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

cmake/cmake.platform

@@ -70,5 +70,46 @@ ELSEIF (${CMAKE_SYSTEM_NAME} MATCHES "Windows")
 
 ENDIF()
 
+IF ("${CPUTYPE}" STREQUAL "")
+  MESSAGE(STATUS "The current platform " ${CMAKE_SYSTEM_PROCESSOR} " is detected")
+
+  IF (CMAKE_SYSTEM_PROCESSOR MATCHES "(amd64)|(AMD64)")
+    MESSAGE(STATUS "The current platform is amd64")
+    SET(PLATFORM_ARCH_STR "amd64")
+  ELSEIF (CMAKE_SYSTEM_PROCESSOR MATCHES "(x86)|(X86)")
+    MESSAGE(STATUS "The current platform is x86")
+    SET(PLATFORM_ARCH_STR "i386")
+  ELSEIF (CMAKE_SYSTEM_PROCESSOR MATCHES "armv7l")
+    MESSAGE(STATUS "The current platform is aarch32")
+    SET(PLATFORM_ARCH_STR "arm")
+  ELSEIF (CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64")
+    MESSAGE(STATUS "The current platform is aarch64")
+    SET(PLATFORM_ARCH_STR "arm64")
+  ENDIF ()
+ELSE ()
+  # if generate ARM version:
+  #   cmake -DCPUTYPE=aarch32 .. or cmake -DCPUTYPE=aarch64
+  IF (${CPUTYPE} MATCHES "aarch32")
+    SET(PLATFORM_ARCH_STR "arm")
+    MESSAGE(STATUS "input cpuType: aarch32")
+  ELSEIF (${CPUTYPE} MATCHES "aarch64")
+    SET(PLATFORM_ARCH_STR "arm64")
+    MESSAGE(STATUS "input cpuType: aarch64")
+  ELSEIF (${CPUTYPE} MATCHES "mips64")
+    SET(PLATFORM_ARCH_STR "mips")
+    MESSAGE(STATUS "input cpuType: mips64")
+  ELSEIF (${CPUTYPE} MATCHES "x64")
+    SET(PLATFORM_ARCH_STR "amd64")
+    MESSAGE(STATUS "input cpuType: x64")
+  ELSEIF (${CPUTYPE} MATCHES "x86")
+    SET(PLATFORM_ARCH_STR "i386")
+    MESSAGE(STATUS "input cpuType: x86")
+  ELSE ()
+    MESSAGE(STATUS "input cpuType unknown " ${CPUTYPE})
+  ENDIF ()
+ENDIF ()
+
+MESSAGE(STATUS "platform arch:" ${PLATFORM_ARCH_STR})
+
 MESSAGE("C Compiler ID: ${CMAKE_C_COMPILER_ID}")
 MESSAGE("CXX Compiler ID: ${CMAKE_CXX_COMPILER_ID}")

docs/en/14-reference/02-rest-api/02-rest-api.mdx

@@ -21,7 +21,7 @@ The following example is in an Ubuntu environment and uses the `curl` tool to ve
 The following example lists all databases on the host h1.taosdata.com. To use it in your environment, replace `h1.taosdata.com` and `6041` (the default port) with the actual running TDengine service FQDN and port number.
 
 ```html
-curl -H 'Authorization: Basic cm9vdDp0YW9zZGF0YQ==' -d 'show databases;' h1.taosdata.com:6041/rest/sql
+curl -L -H "Authorization: Basic cm9vdDp0YW9zZGF0YQ==" -d "show databases;" h1.taosdata.com:6041/rest/sql
 ```
 
 The following return value results indicate that the verification passed.
@@ -106,13 +106,13 @@ The HTTP request's BODY is a complete SQL command, and the data table in the SQL
 Use `curl` to initiate an HTTP request with a custom authentication method, with the following syntax.
 
 ```bash
-curl -H 'Authorization: Basic <TOKEN>' -d '<SQL>' <ip>:<PORT>/rest/sql/[db_name]
+curl -L -H "Authorization: Basic <TOKEN>" -d "<SQL>" <ip>:<PORT>/rest/sql/[db_name]
 ```
 
 Or
 
 ```bash
-curl -u username:password -d '<SQL>' <ip>:<PORT>/rest/sql/[db_name]
+curl -L -u username:password -d "<SQL>" <ip>:<PORT>/rest/sql/[db_name]
 ```
 
 where `TOKEN` is the string after Base64 encoding of `{username}:{password}`, e.g. `root:taosdata` is encoded as `cm9vdDp0YW9zZGF0YQ==`.
@@ -192,7 +192,7 @@ Response body:
 - query all records from table d1001 of database demo
 
   ```bash
-  curl -H 'Authorization: Basic cm9vdDp0YW9zZGF0YQ==' -d 'select * from demo.d1001' 192.168.0.1:6041/rest/sql
+  curl -L -H "Authorization: Basic cm9vdDp0YW9zZGF0YQ==" -d "select * from demo.d1001" 192.168.0.1:6041/rest/sql
   ```
 
   Response body:
@@ -218,7 +218,7 @@ Response body:
 - Create database demo:
 
   ```bash
-  curl -H 'Authorization: Basic cm9vdDp0YW9zZGF0YQ==' -d 'create database demo' 192.168.0.1:6041/rest/sql
+  curl -L -H "Authorization: Basic cm9vdDp0YW9zZGF0YQ==" -d "create database demo" 192.168.0.1:6041/rest/sql
   ```
 
   Response body:
@@ -240,7 +240,7 @@ Response body:
 When the HTTP request URL uses `/rest/sqlt`, the returned result set's timestamp value will be in Unix timestamp format, for example:
 
 ```bash
-curl -H 'Authorization: Basic cm9vdDp0YW9zZGF0YQ==' -d 'select * from demo.d1001' 192.168.0.1:6041/rest/sqlt
+curl -L -H "Authorization: Basic cm9vdDp0YW9zZGF0YQ==" -d "select * from demo.d1001" 192.168.0.1:6041/rest/sqlt
 ```
 
 Response body:
@@ -268,7 +268,7 @@ Response body:
 When the HTTP request URL uses `/rest/sqlutc`, the timestamp of the returned result set will be expressed as a UTC format, for example:
 
 ```bash
-  curl -H 'Authorization: Basic cm9vdDp0YW9zZGF0YQ==' -d 'select * from demo.t1' 192.168.0.1:6041/rest/sqlutc
+  curl -L -H "Authorization: Basic cm9vdDp0YW9zZGF0YQ==" -d "select * from demo.t1" 192.168.0.1:6041/rest/sqlutc
 ```
 
 Response body:

docs/en/14-reference/12-directory.md

@@ -26,7 +26,6 @@ All executable files of TDengine are in the _/usr/local/taos/bin_ directory by d
 - _remove.sh_: script to uninstall TDengine, please execute it carefully, link to the **rmtaos** command in the /usr/bin directory. Will remove the TDengine installation directory `/usr/local/taos`, but will keep `/etc/taos`, `/var/lib/taos`, `/var/log/taos`
 - _taosadapter_: server-side executable that provides RESTful services and accepts writing requests from a variety of other softwares
 - _tarbitrator_: provides arbitration for two-node cluster deployments
-- _run_taosd_and_taosadapter.sh_: script to start both taosd and taosAdapter
 - _TDinsight.sh_: script to download TDinsight and install it
 - _set_core.sh_: script for setting up the system to generate core dump files for easy debugging
 - _taosd-dump-cfg.gdb_: script to facilitate debugging of taosd's gdb execution.

docs/en/27-train-faq/03-docker.md

@@ -109,7 +109,7 @@ taos>
 It's also able to access the REST interface provided by TDengine in container from the host.
 
 ```
-curl -u root:taosdata -d 'show databases' 127.0.0.1:6041/rest/sql
+curl -L -u root:taosdata -d "show databases" 127.0.0.1:6041/rest/sql
 ```
 
