans);
@@ -201,7 +201,7 @@ class MndTestTrans2 : public ::testing::Test {
     }
 
     if (pDb != NULL) {
-      mndTransSetDbInfo(pTrans, pDb);
+      mndTransSetDbName(pTrans, pDb->name);
     }
 
     int32_t code = mndTransPrepare(pMnode, pTrans);

source/dnode/mnode/sdb/inc/sdb.h

@@ -301,6 +301,7 @@ void sdbRelease(SSdb *pSdb, void *pObj);
  * @return void* The next iterator of the table.
  */
 void *sdbFetch(SSdb *pSdb, ESdbType type, void *pIter, void **ppObj);
+void *sdbFetchAll(SSdb *pSdb, ESdbType type, void *pIter, void **ppObj, ESdbStatus *status) ;
 
 /**
  * @brief Cancel a traversal

source/dnode/mnode/sdb/src/sdbHash.c

@@ -368,6 +368,34 @@ void *sdbFetch(SSdb *pSdb, ESdbType type, void *pIter, void **ppObj) {
   return ppRow;
 }
 
+void *sdbFetchAll(SSdb *pSdb, ESdbType type, void *pIter, void **ppObj, ESdbStatus *status) {
+  *ppObj = NULL;
+
+  SHashObj *hash = sdbGetHash(pSdb, type);
+  if (hash == NULL) return NULL;
+
+  TdThreadRwlock *pLock = &pSdb->locks[type];
+  taosThreadRwlockRdlock(pLock);
+
+  SSdbRow **ppRow = taosHashIterate(hash, pIter);
+  while (ppRow != NULL) {
+    SSdbRow *pRow = *ppRow;
+    if (pRow == NULL) {
+      ppRow = taosHashIterate(hash, ppRow);
+      continue;
+    }
+
+    atomic_add_fetch_32(&pRow->refCount, 1);
+    sdbPrintOper(pSdb, pRow, "fetch");
+    *ppObj = pRow->pObj;
+    *status = pRow->status;
+    break;
+  }
+  taosThreadRwlockUnlock(pLock);
+
+  return ppRow;
+}
+
 void sdbCancelFetch(SSdb *pSdb, void *pIter) {
   if (pIter == NULL) return;
   SSdbRow  *pRow = *(SSdbRow **)pIter;

source/dnode/vnode/CMakeLists.txt

@@ -35,7 +35,6 @@ target_sources(
     "src/sma/smaTimeRange.c"
 
     # tsdb
-    # "src/tsdb/tsdbTDBImpl.c"
     "src/tsdb/tsdbCommit.c"
     "src/tsdb/tsdbCommit2.c"
     "src/tsdb/tsdbFile.c"
@@ -45,17 +44,18 @@ target_sources(
     "src/tsdb/tsdbMemTable2.c"
     "src/tsdb/tsdbRead.c"
     "src/tsdb/tsdbReadImpl.c"
-    # "src/tsdb/tsdbSma.c"
     "src/tsdb/tsdbWrite.c"
     "src/tsdb/tsdbSnapshot.c"
 
     # tq
     "src/tq/tq.c"
     "src/tq/tqExec.c"
-    "src/tq/tqCommit.c"
+    "src/tq/tqMeta.c"
+    "src/tq/tqRead.c"
     "src/tq/tqOffset.c"
     "src/tq/tqPush.c"
-    "src/tq/tqRead.c"
+    "src/tq/tqSink.c"
+    "src/tq/tqCommit.c"
 )
 target_include_directories(
     vnode

source/dnode/vnode/inc/vnode.h

@@ -68,6 +68,7 @@ void    vnodeGetInfo(SVnode *pVnode, const char **dbname, int32_t *vgId);
 int32_t vnodeSnapshotReaderOpen(SVnode *pVnode, SVSnapshotReader **ppReader, int64_t sver, int64_t ever);
 int32_t vnodeSnapshotReaderClose(SVSnapshotReader *pReader);
 int32_t vnodeSnapshotRead(SVSnapshotReader *pReader, const void **ppData, uint32_t *nData);
+int32_t vnodeProcessCreateTSma(SVnode *pVnode, void *pCont, uint32_t contLen);
 
 // meta
 typedef struct SMeta       SMeta;  // todo: remove
@@ -78,13 +79,13 @@ void        metaReaderInit(SMetaReader *pReader, SMeta *pMeta, int32_t flags);
 void        metaReaderClear(SMetaReader *pReader);
 int32_t     metaGetTableEntryByUid(SMetaReader *pReader, tb_uid_t uid);
 int32_t     metaReadNext(SMetaReader *pReader);
-const void *metaGetTableTagVal(SMetaEntry *pEntry, int16_t cid);
+const void *metaGetTableTagVal(SMetaEntry *pEntry, int16_t type, STagVal *tagVal);
 
 typedef struct SMetaFltParam {
   tb_uid_t suid;
   int16_t  cid;
   int16_t  type;
-  char *   val;
+  char    *val;
   bool     reverse;
   int (*filterFunc)(void *a, void *b, int16_t type);
 
@@ -118,7 +119,8 @@ tsdbReaderT  tsdbQueryCacheLast(SVnode *pVnode, SQueryTableDataCond *pCond, STab
 int32_t      tsdbGetFileBlocksDistInfo(tsdbReaderT *pReader, STableBlockDistInfo *pTableBlockInfo);
 bool         isTsdbCacheLastRow(tsdbReaderT *pReader);
 int32_t      tsdbGetAllTableList(SMeta *pMeta, uint64_t uid, SArray *list);
-void *       tsdbGetIdx(SMeta *pMeta);
+int32_t      tsdbGetCtbIdList(SMeta *pMeta, int64_t suid, SArray *list);
+void        *tsdbGetIdx(SMeta *pMeta);
 int64_t      tsdbGetNumOfRowsInMemTable(tsdbReaderT *pHandle);
 
 bool    tsdbNextDataBlock(tsdbReaderT pTsdbReadHandle);
@@ -172,7 +174,9 @@ struct SVnodeCfg {
   bool     isHeap;
   bool     isWeak;
   int8_t   isTsma;
+  int8_t   isRsma;
   int8_t   hashMethod;
+  int8_t   standby;
   STsdbCfg tsdbCfg;
   SWalCfg  walCfg;
   SSyncCfg syncCfg;
@@ -189,7 +193,7 @@ struct SMetaEntry {
   int64_t  version;
   int8_t   type;
   tb_uid_t uid;
-  char *   name;
+  char    *name;
   union {
     struct {
       SSchemaWrapper schemaRow;
@@ -217,17 +221,17 @@ struct SMetaEntry {
 
 struct SMetaReader {
   int32_t    flags;
-  SMeta *    pMeta;
+  SMeta     *pMeta;
   SDecoder   coder;
   SMetaEntry me;
-  void *     pBuf;
+  void      *pBuf;
   int32_t    szBuf;
 };
 
 struct SMTbCursor {
-  TBC *       pDbc;
+  TBC        *pDbc;
-  void *      pKey;
+  void       *pKey;
-  void *      pVal;
+  void       *pVal;
   int32_t     kLen;
   int32_t     vLen;
   SMetaReader mr;

source/dnode/vnode/src/inc/meta.h

@@ -117,7 +117,7 @@ typedef struct {
 } SSmaIdxKey;
 
 // metaTable ==================
-int metaCreateTagIdxKey(tb_uid_t suid, int32_t cid, const void* pTagData, int8_t type, tb_uid_t uid,
+int metaCreateTagIdxKey(tb_uid_t suid, int32_t cid, const void* pTagData, int32_t nTagData,  int8_t type, tb_uid_t uid,
                         STagIdxKey** ppTagIdxKey, int32_t* nTagIdxKey);
 
 #ifndef META_REFACT

source/dnode/vnode/src/inc/sma.h

@@ -219,7 +219,7 @@ static int32_t  tdInitSmaEnv(SSma *pSma, int8_t smaType, const char *path, SDisk
 void *tdFreeRSmaInfo(SRSmaInfo *pInfo);
 
 int32_t tdProcessTSmaCreateImpl(SSma *pSma, int64_t version, const char *pMsg);
-int32_t tdUpdateExpiredWindowImpl(SSma *pSma, SSubmitReq *pMsg, int64_t version);
+int32_t tdUpdateExpiredWindowImpl(SSma *pSma, const SSubmitReq *pMsg, int64_t version);
 // TODO: This is the basic params, and should wrap the params to a queryHandle.
 int32_t tdGetTSmaDataImpl(SSma *pSma, char *pData, int64_t indexUid, TSKEY querySKey, int32_t nMaxResult);
 
@@ -227,4 +227,4 @@ int32_t tdGetTSmaDataImpl(SSma *pSma, char *pData, int64_t indexUid, TSKEY query
 }
 #endif
 
-#endif /*_TD_VNODE_SMA_H_*/
+#endif /*_TD_VNODE_SMA_H_*/

source/dnode/vnode/src/inc/tq.h

@@ -66,33 +66,23 @@ struct STqReadHandle {
 // tqPush
 
 typedef struct {
-  int64_t        consumerId;
+  // msg info
-  int32_t        epoch;
-  int32_t        skipLogNum;
-  int64_t        reqOffset;
-  SRpcHandleInfo info;
-  SRWLatch       lock;
-} STqPushHandle;
-
-#if 0
-typedef struct {
-  char    subKey[TSDB_SUBSCRIBE_KEY_LEN];
   int64_t consumerId;
+  int64_t reqOffset;
+  int64_t processedVer;
   int32_t epoch;
-  int8_t  subType;
+  int32_t skipLogNum;
-  // int8_t        withTbName;
+  // rpc info
-  // int8_t        withSchema;
+  int64_t        reqId;
-  // int8_t        withTag;
+  SRpcHandleInfo rpcInfo;
-  char*         qmsg;
+  tmr_h          timerId;
-  SHashObj*     pDropTbUid;
+  int8_t         tmrStopped;
-  STqPushHandle pushHandle;
+  // exec
-  // SRWLatch        lock;
+  int8_t   inputStatus;
-  SWalReadHandle* pWalReader;
+  int8_t   execStatus;
-  // task number should be the same with fetch thread
+  SStreamQ inputQ;
-  STqReadHandle* pExecReader[5];
+  SRWLatch lock;
-  qTaskInfo_t    task[5];
+} STqPushHandle;
-} STqExec;
-#endif
 
 // tqExec
 
@@ -154,28 +144,29 @@ typedef struct {
 
 static STqMgmt tqMgmt = {0};
 
-// init once
+// tqRead
-int  tqInit();
-void tqCleanUp();
-
-// int32_t tEncodeSTqExec(SEncoder* pEncoder, const STqExec* pExec);
-// int32_t tDecodeSTqExec(SDecoder* pDecoder, STqExec* pExec);
-
-int32_t tEncodeSTqHandle(SEncoder* pEncoder, const STqHandle* pHandle);
-int32_t tDecodeSTqHandle(SDecoder* pDecoder, STqHandle* pHandle);
-
 int64_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, SWalHead** pHeadWithCkSum);
 
+// tqExec
 int32_t tqDataExec(STQ* pTq, STqExecHandle* pExec, SSubmitReq* pReq, SMqDataBlkRsp* pRsp, int32_t workerId);
+int32_t tqSendPollRsp(STQ* pTq, const SRpcMsg* pMsg, const SMqPollReq* pReq, const SMqDataBlkRsp* pRsp);
+
+// tqMeta
+int32_t tqMetaOpen(STQ* pTq);
+int32_t tqMetaClose(STQ* pTq);
+int32_t tqMetaSaveHandle(STQ* pTq, const char* key, const STqHandle* pHandle);
+int32_t tqMetaDeleteHandle(STQ* pTq, const char* key);
+
+// tqSink
+void tqTableSink(SStreamTask* pTask, void* vnode, int64_t ver, void* data);
 
 // tqOffset
-STqOffsetStore* STqOffsetOpen(STqOffsetCfg*);
+STqOffsetStore* tqOffsetOpen(STqOffsetCfg*);
-void            STqOffsetClose(STqOffsetStore*);
+void            tqOffsetClose(STqOffsetStore*);
-
+int64_t         tqOffsetFetch(STqOffsetStore* pStore, const char* subscribeKey);
-int64_t tqOffsetFetch(STqOffsetStore* pStore, const char* subscribeKey);
+int32_t         tqOffsetCommit(STqOffsetStore* pStore, const char* subscribeKey, int64_t offset);
-int32_t tqOffsetCommit(STqOffsetStore* pStore, const char* subscribeKey, int64_t offset);
+int32_t         tqOffsetPersist(STqOffsetStore* pStore, const char* subscribeKey);
-int32_t tqOffsetPersist(STqOffsetStore* pStore, const char* subscribeKey);
+int32_t         tqOffsetPersistAll(STqOffsetStore* pStore);
-int32_t tqOffsetPersistAll(STqOffsetStore* pStore);
 
 #ifdef __cplusplus
 }

source/dnode/vnode/src/inc/tsdb.h

@@ -32,14 +32,27 @@ extern "C" {
 #define tsdbTrace(...) do { if (tsdbDebugFlag & DEBUG_TRACE) { taosPrintLog("TSDB ", DEBUG_TRACE, tsdbDebugFlag, __VA_ARGS__); }} while(0)
 // clang-format on
 
+typedef struct TSDBROW TSDBROW;
+typedef struct TSDBKEY TSDBKEY;
+typedef struct SDelOp  SDelOp;
+
+static int tsdbKeyCmprFn(const void *p1, const void *p2);
+
+// tsdbMemTable2.c ==============================================================================================
+typedef struct SMemTable SMemTable;
+
+int32_t tsdbMemTableCreate2(STsdb *pTsdb, SMemTable **ppMemTable);
+void    tsdbMemTableDestroy2(SMemTable *pMemTable);
+
 // tsdbMemTable ================
+typedef struct STsdbRow      STsdbRow;
 typedef struct STbData       STbData;
 typedef struct STsdbMemTable STsdbMemTable;
 typedef struct SMergeInfo    SMergeInfo;
 typedef struct STable        STable;
 
 int  tsdbMemTableCreate(STsdb *pTsdb, STsdbMemTable **ppMemTable);
-void tsdbMemTableDestroy(STsdb *pTsdb, STsdbMemTable *pMemTable);
+void tsdbMemTableDestroy(STsdbMemTable *pMemTable);
 int  tsdbLoadDataFromCache(STsdb *pTsdb, STable *pTable, SSkipListIterator *pIter, TSKEY maxKey, int maxRowsToRead,
                            SDataCols *pCols, TKEY *filterKeys, int nFilterKeys, bool keepDup, SMergeInfo *pMergeInfo);
 
@@ -845,6 +858,42 @@ static FORCE_INLINE int tsdbUnLockFS(STsdbFS *pFs) {
   return 0;
 }
 
+struct TSDBROW {
+  int64_t version;
+  STSRow2 tsRow;
+};
+
+struct TSDBKEY {
+  int64_t version;
+  TSKEY   ts;
+};
+
+struct SDelOp {
+  int64_t version;
+  TSKEY   sKey;  // included
+  TSKEY   eKey;  // included
+  SDelOp *pNext;
+};
+
+static FORCE_INLINE int tsdbKeyCmprFn(const void *p1, const void *p2) {
+  TSDBKEY *pKey1 = (TSDBKEY *)p1;
+  TSDBKEY *pKey2 = (TSDBKEY *)p2;
+
+  if (pKey1->ts < pKey2->ts) {
+    return -1;
+  } else if (pKey1->ts > pKey2->ts) {
+    return 1;
+  }
+
+  if (pKey1->version < pKey2->version) {
+    return -1;
+  } else if (pKey1->version > pKey2->version) {
+    return 1;
+  }
+
+  return 0;
+}
+
 #endif
 
 #ifdef __cplusplus

source/dnode/vnode/src/inc/vnd.h

@@ -81,9 +81,10 @@ int32_t vnodeSyncCommit(SVnode* pVnode);
 int32_t vnodeAsyncCommit(SVnode* pVnode);
 
 // vnodeSync.c
-int32_t   vnodeSyncOpen(SVnode* pVnode, char* path);
+int32_t vnodeSyncOpen(SVnode* pVnode, char* path);
-void      vnodeSyncStart(SVnode* pVnode);
+void    vnodeSyncStart(SVnode* pVnode);
-void      vnodeSyncClose(SVnode* pVnode);
+void    vnodeSyncClose(SVnode* pVnode);
+void    vnodeSyncAlter(SVnode* pVnode, SRpcMsg* pMsg);
 
 #ifdef __cplusplus
 }

source/dnode/vnode/src/inc/vnodeInt.h

@@ -125,6 +125,8 @@ int32_t      tsdbSnapshotReaderClose(STsdbSnapshotReader* pReader);
 int32_t      tsdbSnapshotRead(STsdbSnapshotReader* pReader, void** ppData, uint32_t* nData);
 
 // tq
+int     tqInit();
+void    tqCleanUp();
 STQ*    tqOpen(const char* path, SVnode* pVnode, SWal* pWal);
 void    tqClose(STQ*);
 int     tqPushMsg(STQ*, void* msg, int32_t msgLen, tmsg_t msgType, int64_t ver);
@@ -145,11 +147,11 @@ int32_t tqProcessTaskRecoverRsp(STQ* pTq, SRpcMsg* pMsg);
 int32_t smaOpen(SVnode* pVnode);
 int32_t smaClose(SSma* pSma);
 
-int32_t tdUpdateExpireWindow(SSma* pSma, SSubmitReq* pMsg, int64_t version);
+int32_t tdUpdateExpireWindow(SSma* pSma, const SSubmitReq* pMsg, int64_t version);
 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(SVnode *pVnode, SVCreateStbReq* pReq);
+int32_t tdProcessRSmaCreate(SVnode* pVnode, SVCreateStbReq* pReq);
 int32_t tdProcessRSmaSubmit(SSma* pSma, void* pMsg, int32_t inputType);
 int32_t tdFetchTbUidList(SSma* pSma, STbUidStore** ppStore, tb_uid_t suid, tb_uid_t uid);
 int32_t tdUpdateTbUidList(SSma* pSma, STbUidStore* pUidStore);
@@ -239,6 +241,8 @@ struct SVnode {
 #define VND_RSMA1(vnd)      ((vnd)->pSma->pRSmaTsdb1)
 #define VND_RSMA2(vnd)      ((vnd)->pSma->pRSmaTsdb2)
 #define VND_RETENTIONS(vnd) (&(vnd)->config.tsdbCfg.retentions)
+#define VND_IS_RSMA(v)      ((v)->config.isRsma == 1)
+#define VND_IS_TSMA(v)      ((v)->config.isTsma == 1)
 
 struct STbUidStore {
   tb_uid_t  suid;
@@ -271,11 +275,6 @@ struct SSma {
 #define SMA_RSMA_TSDB1(s) ((s)->pRSmaTsdb1)
 #define SMA_RSMA_TSDB2(s) ((s)->pRSmaTsdb2)
 
