Merge branch '3.0' of https://github.com/taosdata/TDengine into 3.0

README-CN.md

@@ -52,7 +52,7 @@ TDengine 还提供一组辅助工具软件 taosTools，目前它包含 taosBench
 ### Ubuntu 18.04 及以上版本 & Debian：
 
 ```bash
-sudo apt-get install -y gcc cmake build-essential git libssl-dev libgflags2.2 libgflags-dev libgeos-dev
+sudo apt-get install -y gcc cmake build-essential git libssl-dev libgflags2.2 libgflags-dev
 ```
 
 #### 为 taos-tools 安装编译需要的软件
@@ -68,14 +68,14 @@ sudo apt install build-essential libjansson-dev libsnappy-dev liblzma-dev libz-d
 ```bash
 sudo yum install epel-release
 sudo yum update
-sudo yum install -y gcc gcc-c++ make cmake3 git openssl-devel geos geos-devel
+sudo yum install -y gcc gcc-c++ make cmake3 git openssl-devel
 sudo ln -sf /usr/bin/cmake3 /usr/bin/cmake
 ```
 
 ### CentOS 8/Fedora/Rocky Linux
 
 ```bash
-sudo dnf install -y gcc gcc-c++ make cmake epel-release git openssl-devel geos geos-devel
+sudo dnf install -y gcc gcc-c++ make cmake epel-release git openssl-devel
 ```
 
 #### 在 CentOS 上构建 taosTools 安装依赖软件
@@ -117,7 +117,7 @@ scl enable devtoolset-9 -- bash
 ### macOS
 
 ```
-brew install argp-standalone pkgconfig geos
+brew install argp-standalone pkgconfig
 ```
 
 ### 设置 golang 开发环境

README.md

@@ -60,7 +60,7 @@ To build TDengine, use [CMake](https://cmake.org/) 3.0.2 or higher versions in t
 ### Ubuntu 18.04 and above or Debian
 
 ```bash
-sudo apt-get install -y gcc cmake build-essential git libssl-dev libgflags2.2 libgflags-dev libgeos-dev
+sudo apt-get install -y gcc cmake build-essential git libssl-dev libgflags2.2 libgflags-dev
 ```
 
 #### Install build dependencies for taosTools
@@ -76,14 +76,14 @@ sudo apt install build-essential libjansson-dev libsnappy-dev liblzma-dev libz-d
 ```bash
 sudo yum install epel-release
 sudo yum update
-sudo yum install -y gcc gcc-c++ make cmake3 git openssl-devel geos geos-devel
+sudo yum install -y gcc gcc-c++ make cmake3 git openssl-devel
 sudo ln -sf /usr/bin/cmake3 /usr/bin/cmake
 ```
 
 ### CentOS 8/Fedora/Rocky Linux
 
 ```bash
-sudo dnf install -y gcc gcc-c++ make cmake epel-release git openssl-devel geos geos-devel
+sudo dnf install -y gcc gcc-c++ make cmake epel-release git openssl-devel
 ```
 
 #### Install build dependencies for taosTools on CentOS
@@ -124,7 +124,7 @@ scl enable devtoolset-9 -- bash
 ### macOS
 
 ```
-brew install argp-standalone pkgconfig geos
+brew install argp-standalone pkgconfig
 ```
 
 ### Setup golang environment

cmake/cmake.define

@@ -115,18 +115,6 @@ ELSE ()
         SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${GCC_COVERAGE_COMPILE_FLAGS} ${GCC_COVERAGE_LINK_FLAGS}")
     ENDIF ()
 
-    IF (${BUILD_SANITIZER})
-        SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS}     -Werror -Werror=return-type -fPIC -gdwarf-2 -fsanitize=address -fsanitize=undefined -fsanitize-recover=all -fsanitize=float-divide-by-zero -fsanitize=float-cast-overflow -fno-sanitize=shift-base -fno-sanitize=alignment -g3 -Wformat=0")
-        SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-literal-suffix -Werror=return-type -fPIC -gdwarf-2 -fsanitize=address -fsanitize=undefined -fsanitize-recover=all -fsanitize=float-divide-by-zero -fsanitize=float-cast-overflow -fno-sanitize=shift-base -fno-sanitize=alignment -g3 -Wformat=0")
-	      MESSAGE(STATUS "Compile with Address Sanitizer!")
-    ELSEIF (${BUILD_RELEASE})
-        SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Werror -Werror=return-type -fPIC -O3 -Wformat=2 -Wno-format-nonliteral -Wno-format-truncation -Wno-format-y2k")
-        SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Werror -Wno-reserved-user-defined-literal -Wno-literal-suffix -Werror=return-type -fPIC -O3 -Wformat=2 -Wno-format-nonliteral -Wno-format-truncation -Wno-format-y2k")
-    ELSE ()
-        SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Werror -Werror=return-type -fPIC -g3 -gdwarf-2 -Wformat=2 -Wno-format-nonliteral -Wno-format-truncation -Wno-format-y2k")
-        SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-reserved-user-defined-literal -g3 -Wno-literal-suffix -Werror=return-type -fPIC -gdwarf-2 -Wformat=2 -Wno-format-nonliteral -Wno-format-truncation -Wno-format-y2k")
-    ENDIF ()
-
     # disable all assert
     IF ((${DISABLE_ASSERT} MATCHES "true") OR (${DISABLE_ASSERTS} MATCHES "true"))
         ADD_DEFINITIONS(-DDISABLE_ASSERT)
@@ -168,4 +156,20 @@ ELSE ()
         MESSAGE(STATUS "SIMD instructions (FMA/AVX/AVX2) is ACTIVATED")
     ENDIF()
 
+    # build mode
+    SET(CMAKE_C_FLAGS_REL "${CMAKE_C_FLAGS} -Werror -Werror=return-type -fPIC -O3 -Wformat=2 -Wno-format-nonliteral -Wno-format-truncation -Wno-format-y2k")
+    SET(CMAKE_CXX_FLAGS_REL "${CMAKE_CXX_FLAGS} -Werror -Wno-reserved-user-defined-literal -Wno-literal-suffix -Werror=return-type -fPIC -O3 -Wformat=2 -Wno-format-nonliteral -Wno-format-truncation -Wno-format-y2k")
+
+    IF (${BUILD_SANITIZER})
+        SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS}     -Werror -Werror=return-type -fPIC -gdwarf-2 -fsanitize=address -fsanitize=undefined -fsanitize-recover=all -fsanitize=float-divide-by-zero -fsanitize=float-cast-overflow -fno-sanitize=shift-base -fno-sanitize=alignment -g3 -Wformat=0")
+        SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-literal-suffix -Werror=return-type -fPIC -gdwarf-2 -fsanitize=address -fsanitize=undefined -fsanitize-recover=all -fsanitize=float-divide-by-zero -fsanitize=float-cast-overflow -fno-sanitize=shift-base -fno-sanitize=alignment -g3 -Wformat=0")
+	      MESSAGE(STATUS "Compile with Address Sanitizer!")
+    ELSEIF (${BUILD_RELEASE})
+        SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS_REL}")
+        SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS_REL}")
+    ELSE ()
+        SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Werror -Werror=return-type -fPIC -g3 -gdwarf-2 -Wformat=2 -Wno-format-nonliteral -Wno-format-truncation -Wno-format-y2k")
+        SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-reserved-user-defined-literal -g3 -Wno-literal-suffix -Werror=return-type -fPIC -gdwarf-2 -Wformat=2 -Wno-format-nonliteral -Wno-format-truncation -Wno-format-y2k")
+    ENDIF ()
+
 ENDIF ()

contrib/CMakeLists.txt

@@ -229,7 +229,10 @@ endif(${BUILD_WITH_LEVELDB})
 # To support rocksdb build on ubuntu: sudo apt-get install libgflags-dev
 if(${BUILD_WITH_ROCKSDB})
     if(${TD_LINUX})
-        SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-error=maybe-uninitialized -Wno-error=unused-but-set-variable -Wno-error=unused-variable -Wno-error=unused-function -Wno-errno=unused-private-field -Wno-error=unused-result")
+        SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS_REL} -Wno-error=maybe-uninitialized -Wno-error=unused-but-set-variable -Wno-error=unused-variable -Wno-error=unused-function -Wno-errno=unused-private-field -Wno-error=unused-result")
+        IF ("${CMAKE_BUILD_TYPE}" STREQUAL "")
+            SET(CMAKE_BUILD_TYPE Release)
+        endif()
     endif(${TD_LINUX})
     MESSAGE(STATUS "CXXXX STATUS CONFIG: " ${CMAKE_CXX_FLAGS})
 
@@ -265,7 +268,7 @@ if(${BUILD_WITH_ROCKSDB})
     option(WITH_FALLOCATE "" OFF)
     option(WITH_JEMALLOC "" OFF)
     option(WITH_GFLAGS "" OFF)
-    option(PORTABLE "" OFF)
+    option(PORTABLE "" ON)
     option(WITH_LIBURING "" OFF)
     option(FAIL_ON_WARNINGS OFF)
 
@@ -485,6 +488,13 @@ endif(${BUILD_ADDR2LINE})
 
 # geos
 if(${BUILD_GEOS})
+    if(${TD_LINUX})
+        set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS_REL}")
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS_REL}")
+        IF ("${CMAKE_BUILD_TYPE}" STREQUAL "")
+            SET(CMAKE_BUILD_TYPE Release)
+        endif()
+    endif(${TD_LINUX})
     option(BUILD_SHARED_LIBS "Build GEOS with shared libraries" OFF)
     add_subdirectory(geos EXCLUDE_FROM_ALL)
     target_include_directories(

docs/en/14-reference/06-taosdump.md

@@ -79,8 +79,6 @@ Usage: taosdump [OPTION...] dbname [tbname ...]
   -e, --escape-character     Use escaped character for database name
   -N, --without-property     Dump database without its properties.
   -s, --schemaonly           Only dump table schemas.
-  -y, --answer-yes           Input yes for prompt. It will skip data file
-                             checking!
   -d, --avro-codec=snappy    Choose an avro codec among null, deflate, snappy,
                              and lzma.
   -S, --start-time=START_TIME   Start time to dump. Either epoch or

docs/en/14-reference/12-config/index.md

@@ -365,6 +365,16 @@ The charset that takes effect is UTF-8.
 | Unit          | GB                                                                                                |
 | Default Value | 2.0                                                                                               |
 
+### metaCacheMaxSize
+
+| Attribute     | Description                                                                                       |
+| ------------- | ------------------------------------------------------------------------------------------------- |
+| Applicable    | Client Only                                                                                       |
+| Meaning       | Maximum meta cache size in single client process                                                  |
+| Unit          | MB                                                                                                |
+| Default Value | -1 (No limitation)                                                                                |
+
+
 ## Cluster Parameters
 
 ### supportVnodes
@@ -433,6 +443,26 @@ The charset that takes effect is UTF-8.
 | Default Value | 0                                                                                                                                           |
 | Note          | When it's bigger than 0, the log file would be renamed to "taosdlog.xxx" in which "xxx" is the timestamp when the file is changed last time |
 
+### slowLogThreshold
+
+| Attribute     | Description                                                                                              |
+| ------------- | -------------------------------------------------------------------------------------------------------- |
+| Applicable    | Client only                                                                                              |
+| Meaning       | When an operation execution time exceeds this threshold, the operation will be logged in slow log file   |
+| Unit          | second                                                                                                   |
+| Default Value | 3                                                                                                        |
+| Note          | All slow operations will be logged in file "taosSlowLog" in the log directory                            |
+
+### slowLogScope
+
+| Attribute       | Description                                                             |
+| --------------- | ----------------------------------------------------------------------- |
+| Applicable      | Client only                                                             |
+| Meaning         | Slow log type to be logged                                              |
+| Optional Values | ALL, QUERY, INSERT, OTHERS, NONE                                        |
+| Default Value   | ALL                                                                     |
+| Note            | All slow operations will be logged by default, one option could be set  |
+
 ### debugFlag
 
 | Attribute     | Description                                               |

docs/zh/05-get-started/03-package.md

@@ -299,7 +299,7 @@ SELECT COUNT(*) FROM test.meters WHERE location = "California.SanFrancisco";
 SELECT AVG(current), MAX(voltage), MIN(phase) FROM test.meters WHERE groupId = 10;
 ```
 
-对表 `d10` 按 10 每秒进行平均值、最大值和最小值聚合统计：
+对表 `d10` 按每 10 秒进行平均值、最大值和最小值聚合统计：
 
 ```sql
 SELECT FIRST(ts), AVG(current), MAX(voltage), MIN(phase) FROM test.d10 INTERVAL(10s);

docs/zh/14-reference/06-taosdump.md

@@ -82,8 +82,6 @@ Usage: taosdump [OPTION...] dbname [tbname ...]
   -e, --escape-character     Use escaped character for database name
   -N, --without-property     Dump database without its properties.
   -s, --schemaonly           Only dump tables' schema.
-  -y, --answer-yes           Input yes for prompt. It will skip data file
-                             checking!
   -d, --avro-codec=snappy    Choose an avro codec among null, deflate, snappy,
                              and lzma.
   -S, --start-time=START_TIME   Start time to dump. Either epoch or

docs/zh/14-reference/12-config/index.md

@@ -384,6 +384,15 @@ charset 的有效值是 UTF-8。
 | 单位     | GB                                             |
 | 缺省值   | 2.0                                            |
 
+### metaCacheMaxSize
+
+| 属性     | 说明                                           |
+| -------- | ---------------------------------------------- |
+| 适用范围 | 仅客户端适用                                   |
+| 含义     | 指定单个客户端元数据缓存大小的最大值           |
+| 单位     | MB                                             |
+| 缺省值   | -1 (无限制)                                    |
+
 ## 集群相关
 
 ### supportVnodes
@@ -452,6 +461,26 @@ charset 的有效值是 UTF-8。
 | 缺省值   | 0                                                                                   |
 | 补充说明 | 大于 0 时，日志文件会被重命名为 taosdlog.xxx，其中 xxx 为日志文件最后修改的时间戳。 |
 
+### slowLogThreshold
+
+| 属性     | 说明                                                          |
+| -------- | ------------------------------------------------------------- |
+| 适用范围 | 仅客户端适用                                                  |
+| 含义     | 指定慢查询门限值，大于等于门限值认为是慢查询                  |
+| 单位     | 秒                                                            |
+| 缺省值   | 3                                                             |
+| 补充说明 | 每个客户端中所有慢查询会被记录在日志目录下的taosSlowLog文件中 |
+
+### slowLogScope
+
+| 属性     | 说明                                                          |
+| -------- | --------------------------------------------------------------|
+| 适用范围 | 仅客户端适用                                                  |
+| 含义     | 指定启动记录哪些类型的慢查询                                  |
+| 可选值   | ALL, QUERY, INSERT, OTHERS, NONE                              |
+| 缺省值   | ALL                                                           |
+| 补充说明 | 默认记录所有类型的慢查询，可通过配置只记录某一类型的慢查询    |
+
 ### debugFlag
 
 | 属性     | 说明                                                                                              |

examples/JDBC/JDBCDemo/src/main/java/com/taosdata/example/JdbcDemo.java

@@ -51,27 +51,27 @@ public class JdbcDemo {
 
     private void createDatabase() {
         String sql = "create database if not exists " + dbName;
-        exuete(sql);
+        execute(sql);
     }
 
     private void useDatabase() {
         String sql = "use " + dbName;
-        exuete(sql);
+        execute(sql);
     }
 
     private void dropTable() {
         final String sql = "drop table if exists " + dbName + "." + tbName + "";
-        exuete(sql);
+        execute(sql);
     }
 
     private void createTable() {
         final String sql = "create table if not exists " + dbName + "." + tbName + " (ts timestamp, temperature float, humidity int)";
-        exuete(sql);
+        execute(sql);
     }
 
     private void insert() {
         final String sql = "insert into " + dbName + "." + tbName + " (ts, temperature, humidity) values(now, 20.5, 34)";
-        exuete(sql);
+        execute(sql);
     }
 
     private void select() {
@@ -120,7 +120,7 @@ public class JdbcDemo {
         System.out.println("[ " + (succeed ? "OK" : "ERROR!") + " ] time cost: " + cost + " ms, execute statement ====> " + sql);
     }
 
-    private void exuete(String sql) {
+    private void execute(String sql) {
         long start = System.currentTimeMillis();
         try (Statement statement = connection.createStatement()) {
             boolean execute = statement.execute(sql);

examples/JDBC/consumer-demo/pom.xml

@@ -22,7 +22,7 @@
         <dependency>
             <groupId>com.google.guava</groupId>
             <artifactId>guava</artifactId>
-            <version>30.1.1-jre</version>
+            <version>32.0.0-jre</version>
         </dependency>
     </dependencies>
 

include/common/tglobal.h

@@ -82,6 +82,7 @@ extern int64_t tsVndCommitMaxIntervalMs;
 // mnode
 extern int64_t tsMndSdbWriteDelta;
 extern int64_t tsMndLogRetention;
+extern int8_t  tsGrant;
 extern bool    tsMndSkipGrant;
 
 // monitor
@@ -198,6 +199,7 @@ void    taosSetAllDebugFlag(int32_t flag, bool rewrite);
 void    taosSetDebugFlag(int32_t *pFlagPtr, const char *flagName, int32_t flagVal, bool rewrite);
 int32_t taosApplyLocalCfg(SConfig *pCfg, char *name);
 void    taosLocalCfgForbiddenToChange(char *name, bool *forbidden);
+int8_t  taosGranted();
 
 #ifdef __cplusplus
 }

include/common/tmsg.h

@@ -1975,6 +1975,7 @@ typedef struct {
   SArray*  fillNullCols;  // array of SColLocation
   int64_t  deleteMark;
   int8_t   igUpdate;
+  int64_t  lastTs;
 } SCMCreateStreamReq;
 
 typedef struct {
@@ -2035,7 +2036,6 @@ typedef struct {
   SArray* topicNames;  // SArray<char**>
 
   int8_t         withTbName;
-  int8_t         useSnapshot;
   int8_t         autoCommit;
   int32_t        autoCommitInterval;
   int8_t         resetOffsetCfg;
@@ -2055,7 +2055,6 @@ static FORCE_INLINE int32_t tSerializeSCMSubscribeReq(void** buf, const SCMSubsc
   }
 
   tlen += taosEncodeFixedI8(buf, pReq->withTbName);
-  tlen += taosEncodeFixedI8(buf, pReq->useSnapshot);
   tlen += taosEncodeFixedI8(buf, pReq->autoCommit);
   tlen += taosEncodeFixedI32(buf, pReq->autoCommitInterval);
   tlen += taosEncodeFixedI8(buf, pReq->resetOffsetCfg);
@@ -2079,7 +2078,6 @@ static FORCE_INLINE void* tDeserializeSCMSubscribeReq(void* buf, SCMSubscribeReq
   }
 
   buf = taosDecodeFixedI8(buf, &pReq->withTbName);
-  buf = taosDecodeFixedI8(buf, &pReq->useSnapshot);
   buf = taosDecodeFixedI8(buf, &pReq->autoCommit);
   buf = taosDecodeFixedI32(buf, &pReq->autoCommitInterval);
   buf = taosDecodeFixedI8(buf, &pReq->resetOffsetCfg);
@@ -2497,6 +2495,7 @@ typedef struct {
   int64_t  stime;  // timestamp precision ms
   int64_t  reqRid;
   bool     stableQuery;
+  bool     isSubQuery;
   char     fqdn[TSDB_FQDN_LEN];
   int32_t  subPlanNum;
   SArray*  subDesc;  // SArray<SQuerySubDesc>

include/common/tmsgdef.h

@@ -310,6 +310,7 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_VND_TMQ_ADD_CHECKINFO, "vnode-tmq-add-checkinfo", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_TMQ_DEL_CHECKINFO, "vnode-del-checkinfo", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_TMQ_CONSUME, "vnode-tmq-consume", SMqPollReq, SMqDataBlkRsp)
+  TD_DEF_MSG_TYPE(TDMT_VND_TMQ_CONSUME_PUSH, "vnode-tmq-consume-push", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_TMQ_VG_WALINFO, "vnode-tmq-vg-walinfo", SMqPollReq, SMqDataBlkRsp)
   TD_DEF_MSG_TYPE(TDMT_VND_TMQ_MAX_MSG, "vnd-tmq-max", NULL, NULL)
 

include/libs/catalog/catalog.h

@@ -87,6 +87,7 @@ typedef struct SCatalogReq {
   bool    dNodeRequired;  // valid dnode
   bool    svrVerRequired;
   bool    forceUpdate;
+  bool    cloned;
 } SCatalogReq;
 
 typedef struct SMetaRes {

include/libs/function/functionMgt.h

@@ -233,6 +233,7 @@ bool fmIsGroupKeyFunc(int32_t funcId);
 bool fmIsBlockDistFunc(int32_t funcId);
 
 void getLastCacheDataType(SDataType* pType);
+SFunctionNode* createFunction(const char* pName, SNodeList* pParameterList);
 
 int32_t fmGetDistMethod(const SFunctionNode* pFunc, SFunctionNode** pPartialFunc, SFunctionNode** pMergeFunc);
 

include/libs/nodes/cmdnodes.h

@@ -432,9 +432,11 @@ typedef struct SCreateStreamStmt {
   bool                ignoreExists;
   SStreamOptions*     pOptions;
   SNode*              pQuery;
+  SNode*              pPrevQuery;
   SNodeList*          pTags;
   SNode*              pSubtable;
   SNodeList*          pCols;
+  SCMCreateStreamReq* pReq;
 } SCreateStreamStmt;
 
 typedef struct SDropStreamStmt {

include/libs/nodes/plannodes.h

@@ -617,6 +617,7 @@ typedef struct SQueryPlan {
   int32_t      numOfSubplans;
   SNodeList*   pSubplans;  // Element is SNodeListNode. The execution level of subplan, starting from 0.
   SExplainInfo explainInfo;
+  void*        pPostPlan;
 } SQueryPlan;
 
 const char* dataOrderStr(EDataOrderLevel order);

include/libs/nodes/querynodes.h

@@ -441,7 +441,9 @@ typedef struct SQuery {
   EQueryExecStage execStage;
   EQueryExecMode  execMode;
   bool            haveResultSet;
+  SNode*          pPrevRoot;
   SNode*          pRoot;
+  SNode*          pPostRoot;
   int32_t         numOfResCols;
   SSchema*        pResSchema;
   int8_t          precision;

include/libs/parser/parser.h

@@ -74,6 +74,7 @@ int32_t qAnalyseSqlSemantic(SParseContext* pCxt, const struct SCatalogReq* pCata
                             const struct SMetaData* pMetaData, SQuery* pQuery);
 int32_t qContinueParseSql(SParseContext* pCxt, struct SCatalogReq* pCatalogReq, const struct SMetaData* pMetaData,
                           SQuery* pQuery);
+int32_t qContinueParsePostQuery(SParseContext* pCxt, SQuery* pQuery, void** pResRow);
 
 void qDestroyParseContext(SParseContext* pCxt);
 

include/libs/planner/planner.h

@@ -52,6 +52,7 @@ int32_t qCreateQueryPlan(SPlanContext* pCxt, SQueryPlan** pPlan, SArray* pExecNo
 // @groupId id of a group of datasource subplans of this @pSubplan
 // @pSource one execution location of this group of datasource subplans
 int32_t qSetSubplanExecutionNode(SSubplan* pSubplan, int32_t groupId, SDownstreamSourceNode* pSource);
+int32_t qContinuePlanPostQuery(void *pPostPlan);
 
 void qClearSubplanExecutionNode(SSubplan* pSubplan);
 

include/libs/wal/wal.h

@@ -214,7 +214,7 @@ int32_t walSkipFetchBody(SWalReader *pRead, const SWalCkHead *pHead);
 
 void walRefFirstVer(SWal *, SWalRef *);
 void walRefLastVer(SWal *, SWalRef *);
-SWalRef *walRefCommittedVer(SWal *);
+void walRefCommitVer(SWal *, SWalRef *);
 
 SWalRef *walOpenRef(SWal *);
 void     walCloseRef(SWal *pWal, int64_t refId);

include/util/talgo.h

@@ -31,7 +31,7 @@ typedef void *(*__array_item_dup_fn_t)(void *);
 
 typedef void (*FDelete)(void *);
 typedef int32_t (*FEncode)(void **buf, const void *dst);
-typedef void *(*FDecode)(const void *buf, void *dst);
+typedef void *(*FDecode)(const void *buf, void *dst, int8_t sver);
 
 #define TD_EQ 0x1
 #define TD_GT 0x2

include/util/tarray.h

@@ -244,7 +244,7 @@ int32_t taosArraySearchIdx(const SArray* pArray, const void* key, __compar_fn_t
 void taosArraySortPWithExt(SArray* pArray, __ext_compar_fn_t fn, const void* param);
 
 int32_t taosEncodeArray(void** buf, const SArray* pArray, FEncode encode);
-void*   taosDecodeArray(const void* buf, SArray** pArray, FDecode decode, int32_t dataSz);
+void*   taosDecodeArray(const void* buf, SArray** pArray, FDecode decode, int32_t dataSz, int8_t sver);
 
 #ifdef __cplusplus
 }

packaging/checkPackageRuning.py

@@ -42,8 +42,8 @@ else:
 # os.system("rm -rf /var/lib/taos/*")
 # os.system("systemctl restart taosd ")
 
-# wait a moment ,at least 5 seconds
+# wait a moment ,at least 10 seconds
-time.sleep(5)
+time.sleep(10)
 