 Output is like below:
@@ -147,7 +147,7 @@ docker run -d --name tdengine-taosd -p 6030-6042:6030-6042 -p 6030-6042:6030-604
 - Verify the REST interface:
 
 ```bash
-curl -H 'Authorization: Basic cm9vdDp0YW9zZGF0YQ==' -d 'show databases;' 127.0.0.1:6041/rest/sql
+curl -L -H "Authorization: Basic cm9vdDp0YW9zZGF0YQ==" -d "show databases;" 127.0.0.1:6041/rest/sql
 ```
 
 Below is an example output:

docs/zh/14-reference/02-rest-api/02-rest-api.mdx

@@ -21,7 +21,7 @@ RESTful 接口不依赖于任何 TDengine 的库，因此客户端不需要安
 下面示例是列出所有的数据库，请把 h1.taosdata.com 和 6041（缺省值）替换为实际运行的 TDengine 服务 FQDN 和端口号：
 
 ```html
-curl -H 'Authorization: Basic cm9vdDp0YW9zZGF0YQ==' -d 'show databases;' h1.taosdata.com:6041/rest/sql
+curl -L -H "Authorization: Basic cm9vdDp0YW9zZGF0YQ==" -d "show databases;" h1.taosdata.com:6041/rest/sql
 ```
 
 返回值结果如下表示验证通过：
@@ -106,13 +106,13 @@ HTTP 请求的 BODY 里就是一个完整的 SQL 语句，SQL 语句中的数据
 使用 `curl` 通过自定义身份认证方式来发起一个 HTTP Request，语法如下：
 
 ```bash
-curl -H 'Authorization: Basic <TOKEN>' -d '<SQL>' <ip>:<PORT>/rest/sql/[db_name]
+curl -L -H "Authorization: Basic <TOKEN>" -d "<SQL>" <ip>:<PORT>/rest/sql/[db_name]
 ```
 
 或者
 
 ```bash
-curl -u username:password -d '<SQL>' <ip>:<PORT>/rest/sql/[db_name]
+curl -L -u username:password -d "<SQL>" <ip>:<PORT>/rest/sql/[db_name]
 ```
 
 其中，`TOKEN` 为 `{username}:{password}` 经过 Base64 编码之后的字符串，例如 `root:taosdata` 编码后为 `cm9vdDp0YW9zZGF0YQ==`
@@ -192,7 +192,7 @@ curl http://192.168.0.1:6041/rest/login/root/taosdata
 - 在 demo 库里查询表 d1001 的所有记录：
 
   ```bash
-  curl -H 'Authorization: Basic cm9vdDp0YW9zZGF0YQ==' -d 'select * from demo.d1001' 192.168.0.1:6041/rest/sql
+  curl -L -H "Authorization: Basic cm9vdDp0YW9zZGF0YQ==" -d "select * from demo.d1001" 192.168.0.1:6041/rest/sql
   ```
 
   返回值：
@@ -218,7 +218,7 @@ curl http://192.168.0.1:6041/rest/login/root/taosdata
 - 创建库 demo：
 
   ```bash
-  curl -H 'Authorization: Basic cm9vdDp0YW9zZGF0YQ==' -d 'create database demo' 192.168.0.1:6041/rest/sql
+  curl -L -H "Authorization: Basic cm9vdDp0YW9zZGF0YQ==" -d "create database demo" 192.168.0.1:6041/rest/sql
   ```
 
   返回值：
@@ -240,7 +240,7 @@ curl http://192.168.0.1:6041/rest/login/root/taosdata
 HTTP 请求 URL 采用 `/rest/sqlt` 时，返回结果集的时间戳将采用 Unix 时间戳格式表示，例如
 
 ```bash
-curl -H 'Authorization: Basic cm9vdDp0YW9zZGF0YQ==' -d 'select * from demo.d1001' 192.168.0.1:6041/rest/sqlt
+curl -L -H "Authorization: Basic cm9vdDp0YW9zZGF0YQ==" -d "select * from demo.d1001" 192.168.0.1:6041/rest/sqlt
 ```
 
 返回结果：
@@ -268,7 +268,7 @@ curl -H 'Authorization: Basic cm9vdDp0YW9zZGF0YQ==' -d 'select * from demo.d1001
 HTTP 请求 URL 采用 `/rest/sqlutc` 时，返回结果集的时间戳将采用 UTC 时间字符串表示，例如
 
 ```bash
-  curl -H 'Authorization: Basic cm9vdDp0YW9zZGF0YQ==' -d 'select * from demo.t1' 192.168.0.1:6041/rest/sqlutc
+  curl -L -H "Authorization: Basic cm9vdDp0YW9zZGF0YQ==" -d "select * from demo.t1" 192.168.0.1:6041/rest/sqlutc
 ```
 
 返回值：

docs/zh/14-reference/12-directory.md

@@ -26,7 +26,6 @@ TDengine 的所有可执行文件默认存放在 _/usr/local/taos/bin_ 目录下
 - _remove.sh_：卸载 TDengine 的脚本，请谨慎执行，链接到/usr/bin 目录下的**rmtaos**命令。会删除 TDengine 的安装目录/usr/local/taos，但会保留/etc/taos、/var/lib/taos、/var/log/taos
 - _taosadapter_: 提供 RESTful 服务和接受其他多种软件写入请求的服务端可执行文件
 - _tarbitrator_: 提供双节点集群部署的仲裁功能
-- _run_taosd_and_taosadapter.sh_：同时启动 taosd 和 taosAdapter 的脚本
 - _TDinsight.sh_：用于下载 TDinsight 并安装的脚本
 - _set_core.sh_：用于方便调试设置系统生成 core dump 文件的脚本
 - _taosd-dump-cfg.gdb_：用于方便调试 taosd 的 gdb 执行脚本。

docs/zh/27-train-faq/03-docker.md

@@ -108,7 +108,7 @@ taos>
 也可以在宿主机使用 curl 通过 RESTful 端口访问 Docker 容器内的 TDengine server。
 
 ```
-curl -u root:taosdata -d 'show databases' 127.0.0.1:6041/rest/sql
+curl -L -u root:taosdata -d "show databases" 127.0.0.1:6041/rest/sql
 ```
 
 输出示例如下：
@@ -148,7 +148,7 @@ docker run -d --name tdengine-taosd -p 6030-6042:6030-6042 -p 6030-6042:6030-604
 使用 curl 命令验证 RESTful 接口可以正常工作：
 
 ```bash
-curl -H 'Authorization: Basic cm9vdDp0YW9zZGF0YQ==' -d 'show databases;' 127.0.0.1:6041/rest/sql
+curl -L -H "Authorization: Basic cm9vdDp0YW9zZGF0YQ==" -d "show databases;" 127.0.0.1:6041/rest/sql
 ```
 
 输出示例如下：

include/client/taos.h

@@ -252,6 +252,16 @@ DLL_EXPORT void           tmq_conf_set_auto_commit_cb(tmq_conf_t *conf, tmq_comm
 