-static FORCE_INLINE bool vnodeIsRollup(SVnode* pVnode) {
-  SRetention* pRetention = &(pVnode->config.tsdbCfg.retentions[0]);
-  return (pRetention->freq > 0 && pRetention->keep > 0);
-}
-
 // sma
 void smaHandleRes(void* pVnode, int64_t smaId, const SArray* data);
 

source/dnode/vnode/src/meta/metaEntry.c

@@ -30,7 +30,8 @@ int metaEncodeEntry(SEncoder *pCoder, const SMetaEntry *pME) {
     if (tEncodeI64(pCoder, pME->ctbEntry.ctime) < 0) return -1;
     if (tEncodeI32(pCoder, pME->ctbEntry.ttlDays) < 0) return -1;
     if (tEncodeI64(pCoder, pME->ctbEntry.suid) < 0) return -1;
-    if (tEncodeBinary(pCoder, pME->ctbEntry.pTags, kvRowLen(pME->ctbEntry.pTags)) < 0) return -1;
+    debugCheckTags((STag*)pME->ctbEntry.pTags); // TODO: remove after debug
+    if (tEncodeTag(pCoder, (const STag *)pME->ctbEntry.pTags) < 0) return -1;
   } else if (pME->type == TSDB_NORMAL_TABLE) {
     if (tEncodeI64(pCoder, pME->ntbEntry.ctime) < 0) return -1;
     if (tEncodeI32(pCoder, pME->ntbEntry.ttlDays) < 0) return -1;
@@ -47,7 +48,6 @@ int metaEncodeEntry(SEncoder *pCoder, const SMetaEntry *pME) {
 }
 
 int metaDecodeEntry(SDecoder *pCoder, SMetaEntry *pME) {
-  uint32_t len;
   if (tStartDecode(pCoder) < 0) return -1;
 
   if (tDecodeI64(pCoder, &pME->version) < 0) return -1;
@@ -62,7 +62,8 @@ int metaDecodeEntry(SDecoder *pCoder, SMetaEntry *pME) {
     if (tDecodeI64(pCoder, &pME->ctbEntry.ctime) < 0) return -1;
     if (tDecodeI32(pCoder, &pME->ctbEntry.ttlDays) < 0) return -1;
     if (tDecodeI64(pCoder, &pME->ctbEntry.suid) < 0) return -1;
-    if (tDecodeBinary(pCoder, &pME->ctbEntry.pTags, &len) < 0) return -1;  // (TODO)
+    if (tDecodeTag(pCoder, (STag **)&pME->ctbEntry.pTags) < 0) return -1;  // (TODO)
+    debugCheckTags((STag*)pME->ctbEntry.pTags); // TODO: remove after debug
   } else if (pME->type == TSDB_NORMAL_TABLE) {
     if (tDecodeI64(pCoder, &pME->ntbEntry.ctime) < 0) return -1;
     if (tDecodeI32(pCoder, &pME->ntbEntry.ttlDays) < 0) return -1;

source/dnode/vnode/src/meta/metaQuery.c

@@ -573,10 +573,23 @@ SArray *metaGetSmaTbUids(SMeta *pMeta) {
 
 #endif
 
-const void *metaGetTableTagVal(SMetaEntry *pEntry, int16_t cid) {
+const void *metaGetTableTagVal(SMetaEntry *pEntry, int16_t type, STagVal *val) {
   ASSERT(pEntry->type == TSDB_CHILD_TABLE);
-  return tdGetKVRowValOfCol((const SKVRow)pEntry->ctbEntry.pTags, cid);
+  STag *tag = (STag *)pEntry->ctbEntry.pTags;
+  if (type == TSDB_DATA_TYPE_JSON){
+    if(tag->nTag == 0){
+      return NULL;
+    }
+    return tag;
+  }
+  bool find = tTagGet(tag, val);
+
+  if(!find){
+    return NULL;
+  }
+  return val;
 }
+
 typedef struct {
   SMeta *  pMeta;
   TBC *    pCur;
@@ -609,7 +622,13 @@ int32_t metaFilteTableIds(SMeta *pMeta, SMetaFltParam *param, SArray *pUids) {
   STagIdxKey *pKey = NULL;
   int32_t     nKey = 0;
 
-  ret = metaCreateTagIdxKey(pCursor->suid, pCursor->cid, param->val, pCursor->type,
+  int32_t nTagData = 0;
+  if(IS_VAR_DATA_TYPE(param->type)){
+    nTagData = strlen(param->val);
+  }else{
+    nTagData = tDataTypes[param->type].bytes;
+  }
+  ret = metaCreateTagIdxKey(pCursor->suid, pCursor->cid, param->val, nTagData, pCursor->type,
                             param->reverse ? INT64_MAX : INT64_MIN, &pKey, &nKey);
   if (ret != 0) {
     goto END;
@@ -651,4 +670,4 @@ END:
   taosMemoryFree(pCursor);
 
   return ret;
-}
+}

source/dnode/vnode/src/meta/metaTable.c

@@ -49,9 +49,9 @@ int metaCreateSTable(SMeta *pMeta, int64_t version, SVCreateStbReq *pReq) {
   int         vLen = 0;
   const void *pKey = NULL;
   const void *pVal = NULL;
-  void *      pBuf = NULL;
+  void       *pBuf = NULL;
   int32_t     szBuf = 0;
-  void *      p = NULL;
+  void       *p = NULL;
   SMetaReader mr = {0};
 
   // validate req
@@ -105,7 +105,7 @@ int metaDropSTable(SMeta *pMeta, int64_t verison, SVDropStbReq *pReq) {
   }
 
   // drop all child tables
-  TBC *   pCtbIdxc = NULL;
+  TBC    *pCtbIdxc = NULL;
   SArray *pArray = taosArrayInit(8, sizeof(tb_uid_t));
 
   tdbTbcOpen(pMeta->pCtbIdx, &pCtbIdxc, &pMeta->txn);
@@ -160,8 +160,8 @@ _exit:
 int metaAlterSTable(SMeta *pMeta, int64_t version, SVCreateStbReq *pReq) {
   SMetaEntry  oStbEntry = {0};
   SMetaEntry  nStbEntry = {0};
-  TBC *       pUidIdxc = NULL;
+  TBC        *pUidIdxc = NULL;
-  TBC *       pTbDbc = NULL;
+  TBC        *pTbDbc = NULL;
   const void *pData;
   int         nData;
   int64_t     oversion;
@@ -280,7 +280,7 @@ _err:
 }
 
 int metaDropTable(SMeta *pMeta, int64_t version, SVDropTbReq *pReq, SArray *tbUids) {
-  void *   pData = NULL;
+  void    *pData = NULL;
   int      nData = 0;
   int      rc = 0;
   tb_uid_t uid;
@@ -306,7 +306,7 @@ int metaDropTable(SMeta *pMeta, int64_t version, SVDropTbReq *pReq, SArray *tbUi
 }
 
 static int metaDropTableByUid(SMeta *pMeta, tb_uid_t uid, int *type) {
-  void *     pData = NULL;
+  void      *pData = NULL;
   int        nData = 0;
   int        rc = 0;
   int64_t    version;
@@ -344,12 +344,12 @@ static int metaDropTableByUid(SMeta *pMeta, tb_uid_t uid, int *type) {
 static int metaAlterTableColumn(SMeta *pMeta, int64_t version, SVAlterTbReq *pAlterTbReq, STableMetaRsp *pMetaRsp) {
   void *          pVal = NULL;
   int             nVal = 0;
-  const void *    pData = NULL;
+  const void     *pData = NULL;
   int             nData = 0;
   int             ret = 0;
   tb_uid_t        uid;
   int64_t         oversion;
-  SSchema *       pColumn = NULL;
+  SSchema        *pColumn = NULL;
   SMetaEntry      entry = {0};
   SSchemaWrapper *pSchema;
   int             c;
@@ -499,7 +499,7 @@ _err:
 static int metaUpdateTableTagVal(SMeta *pMeta, int64_t version, SVAlterTbReq *pAlterTbReq) {
   SMetaEntry  ctbEntry = {0};
   SMetaEntry  stbEntry = {0};
-  void *      pVal = NULL;
+  void       *pVal = NULL;
   int         nVal = 0;
   int         ret;
   int         c;
@@ -530,7 +530,7 @@ static int metaUpdateTableTagVal(SMeta *pMeta, int64_t version, SVAlterTbReq *pA
   oversion = *(int64_t *)pData;
 
   // search table.db
-  TBC *    pTbDbc = NULL;
+  TBC     *pTbDbc = NULL;
   SDecoder dc1 = {0};
   SDecoder dc2 = {0};
 
@@ -554,7 +554,7 @@ static int metaUpdateTableTagVal(SMeta *pMeta, int64_t version, SVAlterTbReq *pA
   metaDecodeEntry(&dc2, &stbEntry);
 
   SSchemaWrapper *pTagSchema = &stbEntry.stbEntry.schemaTag;
-  SSchema *       pColumn = NULL;
+  SSchema        *pColumn = NULL;
   int32_t         iCol = 0;
   for (;;) {
     pColumn = NULL;
@@ -583,29 +583,39 @@ static int metaUpdateTableTagVal(SMeta *pMeta, int64_t version, SVAlterTbReq *pA
     }
     memcpy((void *)ctbEntry.ctbEntry.pTags, pAlterTbReq->pTagVal, pAlterTbReq->nTagVal);
   } else {
-    SKVRowBuilder kvrb = {0};
+    const STag *pOldTag = (const STag *)ctbEntry.ctbEntry.pTags;
-    const SKVRow  pOldTag = (const SKVRow)ctbEntry.ctbEntry.pTags;
+    STag       *pNewTag = NULL;
-    SKVRow        pNewTag = NULL;
+    SArray     *pTagArray = taosArrayInit(pTagSchema->nCols, sizeof(STagVal));
-
+    if (!pTagArray) {
-    tdInitKVRowBuilder(&kvrb);
+      terrno = TSDB_CODE_OUT_OF_MEMORY;
+      goto _err;
+    }
     for (int32_t i = 0; i < pTagSchema->nCols; i++) {
       SSchema *pCol = &pTagSchema->pSchema[i];
       if (iCol == i) {
-        tdAddColToKVRow(&kvrb, pCol->colId, pAlterTbReq->pTagVal, pAlterTbReq->nTagVal);
+        STagVal val = {0};
+        val.type = pCol->type;
+        val.cid = pCol->colId;
+        if (IS_VAR_DATA_TYPE(pCol->type)) {
+          val.pData = pAlterTbReq->pTagVal;
+          val.nData = pAlterTbReq->nTagVal;
+        } else {
+          memcpy(&val.i64, pAlterTbReq->pTagVal, pAlterTbReq->nTagVal);
+        }
+        taosArrayPush(pTagArray, &val);
       } else {
-        void *p = tdGetKVRowValOfCol(pOldTag, pCol->colId);
+        STagVal val = {0};
-        if (p) {
+        if (tTagGet(pOldTag, &val)) {
-          if (IS_VAR_DATA_TYPE(pCol->type)) {
+          taosArrayPush(pTagArray, &val);
-            tdAddColToKVRow(&kvrb, pCol->colId, p, varDataTLen(p));
-          } else {
-            tdAddColToKVRow(&kvrb, pCol->colId, p, pCol->bytes);
-          }
         }
       }
     }
-
+    if ((terrno = tTagNew(pTagArray, pTagSchema->version, false, &pNewTag)) < 0) {
-    ctbEntry.ctbEntry.pTags = tdGetKVRowFromBuilder(&kvrb);
+      taosArrayDestroy(pTagArray);
-    tdDestroyKVRowBuilder(&kvrb);
+      goto _err;
+    }
+    ctbEntry.ctbEntry.pTags = (uint8_t *)pNewTag;
+    taosArrayDestroy(pTagArray);
   }
 
   // save to table.db
@@ -659,8 +669,8 @@ int metaAlterTable(SMeta *pMeta, int64_t version, SVAlterTbReq *pReq, STableMeta
 
 static int metaSaveToTbDb(SMeta *pMeta, const SMetaEntry *pME) {
   STbDbKey tbDbKey;
-  void *   pKey = NULL;
+  void    *pKey = NULL;
-  void *   pVal = NULL;
+  void    *pVal = NULL;
   int      kLen = 0;
   int      vLen = 0;
   SEncoder coder = {0};
@@ -741,17 +751,17 @@ static int metaUpdateCtbIdx(SMeta *pMeta, const SMetaEntry *pME) {
   return tdbTbInsert(pMeta->pCtbIdx, &ctbIdxKey, sizeof(ctbIdxKey), NULL, 0, &pMeta->txn);
 }
 
-int metaCreateTagIdxKey(tb_uid_t suid, int32_t cid, const void *pTagData, int8_t type, tb_uid_t uid,
+int metaCreateTagIdxKey(tb_uid_t suid, int32_t cid, const void *pTagData, int32_t nTagData, int8_t type, tb_uid_t uid,
                         STagIdxKey **ppTagIdxKey, int32_t *nTagIdxKey) {
-  int32_t nTagData = 0;
+  // int32_t nTagData = 0;
 
-  if (pTagData) {
+  // if (pTagData) {
-    if (IS_VAR_DATA_TYPE(type)) {
+  //   if (IS_VAR_DATA_TYPE(type)) {
-      nTagData = varDataTLen(pTagData);
+  //     nTagData = varDataTLen(pTagData);
-    } else {
+  //   } else {
-      nTagData = tDataTypes[type].bytes;
+  //     nTagData = tDataTypes[type].bytes;
-    }
+  //   }
-  }
+  // }
   *nTagIdxKey = sizeof(STagIdxKey) + nTagData + sizeof(tb_uid_t);
 
   *ppTagIdxKey = (STagIdxKey *)taosMemoryMalloc(*nTagIdxKey);
@@ -775,14 +785,15 @@ static void metaDestroyTagIdxKey(STagIdxKey *pTagIdxKey) {
 }
 
 static int metaUpdateTagIdx(SMeta *pMeta, const SMetaEntry *pCtbEntry) {
-  void *         pData = NULL;
+  void          *pData = NULL;
   int            nData = 0;
   STbDbKey       tbDbKey = {0};
   SMetaEntry     stbEntry = {0};
-  STagIdxKey *   pTagIdxKey = NULL;
+  STagIdxKey    *pTagIdxKey = NULL;
   int32_t        nTagIdxKey;
   const SSchema *pTagColumn;       // = &stbEntry.stbEntry.schema.pSchema[0];
-  const void *   pTagData = NULL;  //
+  const void    *pTagData = NULL;  //
+  int32_t        nTagData = 0;
   SDecoder       dc = {0};
 
   // get super table
@@ -795,7 +806,21 @@ static int metaUpdateTagIdx(SMeta *pMeta, const SMetaEntry *pCtbEntry) {
   metaDecodeEntry(&dc, &stbEntry);
 
   pTagColumn = &stbEntry.stbEntry.schemaTag.pSchema[0];
-  pTagData = tdGetKVRowValOfCol((const SKVRow)pCtbEntry->ctbEntry.pTags, pTagColumn->colId);
+
+  STagVal tagVal = {.cid = pTagColumn->colId};
+  if (pTagColumn->type != TSDB_DATA_TYPE_JSON) {
+    tTagGet((const STag *)pCtbEntry->ctbEntry.pTags, &tagVal);
+    if (IS_VAR_DATA_TYPE(pTagColumn->type)) {
+      pTagData = tagVal.pData;
+      nTagData = (int32_t)tagVal.nData;
+    } else {
+      pTagData = &(tagVal.i64);
+      nTagData = tDataTypes[pTagColumn->type].bytes;
+    }
+  } else {
+    // pTagData = pCtbEntry->ctbEntry.pTags;
+    // nTagData = ((const STag *)pCtbEntry->ctbEntry.pTags)->len;
+  }
 
   // update tag index
 #ifdef USE_INVERTED_INDEX
@@ -810,8 +835,8 @@ static int metaUpdateTagIdx(SMeta *pMeta, const SMetaEntry *pCtbEntry) {
   int ret = indexPut((SIndex *)pMeta->pTagIvtIdx, tmGroup, tuid);
   indexMultiTermDestroy(tmGroup);
 #else
-  if (metaCreateTagIdxKey(pCtbEntry->ctbEntry.suid, pTagColumn->colId, pTagData, pTagColumn->type, pCtbEntry->uid,
+  if (metaCreateTagIdxKey(pCtbEntry->ctbEntry.suid, pTagColumn->colId, pTagData, nTagData, pTagColumn->type,
-                          &pTagIdxKey, &nTagIdxKey) < 0) {
+                          pCtbEntry->uid, &pTagIdxKey, &nTagIdxKey) < 0) {
     return -1;
   }
   tdbTbInsert(pMeta->pTagIdx, pTagIdxKey, nTagIdxKey, NULL, 0, &pMeta->txn);
@@ -824,7 +849,7 @@ static int metaUpdateTagIdx(SMeta *pMeta, const SMetaEntry *pCtbEntry) {
 
 static int metaSaveToSkmDb(SMeta *pMeta, const SMetaEntry *pME) {
   SEncoder              coder = {0};
-  void *                pVal = NULL;
+  void                 *pVal = NULL;
   int                   vLen = 0;
   int                   rcode = 0;
   SSkmDbKey             skmDbKey = {0};

source/dnode/vnode/src/sma/sma.c

@@ -15,7 +15,6 @@
 
 #include "sma.h"
 
-
 // TODO: Who is responsible for resource allocate and release?
 int32_t tdProcessTSmaInsert(SSma* pSma, int64_t indexUid, const char* msg) {
   int32_t code = TSDB_CODE_SUCCESS;
@@ -37,7 +36,7 @@ int32_t tdProcessTSmaCreate(SSma* pSma, int64_t version, const char* msg) {
   return code;
 }
 
-int32_t tdUpdateExpireWindow(SSma* pSma, SSubmitReq* pMsg, int64_t version) {
+int32_t tdUpdateExpireWindow(SSma* pSma, const SSubmitReq* pMsg, int64_t version) {
   int32_t code = TSDB_CODE_SUCCESS;
   if ((code = tdUpdateExpiredWindowImpl(pSma, pMsg, version)) < 0) {
     smaWarn("vgId:%d update expired sma window failed since %s", SMA_VID(pSma), tstrerror(terrno));

source/dnode/vnode/src/sma/smaOpen.c

@@ -104,7 +104,7 @@ int32_t smaOpen(SVnode *pVnode) {
   taosThreadMutexInit(&pSma->mutex, NULL);
   pSma->locked = false;
 
-  if (vnodeIsRollup(pVnode)) {
+  if (VND_IS_RSMA(pVnode)) {
     STsdbKeepCfg keepCfg = {0};
     for (int i = 0; i < TSDB_RETENTION_MAX; ++i) {
       if (i == TSDB_RETENTION_L0) {

source/dnode/vnode/src/sma/smaTimeRange.c

@@ -932,7 +932,7 @@ static int32_t tdSetExpiredWindow(SSma *pSma, SHashObj *pItemsHash, int64_t inde
  * @param msg SSubmitReq
  * @return int32_t
  */
-int32_t tdUpdateExpiredWindowImpl(SSma *pSma, SSubmitReq *pMsg, int64_t version) {
+int32_t tdUpdateExpiredWindowImpl(SSma *pSma, const SSubmitReq *pMsg, int64_t version) {
   // no time-range-sma, just return success
   if (atomic_load_16(&SMA_TSMA_NUM(pSma)) <= 0) {
     smaTrace("vgId:%d not update expire window since no tSma", SMA_VID(pSma));

source/dnode/vnode/src/tq/tq.c

@@ -14,7 +14,6 @@
  */
 
 #include "tq.h"
-#include "tdbInt.h"
 
 int32_t tqInit() {
   int8_t old;
@@ -47,51 +46,6 @@ void tqCleanUp() {
   }
 }
 
-int tqExecKeyCompare(const void* pKey1, int32_t kLen1, const void* pKey2, int32_t kLen2) {
-  return strcmp(pKey1, pKey2);
-}
-
-int32_t tqStoreHandle(STQ* pTq, const char* key, const STqHandle* pHandle) {
-  int32_t code;
-  int32_t vlen;
-  tEncodeSize(tEncodeSTqHandle, pHandle, vlen, code);
-  ASSERT(code == 0);
-
-  void* buf = taosMemoryCalloc(1, vlen);
-  if (buf == NULL) {
-    ASSERT(0);
-  }
-
-  SEncoder encoder;
-  tEncoderInit(&encoder, buf, vlen);
-
-  if (tEncodeSTqHandle(&encoder, pHandle) < 0) {
-    ASSERT(0);
-  }
-
-  TXN txn;
-
-  if (tdbTxnOpen(&txn, 0, tdbDefaultMalloc, tdbDefaultFree, NULL, TDB_TXN_WRITE | TDB_TXN_READ_UNCOMMITTED) < 0) {
-    ASSERT(0);
-  }
-
-  if (tdbBegin(pTq->pMetaStore, &txn) < 0) {
-    ASSERT(0);
-  }
-
-  if (tdbTbUpsert(pTq->pExecStore, key, (int)strlen(key), buf, vlen, &txn) < 0) {
-    ASSERT(0);
-  }
-
-  if (tdbCommit(pTq->pMetaStore, &txn) < 0) {
-    ASSERT(0);
-  }
-
-  tEncoderClear(&encoder);
-  taosMemoryFree(buf);
-  return 0;
-}
-
 STQ* tqOpen(const char* path, SVnode* pVnode, SWal* pWal) {
   STQ* pTq = taosMemoryMalloc(sizeof(STQ));
   if (pTq == NULL) {
@@ -108,60 +62,7 @@ STQ* tqOpen(const char* path, SVnode* pVnode, SWal* pWal) {
 
   pTq->pushMgr = taosHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), true, HASH_ENTRY_LOCK);
 