 # prepare data by taosBenchmark 
 

packaging/deb/makedeb.sh

@@ -124,12 +124,12 @@ if [ -f ${compile_dir}/build/bin/jemalloc-config ]; then
         cp ${compile_dir}/build/lib/libjemalloc.so.2 ${pkg_dir}${install_user_local_path}/lib/
         ln -sf libjemalloc.so.2 ${pkg_dir}${install_user_local_path}/lib/libjemalloc.so
     fi
-    if [ -f ${compile_dir}/build/lib/libjemalloc.a ]; then
+    # if [ -f ${compile_dir}/build/lib/libjemalloc.a ]; then
-        cp ${compile_dir}/build/lib/libjemalloc.a ${pkg_dir}${install_user_local_path}/lib/
+    #     cp ${compile_dir}/build/lib/libjemalloc.a ${pkg_dir}${install_user_local_path}/lib/
-    fi
+    # fi
-    if [ -f ${compile_dir}/build/lib/libjemalloc_pic.a ]; then
+    # if [ -f ${compile_dir}/build/lib/libjemalloc_pic.a ]; then
-        cp ${compile_dir}/build/lib/libjemalloc_pic.a ${pkg_dir}${install_user_local_path}/lib/
+    #     cp ${compile_dir}/build/lib/libjemalloc_pic.a ${pkg_dir}${install_user_local_path}/lib/
-    fi
+    # fi
     if [ -f ${compile_dir}/build/lib/pkgconfig/jemalloc.pc ]; then
         cp ${compile_dir}/build/lib/pkgconfig/jemalloc.pc ${pkg_dir}${install_user_local_path}/lib/pkgconfig/
     fi

packaging/rpm/tdengine.spec

@@ -123,12 +123,12 @@ if [ -f %{_compiledir}/build/bin/jemalloc-config ]; then
         cp %{_compiledir}/build/lib/libjemalloc.so.2 %{buildroot}%{homepath}/jemalloc/lib
         ln -sf libjemalloc.so.2 %{buildroot}%{homepath}/jemalloc/lib/libjemalloc.so
     fi
-    if [ -f %{_compiledir}/build/lib/libjemalloc.a ]; then
+#    if [ -f %{_compiledir}/build/lib/libjemalloc.a ]; then
-        cp %{_compiledir}/build/lib/libjemalloc.a %{buildroot}%{homepath}/jemalloc/lib
+#        cp %{_compiledir}/build/lib/libjemalloc.a %{buildroot}%{homepath}/jemalloc/lib
-    fi
+#    fi
-    if [ -f %{_compiledir}/build/lib/libjemalloc_pic.a ]; then
+#    if [ -f %{_compiledir}/build/lib/libjemalloc_pic.a ]; then
-        cp %{_compiledir}/build/lib/libjemalloc_pic.a %{buildroot}%{homepath}/jemalloc/lib
+#        cp %{_compiledir}/build/lib/libjemalloc_pic.a %{buildroot}%{homepath}/jemalloc/lib
-    fi
+#    fi
     if [ -f %{_compiledir}/build/lib/pkgconfig/jemalloc.pc ]; then
         cp %{_compiledir}/build/lib/pkgconfig/jemalloc.pc %{buildroot}%{homepath}/jemalloc/lib/pkgconfig
     fi

packaging/tools/install.sh

@@ -315,13 +315,13 @@ function install_jemalloc() {
       ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc.so.2 /usr/local/lib
       ${csudo}ln -sf libjemalloc.so.2 /usr/local/lib/libjemalloc.so
       ${csudo}/usr/bin/install -c -d /usr/local/lib
-      if [ -f ${jemalloc_dir}/lib/libjemalloc.a ]; then
+      # if [ -f ${jemalloc_dir}/lib/libjemalloc.a ]; then
-        ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc.a /usr/local/lib
+      #   ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc.a /usr/local/lib
-      fi
+      # fi
-      if [ -f ${jemalloc_dir}/lib/libjemalloc_pic.a ]; then
+      # if [ -f ${jemalloc_dir}/lib/libjemalloc_pic.a ]; then
-        ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc_pic.a /usr/local/lib
+      #   ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc_pic.a /usr/local/lib
-      fi
+      # fi
-      if [ -f ${jemalloc_dir}/lib/libjemalloc_pic.a ]; then
+      if [ -f ${jemalloc_dir}/lib/pkgconfig/jemalloc.pc ]; then
         ${csudo}/usr/bin/install -c -d /usr/local/lib/pkgconfig
         ${csudo}/usr/bin/install -c -m 644 ${jemalloc_dir}/lib/pkgconfig/jemalloc.pc /usr/local/lib/pkgconfig
       fi

packaging/tools/install_client.sh

@@ -214,13 +214,13 @@ function install_jemalloc() {
             ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc.so.2 /usr/local/lib
             ${csudo}ln -sf libjemalloc.so.2 /usr/local/lib/libjemalloc.so
             ${csudo}/usr/bin/install -c -d /usr/local/lib
-            if [ -f ${jemalloc_dir}/lib/libjemalloc.a ]; then
+            # if [ -f ${jemalloc_dir}/lib/libjemalloc.a ]; then
-                ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc.a /usr/local/lib
+            #     ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc.a /usr/local/lib
-            fi
+            # fi
-            if [ -f ${jemalloc_dir}/lib/libjemalloc_pic.a ]; then
+            # if [ -f ${jemalloc_dir}/lib/libjemalloc_pic.a ]; then
-                ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc_pic.a /usr/local/lib
+            #     ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc_pic.a /usr/local/lib
-            fi
+            # fi
-            if [ -f ${jemalloc_dir}/lib/libjemalloc_pic.a ]; then
+            if [ -f ${jemalloc_dir}/lib/pkgconfig/jemalloc.pc ]; then
                 ${csudo}/usr/bin/install -c -d /usr/local/lib/pkgconfig
                 ${csudo}/usr/bin/install -c -m 644 ${jemalloc_dir}/lib/pkgconfig/jemalloc.pc /usr/local/lib/pkgconfig
             fi

packaging/tools/make_install.sh

@@ -241,10 +241,10 @@ function install_jemalloc() {
       ${csudo}/usr/bin/install -c -m 755 ${binary_dir}/build/lib/libjemalloc.so.2 /usr/local/lib
       ${csudo}ln -sf libjemalloc.so.2 /usr/local/lib/libjemalloc.so > /dev/null 2>&1
       ${csudo}/usr/bin/install -c -d /usr/local/lib
-      [ -f ${binary_dir}/build/lib/libjemalloc.a ] &&
+      # [ -f ${binary_dir}/build/lib/libjemalloc.a ] &&
-        ${csudo}/usr/bin/install -c -m 755 ${binary_dir}/build/lib/libjemalloc.a /usr/local/lib
+      #   ${csudo}/usr/bin/install -c -m 755 ${binary_dir}/build/lib/libjemalloc.a /usr/local/lib
-      [ -f ${binary_dir}/build/lib/libjemalloc_pic.a ] &&
+      # [ -f ${binary_dir}/build/lib/libjemalloc_pic.a ] &&
-        ${csudo}/usr/bin/install -c -m 755 ${binary_dir}/build/lib/libjemalloc_pic.a /usr/local/lib
+      #   ${csudo}/usr/bin/install -c -m 755 ${binary_dir}/build/lib/libjemalloc_pic.a /usr/local/lib
       if [ -f "${binary_dir}/build/lib/pkgconfig/jemalloc.pc" ]; then
         ${csudo}/usr/bin/install -c -d /usr/local/lib/pkgconfig
         ${csudo}/usr/bin/install -c -m 644 ${binary_dir}/build/lib/pkgconfig/jemalloc.pc \

packaging/tools/makeclient.sh

@@ -118,12 +118,12 @@ if [ -f ${build_dir}/bin/jemalloc-config ]; then
     cp ${build_dir}/lib/libjemalloc.so.2 ${install_dir}/jemalloc/lib
     ln -sf libjemalloc.so.2 ${install_dir}/jemalloc/lib/libjemalloc.so
   fi
-  if [ -f ${build_dir}/lib/libjemalloc.a ]; then
+  # if [ -f ${build_dir}/lib/libjemalloc.a ]; then
-    cp ${build_dir}/lib/libjemalloc.a ${install_dir}/jemalloc/lib
+  #   cp ${build_dir}/lib/libjemalloc.a ${install_dir}/jemalloc/lib
-  fi
+  # fi
-  if [ -f ${build_dir}/lib/libjemalloc_pic.a ]; then
+  # if [ -f ${build_dir}/lib/libjemalloc_pic.a ]; then
-    cp ${build_dir}/lib/libjemalloc_pic.a ${install_dir}/jemalloc/lib
+  #   cp ${build_dir}/lib/libjemalloc_pic.a ${install_dir}/jemalloc/lib
-  fi
+  # fi
   if [ -f ${build_dir}/lib/pkgconfig/jemalloc.pc ]; then
     cp ${build_dir}/lib/pkgconfig/jemalloc.pc ${install_dir}/jemalloc/lib/pkgconfig
   fi

packaging/tools/makepkg.sh

@@ -217,12 +217,12 @@ if [ -f ${build_dir}/bin/jemalloc-config ]; then
     cp ${build_dir}/lib/libjemalloc.so.2 ${install_dir}/jemalloc/lib
     ln -sf libjemalloc.so.2 ${install_dir}/jemalloc/lib/libjemalloc.so
   fi
-  if [ -f ${build_dir}/lib/libjemalloc.a ]; then
+  # if [ -f ${build_dir}/lib/libjemalloc.a ]; then
-    cp ${build_dir}/lib/libjemalloc.a ${install_dir}/jemalloc/lib
+  #   cp ${build_dir}/lib/libjemalloc.a ${install_dir}/jemalloc/lib
-  fi
+  # fi
-  if [ -f ${build_dir}/lib/libjemalloc_pic.a ]; then
+  # if [ -f ${build_dir}/lib/libjemalloc_pic.a ]; then
-    cp ${build_dir}/lib/libjemalloc_pic.a ${install_dir}/jemalloc/lib
+  #   cp ${build_dir}/lib/libjemalloc_pic.a ${install_dir}/jemalloc/lib
-  fi
+  # fi
   if [ -f ${build_dir}/lib/pkgconfig/jemalloc.pc ]; then
     cp ${build_dir}/lib/pkgconfig/jemalloc.pc ${install_dir}/jemalloc/lib/pkgconfig
   fi

packaging/tools/post.sh

@@ -169,13 +169,13 @@ function install_jemalloc() {
       ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc.so.2 /usr/local/lib
       ${csudo}ln -sf libjemalloc.so.2 /usr/local/lib/libjemalloc.so
       ${csudo}/usr/bin/install -c -d /usr/local/lib
-      if [ -f ${jemalloc_dir}/lib/libjemalloc.a ]; then
+      # if [ -f ${jemalloc_dir}/lib/libjemalloc.a ]; then
-        ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc.a /usr/local/lib
+      #   ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc.a /usr/local/lib
-      fi
+      # fi
-      if [ -f ${jemalloc_dir}/lib/libjemalloc_pic.a ]; then
+      # if [ -f ${jemalloc_dir}/lib/libjemalloc_pic.a ]; then
-        ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc_pic.a /usr/local/lib
+      #   ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc_pic.a /usr/local/lib
-      fi
+      # fi
-      if [ -f ${jemalloc_dir}/lib/libjemalloc_pic.a ]; then
+      if [ -f ${jemalloc_dir}/lib/pkgconfig/jemalloc.pc ]; then
         ${csudo}/usr/bin/install -c -d /usr/local/lib/pkgconfig
         ${csudo}/usr/bin/install -c -m 644 ${jemalloc_dir}/lib/pkgconfig/jemalloc.pc /usr/local/lib/pkgconfig
       fi

source/client/inc/clientInt.h

@@ -227,6 +227,12 @@ typedef struct {
   STaosxRsp      rsp;
 } SMqTaosxRspObj;
 
+typedef struct SReqRelInfo {
+  uint64_t userRefId;
+  uint64_t prevRefId;
+  uint64_t nextRefId;
+} SReqRelInfo;
+
 typedef struct SRequestObj {
   int8_t               resType;  // query or tmq
   uint64_t             requestId;
@@ -250,10 +256,14 @@ typedef struct SRequestObj {
   bool                 validateOnly;  // todo refactor
   bool                 killed;
   bool                 inRetry;
+  bool                 isSubReq;
   uint32_t             prevCode;  // previous error code: todo refactor, add update flag for catalog
   uint32_t             retry;
   int64_t              allocatorRefId;
   SQuery*              pQuery;
+  void*                pPostPlan;
+  SReqRelInfo          relation;
+  void*                pWrapper;
 } SRequestObj;
 
 typedef struct SSyncQueryParam {
@@ -279,6 +289,7 @@ TAOS_RES* taosQueryImplWithReqid(TAOS* taos, const char* sql, bool validateOnly,
 void taosAsyncQueryImpl(uint64_t connId, const char* sql, __taos_async_fn_t fp, void* param, bool validateOnly);
 void taosAsyncQueryImplWithReqid(uint64_t connId, const char* sql, __taos_async_fn_t fp, void* param, bool validateOnly,
                                  int64_t reqid);
+void taosAsyncFetchImpl(SRequestObj *pRequest, __taos_async_fn_t fp, void *param);
 
 int32_t getVersion1BlockMetaSize(const char* p, int32_t numOfCols);
 
@@ -368,6 +379,7 @@ typedef struct SSqlCallbackWrapper {
   SParseContext* pParseCtx;
   SCatalogReq*   pCatalogReq;
   SRequestObj*   pRequest;
+  void*          pPlanInfo;
 } SSqlCallbackWrapper;
 
 SRequestObj* launchQueryImpl(SRequestObj* pRequest, SQuery* pQuery, bool keepQuery, void** res);
@@ -382,6 +394,12 @@ int32_t handleCreateTbExecRes(void* res, SCatalog* pCatalog);
 bool    qnodeRequired(SRequestObj* pRequest);
 void    continueInsertFromCsv(SSqlCallbackWrapper* pWrapper, SRequestObj* pRequest);
 void    destorySqlCallbackWrapper(SSqlCallbackWrapper* pWrapper);
+void    handleQueryAnslyseRes(SSqlCallbackWrapper *pWrapper, SMetaData *pResultMeta, int32_t code);
+void    restartAsyncQuery(SRequestObj *pRequest, int32_t code);
+int32_t buildPreviousRequest(SRequestObj *pRequest, const char* sql, SRequestObj** pNewRequest);
+int32_t prepareAndParseSqlSyntax(SSqlCallbackWrapper **ppWrapper, SRequestObj *pRequest, bool updateMetaForce);
+void    returnToUser(SRequestObj* pRequest);
+void    stopAllQueries(SRequestObj *pRequest);
 
 #ifdef __cplusplus
 }

source/client/src/clientEnv.c

@@ -358,6 +358,49 @@ int32_t releaseRequest(int64_t rid) { return taosReleaseRef(clientReqRefPool, ri
 
 int32_t removeRequest(int64_t rid) { return taosRemoveRef(clientReqRefPool, rid); }
 
+
+void destroySubRequests(SRequestObj *pRequest) {
+  int32_t reqIdx = -1;
+  SRequestObj *pReqList[16] = {NULL};
+  uint64_t tmpRefId = 0;
+
+  if (pRequest->relation.userRefId && pRequest->relation.userRefId != pRequest->self) {
+    return;
+  }
+  
+  SRequestObj* pTmp = pRequest;
+  while (pTmp->relation.prevRefId) {
+    tmpRefId = pTmp->relation.prevRefId;
+    pTmp = acquireRequest(tmpRefId);
+    if (pTmp) {
+      pReqList[++reqIdx] = pTmp;
+      releaseRequest(tmpRefId);
+    } else {
+      tscError("0x%" PRIx64 ", prev req ref 0x%" PRIx64 " is not there, reqId:0x%" PRIx64, pTmp->self,
+               tmpRefId, pTmp->requestId);
+      break;         
+    }
+  }
+
+  for (int32_t i = reqIdx; i >= 0; i--) {
+    removeRequest(pReqList[i]->self);
+  }
+
+  tmpRefId = pRequest->relation.nextRefId;
+  while (tmpRefId) {
+    pTmp = acquireRequest(tmpRefId);
+    if (pTmp) {
+      tmpRefId = pTmp->relation.nextRefId;
+      removeRequest(pTmp->self);      
+      releaseRequest(pTmp->self);
+    } else {
+      tscError("0x%" PRIx64 " is not there", tmpRefId);
+      break;         
+    }
+  }
+}
+
+
 void doDestroyRequest(void *p) {
   if (NULL == p) {
     return;
@@ -368,10 +411,14 @@ void doDestroyRequest(void *p) {
   uint64_t reqId = pRequest->requestId;
   tscTrace("begin to destroy request %" PRIx64 " p:%p", reqId, pRequest);
 
+  destroySubRequests(pRequest);
+  
   taosHashRemove(pRequest->pTscObj->pRequests, &pRequest->self, sizeof(pRequest->self));
 
   schedulerFreeJob(&pRequest->body.queryJob, 0);
 
+  destorySqlCallbackWrapper(pRequest->pWrapper);
+
   taosMemoryFreeClear(pRequest->msgBuf);
   taosMemoryFreeClear(pRequest->pDb);
 
@@ -412,6 +459,63 @@ void destroyRequest(SRequestObj *pRequest) {
   removeRequest(pRequest->self);
 }
 
+void taosStopQueryImpl(SRequestObj *pRequest) {
+  pRequest->killed = true;
+
+  // It is not a query, no need to stop.
+  if (NULL == pRequest->pQuery || QUERY_EXEC_MODE_SCHEDULE != pRequest->pQuery->execMode) {
+    tscDebug("request 0x%" PRIx64 " no need to be killed since not query", pRequest->requestId);
+    return;
+  }
+
+  schedulerFreeJob(&pRequest->body.queryJob, TSDB_CODE_TSC_QUERY_KILLED);
+  tscDebug("request %" PRIx64 " killed", pRequest->requestId);
+}
+
+void stopAllQueries(SRequestObj *pRequest) {
+  int32_t reqIdx = -1;
+  SRequestObj *pReqList[16] = {NULL};
+  uint64_t tmpRefId = 0;
+
+  if (pRequest->relation.userRefId && pRequest->relation.userRefId != pRequest->self) {
+    return;
+  }
+  
+  SRequestObj* pTmp = pRequest;
+  while (pTmp->relation.prevRefId) {
+    tmpRefId = pTmp->relation.prevRefId;
+    pTmp = acquireRequest(tmpRefId);
+    if (pTmp) {
+      pReqList[++reqIdx] = pTmp;
+      releaseRequest(tmpRefId);
+    } else {
+      tscError("0x%" PRIx64 ", prev req ref 0x%" PRIx64 " is not there, reqId:0x%" PRIx64, pTmp->self,
+               tmpRefId, pTmp->requestId);
+      break;         
+    }
+  }
+
+  for (int32_t i = reqIdx; i >= 0; i--) {
+    taosStopQueryImpl(pReqList[i]);
+  }
+
+  taosStopQueryImpl(pRequest);
+
+  tmpRefId = pRequest->relation.nextRefId;
+  while (tmpRefId) {
+    pTmp = acquireRequest(tmpRefId);
+    if (pTmp) {
+      tmpRefId = pTmp->relation.nextRefId;
+      taosStopQueryImpl(pTmp);
+      releaseRequest(pTmp->self);
+    } else {
+      tscError("0x%" PRIx64 " is not there", tmpRefId);
+      break;         
+    }
+  }
+}
+
+
 void crashReportThreadFuncUnexpectedStopped(void) { atomic_store_32(&clientStop, -1); }
 
 static void *tscCrashReportThreadFp(void *param) {

source/client/src/clientHb.c

@@ -464,6 +464,7 @@ int32_t hbBuildQueryDesc(SQueryHbReqBasic *hbBasic, STscObj *pObj) {
     desc.useconds = now - pRequest->metric.start;
     desc.reqRid = pRequest->self;
     desc.stableQuery = pRequest->stableQuery;
+    desc.isSubQuery = pRequest->isSubReq;
     taosGetFqdn(desc.fqdn);
     desc.subPlanNum = pRequest->body.subplanNum;
 

source/client/src/clientImpl.c

@@ -237,6 +237,17 @@ int32_t buildRequest(uint64_t connId, const char* sql, int sqlLen, void* param,
   return TSDB_CODE_SUCCESS;
 }
 
+int32_t buildPreviousRequest(SRequestObj *pRequest, const char* sql, SRequestObj** pNewRequest) {
+  int32_t code = buildRequest(pRequest->pTscObj->id, sql, strlen(sql), pRequest, pRequest->validateOnly, pNewRequest, 0);
+  if (TSDB_CODE_SUCCESS == code) {
+    pRequest->relation.prevRefId = (*pNewRequest)->self;
+    (*pNewRequest)->relation.nextRefId = pRequest->self;
+    (*pNewRequest)->relation.userRefId = pRequest->self;
+    (*pNewRequest)->isSubReq = true;
+  }
+  return code;
+}
+
 int32_t parseSql(SRequestObj* pRequest, bool topicQuery, SQuery** pQuery, SStmtCallback* pStmtCb) {
   STscObj* pTscObj = pRequest->pTscObj;
 
@@ -878,6 +889,81 @@ static bool incompletaFileParsing(SNode* pStmt) {
   return QUERY_NODE_VNODE_MODIFY_STMT != nodeType(pStmt) ? false : ((SVnodeModifyOpStmt*)pStmt)->fileProcessing;
 }
 
+void continuePostSubQuery(SRequestObj* pRequest, TAOS_ROW row) {
+  SSqlCallbackWrapper* pWrapper = pRequest->pWrapper;
+  int32_t code = nodesAcquireAllocator(pWrapper->pParseCtx->allocatorId);
+  if (TSDB_CODE_SUCCESS == code) {
+    int64_t analyseStart = taosGetTimestampUs();
+    code = qContinueParsePostQuery(pWrapper->pParseCtx, pRequest->pQuery, (void**)row);
+    pRequest->metric.analyseCostUs += taosGetTimestampUs() - analyseStart;
+  }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = qContinuePlanPostQuery(pRequest->pPostPlan);
+  }
+  nodesReleaseAllocator(pWrapper->pParseCtx->allocatorId);
+
+  handleQueryAnslyseRes(pWrapper, NULL, code);
+}
+
+void returnToUser(SRequestObj* pRequest) {
+  if (pRequest->relation.userRefId == pRequest->self || 0 == pRequest->relation.userRefId) {
+    // return to client
+    pRequest->body.queryFp(pRequest->body.param, pRequest, pRequest->code);
+    return;
+  }
+
+  SRequestObj* pUserReq = acquireRequest(pRequest->relation.userRefId);
+  if (pUserReq) {
+    pUserReq->code = pRequest->code;
+    // return to client
+    pUserReq->body.queryFp(pUserReq->body.param, pUserReq, pUserReq->code);
+    releaseRequest(pRequest->relation.userRefId);
+    return;
+  } else {
+    tscError("0x%" PRIx64 ", user ref 0x%" PRIx64 " is not there, reqId:0x%" PRIx64, pRequest->self,
+             pRequest->relation.userRefId, pRequest->requestId);
+  }
+}
+
+void postSubQueryFetchCb(void* param, TAOS_RES* res, int32_t rowNum) {
+  SRequestObj* pRequest = (SRequestObj*)res;
+  if (pRequest->code) {
+    returnToUser(pRequest);
+    return;
+  }
+
+  TAOS_ROW row = NULL;
+  if (rowNum > 0) {
+    row = taos_fetch_row(res); // for single row only now
+  }
+
+  SRequestObj* pNextReq = acquireRequest(pRequest->relation.nextRefId);
+  if (pNextReq) {
+    continuePostSubQuery(pNextReq, row);
+    releaseRequest(pRequest->relation.nextRefId);
+  } else {
+    tscError("0x%" PRIx64 ", next req ref 0x%" PRIx64 " is not there, reqId:0x%" PRIx64, pRequest->self,
+             pRequest->relation.nextRefId, pRequest->requestId);
+  }
+}
+
+void handlePostSubQuery(SSqlCallbackWrapper* pWrapper) {
+  SRequestObj* pRequest = pWrapper->pRequest;
+  if (TD_RES_QUERY(pRequest)) {
+    taosAsyncFetchImpl(pRequest, postSubQueryFetchCb, pWrapper);
+    return;
+  }
+
+  SRequestObj* pNextReq = acquireRequest(pRequest->relation.nextRefId);
+  if (pNextReq) {
+    continuePostSubQuery(pNextReq, NULL);
+    releaseRequest(pRequest->relation.nextRefId);
+  } else {
+    tscError("0x%" PRIx64 ", next req ref 0x%" PRIx64 " is not there, reqId:0x%" PRIx64, pRequest->self,
+             pRequest->relation.nextRefId, pRequest->requestId);
+  }
+}
+
 // todo refacto the error code  mgmt
 void schedulerExecCb(SExecResult* pResult, void* param, int32_t code) {
   SSqlCallbackWrapper* pWrapper = param;
@@ -912,12 +998,7 @@ void schedulerExecCb(SExecResult* pResult, void* param, int32_t code) {
   if (code != TSDB_CODE_SUCCESS && NEED_CLIENT_HANDLE_ERROR(code) && pRequest->sqlstr != NULL) {
     tscDebug("0x%" PRIx64 " client retry to handle the error, code:%s, tryCount:%d, reqId:0x%" PRIx64, pRequest->self,
              tstrerror(code), pRequest->retry, pRequest->requestId);
-    pRequest->prevCode = code;
+    restartAsyncQuery(pRequest, code);
-    schedulerFreeJob(&pRequest->body.queryJob, 0);
-    qDestroyQuery(pRequest->pQuery);
-    pRequest->pQuery = NULL;
-    destorySqlCallbackWrapper(pWrapper);
-    doAsyncQuery(pRequest, true);
     return;
   }
 
@@ -938,11 +1019,16 @@ void schedulerExecCb(SExecResult* pResult, void* param, int32_t code) {
     return;
   }
 
+  if (pRequest->relation.nextRefId) {
+    handlePostSubQuery(pWrapper);
+  } else {
     destorySqlCallbackWrapper(pWrapper);
+    pRequest->pWrapper = NULL;
 
     // return to client
     pRequest->body.queryFp(pRequest->body.param, pRequest, code);
   }
+}
 
 SRequestObj* launchQueryImpl(SRequestObj* pRequest, SQuery* pQuery, bool keepQuery, void** res) {
   int32_t code = 0;
@@ -1049,6 +1135,7 @@ static int32_t asyncExecSchQuery(SRequestObj* pRequest, SQuery* pQuery, SMetaDat
              pRequest->requestId);
   } else {
     pRequest->body.subplanNum = pDag->numOfSubplans;
+    TSWAP(pRequest->pPostPlan, pDag->pPostPlan);
   }
 
   pRequest->metric.execStart = taosGetTimestampUs();
@@ -1084,6 +1171,7 @@ static int32_t asyncExecSchQuery(SRequestObj* pRequest, SQuery* pQuery, SMetaDat
     tscDebug("0x%" PRIx64 " plan not executed, code:%s 0x%" PRIx64, pRequest->self, tstrerror(code),
              pRequest->requestId);
     destorySqlCallbackWrapper(pWrapper);
+    pRequest->pWrapper = NULL;
     if (TSDB_CODE_SUCCESS != code) {
       pRequest->code = terrno;
     }
@@ -1103,6 +1191,7 @@ void launchAsyncQuery(SRequestObj* pRequest, SQuery* pQuery, SMetaData* pResultM
   pRequest->body.execMode = pQuery->execMode;
   if (QUERY_EXEC_MODE_SCHEDULE != pRequest->body.execMode) {
     destorySqlCallbackWrapper(pWrapper);
+    pRequest->pWrapper = NULL;
   }
 
   if (pQuery->pRoot && !pRequest->inRetry) {
@@ -2402,3 +2491,90 @@ TAOS_RES* taosQueryImplWithReqid(TAOS* taos, const char* sql, bool validateOnly,
 
   return pRequest;
 }
+
+
+static void fetchCallback(void *pResult, void *param, int32_t code) {
+  SRequestObj *pRequest = (SRequestObj *)param;
+
+  SReqResultInfo *pResultInfo = &pRequest->body.resInfo;
+
+  tscDebug("0x%" PRIx64 " enter scheduler fetch cb, code:%d - %s, reqId:0x%" PRIx64, pRequest->self, code,
+           tstrerror(code), pRequest->requestId);
+
+  pResultInfo->pData = pResult;
+  pResultInfo->numOfRows = 0;
+
+  if (code != TSDB_CODE_SUCCESS) {
+    pRequest->code = code;
+    taosMemoryFreeClear(pResultInfo->pData);
+    pRequest->body.fetchFp(pRequest->body.param, pRequest, 0);
+    return;
+  }
+
+  if (pRequest->code != TSDB_CODE_SUCCESS) {
+    taosMemoryFreeClear(pResultInfo->pData);
+    pRequest->body.fetchFp(pRequest->body.param, pRequest, 0);
+    return;
+  }
+
+  pRequest->code =
+      setQueryResultFromRsp(pResultInfo, (const SRetrieveTableRsp *)pResultInfo->pData, pResultInfo->convertUcs4, true);
+  if (pRequest->code != TSDB_CODE_SUCCESS) {
+    pResultInfo->numOfRows = 0;
+    pRequest->code = code;
+    tscError("0x%" PRIx64 " fetch results failed, code:%s, reqId:0x%" PRIx64, pRequest->self, tstrerror(code),
+             pRequest->requestId);
+  } else {
+    tscDebug("0x%" PRIx64 " fetch results, numOfRows:%" PRId64 " total Rows:%" PRId64 ", complete:%d, reqId:0x%" PRIx64,
+             pRequest->self, pResultInfo->numOfRows, pResultInfo->totalRows, pResultInfo->completed,
+             pRequest->requestId);
+
+    STscObj            *pTscObj = pRequest->pTscObj;
+    SAppClusterSummary *pActivity = &pTscObj->pAppInfo->summary;
+    atomic_add_fetch_64((int64_t *)&pActivity->fetchBytes, pRequest->body.resInfo.payloadLen);
+  }
+
+  pRequest->body.fetchFp(pRequest->body.param, pRequest, pResultInfo->numOfRows);
+}
+
+void taosAsyncFetchImpl(SRequestObj *pRequest, __taos_async_fn_t fp, void *param) {
+  pRequest->body.fetchFp = fp;
+  pRequest->body.param = param;
+
+  SReqResultInfo *pResultInfo = &pRequest->body.resInfo;
+
+  // this query has no results or error exists, return directly
+  if (taos_num_fields(pRequest) == 0 || pRequest->code != TSDB_CODE_SUCCESS) {
+    pResultInfo->numOfRows = 0;
+    pRequest->body.fetchFp(param, pRequest, pResultInfo->numOfRows);
+    return;
+  }
+
+  // all data has returned to App already, no need to try again
+  if (pResultInfo->completed) {
+    // it is a local executed query, no need to do async fetch
+    if (QUERY_EXEC_MODE_SCHEDULE != pRequest->body.execMode) {
+      if (pResultInfo->localResultFetched) {
+        pResultInfo->numOfRows = 0;
+        pResultInfo->current = 0;
+      } else {
+        pResultInfo->localResultFetched = true;
+      }
+    } else {
+      pResultInfo->numOfRows = 0;
+    }
+
+    pRequest->body.fetchFp(param, pRequest, pResultInfo->numOfRows);
+    return;
+  }
+
+  SSchedulerReq req = {
+      .syncReq = false,
+      .fetchFp = fetchCallback,
+      .cbParam = pRequest,
+  };
+
+  schedulerFetchRows(pRequest->body.queryJob, &req);
+}
+
+

source/client/src/clientMain.c

@@ -563,22 +563,13 @@ int taos_select_db(TAOS *taos, const char *db) {
   return code;
 }
 