 /* -------------------------TMQ MSG HANDLE INTERFACE---------------------- */
 
+enum tmq_res_t {
+  TMQ_RES_INVALID = -1,
+  TMQ_RES_DATA = 1,
+  TMQ_RES_TABLE_META = 2,
+};
+
+typedef enum tmq_res_t tmq_res_t;
+
+DLL_EXPORT tmq_res_t   tmq_get_res_type(TAOS_RES *res);
+DLL_EXPORT int32_t     tmq_get_raw_meta(TAOS_RES *res, const void **raw_meta, int32_t *raw_meta_len);
 DLL_EXPORT const char *tmq_get_topic_name(TAOS_RES *res);
 DLL_EXPORT const char *tmq_get_db_name(TAOS_RES *res);
 DLL_EXPORT int32_t     tmq_get_vgroup_id(TAOS_RES *res);

include/common/tcommon.h

@@ -34,6 +34,7 @@ enum {
 enum {
   TMQ_MSG_TYPE__DUMMY = 0,
   TMQ_MSG_TYPE__POLL_RSP,
+  TMQ_MSG_TYPE__POLL_META_RSP,
   TMQ_MSG_TYPE__EP_RSP,
   TMQ_MSG_TYPE__END_RSP,
 };

include/common/tmsg.h

@@ -1560,6 +1560,7 @@ typedef struct {
   char   name[TSDB_TOPIC_FNAME_LEN];  // accout.topic
   int8_t igExists;
   int8_t subType;
+  int8_t withMeta;
   char*  sql;
   char   subDbName[TSDB_DB_FNAME_LEN];
   union {
@@ -2306,6 +2307,7 @@ typedef struct {
   int64_t newConsumerId;
   char    subKey[TSDB_SUBSCRIBE_KEY_LEN];
   int8_t  subType;
+  int8_t  withMeta;
   char*   qmsg;
   int64_t suid;
 } SMqRebVgReq;
@@ -2318,6 +2320,7 @@ static FORCE_INLINE int32_t tEncodeSMqRebVgReq(void** buf, const SMqRebVgReq* pR
   tlen += taosEncodeFixedI64(buf, pReq->newConsumerId);
   tlen += taosEncodeString(buf, pReq->subKey);
   tlen += taosEncodeFixedI8(buf, pReq->subType);
+  tlen += taosEncodeFixedI8(buf, pReq->withMeta);
   if (pReq->subType == TOPIC_SUB_TYPE__COLUMN) {
     tlen += taosEncodeString(buf, pReq->qmsg);
   } else if (pReq->subType == TOPIC_SUB_TYPE__TABLE) {
@@ -2333,6 +2336,7 @@ static FORCE_INLINE void* tDecodeSMqRebVgReq(const void* buf, SMqRebVgReq* pReq)
   buf = taosDecodeFixedI64(buf, &pReq->newConsumerId);
   buf = taosDecodeStringTo(buf, pReq->subKey);
   buf = taosDecodeFixedI8(buf, &pReq->subType);
+  buf = taosDecodeFixedI8(buf, &pReq->withMeta);
   if (pReq->subType == TOPIC_SUB_TYPE__COLUMN) {
     buf = taosDecodeString(buf, &pReq->qmsg);
   } else if (pReq->subType == TOPIC_SUB_TYPE__TABLE) {
@@ -2677,6 +2681,34 @@ static FORCE_INLINE void tDeleteSMqSubTopicEp(SMqSubTopicEp* pSubTopicEp) {
   taosArrayDestroy(pSubTopicEp->vgs);
 }
 
+typedef struct {
+  SMqRspHead head;
+  int64_t    reqOffset;
+  int64_t    rspOffset;
+  int16_t    resMsgType;
+  int32_t    metaRspLen;
+  void*      metaRsp;
+} SMqMetaRsp;
+
+static FORCE_INLINE int32_t tEncodeSMqMetaRsp(void** buf, const SMqMetaRsp* pRsp) {
+  int32_t tlen = 0;
+  tlen += taosEncodeFixedI64(buf, pRsp->reqOffset);
+  tlen += taosEncodeFixedI64(buf, pRsp->rspOffset);
+  tlen += taosEncodeFixedI16(buf, pRsp->resMsgType);
+  tlen += taosEncodeFixedI32(buf, pRsp->metaRspLen);
+  tlen += taosEncodeBinary(buf, pRsp->metaRsp, pRsp->metaRspLen);
+  return tlen;
+}
+
+static FORCE_INLINE void* tDecodeSMqMetaRsp(const void* buf, SMqMetaRsp* pRsp) {
+  buf = taosDecodeFixedI64(buf, &pRsp->reqOffset);
+  buf = taosDecodeFixedI64(buf, &pRsp->rspOffset);
+  buf = taosDecodeFixedI16(buf, &pRsp->resMsgType);
+  buf = taosDecodeFixedI32(buf, &pRsp->metaRspLen);
+  buf = taosDecodeBinary(buf, &pRsp->metaRsp, pRsp->metaRspLen);
+  return (void*)buf;
+}
+
 typedef struct {
   SMqRspHead head;
   int64_t    reqOffset;

include/common/tmsgdef.h

@@ -187,7 +187,6 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_VND_CONSUME, "vnode-consume", SMqPollReq, SMqDataBlkRsp)
   TD_DEF_MSG_TYPE(TDMT_VND_STREAM_TRIGGER, "vnode-stream-trigger", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_STREAM_DISPATCH_WRITE, "vnode-stream-task-dispatch-write", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_STREAM_TASK_DROP, "vnode-stream-task-drop", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_CREATE_SMA, "vnode-create-sma", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_CANCEL_SMA, "vnode-cancel-sma", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_DROP_SMA, "vnode-drop-sma", NULL, NULL)
@@ -204,6 +203,7 @@ enum {
   //shared by snode and vnode
   TD_NEW_MSG_SEG(TDMT_STREAM_MSG)
   TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_DEPLOY, "stream-task-deploy", SStreamTaskDeployReq, SStreamTaskDeployRsp)
+  TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_DROP, "stream-task-drop", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_RUN, "stream-task-run", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_DISPATCH, "stream-task-dispatch", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_RECOVER, "stream-task-recover", NULL, NULL)

include/libs/stream/tstream.h

@@ -30,9 +30,14 @@ extern "C" {
 typedef struct SStreamTask SStreamTask;
 
 enum {
-  TASK_STATUS__IDLE = 1,
-  TASK_STATUS__EXECUTING,
-  TASK_STATUS__CLOSING,
+  TASK_STATUS__NORMAL = 0,
+  TASK_STATUS__DROPPING,
+};
+
+enum {
+  TASK_EXEC_STATUS__IDLE = 1,
+  TASK_EXEC_STATUS__EXECUTING,
+  TASK_EXEC_STATUS__CLOSING,
 };
 
 enum {
@@ -50,16 +55,12 @@ enum {
   TASK_OUTPUT_STATUS__BLOCKED,
 };
 
-enum {
-  STREAM_CREATED_BY__USER = 1,
-  STREAM_CREATED_BY__SMA,
-};
-
 enum {
   STREAM_INPUT__DATA_SUBMIT = 1,
   STREAM_INPUT__DATA_BLOCK,
   STREAM_INPUT__TRIGGER,
   STREAM_INPUT__CHECKPOINT,
+  STREAM_INPUT__DROP,
 };
 
 typedef struct {
@@ -237,7 +238,9 @@ struct SStreamTask {
   int64_t streamId;
   int32_t taskId;
   int8_t  inputType;
-  int8_t  status;
+  int8_t  taskStatus;
+
+  int8_t execStatus;
 
   int8_t  execType;
   int8_t  sinkType;

packaging/tools/makepkg.sh

@@ -88,7 +88,6 @@ else
       ${build_dir}/bin/tarbitrator\
       ${script_dir}/remove.sh \
       ${script_dir}/set_core.sh \
-      ${script_dir}/run_taosd_and_taosadapter.sh \
       ${script_dir}/startPre.sh \
       ${script_dir}/taosd-dump-cfg.gdb"
 fi
@@ -152,7 +151,6 @@ if [ $adapterName != "taosadapter" ]; then
   sed -i "s/taosadapter/${adapterName}/g" ${install_dir}/cfg/$adapterName.service
 
   mv ${install_dir}/bin/taosadapter ${install_dir}/bin/${adapterName}
-  mv ${install_dir}/bin/run_taosd_and_taosadapter.sh ${install_dir}/bin/run_${serverName}_and_${adapterName}.sh
   mv ${install_dir}/bin/taosd-dump-cfg.gdb ${install_dir}/bin/${serverName}-dump-cfg.gdb
 fi
 

packaging/tools/run_taosd_and_taosadapter.sh

-#!/bin/bash
-[[ -x /usr/bin/taosadapter ]] && /usr/bin/taosadapter &
-taosd

source/client/inc/clientInt.h

@@ -20,9 +20,9 @@
 extern "C" {
 #endif
 
+#include "catalog.h"
 #include "parser.h"
 #include "planner.h"
-#include "catalog.h"
 #include "query.h"
 #include "taos.h"
 #include "tcommon.h"
@@ -51,10 +51,12 @@ extern "C" {
 enum {
   RES_TYPE__QUERY = 1,
   RES_TYPE__TMQ,
+  RES_TYPE__TMQ_META,
 };
 