-  if (tdbOpen(path, 16 * 1024, 1, &pTq->pMetaStore) < 0) {
+  if (tqMetaOpen(pTq) < 0) {
-    ASSERT(0);
-  }
-
-  if (tdbTbOpen("handles", -1, -1, tqExecKeyCompare, pTq->pMetaStore, &pTq->pExecStore) < 0) {
-    ASSERT(0);
-  }
-
-  TXN txn;
-
-  if (tdbTxnOpen(&txn, 0, tdbDefaultMalloc, tdbDefaultFree, NULL, 0) < 0) {
-    ASSERT(0);
-  }
-
-  TBC* pCur;
-  if (tdbTbcOpen(pTq->pExecStore, &pCur, &txn) < 0) {
-    ASSERT(0);
-  }
-
-  void* pKey;
-  int   kLen;
-  void* pVal;
-  int   vLen;
-
-  tdbTbcMoveToFirst(pCur);
-  SDecoder decoder;
-
-  while (tdbTbcNext(pCur, &pKey, &kLen, &pVal, &vLen) == 0) {
-    STqHandle handle;
-    tDecoderInit(&decoder, (uint8_t*)pVal, vLen);
-    tDecodeSTqHandle(&decoder, &handle);
-    handle.pWalReader = walOpenReadHandle(pTq->pVnode->pWal);
-    for (int32_t i = 0; i < 5; i++) {
-      handle.execHandle.pExecReader[i] = tqInitSubmitMsgScanner(pTq->pVnode->pMeta);
-    }
-    if (handle.execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
-      for (int32_t i = 0; i < 5; i++) {
-        SReadHandle reader = {
-            .reader = handle.execHandle.pExecReader[i],
-            .meta = pTq->pVnode->pMeta,
-            .pMsgCb = &pTq->pVnode->msgCb,
-        };
-        handle.execHandle.exec.execCol.task[i] =
-            qCreateStreamExecTaskInfo(handle.execHandle.exec.execCol.qmsg, &reader);
-        ASSERT(handle.execHandle.exec.execCol.task[i]);
-      }
-    } else {
-      handle.execHandle.exec.execDb.pFilterOutTbUid =
-          taosHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), false, HASH_NO_LOCK);
-    }
-    taosHashPut(pTq->handles, pKey, kLen, &handle, sizeof(STqHandle));
-  }
-
-  if (tdbTxnClose(&txn) < 0) {
     ASSERT(0);
   }
 
@@ -174,181 +75,47 @@ void tqClose(STQ* pTq) {
     taosHashCleanup(pTq->handles);
     taosHashCleanup(pTq->pStreamTasks);
     taosHashCleanup(pTq->pushMgr);
-    tdbClose(pTq->pMetaStore);
+    tqMetaClose(pTq);
     taosMemoryFree(pTq);
   }
   // TODO
 }
 
-int32_t tEncodeSTqHandle(SEncoder* pEncoder, const STqHandle* pHandle) {
+int32_t tqSendPollRsp(STQ* pTq, const SRpcMsg* pMsg, const SMqPollReq* pReq, const SMqDataBlkRsp* pRsp) {
-  if (tStartEncode(pEncoder) < 0) return -1;
+  int32_t tlen = sizeof(SMqRspHead) + tEncodeSMqDataBlkRsp(NULL, pRsp);
-  if (tEncodeCStr(pEncoder, pHandle->subKey) < 0) return -1;
+  void*   buf = rpcMallocCont(tlen);
-  if (tEncodeI64(pEncoder, pHandle->consumerId) < 0) return -1;
+  if (buf == NULL) {
-  if (tEncodeI32(pEncoder, pHandle->epoch) < 0) return -1;
+    return -1;
-  if (tEncodeI8(pEncoder, pHandle->execHandle.subType) < 0) return -1;
-  if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
-    if (tEncodeCStr(pEncoder, pHandle->execHandle.exec.execCol.qmsg) < 0) return -1;
-  }
-  tEndEncode(pEncoder);
-  return pEncoder->pos;
-}
-
-int32_t tDecodeSTqHandle(SDecoder* pDecoder, STqHandle* pHandle) {
-  if (tStartDecode(pDecoder) < 0) return -1;
-  if (tDecodeCStrTo(pDecoder, pHandle->subKey) < 0) return -1;
-  if (tDecodeI64(pDecoder, &pHandle->consumerId) < 0) return -1;
-  if (tDecodeI32(pDecoder, &pHandle->epoch) < 0) return -1;
-  if (tDecodeI8(pDecoder, &pHandle->execHandle.subType) < 0) return -1;
-  if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
-    if (tDecodeCStrAlloc(pDecoder, &pHandle->execHandle.exec.execCol.qmsg) < 0) return -1;
-  }
-  tEndDecode(pDecoder);
-  return 0;
-}
-
-int32_t tqUpdateTbUidList(STQ* pTq, const SArray* tbUidList, bool isAdd) {
-  void* pIter = NULL;
-  while (1) {
-    pIter = taosHashIterate(pTq->handles, pIter);
-    if (pIter == NULL) break;
-    STqHandle* pExec = (STqHandle*)pIter;
-    if (pExec->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
-      for (int32_t i = 0; i < 5; i++) {
-        int32_t code = qUpdateQualifiedTableId(pExec->execHandle.exec.execCol.task[i], tbUidList, isAdd);
-        ASSERT(code == 0);
-      }
-    } else if (pExec->execHandle.subType == TOPIC_SUB_TYPE__DB) {
-      if (!isAdd) {
-        int32_t sz = taosArrayGetSize(tbUidList);
-        for (int32_t i = 0; i < sz; i++) {
-          int64_t tbUid = *(int64_t*)taosArrayGet(tbUidList, i);
-          taosHashPut(pExec->execHandle.exec.execDb.pFilterOutTbUid, &tbUid, sizeof(int64_t), NULL, 0);
-        }
-      }
-    } else {
-      // tq update id
-    }
-  }
-  while (1) {
-    pIter = taosHashIterate(pTq->pStreamTasks, pIter);
-    if (pIter == NULL) break;
-    SStreamTask* pTask = (SStreamTask*)pIter;
-    if (pTask->inputType == STREAM_INPUT__DATA_SUBMIT) {
-      int32_t code = qUpdateQualifiedTableId(pTask->exec.executor, tbUidList, isAdd);
-      ASSERT(code == 0);
-    }
-  }
-  return 0;
-}
-
-int32_t tqPushMsgNew(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_t ver, SRpcHandleInfo handleInfo) {
-  if (msgType != TDMT_VND_SUBMIT) return 0;
-  void*       pIter = NULL;
-  STqHandle*  pHandle = NULL;
-  SSubmitReq* pReq = (SSubmitReq*)msg;
-  int32_t     workerId = 4;
-  int64_t     fetchOffset = ver;
-
-  while (1) {
-    pIter = taosHashIterate(pTq->pushMgr, pIter);
-    if (pIter == NULL) break;
-    pHandle = *(STqHandle**)pIter;
-
-    taosWLockLatch(&pHandle->pushHandle.lock);
-
-    /*SRpcHandleInfo* pInfo = atomic_load_ptr(&pHandle->pushHandle.pInfo);*/
-    /*ASSERT(pInfo);*/
-
-    SMqDataBlkRsp rsp = {0};
-    rsp.reqOffset = pHandle->pushHandle.reqOffset;
-    rsp.blockData = taosArrayInit(0, sizeof(void*));
-    rsp.blockDataLen = taosArrayInit(0, sizeof(int32_t));
-
-    if (msgType == TDMT_VND_SUBMIT) {
-      tqDataExec(pTq, &pHandle->execHandle, pReq, &rsp, workerId);
-    } else {
-      // TODO
-      ASSERT(0);
-    }
-
-    if (rsp.blockNum == 0) {
-      taosWUnLockLatch(&pHandle->pushHandle.lock);
-      continue;
-    }
-
-    ASSERT(taosArrayGetSize(rsp.blockData) == rsp.blockNum);
-    ASSERT(taosArrayGetSize(rsp.blockDataLen) == rsp.blockNum);
-
-    rsp.rspOffset = fetchOffset;
-
-    int32_t tlen = sizeof(SMqRspHead) + tEncodeSMqDataBlkRsp(NULL, &rsp);
-    void*   buf = rpcMallocCont(tlen);
-    if (buf == NULL) {
-      // todo free
-      return -1;
-    }
-
-    ((SMqRspHead*)buf)->mqMsgType = TMQ_MSG_TYPE__POLL_RSP;
-    ((SMqRspHead*)buf)->epoch = pHandle->pushHandle.epoch;
-    ((SMqRspHead*)buf)->consumerId = pHandle->pushHandle.consumerId;
-
-    void* abuf = POINTER_SHIFT(buf, sizeof(SMqRspHead));
-    tEncodeSMqDataBlkRsp(&abuf, &rsp);
-
-    SRpcMsg resp = {
-        .info = pHandle->pushHandle.info,
-        .pCont = buf,
-        .contLen = tlen,
-        .code = 0,
-    };
-    tmsgSendRsp(&resp);
-
-    /*atomic_store_ptr(&pHandle->pushHandle.pInfo, NULL);*/
-    memset(&pHandle->pushHandle.info, 0, sizeof(SRpcHandleInfo));
-    taosWUnLockLatch(&pHandle->pushHandle.lock);
-
-    tqDebug("vg %d offset %ld from consumer %ld (epoch %d) send rsp, block num: %d, reqOffset: %ld, rspOffset: %ld",
-            TD_VID(pTq->pVnode), fetchOffset, pHandle->pushHandle.consumerId, pHandle->pushHandle.epoch, rsp.blockNum,
-            rsp.reqOffset, rsp.rspOffset);
-
-    // TODO destroy
-    taosArrayDestroy(rsp.blockData);
-    taosArrayDestroy(rsp.blockDataLen);
   }
 
-  return 0;
+  ((SMqRspHead*)buf)->mqMsgType = TMQ_MSG_TYPE__POLL_RSP;
-}
+  ((SMqRspHead*)buf)->epoch = pReq->epoch;
+  ((SMqRspHead*)buf)->consumerId = pReq->consumerId;
 
-int tqPushMsg(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_t ver) {
+  void* abuf = POINTER_SHIFT(buf, sizeof(SMqRspHead));
-  if (msgType == TDMT_VND_SUBMIT) {
+  tEncodeSMqDataBlkRsp(&abuf, pRsp);
-    if (taosHashGetSize(pTq->pStreamTasks) == 0) return 0;
 
-    if (tdUpdateExpireWindow(pTq->pVnode->pSma, msg, ver) != 0) {
+  SRpcMsg resp = {
-      // TODO handle sma error
+      .info = pMsg->info,
-    }
+      .pCont = buf,
-    void* data = taosMemoryMalloc(msgLen);
+      .contLen = tlen,
-    if (data == NULL) {
+      .code = 0,
-      return -1;
+  };
-    }
+  tmsgSendRsp(&resp);
-    memcpy(data, msg, msgLen);
 
-    tqProcessStreamTrigger(pTq, data);
+  tqDebug("vg %d from consumer %ld (epoch %d) send rsp, block num: %d, reqOffset: %ld, rspOffset: %ld",
-  }
+          TD_VID(pTq->pVnode), pReq->consumerId, pReq->epoch, pRsp->blockNum, pRsp->reqOffset, pRsp->rspOffset);
 
   return 0;
 }
 
-int tqCommit(STQ* pTq) {
-  // do nothing
-  return 0;
-}
-
 int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
   SMqPollReq* pReq = pMsg->pCont;
   int64_t     consumerId = pReq->consumerId;
-  int64_t     waitTime = pReq->timeout;
+  int64_t     timeout = pReq->timeout;
   int32_t     reqEpoch = pReq->epoch;
   int64_t     fetchOffset;
+  int32_t     code = 0;
 
   // get offset to fetch message
   if (pReq->currentOffset == TMQ_CONF__RESET_OFFSET__EARLIEAST) {
@@ -389,20 +156,18 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
   }
   if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
     rsp.withSchema = false;
-
     rsp.withTag = false;
   } else {
     rsp.withSchema = true;
-    rsp.blockSchema = taosArrayInit(0, sizeof(void*));
-
     rsp.withTag = false;
+    rsp.blockSchema = taosArrayInit(0, sizeof(void*));
   }
 
   while (1) {
     consumerEpoch = atomic_load_32(&pHandle->epoch);
     if (consumerEpoch > reqEpoch) {
-      tqDebug("tmq poll: consumer %ld (epoch %d) vg %d offset %ld, found new consumer epoch %d discard req epoch %d",
+      tqWarn("tmq poll: consumer %ld (epoch %d) vg %d offset %ld, found new consumer epoch %d, discard req epoch %d",
-              consumerId, pReq->epoch, TD_VID(pTq->pVnode), fetchOffset, consumerEpoch, reqEpoch);
+             consumerId, pReq->epoch, TD_VID(pTq->pVnode), fetchOffset, consumerEpoch, reqEpoch);
       break;
     }
 
@@ -413,34 +178,15 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
 
     SWalReadHead* pHead = &pHeadWithCkSum->head;
 
-#if 0
-      // add to pushMgr
-      taosWLockLatch(&pExec->pushHandle.lock);
-
-      pExec->pushHandle.consumerId = consumerId;
-      pExec->pushHandle.epoch = reqEpoch;
-      pExec->pushHandle.reqOffset = rsp.reqOffset;
-      pExec->pushHandle.skipLogNum = rsp.skipLogNum;
-      pExec->pushHandle.handle = pMsg;
-
-      taosWUnLockLatch(&pExec->pushHandle.lock);
-
-      // TODO add timer
-
-      // TODO: the pointer will always be valid?
-      taosHashPut(pTq->pushMgr, &consumerId, sizeof(int64_t), &pExec, sizeof(void*));
-      taosArrayDestroy(rsp.blockData);
-      taosArrayDestroy(rsp.blockDataLen);
-      return 0;
-#endif
-
     tqDebug("tmq poll: consumer %ld (epoch %d) iter log, vg %d offset %ld msgType %d", consumerId, pReq->epoch,
             TD_VID(pTq->pVnode), fetchOffset, pHead->msgType);
 
     if (pHead->msgType == TDMT_VND_SUBMIT) {
       SSubmitReq* pCont = (SSubmitReq*)&pHead->body;
 
-      tqDataExec(pTq, &pHandle->execHandle, pCont, &rsp, workerId);
+      if (tqDataExec(pTq, &pHandle->execHandle, pCont, &rsp, workerId) < 0) {
+        /*ASSERT(0);*/
+      }
     } else {
       // TODO
       ASSERT(0);
@@ -459,34 +205,16 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
 
   ASSERT(taosArrayGetSize(rsp.blockData) == rsp.blockNum);
   ASSERT(taosArrayGetSize(rsp.blockDataLen) == rsp.blockNum);
+  if (rsp.withSchema) {
+    ASSERT(taosArrayGetSize(rsp.blockSchema) == rsp.blockNum);
+  }
 
   rsp.rspOffset = fetchOffset;
 
-  int32_t tlen = sizeof(SMqRspHead) + tEncodeSMqDataBlkRsp(NULL, &rsp);
+  if (tqSendPollRsp(pTq, pMsg, pReq, &rsp) < 0) {
-  void*   buf = rpcMallocCont(tlen);
+    code = -1;
-  if (buf == NULL) {
-    pMsg->code = -1;
-    return -1;
   }
 
-  ((SMqRspHead*)buf)->mqMsgType = TMQ_MSG_TYPE__POLL_RSP;
-  ((SMqRspHead*)buf)->epoch = pReq->epoch;
-  ((SMqRspHead*)buf)->consumerId = consumerId;
-
-  void* abuf = POINTER_SHIFT(buf, sizeof(SMqRspHead));
-  tEncodeSMqDataBlkRsp(&abuf, &rsp);
-
-  SRpcMsg resp = {
-      .info = pMsg->info,
-      .pCont = buf,
-      .contLen = tlen,
-      .code = 0,
-  };
-  tmsgSendRsp(&resp);
-
-  tqDebug("vg %d offset %ld from consumer %ld (epoch %d) send rsp, block num: %d, reqOffset: %ld, rspOffset: %ld",
-          TD_VID(pTq->pVnode), fetchOffset, consumerId, pReq->epoch, rsp.blockNum, rsp.reqOffset, rsp.rspOffset);
-
   // TODO wrap in destroy func
   taosArrayDestroy(rsp.blockData);
   taosArrayDestroy(rsp.blockDataLen);
@@ -499,7 +227,7 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
     taosArrayDestroyP(rsp.blockTbName, (FDelete)taosMemoryFree);
   }
 
-  return 0;
+  return code;
 }
 
 int32_t tqProcessVgDeleteReq(STQ* pTq, char* msg, int32_t msgLen) {
@@ -508,24 +236,9 @@ int32_t tqProcessVgDeleteReq(STQ* pTq, char* msg, int32_t msgLen) {
   int32_t code = taosHashRemove(pTq->handles, pReq->subKey, strlen(pReq->subKey));
   ASSERT(code == 0);
 
-  TXN txn;
+  if (tqMetaDeleteHandle(pTq, pReq->subKey) < 0) {
-
-  if (tdbTxnOpen(&txn, 0, tdbDefaultMalloc, tdbDefaultFree, NULL, TDB_TXN_WRITE | TDB_TXN_READ_UNCOMMITTED) < 0) {
     ASSERT(0);
   }
-
-  if (tdbBegin(pTq->pMetaStore, &txn) < 0) {
-    ASSERT(0);
-  }
-
-  if (tdbTbDelete(pTq->pExecStore, pReq->subKey, (int)strlen(pReq->subKey), &txn) < 0) {
-    /*ASSERT(0);*/
-  }
-
-  if (tdbCommit(pTq->pMetaStore, &txn) < 0) {
-    ASSERT(0);
-  }
-
   return 0;
 }
 