+
 void taos_stop_query(TAOS_RES *res) {
   if (res == NULL || TD_RES_TMQ(res) || TD_RES_TMQ_META(res) || TD_RES_TMQ_METADATA(res)) {
     return;
   }
 
-  SRequestObj *pRequest = (SRequestObj *)res;
+  stopAllQueries((SRequestObj*)res);
-  pRequest->killed = true;
-
-  // It is not a query, no need to stop.
-  if (NULL == pRequest->pQuery || QUERY_EXEC_MODE_SCHEDULE != pRequest->pQuery->execMode) {
-    tscDebug("request 0x%" PRIx64 " no need to be killed since not query", pRequest->requestId);
-    return;
-  }
-
-  schedulerFreeJob(&pRequest->body.queryJob, TSDB_CODE_TSC_QUERY_KILLED);
-  tscDebug("request %" PRIx64 " killed", pRequest->requestId);
 }
 
 bool taos_is_null(TAOS_RES *res, int32_t row, int32_t col) {
@@ -774,8 +765,13 @@ static void destoryCatalogReq(SCatalogReq *pCatalogReq) {
   taosArrayDestroy(pCatalogReq->pDbVgroup);
   taosArrayDestroy(pCatalogReq->pDbCfg);
   taosArrayDestroy(pCatalogReq->pDbInfo);
+  if (pCatalogReq->cloned) {
+    taosArrayDestroy(pCatalogReq->pTableMeta);
+    taosArrayDestroy(pCatalogReq->pTableHash);
+  } else {
     taosArrayDestroyEx(pCatalogReq->pTableMeta, destoryTablesReq);
     taosArrayDestroyEx(pCatalogReq->pTableHash, destoryTablesReq);
+  }
   taosArrayDestroy(pCatalogReq->pUdf);
   taosArrayDestroy(pCatalogReq->pIndex);
   taosArrayDestroy(pCatalogReq->pUser);
@@ -794,26 +790,108 @@ void destorySqlCallbackWrapper(SSqlCallbackWrapper *pWrapper) {
   taosMemoryFree(pWrapper);
 }
 
+void destroyCtxInRequest(SRequestObj* pRequest) {
+  schedulerFreeJob(&pRequest->body.queryJob, 0);
+  qDestroyQuery(pRequest->pQuery);
+  pRequest->pQuery = NULL;
+  destorySqlCallbackWrapper(pRequest->pWrapper);
+  pRequest->pWrapper = NULL;
+}
+
+
 static void doAsyncQueryFromAnalyse(SMetaData *pResultMeta, void *param, int32_t code) {
   SSqlCallbackWrapper *pWrapper = (SSqlCallbackWrapper *)param;
   SRequestObj         *pRequest = pWrapper->pRequest;
   SQuery              *pQuery = pRequest->pQuery;
 
-  int64_t analyseStart = taosGetTimestampUs();
-  pRequest->metric.ctgCostUs = analyseStart - pRequest->metric.ctgStart;
   qDebug("0x%" PRIx64 " start to semantic analysis, reqId:0x%" PRIx64, pRequest->self, pRequest->requestId);
 
-  if (code == TSDB_CODE_SUCCESS) {
+  int64_t analyseStart = taosGetTimestampUs();
+  pRequest->metric.ctgCostUs = analyseStart - pRequest->metric.ctgStart;
+
+  if (TSDB_CODE_SUCCESS == code) {
     code = qAnalyseSqlSemantic(pWrapper->pParseCtx, pWrapper->pCatalogReq, pResultMeta, pQuery);
+  }
+  
+  pRequest->metric.analyseCostUs += taosGetTimestampUs() - analyseStart;
+  
+  handleQueryAnslyseRes(pWrapper, pResultMeta, code);
+}
+
+int32_t cloneCatalogReq(SCatalogReq*     * ppTarget, SCatalogReq* pSrc) {
+  int32_t code = TSDB_CODE_SUCCESS;
+  SCatalogReq* pTarget = taosMemoryCalloc(1, sizeof(SCatalogReq));
+  if (pTarget == NULL) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+  } else {
+    pTarget->pDbVgroup = taosArrayDup(pSrc->pDbVgroup, NULL);
+    pTarget->pDbCfg = taosArrayDup(pSrc->pDbCfg, NULL);
+    pTarget->pDbInfo = taosArrayDup(pSrc->pDbInfo, NULL);
+    pTarget->pTableMeta = taosArrayDup(pSrc->pTableMeta, NULL);
+    pTarget->pTableHash = taosArrayDup(pSrc->pTableHash, NULL);
+    pTarget->pUdf = taosArrayDup(pSrc->pUdf, NULL);
+    pTarget->pIndex = taosArrayDup(pSrc->pIndex, NULL);
+    pTarget->pUser = taosArrayDup(pSrc->pUser, NULL);
+    pTarget->pTableIndex = taosArrayDup(pSrc->pTableIndex, NULL);
+    pTarget->pTableCfg = taosArrayDup(pSrc->pTableCfg, NULL);
+    pTarget->pTableTag = taosArrayDup(pSrc->pTableTag, NULL);
+    pTarget->qNodeRequired = pSrc->qNodeRequired;
+    pTarget->dNodeRequired = pSrc->dNodeRequired;
+    pTarget->svrVerRequired = pSrc->svrVerRequired;
+    pTarget->forceUpdate = pSrc->forceUpdate;
+    pTarget->cloned = true;
+
+    *ppTarget = pTarget;
+  }
+
+  return code;
+}
+
+
+void handleSubQueryFromAnalyse(SSqlCallbackWrapper *pWrapper, SMetaData *pResultMeta, SNode* pRoot) {
+  SRequestObj* pNewRequest = NULL;
+  SSqlCallbackWrapper* pNewWrapper = NULL;
+  int32_t code = buildPreviousRequest(pWrapper->pRequest, pWrapper->pRequest->sqlstr, &pNewRequest);
+  if (code) {
+    handleQueryAnslyseRes(pWrapper, pResultMeta, code);
+    return;
+  }
+
+  pNewRequest->pQuery = (SQuery*)nodesMakeNode(QUERY_NODE_QUERY);
+  if (NULL == pNewRequest->pQuery) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+  } else {
+    pNewRequest->pQuery->pRoot = pRoot;
+    pRoot = NULL;
+    pNewRequest->pQuery->execStage = QUERY_EXEC_STAGE_ANALYSE;
+  }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = prepareAndParseSqlSyntax(&pNewWrapper, pNewRequest, false);
+  }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = cloneCatalogReq(&pNewWrapper->pCatalogReq, pWrapper->pCatalogReq);
+  }
+  doAsyncQueryFromAnalyse(pResultMeta, pNewWrapper, code);
+  nodesDestroyNode(pRoot);
+}
+
+void handleQueryAnslyseRes(SSqlCallbackWrapper *pWrapper, SMetaData *pResultMeta, int32_t code) {
+  SRequestObj         *pRequest = pWrapper->pRequest;
+  SQuery              *pQuery = pRequest->pQuery;
+
+  if (code == TSDB_CODE_SUCCESS && pQuery->pPrevRoot) {
+    SNode* prevRoot = pQuery->pPrevRoot;
+    pQuery->pPrevRoot = NULL;
+    handleSubQueryFromAnalyse(pWrapper, pResultMeta, prevRoot);
+    return;
+  }
+  
+  if (code == TSDB_CODE_SUCCESS) {
     pRequest->stableQuery = pQuery->stableQuery;
     if (pQuery->pRoot) {
       pRequest->stmtType = pQuery->pRoot->type;
     }
-  }
 
-  pRequest->metric.analyseCostUs = taosGetTimestampUs() - analyseStart;
-
-  if (code == TSDB_CODE_SUCCESS) {
     if (pQuery->haveResultSet) {
       setResSchemaInfo(&pRequest->body.resInfo, pQuery->pResSchema, pQuery->numOfResCols);
       setResPrecision(&pRequest->body.resInfo, pQuery->precision);
@@ -826,14 +904,14 @@ static void doAsyncQueryFromAnalyse(SMetaData *pResultMeta, void *param, int32_t
     launchAsyncQuery(pRequest, pQuery, pResultMeta, pWrapper);
   } else {
     destorySqlCallbackWrapper(pWrapper);
+    pRequest->pWrapper = NULL;
     qDestroyQuery(pRequest->pQuery);
     pRequest->pQuery = NULL;
 
     if (NEED_CLIENT_HANDLE_ERROR(code)) {
       tscDebug("0x%" PRIx64 " client retry to handle the error, code:%d - %s, tryCount:%d, reqId:0x%" PRIx64,
                pRequest->self, code, tstrerror(code), pRequest->retry, pRequest->requestId);
-      pRequest->prevCode = code;
+      restartAsyncQuery(pRequest, code);
-      doAsyncQuery(pRequest, true);
       return;
     }
 
@@ -841,7 +919,7 @@ static void doAsyncQueryFromAnalyse(SMetaData *pResultMeta, void *param, int32_t
     tscError("0x%" PRIx64 " error occurs, code:%s, return to user app, reqId:0x%" PRIx64, pRequest->self,
              tstrerror(code), pRequest->requestId);
     pRequest->code = code;
-    pRequest->body.queryFp(pRequest->body.param, pRequest, code);
+    returnToUser(pRequest);
   }
 }
 
@@ -904,6 +982,7 @@ static void doAsyncQueryFromParse(SMetaData *pResultMeta, void *param, int32_t c
     tscError("0x%" PRIx64 " error happens, code:%d - %s, reqId:0x%" PRIx64, pWrapper->pRequest->self, code,
              tstrerror(code), pWrapper->pRequest->requestId);
     destorySqlCallbackWrapper(pWrapper);
+    pRequest->pWrapper = NULL;
     terrno = code;
     pRequest->code = code;
     pRequest->body.queryFp(pRequest->body.param, pRequest, code);
@@ -920,6 +999,7 @@ void continueInsertFromCsv(SSqlCallbackWrapper *pWrapper, SRequestObj *pRequest)
     tscError("0x%" PRIx64 " error happens, code:%d - %s, reqId:0x%" PRIx64, pWrapper->pRequest->self, code,
              tstrerror(code), pWrapper->pRequest->requestId);
     destorySqlCallbackWrapper(pWrapper);
+    pRequest->pWrapper = NULL;
     terrno = code;
     pRequest->code = code;
     pRequest->body.queryFp(pRequest->body.param, pRequest, code);
@@ -967,27 +1047,16 @@ int32_t createParseContext(const SRequestObj *pRequest, SParseContext **pCxt) {
   return TSDB_CODE_SUCCESS;
 }
 
-void doAsyncQuery(SRequestObj *pRequest, bool updateMetaForce) {
+int32_t prepareAndParseSqlSyntax(SSqlCallbackWrapper **ppWrapper, SRequestObj *pRequest, bool updateMetaForce) {
-  STscObj             *pTscObj = pRequest->pTscObj;
-  SSqlCallbackWrapper *pWrapper = NULL;
   int32_t code = TSDB_CODE_SUCCESS;
-
+  STscObj             *pTscObj = pRequest->pTscObj;
-  if (pRequest->retry++ > REQUEST_TOTAL_EXEC_TIMES) {
+  SSqlCallbackWrapper *pWrapper = taosMemoryCalloc(1, sizeof(SSqlCallbackWrapper));
-    code = pRequest->prevCode;
-    terrno = code;
-    pRequest->code = code;
-    tscDebug("call sync query cb with code: %s", tstrerror(code));
-    pRequest->body.queryFp(pRequest->body.param, pRequest, code);
-    return;
-  }
-
-  if (TSDB_CODE_SUCCESS == code) {
-    pWrapper = taosMemoryCalloc(1, sizeof(SSqlCallbackWrapper));
   if (pWrapper == NULL) {
     code = TSDB_CODE_OUT_OF_MEMORY;
   } else {
     pWrapper->pRequest = pRequest;
-    }
+    pRequest->pWrapper = pWrapper;
+    *ppWrapper = pWrapper;
   }
 
   if (TSDB_CODE_SUCCESS == code) {
@@ -999,7 +1068,7 @@ void doAsyncQuery(SRequestObj *pRequest, bool updateMetaForce) {
     code = catalogGetHandle(pTscObj->pAppInfo->clusterId, &pWrapper->pParseCtx->pCatalog);
   }
 
-  if (TSDB_CODE_SUCCESS == code) {
+  if (TSDB_CODE_SUCCESS == code && NULL == pRequest->pQuery) {
     int64_t syntaxStart = taosGetTimestampUs();
 
     pWrapper->pCatalogReq = taosMemoryCalloc(1, sizeof(SCatalogReq));
@@ -1014,6 +1083,27 @@ void doAsyncQuery(SRequestObj *pRequest, bool updateMetaForce) {
     pRequest->metric.parseCostUs += taosGetTimestampUs() - syntaxStart;
   }
 
+  return code;
+}
+
+
+void doAsyncQuery(SRequestObj *pRequest, bool updateMetaForce) {
+  SSqlCallbackWrapper *pWrapper = NULL;
+  int32_t              code = TSDB_CODE_SUCCESS;
+
+  if (pRequest->retry++ > REQUEST_TOTAL_EXEC_TIMES) {
+    code = pRequest->prevCode;
+    terrno = code;
+    pRequest->code = code;
+    tscDebug("call sync query cb with code: %s", tstrerror(code));
+    pRequest->body.queryFp(pRequest->body.param, pRequest, code);
+    return;
+  }
+
+  if (TSDB_CODE_SUCCESS == code) {
+    code = prepareAndParseSqlSyntax(&pWrapper, pRequest, updateMetaForce);
+  }
+
   if (TSDB_CODE_SUCCESS == code) {
     pRequest->stmtType = pRequest->pQuery->pRoot->type;
     code = phaseAsyncQuery(pWrapper);
@@ -1023,6 +1113,7 @@ void doAsyncQuery(SRequestObj *pRequest, bool updateMetaForce) {
     tscError("0x%" PRIx64 " error happens, code:%d - %s, reqId:0x%" PRIx64, pRequest->self, code, tstrerror(code),
              pRequest->requestId);
     destorySqlCallbackWrapper(pWrapper);
+    pRequest->pWrapper = NULL;
     qDestroyQuery(pRequest->pQuery);
     pRequest->pQuery = NULL;
 
@@ -1040,48 +1131,57 @@ void doAsyncQuery(SRequestObj *pRequest, bool updateMetaForce) {
   }
 }
 
-static void fetchCallback(void *pResult, void *param, int32_t code) {
+void restartAsyncQuery(SRequestObj *pRequest, int32_t code) {
-  SRequestObj *pRequest = (SRequestObj *)param;
+  int32_t reqIdx = 0;
-
+  SRequestObj *pReqList[16] = {NULL};
-  SReqResultInfo *pResultInfo = &pRequest->body.resInfo;
+  SRequestObj *pUserReq = NULL;
-
+  pReqList[0] = pRequest;
-  tscDebug("0x%" PRIx64 " enter scheduler fetch cb, code:%d - %s, reqId:0x%" PRIx64, pRequest->self, code,
+  uint64_t tmpRefId = 0;
-           tstrerror(code), pRequest->requestId);
+  SRequestObj* pTmp = pRequest;
-
+  while (pTmp->relation.prevRefId) {
-  pResultInfo->pData = pResult;
+    tmpRefId = pTmp->relation.prevRefId;
-  pResultInfo->numOfRows = 0;
+    pTmp = acquireRequest(tmpRefId);
-
+    if (pTmp) {
-  if (code != TSDB_CODE_SUCCESS) {
+      pReqList[++reqIdx] = pTmp;
-    pRequest->code = code;
+      releaseRequest(tmpRefId);
-    taosMemoryFreeClear(pResultInfo->pData);
-    pRequest->body.fetchFp(pRequest->body.param, pRequest, 0);
-    return;
-  }
-
-  if (pRequest->code != TSDB_CODE_SUCCESS) {
-    taosMemoryFreeClear(pResultInfo->pData);
-    pRequest->body.fetchFp(pRequest->body.param, pRequest, 0);
-    return;
-  }
-
-  pRequest->code =
-      setQueryResultFromRsp(pResultInfo, (const SRetrieveTableRsp *)pResultInfo->pData, pResultInfo->convertUcs4, true);
-  if (pRequest->code != TSDB_CODE_SUCCESS) {
-    pResultInfo->numOfRows = 0;
-    pRequest->code = code;
-    tscError("0x%" PRIx64 " fetch results failed, code:%s, reqId:0x%" PRIx64, pRequest->self, tstrerror(code),
-             pRequest->requestId);
     } else {
-    tscDebug("0x%" PRIx64 " fetch results, numOfRows:%" PRId64 " total Rows:%" PRId64 ", complete:%d, reqId:0x%" PRIx64,
+      tscError("0x%" PRIx64 ", prev req ref 0x%" PRIx64 " is not there, reqId:0x%" PRIx64, pTmp->self,
-             pRequest->self, pResultInfo->numOfRows, pResultInfo->totalRows, pResultInfo->completed,
+               tmpRefId, pTmp->requestId);
-             pRequest->requestId);
+      break;         
-
+    }
-    STscObj            *pTscObj = pRequest->pTscObj;
-    SAppClusterSummary *pActivity = &pTscObj->pAppInfo->summary;
-    atomic_add_fetch_64((int64_t *)&pActivity->fetchBytes, pRequest->body.resInfo.payloadLen);
   }
 
-  pRequest->body.fetchFp(pRequest->body.param, pRequest, pResultInfo->numOfRows);
+  tmpRefId = pRequest->relation.nextRefId;
+  while (tmpRefId) {
+    pTmp = acquireRequest(tmpRefId);
+    if (pTmp) {
+      tmpRefId = pTmp->relation.nextRefId;
+      removeRequest(pTmp->self);      
+      releaseRequest(pTmp->self);
+    } else {
+      tscError("0x%" PRIx64 " is not there", tmpRefId);
+      break;         
+    }
+  }
+
+  for (int32_t i = reqIdx; i >= 0; i--) {
+    destroyCtxInRequest(pReqList[i]);
+    if (pReqList[i]->relation.userRefId == pReqList[i]->self || 0 == pReqList[i]->relation.userRefId) {
+      pUserReq = pReqList[i];
+    } else {
+      removeRequest(pReqList[i]->self);
+    }
+  }
+
+  if (pUserReq) {
+    pUserReq->prevCode = code;
+    memset(&pUserReq->relation, 0, sizeof(pUserReq->relation));
+  } else {
+    tscError("user req is missing");
+    return;
+  }
+
+  doAsyncQuery(pUserReq, true);
 }
 
 void taos_fetch_rows_a(TAOS_RES *res, __taos_async_fn_t fp, void *param) {
@@ -1095,43 +1195,8 @@ void taos_fetch_rows_a(TAOS_RES *res, __taos_async_fn_t fp, void *param) {
   }
 
   SRequestObj *pRequest = res;
-  pRequest->body.fetchFp = fp;
-  pRequest->body.param = param;
 
-  SReqResultInfo *pResultInfo = &pRequest->body.resInfo;
+  taosAsyncFetchImpl(pRequest, fp, param);
-
-  // this query has no results or error exists, return directly
-  if (taos_num_fields(pRequest) == 0 || pRequest->code != TSDB_CODE_SUCCESS) {
-    pResultInfo->numOfRows = 0;
-    pRequest->body.fetchFp(param, pRequest, pResultInfo->numOfRows);
-    return;
-  }
-
-  // all data has returned to App already, no need to try again
-  if (pResultInfo->completed) {
-    // it is a local executed query, no need to do async fetch
-    if (QUERY_EXEC_MODE_SCHEDULE != pRequest->body.execMode) {
-      if (pResultInfo->localResultFetched) {
-        pResultInfo->numOfRows = 0;
-        pResultInfo->current = 0;
-      } else {
-        pResultInfo->localResultFetched = true;
-      }
-    } else {
-      pResultInfo->numOfRows = 0;
-    }
-
-    pRequest->body.fetchFp(param, pRequest, pResultInfo->numOfRows);
-    return;
-  }
-
-  SSchedulerReq req = {
-      .syncReq = false,
-      .fetchFp = fetchCallback,
-      .cbParam = pRequest,
-  };
-
-  schedulerFetchRows(pRequest->body.queryJob, &req);
 }
 
 void taos_fetch_raw_block_a(TAOS_RES *res, __taos_async_fn_t fp, void *param) {

source/client/src/clientMsgHandler.c

@@ -77,6 +77,7 @@ int32_t processConnectRsp(void* param, SDataBuf* pMsg, int32_t code) {
   }
 
   if ((code = taosCheckVersionCompatibleFromStr(version, connectRsp.sVer, 3)) != 0) {
+    tscError("version not compatible. client version: %s, server version: %s", version, connectRsp.sVer);
     setErrno(pRequest, code);
     tsem_post(&pRequest->body.rspSem);
     goto End;

source/client/src/clientSml.c

@@ -1553,17 +1553,8 @@ static int32_t smlParseLine(SSmlHandle *info, char *lines[], char *rawLine, char
       }
     }
 
-    char cTmp = 0;  // for print tmp if is raw
-    if (info->isRawLine) {
-      cTmp = tmp[len];
-      tmp[len] = '\0';
-    }
-
     uDebug("SML:0x%" PRIx64 " smlParseLine israw:%d, numLines:%d, protocol:%d, len:%d, sql:%s", info->id,
-           info->isRawLine, numLines, info->protocol, len, tmp);
+           info->isRawLine, numLines, info->protocol, len, info->isRawLine ? "rawdata" : tmp);
-    if (info->isRawLine) {
-      tmp[len] = cTmp;
-    }
 
     if (info->protocol == TSDB_SML_LINE_PROTOCOL) {
       if (info->dataFormat) {
@@ -1584,8 +1575,7 @@ static int32_t smlParseLine(SSmlHandle *info, char *lines[], char *rawLine, char
       code = TSDB_CODE_SML_INVALID_PROTOCOL_TYPE;
     }
     if (code != TSDB_CODE_SUCCESS) {
-      tmp[len] = '\0';
+      uError("SML:0x%" PRIx64 " smlParseLine failed. line %d : %s", info->id, i, info->isRawLine ? "rawdata" : tmp);
-      uError("SML:0x%" PRIx64 " smlParseLine failed. line %d : %s", info->id, i, tmp);
       return code;
     }
     if (info->reRun) {
@@ -1756,9 +1746,8 @@ TAOS_RES *taos_schemaless_insert_inner(TAOS *taos, char *lines[], char *rawLine,
     request->code = code;
     info->cost.endTime = taosGetTimestampUs();
     info->cost.code = code;
-    if (code == TSDB_CODE_TDB_INVALID_TABLE_SCHEMA_VER || code == TSDB_CODE_SDB_OBJ_CREATING ||
+    if (NEED_CLIENT_HANDLE_ERROR(code) || code == TSDB_CODE_SDB_OBJ_CREATING ||
-        code == TSDB_CODE_PAR_VALUE_TOO_LONG || code == TSDB_CODE_MND_TRANS_CONFLICT ||
+        code == TSDB_CODE_PAR_VALUE_TOO_LONG || code == TSDB_CODE_MND_TRANS_CONFLICT) {
-        code == TSDB_CODE_PAR_TABLE_NOT_EXIST) {
       if (cnt++ >= 10) {
         uInfo("SML:%" PRIx64 " retry:%d/10 end code:%d, msg:%s", info->id, cnt, code, tstrerror(code));
         break;

source/client/src/clientTmq.c

@@ -358,7 +358,7 @@ tmq_conf_res_t tmq_conf_set(tmq_conf_t* conf, const char* key, const char* value
     return TMQ_CONF_OK;
   }
 
-  if (strcasecmp(key, "enable.heartbeat.background") == 0) {
+//  if (strcasecmp(key, "enable.heartbeat.background") == 0) {
     //    if (strcasecmp(value, "true") == 0) {
     //      conf->hbBgEnable = true;
     //      return TMQ_CONF_OK;
@@ -366,10 +366,10 @@ tmq_conf_res_t tmq_conf_set(tmq_conf_t* conf, const char* key, const char* value
     //      conf->hbBgEnable = false;
     //      return TMQ_CONF_OK;
     //    } else {
-    tscError("the default value of enable.heartbeat.background is true, can not be seted");
+//    tscError("the default value of enable.heartbeat.background is true, can not be seted");
-    return TMQ_CONF_INVALID;
+//    return TMQ_CONF_INVALID;
+    //    }
 //  }
-  }
 
   if (strcasecmp(key, "td.connect.ip") == 0) {
     conf->ip = taosStrdup(value);
@@ -423,30 +423,30 @@ char** tmq_list_to_c_array(const tmq_list_t* list) {
   return container->pData;
 }
 
-static SMqClientVg* foundClientVg(SArray* pTopicList, const char* pName, int32_t vgId, int32_t* index,
+//static SMqClientVg* foundClientVg(SArray* pTopicList, const char* pName, int32_t vgId, int32_t* index,
-                                  int32_t* numOfVgroups) {
+//                                  int32_t* numOfVgroups) {
-  int32_t numOfTopics = taosArrayGetSize(pTopicList);
+//  int32_t numOfTopics = taosArrayGetSize(pTopicList);
-  *index = -1;
+//  *index = -1;
-  *numOfVgroups = 0;
+//  *numOfVgroups = 0;
-
+//
-  for (int32_t i = 0; i < numOfTopics; ++i) {
+//  for (int32_t i = 0; i < numOfTopics; ++i) {
-    SMqClientTopic* pTopic = taosArrayGet(pTopicList, i);
+//    SMqClientTopic* pTopic = taosArrayGet(pTopicList, i);
-    if (strcmp(pTopic->topicName, pName) != 0) {
+//    if (strcmp(pTopic->topicName, pName) != 0) {
-      continue;
+//      continue;
-    }
+//    }
-
+//
-    *numOfVgroups = taosArrayGetSize(pTopic->vgs);
+//    *numOfVgroups = taosArrayGetSize(pTopic->vgs);
-    for (int32_t j = 0; j < (*numOfVgroups); ++j) {
+//    for (int32_t j = 0; j < (*numOfVgroups); ++j) {
-      SMqClientVg* pClientVg = taosArrayGet(pTopic->vgs, j);
+//      SMqClientVg* pClientVg = taosArrayGet(pTopic->vgs, j);
-      if (pClientVg->vgId == vgId) {
+//      if (pClientVg->vgId == vgId) {
-        *index = j;
+//        *index = j;
-        return pClientVg;
+//        return pClientVg;
-      }
+//      }
-    }
+//    }
-  }
+//  }
-
+//
-  return NULL;
+//  return NULL;
-}
+//}
 
 // Two problems do not need to be addressed here
 // 1. update to of epset. the response of poll request will automatically handle this problem
@@ -573,7 +573,7 @@ static int32_t doSendCommitMsg(tmq_t* tmq, SMqClientVg* pVg, const char* pTopicN
 
   char commitBuf[TSDB_OFFSET_LEN] = {0};
   tFormatOffset(commitBuf, tListLen(commitBuf), &pVg->offsetInfo.committedOffset);
-  tscDebug("consumer:0x%" PRIx64 " topic:%s on vgId:%d send offset:%s prev:%s, ep:%s:%d, ordinal:%d/%d, req:0x%" PRIx64,
+  tscInfo("consumer:0x%" PRIx64 " topic:%s on vgId:%d send offset:%s prev:%s, ep:%s:%d, ordinal:%d/%d, req:0x%" PRIx64,
            tmq->consumerId, pOffset->offset.subKey, pVg->vgId, offsetBuf, commitBuf, pEp->fqdn, pEp->port, index + 1,
            totalVgroups, pMsgSendInfo->requestId);
 