 #define TD_RES_QUERY(res)    (*(int8_t*)res == RES_TYPE__QUERY)
 #define TD_RES_TMQ(res)      (*(int8_t*)res == RES_TYPE__TMQ)
+#define TD_RES_TMQ_META(res) (*(int8_t*)res == RES_TYPE__TMQ_META)
 
 typedef struct SAppInstInfo SAppInstInfo;
 
@@ -82,7 +84,7 @@ typedef struct {
   int8_t        threadStop;
   TdThread      thread;
   TdThreadMutex lock;  // used when app init and cleanup
-  SHashObj     *appSummary;
+  SHashObj*     appSummary;
   SArray*       appHbMgrs;  // SArray<SAppHbMgr*> one for each cluster
   FHbReqHandle  reqHandle[CONN_TYPE__MAX];
   FHbRspHandle  rspHandle[CONN_TYPE__MAX];
@@ -129,7 +131,7 @@ typedef struct STscObj {
   int8_t        connType;
   int32_t       acctId;
   uint32_t      connId;
-  TAOS         *id;         // ref ID returned by taosAddRef
+  TAOS*         id;         // ref ID returned by taosAddRef
   TdThreadMutex mutex;      // used to protect the operation on db
   int32_t       numOfReqs;  // number of sqlObj bound to this connection
   SAppInstInfo* pAppInfo;
@@ -188,6 +190,14 @@ typedef struct {
   SReqResultInfo resInfo;
 } SMqRspObj;
 
+typedef struct {
+  int8_t     resType;
+  char       topic[TSDB_TOPIC_FNAME_LEN];
+  char       db[TSDB_DB_FNAME_LEN];
+  int32_t    vgId;
+  SMqMetaRsp metaRsp;
+} SMqMetaRspObj;
+
 typedef struct SRequestObj {
   int8_t               resType;  // query or tmq
   uint64_t             requestId;
@@ -207,7 +217,7 @@ typedef struct SRequestObj {
   bool                 stableQuery;
 
   bool     killed;
-  uint32_t             prevCode; //previous error code: todo refactor, add update flag for catalog
+  uint32_t prevCode;  // previous error code: todo refactor, add update flag for catalog
   uint32_t retry;
 } SRequestObj;
 
@@ -299,7 +309,7 @@ int32_t getPlan(SRequestObj* pRequest, SQuery* pQuery, SQueryPlan** pPlan, SArra
 
 int32_t buildRequest(STscObj* pTscObj, const char* sql, int sqlLen, SRequestObj** pRequest);
 
-void taos_close_internal(void *taos);
+void taos_close_internal(void* taos);
 
 // --- heartbeat
 // global, called by mgmt
@@ -320,12 +330,12 @@ void hbMgrInitMqHbRspHandle();
 
 SRequestObj* launchQueryImpl(SRequestObj* pRequest, SQuery* pQuery, bool keepQuery, void** res);
 int32_t      scheduleQuery(SRequestObj* pRequest, SQueryPlan* pDag, SArray* pNodeList);
-void         launchAsyncQuery(SRequestObj* pRequest, SQuery* pQuery, SMetaData *pResultMeta);
+void         launchAsyncQuery(SRequestObj* pRequest, SQuery* pQuery, SMetaData* pResultMeta);
 int32_t      refreshMeta(STscObj* pTscObj, SRequestObj* pRequest);
 int32_t      updateQnodeList(SAppInstInfo* pInfo, SArray* pNodeList);
 void         doAsyncQuery(SRequestObj* pRequest, bool forceUpdateMeta);
-int32_t      removeMeta(STscObj* pTscObj, SArray* tbList);// todo move to clientImpl.c and become a static function
-int32_t      handleAlterTbExecRes(void* res, struct SCatalog* pCatalog);// todo move to xxx
+int32_t      removeMeta(STscObj* pTscObj, SArray* tbList);  // todo move to clientImpl.c and become a static function
+int32_t      handleAlterTbExecRes(void* res, struct SCatalog* pCatalog);  // todo move to xxx
 bool         qnodeRequired(SRequestObj* pRequest);
 
 #ifdef __cplusplus

source/client/src/clientMain.c

@@ -17,6 +17,7 @@
 #include "clientInt.h"
 #include "clientLog.h"
 #include "clientStmt.h"
+#include "functionMgt.h"
 #include "os.h"
 #include "query.h"
 #include "scheduler.h"
@@ -25,7 +26,6 @@
 #include "tref.h"
 #include "trpc.h"
 #include "version.h"
-#include "functionMgt.h"
 
 #define TSC_VAR_NOT_RELEASE 1
 #define TSC_VAR_RELEASED    0
@@ -97,7 +97,7 @@ TAOS *taos_connect(const char *ip, const char *user, const char *pass, const cha
     pass = TSDB_DEFAULT_PASS;
   }
 
-  STscObj* pObj = taos_connect_internal(ip, user, pass, NULL, db, port, CONN_TYPE__QUERY);
+  STscObj *pObj = taos_connect_internal(ip, user, pass, NULL, db, port, CONN_TYPE__QUERY);
   if (pObj) {
     return pObj->id;
   }
@@ -111,9 +111,9 @@ void taos_close_internal(void *taos) {
   }
 
   STscObj *pTscObj = (STscObj *)taos;
-  tscDebug("0x%" PRIx64 " try to close connection, numOfReq:%d", *(int64_t*)pTscObj->id, pTscObj->numOfReqs);
+  tscDebug("0x%" PRIx64 " try to close connection, numOfReq:%d", *(int64_t *)pTscObj->id, pTscObj->numOfReqs);
 
-  taosRemoveRef(clientConnRefPool, *(int64_t*)pTscObj->id);
+  taosRemoveRef(clientConnRefPool, *(int64_t *)pTscObj->id);
 }
 
 void taos_close(TAOS *taos) {
@@ -121,31 +121,30 @@ void taos_close(TAOS *taos) {
     return;
   }
 
-  STscObj* pObj = acquireTscObj(*(int64_t*)taos);
+  STscObj *pObj = acquireTscObj(*(int64_t *)taos);
   if (NULL == pObj) {
     return;
   }
 
   taos_close_internal(pObj);
-  releaseTscObj(*(int64_t*)taos);
+  releaseTscObj(*(int64_t *)taos);
   taosMemoryFree(taos);
 }
 
-
-int taos_errno(TAOS_RES *tres) {
-  if (tres == NULL) {
+int taos_errno(TAOS_RES *res) {
+  if (res == NULL || TD_RES_TMQ_META(res)) {
     return terrno;
   }
 
-  if (TD_RES_TMQ(tres)) {
+  if (TD_RES_TMQ(res)) {
     return 0;
   }
 
-  return ((SRequestObj *)tres)->code;
+  return ((SRequestObj *)res)->code;
 }
 
 const char *taos_errstr(TAOS_RES *res) {
-  if (res == NULL) {
+  if (res == NULL || TD_RES_TMQ_META(res)) {
     return (const char *)tstrerror(terrno);
   }
 