@@ -547,18 +260,14 @@ int32_t tqProcessVgChangeReq(STQ* pTq, char* msg, int32_t msgLen) {
     pHandle->epoch = -1;
 
     pHandle->execHandle.subType = req.subType;
-    /*pExec->withTbName = req.withTbName;*/
-    /*pExec->withSchema = req.withSchema;*/
-    /*pExec->withTag = req.withTag;*/
-
-    pHandle->execHandle.exec.execCol.qmsg = req.qmsg;
-    req.qmsg = NULL;
 
     pHandle->pWalReader = walOpenReadHandle(pTq->pVnode->pWal);
     for (int32_t i = 0; i < 5; i++) {
       pHandle->execHandle.pExecReader[i] = tqInitSubmitMsgScanner(pTq->pVnode->pMeta);
     }
     if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
+      pHandle->execHandle.exec.execCol.qmsg = req.qmsg;
+      req.qmsg = NULL;
       for (int32_t i = 0; i < 5; i++) {
         SReadHandle handle = {
             .reader = pHandle->execHandle.pExecReader[i],
@@ -573,6 +282,18 @@ int32_t tqProcessVgChangeReq(STQ* pTq, char* msg, int32_t msgLen) {
       pHandle->execHandle.exec.execDb.pFilterOutTbUid =
           taosHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), false, HASH_NO_LOCK);
     } else if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__TABLE) {
+      pHandle->execHandle.exec.execTb.suid = req.suid;
+      SArray* tbUidList = taosArrayInit(0, sizeof(int64_t));
+      tsdbGetCtbIdList(pTq->pVnode->pMeta, req.suid, tbUidList);
+      tqDebug("vg %d, tq try get suid: %ld", pTq->pVnode->config.vgId, req.suid);
+      for (int32_t i = 0; i < taosArrayGetSize(tbUidList); i++) {
+        int64_t tbUid = *(int64_t*)taosArrayGet(tbUidList, i);
+        tqDebug("vg %d, idx %d, uid: %ld", pTq->pVnode->config.vgId, i, tbUid);
+      }
+      for (int32_t i = 0; i < 5; i++) {
+        tqReadHandleSetTbUidList(pHandle->execHandle.pExecReader[i], tbUidList);
+      }
+      taosArrayDestroy(tbUidList);
     }
     taosHashPut(pTq->handles, req.subKey, strlen(req.subKey), pHandle, sizeof(STqHandle));
   } else {
@@ -583,30 +304,12 @@ int32_t tqProcessVgChangeReq(STQ* pTq, char* msg, int32_t msgLen) {
     atomic_add_fetch_32(&pHandle->epoch, 1);
   }
 
-  if (tqStoreHandle(pTq, req.subKey, pHandle) < 0) {
+  if (tqMetaSaveHandle(pTq, req.subKey, pHandle) < 0) {
     // TODO
   }
   return 0;
 }
 
-void tqTableSink(SStreamTask* pTask, void* vnode, int64_t ver, void* data) {
-  const SArray* pRes = (const SArray*)data;
-  SVnode*       pVnode = (SVnode*)vnode;
-
-  ASSERT(pTask->tbSink.pTSchema);
-  SSubmitReq* pReq = tdBlockToSubmit(pRes, pTask->tbSink.pTSchema, true, pTask->tbSink.stbUid,
-                                     pTask->tbSink.stbFullName, pVnode->config.vgId);
-  /*tPrintFixedSchemaSubmitReq(pReq, pTask->tbSink.pTSchema);*/
-  // build write msg
-  SRpcMsg msg = {
-      .msgType = TDMT_VND_SUBMIT,
-      .pCont = pReq,
-      .contLen = ntohl(pReq->length),
-  };
-
-  ASSERT(tmsgPutToQueue(&pVnode->msgCb, WRITE_QUEUE, &msg) == 0);
-}
-
 int32_t tqProcessTaskDeploy(STQ* pTq, char* msg, int32_t msgLen) {
   SStreamTask* pTask = taosMemoryCalloc(1, sizeof(SStreamTask));
   if (pTask == NULL) {
@@ -697,9 +400,11 @@ int32_t tqProcessStreamTrigger(STQ* pTq, SSubmitReq* pReq) {
         continue;
       }
 
-      streamDataSubmitRefInc(pSubmit);
+      SStreamDataSubmit* pSubmitClone = streamSubmitRefClone(pSubmit);
-      SStreamDataSubmit* pSubmitClone = taosAllocateQitem(sizeof(SStreamDataSubmit), DEF_QITEM);
+      if (pSubmitClone == NULL) {
-      memcpy(pSubmitClone, pSubmit, sizeof(SStreamDataSubmit));
+        atomic_store_8(&pTask->inputStatus, TASK_INPUT_STATUS__FAILED);
+        continue;
+      }
       taosWriteQitem(pTask->inputQ, pSubmitClone);
 
       int8_t execStatus = atomic_load_8(&pTask->status);

source/dnode/vnode/src/tq/tqCommit.c

@@ -14,3 +14,8 @@
  */
 
 #include "tq.h"
+
+int tqCommit(STQ* pTq) {
+  // do nothing
+  return 0;
+}

source/dnode/vnode/src/tq/tqMeta.c

@@ -12,3 +12,163 @@
  * 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 "tdbInt.h"
+#include "tq.h"
+
+static int32_t tEncodeSTqHandle(SEncoder* pEncoder, const STqHandle* pHandle) {
+  if (tStartEncode(pEncoder) < 0) return -1;
+  if (tEncodeCStr(pEncoder, pHandle->subKey) < 0) return -1;
+  if (tEncodeI64(pEncoder, pHandle->consumerId) < 0) return -1;
+  if (tEncodeI32(pEncoder, pHandle->epoch) < 0) return -1;
+  if (tEncodeI8(pEncoder, pHandle->execHandle.subType) < 0) return -1;
+  if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
+    if (tEncodeCStr(pEncoder, pHandle->execHandle.exec.execCol.qmsg) < 0) return -1;
+  }
+  tEndEncode(pEncoder);
+  return pEncoder->pos;
+}
+
+static int32_t tDecodeSTqHandle(SDecoder* pDecoder, STqHandle* pHandle) {
+  if (tStartDecode(pDecoder) < 0) return -1;
+  if (tDecodeCStrTo(pDecoder, pHandle->subKey) < 0) return -1;
+  if (tDecodeI64(pDecoder, &pHandle->consumerId) < 0) return -1;
+  if (tDecodeI32(pDecoder, &pHandle->epoch) < 0) return -1;
+  if (tDecodeI8(pDecoder, &pHandle->execHandle.subType) < 0) return -1;
+  if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
+    if (tDecodeCStrAlloc(pDecoder, &pHandle->execHandle.exec.execCol.qmsg) < 0) return -1;
+  }
+  tEndDecode(pDecoder);
+  return 0;
+}
+
+int tqExecKeyCompare(const void* pKey1, int32_t kLen1, const void* pKey2, int32_t kLen2) {
+  return strcmp(pKey1, pKey2);
+}
+
+int32_t tqMetaOpen(STQ* pTq) {
+  if (tdbOpen(pTq->path, 16 * 1024, 1, &pTq->pMetaStore) < 0) {
+    ASSERT(0);
+  }
+
+  if (tdbTbOpen("handles", -1, -1, tqExecKeyCompare, pTq->pMetaStore, &pTq->pExecStore) < 0) {
+    ASSERT(0);
+  }
+
+  TXN txn;
+
+  if (tdbTxnOpen(&txn, 0, tdbDefaultMalloc, tdbDefaultFree, NULL, 0) < 0) {
+    ASSERT(0);
+  }
+
+  TBC* pCur;
+  if (tdbTbcOpen(pTq->pExecStore, &pCur, &txn) < 0) {
+    ASSERT(0);
+  }
+
+  void* pKey;
+  int   kLen;
+  void* pVal;
+  int   vLen;
+
+  tdbTbcMoveToFirst(pCur);
+  SDecoder decoder;
+
+  while (tdbTbcNext(pCur, &pKey, &kLen, &pVal, &vLen) == 0) {
+    STqHandle handle;
+    tDecoderInit(&decoder, (uint8_t*)pVal, vLen);
+    tDecodeSTqHandle(&decoder, &handle);
+    handle.pWalReader = walOpenReadHandle(pTq->pVnode->pWal);
+    for (int32_t i = 0; i < 5; i++) {
+      handle.execHandle.pExecReader[i] = tqInitSubmitMsgScanner(pTq->pVnode->pMeta);
+    }
+    if (handle.execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
+      for (int32_t i = 0; i < 5; i++) {
+        SReadHandle reader = {
+            .reader = handle.execHandle.pExecReader[i],
+            .meta = pTq->pVnode->pMeta,
+            .pMsgCb = &pTq->pVnode->msgCb,
+        };
+        handle.execHandle.exec.execCol.task[i] =
+            qCreateStreamExecTaskInfo(handle.execHandle.exec.execCol.qmsg, &reader);
+        ASSERT(handle.execHandle.exec.execCol.task[i]);
+      }
+    } else {
+      handle.execHandle.exec.execDb.pFilterOutTbUid =
+          taosHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), false, HASH_NO_LOCK);
+    }
+    taosHashPut(pTq->handles, pKey, kLen, &handle, sizeof(STqHandle));
+  }
+
+  if (tdbTxnClose(&txn) < 0) {
+    ASSERT(0);
+  }
+  return 0;
+}
+
+int32_t tqMetaClose(STQ* pTq) {
+  tdbClose(pTq->pMetaStore);
+  return 0;
+}
+
+int32_t tqMetaSaveHandle(STQ* pTq, const char* key, const STqHandle* pHandle) {
+  int32_t code;
+  int32_t vlen;
+  tEncodeSize(tEncodeSTqHandle, pHandle, vlen, code);
+  ASSERT(code == 0);
+
+  void* buf = taosMemoryCalloc(1, vlen);
+  if (buf == NULL) {
+    ASSERT(0);
+  }
+
+  SEncoder encoder;
+  tEncoderInit(&encoder, buf, vlen);
+
+  if (tEncodeSTqHandle(&encoder, pHandle) < 0) {
+    ASSERT(0);
+  }
+
+  TXN txn;
+
+  if (tdbTxnOpen(&txn, 0, tdbDefaultMalloc, tdbDefaultFree, NULL, TDB_TXN_WRITE | TDB_TXN_READ_UNCOMMITTED) < 0) {
+    ASSERT(0);
+  }
+
+  if (tdbBegin(pTq->pMetaStore, &txn) < 0) {
+    ASSERT(0);
+  }
+
+  if (tdbTbUpsert(pTq->pExecStore, key, (int)strlen(key), buf, vlen, &txn) < 0) {
+    ASSERT(0);
+  }
+
+  if (tdbCommit(pTq->pMetaStore, &txn) < 0) {
+    ASSERT(0);
+  }
+
+  tEncoderClear(&encoder);
+  taosMemoryFree(buf);
+  return 0;
+}
+
+int32_t tqMetaDeleteHandle(STQ* pTq, const char* key) {
+  TXN txn;
+
+  if (tdbTxnOpen(&txn, 0, tdbDefaultMalloc, tdbDefaultFree, NULL, TDB_TXN_WRITE | TDB_TXN_READ_UNCOMMITTED) < 0) {
+    ASSERT(0);
+  }
+
+  if (tdbBegin(pTq->pMetaStore, &txn) < 0) {
+    ASSERT(0);
+  }
+
+  if (tdbTbDelete(pTq->pExecStore, key, (int)strlen(key), &txn) < 0) {
+    /*ASSERT(0);*/
+  }
+
+  if (tdbCommit(pTq->pMetaStore, &txn) < 0) {
+    ASSERT(0);
+  }
+
+  return 0;
+}

source/dnode/vnode/src/tq/tqOffset.c

@@ -30,7 +30,7 @@ struct STqOffsetStore {
   SHashObj*    pHash;  // SHashObj<subscribeKey, offset>
 };
 
-STqOffsetStore* STqOffsetOpen(STqOffsetCfg* pCfg) {
+STqOffsetStore* tqOffsetOpen(STqOffsetCfg* pCfg) {
   STqOffsetStore* pStore = taosMemoryMalloc(sizeof(STqOffsetStore));
   if (pStore == NULL) {
     return NULL;

source/dnode/vnode/src/tq/tqPush.c

@@ -12,3 +12,246 @@
  * 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 "tq.h"
+
+void tqTmrRspFunc(void* param, void* tmrId) {
+  STqHandle* pHandle = (STqHandle*)param;
+  atomic_store_8(&pHandle->pushHandle.tmrStopped, 1);
+}
+
+int32_t tqExecFromInputQ(STQ* pTq, STqHandle* pHandle) {
+  SMqDataBlkRsp rsp = {0};
+  // 1. guard and set status executing
+  int8_t execStatus =
+      atomic_val_compare_exchange_8(&pHandle->pushHandle.execStatus, TASK_STATUS__IDLE, TASK_STATUS__EXECUTING);
+  if (execStatus == TASK_STATUS__IDLE) {
+    SStreamDataSubmit* pSubmit = NULL;
+    // 2. check processedVer
+    // 2.1. if not missed, get msg from queue
+    // 2.2. if missed, scan wal
+    pSubmit = streamQNextItem(&pHandle->pushHandle.inputQ);
+    while (pHandle->pushHandle.processedVer <= pSubmit->ver) {
+      // read from wal
+    }
+    while (pHandle->pushHandle.processedVer > pSubmit->ver + 1) {
+      streamQSetSuccess(&pHandle->pushHandle.inputQ);
+      streamDataSubmitRefDec(pSubmit);
+      pSubmit = streamQNextItem(&pHandle->pushHandle.inputQ);
+      if (pSubmit == NULL) break;
+    }
+    // 3. exec, after each success, update processed ver
+    // first run
+    while (pSubmit != NULL) {
+      ASSERT(pSubmit->ver == pHandle->pushHandle.processedVer + 1);
+      if (tqDataExec(pTq, &pHandle->execHandle, pSubmit->data, &rsp, 0) < 0) {
+        /*ASSERT(0);*/
+      }
+      // update processed
+      atomic_store_64(&pHandle->pushHandle.processedVer, pSubmit->ver);
+      streamQSetSuccess(&pHandle->pushHandle.inputQ);
+      streamDataSubmitRefDec(pSubmit);
+      if (rsp.blockNum > 0) {
+        goto SEND_RSP;
+      } else {
+        pSubmit = streamQNextItem(&pHandle->pushHandle.inputQ);
+      }
+    }
+    // set exec status closing
+    atomic_store_8(&pHandle->pushHandle.execStatus, TASK_STATUS__CLOSING);
+    // second run
+    while (pSubmit != NULL) {
+      ASSERT(pSubmit->ver == pHandle->pushHandle.processedVer + 1);
+      if (tqDataExec(pTq, &pHandle->execHandle, pSubmit->data, &rsp, 0) < 0) {
+        /*ASSERT(0);*/
+      }
+      // update processed
+      atomic_store_64(&pHandle->pushHandle.processedVer, pSubmit->ver);
+      streamQSetSuccess(&pHandle->pushHandle.inputQ);
+      streamDataSubmitRefDec(pSubmit);
+      if (rsp.blockNum > 0) {
+        goto SEND_RSP;
+      } else {
+        pSubmit = streamQNextItem(&pHandle->pushHandle.inputQ);
+      }
+    }
+    // set exec status idle
+    atomic_store_8(&pHandle->pushHandle.execStatus, TASK_STATUS__IDLE);
+  }
+SEND_RSP:
+  // 4. if get result
+  // 4.1 set exec input status blocked and exec status idle
+  atomic_store_8(&pHandle->pushHandle.execStatus, TASK_STATUS__IDLE);
+  // 4.2 rpc send
+  rsp.rspOffset = pHandle->pushHandle.processedVer;
+  /*if (tqSendPollRsp(pTq, pMsg, pReq, &rsp) < 0) {*/
+  /*return -1;*/
+  /*}*/
+  // 4.3 clear rpc info
+  memset(&pHandle->pushHandle.rpcInfo, 0, sizeof(SRpcHandleInfo));
+  return 0;
+}
+
+int32_t tqOpenPushHandle(STQ* pTq, STqHandle* pHandle) {
+  memset(&pHandle->pushHandle, 0, sizeof(STqPushHandle));
+  pHandle->pushHandle.inputQ.queue = taosOpenQueue();
+  pHandle->pushHandle.inputQ.qall = taosAllocateQall();
+  if (pHandle->pushHandle.inputQ.queue == NULL || pHandle->pushHandle.inputQ.qall == NULL) {
+    if (pHandle->pushHandle.inputQ.queue) {
+      taosCloseQueue(pHandle->pushHandle.inputQ.queue);
+    }
+    if (pHandle->pushHandle.inputQ.qall) {
+      taosFreeQall(pHandle->pushHandle.inputQ.qall);
+    }
+    return -1;
+  }
+  return 0;
+}
+
+int32_t tqPreparePush(STQ* pTq, STqHandle* pHandle, int64_t reqId, const SRpcHandleInfo* pInfo, int64_t processedVer,
+                      int64_t timeout) {
+  memcpy(&pHandle->pushHandle.rpcInfo, pInfo, sizeof(SRpcHandleInfo));
+  atomic_store_64(&pHandle->pushHandle.reqId, reqId);
+  atomic_store_64(&pHandle->pushHandle.processedVer, processedVer);
+  atomic_store_8(&pHandle->pushHandle.inputStatus, TASK_INPUT_STATUS__NORMAL);
+  atomic_store_8(&pHandle->pushHandle.tmrStopped, 0);
+  taosTmrReset(tqTmrRspFunc, (int32_t)timeout, pHandle, tqMgmt.timer, &pHandle->pushHandle.timerId);
+  return 0;
+}
+
+int32_t tqEnqueue(STqHandle* pHandle, SStreamDataSubmit* pSubmit) {
+  int8_t inputStatus = atomic_load_8(&pHandle->pushHandle.inputStatus);
+  if (inputStatus == TASK_INPUT_STATUS__NORMAL) {
+    SStreamDataSubmit* pSubmitClone = streamSubmitRefClone(pSubmit);
+    if (pSubmitClone == NULL) {
+      return -1;
+    }
+    taosWriteQitem(pHandle->pushHandle.inputQ.queue, pSubmitClone);
+    return 0;
+  }
+  return -1;
+}
+
+int32_t tqSendExecReq(STQ* pTq, STqHandle* pHandle) {
+  //
+  return 0;
+}
+
+int32_t tqEnqueueAll(STQ* pTq, SSubmitReq* pReq) {
+  void*              pIter = NULL;
+  SStreamDataSubmit* pSubmit = streamDataSubmitNew(pReq);
+  if (pSubmit == NULL) {
+    return -1;
+  }
+
+  while (1) {
+    pIter = taosHashIterate(pTq->handles, pIter);
+    if (pIter == NULL) break;
+    STqHandle* pHandle = (STqHandle*)pIter;
+    if (tqEnqueue(pHandle, pSubmit) < 0) {
+      continue;
+    }
+    int8_t execStatus = atomic_load_8(&pHandle->pushHandle.execStatus);
+    if (execStatus == TASK_STATUS__IDLE || execStatus == TASK_STATUS__CLOSING) {
+      tqSendExecReq(pTq, pHandle);
+    }
+  }
+
+  streamDataSubmitRefDec(pSubmit);
+
+  return 0;
+}
+
+int32_t tqPushMsgNew(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_t ver, SRpcHandleInfo handleInfo) {
+  if (msgType != TDMT_VND_SUBMIT) return 0;
+  void*       pIter = NULL;
+  STqHandle*  pHandle = NULL;
+  SSubmitReq* pReq = (SSubmitReq*)msg;
+  int32_t     workerId = 4;
+  int64_t     fetchOffset = ver;
+
+  while (1) {
+    pIter = taosHashIterate(pTq->pushMgr, pIter);
+    if (pIter == NULL) break;
+    pHandle = *(STqHandle**)pIter;
+
+    taosWLockLatch(&pHandle->pushHandle.lock);
+
+    SMqDataBlkRsp rsp = {0};
+    rsp.reqOffset = pHandle->pushHandle.reqOffset;
+    rsp.blockData = taosArrayInit(0, sizeof(void*));
+    rsp.blockDataLen = taosArrayInit(0, sizeof(int32_t));
+
+    if (msgType == TDMT_VND_SUBMIT) {
+      tqDataExec(pTq, &pHandle->execHandle, pReq, &rsp, workerId);
+    } else {
+      // TODO
+      ASSERT(0);
+    }
+
+    if (rsp.blockNum == 0) {
+      taosWUnLockLatch(&pHandle->pushHandle.lock);
+      continue;
+    }
+
+    ASSERT(taosArrayGetSize(rsp.blockData) == rsp.blockNum);
+    ASSERT(taosArrayGetSize(rsp.blockDataLen) == rsp.blockNum);
+
+    rsp.rspOffset = fetchOffset;
+
+    int32_t tlen = sizeof(SMqRspHead) + tEncodeSMqDataBlkRsp(NULL, &rsp);
+    void*   buf = rpcMallocCont(tlen);
+    if (buf == NULL) {
+      // todo free
+      return -1;
+    }
+
+    ((SMqRspHead*)buf)->mqMsgType = TMQ_MSG_TYPE__POLL_RSP;
+    ((SMqRspHead*)buf)->epoch = pHandle->pushHandle.epoch;
+    ((SMqRspHead*)buf)->consumerId = pHandle->pushHandle.consumerId;
+
+    void* abuf = POINTER_SHIFT(buf, sizeof(SMqRspHead));
+    tEncodeSMqDataBlkRsp(&abuf, &rsp);
+
+    SRpcMsg resp = {
+        .info = pHandle->pushHandle.rpcInfo,
+        .pCont = buf,
+        .contLen = tlen,
+        .code = 0,
+    };
+    tmsgSendRsp(&resp);
+
+    memset(&pHandle->pushHandle.rpcInfo, 0, sizeof(SRpcHandleInfo));
+    taosWUnLockLatch(&pHandle->pushHandle.lock);
+
+    tqDebug("vg %d offset %ld from consumer %ld (epoch %d) send rsp, block num: %d, reqOffset: %ld, rspOffset: %ld",
+            TD_VID(pTq->pVnode), fetchOffset, pHandle->pushHandle.consumerId, pHandle->pushHandle.epoch, rsp.blockNum,
+            rsp.reqOffset, rsp.rspOffset);
+
+    // TODO destroy
+    taosArrayDestroy(rsp.blockData);
+    taosArrayDestroy(rsp.blockDataLen);
+  }
+
+  return 0;
+}
+
+int tqPushMsg(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_t ver) {
+  if (msgType == TDMT_VND_SUBMIT) {
+    if (taosHashGetSize(pTq->pStreamTasks) == 0) return 0;
+
+    if (tdUpdateExpireWindow(pTq->pVnode->pSma, msg, ver) != 0) {
+      // TODO handle sma error
+    }
+    void* data = taosMemoryMalloc(msgLen);
+    if (data == NULL) {
+      return -1;
+    }
+    memcpy(data, msg, msgLen);
+
+    tqProcessStreamTrigger(pTq, data);
+  }
+
+  return 0;
+}
+