@@ -811,7 +811,9 @@ void tmqSendHbReq(void* param, void* tmrId) {
         offRows->vgId = pVg->vgId;
         offRows->rows = pVg->numOfRows;
         offRows->offset = pVg->offsetInfo.committedOffset;
-        tscDebug("report offset: %d", offRows->offset.type);
+        char buf[TSDB_OFFSET_LEN] = {0};
+        tFormatOffset(buf, TSDB_OFFSET_LEN, &offRows->offset);
+        tscInfo("report offset: vgId:%d, offset:%s, rows:%"PRId64, offRows->vgId, buf, offRows->rows);
       }
     }
     tmq->needReportOffsetRows = false;
@@ -862,7 +864,7 @@ OVER:
 
 static void defaultCommitCbFn(tmq_t* pTmq, int32_t code, void* param) {
   if (code != 0) {
-    tscDebug("consumer:0x%" PRIx64 ", failed to commit offset, code:%s", pTmq->consumerId, tstrerror(code));
+    tscError("consumer:0x%" PRIx64 ", failed to commit offset, code:%s", pTmq->consumerId, tstrerror(code));
   }
 }
 
@@ -1161,7 +1163,7 @@ int32_t tmq_subscribe(tmq_t* tmq, const tmq_list_t* topic_list) {
   SCMSubscribeReq req = {0};
   int32_t         code = 0;
 
-  tscDebug("consumer:0x%" PRIx64 " cgroup:%s, subscribe %d topics", tmq->consumerId, tmq->groupId, sz);
+  tscInfo("consumer:0x%" PRIx64 " cgroup:%s, subscribe %d topics", tmq->consumerId, tmq->groupId, sz);
 
   req.consumerId = tmq->consumerId;
   tstrncpy(req.clientId, tmq->clientId, 256);
@@ -1174,7 +1176,6 @@ int32_t tmq_subscribe(tmq_t* tmq, const tmq_list_t* topic_list) {
   }
 
   req.withTbName = tmq->withTbName;
-  req.useSnapshot = tmq->useSnapshot;
   req.autoCommit = tmq->autoCommit;
   req.autoCommitInterval = tmq->autoCommitInterval;
   req.resetOffsetCfg = tmq->resetOffsetCfg;
@@ -1190,7 +1191,7 @@ int32_t tmq_subscribe(tmq_t* tmq, const tmq_list_t* topic_list) {
     }
 
     tNameExtractFullName(&name, topicFName);
-    tscDebug("consumer:0x%" PRIx64 " subscribe topic:%s", tmq->consumerId, topicFName);
+    tscInfo("consumer:0x%" PRIx64 " subscribe topic:%s", tmq->consumerId, topicFName);
 
     taosArrayPush(req.topicNames, &topicFName);
   }
@@ -1251,7 +1252,7 @@ int32_t tmq_subscribe(tmq_t* tmq, const tmq_list_t* topic_list) {
       goto FAIL;
     }
 
-    tscDebug("consumer:0x%" PRIx64 ", mnd not ready for subscribe, retry:%d in 500ms", tmq->consumerId, retryCnt);
+    tscInfo("consumer:0x%" PRIx64 ", mnd not ready for subscribe, retry:%d in 500ms", tmq->consumerId, retryCnt);
     taosMsleep(500);
   }
 
@@ -1478,7 +1479,7 @@ static void initClientTopicFromRsp(SMqClientTopic* pTopic, SMqSubTopicEp* pTopic
   tstrncpy(pTopic->topicName, pTopicEp->topic, TSDB_TOPIC_FNAME_LEN);
   tstrncpy(pTopic->db, pTopicEp->db, TSDB_DB_FNAME_LEN);
 
-  tscDebug("consumer:0x%" PRIx64 ", update topic:%s, new numOfVgs:%d", tmq->consumerId, pTopic->topicName, vgNumGet);
+  tscInfo("consumer:0x%" PRIx64 ", update topic:%s, new numOfVgs:%d", tmq->consumerId, pTopic->topicName, vgNumGet);
   pTopic->vgs = taosArrayInit(vgNumGet, sizeof(SMqClientVg));
 
   for (int32_t j = 0; j < vgNumGet; j++) {
@@ -1531,7 +1532,7 @@ static bool doUpdateLocalEp(tmq_t* tmq, int32_t epoch, const SMqAskEpRsp* pRsp)
   int32_t topicNumGet = taosArrayGetSize(pRsp->topics);
 
   char vgKey[TSDB_TOPIC_FNAME_LEN + 22];
-  tscDebug("consumer:0x%" PRIx64 " update ep epoch from %d to epoch %d, incoming topics:%d, existed topics:%d",
+  tscInfo("consumer:0x%" PRIx64 " update ep epoch from %d to epoch %d, incoming topics:%d, existed topics:%d",
            tmq->consumerId, tmq->epoch, epoch, topicNumGet, topicNumCur);
   if (epoch <= tmq->epoch) {
     return false;
@@ -1554,14 +1555,14 @@ static bool doUpdateLocalEp(tmq_t* tmq, int32_t epoch, const SMqAskEpRsp* pRsp)
     SMqClientTopic* pTopicCur = taosArrayGet(tmq->clientTopics, i);
     if (pTopicCur->vgs) {
       int32_t vgNumCur = taosArrayGetSize(pTopicCur->vgs);
-      tscDebug("consumer:0x%" PRIx64 ", current vg num: %d", tmq->consumerId, vgNumCur);
+      tscInfo("consumer:0x%" PRIx64 ", current vg num: %d", tmq->consumerId, vgNumCur);
       for (int32_t j = 0; j < vgNumCur; j++) {
         SMqClientVg* pVgCur = taosArrayGet(pTopicCur->vgs, j);
         makeTopicVgroupKey(vgKey, pTopicCur->topicName, pVgCur->vgId);
 
         char buf[TSDB_OFFSET_LEN];
         tFormatOffset(buf, TSDB_OFFSET_LEN, &pVgCur->offsetInfo.currentOffset);
-        tscDebug("consumer:0x%" PRIx64 ", epoch:%d vgId:%d vgKey:%s, offset:%s", tmq->consumerId, epoch, pVgCur->vgId,
+        tscInfo("consumer:0x%" PRIx64 ", epoch:%d vgId:%d vgKey:%s, offset:%s", tmq->consumerId, epoch, pVgCur->vgId,
                  vgKey, buf);
 
         SVgroupSaveInfo info = {.offset = pVgCur->offsetInfo.currentOffset, .numOfRows = pVgCur->numOfRows};
@@ -1591,7 +1592,7 @@ static bool doUpdateLocalEp(tmq_t* tmq, int32_t epoch, const SMqAskEpRsp* pRsp)
   atomic_store_8(&tmq->status, flag);
   atomic_store_32(&tmq->epoch, epoch);
 
-  tscDebug("consumer:0x%" PRIx64 " update topic info completed", tmq->consumerId);
+  tscInfo("consumer:0x%" PRIx64 " update topic info completed", tmq->consumerId);
   return set;
 }
 
@@ -1627,7 +1628,7 @@ int32_t askEpCallbackFn(void* param, SDataBuf* pMsg, int32_t code) {
   SMqRspHead* head = pMsg->pData;
   int32_t     epoch = atomic_load_32(&tmq->epoch);
   if (head->epoch <= epoch) {
-    tscDebug("consumer:0x%" PRIx64 ", recv ep, msg epoch %d, current epoch %d, no need to update local ep",
+    tscInfo("consumer:0x%" PRIx64 ", recv ep, msg epoch %d, current epoch %d, no need to update local ep",
              tmq->consumerId, head->epoch, epoch);
 
     if (tmq->status == TMQ_CONSUMER_STATUS__RECOVER) {
@@ -1639,7 +1640,7 @@ int32_t askEpCallbackFn(void* param, SDataBuf* pMsg, int32_t code) {
     }
 
   } else {
-    tscDebug("consumer:0x%" PRIx64 ", recv ep, msg epoch %d, current epoch %d, update local ep", tmq->consumerId,
+    tscInfo("consumer:0x%" PRIx64 ", recv ep, msg epoch %d, current epoch %d, update local ep", tmq->consumerId,
              head->epoch, epoch);
   }
 
@@ -2067,12 +2068,12 @@ TAOS_RES* tmq_consumer_poll(tmq_t* tmq, int64_t timeout) {
   void*   rspObj;
   int64_t startTime = taosGetTimestampMs();
 
-  tscDebug("consumer:0x%" PRIx64 " start to poll at %" PRId64 ", timeout:%" PRId64, tmq->consumerId, startTime,
+  tscInfo("consumer:0x%" PRIx64 " start to poll at %" PRId64 ", timeout:%" PRId64, tmq->consumerId, startTime,
            timeout);
 
   // in no topic status, delayed task also need to be processed
   if (atomic_load_8(&tmq->status) == TMQ_CONSUMER_STATUS__INIT) {
-    tscDebug("consumer:0x%" PRIx64 " poll return since consumer is init", tmq->consumerId);
+    tscInfo("consumer:0x%" PRIx64 " poll return since consumer is init", tmq->consumerId);
     taosMsleep(500);  //     sleep for a while
     return NULL;
   }
@@ -2084,7 +2085,7 @@ TAOS_RES* tmq_consumer_poll(tmq_t* tmq, int64_t timeout) {
         return NULL;
       }
 
-      tscDebug("consumer:0x%" PRIx64 " not ready, retry:%d/40 in 500ms", tmq->consumerId, retryCnt);
+      tscInfo("consumer:0x%" PRIx64 " not ready, retry:%d/40 in 500ms", tmq->consumerId, retryCnt);
       taosMsleep(500);
     }
   }
@@ -2093,7 +2094,7 @@ TAOS_RES* tmq_consumer_poll(tmq_t* tmq, int64_t timeout) {
     tmqHandleAllDelayedTask(tmq);
 
     if (tmqPollImpl(tmq, timeout) < 0) {
-      tscDebug("consumer:0x%" PRIx64 " return due to poll error", tmq->consumerId);
+      tscError("consumer:0x%" PRIx64 " return due to poll error", tmq->consumerId);
     }
 
     rspObj = tmqHandleAllRsp(tmq, timeout, false);
@@ -2101,7 +2102,7 @@ TAOS_RES* tmq_consumer_poll(tmq_t* tmq, int64_t timeout) {
       tscDebug("consumer:0x%" PRIx64 " return rsp %p", tmq->consumerId, rspObj);
       return (TAOS_RES*)rspObj;
     } else if (terrno == TSDB_CODE_TQ_NO_COMMITTED_OFFSET) {
-      tscDebug("consumer:0x%" PRIx64 " return null since no committed offset", tmq->consumerId);
+      tscInfo("consumer:0x%" PRIx64 " return null since no committed offset", tmq->consumerId);
       return NULL;
     }
 
@@ -2109,7 +2110,7 @@ TAOS_RES* tmq_consumer_poll(tmq_t* tmq, int64_t timeout) {
       int64_t currentTime = taosGetTimestampMs();
       int64_t elapsedTime = currentTime - startTime;
       if (elapsedTime > timeout) {
-        tscDebug("consumer:0x%" PRIx64 " (epoch %d) timeout, no rsp, start time %" PRId64 ", current time %" PRId64,
+        tscInfo("consumer:0x%" PRIx64 " (epoch %d) timeout, no rsp, start time %" PRId64 ", current time %" PRId64,
                  tmq->consumerId, tmq->epoch, startTime, currentTime);
         return NULL;
       }
@@ -2142,7 +2143,7 @@ static void displayConsumeStatistics(const tmq_t* pTmq) {
 }
 
 int32_t tmq_consumer_close(tmq_t* tmq) {
-  tscDebug("consumer:0x%" PRIx64 " start to close consumer, status:%d", tmq->consumerId, tmq->status);
+  tscInfo("consumer:0x%" PRIx64 " start to close consumer, status:%d", tmq->consumerId, tmq->status);
   displayConsumeStatistics(tmq);
 
   if (tmq->status == TMQ_CONSUMER_STATUS__READY) {
@@ -2169,7 +2170,7 @@ int32_t tmq_consumer_close(tmq_t* tmq) {
 
     tmq_list_destroy(lst);
   } else {
-    tscWarn("consumer:0x%" PRIx64 " not in ready state, close it directly", tmq->consumerId);
+    tscInfo("consumer:0x%" PRIx64 " not in ready state, close it directly", tmq->consumerId);
   }
 
   taosRemoveRef(tmqMgmt.rsetId, tmq->refId);
@@ -2432,7 +2433,7 @@ void asyncAskEp(tmq_t* pTmq, __tmq_askep_fn_t askEpFn, void* param) {
   sendInfo->msgType = TDMT_MND_TMQ_ASK_EP;
 
   SEpSet epSet = getEpSet_s(&pTmq->pTscObj->pAppInfo->mgmtEp);
-  tscDebug("consumer:0x%" PRIx64 " ask ep from mnode, reqId:0x%" PRIx64, pTmq->consumerId, sendInfo->requestId);
+  tscInfo("consumer:0x%" PRIx64 " ask ep from mnode, reqId:0x%" PRIx64, pTmq->consumerId, sendInfo->requestId);
 
   int64_t transporterId = 0;
   asyncSendMsgToServer(pTmq->pTscObj->pAppInfo->pTransporter, &epSet, &transporterId, sendInfo);
@@ -2656,7 +2657,7 @@ int32_t tmq_get_topic_assignment(tmq_t* tmq, const char* pTopicName, tmq_topic_a
       char    offsetFormatBuf[TSDB_OFFSET_LEN];
       tFormatOffset(offsetFormatBuf, tListLen(offsetFormatBuf), &pClientVg->offsetInfo.currentOffset);
 
-      tscDebug("consumer:0x%" PRIx64 " %s retrieve wal info vgId:%d, epoch %d, req:%s, reqId:0x%" PRIx64,
+      tscInfo("consumer:0x%" PRIx64 " %s retrieve wal info vgId:%d, epoch %d, req:%s, reqId:0x%" PRIx64,
                tmq->consumerId, pTopic->topicName, pClientVg->vgId, tmq->epoch, offsetFormatBuf, req.reqId);
       asyncSendMsgToServer(tmq->pTscObj->pAppInfo->pTransporter, &pClientVg->epSet, &transporterId, sendInfo);
     }
@@ -2693,7 +2694,7 @@ int32_t tmq_get_topic_assignment(tmq_t* tmq, const char* pTopicName, tmq_topic_a
         char offsetBuf[TSDB_OFFSET_LEN] = {0};
         tFormatOffset(offsetBuf, tListLen(offsetBuf), &pOffsetInfo->currentOffset);
 
-        tscDebug("vgId:%d offset is update to:%s", p->vgId, offsetBuf);
+        tscInfo("vgId:%d offset is update to:%s", p->vgId, offsetBuf);
 
         pOffsetInfo->walVerBegin = p->begin;
         pOffsetInfo->walVerEnd = p->end;
@@ -2772,7 +2773,7 @@ int32_t tmq_offset_seek(tmq_t* tmq, const char* pTopicName, int32_t vgId, int64_
   SMqRspObj rspObj = {.resType = RES_TYPE__TMQ, .vgId = pVg->vgId};
   tstrncpy(rspObj.topic, tname, tListLen(rspObj.topic));
 
-  tscDebug("consumer:0x%" PRIx64 " seek to %" PRId64 " on vgId:%d", tmq->consumerId, offset, pVg->vgId);
+  tscInfo("consumer:0x%" PRIx64 " seek to %" PRId64 " on vgId:%d", tmq->consumerId, offset, pVg->vgId);
 
   SSyncCommitInfo* pInfo = taosMemoryMalloc(sizeof(SSyncCommitInfo));
   if (pInfo == NULL) {

source/common/src/systable.c

@@ -280,7 +280,7 @@ static const SSysDbTableSchema topicSchema[] = {
     {.name = "db_name", .bytes = SYSTABLE_SCH_DB_NAME_LEN, .type = TSDB_DATA_TYPE_BINARY, .sysInfo = false},
     {.name = "create_time", .bytes = 8, .type = TSDB_DATA_TYPE_TIMESTAMP, .sysInfo = false},
     {.name = "sql", .bytes = TSDB_SHOW_SQL_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_BINARY, .sysInfo = false},
-    {.name = "schema", .bytes = TSDB_SHOW_SCHEMA_JSON_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_BINARY, .sysInfo = false},
+    {.name = "schema", .bytes = TSDB_MAX_BINARY_LEN, .type = TSDB_DATA_TYPE_BINARY, .sysInfo = false},
     {.name = "meta", .bytes = 4 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_BINARY, .sysInfo = false},
     {.name = "type", .bytes = 8 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_BINARY, .sysInfo = false},
 };
@@ -361,11 +361,7 @@ static const SSysDbTableSchema consumerSchema[] = {
     {.name = "up_time", .bytes = 8, .type = TSDB_DATA_TYPE_TIMESTAMP, .sysInfo = false},
     {.name = "subscribe_time", .bytes = 8, .type = TSDB_DATA_TYPE_TIMESTAMP, .sysInfo = false},
     {.name = "rebalance_time", .bytes = 8, .type = TSDB_DATA_TYPE_TIMESTAMP, .sysInfo = false},
-    {.name = "msg.with.table.name", .bytes = 1, .type = TSDB_DATA_TYPE_SMALLINT, .sysInfo = false},
+    {.name = "parameters", .bytes = 64 + TSDB_OFFSET_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_BINARY, .sysInfo = false},
-    {.name = "experimental.snapshot.enable", .bytes = 1, .type = TSDB_DATA_TYPE_SMALLINT, .sysInfo = false},
-    {.name = "enable.auto.commit", .bytes = 1, .type = TSDB_DATA_TYPE_SMALLINT, .sysInfo = false},
-    {.name = "auto.commit.interval.ms", .bytes = 4, .type = TSDB_DATA_TYPE_INT, .sysInfo = false},
-    {.name = "auto.offset.reset", .bytes = TSDB_OFFSET_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_BINARY, .sysInfo = false},
 };
 
 static const SSysDbTableSchema offsetSchema[] = {
@@ -388,6 +384,7 @@ static const SSysDbTableSchema querySchema[] = {
     {.name = "create_time", .bytes = 8, .type = TSDB_DATA_TYPE_TIMESTAMP, .sysInfo = false},
     {.name = "exec_usec", .bytes = 8, .type = TSDB_DATA_TYPE_BIGINT, .sysInfo = false},
     {.name = "stable_query", .bytes = 1, .type = TSDB_DATA_TYPE_BOOL, .sysInfo = false},
+    {.name = "sub_query", .bytes = 1, .type = TSDB_DATA_TYPE_BOOL, .sysInfo = false},
     {.name = "sub_num", .bytes = 4, .type = TSDB_DATA_TYPE_INT, .sysInfo = false},
     {.name = "sub_status", .bytes = TSDB_SHOW_SUBQUERY_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR, .sysInfo = false},
     {.name = "sql", .bytes = TSDB_SHOW_SQL_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR, .sysInfo = false},

source/common/src/tdatablock.c

@@ -1590,18 +1590,35 @@ size_t blockDataGetCapacityInRow(const SSDataBlock* pBlock, size_t pageSize, int
   int32_t nRows = payloadSize / rowSize;
   ASSERT(nRows >= 1);
 
-  // the true value must be less than the value of nRows
+  int32_t numVarCols = 0;
-  int32_t additional = 0;
+  int32_t numFixCols = 0;
   for (int32_t i = 0; i < numOfCols; ++i) {
     SColumnInfoData* pCol = taosArrayGet(pBlock->pDataBlock, i);
     if (IS_VAR_DATA_TYPE(pCol->info.type)) {
-      additional += nRows * sizeof(int32_t);
+      ++numVarCols;
     } else {
-      additional += BitmapLen(nRows);
+      ++numFixCols;
     }
   }
 
-  int32_t newRows = (payloadSize - additional) / rowSize;
+  // find the data payload whose size is greater than payloadSize
+  int result = -1;
+  int start = 1;
+  int end = nRows;
+  while (start <= end) {
+    int mid = start + (end - start) / 2;
+    //data size + var data type columns offset + fixed data type columns bitmap len 
+    int midSize = rowSize * mid + numVarCols * sizeof(int32_t) * mid + numFixCols * BitmapLen(mid); 
+    if (midSize > payloadSize) {
+      result = mid;
+      end = mid - 1;
+    } else {
+      start = mid + 1;
+    }
+  }
+
+  int32_t newRows = (result != -1) ? result - 1 : nRows;
+  // the true value must be less than the value of nRows
   ASSERT(newRows <= nRows && newRows >= 1);
 
   return newRows;

source/common/src/tglobal.c

@@ -14,6 +14,7 @@
  */
 
 #define _DEFAULT_SOURCE
+#include "os.h"
 #include "tglobal.h"
 #include "tconfig.h"
 #include "tgrant.h"
@@ -73,6 +74,7 @@ int64_t tsVndCommitMaxIntervalMs = 600 * 1000;
 // mnode
 int64_t tsMndSdbWriteDelta = 200;
 int64_t tsMndLogRetention = 2000;
+int8_t  tsGrant = 1;
 bool    tsMndSkipGrant = false;
 
 // monitor
@@ -1525,3 +1527,5 @@ void taosSetAllDebugFlag(int32_t flag, bool rewrite) {
   taosSetDebugFlag(&metaDebugFlag, "metaDebugFlag", flag, rewrite);
   uInfo("all debug flag are set to %d", flag);
 }
+
+int8_t taosGranted() { return atomic_load_8(&tsGrant); }

source/common/src/tmsg.c

@@ -224,6 +224,7 @@ static int32_t tSerializeSClientHbReq(SEncoder *pEncoder, const SClientHbReq *pR
         if (tEncodeI64(pEncoder, desc->stime) < 0) return -1;
         if (tEncodeI64(pEncoder, desc->reqRid) < 0) return -1;
         if (tEncodeI8(pEncoder, desc->stableQuery) < 0) return -1;
+        if (tEncodeI8(pEncoder, desc->isSubQuery) < 0) return -1;
         if (tEncodeCStr(pEncoder, desc->fqdn) < 0) return -1;
         if (tEncodeI32(pEncoder, desc->subPlanNum) < 0) return -1;
 
@@ -291,6 +292,7 @@ static int32_t tDeserializeSClientHbReq(SDecoder *pDecoder, SClientHbReq *pReq)
           if (tDecodeI64(pDecoder, &desc.stime) < 0) return -1;
           if (tDecodeI64(pDecoder, &desc.reqRid) < 0) return -1;
           if (tDecodeI8(pDecoder, (int8_t *)&desc.stableQuery) < 0) return -1;
+          if (tDecodeI8(pDecoder, (int8_t *)&desc.isSubQuery) < 0) return -1;
           if (tDecodeCStrTo(pDecoder, desc.fqdn) < 0) return -1;
           if (tDecodeI32(pDecoder, &desc.subPlanNum) < 0) return -1;
 
@@ -6167,6 +6169,7 @@ int32_t tSerializeSCMCreateStreamReq(void *buf, int32_t bufLen, const SCMCreateS
   }
   if (tEncodeI64(&encoder, pReq->deleteMark) < 0) return -1;
   if (tEncodeI8(&encoder, pReq->igUpdate) < 0) return -1;
+  if (tEncodeI64(&encoder, pReq->lastTs) < 0) return -1;
 
   tEndEncode(&encoder);
 
@@ -6252,6 +6255,7 @@ int32_t tDeserializeSCMCreateStreamReq(void *buf, int32_t bufLen, SCMCreateStrea
 
   if (tDecodeI64(&decoder, &pReq->deleteMark) < 0) return -1;
   if (tDecodeI8(&decoder, &pReq->igUpdate) < 0) return -1;
+  if (tDecodeI64(&decoder, &pReq->lastTs) < 0) return -1;
 
   tEndDecode(&decoder);
 
@@ -6318,6 +6322,9 @@ int32_t tDeserializeSMRecoverStreamReq(void *buf, int32_t bufLen, SMRecoverStrea
 }
 
 void tFreeSCMCreateStreamReq(SCMCreateStreamReq *pReq) {
+  if (NULL == pReq) {
+    return;
+  }
   taosArrayDestroy(pReq->pTags);
   taosMemoryFreeClear(pReq->sql);
   taosMemoryFreeClear(pReq->ast);

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

@@ -265,6 +265,7 @@ int32_t vmProcessCreateVnodeReq(SVnodeMgmt *pMgmt, SRpcMsg *pMsg) {
 
   snprintf(path, TSDB_FILENAME_LEN, "vnode%svnode%d", TD_DIRSEP, vnodeCfg.vgId);
 
+#if 0
   if (pMgmt->pTfs) {
     if (tfsDirExistAt(pMgmt->pTfs, path, (SDiskID){0})) {
       terrno = TSDB_CODE_VND_DIR_ALREADY_EXIST;
@@ -278,6 +279,7 @@ int32_t vmProcessCreateVnodeReq(SVnodeMgmt *pMgmt, SRpcMsg *pMsg) {
       return -1;
     }
   }
+#endif
 
   if (vnodeCreate(path, &vnodeCfg, pMgmt->pTfs) < 0) {
     tFreeSCreateVnodeReq(&req);
@@ -712,6 +714,7 @@ SArray *vmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_ADD_CHECKINFO, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_DEL_CHECKINFO, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_CONSUME, vmPutMsgToQueryQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_CONSUME_PUSH, vmPutMsgToQueryQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_VG_WALINFO, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_DELETE, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_BATCH_DEL, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;

source/dnode/mgmt/node_mgmt/src/dmTransport.c

@@ -23,10 +23,6 @@ static inline void dmBuildMnodeRedirectRsp(SDnode *pDnode, SRpcMsg *pMsg) {
   SEpSet epSet = {0};
   dmGetMnodeEpSetForRedirect(&pDnode->data, pMsg, &epSet);
 
-  if (epSet.numOfEps == 1) {
-    return;
-  }
-
   const int32_t contLen = tSerializeSEpSet(NULL, 0, &epSet);
   pMsg->pCont = rpcMallocCont(contLen);
   if (pMsg->pCont == NULL) {

source/dnode/mgmt/node_util/inc/dmUtil.h

@@ -105,6 +105,7 @@ typedef struct {
   SHashObj      *dnodeHash;
   TdThreadRwlock lock;
   SMsgCb         msgCb;
+  bool           validMnodeEps;
 } SDnodeData;
 
 typedef struct {

source/dnode/mgmt/node_util/src/dmEps.c

@@ -288,6 +288,8 @@ static void dmResetEps(SDnodeData *pData, SArray *dnodeEps) {
     taosHashPut(pData->dnodeHash, &pDnodeEp->id, sizeof(int32_t), pDnodeEp, sizeof(SDnodeEp));
   }
 
+  pData->validMnodeEps = true;
+
   dmPrintEps(pData);
 }
 
@@ -348,6 +350,7 @@ void dmRotateMnodeEpSet(SDnodeData *pData) {
 }
 
 void dmGetMnodeEpSetForRedirect(SDnodeData *pData, SRpcMsg *pMsg, SEpSet *pEpSet) {
+  if(!pData->validMnodeEps) return;
   dmGetMnodeEpSet(pData, pEpSet);
   dTrace("msg is redirected, handle:%p num:%d use:%d", pMsg->info.handle, pEpSet->numOfEps, pEpSet->inUse);
   for (int32_t i = 0; i < pEpSet->numOfEps; ++i) {

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

@@ -553,7 +553,6 @@ typedef struct {
   int64_t rebalanceTime;
 
   int8_t         withTbName;
-  int8_t         useSnapshot;
   int8_t         autoCommit;
   int32_t        autoCommitInterval;
   int32_t        resetOffsetCfg;
@@ -566,14 +565,14 @@ void*           tDecodeSMqConsumerObj(const void* buf, SMqConsumerObj* pConsumer
 
 typedef struct {
   int32_t vgId;
-  char*   qmsg;  // SubPlanToString
+//  char*   qmsg;  // SubPlanToString
   SEpSet  epSet;
 } SMqVgEp;
 
 SMqVgEp* tCloneSMqVgEp(const SMqVgEp* pVgEp);
 void     tDeleteSMqVgEp(SMqVgEp* pVgEp);
 int32_t  tEncodeSMqVgEp(void** buf, const SMqVgEp* pVgEp);
-void*    tDecodeSMqVgEp(const void* buf, SMqVgEp* pVgEp);
+void*    tDecodeSMqVgEp(const void* buf, SMqVgEp* pVgEp, int8_t sver);
 
 typedef struct {
   int64_t consumerId;  // -1 for unassigned
@@ -598,6 +597,7 @@ typedef struct {
   SArray*   unassignedVgs;  // SArray<SMqVgEp*>
   SArray*   offsetRows;
   char      dbName[TSDB_DB_FNAME_LEN];
+  char*     qmsg;           // SubPlanToString
 } SMqSubscribeObj;
 