@@ -179,11 +178,15 @@ void taos_free_result(TAOS_RES *res) {
     if (pRsp->rsp.withSchema) taosArrayDestroyP(pRsp->rsp.blockSchema, (FDelete)tDeleteSSchemaWrapper);
     pRsp->resInfo.pRspMsg = NULL;
     doFreeReqResultInfo(&pRsp->resInfo);
+  } else if (TD_RES_TMQ_META(res)) {
+    SMqMetaRspObj *pRspObj = (SMqMetaRspObj *)res;
+    taosMemoryFree(pRspObj->metaRsp.metaRsp);
+    taosMemoryFree(pRspObj);
   }
 }
 
 int taos_field_count(TAOS_RES *res) {
-  if (res == NULL) {
+  if (res == NULL || TD_RES_TMQ_META(res)) {
     return 0;
   }
 
@@ -194,7 +197,7 @@ int taos_field_count(TAOS_RES *res) {
 int taos_num_fields(TAOS_RES *res) { return taos_field_count(res); }
 
 TAOS_FIELD *taos_fetch_fields(TAOS_RES *res) {
-  if (taos_num_fields(res) == 0) {
+  if (taos_num_fields(res) == 0 || TD_RES_TMQ_META(res)) {
     return NULL;
   }
 
@@ -216,7 +219,7 @@ TAOS_RES *taos_query(TAOS *taos, const char *sql) {
     return NULL;
   }
 
-  STscObj* pTscObj = acquireTscObj(*(int64_t*)taos);
+  STscObj *pTscObj = acquireTscObj(*(int64_t *)taos);
   if (pTscObj == NULL || sql == NULL) {
     terrno = TSDB_CODE_TSC_DISCONNECTED;
     return NULL;
@@ -229,21 +232,21 @@ TAOS_RES *taos_query(TAOS *taos, const char *sql) {
   taos_query_a(taos, sql, syncQueryFn, param);
   tsem_wait(&param->sem);
 
-  releaseTscObj(*(int64_t*)taos);
+  releaseTscObj(*(int64_t *)taos);
 
   return param->pRequest;
 #else
   size_t sqlLen = strlen(sql);
   if (sqlLen > (size_t)TSDB_MAX_ALLOWED_SQL_LEN) {
-    releaseTscObj(*(int64_t*)taos);
+    releaseTscObj(*(int64_t *)taos);
     tscError("sql string exceeds max length:%d", TSDB_MAX_ALLOWED_SQL_LEN);
     terrno = TSDB_CODE_TSC_EXCEED_SQL_LIMIT;
     return NULL;
   }
 
-  TAOS_RES* pRes = execQuery(pTscObj, sql, sqlLen);
+  TAOS_RES *pRes = execQuery(pTscObj, sql, sqlLen);
 
-  releaseTscObj(*(int64_t*)taos);
+  releaseTscObj(*(int64_t *)taos);
 
   return pRes;
 #endif
@@ -380,7 +383,7 @@ int taos_print_row(char *str, TAOS_ROW row, TAOS_FIELD *fields, int num_fields)
 }
 
 int *taos_fetch_lengths(TAOS_RES *res) {
-  if (res == NULL) {
+  if (res == NULL || TD_RES_TMQ_META(res)) {
     return NULL;
   }
 
@@ -389,7 +392,7 @@ int *taos_fetch_lengths(TAOS_RES *res) {
 }
 
 TAOS_ROW *taos_result_block(TAOS_RES *res) {
-  if (res == NULL) {
+  if (res == NULL || TD_RES_TMQ_META(res)) {
     terrno = TSDB_CODE_INVALID_PARA;
     return NULL;
   }
@@ -438,7 +441,7 @@ const char *taos_data_type(int type) {
 const char *taos_get_client_info() { return version; }
 
 int taos_affected_rows(TAOS_RES *res) {
-  if (res == NULL || TD_RES_TMQ(res)) {
+  if (res == NULL || TD_RES_TMQ(res) || TD_RES_TMQ_META(res)) {
     return 0;
   }
 
@@ -448,7 +451,7 @@ int taos_affected_rows(TAOS_RES *res) {
 }
 
 int taos_result_precision(TAOS_RES *res) {
-  if (res == NULL) {
+  if (res == NULL || TD_RES_TMQ_META(res)) {
     return TSDB_TIME_PRECISION_MILLI;
   }
 
@@ -463,15 +466,15 @@ int taos_result_precision(TAOS_RES *res) {
 }
 
 int taos_select_db(TAOS *taos, const char *db) {
-  STscObj* pObj = acquireTscObj(*(int64_t*)taos);
+  STscObj *pObj = acquireTscObj(*(int64_t *)taos);
   if (pObj == NULL) {
-    releaseTscObj(*(int64_t*)taos);
+    releaseTscObj(*(int64_t *)taos);
     terrno = TSDB_CODE_TSC_DISCONNECTED;
     return TSDB_CODE_TSC_DISCONNECTED;
   }
 
   if (db == NULL || strlen(db) == 0) {
-    releaseTscObj(*(int64_t*)taos);
+    releaseTscObj(*(int64_t *)taos);
     terrno = TSDB_CODE_TSC_INVALID_INPUT;
     return terrno;
   }
@@ -483,12 +486,12 @@ int taos_select_db(TAOS *taos, const char *db) {
   int32_t   code = taos_errno(pRequest);
 
   taos_free_result(pRequest);
-  releaseTscObj(*(int64_t*)taos);
+  releaseTscObj(*(int64_t *)taos);
   return code;
 }
 
 void taos_stop_query(TAOS_RES *res) {
-  if (res == NULL) {
+  if (res == NULL || TD_RES_TMQ(res) || TD_RES_TMQ_META(res)) {
     return;
   }
 
@@ -510,6 +513,9 @@ void taos_stop_query(TAOS_RES *res) {
 }
 
 bool taos_is_null(TAOS_RES *res, int32_t row, int32_t col) {
+  if (res == NULL || TD_RES_TMQ_META(res)) {
+    return true;
+  }
   SReqResultInfo *pResultInfo = tscGetCurResInfo(res);
   if (col >= pResultInfo->numOfCols || col < 0 || row >= pResultInfo->numOfRows || row < 0) {
     return true;
@@ -532,7 +538,7 @@ int taos_fetch_block(TAOS_RES *res, TAOS_ROW *rows) {
 }
 
 int taos_fetch_block_s(TAOS_RES *res, int *numOfRows, TAOS_ROW *rows) {
-  if (res == NULL) {
+  if (res == NULL || TD_RES_TMQ_META(res)) {
     return 0;
   }
 
@@ -575,7 +581,7 @@ int taos_fetch_block_s(TAOS_RES *res, int *numOfRows, TAOS_ROW *rows) {
 }
 
 int taos_fetch_raw_block(TAOS_RES *res, int *numOfRows, void **pData) {
-  if (res == NULL) {
+  if (res == NULL || TD_RES_TMQ_META(res)) {
     return 0;
   }
 
@@ -615,7 +621,7 @@ int taos_fetch_raw_block(TAOS_RES *res, int *numOfRows, void **pData) {
 }
 
 int *taos_get_column_data_offset(TAOS_RES *res, int columnIndex) {
-  if (res == NULL) {
+  if (res == NULL || TD_RES_TMQ_META(res)) {
     return 0;
   }
 
@@ -636,7 +642,7 @@ int *taos_get_column_data_offset(TAOS_RES *res, int columnIndex) {
 int taos_validate_sql(TAOS *taos, const char *sql) { return true; }
 
 void taos_reset_current_db(TAOS *taos) {
-  STscObj* pTscObj = acquireTscObj(*(int64_t*)taos);
+  STscObj *pTscObj = acquireTscObj(*(int64_t *)taos);
   if (pTscObj == NULL) {
     terrno = TSDB_CODE_TSC_DISCONNECTED;
     return;
@@ -644,17 +650,17 @@ void taos_reset_current_db(TAOS *taos) {
 
   resetConnectDB(pTscObj);
 