source/dnode/vnode/src/tq/tqRead.c

@@ -298,3 +298,38 @@ int tqReadHandleRemoveTbUidList(STqReadHandle* pHandle, const SArray* tbUidList)
 
   return 0;
 }
+
+int32_t tqUpdateTbUidList(STQ* pTq, const SArray* tbUidList, bool isAdd) {
+  void* pIter = NULL;
+  while (1) {
+    pIter = taosHashIterate(pTq->handles, pIter);
+    if (pIter == NULL) break;
+    STqHandle* pExec = (STqHandle*)pIter;
+    if (pExec->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
+      for (int32_t i = 0; i < 5; i++) {
+        int32_t code = qUpdateQualifiedTableId(pExec->execHandle.exec.execCol.task[i], tbUidList, isAdd);
+        ASSERT(code == 0);
+      }
+    } else if (pExec->execHandle.subType == TOPIC_SUB_TYPE__DB) {
+      if (!isAdd) {
+        int32_t sz = taosArrayGetSize(tbUidList);
+        for (int32_t i = 0; i < sz; i++) {
+          int64_t tbUid = *(int64_t*)taosArrayGet(tbUidList, i);
+          taosHashPut(pExec->execHandle.exec.execDb.pFilterOutTbUid, &tbUid, sizeof(int64_t), NULL, 0);
+        }
+      }
+    } else {
+      // tq update id
+    }
+  }
+  while (1) {
+    pIter = taosHashIterate(pTq->pStreamTasks, pIter);
+    if (pIter == NULL) break;
+    SStreamTask* pTask = (SStreamTask*)pIter;
+    if (pTask->inputType == STREAM_INPUT__DATA_SUBMIT) {
+      int32_t code = qUpdateQualifiedTableId(pTask->exec.executor, tbUidList, isAdd);
+      ASSERT(code == 0);
+    }
+  }
+  return 0;
+}

source/dnode/vnode/src/tq/tqSink.c

@@ -0,0 +1,34 @@
+/*
+ * 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 "tq.h"
+
+void tqTableSink(SStreamTask* pTask, void* vnode, int64_t ver, void* data) {
+  const SArray* pRes = (const SArray*)data;
+  SVnode*       pVnode = (SVnode*)vnode;
+
+  ASSERT(pTask->tbSink.pTSchema);
+  SSubmitReq* pReq = tdBlockToSubmit(pRes, pTask->tbSink.pTSchema, true, pTask->tbSink.stbUid,
+                                     pTask->tbSink.stbFullName, pVnode->config.vgId);
+  /*tPrintFixedSchemaSubmitReq(pReq, pTask->tbSink.pTSchema);*/
+  // build write msg
+  SRpcMsg msg = {
+      .msgType = TDMT_VND_SUBMIT,
+      .pCont = pReq,
+      .contLen = ntohl(pReq->length),
+  };
+
+  ASSERT(tmsgPutToQueue(&pVnode->msgCb, WRITE_QUEUE, &msg) == 0);
+}

source/dnode/vnode/src/tsdb/tsdbCommit.c

@@ -238,7 +238,7 @@ static void tsdbStartCommit(STsdb *pRepo) {
 
 static void tsdbEndCommit(STsdb *pTsdb, int eno) {
   tsdbEndFSTxn(pTsdb);
-  tsdbMemTableDestroy(pTsdb, pTsdb->imem);
+  tsdbMemTableDestroy(pTsdb->imem);
   pTsdb->imem = NULL;
   tsdbInfo("vgId:%d commit over, %s", REPO_ID(pTsdb), (eno == TSDB_CODE_SUCCESS) ? "succeed" : "failed");
 }

source/dnode/vnode/src/tsdb/tsdbDelete.c

@@ -0,0 +1,14 @@
+/*
+ * 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/>.
+ */

source/dnode/vnode/src/tsdb/tsdbMemTable.c

@@ -60,7 +60,7 @@ int tsdbMemTableCreate(STsdb *pTsdb, STsdbMemTable **ppMemTable) {
   return 0;
 }
 
-void tsdbMemTableDestroy(STsdb *pTsdb, STsdbMemTable *pMemTable) {
+void tsdbMemTableDestroy(STsdbMemTable *pMemTable) {
   if (pMemTable) {
     taosHashCleanup(pMemTable->pHashIdx);
     SSkipListIterator *pIter = tSkipListCreateIter(pMemTable->pSlIdx);
@@ -142,69 +142,6 @@ int tsdbLoadDataFromCache(STsdb *pTsdb, STable *pTable, SSkipListIterator *pIter
       } else {
         fKey = tdGetKey(filterKeys[filterIter]);
       }
-#if 0
-    } else if (fKey > rowKey) {
-      if (isRowDel) {
-        pMergeInfo->rowsDeleteFailed++;
-      } else {
-        if (pMergeInfo->rowsInserted - pMergeInfo->rowsDeleteSucceed >= maxRowsToRead) break;
-        if (pCols && pMergeInfo->nOperations >= pCols->maxPoints) break;
-
-        pMergeInfo->rowsInserted++;
-        pMergeInfo->nOperations++;
-        pMergeInfo->keyFirst = TMIN(pMergeInfo->keyFirst, rowKey);
-        pMergeInfo->keyLast = TMAX(pMergeInfo->keyLast, rowKey);
-        tsdbAppendTableRowToCols(pTable, pCols, &pSchema, row);
-      }
-
-      tSkipListIterNext(pIter);
-      row = tsdbNextIterRow(pIter);
-      if (row == NULL || TD_ROW_KEY(row) > maxKey) {
-        rowKey = INT64_MAX;
-        isRowDel = false;
-      } else {
-        rowKey = TD_ROW_KEY(row);
-        isRowDel = TD_ROW_IS_DELETED(row);
-      }
-    } else {
-      if (isRowDel) {
-        ASSERT(!keepDup);
-        if (pCols && pMergeInfo->nOperations >= pCols->maxPoints) break;
-        pMergeInfo->rowsDeleteSucceed++;
-        pMergeInfo->nOperations++;
-        tsdbAppendTableRowToCols(pTable, pCols, &pSchema, row);
-      } else {
-        if (keepDup) {
-          if (pCols && pMergeInfo->nOperations >= pCols->maxPoints) break;
-          pMergeInfo->rowsUpdated++;
-          pMergeInfo->nOperations++;
-          pMergeInfo->keyFirst = TMIN(pMergeInfo->keyFirst, rowKey);
-          pMergeInfo->keyLast = TMAX(pMergeInfo->keyLast, rowKey);
-          tsdbAppendTableRowToCols(pTable, pCols, &pSchema, row);
-        } else {
-          pMergeInfo->keyFirst = TMIN(pMergeInfo->keyFirst, fKey);
-          pMergeInfo->keyLast = TMAX(pMergeInfo->keyLast, fKey);
-        }
-      }
-
-      tSkipListIterNext(pIter);
-      row = tsdbNextIterRow(pIter);
-      if (row == NULL || TD_ROW_KEY(row) > maxKey) {
-        rowKey = INT64_MAX;
-        isRowDel = false;
-      } else {
-        rowKey = TD_ROW_KEY(row);
-        isRowDel = TD_ROW_IS_DELETED(row);
-      }
-
-      filterIter++;
-      if (filterIter >= nFilterKeys) {
-        fKey = INT64_MAX;
-      } else {
-        fKey = tdGetKey(filterKeys[filterIter]);
-      }
-    }
-#endif
 #if 1
     } else if (fKey > rowKey) {
       if (isRowDel) {
@@ -321,7 +258,7 @@ int tsdbInsertTableData(STsdb *pTsdb, SSubmitMsgIter *pMsgIter, SSubmitBlk *pBlo
     terrno = TSDB_CODE_PAR_TABLE_NOT_EXIST;
     return -1;
   }
-  if(pRsp->tblFName) strcat(pRsp->tblFName, mr.me.name);
+  if (pRsp->tblFName) strcat(pRsp->tblFName, mr.me.name);
 
   if (mr.me.type == TSDB_NORMAL_TABLE) {
     sverNew = mr.me.ntbEntry.schemaRow.version;

source/dnode/vnode/src/tsdb/tsdbMemTable2.c

@@ -15,52 +15,308 @@
 
 #include "tsdb.h"
 
-typedef struct SMemTable          SMemTable;
+typedef struct SMemData         SMemData;
-typedef struct SMemData           SMemData;
+typedef struct SMemSkipList     SMemSkipList;
-typedef struct SMemSkipList       SMemSkipList;
+typedef struct SMemSkipListNode SMemSkipListNode;
-typedef struct SMemSkipListNode   SMemSkipListNode;
-typedef struct SMemSkipListCurosr SMemSkipListCurosr;
-
-#define SL_MAX_LEVEL 5
-
-struct SMemTable {
-  STsdb              *pTsdb;
-  TSKEY               minKey;
-  TSKEY               maxKey;
-  int64_t             minVer;
-  int64_t             maxVer;
-  int64_t             nRows;
-  int32_t             nHash;
-  int32_t             nBucket;
-  SMemData          **pBuckets;
-  SMemSkipListCurosr *pSlc;
-};
 
 struct SMemSkipListNode {
   int8_t            level;
-  SMemSkipListNode *forwards[1];  // Windows does not allow 0
+  SMemSkipListNode *forwards[0];
 };
 
 struct SMemSkipList {
-  uint32_t         seed;
+  uint32_t          seed;
-  int8_t           maxLevel;
+  int32_t           size;
-  int8_t           level;
+  int8_t            maxLevel;
-  int32_t          size;
+  int8_t            level;
-  SMemSkipListNode pHead[1];  // Windows does not allow 0
+  SMemSkipListNode *pHead;
+  SMemSkipListNode *pTail;
 };
 
 struct SMemData {
-  SMemData    *pHashNext;
   tb_uid_t     suid;
   tb_uid_t     uid;
-  TSKEY        minKey;
+  TSDBKEY      minKey;
-  TSKEY        maxKey;
+  TSDBKEY      maxKey;
-  int64_t      minVer;
+  SDelOp      *delOpHead;
-  int64_t      maxVer;
+  SDelOp      *delOpTail;
-  int64_t      nRows;
   SMemSkipList sl;
 };
 
+struct SMemTable {
+  STsdb  *pTsdb;
+  int32_t nRef;
+  TSDBKEY minKey;
+  TSDBKEY maxKey;
+  int64_t nRows;
+  SArray *pArray;  // SArray<SMemData>
+};
+
+#define SL_NODE_SIZE(l)        (sizeof(SMemSkipListNode) + sizeof(SMemSkipListNode *) * (l)*2)
+#define SL_NODE_HALF_SIZE(l)   (sizeof(SMemSkipListNode) + sizeof(SMemSkipListNode *) * (l))
+#define SL_NODE_FORWARD(n, l)  ((n)->forwards[l])
+#define SL_NODE_BACKWARD(n, l) ((n)->forwards[(n)->level + (l)])
+#define SL_NODE_DATA(n)        (&SL_NODE_BACKWARD(n, (n)->level))
+
+#define SL_HEAD_FORWARD(sl, l)  SL_NODE_FORWARD((sl)->pHead, l)
+#define SL_TAIL_BACKWARD(sl, l) SL_NODE_FORWARD((sl)->pTail, l)
+
+static int32_t tsdbGetOrCreateMemData(SMemTable *pMemTable, tb_uid_t suid, tb_uid_t uid, SMemData **ppMemData);
+static int     memDataPCmprFn(const void *p1, const void *p2);
+static int32_t tPutTSDBRow(uint8_t *p, TSDBROW *pRow);
+static int32_t tGetTSDBRow(uint8_t *p, TSDBROW *pRow);
+static int8_t  tsdbMemSkipListRandLevel(SMemSkipList *pSl);
+
+// SMemTable ==============================================
+int32_t tsdbMemTableCreate2(STsdb *pTsdb, SMemTable **ppMemTable) {
+  int32_t    code = 0;
+  SMemTable *pMemTable = NULL;
+
+  pMemTable = (SMemTable *)taosMemoryCalloc(1, sizeof(*pMemTable));
+  if (pMemTable == NULL) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+    goto _err;
+  }
+  pMemTable->pTsdb = pTsdb;
+  pMemTable->nRef = 1;
+  pMemTable->minKey = (TSDBKEY){.version = INT64_MAX, .ts = TSKEY_MAX};
+  pMemTable->maxKey = (TSDBKEY){.version = -1, .ts = TSKEY_MIN};
+  pMemTable->nRows = 0;
+  pMemTable->pArray = taosArrayInit(512, sizeof(SMemData *));
+  if (pMemTable->pArray == NULL) {
+    taosMemoryFree(pMemTable);
+    code = TSDB_CODE_OUT_OF_MEMORY;
+    goto _err;
+  }
+
+  *ppMemTable = pMemTable;
+  return code;
+
+_err:
+  *ppMemTable = NULL;
+  return code;
+}
+
+void tsdbMemTableDestroy2(SMemTable *pMemTable) {
+  taosArrayDestroyEx(pMemTable->pArray, NULL /*TODO*/);
+  taosMemoryFree(pMemTable);
+}
+
+int32_t tsdbInsertTableData2(STsdb *pTsdb, int64_t version, SVSubmitBlk *pSubmitBlk) {
+  int32_t    code = 0;
+  SMemTable *pMemTable = (SMemTable *)pTsdb->mem;  // TODO
+  SMemData  *pMemData;
+  TSDBROW    row = {.version = version};
+
+  ASSERT(pMemTable);
+
+  {
+    // check if table exists (todo)
+  }
+
+  code = tsdbGetOrCreateMemData(pMemTable, pSubmitBlk->suid, pSubmitBlk->uid, &pMemData);
+  if (code) {
+    tsdbError("vgId:%d failed to create/get table data since %s", TD_VID(pTsdb->pVnode), tstrerror(code));
+    goto _err;
+  }
+
+  // do insert
+  int32_t           nt;
+  uint8_t          *pt;
+  int32_t           n = 0;
+  uint8_t          *p = pSubmitBlk->pData;
+  SVBufPool        *pPool = pTsdb->pVnode->inUse;
+  int8_t            level;
+  SMemSkipListNode *pNode;
+  while (n < pSubmitBlk->nData) {
+    nt = tGetTSRow(p + n, &row.tsRow);
+    n += nt;
+
+    ASSERT(n <= pSubmitBlk->nData);
+
+    // build the node
+    level = tsdbMemSkipListRandLevel(&pMemData->sl);
+    pNode = (SMemSkipListNode *)vnodeBufPoolMalloc(pPool, SL_NODE_SIZE(level) + nt + sizeof(version));
+    if (pNode == NULL) {
+      code = TSDB_CODE_OUT_OF_MEMORY;
+      goto _err;
+    }
+    pNode->level = level;
+    tPutTSDBRow((uint8_t *)SL_NODE_DATA(pNode), &row);
+
+    // put the node (todo)
+
+    // set info
+    if (tsdbKeyCmprFn(&row, &pMemData->minKey) < 0) pMemData->minKey = *(TSDBKEY *)&row;
+    if (tsdbKeyCmprFn(&row, &pMemData->maxKey) > 0) pMemData->maxKey = *(TSDBKEY *)&row;
+  }
+
+  if (tsdbKeyCmprFn(&pMemTable->minKey, &pMemData->minKey) < 0) pMemTable->minKey = pMemData->minKey;
+  if (tsdbKeyCmprFn(&pMemTable->maxKey, &pMemData->maxKey) > 0) pMemTable->maxKey = pMemData->maxKey;
+
+  return code;
+
+_err:
+  return code;
+}
+
+int32_t tsdbDeleteTableData2(STsdb *pTsdb, int64_t version, tb_uid_t suid, tb_uid_t uid, TSKEY sKey, TSKEY eKey) {
+  int32_t    code = 0;
+  SMemTable *pMemTable = (SMemTable *)pTsdb->mem;  // TODO
+  SMemData  *pMemData;
+  SVBufPool *pPool = pTsdb->pVnode->inUse;
+
+  ASSERT(pMemTable);
+
+  {
+    // check if table exists (todo)
+  }
+
+  code = tsdbGetOrCreateMemData(pMemTable, suid, uid, &pMemData);
+  if (code) {
+    goto _err;
+  }
+
+  // do delete
+  SDelOp *pDelOp = (SDelOp *)vnodeBufPoolMalloc(pPool, sizeof(*pDelOp));
+  if (pDelOp == NULL) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+    goto _err;
+  }
+  pDelOp->version = version;
+  pDelOp->sKey = sKey;
+  pDelOp->eKey = eKey;
+  pDelOp->pNext = NULL;
+  if (pMemData->delOpHead == NULL) {
+    ASSERT(pMemData->delOpTail == NULL);
+    pMemData->delOpHead = pMemData->delOpTail = pDelOp;
+  } else {
+    pMemData->delOpTail->pNext = pDelOp;
+    pMemData->delOpTail = pDelOp;
+  }
+
+  {
+    // update the state of pMemTable, pMemData, last and lastrow (todo)
+  }
+
+  tsdbDebug("vgId:%d delete data from table suid:%" PRId64 " uid:%" PRId64 " sKey:%" PRId64 " eKey:%" PRId64
+            " since %s",
+            TD_VID(pTsdb->pVnode), suid, uid, sKey, eKey, tstrerror(code));
+  return code;
+
+_err:
+  tsdbError("vgId:%d failed to delete data from table suid:%" PRId64 " uid:%" PRId64 " sKey:%" PRId64 " eKey:%" PRId64
+            " since %s",
+            TD_VID(pTsdb->pVnode), suid, uid, sKey, eKey, tstrerror(code));
+  return code;
+}
+
+static int32_t tsdbGetOrCreateMemData(SMemTable *pMemTable, tb_uid_t suid, tb_uid_t uid, SMemData **ppMemData) {
+  int32_t    code = 0;
+  int32_t    idx = 0;
+  SMemData  *pMemDataT = &(SMemData){.suid = suid, .uid = uid};
+  SMemData  *pMemData = NULL;
+  SVBufPool *pPool = pMemTable->pTsdb->pVnode->inUse;
+  int8_t     maxLevel = pMemTable->pTsdb->pVnode->config.tsdbCfg.slLevel;
+
+  // get
+  idx = taosArraySearchIdx(pMemTable->pArray, &pMemDataT, memDataPCmprFn, TD_GE);
+  if (idx >= 0) {
+    pMemData = (SMemData *)taosArrayGet(pMemTable->pArray, idx);
+    if (memDataPCmprFn(&pMemDataT, &pMemData) == 0) goto _exit;
+  }
+
+  // create
+  pMemData = vnodeBufPoolMalloc(pPool, sizeof(*pMemData) + SL_NODE_HALF_SIZE(maxLevel) * 2);
+  if (pMemData == NULL) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+    goto _err;
+  }
+  pMemData->suid = suid;
+  pMemData->uid = uid;
+  pMemData->minKey = (TSDBKEY){.version = INT64_MAX, .ts = TSKEY_MAX};
+  pMemData->maxKey = (TSDBKEY){.version = -1, .ts = TSKEY_MIN};
+  pMemData->delOpHead = pMemData->delOpTail = NULL;
+  pMemData->sl.seed = taosRand();
+  pMemData->sl.size = 0;
+  pMemData->sl.maxLevel = maxLevel;
+  pMemData->sl.level = 0;
+  pMemData->sl.pHead = (SMemSkipListNode *)&pMemData[1];
+  pMemData->sl.pTail = (SMemSkipListNode *)POINTER_SHIFT(pMemData->sl.pHead, SL_NODE_HALF_SIZE(maxLevel));
+
+  for (int8_t iLevel = 0; iLevel < pMemData->sl.maxLevel; iLevel++) {
+    SL_HEAD_FORWARD(&pMemData->sl, iLevel) = pMemData->sl.pTail;
+    SL_TAIL_BACKWARD(&pMemData->sl, iLevel) = pMemData->sl.pHead;
+  }
+
+  if (idx < 0) idx = 0;
+  if (taosArrayInsert(pMemTable->pArray, idx, &pMemData) == NULL) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+    goto _err;
+  }
+
+_exit:
+  *ppMemData = pMemData;
+  return code;
+
+_err:
+  *ppMemData = NULL;
+  return code;
+}
+
+static int memDataPCmprFn(const void *p1, const void *p2) {
+  SMemData *pMemData1 = *(SMemData **)p1;
+  SMemData *pMemData2 = *(SMemData **)p2;
+
+  if (pMemData1->suid < pMemData2->suid) {
+    return -1;
+  } else if (pMemData1->suid > pMemData2->suid) {
+    return 1;
+  }
+
+  if (pMemData1->uid < pMemData2->uid) {
+    return -1;
+  } else if (pMemData1->uid > pMemData2->uid) {
+    return 1;
+  }
+
+  return 0;
+}
+
+static int32_t tPutTSDBRow(uint8_t *p, TSDBROW *pRow) {
+  int32_t n = 0;
+
+  n += tPutI64(p ? p + n : p, pRow->version);
+  n += tPutTSRow(p ? p + n : p, &pRow->tsRow);
+
+  return n;
+}
+
+static int32_t tGetTSDBRow(uint8_t *p, TSDBROW *pRow) {
+  int32_t n = 0;
+
+  n += tGetI64(p + n, &pRow->version);
+  n += tGetTSRow(p + n, &pRow->tsRow);
+
+  return n;
+}
+
+static FORCE_INLINE int8_t tsdbMemSkipListRandLevel(SMemSkipList *pSl) {
+  int8_t         level = 1;
+  int8_t         tlevel = TMIN(pSl->maxLevel, pSl->level + 1);
+  const uint32_t factor = 4;
+
+  while ((taosRandR(&pSl->seed) % factor) == 0 && level < tlevel) {
+    level++;
+  }
+
+  return level;
+}
+
+#if 0  //====================================================================================
+
+#define SL_MAX_LEVEL 5
+
 struct SMemSkipListCurosr {
   SMemSkipList     *pSl;
   SMemSkipListNode *pNodes[SL_MAX_LEVEL];
@@ -74,12 +330,6 @@ typedef struct {
 