 SMqSubscribeObj* tNewSubscribeObj(const char key[TSDB_SUBSCRIBE_KEY_LEN]);
@@ -696,12 +696,12 @@ int32_t tEncodeSStreamObj(SEncoder* pEncoder, const SStreamObj* pObj);
 int32_t tDecodeSStreamObj(SDecoder* pDecoder, SStreamObj* pObj, int32_t sver);
 void    tFreeStreamObj(SStreamObj* pObj);
 
-typedef struct {
+//typedef struct {
-  char    streamName[TSDB_STREAM_FNAME_LEN];
+//  char    streamName[TSDB_STREAM_FNAME_LEN];
-  int64_t uid;
+//  int64_t uid;
-  int64_t streamUid;
+//  int64_t streamUid;
-  SArray* childInfo;  // SArray<SStreamChildEpInfo>
+//  SArray* childInfo;  // SArray<SStreamChildEpInfo>
-} SStreamCheckpointObj;
+//} SStreamCheckpointObj;
 
 #ifdef __cplusplus
 }

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

@@ -665,7 +665,7 @@ int32_t mndProcessSubscribeReq(SRpcMsg *pMsg) {
   SCMSubscribeReq subscribe = {0};
   tDeserializeSCMSubscribeReq(msgStr, &subscribe);
 
-  uint64_t        consumerId = subscribe.consumerId;
+  int64_t        consumerId = subscribe.consumerId;
   char           *cgroup = subscribe.cgroup;
   SMqConsumerObj *pExistedConsumer = NULL;
   SMqConsumerObj *pConsumerNew = NULL;
@@ -697,7 +697,6 @@ int32_t mndProcessSubscribeReq(SRpcMsg *pMsg) {
     tstrncpy(pConsumerNew->clientId, subscribe.clientId, tListLen(pConsumerNew->clientId));
 
     pConsumerNew->withTbName = subscribe.withTbName;
-    pConsumerNew->useSnapshot = subscribe.useSnapshot;
     pConsumerNew->autoCommit = subscribe.autoCommit;
     pConsumerNew->autoCommitInterval = subscribe.autoCommitInterval;
     pConsumerNew->resetOffsetCfg = subscribe.resetOffsetCfg;
@@ -1186,25 +1185,16 @@ static int32_t mndRetrieveConsumer(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *
       pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
       colDataSetVal(pColInfo, numOfRows, (const char *)&pConsumer->rebalanceTime, pConsumer->rebalanceTime == 0);
 
-      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+      char buf[TSDB_OFFSET_LEN] = {0};
-      colDataSetVal(pColInfo, numOfRows, (const char *)&pConsumer->withTbName, false);
-
-      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
-      colDataSetVal(pColInfo, numOfRows, (const char *)&pConsumer->useSnapshot, false);
-
-      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
-      colDataSetVal(pColInfo, numOfRows, (const char *)&pConsumer->autoCommit, false);
-
-      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
-      colDataSetVal(pColInfo, numOfRows, (const char *)&pConsumer->autoCommitInterval, false);
-
-      char buf[TSDB_OFFSET_LEN + VARSTR_HEADER_SIZE] = {0};
       STqOffsetVal pVal = {.type = pConsumer->resetOffsetCfg};
-      tFormatOffset(varDataVal(buf), TSDB_OFFSET_LEN, &pVal);
+      tFormatOffset(buf, TSDB_OFFSET_LEN, &pVal);
-      varDataSetLen(buf, strlen(varDataVal(buf)));
+
+      char parasStr[64 + TSDB_OFFSET_LEN + VARSTR_HEADER_SIZE] = {0};
+      sprintf(varDataVal(parasStr), "tbname:%d,commit:%d,interval:%d,reset:%s", pConsumer->withTbName, pConsumer->autoCommit, pConsumer->autoCommitInterval, buf);
+      varDataSetLen(parasStr, strlen(varDataVal(parasStr)));
 
       pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
-      colDataSetVal(pColInfo, numOfRows, (const char *)buf, false);
+      colDataSetVal(pColInfo, numOfRows, (const char *)parasStr, false);
 
       numOfRows++;
     }

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

@@ -187,14 +187,14 @@ SMqVgEp *tCloneSMqVgEp(const SMqVgEp *pVgEp) {
   SMqVgEp *pVgEpNew = taosMemoryMalloc(sizeof(SMqVgEp));
   if (pVgEpNew == NULL) return NULL;
   pVgEpNew->vgId = pVgEp->vgId;
-  pVgEpNew->qmsg = taosStrdup(pVgEp->qmsg);
+//  pVgEpNew->qmsg = taosStrdup(pVgEp->qmsg);
   pVgEpNew->epSet = pVgEp->epSet;
   return pVgEpNew;
 }
 
 void tDeleteSMqVgEp(SMqVgEp *pVgEp) {
   if (pVgEp) {
-    taosMemoryFreeClear(pVgEp->qmsg);
+//    taosMemoryFreeClear(pVgEp->qmsg);
     taosMemoryFree(pVgEp);
   }
 }
@@ -202,14 +202,18 @@ void tDeleteSMqVgEp(SMqVgEp *pVgEp) {
 int32_t tEncodeSMqVgEp(void **buf, const SMqVgEp *pVgEp) {
   int32_t tlen = 0;
   tlen += taosEncodeFixedI32(buf, pVgEp->vgId);
-  tlen += taosEncodeString(buf, pVgEp->qmsg);
+//  tlen += taosEncodeString(buf, pVgEp->qmsg);
   tlen += taosEncodeSEpSet(buf, &pVgEp->epSet);
   return tlen;
 }
 
-void *tDecodeSMqVgEp(const void *buf, SMqVgEp *pVgEp) {
+void *tDecodeSMqVgEp(const void *buf, SMqVgEp *pVgEp, int8_t sver) {
   buf = taosDecodeFixedI32(buf, &pVgEp->vgId);
-  buf = taosDecodeString(buf, &pVgEp->qmsg);
+  if(sver == 1){
+    uint64_t size = 0;
+    buf = taosDecodeVariantU64(buf, &size);
+    buf = POINTER_SHIFT(buf, size);
+  }
   buf = taosDecodeSEpSet(buf, &pVgEp->epSet);
   return (void *)buf;
 }
@@ -322,7 +326,6 @@ int32_t tEncodeSMqConsumerObj(void **buf, const SMqConsumerObj *pConsumer) {
   }
 
   tlen += taosEncodeFixedI8(buf, pConsumer->withTbName);
-  tlen += taosEncodeFixedI8(buf, pConsumer->useSnapshot);
   tlen += taosEncodeFixedI8(buf, pConsumer->autoCommit);
   tlen += taosEncodeFixedI32(buf, pConsumer->autoCommitInterval);
   tlen += taosEncodeFixedI32(buf, pConsumer->resetOffsetCfg);
@@ -382,12 +385,10 @@ void *tDecodeSMqConsumerObj(const void *buf, SMqConsumerObj *pConsumer, int8_t s
 
   if(sver > 1){
     buf = taosDecodeFixedI8(buf, &pConsumer->withTbName);
-    buf = taosDecodeFixedI8(buf, &pConsumer->useSnapshot);
     buf = taosDecodeFixedI8(buf, &pConsumer->autoCommit);
     buf = taosDecodeFixedI32(buf, &pConsumer->autoCommitInterval);
     buf = taosDecodeFixedI32(buf, &pConsumer->resetOffsetCfg);
   }
-
   return (void *)buf;
 }
 
@@ -395,13 +396,13 @@ void *tDecodeSMqConsumerObj(const void *buf, SMqConsumerObj *pConsumer, int8_t s
 //  SMqConsumerEp *pConsumerEpNew = taosMemoryMalloc(sizeof(SMqConsumerEp));
 //  if (pConsumerEpNew == NULL) return NULL;
 //  pConsumerEpNew->consumerId = pConsumerEpOld->consumerId;
-//  pConsumerEpNew->vgs = taosArrayDup(pConsumerEpOld->vgs, (__array_item_dup_fn_t)tCloneSMqVgEp);
+//  pConsumerEpNew->vgs = taosArrayDup(pConsumerEpOld->vgs, NULL);
 //  return pConsumerEpNew;
 //}
 //
 //void tDeleteSMqConsumerEp(void *data) {
 //  SMqConsumerEp *pConsumerEp = (SMqConsumerEp *)data;
-//  taosArrayDestroyP(pConsumerEp->vgs, (FDelete)tDeleteSMqVgEp);
+//  taosArrayDestroy(pConsumerEp->vgs);
 //}
 
 int32_t tEncodeSMqConsumerEp(void **buf, const SMqConsumerEp *pConsumerEp) {
@@ -437,7 +438,7 @@ int32_t tEncodeSMqConsumerEp(void **buf, const SMqConsumerEp *pConsumerEp) {
 
 void *tDecodeSMqConsumerEp(const void *buf, SMqConsumerEp *pConsumerEp, int8_t sver) {
   buf = taosDecodeFixedI64(buf, &pConsumerEp->consumerId);
-  buf = taosDecodeArray(buf, &pConsumerEp->vgs, (FDecode)tDecodeSMqVgEp, sizeof(SMqVgEp));
+  buf = taosDecodeArray(buf, &pConsumerEp->vgs, (FDecode)tDecodeSMqVgEp, sizeof(SMqVgEp), sver);
   if (sver > 1){
     int32_t szVgs = 0;
     buf = taosDecodeFixedI32(buf, &szVgs);
@@ -521,6 +522,7 @@ SMqSubscribeObj *tCloneSubscribeObj(const SMqSubscribeObj *pSub) {
   pSubNew->unassignedVgs = taosArrayDup(pSub->unassignedVgs, (__array_item_dup_fn_t)tCloneSMqVgEp);
   pSubNew->offsetRows = taosArrayDup(pSub->offsetRows, NULL);
   memcpy(pSubNew->dbName, pSub->dbName, TSDB_DB_FNAME_LEN);
+  pSubNew->qmsg = taosStrdup(pSub->qmsg);
   return pSubNew;
 }
 
@@ -535,6 +537,7 @@ void tDeleteSubscribeObj(SMqSubscribeObj *pSub) {
   }
   taosHashCleanup(pSub->consumerHash);
   taosArrayDestroyP(pSub->unassignedVgs, (FDelete)tDeleteSMqVgEp);
+  taosMemoryFreeClear(pSub->qmsg);
   taosArrayDestroy(pSub->offsetRows);
 }
 
@@ -579,6 +582,7 @@ int32_t tEncodeSubscribeObj(void **buf, const SMqSubscribeObj *pSub) {
       // do nothing
     }
   }
+  tlen += taosEncodeString(buf, pSub->qmsg);
   return tlen;
 }
 
@@ -601,7 +605,7 @@ void *tDecodeSubscribeObj(const void *buf, SMqSubscribeObj *pSub, int8_t sver) {
     taosHashPut(pSub->consumerHash, &consumerEp.consumerId, sizeof(int64_t), &consumerEp, sizeof(SMqConsumerEp));
   }
 
-  buf = taosDecodeArray(buf, &pSub->unassignedVgs, (FDecode)tDecodeSMqVgEp, sizeof(SMqVgEp));
+  buf = taosDecodeArray(buf, &pSub->unassignedVgs, (FDecode)tDecodeSMqVgEp, sizeof(SMqVgEp), sver);
   buf = taosDecodeStringTo(buf, pSub->dbName);
 
   if (sver > 1){
@@ -625,6 +629,7 @@ void *tDecodeSubscribeObj(const void *buf, SMqSubscribeObj *pSub, int8_t sver) {
         }
       }
     }
+    buf = taosDecodeString(buf, &pSub->qmsg);
   }
   return (void *)buf;
 }

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

@@ -227,6 +227,7 @@ static int32_t mndProcessConnectReq(SRpcMsg *pReq) {
   }
 
   if ((code = taosCheckVersionCompatibleFromStr(connReq.sVer, version, 3)) != 0) {
+    mGError("version not compatible. client version: %s, server version: %s", connReq.sVer, version);
     terrno = code;
     goto _OVER;
   }
@@ -834,6 +835,9 @@ static int32_t mndRetrieveQueries(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *p
       pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
       colDataSetVal(pColInfo, numOfRows, (const char *)&pQuery->stableQuery, false);
 
+      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+      colDataSetVal(pColInfo, numOfRows, (const char *)&pQuery->isSubQuery, false);
+
       pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
       colDataSetVal(pColInfo, numOfRows, (const char *)&pQuery->subPlanNum, false);
 

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

@@ -570,23 +570,19 @@ int32_t mndSchedInitSubEp(SMnode* pMnode, const SMqTopicObj* pTopic, SMqSubscrib
 
     mDebug("init subscription %s for topic:%s assign vgId:%d", pSub->key, pTopic->name, pVgEp->vgId);
 
+    sdbRelease(pSdb, pVgroup);
+  }
+
   if (pSubplan) {
     int32_t msgLen;
 
-      pSubplan->execNode.epSet = pVgEp->epSet;
+    if (qSubPlanToString(pSubplan, &pSub->qmsg, &msgLen) < 0) {
-      pSubplan->execNode.nodeId = pVgEp->vgId;
-
-      if (qSubPlanToString(pSubplan, &pVgEp->qmsg, &msgLen) < 0) {
-        sdbRelease(pSdb, pVgroup);
       qDestroyQueryPlan(pPlan);
       terrno = TSDB_CODE_QRY_INVALID_INPUT;
       return -1;
     }
   } else {
-      pVgEp->qmsg = taosStrdup("");
+    pSub->qmsg = taosStrdup("");
-    }
-
-    sdbRelease(pSdb, pVgroup);
   }
 
   qDestroyQueryPlan(pPlan);

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

@@ -3159,8 +3159,14 @@ static int32_t mndRetrieveStbCol(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pB
   SSdb    *pSdb = pMnode->pSdb;
   SStbObj *pStb = NULL;
 
-  int32_t numOfRows = buildSysDbColsInfo(pBlock, pShow->db, pShow->filterTb);
+
+  int32_t numOfRows = 0;
+  if (!pShow->sysDbRsp) {
+    numOfRows = buildSysDbColsInfo(pBlock, pShow->db, pShow->filterTb);
     mDebug("mndRetrieveStbCol get system table cols, rows:%d, db:%s", numOfRows, pShow->db);
+    pShow->sysDbRsp = true;
+  }
+
   SDbObj *pDb = NULL;
   if (strlen(pShow->db) > 0) {
     pDb = mndAcquireDb(pMnode, pShow->db);

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

@@ -99,13 +99,23 @@ static SMqSubscribeObj *mndCreateSubscription(SMnode *pMnode, const SMqTopicObj
   return pSub;
 }
 
-static int32_t mndBuildSubChangeReq(void **pBuf, int32_t *pLen, const SMqSubscribeObj *pSub,
+static int32_t mndBuildSubChangeReq(void **pBuf, int32_t *pLen, SMqSubscribeObj *pSub,
-                                    const SMqRebOutputVg *pRebVg) {
+                                    const SMqRebOutputVg *pRebVg, SSubplan* pPlan) {
   SMqRebVgReq req = {0};
   req.oldConsumerId = pRebVg->oldConsumerId;
   req.newConsumerId = pRebVg->newConsumerId;
   req.vgId = pRebVg->pVgEp->vgId;
-  req.qmsg = pRebVg->pVgEp->qmsg;
+  if(pPlan){
+    pPlan->execNode.epSet = pRebVg->pVgEp->epSet;
+    pPlan->execNode.nodeId = pRebVg->pVgEp->vgId;
+    int32_t msgLen;
+    if (qSubPlanToString(pPlan, &req.qmsg, &msgLen) < 0) {
+      terrno = TSDB_CODE_QRY_INVALID_INPUT;
+      return -1;
+    }
+  }else{
+    req.qmsg = taosStrdup("");
+  }
   req.subType = pSub->subType;
   req.withMeta = pSub->withMeta;
   req.suid = pSub->stbUid;
@@ -115,6 +125,7 @@ static int32_t mndBuildSubChangeReq(void **pBuf, int32_t *pLen, const SMqSubscri
   int32_t ret = 0;
   tEncodeSize(tEncodeSMqRebVgReq, &req, tlen, ret);
   if (ret < 0) {
+    taosMemoryFree(req.qmsg);
     return -1;
   }
 
@@ -122,6 +133,7 @@ static int32_t mndBuildSubChangeReq(void **pBuf, int32_t *pLen, const SMqSubscri
   void   *buf = taosMemoryMalloc(tlen);
   if (buf == NULL) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
+    taosMemoryFree(req.qmsg);
     return -1;
   }
 
@@ -135,17 +147,19 @@ static int32_t mndBuildSubChangeReq(void **pBuf, int32_t *pLen, const SMqSubscri
   if (tEncodeSMqRebVgReq(&encoder, &req) < 0) {
     taosMemoryFreeClear(buf);
     tEncoderClear(&encoder);
+    taosMemoryFree(req.qmsg);
     return -1;
   }
   tEncoderClear(&encoder);
   *pBuf = buf;
   *pLen = tlen;
 
+  taosMemoryFree(req.qmsg);
   return 0;
 }
 
-static int32_t mndPersistSubChangeVgReq(SMnode *pMnode, STrans *pTrans, const SMqSubscribeObj *pSub,
+static int32_t mndPersistSubChangeVgReq(SMnode *pMnode, STrans *pTrans, SMqSubscribeObj *pSub,
-                                        const SMqRebOutputVg *pRebVg) {
+                                        const SMqRebOutputVg *pRebVg, SSubplan* pPlan) {
 //  if (pRebVg->oldConsumerId == pRebVg->newConsumerId) {
 //    terrno = TSDB_CODE_MND_INVALID_SUB_OPTION;
 //    return -1;
@@ -153,7 +167,7 @@ static int32_t mndPersistSubChangeVgReq(SMnode *pMnode, STrans *pTrans, const SM
 
   void   *buf;
   int32_t tlen;
-  if (mndBuildSubChangeReq(&buf, &tlen, pSub, pRebVg) < 0) {
+  if (mndBuildSubChangeReq(&buf, &tlen, pSub, pRebVg, pPlan) < 0) {
     return -1;
   }
 
@@ -519,14 +533,25 @@ static int32_t mndDoRebalance(SMnode *pMnode, const SMqRebInputObj *pInput, SMqR
 }
 
 static int32_t mndPersistRebResult(SMnode *pMnode, SRpcMsg *pMsg, const SMqRebOutputObj *pOutput) {
+  struct SSubplan* pPlan = NULL;
+  if(strcmp(pOutput->pSub->qmsg, "") != 0){
+    int32_t code = qStringToSubplan(pOutput->pSub->qmsg, &pPlan);
+    if (code != TSDB_CODE_SUCCESS) {
+      terrno = code;
+      return -1;
+    }
+  }
+
   STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_DB_INSIDE, pMsg, "tmq-reb");
   if (pTrans == NULL) {
+    nodesDestroyNode((SNode*)pPlan);
     return -1;
   }
 
   mndTransSetDbName(pTrans, pOutput->pSub->dbName, NULL);
   if (mndTransCheckConflict(pMnode, pTrans) != 0) {
     mndTransDrop(pTrans);
+    nodesDestroyNode((SNode*)pPlan);
     return -1;
   }
 
@@ -536,11 +561,13 @@ static int32_t mndPersistRebResult(SMnode *pMnode, SRpcMsg *pMsg, const SMqRebOu
   int32_t       vgNum = taosArrayGetSize(rebVgs);
   for (int32_t i = 0; i < vgNum; i++) {
     SMqRebOutputVg *pRebVg = taosArrayGet(rebVgs, i);
-    if (mndPersistSubChangeVgReq(pMnode, pTrans, pOutput->pSub, pRebVg) < 0) {
+    if (mndPersistSubChangeVgReq(pMnode, pTrans, pOutput->pSub, pRebVg, pPlan) < 0) {
       mndTransDrop(pTrans);
+      nodesDestroyNode((SNode*)pPlan);
       return -1;
     }
   }
+  nodesDestroyNode((SNode*)pPlan);
 
   // 2. redo log: subscribe and vg assignment
   // subscribe

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

@@ -912,12 +912,14 @@ static int32_t mndRetrieveTopic(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBl
     pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
     colDataSetVal(pColInfo, numOfRows, (const char *)&pTopic->createTime, false);
 
-    char sql[TSDB_SHOW_SQL_LEN + VARSTR_HEADER_SIZE] = {0};
+    char *sql = taosMemoryMalloc(strlen(pTopic->sql) + VARSTR_HEADER_SIZE);
     STR_TO_VARSTR(sql, pTopic->sql);
 
     pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
     colDataSetVal(pColInfo, numOfRows, (const char *)sql, false);
 
+    taosMemoryFree(sql);
+
     char *schemaJson = taosMemoryMalloc(TSDB_SHOW_SCHEMA_JSON_LEN + VARSTR_HEADER_SIZE);
     if(pTopic->subType == TOPIC_SUB_TYPE__COLUMN){
       schemaToJson(pTopic->schema.pSchema, pTopic->schema.nCols, schemaJson);

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

@@ -139,6 +139,7 @@ static STqMgmt tqMgmt = {0};
 
 int32_t tEncodeSTqHandle(SEncoder* pEncoder, const STqHandle* pHandle);
 int32_t tDecodeSTqHandle(SDecoder* pDecoder, STqHandle* pHandle);
+void    tqDestroyTqHandle(void* data);
 
 // tqRead
 int32_t tqScanTaosx(STQ* pTq, const STqHandle* pHandle, STaosxRsp* pRsp, SMqMetaRsp* pMetaRsp, STqOffsetVal* offset);
@@ -161,6 +162,8 @@ int32_t tqMetaRestoreHandle(STQ* pTq);
 int32_t tqMetaSaveCheckInfo(STQ* pTq, const char* key, const void* value, int32_t vLen);
 int32_t tqMetaDeleteCheckInfo(STQ* pTq, const char* key);
 int32_t tqMetaRestoreCheckInfo(STQ* pTq);
+int32_t tqMetaGetHandle(STQ* pTq, const char* key);
+int32_t tqCreateHandle(STQ* pTq, SMqRebVgReq* req, STqHandle* handle);
 
 STqOffsetStore* tqOffsetOpen(STQ* pTq);
 void            tqOffsetClose(STqOffsetStore*);

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

@@ -229,6 +229,7 @@ int32_t tqProcessDeleteSubReq(STQ* pTq, int64_t version, char* msg, int32_t msgL
 int32_t tqProcessOffsetCommitReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
 int32_t tqProcessSeekReq(STQ* pTq, int64_t sversion, char* msg, int32_t msgLen);
 int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg);
+int32_t tqProcessPollPush(STQ* pTq, SRpcMsg* pMsg);
 int32_t tqProcessVgWalInfoReq(STQ* pTq, SRpcMsg* pMsg);
 
 // tq-stream

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

@@ -62,7 +62,7 @@ void tqCleanUp() {
   }
 }
 
-static void destroyTqHandle(void* data) {
+void tqDestroyTqHandle(void* data) {
   STqHandle* pData = (STqHandle*)data;
   qDestroyTask(pData->execHandle.task);
 
@@ -102,7 +102,7 @@ STQ* tqOpen(const char* path, SVnode* pVnode) {
   pTq->walLogLastVer = pVnode->pWal->vers.lastVer;
 
   pTq->pHandle = taosHashInit(64, MurmurHash3_32, true, HASH_ENTRY_LOCK);
-  taosHashSetFreeFp(pTq->pHandle, destroyTqHandle);
+  taosHashSetFreeFp(pTq->pHandle, tqDestroyTqHandle);
 
   taosInitRWLatch(&pTq->lock);
   pTq->pPushMgr = taosHashInit(64, MurmurHash3_32, false, HASH_NO_LOCK);
@@ -264,7 +264,7 @@ int32_t tqSendDataRsp(STqHandle* pHandle, const SRpcMsg* pMsg, const SMqPollReq*
   tFormatOffset(buf1, TSDB_OFFSET_LEN, &pRsp->reqOffset);
   tFormatOffset(buf2, TSDB_OFFSET_LEN, &pRsp->rspOffset);
 
-  tqDebug("vgId:%d consumer:0x%" PRIx64 " (epoch %d) send rsp, block num:%d, req:%s, rsp:%s, reqId:0x%" PRIx64, vgId,
+  tqDebug("tmq poll vgId:%d consumer:0x%" PRIx64 " (epoch %d) send rsp, block num:%d, req:%s, rsp:%s, reqId:0x%" PRIx64, vgId,
           pReq->consumerId, pReq->epoch, pRsp->blockNum, buf1, buf2, pReq->reqId);
 
   return 0;
@@ -421,6 +421,35 @@ int32_t tqCheckColModifiable(STQ* pTq, int64_t tbUid, int32_t colId) {
   return 0;
 }
 
+int32_t tqProcessPollPush(STQ* pTq, SRpcMsg* pMsg) {
+  int32_t vgId = TD_VID(pTq->pVnode);
+  taosWLockLatch(&pTq->lock);
+  if (taosHashGetSize(pTq->pPushMgr) > 0) {
+    void* pIter = taosHashIterate(pTq->pPushMgr, NULL);
+
+    while (pIter) {
+      STqHandle* pHandle = *(STqHandle**)pIter;
+      tqDebug("vgId:%d start set submit for pHandle:%p, consumer:0x%" PRIx64, vgId, pHandle, pHandle->consumerId);
+
+      if (ASSERT(pHandle->msg != NULL)) {
+        tqError("pHandle->msg should not be null");
+        break;
+      }else{
+        SRpcMsg msg = {.msgType = TDMT_VND_TMQ_CONSUME, .pCont = pHandle->msg->pCont, .contLen = pHandle->msg->contLen, .info = pHandle->msg->info};
+        tmsgPutToQueue(&pTq->pVnode->msgCb, QUERY_QUEUE, &msg);
+        taosMemoryFree(pHandle->msg);
+        pHandle->msg = NULL;
+      }
+
+      pIter = taosHashIterate(pTq->pPushMgr, pIter);
+    }
+
+    taosHashClear(pTq->pPushMgr);
+  }
+  taosWUnLockLatch(&pTq->lock);
+  return 0;
+}
+
 int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg) {
   SMqPollReq req = {0};
   int        code = 0;
@@ -661,13 +690,17 @@ int32_t tqProcessSubscribeReq(STQ* pTq, int64_t sversion, char* msg, int32_t msg
     return -1;
   }
 
-  SVnode* pVnode = pTq->pVnode;
+  tqDebug("vgId:%d, tq process sub req:%s, Id:0x%" PRIx64 " -> Id:0x%" PRIx64, pTq->pVnode->config.vgId, req.subKey,
-  int32_t vgId = TD_VID(pVnode);
-
-  tqDebug("vgId:%d, tq process sub req:%s, Id:0x%" PRIx64 " -> Id:0x%" PRIx64, pVnode->config.vgId, req.subKey,
           req.oldConsumerId, req.newConsumerId);
 