-  releaseTscObj(*(int64_t*)taos);
+  releaseTscObj(*(int64_t *)taos);
 }
 
 const char *taos_get_server_info(TAOS *taos) {
-  STscObj* pTscObj = acquireTscObj(*(int64_t*)taos);
+  STscObj *pTscObj = acquireTscObj(*(int64_t *)taos);
   if (pTscObj == NULL) {
     terrno = TSDB_CODE_TSC_DISCONNECTED;
     return NULL;
   }
 
-  releaseTscObj(*(int64_t*)taos);
+  releaseTscObj(*(int64_t *)taos);
 
   return pTscObj->ver;
 }
@@ -723,11 +729,11 @@ void retrieveMetaCallback(SMetaData *pResultMeta, void *param, int32_t code) {
 }
 
 void taos_query_a(TAOS *taos, const char *sql, __taos_async_fn_t fp, void *param) {
-  STscObj* pTscObj = acquireTscObj(*(int64_t*)taos);
+  STscObj *pTscObj = acquireTscObj(*(int64_t *)taos);
   if (pTscObj == NULL || sql == NULL || NULL == fp) {
     terrno = TSDB_CODE_INVALID_PARA;
     if (pTscObj) {
-      releaseTscObj(*(int64_t*)taos);
+      releaseTscObj(*(int64_t *)taos);
     } else {
       terrno = TSDB_CODE_TSC_DISCONNECTED;
     }
@@ -849,8 +855,8 @@ static void fetchCallback(void *pResult, void *param, int32_t code) {
 
   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);
+  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;
@@ -884,6 +890,7 @@ static void fetchCallback(void *pResult, void *param, int32_t code) {
 
 void taos_fetch_rows_a(TAOS_RES *res, __taos_async_fn_t fp, void *param) {
   ASSERT(res != NULL && fp != NULL);
+  ASSERT(TD_RES_QUERY(res));
 
   SRequestObj *pRequest = res;
   pRequest->body.fetchFp = fp;
@@ -910,6 +917,7 @@ void taos_fetch_rows_a(TAOS_RES *res, __taos_async_fn_t fp, void *param) {
 
 void taos_fetch_raw_block_a(TAOS_RES *res, __taos_async_fn_t fp, void *param) {
   ASSERT(res != NULL && fp != NULL);
+  ASSERT(TD_RES_QUERY(res));
   SRequestObj *pRequest = res;
 
   pRequest->body.resInfo.convertUcs4 = false;
@@ -923,6 +931,7 @@ void taos_fetch_raw_block_a(TAOS_RES *res, __taos_async_fn_t fp, void *param) {
 
 const void *taos_get_raw_block(TAOS_RES *res) {
   ASSERT(res != NULL);
+  ASSERT(TD_RES_QUERY(res));
   SRequestObj *pRequest = res;
 
   return pRequest->body.resInfo.pData;
@@ -949,16 +958,16 @@ int taos_load_table_info(TAOS *taos, const char *tableNameList) {
 }
 
 TAOS_STMT *taos_stmt_init(TAOS *taos) {
-  STscObj* pObj = acquireTscObj(*(int64_t*)taos);
+  STscObj *pObj = acquireTscObj(*(int64_t *)taos);
   if (NULL == pObj) {
     tscError("invalid parameter for %s", __FUNCTION__);
     terrno = TSDB_CODE_TSC_DISCONNECTED;
     return NULL;
   }
 
-  TAOS_STMT* pStmt = stmtInit(pObj);
+  TAOS_STMT *pStmt = stmtInit(pObj);
 
-  releaseTscObj(*(int64_t*)taos);
+  releaseTscObj(*(int64_t *)taos);
 
   return pStmt;
 }

source/client/src/tmq.c

@@ -149,7 +149,10 @@ typedef struct {
   int32_t         epoch;
   SMqClientVg*    vgHandle;
   SMqClientTopic* topicHandle;
-  SMqDataBlkRsp   msg;
+  union {
+    SMqDataBlkRsp dataRsp;
+    SMqMetaRsp    metaRsp;
+  };
 } SMqPollRspWrapper;
 
 typedef struct {
@@ -1131,6 +1134,11 @@ int32_t tmqPollCb(void* param, const SDataBuf* pMsg, int32_t code) {
     tscWarn("mismatch rsp from vg %d, epoch %d, current epoch %d", vgId, msgEpoch, tmqEpoch);
   }
 
+  // handle meta rsp
+  int8_t rspType = ((SMqRspHead*)pMsg->pData)->mqMsgType;
+  if (rspType == TMQ_MSG_TYPE__POLL_META_RSP) {
+  }
+
   SMqPollRspWrapper* pRspWrapper = taosAllocateQitem(sizeof(SMqPollRspWrapper), DEF_QITEM);
   if (pRspWrapper == NULL) {
     taosMemoryFree(pMsg->pData);
@@ -1138,17 +1146,23 @@ int32_t tmqPollCb(void* param, const SDataBuf* pMsg, int32_t code) {
     goto CREATE_MSG_FAIL;
   }
 
-  pRspWrapper->tmqRspType = TMQ_MSG_TYPE__POLL_RSP;
+  pRspWrapper->tmqRspType = rspType;
   pRspWrapper->vgHandle = pVg;
   pRspWrapper->topicHandle = pTopic;
 
-  memcpy(&pRspWrapper->msg, pMsg->pData, sizeof(SMqRspHead));
+  memcpy(&pRspWrapper->dataRsp, pMsg->pData, sizeof(SMqRspHead));
+
+  if (rspType == TMQ_MSG_TYPE__POLL_RSP) {
+    tDecodeSMqDataBlkRsp(POINTER_SHIFT(pMsg->pData, sizeof(SMqRspHead)), &pRspWrapper->dataRsp);
+  } else {
+    ASSERT(rspType == TMQ_MSG_TYPE__POLL_META_RSP);
+    tDecodeSMqMetaRsp(POINTER_SHIFT(pMsg->pData, sizeof(SMqRspHead)), &pRspWrapper->metaRsp);
+  }
 
-  tDecodeSMqDataBlkRsp(POINTER_SHIFT(pMsg->pData, sizeof(SMqRspHead)), &pRspWrapper->msg);
   taosMemoryFree(pMsg->pData);
 
   tscDebug("consumer %ld recv poll: vg %d, req offset %ld, rsp offset %ld", tmq->consumerId, pVg->vgId,
-           pRspWrapper->msg.reqOffset, pRspWrapper->msg.rspOffset);
+           pRspWrapper->dataRsp.reqOffset, pRspWrapper->dataRsp.rspOffset);
 
   taosWriteQitem(tmq->mqueue, pRspWrapper);
   tsem_post(&tmq->rspSem);
@@ -1516,6 +1530,17 @@ SMqPollReq* tmqBuildConsumeReqImpl(tmq_t* tmq, int64_t timeout, SMqClientTopic*
   return pReq;
 }
 