 #define HASH_BUCKET(SUID, UID, NBUCKET) (TABS((SUID) + (UID)) % (NBUCKET))
 
-#define SL_NODE_SIZE(l)        (sizeof(SMemSkipListNode) + sizeof(SMemSkipListNode *) * (l)*2)
-#define SL_NODE_HALF_SIZE(l)   (sizeof(SMemSkipListNode) + sizeof(SMemSkipListNode *) * (l))
-#define SL_NODE_FORWARD(n, l)  ((n)->forwards[l])
-#define SL_NODE_BACKWARD(n, l) ((n)->forwards[(n)->level + (l)])
-#define SL_NODE_DATA(n)        (&SL_NODE_BACKWARD(n, (n)->level))
-
 #define SL_HEAD_NODE(sl)            ((sl)->pHead)
 #define SL_TAIL_NODE(sl)            ((SMemSkipListNode *)&SL_NODE_FORWARD(SL_HEAD_NODE(sl), (sl)->maxLevel))
 #define SL_HEAD_NODE_FORWARD(n, l)  SL_NODE_FORWARD(n, l)
@@ -99,50 +349,7 @@ static int32_t tsdbMemSkipListCursorMoveToNext(SMemSkipListCurosr *pSlc);
 static int32_t tsdbMemSkipListCursorMoveToPrev(SMemSkipListCurosr *pSlc);
 static SMemSkipListNode *tsdbMemSkipListNodeCreate(SVBufPool *pPool, SMemSkipList *pSl, const STsdbRow *pTRow);
 
-// SMemTable
+// SMemTable ========================
-int32_t tsdbMemTableCreate2(STsdb *pTsdb, SMemTable **ppMemTb) {
-  SMemTable *pMemTb = NULL;
-
-  pMemTb = taosMemoryCalloc(1, sizeof(*pMemTb));
-  if (pMemTb == NULL) {
-    terrno = TSDB_CODE_OUT_OF_MEMORY;
-    return -1;
-  }
-
-  pMemTb->pTsdb = pTsdb;
-  pMemTb->minKey = TSKEY_MAX;
-  pMemTb->maxKey = TSKEY_MIN;
-  pMemTb->minVer = -1;
-  pMemTb->maxVer = -1;
-  pMemTb->nRows = 0;
-  pMemTb->nHash = 0;
-  pMemTb->nBucket = 1024;
-  pMemTb->pBuckets = taosMemoryCalloc(pMemTb->nBucket, sizeof(*pMemTb->pBuckets));
-  if (pMemTb->pBuckets == NULL) {
-    terrno = TSDB_CODE_OUT_OF_MEMORY;
-    taosMemoryFree(pMemTb);
-    return -1;
-  }
-  if (tsdbMemSkipListCursorCreate(pTsdb->pVnode->config.tsdbCfg.slLevel, &pMemTb->pSlc) < 0) {
-    terrno = TSDB_CODE_OUT_OF_MEMORY;
-    taosMemoryFree(pMemTb->pBuckets);
-    taosMemoryFree(pMemTb);
-  }
-
-  *ppMemTb = pMemTb;
-  return 0;
-}
-
-int32_t tsdbMemTableDestroy2(STsdb *pTsdb, SMemTable *pMemTb) {
-  if (pMemTb) {
-    // loop to destroy the contents (todo)
-    tsdbMemSkipListCursorDestroy(pMemTb->pSlc);
-    taosMemoryFree(pMemTb->pBuckets);
-    taosMemoryFree(pMemTb);
-  }
-  return 0;
-}
-
 int32_t tsdbInsertData2(SMemTable *pMemTb, int64_t version, const SVSubmitBlk *pSubmitBlk) {
   SMemData  *pMemData;
   STsdb     *pTsdb = pMemTb->pTsdb;
@@ -253,18 +460,6 @@ int32_t tsdbInsertData2(SMemTable *pMemTb, int64_t version, const SVSubmitBlk *p
   return 0;
 }
 
-static FORCE_INLINE int8_t tsdbMemSkipListRandLevel(SMemSkipList *pSl) {
-  int8_t         level = 1;
-  int8_t         tlevel = TMIN(pSl->maxLevel, pSl->level + 1);
-  const uint32_t factor = 4;
-
-  while ((taosRandR(&pSl->seed) % factor) == 0 && level < tlevel) {
-    level++;
-  }
-
-  return level;
-}
-
 static FORCE_INLINE int32_t tsdbEncodeRow(SEncoder *pEncoder, const STsdbRow *pRow) {
   if (tEncodeI64(pEncoder, pRow->version) < 0) return -1;
   if (tEncodeBinary(pEncoder, (const uint8_t *)pRow->pRow, pRow->szRow) < 0) return -1;
@@ -377,4 +572,5 @@ static SMemSkipListNode *tsdbMemSkipListNodeCreate(SVBufPool *pPool, SMemSkipLis
   }
 
   return pNode;
-}
+}
+#endif

source/dnode/vnode/src/tsdb/tsdbRead.c

@@ -13,8 +13,8 @@
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */
 
-#include "vnode.h"
 #include "tsdb.h"
+#include "vnode.h"
 
 #define EXTRA_BYTES                2
 #define ASCENDING_TRAVERSE(o)      (o == TSDB_ORDER_ASC)
@@ -327,13 +327,13 @@ static void setQueryTimewindow(STsdbReadHandle* pTsdbReadHandle, SQueryTableData
 
   if (updateTs) {
     tsdbDebug("%p update the query time window, old:%" PRId64 " - %" PRId64 ", new:%" PRId64 " - %" PRId64 ", %s",
-              pTsdbReadHandle, pCond->twindows[tWinIdx].skey, pCond->twindows[tWinIdx].ekey, pTsdbReadHandle->window.skey,
+              pTsdbReadHandle, pCond->twindows[tWinIdx].skey, pCond->twindows[tWinIdx].ekey,
-              pTsdbReadHandle->window.ekey, pTsdbReadHandle->idStr);
+              pTsdbReadHandle->window.skey, pTsdbReadHandle->window.ekey, pTsdbReadHandle->idStr);
   }
 }
 
 static STsdb* getTsdbByRetentions(SVnode* pVnode, STsdbReadHandle* pReadHandle, TSKEY winSKey, SRetention* retentions) {
-  if (vnodeIsRollup(pVnode)) {
+  if (VND_IS_RSMA(pVnode)) {
     int     level = 0;
     int64_t now = taosGetTimestamp(pVnode->config.tsdbCfg.precision);
 
@@ -586,7 +586,8 @@ void tsdbResetReadHandle(tsdbReaderT queryHandle, SQueryTableDataCond* pCond, in
   resetCheckInfo(pTsdbReadHandle);
 }
 
-void tsdbResetQueryHandleForNewTable(tsdbReaderT queryHandle, SQueryTableDataCond* pCond, STableListInfo* tableList, int32_t tWinIdx) {
+void tsdbResetQueryHandleForNewTable(tsdbReaderT queryHandle, SQueryTableDataCond* pCond, STableListInfo* tableList,
+                                     int32_t tWinIdx) {
   STsdbReadHandle* pTsdbReadHandle = queryHandle;
 
   pTsdbReadHandle->order = pCond->order;
@@ -2845,6 +2846,22 @@ int32_t tsdbGetAllTableList(SMeta* pMeta, uint64_t uid, SArray* list) {
   return TSDB_CODE_SUCCESS;
 }
 
+int32_t tsdbGetCtbIdList(SMeta* pMeta, int64_t suid, SArray* list) {
+  SMCtbCursor* pCur = metaOpenCtbCursor(pMeta, suid);
+
+  while (1) {
+    tb_uid_t id = metaCtbCursorNext(pCur);
+    if (id == 0) {
+      break;
+    }
+
+    taosArrayPush(list, &id);
+  }
+
+  metaCloseCtbCursor(pCur);
+  return TSDB_CODE_SUCCESS;
+}
+
 static void destroyHelper(void* param) {
   if (param == NULL) {
     return;

source/dnode/vnode/src/tsdb/tsdbTDBImpl.c

@@ -1,128 +0,0 @@
-/*
- * 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/>.
- */
-
-#define ALLOW_FORBID_FUNC
-
-#include "tsdb.h"
-
-int32_t tsdbOpenDBEnv(TDB **ppEnv, const char *path) {
-  int ret = 0;
-
-  if (path == NULL) return -1;
-
-  ret = tdbOpen(path, 4096, 256, ppEnv);  // use as param
-
-  if (ret != 0) {
-    tsdbError("Failed to create tsdb db env, ret = %d", ret);
-    return -1;
-  }
-
-  return 0;
-}
-
-int32_t tsdbCloseDBEnv(TDB *pEnv) { return tdbClose(pEnv); }
-
-static inline int tsdbSmaKeyCmpr(const void *arg1, int len1, const void *arg2, int len2) {
-  const SSmaKey *pKey1 = (const SSmaKey *)arg1;
-  const SSmaKey *pKey2 = (const SSmaKey *)arg2;
-
-  ASSERT(len1 == len2 && len1 == sizeof(SSmaKey));
-
-  if (pKey1->skey < pKey2->skey) {
-    return -1;
-  } else if (pKey1->skey > pKey2->skey) {
-    return 1;
-  }
-  if (pKey1->groupId < pKey2->groupId) {
-    return -1;
-  } else if (pKey1->groupId > pKey2->groupId) {
-    return 1;
-  }
-
-  return 0;
-}
-
-static int32_t tsdbOpenDBDb(TTB **ppDB, TDB *pEnv, const char *pFName) {
-  int           ret;
-  tdb_cmpr_fn_t compFunc;
-
-  // Create a database
-  compFunc = tsdbSmaKeyCmpr;
-  ret = tdbTbOpen(pFName, -1, -1, compFunc, pEnv, ppDB);
-
-  return 0;
-}
-
-static int32_t tsdbCloseDBDb(TTB *pDB) { return tdbTbClose(pDB); }
-
-int32_t tsdbOpenDBF(TDB *pEnv, SDBFile *pDBF) {
-  // TEnv is shared by a group of SDBFile
-  if (!pEnv || !pDBF) {
-    terrno = TSDB_CODE_INVALID_PTR;
-    return -1;
-  }
-
-  // Open DBF
-  if (tsdbOpenDBDb(&(pDBF->pDB), pEnv, pDBF->path) < 0) {
-    terrno = TSDB_CODE_TDB_INIT_FAILED;
-    tsdbCloseDBDb(pDBF->pDB);
-    return -1;
-  }
-
-  return 0;
-}
-
-int32_t tsdbCloseDBF(SDBFile *pDBF) {
-  int32_t ret = 0;
-  if (pDBF->pDB) {
-    ret = tsdbCloseDBDb(pDBF->pDB);
-    pDBF->pDB = NULL;
-  }
-  taosMemoryFreeClear(pDBF->path);
-  return ret;
-}
-
-int32_t tsdbSaveSmaToDB(SDBFile *pDBF, void *pKey, int32_t keyLen, void *pVal, int32_t valLen, TXN *txn) {
-  int32_t ret;
-
-  ret = tdbTbInsert(pDBF->pDB, pKey, keyLen, pVal, valLen, txn);
-  if (ret < 0) {
-    tsdbError("Failed to create insert sma data into db, ret = %d", ret);
-    return -1;
-  }
-
-  return 0;
-}
-
-void *tsdbGetSmaDataByKey(SDBFile *pDBF, const void *pKey, int32_t keyLen, int32_t *valLen) {
-  void *pVal = NULL;
-  int   ret;
-
-  ret = tdbTbGet(pDBF->pDB, pKey, keyLen, &pVal, valLen);
-
-  if (ret < 0) {
-    tsdbError("Failed to get sma data from db, ret = %d", ret);
-    return NULL;
-  }
-
-  ASSERT(*valLen >= 0);
-
-  // TODO: lock?
-  // TODO: Would the key/value be destoryed during return the data?
-  // TODO: How about the key is updated while value length is changed? The original value buffer would be freed
-  // automatically?
-
-  return pVal;
-}

source/dnode/vnode/src/vnd/vnodeCfg.c

@@ -57,6 +57,7 @@ int vnodeEncodeConfig(const void *pObj, SJson *pJson) {
   if (tjsonAddIntegerToObject(pJson, "isHeap", pCfg->isHeap) < 0) return -1;
   if (tjsonAddIntegerToObject(pJson, "isWeak", pCfg->isWeak) < 0) return -1;
   if (tjsonAddIntegerToObject(pJson, "isTsma", pCfg->isTsma) < 0) return -1;
+  if (tjsonAddIntegerToObject(pJson, "isRsma", pCfg->isRsma) < 0) return -1;
   if (tjsonAddIntegerToObject(pJson, "precision", pCfg->tsdbCfg.precision) < 0) return -1;
   if (tjsonAddIntegerToObject(pJson, "update", pCfg->tsdbCfg.update) < 0) return -1;
   if (tjsonAddIntegerToObject(pJson, "compression", pCfg->tsdbCfg.compression) < 0) return -1;
@@ -133,6 +134,8 @@ int vnodeDecodeConfig(const SJson *pJson, void *pObj) {
   if(code < 0) return -1;
   tjsonGetNumberValue(pJson, "isTsma", pCfg->isTsma, code);
   if(code < 0) return -1;
+  tjsonGetNumberValue(pJson, "isRsma", pCfg->isRsma, code);
+  if(code < 0) return -1;
   tjsonGetNumberValue(pJson, "precision", pCfg->tsdbCfg.precision, code);
   if(code < 0) return -1;
   tjsonGetNumberValue(pJson, "update", pCfg->tsdbCfg.update, code);

source/dnode/vnode/src/vnd/vnodeCommit.c

@@ -230,7 +230,7 @@ int vnodeCommit(SVnode *pVnode) {
     return -1;
   }
 
-  if(vnodeIsRollup(pVnode)) {
+  if (VND_IS_RSMA(pVnode)) {
     if (tsdbCommit(VND_RSMA0(pVnode)) < 0) {
       ASSERT(0);
       return -1;
@@ -250,7 +250,6 @@ int vnodeCommit(SVnode *pVnode) {
     }
   }
 
-  
   if (tqCommit(pVnode->pTq) < 0) {
     ASSERT(0);
     return -1;

source/dnode/vnode/src/vnd/vnodeModule.c

@@ -69,6 +69,9 @@ int vnodeInit(int nthreads) {
   if (walInit() < 0) {
     return -1;
   }
+  if (tqInit() < 0) {
+    return -1;
+  }
 
   return 0;
 }
@@ -94,6 +97,9 @@ void vnodeCleanup() {
   taosMemoryFreeClear(vnodeGlobal.threads);
   taosThreadCondDestroy(&(vnodeGlobal.hasTask));
   taosThreadMutexDestroy(&(vnodeGlobal.mutex));
+
+  walCleanUp();
+  tqCleanUp();
 }
 
 int vnodeScheduleTask(int (*execute)(void*), void* arg) {
@@ -155,4 +161,4 @@ static void* loop(void* arg) {
   }
 
   return NULL;
-}
+}

source/dnode/vnode/src/vnd/vnodeOpen.c

@@ -97,7 +97,7 @@ SVnode *vnodeOpen(const char *path, STfs *pTfs, SMsgCb msgCb) {
   }
 