-  STqHandle* pHandle = taosHashGet(pTq->pHandle, req.subKey, strlen(req.subKey));
+  STqHandle* pHandle = NULL;
+  while(1){
+    pHandle = taosHashGet(pTq->pHandle, req.subKey, strlen(req.subKey));
+    if (pHandle || tqMetaGetHandle(pTq, req.subKey) < 0){
+      break;
+    }
+  }
+
   if (pHandle == NULL) {
     if (req.oldConsumerId != -1) {
       tqError("vgId:%d, build new consumer handle %s for consumer:0x%" PRIx64 ", but old consumerId:0x%" PRIx64,
@@ -678,86 +711,13 @@ int32_t tqProcessSubscribeReq(STQ* pTq, int64_t sversion, char* msg, int32_t msg
       tqError("vgId:%d, tq invalid re-balance request, new consumerId %" PRId64 "", req.vgId, req.newConsumerId);
       goto end;
     }
-
+    STqHandle handle = {0};
-    STqHandle tqHandle = {0};
+    ret = tqCreateHandle(pTq, &req, &handle);
-    pHandle = &tqHandle;
+    if(ret < 0){
-
+      tqDestroyTqHandle(&handle);
-    memcpy(pHandle->subKey, req.subKey, TSDB_SUBSCRIBE_KEY_LEN);
-    pHandle->consumerId = req.newConsumerId;
-    pHandle->epoch = -1;
-
-    pHandle->execHandle.subType = req.subType;
-    pHandle->fetchMeta = req.withMeta;
-
-    // TODO version should be assigned and refed during preprocess
-    SWalRef* pRef = walRefCommittedVer(pVnode->pWal);
-    if (pRef == NULL) {
-      ret = -1;
       goto end;
     }
-
+    ret = tqMetaSaveHandle(pTq, req.subKey, &handle);
-    int64_t ver = pRef->refVer;
-    pHandle->pRef = pRef;
-
-    SReadHandle handle = {.vnode = pVnode, .initTableReader = true, .initTqReader = true, .version = ver};
-    initStorageAPI(&handle.api);
-
-    pHandle->snapshotVer = ver;
-
-    if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
-      pHandle->execHandle.execCol.qmsg = taosStrdup(req.qmsg);
-
-      pHandle->execHandle.task = qCreateQueueExecTaskInfo(pHandle->execHandle.execCol.qmsg, &handle, vgId,
-                                                          &pHandle->execHandle.numOfCols, req.newConsumerId);
-      void* scanner = NULL;
-      qExtractStreamScanner(pHandle->execHandle.task, &scanner);
-      pHandle->execHandle.pTqReader = qExtractReaderFromStreamScanner(scanner);
-    } else if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__DB) {
-      pHandle->pWalReader = walOpenReader(pVnode->pWal, NULL);
-      pHandle->execHandle.pTqReader = tqReaderOpen(pVnode);
-
-      pHandle->execHandle.execDb.pFilterOutTbUid =
-          taosHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), false, HASH_ENTRY_LOCK);
-      buildSnapContext(handle.vnode, handle.version, 0, pHandle->execHandle.subType, pHandle->fetchMeta,
-                       (SSnapContext**)(&handle.sContext));
-
-      pHandle->execHandle.task = qCreateQueueExecTaskInfo(NULL, &handle, vgId, NULL, req.newConsumerId);
-    } else if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__TABLE) {
-      pHandle->pWalReader = walOpenReader(pVnode->pWal, NULL);
-      pHandle->execHandle.execTb.suid = req.suid;
-      pHandle->execHandle.execTb.qmsg = taosStrdup(req.qmsg);
-
-      if (strcmp(pHandle->execHandle.execTb.qmsg, "") != 0) {
-        if (nodesStringToNode(pHandle->execHandle.execTb.qmsg, &pHandle->execHandle.execTb.node) != 0) {
-          tqError("nodesStringToNode error in sub stable, since %s, vgId:%d, subkey:%s consumer:0x%" PRIx64, terrstr(),
-                  pVnode->config.vgId, req.subKey, pHandle->consumerId);
-          return -1;
-        }
-      }
-
-      buildSnapContext(handle.vnode, handle.version, req.suid, pHandle->execHandle.subType, pHandle->fetchMeta,
-                       (SSnapContext**)(&handle.sContext));
-      pHandle->execHandle.task = qCreateQueueExecTaskInfo(NULL, &handle, vgId, NULL, req.newConsumerId);
-
-      SArray* tbUidList = NULL;
-      ret = qGetTableList(req.suid, pVnode, pHandle->execHandle.execTb.node, &tbUidList, pHandle->execHandle.task);
-      if (ret != TDB_CODE_SUCCESS) {
-        tqError("qGetTableList error:%d vgId:%d, subkey:%s consumer:0x%" PRIx64, ret, pVnode->config.vgId, req.subKey,
-                pHandle->consumerId);
-        taosArrayDestroy(tbUidList);
-        goto end;
-      }
-      tqDebug("tq try to get ctb for stb subscribe, vgId:%d, subkey:%s consumer:0x%" PRIx64 " suid:%" PRId64,
-              pVnode->config.vgId, req.subKey, pHandle->consumerId, req.suid);
-      pHandle->execHandle.pTqReader = tqReaderOpen(pVnode);
-      tqReaderSetTbUidList(pHandle->execHandle.pTqReader, tbUidList, NULL);
-      taosArrayDestroy(tbUidList);
-    }
-
-    taosHashPut(pTq->pHandle, req.subKey, strlen(req.subKey), pHandle, sizeof(STqHandle));
-    tqDebug("try to persist handle %s consumer:0x%" PRIx64, req.subKey, pHandle->consumerId);
-    ret = tqMetaSaveHandle(pTq, req.subKey, pHandle);
-    goto end;
   } else {
     taosWLockLatch(&pTq->lock);
 

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

@@ -88,9 +88,9 @@ int32_t tqMetaOpen(STQ* pTq) {
     return -1;
   }
 
-  if (tqMetaRestoreHandle(pTq) < 0) {
+//  if (tqMetaRestoreHandle(pTq) < 0) {
-    return -1;
+//    return -1;
-  }
+//  }
 
   if (tqMetaRestoreCheckInfo(pTq) < 0) {
     return -1;
@@ -274,6 +274,120 @@ int32_t tqMetaDeleteHandle(STQ* pTq, const char* key) {
   return 0;
 }
 
+static int buildHandle(STQ* pTq, STqHandle* handle){
+  SVnode* pVnode = pTq->pVnode;
+  int32_t vgId = TD_VID(pVnode);
+
+  handle->pRef = walOpenRef(pVnode->pWal);
+  if (handle->pRef == NULL) {
+    return -1;
+  }
+  walSetRefVer(handle->pRef, handle->snapshotVer);
+
+  SReadHandle reader = {
+      .vnode = pVnode,
+      .initTableReader = true,
+      .initTqReader = true,
+      .version = handle->snapshotVer,
+  };
+
+  initStorageAPI(&reader.api);
+
+  if (handle->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
+    handle->execHandle.task =
+        qCreateQueueExecTaskInfo(handle->execHandle.execCol.qmsg, &reader, vgId, &handle->execHandle.numOfCols, handle->consumerId);
+    if (handle->execHandle.task == NULL) {
+      tqError("cannot create exec task for %s", handle->subKey);
+      return -1;
+    }
+    void* scanner = NULL;
+    qExtractStreamScanner(handle->execHandle.task, &scanner);
+    if (scanner == NULL) {
+      tqError("cannot extract stream scanner for %s", handle->subKey);
+      return -1;
+    }
+    handle->execHandle.pTqReader = qExtractReaderFromStreamScanner(scanner);
+    if (handle->execHandle.pTqReader == NULL) {
+      tqError("cannot extract exec reader for %s", handle->subKey);
+      return -1;
+    }
+  } else if (handle->execHandle.subType == TOPIC_SUB_TYPE__DB) {
+    handle->pWalReader = walOpenReader(pVnode->pWal, NULL);
+    handle->execHandle.pTqReader = tqReaderOpen(pVnode);
+
+    buildSnapContext(reader.vnode, reader.version, 0, handle->execHandle.subType, handle->fetchMeta,
+                     (SSnapContext**)(&reader.sContext));
+    handle->execHandle.task = qCreateQueueExecTaskInfo(NULL, &reader, vgId, NULL, handle->consumerId);
+  } else if (handle->execHandle.subType == TOPIC_SUB_TYPE__TABLE) {
+    handle->pWalReader = walOpenReader(pVnode->pWal, NULL);
+
+    if(handle->execHandle.execTb.qmsg != NULL && strcmp(handle->execHandle.execTb.qmsg, "") != 0) {
+      if (nodesStringToNode(handle->execHandle.execTb.qmsg, &handle->execHandle.execTb.node) != 0) {
+        tqError("nodesStringToNode error in sub stable, since %s", terrstr());
+        return -1;
+      }
+    }
+    buildSnapContext(reader.vnode, reader.version, handle->execHandle.execTb.suid, handle->execHandle.subType,
+                     handle->fetchMeta, (SSnapContext**)(&reader.sContext));
+    handle->execHandle.task = qCreateQueueExecTaskInfo(NULL, &reader, vgId, NULL, handle->consumerId);
+
+    SArray* tbUidList = NULL;
+    int ret = qGetTableList(handle->execHandle.execTb.suid, pVnode, handle->execHandle.execTb.node, &tbUidList, handle->execHandle.task);
+    if(ret != TDB_CODE_SUCCESS) {
+      tqError("qGetTableList error:%d handle %s consumer:0x%" PRIx64, ret, handle->subKey, handle->consumerId);
+      taosArrayDestroy(tbUidList);
+      return -1;
+    }
+    tqDebug("vgId:%d, tq try to get ctb for stb subscribe, suid:%" PRId64, pVnode->config.vgId, handle->execHandle.execTb.suid);
+    handle->execHandle.pTqReader = tqReaderOpen(pVnode);
+    tqReaderSetTbUidList(handle->execHandle.pTqReader, tbUidList, NULL);
+    taosArrayDestroy(tbUidList);
+  }
+  return 0;
+}
+
+static int restoreHandle(STQ* pTq, void* pVal, int vLen, STqHandle* handle){
+  int32_t  vgId = TD_VID(pTq->pVnode);
+  SDecoder decoder;
+  tDecoderInit(&decoder, (uint8_t*)pVal, vLen);
+  tDecodeSTqHandle(&decoder, handle);
+  tDecoderClear(&decoder);
+
+  if(buildHandle(pTq, handle) < 0){
+    return -1;
+  }
+  tqDebug("tq restore %s consumer %" PRId64 " vgId:%d", handle->subKey, handle->consumerId, vgId);
+  return taosHashPut(pTq->pHandle, handle->subKey, strlen(handle->subKey), handle, sizeof(STqHandle));
+}
+
+int32_t tqCreateHandle(STQ* pTq, SMqRebVgReq* req, STqHandle* handle){
+  int32_t  vgId = TD_VID(pTq->pVnode);
+
+  memcpy(handle->subKey, req->subKey, TSDB_SUBSCRIBE_KEY_LEN);
+  handle->consumerId = req->newConsumerId;
+  handle->epoch = -1;
+
+  handle->execHandle.subType = req->subType;
+  handle->fetchMeta = req->withMeta;
+  if(req->subType == TOPIC_SUB_TYPE__COLUMN){
+    handle->execHandle.execCol.qmsg = taosStrdup(req->qmsg);
+  }else if(req->subType == TOPIC_SUB_TYPE__DB){
+    handle->execHandle.execDb.pFilterOutTbUid =
+        taosHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), false, HASH_ENTRY_LOCK);
+  }else if(req->subType == TOPIC_SUB_TYPE__TABLE){
+    handle->execHandle.execTb.suid = req->suid;
+    handle->execHandle.execTb.qmsg = taosStrdup(req->qmsg);
+  }
+
+  handle->snapshotVer = walGetLastVer(pTq->pVnode->pWal);
+
+  if(buildHandle(pTq, handle) < 0){
+    return -1;
+  }
+  tqDebug("tq restore %s consumer %" PRId64 " vgId:%d", handle->subKey, handle->consumerId, vgId);
+  return taosHashPut(pTq->pHandle, handle->subKey, strlen(handle->subKey), handle, sizeof(STqHandle));
+}
+
 int32_t tqMetaRestoreHandle(STQ* pTq) {
   int  code = 0;
   TBC* pCur = NULL;
@@ -281,97 +395,40 @@ int32_t tqMetaRestoreHandle(STQ* pTq) {
     return -1;
   }
 
-  int32_t  vgId = TD_VID(pTq->pVnode);
   void*    pKey = NULL;
   int      kLen = 0;
   void*    pVal = NULL;
   int      vLen = 0;
-  SDecoder decoder;
 
   tdbTbcMoveToFirst(pCur);
 
   while (tdbTbcNext(pCur, &pKey, &kLen, &pVal, &vLen) == 0) {
     STqHandle handle = {0};
-    tDecoderInit(&decoder, (uint8_t*)pVal, vLen);
+    code = restoreHandle(pTq, pVal, vLen, &handle);
-    tDecodeSTqHandle(&decoder, &handle);
+    if (code < 0){
-    tDecoderClear(&decoder);
+      tqDestroyTqHandle(&handle);
-
+      break;
-    handle.pRef = walOpenRef(pTq->pVnode->pWal);
-    if (handle.pRef == NULL) {
-      code = -1;
-      goto end;
     }
-    walSetRefVer(handle.pRef, handle.snapshotVer);
-
-    SReadHandle reader = {
-        .vnode = pTq->pVnode,
-        .initTableReader = true,
-        .initTqReader = true,
-        .version = handle.snapshotVer
-    };
-
-    initStorageAPI(&reader.api);
-
-    if (handle.execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
-      handle.execHandle.task =
-          qCreateQueueExecTaskInfo(handle.execHandle.execCol.qmsg, &reader, vgId, &handle.execHandle.numOfCols, 0);
-      if (handle.execHandle.task == NULL) {
-        tqError("cannot create exec task for %s", handle.subKey);
-        code = -1;
-        goto end;
-      }
-      void* scanner = NULL;
-      qExtractStreamScanner(handle.execHandle.task, &scanner);
-      if (scanner == NULL) {
-        tqError("cannot extract stream scanner for %s", handle.subKey);
-        code = -1;
-        goto end;
-      }
-      handle.execHandle.pTqReader = qExtractReaderFromStreamScanner(scanner);
-      if (handle.execHandle.pTqReader == NULL) {
-        tqError("cannot extract exec reader for %s", handle.subKey);
-        code = -1;
-        goto end;
-      }
-    } else if (handle.execHandle.subType == TOPIC_SUB_TYPE__DB) {
-      handle.pWalReader = walOpenReader(pTq->pVnode->pWal, NULL);
-      handle.execHandle.pTqReader = tqReaderOpen(pTq->pVnode);
-
-      buildSnapContext(reader.vnode, reader.version, 0, handle.execHandle.subType, handle.fetchMeta,
-                       (SSnapContext**)(&reader.sContext));
-      handle.execHandle.task = qCreateQueueExecTaskInfo(NULL, &reader, vgId, NULL, 0);
-    } else if (handle.execHandle.subType == TOPIC_SUB_TYPE__TABLE) {
-      handle.pWalReader = walOpenReader(pTq->pVnode->pWal, NULL);
-
-      if(handle.execHandle.execTb.qmsg != NULL && strcmp(handle.execHandle.execTb.qmsg, "") != 0) {
-        if (nodesStringToNode(handle.execHandle.execTb.qmsg, &handle.execHandle.execTb.node) != 0) {
-          tqError("nodesStringToNode error in sub stable, since %s", terrstr());
-          return -1;
-        }
-      }
-      buildSnapContext(reader.vnode, reader.version, handle.execHandle.execTb.suid, handle.execHandle.subType,
-                       handle.fetchMeta, (SSnapContext**)(&reader.sContext));
-      handle.execHandle.task = qCreateQueueExecTaskInfo(NULL, &reader, vgId, NULL, 0);
-
-      SArray* tbUidList = NULL;
-      int ret = qGetTableList(handle.execHandle.execTb.suid, pTq->pVnode, handle.execHandle.execTb.node, &tbUidList, handle.execHandle.task);
-      if(ret != TDB_CODE_SUCCESS) {
-        tqError("qGetTableList error:%d handle %s consumer:0x%" PRIx64, ret, handle.subKey, handle.consumerId);
-        taosArrayDestroy(tbUidList);
-        goto end;
-      }
-      tqDebug("vgId:%d, tq try to get ctb for stb subscribe, suid:%" PRId64, pTq->pVnode->config.vgId, handle.execHandle.execTb.suid);
-      handle.execHandle.pTqReader = tqReaderOpen(pTq->pVnode);
-      tqReaderSetTbUidList(handle.execHandle.pTqReader, tbUidList, NULL);
-      taosArrayDestroy(tbUidList);
-    }
-    tqDebug("tq restore %s consumer %" PRId64 " vgId:%d", handle.subKey, handle.consumerId, vgId);
-    taosHashPut(pTq->pHandle, pKey, kLen, &handle, sizeof(STqHandle));
   }
 
-end:
   tdbFree(pKey);
   tdbFree(pVal);
   tdbTbcClose(pCur);
   return code;
 }
+
+int32_t tqMetaGetHandle(STQ* pTq, const char* key) {
+  void*    pVal = NULL;
+  int      vLen = 0;
+
+  if (tdbTbGet(pTq->pExecStore, key, (int)strlen(key), &pVal, &vLen) < 0) {
+    return -1;
+  }
+  STqHandle handle = {0};
+  int code = restoreHandle(pTq, pVal, vLen, &handle);
+  if (code < 0){
+    tqDestroyTqHandle(&handle);
+  }
+  tdbFree(pVal);
+  return code;
+}

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

@@ -17,35 +17,16 @@
 #include "vnd.h"
 
 int32_t tqProcessSubmitReqForSubscribe(STQ* pTq) {
-  int32_t vgId = TD_VID(pTq->pVnode);
+  if (taosHashGetSize(pTq->pPushMgr) <= 0) {
-
+    return 0;
-  taosWLockLatch(&pTq->lock);
+  }
-
+  SRpcMsg msg = {.msgType = TDMT_VND_TMQ_CONSUME_PUSH};
-  if (taosHashGetSize(pTq->pPushMgr) > 0) {
+  msg.pCont = rpcMallocCont(sizeof(SMsgHead));
-    void* pIter = taosHashIterate(pTq->pPushMgr, NULL);
+  msg.contLen = sizeof(SMsgHead);
-
+  SMsgHead *pHead = msg.pCont;
-    while (pIter) {
+  pHead->vgId = TD_VID(pTq->pVnode);
-      STqHandle* pHandle = *(STqHandle**)pIter;
+  pHead->contLen = msg.contLen;
-      tqDebug("vgId:%d start set submit for pHandle:%p, consumer:0x%" PRIx64, vgId, pHandle, pHandle->consumerId);
-
-      if (ASSERT(pHandle->msg != NULL)) {
-        tqError("pHandle->msg should not be null");
-        break;
-      }else{
-        SRpcMsg msg = {.msgType = TDMT_VND_TMQ_CONSUME, .pCont = pHandle->msg->pCont, .contLen = pHandle->msg->contLen, .info = pHandle->msg->info};
   tmsgPutToQueue(&pTq->pVnode->msgCb, QUERY_QUEUE, &msg);
-        taosMemoryFree(pHandle->msg);
-        pHandle->msg = NULL;
-      }
-
-      pIter = taosHashIterate(pTq->pPushMgr, pIter);
-    }
-
-    taosHashClear(pTq->pPushMgr);
-  }
-
-  // unlock
-  taosWUnLockLatch(&pTq->lock);
   return 0;
 }
 

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

@@ -176,13 +176,17 @@ static int32_t extractDataAndRspForNormalSubscribe(STQ* pTq, STqHandle* pHandle,
   if (terrno == TSDB_CODE_WAL_LOG_NOT_EXIST && dataRsp.blockNum == 0) {
     // lock
     taosWLockLatch(&pTq->lock);
+    int64_t ver = walGetCommittedVer(pTq->pVnode->pWal);
+    if (pOffset->version >= ver ||
+        dataRsp.rspOffset.version >= ver) {  // check if there are data again to avoid lost data
       code = tqRegisterPushHandle(pTq, pHandle, pMsg);
       taosWUnLockLatch(&pTq->lock);
-    tDeleteMqDataRsp(&dataRsp);
+      goto end;
-    return code;
+    } else {
+      taosWUnLockLatch(&pTq->lock);
+    }
   }
 
-  // NOTE: this pHandle->consumerId may have been changed already.
   code = tqSendDataRsp(pHandle, pMsg, pRequest, (SMqDataRsp*)&dataRsp, TMQ_MSG_TYPE__POLL_RSP, vgId);
 
 end : {
@@ -192,10 +196,9 @@ end : {
           " code:%d",
           consumerId, pHandle->subKey, vgId, dataRsp.blockNum, buf, pRequest->reqId, code);
   tDeleteMqDataRsp(&dataRsp);
-}
-
   return code;
   }
+}
 
 static int32_t extractDataAndRspForDbStbSubscribe(STQ* pTq, STqHandle* pHandle, const SMqPollReq* pRequest,
                                                   SRpcMsg* pMsg, STqOffsetVal* offset) {

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

@@ -691,6 +691,7 @@ static int32_t tsdbCacheLoadFromRaw(STsdb *pTsdb, tb_uid_t uid, SArray *pLastArr
                         .colVal = COL_VAL_NONE(idxKey->key.cid, pr->pSchema->columns[slotIds[i]].type)};
     if (!pLastCol) {
       pLastCol = &noneCol;
+      reallocVarData(&pLastCol->colVal);
     }
 
     taosArraySet(pLastArray, idxKey->idx, pLastCol);
@@ -2848,15 +2849,17 @@ static int32_t mergeLastCid(tb_uid_t uid, STsdb *pTsdb, SArray **ppLastArray, SC
         tsdbRowGetColVal(pRow, pTSchema, slotIds[iCol], pColVal);
 
         *pCol = (SLastCol){.ts = rowTs, .colVal = *pColVal};
-        if (IS_VAR_DATA_TYPE(pColVal->type) && pColVal->value.nData > 0) {
+        if (IS_VAR_DATA_TYPE(pColVal->type) /*&& pColVal->value.nData > 0*/) {
           pCol->colVal.value.pData = taosMemoryMalloc(pCol->colVal.value.nData);
           if (pCol->colVal.value.pData == NULL) {
             terrno = TSDB_CODE_OUT_OF_MEMORY;
             code = TSDB_CODE_OUT_OF_MEMORY;
             goto _err;
           }
+          if (pColVal->value.nData > 0) {
             memcpy(pCol->colVal.value.pData, pColVal->value.pData, pColVal->value.nData);
           }
+        }
 
         if (!COL_VAL_IS_VALUE(pColVal)) {
           if (!setNoneCol) {
@@ -3016,15 +3019,17 @@ static int32_t mergeLastRowCid(tb_uid_t uid, STsdb *pTsdb, SArray **ppLastArray,
         tsdbRowGetColVal(pRow, pTSchema, slotIds[iCol], pColVal);
 
         *pCol = (SLastCol){.ts = rowTs, .colVal = *pColVal};
-        if (IS_VAR_DATA_TYPE(pColVal->type) && pColVal->value.nData > 0) {
+        if (IS_VAR_DATA_TYPE(pColVal->type) /*&& pColVal->value.nData > 0*/) {
           pCol->colVal.value.pData = taosMemoryMalloc(pCol->colVal.value.nData);
           if (pCol->colVal.value.pData == NULL) {
             terrno = TSDB_CODE_OUT_OF_MEMORY;
             code = TSDB_CODE_OUT_OF_MEMORY;
             goto _err;
           }
+          if (pColVal->value.nData > 0) {
             memcpy(pCol->colVal.value.pData, pColVal->value.pData, pColVal->value.nData);
           }
+        }
 
         /*if (COL_VAL_IS_NONE(pColVal)) {
           if (!setNoneCol) {

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

@@ -129,6 +129,12 @@ int32_t vnodeAlterReplica(const char *path, SAlterVnodeReplicaReq *pReq, STfs *p
   return 0;
 }
 
+static int32_t vnodeVgroupIdLen(int32_t vgId) {
+  char tmp[TSDB_FILENAME_LEN];
+  sprintf(tmp, "%d", vgId);
+  return strlen(tmp);
+}
+
 int32_t vnodeRenameVgroupId(const char *srcPath, const char *dstPath, int32_t srcVgId, int32_t dstVgId, STfs *pTfs) {
   int32_t ret = tfsRename(pTfs, srcPath, dstPath);
   if (ret != 0) return ret;
@@ -154,8 +160,7 @@ int32_t vnodeRenameVgroupId(const char *srcPath, const char *dstPath, int32_t sr
 
     int32_t tsdbFileVgId = atoi(tsdbFilePrefixPos + 6);
     if (tsdbFileVgId == srcVgId) {
-      char *tsdbFileSurfixPos = strstr(tsdbFilePrefixPos, "f");
+      char *tsdbFileSurfixPos = tsdbFilePrefixPos + 6 + vnodeVgroupIdLen(srcVgId);
-      if (tsdbFileSurfixPos == NULL) continue;
 
       tsdbFilePrefixPos[6] = 0;
       snprintf(newRname, TSDB_FILENAME_LEN, "%s%d%s", oldRname, dstVgId, tsdbFileSurfixPos);

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

@@ -238,6 +238,10 @@ static int32_t vnodePreProcessSubmitMsg(SVnode *pVnode, SRpcMsg *pMsg) {
   tEndDecode(pCoder);
 
 _exit:
+  if (code) {
+    vError("vgId:%d, failed to preprocess submit request since %s, msg type:%d", TD_VID(pVnode), tstrerror(code),
+           pMsg->msgType);
+  }
   tDecoderClear(pCoder);
   return code;
 }
@@ -297,7 +301,7 @@ int32_t vnodePreProcessWriteMsg(SVnode *pVnode, SRpcMsg *pMsg) {
 
 _exit:
   if (code) {
-    vError("vgId%d failed to preprocess write request since %s, msg type:%d", TD_VID(pVnode), tstrerror(code),
+    vError("vgId:%d, failed to preprocess write request since %s, msg type:%d", TD_VID(pVnode), tstrerror(code),
            pMsg->msgType);
   }
   return code;
@@ -505,7 +509,7 @@ int32_t vnodePreprocessQueryMsg(SVnode *pVnode, SRpcMsg *pMsg) {
 
 int32_t vnodeProcessQueryMsg(SVnode *pVnode, SRpcMsg *pMsg) {
   vTrace("message in vnode query queue is processing");
-  if ((pMsg->msgType == TDMT_SCH_QUERY || pMsg->msgType == TDMT_VND_TMQ_CONSUME) && !syncIsReadyForRead(pVnode->sync)) {
+  if ((pMsg->msgType == TDMT_SCH_QUERY || pMsg->msgType == TDMT_VND_TMQ_CONSUME || pMsg->msgType == TDMT_VND_TMQ_CONSUME_PUSH) && !syncIsReadyForRead(pVnode->sync)) {
     vnodeRedirectRpcMsg(pVnode, pMsg, terrno);
     return 0;
   }
@@ -526,6 +530,8 @@ int32_t vnodeProcessQueryMsg(SVnode *pVnode, SRpcMsg *pMsg) {
       return qWorkerProcessCQueryMsg(&handle, pVnode->pQuery, pMsg, 0);
     case TDMT_VND_TMQ_CONSUME:
       return tqProcessPollReq(pVnode->pTq, pMsg);
+    case TDMT_VND_TMQ_CONSUME_PUSH:
+      return tqProcessPollPush(pVnode->pTq, pMsg);
     default:
       vError("unknown msg type:%d in query queue", pMsg->msgType);
       return TSDB_CODE_APP_ERROR;
@@ -559,8 +565,8 @@ int32_t vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg, SQueueInfo *pInfo) {
       return vnodeGetTableCfg(pVnode, pMsg, true);
     case TDMT_VND_BATCH_META:
       return vnodeGetBatchMeta(pVnode, pMsg);
-    case TDMT_VND_TMQ_CONSUME:
+//    case TDMT_VND_TMQ_CONSUME:
-      return tqProcessPollReq(pVnode->pTq, pMsg);
+//      return tqProcessPollReq(pVnode->pTq, pMsg);
     case TDMT_VND_TMQ_VG_WALINFO:
       return tqProcessVgWalInfoReq(pVnode->pTq, pMsg);
     case TDMT_STREAM_TASK_RUN:

source/libs/catalog/inc/catalogInt.h

@@ -950,8 +950,8 @@ int32_t ctgCloneMetaOutput(STableMetaOutput* output, STableMetaOutput** pOutput)
 int32_t ctgGenerateVgList(SCatalog* pCtg, SHashObj* vgHash, SArray** pList);
 void    ctgFreeJob(void* job);
 void    ctgFreeHandleImpl(SCatalog* pCtg);
-int32_t ctgGetVgInfoFromHashValue(SCatalog* pCtg, SDBVgInfo* dbInfo, const SName* pTableName, SVgroupInfo* pVgroup);
+int32_t ctgGetVgInfoFromHashValue(SCatalog* pCtg, SEpSet* pMgmtEps, SDBVgInfo* dbInfo, const SName* pTableName, SVgroupInfo* pVgroup);
-int32_t ctgGetVgInfosFromHashValue(SCatalog* pCtg, SCtgTaskReq* tReq, SDBVgInfo* dbInfo, SCtgTbHashsCtx* pCtx,
+int32_t ctgGetVgInfosFromHashValue(SCatalog* pCtg, SEpSet* pMgmgEpSet, SCtgTaskReq* tReq, SDBVgInfo* dbInfo, SCtgTbHashsCtx* pCtx,
                                    char* dbFName, SArray* pNames, bool update);
 int32_t ctgGetVgIdsFromHashValue(SCatalog* pCtg, SDBVgInfo* dbInfo, char* dbFName, const char* pTbs[], int32_t tbNum,
                                  int32_t* vgId);

source/libs/catalog/src/catalog.c

@@ -568,7 +568,7 @@ int32_t ctgGetTbHashVgroup(SCatalog* pCtg, SRequestConnInfo* pConn, const SName*
     return TSDB_CODE_SUCCESS;
   }
   