+SMqMetaRspObj* tmqBuildMetaRspFromWrapper(SMqPollRspWrapper* pWrapper) {
+  SMqMetaRspObj* pRspObj = taosMemoryCalloc(1, sizeof(SMqMetaRspObj));
+  pRspObj->resType = RES_TYPE__TMQ;
+  tstrncpy(pRspObj->topic, pWrapper->topicHandle->topicName, TSDB_TOPIC_FNAME_LEN);
+  tstrncpy(pRspObj->db, pWrapper->topicHandle->db, TSDB_DB_FNAME_LEN);
+  pRspObj->vgId = pWrapper->vgHandle->vgId;
+
+  memcpy(&pRspObj->metaRsp, &pWrapper->metaRsp, sizeof(SMqMetaRsp));
+  return pRspObj;
+}
+
 SMqRspObj* tmqBuildRspFromWrapper(SMqPollRspWrapper* pWrapper) {
   SMqRspObj* pRspObj = taosMemoryCalloc(1, sizeof(SMqRspObj));
   pRspObj->resType = RES_TYPE__TMQ;
@@ -1523,11 +1548,11 @@ SMqRspObj* tmqBuildRspFromWrapper(SMqPollRspWrapper* pWrapper) {
   tstrncpy(pRspObj->db, pWrapper->topicHandle->db, TSDB_DB_FNAME_LEN);
   pRspObj->vgId = pWrapper->vgHandle->vgId;
   pRspObj->resIter = -1;
-  memcpy(&pRspObj->rsp, &pWrapper->msg, sizeof(SMqDataBlkRsp));
+  memcpy(&pRspObj->rsp, &pWrapper->dataRsp, sizeof(SMqDataBlkRsp));
 
   pRspObj->resInfo.totalRows = 0;
   pRspObj->resInfo.precision = TSDB_TIME_PRECISION_MILLI;
-  if (!pWrapper->msg.withSchema) {
+  if (!pWrapper->dataRsp.withSchema) {
     setResSchemaInfo(&pRspObj->resInfo, pWrapper->topicHandle->schema.pSchema, pWrapper->topicHandle->schema.nCols);
   }
 
@@ -1643,12 +1668,12 @@ SMqRspObj* tmqHandleAllRsp(tmq_t* tmq, int64_t timeout, bool pollIfReset) {
       SMqPollRspWrapper* pollRspWrapper = (SMqPollRspWrapper*)rspWrapper;
       /*atomic_sub_fetch_32(&tmq->readyRequest, 1);*/
       int32_t consumerEpoch = atomic_load_32(&tmq->epoch);
-      if (pollRspWrapper->msg.head.epoch == consumerEpoch) {
+      if (pollRspWrapper->dataRsp.head.epoch == consumerEpoch) {
         SMqClientVg* pVg = pollRspWrapper->vgHandle;
         /*printf("vg %d offset %ld up to %ld\n", pVg->vgId, pVg->currentOffset, rspMsg->msg.rspOffset);*/
-        pVg->currentOffset = pollRspWrapper->msg.rspOffset;
+        pVg->currentOffset = pollRspWrapper->dataRsp.rspOffset;
         atomic_store_32(&pVg->vgStatus, TMQ_VG_STATUS__IDLE);
-        if (pollRspWrapper->msg.blockNum == 0) {
+        if (pollRspWrapper->dataRsp.blockNum == 0) {
           taosFreeQitem(pollRspWrapper);
           rspWrapper = NULL;
           continue;
@@ -1658,8 +1683,25 @@ SMqRspObj* tmqHandleAllRsp(tmq_t* tmq, int64_t timeout, bool pollIfReset) {
         taosFreeQitem(pollRspWrapper);
         return pRsp;
       } else {
-        tscDebug("msg discard since epoch mismatch: msg epoch %d, consumer epoch %d\n", pollRspWrapper->msg.head.epoch,
-                 consumerEpoch);
+        tscDebug("msg discard since epoch mismatch: msg epoch %d, consumer epoch %d\n",
+                 pollRspWrapper->dataRsp.head.epoch, consumerEpoch);
+        taosFreeQitem(pollRspWrapper);
+      }
+    } else if (rspWrapper->tmqRspType == TMQ_MSG_TYPE__POLL_META_RSP) {
+      SMqPollRspWrapper* pollRspWrapper = (SMqPollRspWrapper*)rspWrapper;
+      int32_t            consumerEpoch = atomic_load_32(&tmq->epoch);
+      if (pollRspWrapper->dataRsp.head.epoch == consumerEpoch) {
+        SMqClientVg* pVg = pollRspWrapper->vgHandle;
+        /*printf("vg %d offset %ld up to %ld\n", pVg->vgId, pVg->currentOffset, rspMsg->msg.rspOffset);*/
+        pVg->currentOffset = pollRspWrapper->dataRsp.rspOffset;
+        atomic_store_32(&pVg->vgStatus, TMQ_VG_STATUS__IDLE);
+        // build rsp
+        SMqRspObj* pRsp = tmqBuildRspFromWrapper(pollRspWrapper);
+        taosFreeQitem(pollRspWrapper);
+        return pRsp;
+      } else {
+        tscDebug("msg discard since epoch mismatch: msg epoch %d, consumer epoch %d\n",
+                 pollRspWrapper->dataRsp.head.epoch, consumerEpoch);
         taosFreeQitem(pollRspWrapper);
       }
     } else {
@@ -1747,10 +1789,23 @@ const char* tmq_err2str(int32_t err) {
   }
 }
 
+tmq_res_t tmq_get_res_type(TAOS_RES* res) {
+  if (TD_RES_TMQ(res)) {
+    return TMQ_RES_DATA;
+  } else if (TD_RES_TMQ_META(res)) {
+    return TMQ_RES_TABLE_META;
+  } else {
+    return TMQ_RES_INVALID;
+  }
+}
+
 const char* tmq_get_topic_name(TAOS_RES* res) {
   if (TD_RES_TMQ(res)) {
     SMqRspObj* pRspObj = (SMqRspObj*)res;
     return strchr(pRspObj->topic, '.') + 1;
+  } else if (TD_RES_TMQ_META(res)) {
+    SMqMetaRspObj* pMetaRspObj = (SMqMetaRspObj*)res;
+    return strchr(pMetaRspObj->topic, '.') + 1;
   } else {
     return NULL;
   }
@@ -1760,6 +1815,9 @@ const char* tmq_get_db_name(TAOS_RES* res) {
   if (TD_RES_TMQ(res)) {
     SMqRspObj* pRspObj = (SMqRspObj*)res;
     return strchr(pRspObj->db, '.') + 1;
+  } else if (TD_RES_TMQ_META(res)) {
+    SMqMetaRspObj* pMetaRspObj = (SMqMetaRspObj*)res;
+    return strchr(pMetaRspObj->db, '.') + 1;
   } else {
     return NULL;
   }
@@ -1769,6 +1827,9 @@ int32_t tmq_get_vgroup_id(TAOS_RES* res) {
   if (TD_RES_TMQ(res)) {
     SMqRspObj* pRspObj = (SMqRspObj*)res;
     return pRspObj->vgId;
+  } else if (TD_RES_TMQ_META(res)) {
+    SMqMetaRspObj* pMetaRspObj = (SMqMetaRspObj*)res;
+    return pMetaRspObj->vgId;
   } else {
     return -1;
   }
@@ -1786,6 +1847,16 @@ const char* tmq_get_table_name(TAOS_RES* res) {
   return NULL;
 }
 
+int32_t tmq_get_raw_meta(TAOS_RES* res, const void** raw_meta, int32_t* raw_meta_len) {
+  if (TD_RES_TMQ_META(res)) {
+    SMqMetaRspObj* pMetaRspObj = (SMqMetaRspObj*)res;
+    *raw_meta = pMetaRspObj->metaRsp.metaRsp;
+    *raw_meta_len = pMetaRspObj->metaRsp.metaRspLen;
+    return 0;
+  }
+  return -1;
+}
+
 void tmq_commit_async(tmq_t* tmq, const TAOS_RES* msg, tmq_commit_cb* cb, void* param) {
   tmqCommitInner2(tmq, msg, 0, 1, cb, param);
 }

source/common/src/tmsg.c

@@ -2956,6 +2956,7 @@ int32_t tSerializeSCMCreateTopicReq(void *buf, int32_t bufLen, const SCMCreateTo
   if (tEncodeCStr(&encoder, pReq->name) < 0) return -1;
   if (tEncodeI8(&encoder, pReq->igExists) < 0) return -1;
   if (tEncodeI8(&encoder, pReq->subType) < 0) return -1;
+  if (tEncodeI8(&encoder, pReq->withMeta) < 0) return -1;
   if (tEncodeCStr(&encoder, pReq->subDbName) < 0) return -1;
   if (TOPIC_SUB_TYPE__DB == pReq->subType) {
   } else if (TOPIC_SUB_TYPE__TABLE == pReq->subType) {
@@ -2985,6 +2986,7 @@ int32_t tDeserializeSCMCreateTopicReq(void *buf, int32_t bufLen, SCMCreateTopicR
   if (tDecodeCStrTo(&decoder, pReq->name) < 0) return -1;
   if (tDecodeI8(&decoder, &pReq->igExists) < 0) return -1;
   if (tDecodeI8(&decoder, &pReq->subType) < 0) return -1;
+  if (tDecodeI8(&decoder, &pReq->withMeta) < 0) return -1;
   if (tDecodeCStrTo(&decoder, pReq->subDbName) < 0) return -1;
   if (TOPIC_SUB_TYPE__DB == pReq->subType) {
   } else if (TOPIC_SUB_TYPE__TABLE == pReq->subType) {