   // open tsdb
-  if (!vnodeIsRollup(pVnode) && tsdbOpen(pVnode, &VND_TSDB(pVnode), VNODE_TSDB_DIR, NULL) < 0) {
+  if (!VND_IS_RSMA(pVnode) && tsdbOpen(pVnode, &VND_TSDB(pVnode), VNODE_TSDB_DIR, NULL) < 0) {
     vError("vgId:%d failed to open vnode tsdb since %s", TD_VID(pVnode), tstrerror(terrno));
     goto _err;
   }
@@ -189,4 +189,4 @@ void vnodeStop(SVnode *pVnode) {}
 
 int64_t vnodeGetSyncHandle(SVnode *pVnode) { return pVnode->sync; }
 
-void vnodeGetSnapshot(SVnode *pVnode, SSnapshot *pSnapshot) { pSnapshot->lastApplyIndex = pVnode->state.committed; }
+void vnodeGetSnapshot(SVnode *pVnode, SSnapshot *pSnapshot) { pSnapshot->lastApplyIndex = pVnode->state.committed; }

source/dnode/vnode/src/vnd/vnodeSvr.c

@@ -22,7 +22,7 @@ static int vnodeProcessCreateTbReq(SVnode *pVnode, int64_t version, void *pReq,
 static int vnodeProcessAlterTbReq(SVnode *pVnode, int64_t version, void *pReq, int32_t len, SRpcMsg *pRsp);
 static int vnodeProcessDropTbReq(SVnode *pVnode, int64_t version, void *pReq, int32_t len, SRpcMsg *pRsp);
 static int vnodeProcessSubmitReq(SVnode *pVnode, int64_t version, void *pReq, int32_t len, SRpcMsg *pRsp);
-static int vnodeProcessCreateTSmaReq(SVnode *pVnode, int64_t version, void *pReq, int len, SRpcMsg *pRsp);
+static int vnodeProcessCreateTSmaReq(SVnode *pVnode, int64_t version, void *pReq, int32_t len, SRpcMsg *pRsp);
 
 int32_t vnodePreprocessReq(SVnode *pVnode, SRpcMsg *pMsg) {
   SDecoder dc = {0};
@@ -88,6 +88,9 @@ int32_t vnodePreprocessReq(SVnode *pVnode, SRpcMsg *pMsg) {
       }
 
     } break;
+    case TDMT_VND_ALTER_REPLICA: {
+      vnodeSyncAlter(pVnode, pMsg);
+    } break;
     default:
       break;
   }
@@ -154,7 +157,7 @@ int vnodeProcessWriteReq(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRpcMsg
                               pMsg->contLen - sizeof(SMsgHead)) < 0) {
       }
     } break;
-    case TDMT_VND_ALTER_VNODE:
+    case TDMT_VND_ALTER_CONFIG:
       break;
     default:
       ASSERT(0);
@@ -789,28 +792,30 @@ _exit:
 
   // TODO: the partial success scenario and the error case
   // TODO: refactor
-  if ((terrno == TSDB_CODE_SUCCESS || terrno == TSDB_CODE_TDB_TABLE_ALREADY_EXIST) &&
+  if ((terrno == TSDB_CODE_SUCCESS) && (pRsp->code == TSDB_CODE_SUCCESS)) {
-      (pRsp->code == TSDB_CODE_SUCCESS)) {
     tdProcessRSmaSubmit(pVnode->pSma, pReq, STREAM_DATA_TYPE_SUBMIT_BLOCK);
   }
 
   return 0;
 }
 
-static int vnodeProcessCreateTSmaReq(SVnode *pVnode, int64_t version, void *pReq, int len, SRpcMsg *pRsp) {
+static int vnodeProcessCreateTSmaReq(SVnode *pVnode, int64_t version, void *pReq, int32_t len, SRpcMsg *pRsp) {
   SVCreateTSmaReq req = {0};
   SDecoder        coder;
 
-  pRsp->msgType = TDMT_VND_CREATE_SMA_RSP;
+  if (pRsp) {
-  pRsp->code = TSDB_CODE_SUCCESS;
+    pRsp->msgType = TDMT_VND_CREATE_SMA_RSP;
-  pRsp->pCont = NULL;
+    pRsp->code = TSDB_CODE_SUCCESS;
-  pRsp->contLen = 0;
+    pRsp->pCont = NULL;
+    pRsp->contLen = 0;
+  }
 
   // decode and process req
   tDecoderInit(&coder, pReq, len);
 
   if (tDecodeSVCreateTSmaReq(&coder, &req) < 0) {
-    pRsp->code = terrno;
+    terrno = TSDB_CODE_MSG_DECODE_ERROR;
+    if (pRsp) pRsp->code = terrno;
     goto _err;
   }
 
@@ -818,18 +823,30 @@ static int vnodeProcessCreateTSmaReq(SVnode *pVnode, int64_t version, void *pReq
   req.timezoneInt = tsTimezone;
 
   if (tdProcessTSmaCreate(pVnode->pSma, version, (const char *)&req) < 0) {
-    pRsp->code = terrno;
+    if (pRsp) pRsp->code = terrno;
     goto _err;
   }
 
   tDecoderClear(&coder);
-  vDebug("vgId:%d success to create tsma %s:%" PRIi64 " for table %" PRIi64, TD_VID(pVnode), req.indexName,
+  vDebug("vgId:%d success to create tsma %s:%" PRIi64 " version %" PRIi64 " for table %" PRIi64, TD_VID(pVnode),
-         req.indexUid, req.tableUid);
+         req.indexName, req.indexUid, version, req.tableUid);
   return 0;
 
 _err:
   tDecoderClear(&coder);
-  vError("vgId:%d failed to create tsma %s:%" PRIi64 " for table %" PRIi64 " since %s", TD_VID(pVnode), req.indexName,
+  vError("vgId:%d failed to create tsma %s:%" PRIi64 " version %" PRIi64 "for table %" PRIi64 " since %s",
-         req.indexUid, req.tableUid, terrstr(terrno));
+         TD_VID(pVnode), req.indexName, req.indexUid, version, req.tableUid, terrstr(terrno));
   return -1;
 }
+
+/**
+ * @brief specific for smaDstVnode
+ *
+ * @param pVnode
+ * @param pCont
+ * @param contLen
+ * @return int32_t
+ */
+int32_t vnodeProcessCreateTSma(SVnode *pVnode, void *pCont, uint32_t contLen) {
+  return vnodeProcessCreateTSmaReq(pVnode, 1, pCont, contLen, NULL);
+}

source/dnode/vnode/src/vnd/vnodeSync.c

@@ -27,6 +27,7 @@ static int32_t   vnodeSyncGetSnapshot(SSyncFSM *pFsm, SSnapshot *pSnapshot);
 int32_t vnodeSyncOpen(SVnode *pVnode, char *path) {
   SSyncInfo syncInfo = {
       .vgId = pVnode->config.vgId,
+      .isStandBy = pVnode->config.standby,
       .syncCfg = pVnode->config.syncCfg,
       .pWal = pVnode->pWal,
       .msgcb = NULL,
@@ -49,28 +50,72 @@ int32_t vnodeSyncOpen(SVnode *pVnode, char *path) {
   return 0;
 }
 
+void vnodeSyncAlter(SVnode *pVnode, SRpcMsg *pMsg) {
+  SAlterVnodeReq req = {0};
+  if (tDeserializeSAlterVnodeReq((char *)pMsg->pCont + sizeof(SMsgHead), pMsg->contLen - sizeof(SMsgHead), &req) != 0) {
+    terrno = TSDB_CODE_INVALID_MSG;
+    vError("vgId:%d, failed to alter replica since %s", TD_VID(pVnode), terrstr());
+    SRpcMsg rsp = {.info = pMsg->info, .code = terrno};
+    tmsgSendRsp(&rsp);
+  }
+
+  vInfo("vgId:%d, start to alter vnode replica to %d", TD_VID(pVnode), req.replica);
+  SSyncCfg cfg = {.replicaNum = req.replica, .myIndex = req.selfIndex};
+  for (int32_t r = 0; r < req.replica; ++r) {
+    SNodeInfo *pNode = &cfg.nodeInfo[r];
+    tstrncpy(pNode->nodeFqdn, req.replicas[r].fqdn, sizeof(pNode->nodeFqdn));
+    pNode->nodePort = req.replicas[r].port;
+    vInfo("vgId:%d, replica:%d %s:%u", TD_VID(pVnode), r, pNode->nodeFqdn, pNode->nodePort);
+  }
+
+  if (syncReconfig(pVnode->sync, &cfg) != 0) {
+    vError("vgId:%d, failed to propose sync reconfig since %s", TD_VID(pVnode), terrstr());
+    SRpcMsg rsp = {.info = pMsg->info, .code = terrno};
+    tmsgSendRsp(&rsp);
+  }
+}
+
 void vnodeSyncStart(SVnode *pVnode) {
   syncSetMsgCb(pVnode->sync, &pVnode->msgCb);
-  syncStart(pVnode->sync);
+  if (pVnode->config.standby) {
+    syncStartStandBy(pVnode->sync);
+  } else {
+    syncStart(pVnode->sync);
+  }
 }
 
 void vnodeSyncClose(SVnode *pVnode) { syncStop(pVnode->sync); }
 
-int32_t vnodeSyncEqMsg(const SMsgCb *msgcb, SRpcMsg *pMsg) { 
+int32_t vnodeSyncEqMsg(const SMsgCb *msgcb, SRpcMsg *pMsg) {
   int32_t code = tmsgPutToQueue(msgcb, SYNC_QUEUE, pMsg);
   if (code != 0) {
     rpcFreeCont(pMsg->pCont);
+    pMsg->pCont = NULL;
   }
   return code;
 }
 
-int32_t vnodeSyncSendMsg(const SEpSet *pEpSet, SRpcMsg *pMsg) { return tmsgSendReq(pEpSet, pMsg); }
+int32_t vnodeSyncSendMsg(const SEpSet *pEpSet, SRpcMsg *pMsg) {
+  int32_t code = tmsgSendReq(pEpSet, pMsg);
+  if (code != 0) {
+    rpcFreeCont(pMsg->pCont);
+    pMsg->pCont = NULL;
+  }
+  return code;
+}
 
 int32_t vnodeSyncGetSnapshot(SSyncFSM *pFsm, SSnapshot *pSnapshot) {
   vnodeGetSnapshot(pFsm->data, pSnapshot);
   return 0;
 }
 
+void vnodeSyncReconfig(struct SSyncFSM *pFsm, SSyncCfg newCfg, SReConfigCbMeta cbMeta) {
+  SVnode *pVnode = pFsm->data;
+  vInfo("vgId:%d, sync reconfig is confirmed", TD_VID(pVnode));
+
+  // todo rpc response here
+}
+
 void vnodeSyncCommitMsg(SSyncFSM *pFsm, const SRpcMsg *pMsg, SFsmCbMeta cbMeta) {
   SyncIndex beginIndex = SYNC_INDEX_INVALID;
   if (pFsm->FpGetSnapshot != NULL) {
@@ -87,20 +132,12 @@ void vnodeSyncCommitMsg(SSyncFSM *pFsm, const SRpcMsg *pMsg, SFsmCbMeta cbMeta)
         pFsm, cbMeta.index, cbMeta.isWeak, cbMeta.code, cbMeta.state, syncUtilState2String(cbMeta.state), beginIndex);
     syncRpcMsgLog2(logBuf, (SRpcMsg *)pMsg);
 
-    SVnode       *pVnode = (SVnode *)(pFsm->data);
+    SVnode       *pVnode = pFsm->data;
     SyncApplyMsg *pSyncApplyMsg = syncApplyMsgBuild2(pMsg, pVnode->config.vgId, &cbMeta);
     SRpcMsg       applyMsg;
     syncApplyMsg2RpcMsg(pSyncApplyMsg, &applyMsg);
     syncApplyMsgDestroy(pSyncApplyMsg);
 
-    /*
-        SRpcMsg applyMsg;
-        applyMsg = *pMsg;
-        applyMsg.pCont = rpcMallocCont(applyMsg.contLen);
-        assert(applyMsg.contLen == pMsg->contLen);
-        memcpy(applyMsg.pCont, pMsg->pCont, applyMsg.contLen);
-    */
-
     // recover handle for response
     SRpcMsg saveRpcMsg;
     int32_t ret = syncGetAndDelRespRpc(pVnode->sync, cbMeta.seqNum, &saveRpcMsg);
@@ -142,14 +179,13 @@ void vnodeSyncRollBackMsg(SSyncFSM *pFsm, const SRpcMsg *pMsg, SFsmCbMeta cbMeta
 
 SSyncFSM *vnodeSyncMakeFsm(SVnode *pVnode) {
   SSyncFSM *pFsm = taosMemoryCalloc(1, sizeof(SSyncFSM));
-  memset(pFsm, 0, sizeof(*pFsm));
   pFsm->data = pVnode;
   pFsm->FpCommitCb = vnodeSyncCommitMsg;
   pFsm->FpPreCommitCb = vnodeSyncPreCommitMsg;
   pFsm->FpRollBackCb = vnodeSyncRollBackMsg;
   pFsm->FpGetSnapshot = vnodeSyncGetSnapshot;
   pFsm->FpRestoreFinishCb = NULL;
-  pFsm->FpReConfigCb = NULL;
+  pFsm->FpReConfigCb = vnodeSyncReconfig;
 
   return pFsm;
 }

source/dnode/vnode/test/tsdbSmaTest.cpp

@@ -368,7 +368,7 @@ TEST(testCase, tSma_Data_Insert_Query_Test) {
   SDiskCfg pDisks = {0};
   pDisks.level = 0;
   pDisks.primary = 1;
-  strncpy(pDisks.dir, "/var/lib/taos", TSDB_FILENAME_LEN);
+  strncpy(pDisks.dir, TD_DATA_DIR_PATH, TSDB_FILENAME_LEN);
   int32_t numOfDisks = 1;
   pTsdb->pTfs = tfsOpen(&pDisks, numOfDisks);
   EXPECT_NE(pTsdb->pTfs, nullptr);

source/libs/catalog/test/catalogTests.cpp

@@ -137,7 +137,7 @@ void ctgTestInitLogFile() {
 
   tsAsyncLog = 0;
   qDebugFlag = 159;
-  strcpy(tsLogDir, "/var/log/taos");
+  strcpy(tsLogDir, TD_LOG_DIR_PATH);
 
   ctgdEnableDebug("api");
   ctgdEnableDebug("meta");

source/libs/executor/src/scanoperator.c

@@ -313,29 +313,23 @@ void addTagPseudoColumnData(SReadHandle *pHandle, SExprInfo* pPseudoExpr, int32_
     if (fmIsScanPseudoColumnFunc(functionId)) {
       setTbNameColData(pHandle->meta, pBlock, pColInfoData, functionId);
     } else {  // these are tags
-      const char* p = NULL;
+      STagVal tagVal = {0};
-      if (pColInfoData->info.type == TSDB_DATA_TYPE_JSON) {
+      tagVal.cid = pExpr->base.pParam[0].pCol->colId;
-        const uint8_t* tmp = mr.me.ctbEntry.pTags;
+      const char *p = metaGetTableTagVal(&mr.me, pColInfoData->info.type, &tagVal);
 
-        char* data = taosMemoryCalloc(kvRowLen(tmp) + 1, 1);
+      char *data = NULL;
-        if (data == NULL) {
+      if(pColInfoData->info.type != TSDB_DATA_TYPE_JSON && p != NULL){
-          metaReaderClear(&mr);
+        data = tTagValToData((const STagVal *)p, false);
-          qError("doTagScan calloc error:%d", kvRowLen(tmp) + 1);
+      }else {
-          return;
+        data = (char*)p;
-        }
-
-        *data = TSDB_DATA_TYPE_JSON;
-        memcpy(data + 1, tmp, kvRowLen(tmp));
-        p = data;
-      } else {
-        p = metaGetTableTagVal(&mr.me, pExpr->base.pParam[0].pCol->colId);
       }
 
       for (int32_t i = 0; i < pBlock->info.rows; ++i) {
-        colDataAppend(pColInfoData, i, p, (p == NULL));
+        colDataAppend(pColInfoData, i, data, (data == NULL));
       }
-      if (pColInfoData->info.type == TSDB_DATA_TYPE_JSON) {
+      if (data && (pColInfoData->info.type != TSDB_DATA_TYPE_JSON) && p != NULL &&
-        taosMemoryFree((void*)p);
+          IS_VAR_DATA_TYPE(((const STagVal*)p)->type)) {
+        taosMemoryFree(data);
       }
     }
   }
@@ -1589,22 +1583,21 @@ static SSDataBlock* doTagScan(SOperatorInfo* pOperator) {
         STR_TO_VARSTR(str, mr.me.name);
         colDataAppend(pDst, count, str, false);
       } else {  // it is a tag value
-        if (pDst->info.type == TSDB_DATA_TYPE_JSON) {
+        STagVal val = {0};
-          const uint8_t* tmp = mr.me.ctbEntry.pTags;
+        val.cid = pExprInfo[j].base.pParam[0].pCol->colId;
-          // TODO opt perf by realloc memory
+        const char* p = metaGetTableTagVal(&mr.me, pDst->info.type, &val);
-          char* data = taosMemoryCalloc(kvRowLen(tmp) + 1, 1);
-          if (data == NULL) {
-            qError("%s failed to malloc memory, size:%d", GET_TASKID(pTaskInfo), kvRowLen(tmp) + 1);
-            longjmp(pTaskInfo->env, TSDB_CODE_OUT_OF_MEMORY);
-          }
 
-          *data = TSDB_DATA_TYPE_JSON;
+        char *data = NULL;
-          memcpy(data + 1, tmp, kvRowLen(tmp));
+        if(pDst->info.type != TSDB_DATA_TYPE_JSON && p != NULL){
-          colDataAppend(pDst, count, data, false);
+          data = tTagValToData((const STagVal *)p, false);
+        }else {
+          data = (char*)p;
+        }
+        colDataAppend(pDst, count, data, (data == NULL));
+
+        if(pDst->info.type != TSDB_DATA_TYPE_JSON && p != NULL
+            && IS_VAR_DATA_TYPE(((const STagVal *)p)->type) && data != NULL){
           taosMemoryFree(data);
-        } else {
-          const char* p = metaGetTableTagVal(&mr.me, pExprInfo[j].base.pParam[0].pCol->colId);
-          colDataAppend(pDst, count, p, (p == NULL));
         }
       }
     }
@@ -1634,8 +1627,8 @@ static void destroyTagScanOperatorInfo(void* param, int32_t numOfOutput) {
 }
 