-  CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, vgInfo ? vgInfo : dbCache->vgCache.vgInfo, pTableName, pVgroup));
+  CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, pConn ? &pConn->mgmtEps : NULL, vgInfo ? vgInfo : dbCache->vgCache.vgInfo, pTableName, pVgroup));
 
 _return:
 
@@ -629,7 +629,7 @@ int32_t ctgGetCachedTbVgMeta(SCatalog* pCtg, const SName* pTableName, SVgroupInf
     return TSDB_CODE_SUCCESS;
   }
 
-  CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, dbCache->vgCache.vgInfo, pTableName, pVgroup));
+  CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, NULL, dbCache->vgCache.vgInfo, pTableName, pVgroup));
 
   ctgRUnlockVgInfo(dbCache);
 

source/libs/catalog/src/ctgAsync.c

@@ -1112,7 +1112,7 @@ int32_t ctgHandleGetTbMetaRsp(SCtgTaskReq* tReq, int32_t reqType, const SDataBuf
       SUseDbOutput* pOut = (SUseDbOutput*)pMsgCtx->out;
 
       SVgroupInfo vgInfo = {0};
-      CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, pOut->dbVgroup, pName, &vgInfo));
+      CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, &pConn->mgmtEps, pOut->dbVgroup, pName, &vgInfo));
 
       ctgDebug("will refresh tbmeta, not supposed to be stb, tbName:%s, flag:%d", tNameGetTableName(pName), flag);
 
@@ -1132,7 +1132,7 @@ int32_t ctgHandleGetTbMetaRsp(SCtgTaskReq* tReq, int32_t reqType, const SDataBuf
           CTG_ERR_RET(ctgAcquireVgInfoFromCache(pCtg, dbFName, &dbCache));
           if (NULL != dbCache) {
             SVgroupInfo vgInfo = {0};
-            CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, dbCache->vgCache.vgInfo, pName, &vgInfo));
+            CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, &pConn->mgmtEps, dbCache->vgCache.vgInfo, pName, &vgInfo));
 
             ctgDebug("will refresh tbmeta, supposed to be stb, tbName:%s, flag:%d", tNameGetTableName(pName), flag);
 
@@ -1282,7 +1282,7 @@ int32_t ctgHandleGetTbMetasRsp(SCtgTaskReq* tReq, int32_t reqType, const SDataBu
       SUseDbOutput* pOut = (SUseDbOutput*)pMsgCtx->out;
 
       SVgroupInfo vgInfo = {0};
-      CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, pOut->dbVgroup, pName, &vgInfo));
+      CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, &pConn->mgmtEps, pOut->dbVgroup, pName, &vgInfo));
 
       ctgTaskDebug("will refresh tbmeta, not supposed to be stb, tbName:%s, flag:%d", tNameGetTableName(pName), flag);
 
@@ -1302,7 +1302,7 @@ int32_t ctgHandleGetTbMetasRsp(SCtgTaskReq* tReq, int32_t reqType, const SDataBu
           CTG_ERR_RET(ctgAcquireVgInfoFromCache(pCtg, dbFName, &dbCache));
           if (NULL != dbCache) {
             SVgroupInfo vgInfo = {0};
-            CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, dbCache->vgCache.vgInfo, pName, &vgInfo));
+            CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, &pConn->mgmtEps, dbCache->vgCache.vgInfo, pName, &vgInfo));
 
             ctgTaskDebug("will refresh tbmeta, supposed to be stb, tbName:%s, flag:%d", tNameGetTableName(pName), flag);
 
@@ -1501,7 +1501,7 @@ int32_t ctgHandleGetTbHashRsp(SCtgTaskReq* tReq, int32_t reqType, const SDataBuf
         CTG_ERR_JRET(TSDB_CODE_OUT_OF_MEMORY);
       }
 
-      CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, pOut->dbVgroup, ctx->pName, (SVgroupInfo*)pTask->res));
+      CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, &pTask->pJob->conn.mgmtEps, pOut->dbVgroup, ctx->pName, (SVgroupInfo*)pTask->res));
 
       CTG_ERR_JRET(ctgUpdateVgroupEnqueue(pCtg, ctx->dbFName, pOut->dbId, pOut->dbVgroup, false));
       pOut->dbVgroup = NULL;
@@ -1536,7 +1536,7 @@ int32_t ctgHandleGetTbHashsRsp(SCtgTaskReq* tReq, int32_t reqType, const SDataBu
       SUseDbOutput* pOut = (SUseDbOutput*)pMsgCtx->out;
 
       STablesReq* pReq = taosArrayGet(ctx->pNames, pFetch->dbIdx);
-      CTG_ERR_JRET(ctgGetVgInfosFromHashValue(pCtg, tReq, pOut->dbVgroup, ctx, pMsgCtx->target, pReq->pTables, true));
+      CTG_ERR_JRET(ctgGetVgInfosFromHashValue(pCtg, &pTask->pJob->conn.mgmtEps, tReq, pOut->dbVgroup, ctx, pMsgCtx->target, pReq->pTables, true));
 
       CTG_ERR_JRET(ctgUpdateVgroupEnqueue(pCtg, pMsgCtx->target, pOut->dbId, pOut->dbVgroup, false));
       pOut->dbVgroup = NULL;
@@ -1799,7 +1799,7 @@ int32_t ctgAsyncRefreshTbMeta(SCtgTaskReq* tReq, int32_t flag, SName* pName, int
   CTG_ERR_RET(ctgAcquireVgInfoFromCache(pCtg, dbFName, &dbCache));
   if (dbCache) {
     SVgroupInfo vgInfo = {0};
-    CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, dbCache->vgCache.vgInfo, pName, &vgInfo));
+    CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, &pConn->mgmtEps, dbCache->vgCache.vgInfo, pName, &vgInfo));
 
     ctgDebug("will refresh tbmeta, not supposed to be stb, tbName:%s, flag:%d", tNameGetTableName(pName), flag);
 
@@ -1948,7 +1948,7 @@ int32_t ctgLaunchGetTbHashTask(SCtgTask* pTask) {
     if (NULL == pTask->res) {
       CTG_ERR_JRET(TSDB_CODE_OUT_OF_MEMORY);
     }
-    CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, dbCache->vgCache.vgInfo, pCtx->pName, (SVgroupInfo*)pTask->res));
+    CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, &pConn->mgmtEps, dbCache->vgCache.vgInfo, pCtx->pName, (SVgroupInfo*)pTask->res));
 
     ctgReleaseVgInfoToCache(pCtg, dbCache);
     dbCache = NULL;
@@ -1996,7 +1996,7 @@ int32_t ctgLaunchGetTbHashsTask(SCtgTask* pTask) {
       tReq.pTask = pTask;
       tReq.msgIdx = -1;
       CTG_ERR_JRET(
-          ctgGetVgInfosFromHashValue(pCtg, &tReq, dbCache->vgCache.vgInfo, pCtx, pReq->dbFName, pReq->pTables, false));
+          ctgGetVgInfosFromHashValue(pCtg, &pConn->mgmtEps, &tReq, dbCache->vgCache.vgInfo, pCtx, pReq->dbFName, pReq->pTables, false));
 
       ctgReleaseVgInfoToCache(pCtg, dbCache);
       dbCache = NULL;
@@ -2375,7 +2375,7 @@ int32_t ctgGetTbCfgCb(SCtgTask* pTask) {
     SDBVgInfo* pDb = (SDBVgInfo*)pTask->subRes.res;
 
     pCtx->pVgInfo = taosMemoryCalloc(1, sizeof(SVgroupInfo));
-    CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pTask->pJob->pCtg, pDb, pCtx->pName, pCtx->pVgInfo));
+    CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pTask->pJob->pCtg, &pTask->pJob->conn.mgmtEps, pDb, pCtx->pName, pCtx->pVgInfo));
   }
 
   CTG_RET(ctgLaunchGetTbCfgTask(pTask));
@@ -2395,7 +2395,7 @@ int32_t ctgGetTbTagCb(SCtgTask* pTask) {
 
   if (NULL == pCtx->pVgInfo) {
     pCtx->pVgInfo = taosMemoryCalloc(1, sizeof(SVgroupInfo));
-    CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pTask->pJob->pCtg, pDb, pCtx->pName, pCtx->pVgInfo));
+    CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pTask->pJob->pCtg, &pTask->pJob->conn.mgmtEps, pDb, pCtx->pName, pCtx->pVgInfo));
   }
   
   CTG_RET(ctgLaunchGetTbTagTask(pTask));

source/libs/catalog/src/ctgCache.c

@@ -2989,7 +2989,7 @@ int32_t ctgGetTbHashVgroupFromCache(SCatalog *pCtg, const SName *pTableName, SVg
   }
 
   *pVgroup = taosMemoryCalloc(1, sizeof(SVgroupInfo));
-  CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, dbCache->vgCache.vgInfo, pTableName, *pVgroup));
+  CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, NULL, dbCache->vgCache.vgInfo, pTableName, *pVgroup));
 
 _return:
 

source/libs/catalog/src/ctgUtil.c

@@ -969,7 +969,7 @@ int32_t ctgHashValueComp(void const* lp, void const* rp) {
   return 0;
 }
 
-int32_t ctgGetVgInfoFromHashValue(SCatalog* pCtg, SDBVgInfo* dbInfo, const SName* pTableName, SVgroupInfo* pVgroup) {
+int32_t ctgGetVgInfoFromHashValue(SCatalog* pCtg, SEpSet* pMgmtEps, SDBVgInfo* dbInfo, const SName* pTableName, SVgroupInfo* pVgroup) {
   int32_t code = 0;
   CTG_ERR_RET(ctgMakeVgArray(dbInfo));
 
@@ -977,6 +977,14 @@ int32_t ctgGetVgInfoFromHashValue(SCatalog* pCtg, SDBVgInfo* dbInfo, const SName
   char    db[TSDB_DB_FNAME_LEN] = {0};
   tNameGetFullDbName(pTableName, db);
 
+  if (IS_SYS_DBNAME(pTableName->dbname)) {
+    pVgroup->vgId = MNODE_HANDLE;
+    if (pMgmtEps) {
+      memcpy(&pVgroup->epSet, pMgmtEps, sizeof(pVgroup->epSet));
+    }
+    return TSDB_CODE_SUCCESS;
+  }
+
   if (vgNum <= 0) {
     ctgError("db vgroup cache invalid, db:%s, vgroup number:%d", db, vgNum);
     CTG_ERR_RET(TSDB_CODE_TSC_DB_NOT_SELECTED);
@@ -1020,23 +1028,53 @@ int32_t ctgGetVgInfoFromHashValue(SCatalog* pCtg, SDBVgInfo* dbInfo, const SName
   CTG_RET(code);
 }
 
-int32_t ctgGetVgInfosFromHashValue(SCatalog* pCtg, SCtgTaskReq* tReq, SDBVgInfo* dbInfo, SCtgTbHashsCtx* pCtx,
+int32_t ctgGetVgInfosFromHashValue(SCatalog* pCtg, SEpSet* pMgmgEpSet, SCtgTaskReq* tReq, SDBVgInfo* dbInfo, SCtgTbHashsCtx* pCtx,
                                    char* dbFName, SArray* pNames, bool update) {
   int32_t   code = 0;
   SCtgTask* pTask = tReq->pTask;
   SMetaRes  res = {0};
+  SVgroupInfo* vgInfo = NULL;
 
   CTG_ERR_RET(ctgMakeVgArray(dbInfo));
 
+  int32_t      tbNum = taosArrayGetSize(pNames);
+
+  char* pSep = strchr(dbFName, '.');
+  if (pSep && IS_SYS_DBNAME(pSep + 1)) {
+    SVgroupInfo mgmtInfo = {0};
+    mgmtInfo.vgId = MNODE_HANDLE;
+    if (pMgmgEpSet) {
+      memcpy(&mgmtInfo.epSet, pMgmgEpSet, sizeof(mgmtInfo.epSet));
+    }
+    for (int32_t i = 0; i < tbNum; ++i) {
+      vgInfo = taosMemoryMalloc(sizeof(SVgroupInfo));
+      if (NULL == vgInfo) {
+        CTG_ERR_RET(TSDB_CODE_OUT_OF_MEMORY);
+      }
+
+      memcpy(vgInfo, &mgmtInfo, sizeof(mgmtInfo));
+
+      ctgDebug("Got tb hash vgroup, vgId:%d, epNum %d, current %s port %d", vgInfo->vgId, vgInfo->epSet.numOfEps,
+               vgInfo->epSet.eps[vgInfo->epSet.inUse].fqdn, vgInfo->epSet.eps[vgInfo->epSet.inUse].port);
+
+      if (update) {
+        SCtgFetch* pFetch = taosArrayGet(pCtx->pFetchs, tReq->msgIdx);
+        SMetaRes*  pRes = taosArrayGet(pCtx->pResList, pFetch->resIdx + i);
+        pRes->pRes = vgInfo;
+      } else {
+        res.pRes = vgInfo;
+        taosArrayPush(pCtx->pResList, &res);
+      }
+    }
+    return TSDB_CODE_SUCCESS;
+  }
+
   int32_t vgNum = taosArrayGetSize(dbInfo->vgArray);
   if (vgNum <= 0) {
     ctgError("db vgroup cache invalid, db:%s, vgroup number:%d", dbFName, vgNum);
     CTG_ERR_RET(TSDB_CODE_CTG_INTERNAL_ERROR);
   }
 
-  SVgroupInfo* vgInfo = NULL;
-  int32_t      tbNum = taosArrayGetSize(pNames);
-
   if (1 == vgNum) {
     for (int32_t i = 0; i < tbNum; ++i) {
       vgInfo = taosMemoryMalloc(sizeof(SVgroupInfo));

source/libs/executor/src/aggregateoperator.c

@@ -461,7 +461,11 @@ int32_t doInitAggInfoSup(SAggSupporter* pAggSup, SqlFunctionCtx* pCtx, int32_t n
 
   uint32_t defaultPgsz = 0;
   uint32_t defaultBufsz = 0;
-  getBufferPgSize(pAggSup->resultRowSize, &defaultPgsz, &defaultBufsz);
+  code = getBufferPgSize(pAggSup->resultRowSize, &defaultPgsz, &defaultBufsz);
+  if (code) {
+    qError("failed to get buff page size, rowSize:%d", pAggSup->resultRowSize);
+    return code;
+  }
   
   if (!osTempSpaceAvailable()) {
     code = TSDB_CODE_NO_DISKSPACE;

source/libs/executor/src/eventwindowoperator.c

@@ -174,6 +174,7 @@ void destroyEWindowOperatorInfo(void* param) {
   colDataDestroy(&pInfo->twAggSup.timeWindowData);
 
   cleanupAggSup(&pInfo->aggSup);
+  cleanupExprSupp(&pInfo->scalarSup);
   taosMemoryFreeClear(param);
 }
 

source/libs/executor/src/executorInt.c

@@ -922,8 +922,13 @@ void destroyExprInfo(SExprInfo* pExpr, int32_t numOfExprs) {
 
 int32_t getBufferPgSize(int32_t rowSize, uint32_t* defaultPgsz, uint32_t* defaultBufsz) {
   *defaultPgsz = 4096;
+  uint32_t last = *defaultPgsz;
   while (*defaultPgsz < rowSize * 4) {
     *defaultPgsz <<= 1u;
+    if (*defaultPgsz < last) {
+      return TSDB_CODE_INVALID_PARA;
+    }
+    last = *defaultPgsz;
   }
 
   // The default buffer for each operator in query is 10MB.
@@ -932,6 +937,9 @@ int32_t getBufferPgSize(int32_t rowSize, uint32_t* defaultPgsz, uint32_t* defaul
   *defaultBufsz = 4096 * 2560;
   if ((*defaultBufsz) <= (*defaultPgsz)) {
     (*defaultBufsz) = (*defaultPgsz) * 4;
+    if (*defaultBufsz < ((int64_t)(*defaultPgsz)) * 4) {
+      return TSDB_CODE_INVALID_PARA;
+    }
   }
 
   return 0;

source/libs/executor/src/groupoperator.c

@@ -871,7 +871,12 @@ SOperatorInfo* createPartitionOperatorInfo(SOperatorInfo* downstream, SPartition
   uint32_t defaultBufsz = 0;
 
   pInfo->binfo.pRes = createDataBlockFromDescNode(pPartNode->node.pOutputDataBlockDesc);
-  getBufferPgSize(pInfo->binfo.pRes->info.rowSize, &defaultPgsz, &defaultBufsz);
+  int32_t code = getBufferPgSize(pInfo->binfo.pRes->info.rowSize, &defaultPgsz, &defaultBufsz);
+  if (code != TSDB_CODE_SUCCESS) {
+    terrno = code;
+    pTaskInfo->code = code;
+    goto _error;
+  }
 
   if (!osTempSpaceAvailable()) {
     terrno = TSDB_CODE_NO_DISKSPACE;
@@ -880,7 +885,7 @@ SOperatorInfo* createPartitionOperatorInfo(SOperatorInfo* downstream, SPartition
     goto _error;
   }
 
-  int32_t code = createDiskbasedBuf(&pInfo->pBuf, defaultPgsz, defaultBufsz, pTaskInfo->id.str, tsTempDir);
+  code = createDiskbasedBuf(&pInfo->pBuf, defaultPgsz, defaultBufsz, pTaskInfo->id.str, tsTempDir);
   if (code != TSDB_CODE_SUCCESS) {
     terrno = code;
     pTaskInfo->code = code;

source/libs/executor/src/tsort.c

@@ -101,7 +101,11 @@ static int32_t sortComparCleanup(SMsortComparParam* cmpParam) {
   for (int32_t i = 0; i < cmpParam->numOfSources; ++i) {
     SSortSource* pSource = cmpParam->pSources[i];
     blockDataDestroy(pSource->src.pBlock);
+    if (pSource->pageIdList) {
+      taosArrayDestroy(pSource->pageIdList);
+    }
     taosMemoryFreeClear(pSource);
+    cmpParam->pSources[i] = NULL;
   }
 
   cmpParam->numOfSources = 0;
@@ -123,9 +127,11 @@ void tsortClearOrderdSource(SArray* pOrderedSource, int64_t *fetchUs, int64_t *f
     // release pageIdList
     if ((*pSource)->pageIdList) {
       taosArrayDestroy((*pSource)->pageIdList);
+      (*pSource)->pageIdList = NULL;
     }
     if ((*pSource)->param && !(*pSource)->onlyRef) {
       taosMemoryFree((*pSource)->param);
+      (*pSource)->param = NULL;
     }
 
     if (!(*pSource)->onlyRef && (*pSource)->src.pBlock) {

source/libs/function/src/builtinsimpl.c

@@ -2660,7 +2660,7 @@ bool diffFunctionSetup(SqlFunctionCtx* pCtx, SResultRowEntryInfo* pResInfo) {
   } else {
     pDiffInfo->ignoreNegative = false;
   }
-  pDiffInfo->includeNull = false;
+  pDiffInfo->includeNull = true;
   pDiffInfo->firstOutput = false;
   return true;
 }

source/libs/function/src/functionMgt.c

@@ -344,7 +344,7 @@ static int32_t getFuncInfo(SFunctionNode* pFunc) {
   return fmGetFuncInfo(pFunc, msg, sizeof(msg));
 }
 
-static SFunctionNode* createFunction(const char* pName, SNodeList* pParameterList) {
+SFunctionNode* createFunction(const char* pName, SNodeList* pParameterList) {
   SFunctionNode* pFunc = (SFunctionNode*)nodesMakeNode(QUERY_NODE_FUNCTION);
   if (NULL == pFunc) {
     return NULL;

source/libs/nodes/src/nodesUtilFuncs.c

@@ -953,6 +953,8 @@ void nodesDestroyNode(SNode* pNode) {
       nodesDestroyNode(pStmt->pQuery);
       nodesDestroyList(pStmt->pTags);
       nodesDestroyNode(pStmt->pSubtable);
+      tFreeSCMCreateStreamReq(pStmt->pReq);
+      taosMemoryFreeClear(pStmt->pReq);
       break;
     }
     case QUERY_NODE_DROP_STREAM_STMT:            // no pointer field
@@ -1052,6 +1054,7 @@ void nodesDestroyNode(SNode* pNode) {
     case QUERY_NODE_QUERY: {
       SQuery* pQuery = (SQuery*)pNode;
       nodesDestroyNode(pQuery->pRoot);
+      nodesDestroyNode(pQuery->pPostRoot);
       taosMemoryFreeClear(pQuery->pResSchema);
       if (NULL != pQuery->pCmdMsg) {
         taosMemoryFreeClear(pQuery->pCmdMsg->pMsg);
@@ -1953,9 +1956,9 @@ static uint32_t funcNodeHash(const char* pKey, uint32_t len) {
 }
 
 static int32_t funcNodeEqual(const void* pLeft, const void* pRight, size_t len) {
-  if (0 != strcmp((*(const SExprNode**)pLeft)->aliasName, (*(const SExprNode**)pRight)->aliasName)) {
+  // if (0 != strcmp((*(const SExprNode**)pLeft)->aliasName, (*(const SExprNode**)pRight)->aliasName)) {
-    return 1;
+  //   return 1;
-  }
+  // }
   return nodesEqualNode(*(const SNode**)pLeft, *(const SNode**)pRight) ? 0 : 1;
 }
 

source/libs/parser/inc/parAst.h

@@ -210,7 +210,7 @@ SNode* createCreateTopicStmtUseDb(SAstCreateContext* pCxt, bool ignoreExists, ST
 SNode* createCreateTopicStmtUseTable(SAstCreateContext* pCxt, bool ignoreExists, SToken* pTopicName, SNode* pRealTable,
                                      bool withMeta, SNode* pWhere);
 SNode* createDropTopicStmt(SAstCreateContext* pCxt, bool ignoreNotExists, SToken* pTopicName);
-SNode* createDropCGroupStmt(SAstCreateContext* pCxt, bool ignoreNotExists, const SToken* pCGroupId, SToken* pTopicName);
+SNode* createDropCGroupStmt(SAstCreateContext* pCxt, bool ignoreNotExists, SToken* pCGroupId, SToken* pTopicName);
 SNode* createAlterLocalStmt(SAstCreateContext* pCxt, const SToken* pConfig, const SToken* pValue);
 SNode* createDefaultExplainOptions(SAstCreateContext* pCxt);
 SNode* setExplainVerbose(SAstCreateContext* pCxt, SNode* pOptions, const SToken* pVal);

source/libs/parser/inc/parInt.h

@@ -34,6 +34,7 @@ int32_t authenticate(SParseContext* pParseCxt, SQuery* pQuery, SParseMetaCache*
 int32_t translate(SParseContext* pParseCxt, SQuery* pQuery, SParseMetaCache* pMetaCache);
 int32_t extractResultSchema(const SNode* pRoot, int32_t* numOfCols, SSchema** pSchema);
 int32_t calculateConstant(SParseContext* pParseCxt, SQuery* pQuery);
+int32_t translatePostCreateStream(SParseContext* pParseCxt, SQuery* pQuery, void** pResRow);
 
 #ifdef __cplusplus
 }

source/libs/parser/src/parAstCreater.c

@@ -210,6 +210,15 @@ static bool checkTopicName(SAstCreateContext* pCxt, SToken* pTopicName) {
   return true;
 }
 
+static bool checkCGroupName(SAstCreateContext* pCxt, SToken* pCGroup) {
+  trimEscape(pCGroup);
+  if (pCGroup->n >= TSDB_CGROUP_LEN) {
+    pCxt->errCode = generateSyntaxErrMsg(&pCxt->msgBuf, TSDB_CODE_PAR_INVALID_IDENTIFIER_NAME, pCGroup->z);
+    return false;
+  }
+  return true;
+}
+
 static bool checkStreamName(SAstCreateContext* pCxt, SToken* pStreamName) {
   trimEscape(pStreamName);
   if (pStreamName->n >= TSDB_STREAM_NAME_LEN) {
@@ -1751,12 +1760,15 @@ SNode* createDropTopicStmt(SAstCreateContext* pCxt, bool ignoreNotExists, SToken
   return (SNode*)pStmt;
 }
 
-SNode* createDropCGroupStmt(SAstCreateContext* pCxt, bool ignoreNotExists, const SToken* pCGroupId,
+SNode* createDropCGroupStmt(SAstCreateContext* pCxt, bool ignoreNotExists, SToken* pCGroupId,
                             SToken* pTopicName) {
   CHECK_PARSER_STATUS(pCxt);
   if (!checkTopicName(pCxt, pTopicName)) {
     return NULL;
   }
+  if (!checkCGroupName(pCxt, pCGroupId)) {
+    return NULL;
+  }
   SDropCGroupStmt* pStmt = (SDropCGroupStmt*)nodesMakeNode(QUERY_NODE_DROP_CGROUP_STMT);
   CHECK_OUT_OF_MEM(pStmt);
   pStmt->ignoreNotExists = ignoreNotExists;

source/libs/parser/src/parAstParser.c

@@ -384,6 +384,10 @@ static int32_t collectMetaKeyFromCreateStream(SCollectMetaKeyCxt* pCxt, SCreateS
   if (TSDB_CODE_SUCCESS == code) {
     code = collectMetaKeyFromQuery(pCxt, pStmt->pQuery);
   }
+  if (TSDB_CODE_SUCCESS == code && pStmt->pOptions->fillHistory) {
+    SSelectStmt* pSelect = (SSelectStmt*)pStmt->pQuery;
+    code = reserveDbCfgForLastRow(pCxt, pSelect->pFromTable);
+  }  
   return code;
 }
 

source/libs/parser/src/parCalcConst.c

@@ -311,6 +311,9 @@ static int32_t calcConstDelete(SCalcConstContext* pCxt, SDeleteStmt* pDelete) {
   if (TSDB_CODE_SUCCESS == code) {
     code = calcConstStmtCondition(pCxt, &pDelete->pWhere, &pDelete->deleteZeroRows);
   }
+  if (code == TSDB_CODE_SUCCESS && pDelete->timeRange.skey > pDelete->timeRange.ekey) {
+    pDelete->deleteZeroRows = true;
+  }
   return code;
 }
 
@@ -465,6 +468,9 @@ static bool isEmptyResultQuery(SNode* pStmt) {
       }
       break;
     }
+    case QUERY_NODE_DELETE_STMT:
+      isEmptyResult = ((SDeleteStmt*)pStmt)->deleteZeroRows;
+      break;
     default:
       break;
   }

source/libs/parser/src/parTranslater.c

@@ -53,6 +53,8 @@ typedef struct STranslateContext {
   bool             createStream;
   bool             stableQuery;
   bool             showRewrite;
+  SNode*           pPrevRoot;
+  SNode*           pPostRoot;  
 } STranslateContext;
 
 typedef struct SBuildTopicContext {
@@ -276,6 +278,10 @@ static const SSysTableShowAdapter sysTableShowAdapter[] = {
 static int32_t  translateSubquery(STranslateContext* pCxt, SNode* pNode);
 static int32_t  translateQuery(STranslateContext* pCxt, SNode* pNode);
 static EDealRes translateValue(STranslateContext* pCxt, SValueNode* pVal);
+static int32_t createSimpleSelectStmtFromProjList(const char* pDb, const char* pTable, SNodeList* pProjectionList, SSelectStmt** pStmt);
+static int32_t createLastTsSelectStmt(char* pDb, char* pTable, STableMeta* pMeta, SNode** pQuery);
+static int32_t setQuery(STranslateContext* pCxt, SQuery* pQuery);
+static int32_t setRefreshMate(STranslateContext* pCxt, SQuery* pQuery);
 
 static bool afterGroupBy(ESqlClause clause) { return clause > SQL_CLAUSE_GROUP_BY; }
 
@@ -1378,13 +1384,33 @@ static bool isCountStar(SFunctionNode* pFunc) {
   return (QUERY_NODE_COLUMN == nodeType(pPara) && 0 == strcmp(((SColumnNode*)pPara)->colName, "*"));
 }
 
+static int32_t rewriteCountStarAsCount1(STranslateContext* pCxt, SFunctionNode* pCount) {
+  int32_t code = TSDB_CODE_SUCCESS;
+  SValueNode* pVal = (SValueNode*)nodesMakeNode(QUERY_NODE_VALUE);
+  if (NULL == pVal) {
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+  pVal->node.resType.type = TSDB_DATA_TYPE_INT;
+  pVal->node.resType.bytes = tDataTypes[TSDB_DATA_TYPE_INT].bytes;
+  const int32_t val = 1;
+  nodesSetValueNodeValue(pVal, (void*)&val);
+  pVal->translate = true;
+  nodesListErase(pCount->pParameterList, nodesListGetCell(pCount->pParameterList, 0));
+  code = nodesListAppend(pCount->pParameterList, (SNode*)pVal);
+  return code;
+}
+
 // count(*) is rewritten as count(ts) for scannning optimization
 static int32_t rewriteCountStar(STranslateContext* pCxt, SFunctionNode* pCount) {
   SColumnNode* pCol = (SColumnNode*)nodesListGetNode(pCount->pParameterList, 0);
   STableNode*  pTable = NULL;
   int32_t      code = findTable(pCxt, ('\0' == pCol->tableAlias[0] ? NULL : pCol->tableAlias), &pTable);
-  if (TSDB_CODE_SUCCESS == code && QUERY_NODE_REAL_TABLE == nodeType(pTable)) {
+  if (TSDB_CODE_SUCCESS == code) {
+    if (QUERY_NODE_REAL_TABLE == nodeType(pTable)) {
       setColumnInfoBySchema((SRealTableNode*)pTable, ((SRealTableNode*)pTable)->pMeta->schema, -1, pCol);
+    } else {
+      code = rewriteCountStarAsCount1(pCxt, pCount);
+    }
   }
   return code;
 }
@@ -4973,6 +4999,7 @@ static int32_t buildTableForSampleAst(SSampleAstInfo* pInfo, SNode** pOutput) {
   }
   snprintf(pTable->table.dbName, sizeof(pTable->table.dbName), "%s", pInfo->pDbName);
   snprintf(pTable->table.tableName, sizeof(pTable->table.tableName), "%s", pInfo->pTableName);
+  snprintf(pTable->table.tableAlias, sizeof(pTable->table.tableAlias), "%s", pInfo->pTableName);
   TSWAP(pTable->pMeta, pInfo->pRollupTableMeta);
   *pOutput = (SNode*)pTable;
   return TSDB_CODE_SUCCESS;
@@ -6763,6 +6790,54 @@ static int32_t translateStreamTargetTable(STranslateContext* pCxt, SCreateStream
   return code;
 }
 
+static int32_t createLastTsSelectStmt(char* pDb, char* pTable, STableMeta* pMeta, SNode** pQuery) {
+  SColumnNode* col = (SColumnNode*)nodesMakeNode(QUERY_NODE_COLUMN);
+  if (NULL == col) {
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+  
+  strcpy(col->tableAlias, pTable);
+  strcpy(col->colName, pMeta->schema[0].name);
+  SNodeList* pParamterList = nodesMakeList();
+  if (NULL == pParamterList) {
+    nodesDestroyNode((SNode *)col);
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+  
+  int32_t code = nodesListStrictAppend(pParamterList, (SNode *)col);
+  if (code) {
+    nodesDestroyNode((SNode *)col);
+    nodesDestroyList(pParamterList);
+    return code;
+  }
+  
+  SNode* pFunc = (SNode*)createFunction("last", pParamterList);
+  if (NULL == pFunc) {
+    nodesDestroyList(pParamterList);
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+  
+  SNodeList* pProjectionList = nodesMakeList();
+  if (NULL == pProjectionList) {
+    nodesDestroyList(pParamterList);
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+  code = nodesListStrictAppend(pProjectionList, pFunc);
+  if (code) {
+    nodesDestroyNode(pFunc);
+    nodesDestroyList(pProjectionList);
+    return code;
+  }
+  
+  code = createSimpleSelectStmtFromProjList(pDb, pTable, pProjectionList, (SSelectStmt **)pQuery);
+  if (code) {
+    nodesDestroyList(pProjectionList);
+    return code;
+  }
+
+  return code;
+}
+
 static int32_t buildCreateStreamQuery(STranslateContext* pCxt, SCreateStreamStmt* pStmt, SCMCreateStreamReq* pReq) {
   pCxt->createStream = true;
   STableMeta* pMeta = NULL;
@@ -6789,6 +6864,18 @@ static int32_t buildCreateStreamQuery(STranslateContext* pCxt, SCreateStreamStmt
     getSourceDatabase(pStmt->pQuery, pCxt->pParseCxt->acctId, pReq->sourceDB);
     code = nodesNodeToString(pStmt->pQuery, false, &pReq->ast, NULL);
   }
+  if (TSDB_CODE_SUCCESS == code && pStmt->pOptions->fillHistory) {
+    SRealTableNode* pTable = (SRealTableNode*)(((SSelectStmt*)pStmt->pQuery)->pFromTable);
+    code = createLastTsSelectStmt(pTable->table.dbName, pTable->table.tableName, pTable->pMeta, &pStmt->pPrevQuery);
+/*
+    if (TSDB_CODE_SUCCESS == code) {
+      STranslateContext cxt = {0};
+      int32_t code = initTranslateContext(pCxt->pParseCxt, pCxt->pMetaCache, &cxt);
+      code = translateQuery(&cxt, pStmt->pPrevQuery);
+      destroyTranslateContext(&cxt);
+    }
+*/    
+  }
   taosMemoryFree(pMeta);
   return code;
 }
@@ -6855,13 +6942,86 @@ static int32_t translateCreateStream(STranslateContext* pCxt, SCreateStreamStmt*
     code = buildCreateStreamReq(pCxt, pStmt, &createReq);
   }
   if (TSDB_CODE_SUCCESS == code) {
+    if (NULL == pStmt->pPrevQuery) {
       code = buildCmdMsg(pCxt, TDMT_MND_CREATE_STREAM, (FSerializeFunc)tSerializeSCMCreateStreamReq, &createReq);
+    } else {
+      pStmt->pReq = taosMemoryMalloc(sizeof(createReq));
+      if (NULL == pStmt->pReq) {
+        code = TSDB_CODE_OUT_OF_MEMORY;
+      } else {
+        memcpy(pStmt->pReq, &createReq, sizeof(createReq));
+        memset(&createReq, 0, sizeof(createReq));
+        TSWAP(pCxt->pPrevRoot, pStmt->pPrevQuery);
+      }
+    }
   }
 
   tFreeSCMCreateStreamReq(&createReq);
   return code;
 }
 
+int32_t buildIntervalForCreateStream(SCreateStreamStmt* pStmt, SInterval* pInterval) {
+  int32_t code = TSDB_CODE_SUCCESS;
+  if (QUERY_NODE_SELECT_STMT != nodeType(pStmt->pQuery)) {
+    return code;
+  }
+  SSelectStmt* pSelect = (SSelectStmt*)pStmt->pQuery;
+  if (NULL == pSelect->pWindow || QUERY_NODE_INTERVAL_WINDOW != nodeType(pSelect->pWindow)) {
+    return code;
+  }
+  
+  SIntervalWindowNode* pWindow = (SIntervalWindowNode*)pSelect->pWindow;
+  pInterval->interval = ((SValueNode*)pWindow->pInterval)->datum.i;
+  pInterval->intervalUnit = ((SValueNode*)pWindow->pInterval)->unit;
+  pInterval->offset = (NULL != pWindow->pOffset ? ((SValueNode*)pWindow->pOffset)->datum.i : 0);
+  pInterval->sliding = (NULL != pWindow->pSliding ? ((SValueNode*)pWindow->pSliding)->datum.i : pInterval->interval);
+  pInterval->slidingUnit =
+      (NULL != pWindow->pSliding ? ((SValueNode*)pWindow->pSliding)->unit : pInterval->intervalUnit);
+  pInterval->precision = ((SColumnNode*)pWindow->pCol)->node.resType.precision;    
+
+  return code;
+}
+
+int32_t translatePostCreateStream(SParseContext* pParseCxt, SQuery* pQuery, void** pResRow) {
+  SCreateStreamStmt* pStmt = (SCreateStreamStmt*)pQuery->pRoot;
+  STranslateContext cxt = {0};
+  SInterval interval = {0};
+  int64_t lastTs = 0;
+
+  int32_t code = initTranslateContext(pParseCxt, NULL, &cxt);
+  if (TSDB_CODE_SUCCESS == code) {
+    code = buildIntervalForCreateStream(pStmt, &interval);
+  }
+  if (TSDB_CODE_SUCCESS == code) {
+    if (pResRow && pResRow[0]) {
+      lastTs = *(int64_t*)pResRow[0];
+    } else if (interval.interval > 0) {
+      lastTs = convertTimePrecision(taosGetTimestampMs(), TSDB_TIME_PRECISION_MILLI, interval.precision);
+    } else {
+      lastTs = taosGetTimestampMs();
+    }
+  }
+  if (TSDB_CODE_SUCCESS == code) {
+    if (interval.interval > 0) {
+      pStmt->pReq->lastTs = taosTimeTruncate(lastTs, &interval);
+    } else {
+      pStmt->pReq->lastTs = lastTs;
+    }
+    code = buildCmdMsg(&cxt, TDMT_MND_CREATE_STREAM, (FSerializeFunc)tSerializeSCMCreateStreamReq, pStmt->pReq);
+  }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = setQuery(&cxt, pQuery);
+  }
+  setRefreshMate(&cxt, pQuery);
+  destroyTranslateContext(&cxt);
+
+  tFreeSCMCreateStreamReq(pStmt->pReq);
+  taosMemoryFreeClear(pStmt->pReq);
+
+  return code;
+}
+
+
 static int32_t translateDropStream(STranslateContext* pCxt, SDropStreamStmt* pStmt) {
   SMDropStreamReq dropReq = {0};
   SName           name;
@@ -7542,8 +7702,7 @@ static SNodeList* createProjectCols(int32_t ncols, const char* const pCols[]) {
   return pProjections;
 }
 
-static int32_t createSimpleSelectStmt(const char* pDb, const char* pTable, int32_t numOfProjs,
+static int32_t createSimpleSelectStmtImpl(const char* pDb, const char* pTable, SNodeList* pProjectionList, SSelectStmt** pStmt) {
-                                      const char* const pProjCol[], SSelectStmt** pStmt) {
   SSelectStmt* pSelect = (SSelectStmt*)nodesMakeNode(QUERY_NODE_SELECT_STMT);
   if (NULL == pSelect) {
     return TSDB_CODE_OUT_OF_MEMORY;
@@ -7559,27 +7718,38 @@ static int32_t createSimpleSelectStmt(const char* pDb, const char* pTable, int32
   snprintf(pRealTable->table.tableName, sizeof(pRealTable->table.tableName), "%s", pTable);
   snprintf(pRealTable->table.tableAlias, sizeof(pRealTable->table.tableAlias), "%s", pTable);
   pSelect->pFromTable = (SNode*)pRealTable;
-
+  pSelect->pProjectionList = pProjectionList;
-  if (numOfProjs >= 0) {
-    pSelect->pProjectionList = createProjectCols(numOfProjs, pProjCol);
-    if (NULL == pSelect->pProjectionList) {
-      nodesDestroyNode((SNode*)pSelect);
-      return TSDB_CODE_OUT_OF_MEMORY;
-    }
-  }
 
   *pStmt = pSelect;
 
   return TSDB_CODE_SUCCESS;
 }
 
+
+static int32_t createSimpleSelectStmtFromCols(const char* pDb, const char* pTable, int32_t numOfProjs,
+                                      const char* const pProjCol[], SSelectStmt** pStmt) {
+  SNodeList* pProjectionList = NULL;
+  if (numOfProjs >= 0) {
+    pProjectionList = createProjectCols(numOfProjs, pProjCol);
+    if (NULL == pProjectionList) {
+      return TSDB_CODE_OUT_OF_MEMORY;
+    }
+  }
+
+  return createSimpleSelectStmtImpl(pDb, pTable, pProjectionList, pStmt);
+}
+
+static int32_t createSimpleSelectStmtFromProjList(const char* pDb, const char* pTable, SNodeList* pProjectionList, SSelectStmt** pStmt) {
+  return createSimpleSelectStmtImpl(pDb, pTable, pProjectionList, pStmt);
+}
+
 static int32_t createSelectStmtForShow(ENodeType showType, SSelectStmt** pStmt) {
   const SSysTableShowAdapter* pShow = &sysTableShowAdapter[showType - SYSTABLE_SHOW_TYPE_OFFSET];
-  return createSimpleSelectStmt(pShow->pDbName, pShow->pTableName, pShow->numOfShowCols, pShow->pShowCols, pStmt);
+  return createSimpleSelectStmtFromCols(pShow->pDbName, pShow->pTableName, pShow->numOfShowCols, pShow->pShowCols, pStmt);
 }
 
 static int32_t createSelectStmtForShowTableDist(SShowTableDistributedStmt* pStmt, SSelectStmt** pOutput) {
-  return createSimpleSelectStmt(pStmt->dbName, pStmt->tableName, 0, NULL, pOutput);
+  return createSimpleSelectStmtFromCols(pStmt->dbName, pStmt->tableName, 0, NULL, pOutput);
 }
 
 static int32_t createOperatorNode(EOperatorType opType, const char* pColName, SNode* pRight, SNode** pOp) {
@@ -7713,7 +7883,7 @@ static int32_t createShowTableTagsProjections(SNodeList** pProjections, SNodeLis
 static int32_t rewriteShowStableTags(STranslateContext* pCxt, SQuery* pQuery) {
   SShowTableTagsStmt* pShow = (SShowTableTagsStmt*)pQuery->pRoot;
   SSelectStmt*        pSelect = NULL;
-  int32_t code = createSimpleSelectStmt(((SValueNode*)pShow->pDbName)->literal, ((SValueNode*)pShow->pTbName)->literal,
+  int32_t code = createSimpleSelectStmtFromCols(((SValueNode*)pShow->pDbName)->literal, ((SValueNode*)pShow->pTbName)->literal,
                                         -1, NULL, &pSelect);
   if (TSDB_CODE_SUCCESS == code) {
     code = createShowTableTagsProjections(&pSelect->pProjectionList, &pShow->pTags);
@@ -9038,6 +9208,7 @@ static int32_t setQuery(STranslateContext* pCxt, SQuery* pQuery) {
       }
       break;
     default:
+      pQuery->haveResultSet = false;
       pQuery->execMode = QUERY_EXEC_MODE_RPC;
       if (NULL != pCxt->pCmdMsg) {
         TSWAP(pQuery->pCmdMsg, pCxt->pCmdMsg);
@@ -9072,6 +9243,10 @@ int32_t translate(SParseContext* pParseCxt, SQuery* pQuery, SParseMetaCache* pMe
   if (TSDB_CODE_SUCCESS == code) {
     code = translateQuery(&cxt, pQuery->pRoot);
   }
+  if (TSDB_CODE_SUCCESS == code && (cxt.pPrevRoot || cxt.pPostRoot)) {
+    pQuery->pPrevRoot = cxt.pPrevRoot;
+    pQuery->pPostRoot = cxt.pPostRoot;
+  }
   if (TSDB_CODE_SUCCESS == code) {
     code = setQuery(&cxt, pQuery);
   }

source/libs/parser/src/parser.c

@@ -204,7 +204,7 @@ int32_t qAnalyseSqlSemantic(SParseContext* pCxt, const struct SCatalogReq* pCata
                             const struct SMetaData* pMetaData, SQuery* pQuery) {
   SParseMetaCache metaCache = {0};
   int32_t         code = nodesAcquireAllocator(pCxt->allocatorId);
-  if (TSDB_CODE_SUCCESS == code) {
+  if (TSDB_CODE_SUCCESS == code && pCatalogReq) {
     code = putMetaDataToCache(pCatalogReq, pMetaData, &metaCache);
   }
   if (TSDB_CODE_SUCCESS == code) {
@@ -221,6 +221,19 @@ int32_t qContinueParseSql(SParseContext* pCxt, struct SCatalogReq* pCatalogReq,
   return parseInsertSql(pCxt, &pQuery, pCatalogReq, pMetaData);
 }
 
+int32_t qContinueParsePostQuery(SParseContext* pCxt, SQuery* pQuery, void** pResRow) {
+  int32_t code = TSDB_CODE_SUCCESS;
+  switch (nodeType(pQuery->pRoot)) {
+    case QUERY_NODE_CREATE_STREAM_STMT:
+      code = translatePostCreateStream(pCxt, pQuery, pResRow);
+      break;
+    default:
+      break;
+  }
+
+  return code;
+}
+
 void qDestroyParseContext(SParseContext* pCxt) {
   if (NULL == pCxt) {
     return;

source/libs/parser/test/parInitialCTest.cpp

@@ -885,12 +885,12 @@ TEST_F(ParserInitialCTest, createStream) {
 
   setCreateStreamReq(
       "s1", "test",
-      "create stream if not exists s1 trigger max_delay 20s watermark 10s ignore expired 0 fill_history 1 ignore "
+      "create stream if not exists s1 trigger max_delay 20s watermark 10s ignore expired 0 fill_history 0 ignore "
       "update 1 into st3 as select count(*) from t1 interval(10s)",
       "st3", 1);
   setStreamOptions(STREAM_CREATE_STABLE_TRUE, STREAM_TRIGGER_MAX_DELAY, 20 * MILLISECOND_PER_SECOND,
-                   10 * MILLISECOND_PER_SECOND, 0, 1, 1);
+                   10 * MILLISECOND_PER_SECOND, 0, 0, 1);
-  run("CREATE STREAM IF NOT EXISTS s1 TRIGGER MAX_DELAY 20s WATERMARK 10s IGNORE EXPIRED 0 FILL_HISTORY 1 IGNORE "
+  run("CREATE STREAM IF NOT EXISTS s1 TRIGGER MAX_DELAY 20s WATERMARK 10s IGNORE EXPIRED 0 FILL_HISTORY 0 IGNORE "
       "UPDATE 1 INTO st3 AS SELECT COUNT(*) FROM t1 INTERVAL(10S)");
   clearCreateStreamReq();
 

source/libs/planner/src/planner.c

@@ -97,6 +97,12 @@ static int32_t setSubplanExecutionNode(SPhysiNode* pNode, int32_t groupId, SDown
   return TSDB_CODE_SUCCESS;
 }
 
+int32_t qContinuePlanPostQuery(void *pPostPlan) {
+  //TODO
+  return TSDB_CODE_SUCCESS;
+}
+
+
 int32_t qSetSubplanExecutionNode(SSubplan* subplan, int32_t groupId, SDownstreamSourceNode* pSource) {
   planDebug("QID:0x%" PRIx64 " set subplan execution node, groupId:%d", subplan->id.queryId, groupId);
   return setSubplanExecutionNode(subplan->pNode, groupId, pSource);

source/libs/qworker/CMakeLists.txt

@@ -7,15 +7,9 @@ target_include_directories(
     PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
 )
 
-IF (TD_GRANT)
-    TARGET_LINK_LIBRARIES(qworker
-            PRIVATE os util transport nodes planner qcom executor index grant
-            )
-ELSE ()
 TARGET_LINK_LIBRARIES(qworker
         PRIVATE os util transport nodes planner qcom executor index
         )
-ENDIF()
 
 if(${BUILD_TEST})
     ADD_SUBDIRECTORY(test)

source/libs/qworker/src/qwMsg.c

@@ -366,7 +366,7 @@ int32_t qWorkerPreprocessQueryMsg(void *qWorkerMgmt, SRpcMsg *pMsg, bool chkGran
     QW_ERR_RET(TSDB_CODE_QRY_INVALID_INPUT);
   }
 
-  if (chkGrant && (!TEST_SHOW_REWRITE_MASK(msg.msgMask)) && (grantCheck(TSDB_GRANT_TIME) != TSDB_CODE_SUCCESS)) {
+  if (chkGrant && (!TEST_SHOW_REWRITE_MASK(msg.msgMask)) && !taosGranted()) {
     QW_ELOG("query failed cause of grant expired, msgMask:%d", msg.msgMask);
     tFreeSSubQueryMsg(&msg);
     QW_ERR_RET(TSDB_CODE_GRANT_EXPIRED);

source/libs/qworker/src/qworker.c

@@ -824,6 +824,7 @@ int32_t qwProcessCQuery(QW_FPARAMS_DEF, SQWMsg *qwMsg) {
       break;
     }
     QW_UNLOCK(QW_WRITE, &ctx->lock);
+    queryStop = false;
   } while (true);
 
   input.code = code;

source/libs/stream/src/streamMeta.c

@@ -89,6 +89,9 @@ SStreamMeta* streamMetaOpen(const char* path, void* ahandle, FTaskExpand expandF
   }
 
   pMeta->streamBackend = streamBackendInit(streamPath);
+  if (pMeta->streamBackend == NULL) {
+    goto _err;
+  }
   pMeta->streamBackendRid = taosAddRef(streamBackendId, pMeta->streamBackend);
 
   taosMemoryFree(streamPath);

source/libs/stream/test/CMakeLists.txt

@@ -8,20 +8,9 @@ AUX_SOURCE_DIRECTORY(${CMAKE_CURRENT_SOURCE_DIR} SOURCE_LIST)
 # bloomFilterTest
 ADD_EXECUTABLE(streamUpdateTest "tstreamUpdateTest.cpp")
 
-#TARGET_LINK_LIBRARIES(
-#  streamUpdateTest
-#  PUBLIC os util common gtest gtest_main stream executor
-#)
-
-IF (TD_GRANT)
-    TARGET_LINK_LIBRARIES(streamUpdateTest
-            PUBLIC os util common gtest gtest_main stream executor index grant
-            )
-ELSE ()
 TARGET_LINK_LIBRARIES(streamUpdateTest
         PUBLIC os util common gtest gtest_main stream executor index
         )
-ENDIF()
 
 TARGET_INCLUDE_DIRECTORIES(
   streamUpdateTest

source/libs/wal/src/walRef.c

@@ -81,26 +81,11 @@ void walRefLastVer(SWal *pWal, SWalRef *pRef) {
   wDebug("vgId:%d, wal ref version %" PRId64 " for last", pWal->cfg.vgId, ver);
 }
 
-SWalRef *walRefCommittedVer(SWal *pWal) {
+void walRefCommitVer(SWal *pWal, SWalRef *pRef) {
-  SWalRef *pRef = walOpenRef(pWal);
-  if (pRef == NULL) {
-    terrno = TSDB_CODE_OUT_OF_MEMORY;
-    return NULL;
-  }
   taosThreadMutexLock(&pWal->mutex);
-
   int64_t ver = walGetCommittedVer(pWal);
-
-  wDebug("vgId:%d, wal ref version %" PRId64 " for committed", pWal->cfg.vgId, ver);
-
   pRef->refVer = ver;
-  // bsearch in fileSet
-  SWalFileInfo tmpInfo;
-  tmpInfo.firstVer = ver;
-  SWalFileInfo *pRet = taosArraySearch(pWal->fileInfoSet, &tmpInfo, compareWalFileInfo, TD_LE);
-  ASSERT(pRet != NULL);
-  //  pRef->refFile = pRet->firstVer;
 
   taosThreadMutexUnlock(&pWal->mutex);
-  return pRef;
+  wDebug("vgId:%d, wal ref version %" PRId64 " for committed", pWal->cfg.vgId, ver);
 }

source/os/CMakeLists.txt

@@ -64,7 +64,7 @@ else()
 endif()
 
 IF (JEMALLOC_ENABLED)
-    target_link_libraries(os PUBLIC -ljemalloc)
+    target_link_libraries(os PUBLIC -L${CMAKE_BINARY_DIR}/build/lib -ljemalloc)
 ENDIF ()
 
 if(${BUILD_TEST})

source/util/src/tarray.c

@@ -476,13 +476,13 @@ int32_t taosEncodeArray(void** buf, const SArray* pArray, FEncode encode) {
   return tlen;
 }
 
-void* taosDecodeArray(const void* buf, SArray** pArray, FDecode decode, int32_t dataSz) {
+void* taosDecodeArray(const void* buf, SArray** pArray, FDecode decode, int32_t dataSz, int8_t sver) {
   int32_t sz;
   buf = taosDecodeFixedI32(buf, &sz);
   *pArray = taosArrayInit(sz, sizeof(void*));
   for (int32_t i = 0; i < sz; i++) {
     void* data = taosMemoryCalloc(1, dataSz);
-    buf = decode(buf, data);
+    buf = decode(buf, data, sver);
     taosArrayPush(*pArray, &data);
   }
   return (void*)buf;

tests/parallel_test/cases.task

@@ -449,6 +449,7 @@
 ,,y,system-test,./pytest.sh python3 ./test.py -f 7-tmq/subscribeStb2.py
 ,,y,system-test,./pytest.sh python3 ./test.py -f 7-tmq/subscribeStb3.py
 ,,y,system-test,./pytest.sh python3 ./test.py -f 7-tmq/subscribeDb0.py -N 3 -n 3
+,,y,system-test,./pytest.sh python3 ./test.py -f 7-tmq/ins_topics_test.py
 
 
 ,,y,system-test,./pytest.sh python3 ./test.py -f 1-insert/delete_stable.py
@@ -823,6 +824,8 @@
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/tagFilter.py
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/projectionDesc.py
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/ts_3398.py -N 3 -n 3
+,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/ts_3405.py -N 3 -n 3
+,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/ts_3423.py -N 3 -n 3
 
 ,,n,system-test,python3 ./test.py -f 2-query/queryQnode.py
 ,,y,system-test,./pytest.sh python3 ./test.py -f 6-cluster/5dnode1mnode.py
@@ -932,7 +935,7 @@
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/csum.py -Q 2
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/mavg.py -Q 2
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/sample.py -Q 2
-,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/function_diff.py -Q 2
+#,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/function_diff.py -Q 2
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/unique.py -Q 2
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/stateduration.py -Q 2
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/function_stateduration.py -Q 2
@@ -1027,7 +1030,7 @@
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/csum.py -Q 3
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/mavg.py -Q 3
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/sample.py -Q 3
-,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/function_diff.py -Q 3
+#,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/function_diff.py -Q 3
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/unique.py -Q 3
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/stateduration.py -Q 3
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/function_stateduration.py -Q 3
@@ -1123,7 +1126,7 @@
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/mavg.py -Q 4
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/sample.py -Q 4
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/cast.py -Q 4
-,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/function_diff.py -Q 4
+#,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/function_diff.py -Q 4
 ,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/unique.py -Q 4
 #,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/stateduration.py -Q 4
 #,,y,system-test,./pytest.sh python3 ./test.py -f 2-query/function_stateduration.py -Q 4

tests/script/tsim/parser/projection_limit_offset.sim

@@ -380,10 +380,10 @@ if $row != 8 then
 endi
 
 sql select diff(k) from tm0
-if $row != 3 then
+if $row != 4 then
    return -1
 endi
-if $data20 != -1 then
+if $data20 != NULL then
    return -1
 endi
 

tests/script/tsim/query/delete_and_query.sim

@@ -0,0 +1,25 @@
+system sh/stop_dnodes.sh
+system sh/deploy.sh -n dnode1 -i 1
+system sh/exec.sh -n dnode1 -s start
+sql connect
+
+sql create database if not exists test
+sql use test
+sql create table t1 (ts timestamp, c2 int)
+sql insert into t1 values(now, 1)
+
+sql delete from t1 where ts is null
+sql delete from t1 where ts < now
+sql select ts from t1 order by ts asc
+
+print ----------rows:  $rows
+if $rows != 0 then
+  return -1
+endi
+
+sql select ts from t1 order by ts desc
+print ----------rows:  $rows
+if $rows != 0 then
+  return -1
+endi
+

tests/script/tsim/query/sys_tbname.sim

@@ -131,4 +131,8 @@ print $rows
 if $rows != 9 then
   return -1
 endi
+
+print =========================== td-24781
+sql select DISTINCT (`precision`) from `information_schema`.`ins_databases` PARTITION BY `precision`
+
 #system sh/exec.sh -n dnode1 -s stop -x SIGINT