 SOperatorInfo* createTagScanOperatorInfo(SReadHandle* pReadHandle, SExprInfo* pExpr, int32_t numOfOutput,
-                                         SSDataBlock* pResBlock, SArray* pColMatchInfo,
+                                         SSDataBlock* pResBlock, SArray* pColMatchInfo, STableListInfo* pTableListInfo,
-                                         STableListInfo* pTableListInfo, SExecTaskInfo* pTaskInfo) {
+                                         SExecTaskInfo* pTaskInfo) {
   STagScanInfo*  pInfo = taosMemoryCalloc(1, sizeof(STagScanInfo));
   SOperatorInfo* pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
   if (pInfo == NULL || pOperator == NULL) {

source/libs/function/inc/builtins.h

@@ -26,22 +26,24 @@ typedef int32_t (*FTranslateFunc)(SFunctionNode* pFunc, char* pErrBuf, int32_t l
 typedef EFuncDataRequired (*FFuncDataRequired)(SFunctionNode* pFunc, STimeWindow* pTimeWindow);
 
 typedef struct SBuiltinFuncDefinition {
-  char name[FUNCTION_NAME_MAX_LENGTH];
+  const char*        name;
-  EFunctionType type;
+  EFunctionType      type;
-  uint64_t classification;
+  uint64_t           classification;
-  FTranslateFunc translateFunc;
+  FTranslateFunc     translateFunc;
-  FFuncDataRequired dataRequiredFunc;
+  FFuncDataRequired  dataRequiredFunc;
-  FExecGetEnv getEnvFunc;
+  FExecGetEnv        getEnvFunc;
-  FExecInit initFunc;
+  FExecInit          initFunc;
-  FExecProcess processFunc;
+  FExecProcess       processFunc;
   FScalarExecProcess sprocessFunc;
-  FExecFinalize finalizeFunc;
+  FExecFinalize      finalizeFunc;
-  FExecProcess invertFunc;
+  FExecProcess       invertFunc;
-  FExecCombine combineFunc;
+  FExecCombine       combineFunc;
+  const char*        pPartialFunc;
+  const char*        pMergeFunc;
 } SBuiltinFuncDefinition;
 
 extern const SBuiltinFuncDefinition funcMgtBuiltins[];
-extern const int funcMgtBuiltinsNum;
+extern const int                    funcMgtBuiltinsNum;
 
 #ifdef __cplusplus
 }

source/libs/function/src/builtins.c

@@ -103,6 +103,28 @@ static int32_t translateInOutStr(SFunctionNode* pFunc, char* pErrBuf, int32_t le
   return TSDB_CODE_SUCCESS;
 }
 
+static int32_t translateLogarithm(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
+  int32_t numOfParams = LIST_LENGTH(pFunc->pParameterList);
+  if (1 != numOfParams && 2 != numOfParams) {
+    return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
+  }
+
+  uint8_t para1Type = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType.type;
+  if (!IS_NUMERIC_TYPE(para1Type)) {
+    return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
+  }
+
+  if (2 == numOfParams) {
+    uint8_t para2Type = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 1))->resType.type;
+    if (!IS_NUMERIC_TYPE(para2Type)) {
+      return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
+    }
+  }
+
+  pFunc->node.resType = (SDataType){.bytes = tDataTypes[TSDB_DATA_TYPE_DOUBLE].bytes, .type = TSDB_DATA_TYPE_DOUBLE};
+  return TSDB_CODE_SUCCESS;
+}
+
 static int32_t translateCount(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
   if (1 != LIST_LENGTH(pFunc->pParameterList)) {
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
@@ -156,14 +178,14 @@ static int32_t translatePercentile(SFunctionNode* pFunc, char* pErrBuf, int32_t
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  //param0
+  // param0
   SNode* pParamNode0 = nodesListGetNode(pFunc->pParameterList, 0);
   if (nodeType(pParamNode0) != QUERY_NODE_COLUMN) {
     return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
                            "The first parameter of PERCENTILE function can only be column");
   }
 
-  //param1
+  // param1
   SValueNode* pValue = (SValueNode*)nodesListGetNode(pFunc->pParameterList, 1);
 
   if (pValue->datum.i < 0 || pValue->datum.i > 100) {
@@ -178,7 +200,7 @@ static int32_t translatePercentile(SFunctionNode* pFunc, char* pErrBuf, int32_t
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  //set result type
+  // set result type
   pFunc->node.resType = (SDataType){.bytes = tDataTypes[TSDB_DATA_TYPE_DOUBLE].bytes, .type = TSDB_DATA_TYPE_DOUBLE};
   return TSDB_CODE_SUCCESS;
 }
@@ -197,14 +219,14 @@ static int32_t translateApercentile(SFunctionNode* pFunc, char* pErrBuf, int32_t
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  //param0
+  // param0
   SNode* pParamNode0 = nodesListGetNode(pFunc->pParameterList, 0);
   if (nodeType(pParamNode0) != QUERY_NODE_COLUMN) {
     return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
                            "The first parameter of APERCENTILE function can only be column");
   }
 
-  //param1
+  // param1
   SNode* pParamNode1 = nodesListGetNode(pFunc->pParameterList, 1);
   if (nodeType(pParamNode1) != QUERY_NODE_VALUE) {
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
@@ -223,7 +245,7 @@ static int32_t translateApercentile(SFunctionNode* pFunc, char* pErrBuf, int32_t
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  //param2
+  // param2
   if (3 == numOfParams) {
     uint8_t para3Type = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 2))->resType.type;
     if (!IS_VAR_DATA_TYPE(para3Type)) {
@@ -263,14 +285,14 @@ static int32_t translateTop(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  //param0
+  // param0
   SNode* pParamNode0 = nodesListGetNode(pFunc->pParameterList, 0);
   if (nodeType(pParamNode0) != QUERY_NODE_COLUMN) {
     return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
                            "The first parameter of TOP/BOTTOM function can only be column");
   }
 
-  //param1
+  // param1
   SNode* pParamNode1 = nodesListGetNode(pFunc->pParameterList, 1);
   if (nodeType(pParamNode1) != QUERY_NODE_VALUE) {
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
@@ -287,7 +309,7 @@ static int32_t translateTop(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
 
   pValue->notReserved = true;
 
-  //set result type
+  // set result type
   SDataType* pType = &((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType;
   pFunc->node.resType = (SDataType){.bytes = pType->bytes, .type = pType->type};
   return TSDB_CODE_SUCCESS;
@@ -687,8 +709,8 @@ static int32_t translateTail(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
 
     if (pValue->datum.i < ((i > 1) ? 0 : 1) || pValue->datum.i > 100) {
       return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-                            "TAIL function second parameter should be in range [1, 100], "
+                             "TAIL function second parameter should be in range [1, 100], "
-                            "third parameter should be in range [0, 100]");
+                             "third parameter should be in range [0, 100]");
     }
 
     pValue->notReserved = true;
@@ -749,7 +771,7 @@ static int32_t translateDiff(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  //param0
+  // param0
   SNode* pParamNode0 = nodesListGetNode(pFunc->pParameterList, 0);
   if (nodeType(pParamNode0) != QUERY_NODE_COLUMN) {
     return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
@@ -757,12 +779,11 @@ static int32_t translateDiff(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
   }
 
   uint8_t colType = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType.type;
-  if (!IS_SIGNED_NUMERIC_TYPE(colType) && !IS_FLOAT_TYPE(colType) &&
+  if (!IS_SIGNED_NUMERIC_TYPE(colType) && !IS_FLOAT_TYPE(colType) && TSDB_DATA_TYPE_BOOL != colType) {
-      TSDB_DATA_TYPE_BOOL != colType) {
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  //param1
+  // param1
   if (numOfParams == 2) {
     uint8_t paraType = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 1))->resType.type;
     if (!IS_INTEGER_TYPE(paraType)) {
@@ -817,11 +838,20 @@ static int32_t translateConcatImpl(SFunctionNode* pFunc, char* pErrBuf, int32_t
   int32_t resultBytes = 0;
   int32_t sepBytes = 0;
 
+  //concat_ws separator should be constant string
+  if (hasSep) {
+    SNode* pPara = nodesListGetNode(pFunc->pParameterList, 0);
+    if (nodeType(pPara) != QUERY_NODE_VALUE) {
+      return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
+                             "The first parameter of CONCAT_WS function can only be constant string");
+    }
+  }
+
   /* For concat/concat_ws function, if params have NCHAR type, promote the final result to NCHAR */
   for (int32_t i = 0; i < numOfParams; ++i) {
     SNode*  pPara = nodesListGetNode(pFunc->pParameterList, i);
     uint8_t paraType = ((SExprNode*)pPara)->resType.type;
-    if (!IS_VAR_DATA_TYPE(paraType)) {
+    if (!IS_VAR_DATA_TYPE(paraType) && TSDB_DATA_TYPE_NULL != paraType) {
       return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
     }
     if (TSDB_DATA_TYPE_NCHAR == paraType) {
@@ -880,7 +910,7 @@ static int32_t translateSubstr(SFunctionNode* pFunc, char* pErrBuf, int32_t len)
 
   if (3 == numOfParams) {
     SExprNode* p2 = (SExprNode*)nodesListGetNode(pFunc->pParameterList, 2);
-    uint8_t para2Type = p2->resType.type;
+    uint8_t    para2Type = p2->resType.type;
     if (!IS_INTEGER_TYPE(para2Type)) {
       return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
     }
@@ -921,16 +951,108 @@ static int32_t translateCast(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
   return TSDB_CODE_SUCCESS;
 }
 
+/* Following are valid ISO-8601 timezone format:
+ * 1 z/Z
+ * 2 ±hh:mm
+ * 3 ±hhmm
+ * 4 ±hh
+ *
+ */
+
+static bool validateTimezoneFormat(const SValueNode* pVal) {
+  if (TSDB_DATA_TYPE_BINARY != pVal->node.resType.type) {
+    return false;
+  }
+
+  char *tz    = varDataVal(pVal->datum.p);
+  int32_t len = varDataLen(pVal->datum.p);
+
+  if (len == 0) {
+    return false;
+  } else if (len == 1 && (tz[0] == 'z' || tz[0] == 'Z')) {
+    return true;
+  } else if ((tz[0] == '+' || tz[0] == '-')) {
+    switch (len) {
+      case 3:
+      case 5: {
+        for (int32_t i = 1; i < len; ++i) {
+          if (!isdigit(tz[i])) {
+            return false;
+          }
+        }
+        break;
+      }
+      case 6: {
+        for (int32_t i = 1; i < len; ++i) {
+          if (i == 3) {
+            if (tz[i] != ':') {
+              return false;
+            }
+            continue;
+          }
+          if (!isdigit(tz[i])) {
+            return false;
+          }
+        }
+        break;
+      }
+      default: {
+        return false;
+      }
+    }
+  } else {
+    return false;
+  }
+
+  return true;
+}
+
+void static addTimezoneParam(SNodeList* pList) {
+  char buf[6] = {0};
+  time_t t = taosTime(NULL);
+  struct tm *tmInfo = taosLocalTime(&t, NULL);
+  strftime(buf, sizeof(buf), "%z", tmInfo);
+  int32_t len = (int32_t)strlen(buf);
+
+  SValueNode* pVal = (SValueNode*)nodesMakeNode(QUERY_NODE_VALUE);
+  pVal->literal = strndup(buf, len);
+  pVal->isDuration =false;
+  pVal->translate = true;
+  pVal->node.resType.type = TSDB_DATA_TYPE_BINARY;
+  pVal->node.resType.bytes = len + VARSTR_HEADER_SIZE;
+  pVal->node.resType.precision = TSDB_TIME_PRECISION_MILLI;
+  pVal->datum.p = taosMemoryCalloc(1, len + VARSTR_HEADER_SIZE +1);
+  varDataSetLen(pVal->datum.p, len);
+  strncpy(varDataVal(pVal->datum.p), pVal->literal, len);
+
+  nodesListAppend(pList, pVal);
+}
+
 static int32_t translateToIso8601(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
-  if (1 != LIST_LENGTH(pFunc->pParameterList)) {
+  int32_t numOfParams = LIST_LENGTH(pFunc->pParameterList);
+  if (1 != numOfParams && 2 != numOfParams) {
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
+  //param0
   uint8_t paraType = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType.type;
   if (!IS_INTEGER_TYPE(paraType) && TSDB_DATA_TYPE_TIMESTAMP != paraType) {
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
+  //param1
+  if (numOfParams == 2) {
+    SValueNode* pValue = (SValueNode*)nodesListGetNode(pFunc->pParameterList, 1);
+
+    if (!validateTimezoneFormat(pValue)) {
+      return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
+                             "Invalid timzone format");
+    }
+  } else { //add default client timezone
+    addTimezoneParam(pFunc->pParameterList);
+  }
+
+  //set result type
   pFunc->node.resType = (SDataType){.bytes = 64, .type = TSDB_DATA_TYPE_BINARY};
   return TSDB_CODE_SUCCESS;
 }
@@ -1021,6 +1143,8 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
     .finalizeFunc = functionFinalize,
     .invertFunc   = countInvertFunction,
     .combineFunc = combineFunction,
+    // .pPartialFunc = "count",
+    // .pMergeFunc   = "sum"
   },
   {
     .name = "sum",
@@ -1302,7 +1426,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
     .name = "log",
     .type = FUNCTION_TYPE_LOG,
     .classification = FUNC_MGT_SCALAR_FUNC,
-    .translateFunc = translateIn2NumOutDou,
+    .translateFunc = translateLogarithm,
     .getEnvFunc   = NULL,
     .initFunc     = NULL,
     .sprocessFunc = logFunction,

source/libs/function/src/functionMgt.c

@@ -199,3 +199,81 @@ bool fmIsInvertible(int32_t funcId) {
   }
   return res;
 }
+
+static SFunctionNode* createFunction(const char* pName, SNodeList* pParameterList) {
+  SFunctionNode* pFunc = nodesMakeNode(QUERY_NODE_FUNCTION);
+  if (NULL == pFunc) {
+    return NULL;
+  }
+  strcpy(pFunc->functionName, pName);
+  pFunc->pParameterList = pParameterList;
+  char msg[64] = {0};
+  if (TSDB_CODE_SUCCESS != fmGetFuncInfo(pFunc, msg, sizeof(msg))) {
+    nodesDestroyNode(pFunc);
+    return NULL;
+  }
+  return pFunc;
+}
+
+static SColumnNode* createColumnByFunc(const SFunctionNode* pFunc) {
+  SColumnNode* pCol = nodesMakeNode(QUERY_NODE_COLUMN);
+  if (NULL == pCol) {
+    return NULL;
+  }
+  strcpy(pCol->colName, pFunc->node.aliasName);
+  pCol->node.resType = pFunc->node.resType;
+  return pCol;
+}
+
+bool fmIsDistExecFunc(int32_t funcId) {
+  if (!fmIsVectorFunc(funcId)) {
+    return true;
+  }
+  return (NULL != funcMgtBuiltins[funcId].pPartialFunc && NULL != funcMgtBuiltins[funcId].pMergeFunc);
+}
+
+static int32_t createPartialFunction(const SFunctionNode* pSrcFunc, SFunctionNode** pPartialFunc) {
+  SNodeList* pParameterList = nodesCloneList(pSrcFunc->pParameterList);
+  if (NULL == pParameterList) {
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+  *pPartialFunc = createFunction(funcMgtBuiltins[pSrcFunc->funcId].pPartialFunc, pParameterList);
+  if (NULL == *pPartialFunc) {
+    nodesDestroyList(pParameterList);
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+  snprintf((*pPartialFunc)->node.aliasName, sizeof((*pPartialFunc)->node.aliasName), "%s.%p",
+           (*pPartialFunc)->functionName, pSrcFunc);
+  return TSDB_CODE_SUCCESS;
+}
+
+static int32_t createMergeFunction(const SFunctionNode* pSrcFunc, const SFunctionNode* pPartialFunc,
+                                   SFunctionNode** pMergeFunc) {
+  SNodeList* pParameterList = NULL;
+  nodesListMakeStrictAppend(&pParameterList, createColumnByFunc(pPartialFunc));
+  *pMergeFunc = createFunction(funcMgtBuiltins[pSrcFunc->funcId].pMergeFunc, pParameterList);
+  if (NULL == *pMergeFunc) {
+    nodesDestroyList(pParameterList);
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+  strcpy((*pMergeFunc)->node.aliasName, pSrcFunc->node.aliasName);
+  return TSDB_CODE_SUCCESS;
+}
+
+int32_t fmGetDistMethod(const SFunctionNode* pFunc, SFunctionNode** pPartialFunc, SFunctionNode** pMergeFunc) {
+  if (!fmIsDistExecFunc(pFunc->funcId)) {
+    return TSDB_CODE_FAILED;
+  }
+
+  int32_t code = createPartialFunction(pFunc, pPartialFunc);
+  if (TSDB_CODE_SUCCESS == code) {
+    code = createMergeFunction(pFunc, *pPartialFunc, pMergeFunc);
+  }
+
+  if (TSDB_CODE_SUCCESS != code) {
+    nodesDestroyNode(*pPartialFunc);
+    nodesDestroyNode(*pMergeFunc);
+  }
+
+  return code;
+}

source/libs/index/inc/indexInt.h

@@ -131,8 +131,7 @@ typedef struct TFileCacheKey {
   char*    colName;
   int32_t  nColName;
 } ICacheKey;
-
+int indexFlushCacheToTFile(SIndex* sIdx, void*, bool quit);
-int indexFlushCacheToTFile(SIndex* sIdx, void*);
 
 int64_t indexAddRef(void* p);
 int32_t indexRemoveRef(int64_t ref);

source/libs/index/src/index.c

@@ -150,6 +150,7 @@ void indexClose(SIndex* sIdx) {
       indexCacheForceToMerge((void*)(*pCache));
       indexInfo("%s wait to merge", (*pCache)->colName);
       indexWait((void*)(sIdx));
+      indexInfo("%s finish to wait", (*pCache)->colName);
       iter = taosHashIterate(sIdx->colObj, iter);
       indexCacheUnRef(*pCache);
     }
@@ -454,7 +455,7 @@ static void indexDestroyFinalResult(SArray* result) {
   taosArrayDestroy(result);
 }
 
-int indexFlushCacheToTFile(SIndex* sIdx, void* cache) {
+int indexFlushCacheToTFile(SIndex* sIdx, void* cache, bool quit) {
   if (sIdx == NULL) {
     return -1;
   }
@@ -464,7 +465,7 @@ int indexFlushCacheToTFile(SIndex* sIdx, void* cache) {
 
   IndexCache* pCache = (IndexCache*)cache;
 
-  while (sIdx->quit && atomic_load_32(&pCache->merging) == 1) {
+  while (quit && atomic_load_32(&pCache->merging) == 1) {
   }
   TFileReader* pReader = tfileGetReaderByCol(sIdx->tindex, pCache->suid, pCache->colName);
   if (pReader == NULL) {
@@ -476,11 +477,11 @@ int indexFlushCacheToTFile(SIndex* sIdx, void* cache) {
     indexError("%p immtable is empty, ignore merge opera", pCache);
     indexCacheDestroyImm(pCache);
     tfileReaderUnRef(pReader);
-    if (sIdx->quit) {
+    atomic_store_32(&pCache->merging, 0);
+    if (quit) {
       indexPost(sIdx);
     }
     indexReleaseRef(sIdx->refId);
-    atomic_store_32(&pCache->merging, 0);
     return 0;
   }
 
@@ -539,10 +540,10 @@ int indexFlushCacheToTFile(SIndex* sIdx, void* cache) {
   } else {
     indexInfo("success to merge , time cost: %" PRId64 "ms", cost / 1000);
   }
-  if (sIdx->quit) {
+  atomic_store_32(&pCache->merging, 0);
+  if (quit) {
     indexPost(sIdx);
   }
-  atomic_store_32(&pCache->merging, 0);
   indexReleaseRef(sIdx->refId);
 
   return ret;

source/libs/index/src/indexCache.c

@@ -728,9 +728,9 @@ static void doMergeWork(SSchedMsg* msg) {
   IndexCache* pCache = msg->ahandle;
   SIndex*     sidx = (SIndex*)pCache->index;
 
-  sidx->quit = msg->thandle ? true : false;
+  int quit = msg->thandle ? true : false;
   taosMemoryFree(msg->thandle);
-  indexFlushCacheToTFile(sidx, pCache);
+  indexFlushCacheToTFile(sidx, pCache, quit);
 }
 static bool indexCacheIteratorNext(Iterate* itera) {
   SSkipListIterator* iter = itera->iter;

source/libs/index/test/jsonUT.cc

@@ -51,6 +51,7 @@ class JsonEnv : public ::testing::Test {
     tIndexJsonClose(index);
     indexOptsDestroy(opts);
     printf("destory\n");
+    taosMsleep(1000);
   }
   SIndexJsonOpts* opts;
   SIndexJson*     index;