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

@@ -73,6 +73,7 @@ int32_t mmProcessGetLoadsReq(SMnodeMgmt *pMgmt, SRpcMsg *pMsg) {
 }
 
 int32_t mmProcessCreateReq(const SMgmtInputOpt *pInput, SRpcMsg *pMsg) {
+  const STraceId  *trace = &pMsg->info.traceId;
   SDCreateMnodeReq createReq = {0};
   if (tDeserializeSDCreateMnodeReq(pMsg->pCont, pMsg->contLen, &createReq) != 0) {
     terrno = TSDB_CODE_INVALID_MSG;
@@ -81,7 +82,7 @@ int32_t mmProcessCreateReq(const SMgmtInputOpt *pInput, SRpcMsg *pMsg) {
 
   if (createReq.replica != 1) {
     terrno = TSDB_CODE_INVALID_OPTION;
-    dError("failed to create mnode since %s", terrstr());
+    dGError("failed to create mnode since %s", terrstr());
     return -1;
   }
 
@@ -91,7 +92,7 @@ int32_t mmProcessCreateReq(const SMgmtInputOpt *pInput, SRpcMsg *pMsg) {
   mgmt.path = pInput->path;
   mgmt.name = pInput->name;
   if (mmWriteFile(&mgmt, &createReq.replicas[0], deployed) != 0) {
-    dError("failed to write mnode file since %s", terrstr());
+    dGError("failed to write mnode file since %s", terrstr());
     return -1;
   }
 
@@ -99,6 +100,7 @@ int32_t mmProcessCreateReq(const SMgmtInputOpt *pInput, SRpcMsg *pMsg) {
 }
 
 int32_t mmProcessDropReq(const SMgmtInputOpt *pInput, SRpcMsg *pMsg) {
+  const STraceId *trace = &pMsg->info.traceId;
   SDDropMnodeReq  dropReq = {0};
   if (tDeserializeSCreateDropMQSBNodeReq(pMsg->pCont, pMsg->contLen, &dropReq) != 0) {
     terrno = TSDB_CODE_INVALID_MSG;
@@ -107,7 +109,7 @@ int32_t mmProcessDropReq(const SMgmtInputOpt *pInput, SRpcMsg *pMsg) {
 
   if (pInput->pData->dnodeId != 0 && dropReq.dnodeId != pInput->pData->dnodeId) {
     terrno = TSDB_CODE_INVALID_OPTION;
-    dError("failed to drop mnode since %s", terrstr());
+    dGError("failed to drop mnode since %s", terrstr());
     return -1;
   }
 
@@ -117,7 +119,7 @@ int32_t mmProcessDropReq(const SMgmtInputOpt *pInput, SRpcMsg *pMsg) {
   mgmt.path = pInput->path;
   mgmt.name = pInput->name;
   if (mmWriteFile(&mgmt, NULL, deployed) != 0) {
-    dError("failed to write mnode file since %s", terrstr());
+    dGError("failed to write mnode file since %s", terrstr());
     return -1;
   }
 
@@ -216,7 +218,7 @@ SArray *mmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_VG_DELETE_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_DROP_TASK, mmPutMsgToFetchQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DEPLOY_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_STREAM_TASK_DROP_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DROP_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_CONFIG_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_REPLICA_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_CONFIRM_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;

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

@@ -18,7 +18,6 @@
 
 static inline int32_t mmAcquire(SMnodeMgmt *pMgmt) {
   int32_t code = 0;
-
   taosThreadRwlockRdlock(&pMgmt->lock);
   if (pMgmt->stopped) {
     code = -1;
@@ -48,7 +47,8 @@ static inline void mmSendRsp(SRpcMsg *pMsg, int32_t code) {
 static void mmProcessRpcMsg(SQueueInfo *pInfo, SRpcMsg *pMsg) {
   SMnodeMgmt *pMgmt = pInfo->ahandle;
   int32_t     code = -1;
-  STraceId *  trace = &pMsg->info.traceId;
+
+  const STraceId *trace = &pMsg->info.traceId;
   dGTrace("msg:%p, get from mnode queue", pMsg);
 
   switch (pMsg->msgType) {
@@ -72,7 +72,7 @@ static void mmProcessRpcMsg(SQueueInfo *pInfo, SRpcMsg *pMsg) {
     mndPostProcessQueryMsg(pMsg);
   }
 
-  dTrace("msg:%p, is freed, code:0x%x", pMsg, code);
+  dGTrace("msg:%p, is freed, code:0x%x", pMsg, code);
   rpcFreeCont(pMsg->pCont);
   taosFreeQitem(pMsg);
 }
@@ -80,7 +80,9 @@ static void mmProcessRpcMsg(SQueueInfo *pInfo, SRpcMsg *pMsg) {
 static void mmProcessSyncMsg(SQueueInfo *pInfo, SRpcMsg *pMsg) {
   SMnodeMgmt *pMgmt = pInfo->ahandle;
   pMsg->info.node = pMgmt->pMnode;
-  dTrace("msg:%p, get from mnode-sync queue", pMsg);
+
+  const STraceId *trace = &pMsg->info.traceId;
+  dGTrace("msg:%p, get from mnode-sync queue", pMsg);
 
   SMsgHead *pHead = pMsg->pCont;
   pHead->contLen = ntohl(pHead->contLen);
@@ -88,19 +90,21 @@ static void mmProcessSyncMsg(SQueueInfo *pInfo, SRpcMsg *pMsg) {
 
   int32_t code = mndProcessSyncMsg(pMsg);
 
-  dTrace("msg:%p, is freed, code:0x%x", pMsg, code);
+  dGTrace("msg:%p, is freed, code:0x%x", pMsg, code);
   rpcFreeCont(pMsg->pCont);
   taosFreeQitem(pMsg);
 }
 
 static inline int32_t mmPutMsgToWorker(SMnodeMgmt *pMgmt, SSingleWorker *pWorker, SRpcMsg *pMsg) {
+  const STraceId *trace = &pMsg->info.traceId;
+
   if (mmAcquire(pMgmt) == 0) {
-    dTrace("msg:%p, put into %s queue, type:%s", pMsg, pWorker->name, TMSG_INFO(pMsg->msgType));
+    dGTrace("msg:%p, put into %s queue, type:%s", pMsg, pWorker->name, TMSG_INFO(pMsg->msgType));
     taosWriteQitem(pWorker->queue, pMsg);
     mmRelease(pMgmt);
     return 0;
   } else {
-    dTrace("msg:%p, failed to put into %s queue since %s, type:%s", pMsg, pWorker->name, terrstr(),
+    dGTrace("msg:%p, failed to put into %s queue since %s, type:%s", pMsg, pWorker->name, terrstr(),
             TMSG_INFO(pMsg->msgType));
     return -1;
   }
@@ -121,19 +125,17 @@ int32_t mmPutMsgToReadQueue(SMnodeMgmt *pMgmt, SRpcMsg *pMsg) {
 int32_t mmPutMsgToQueryQueue(SMnodeMgmt *pMgmt, SRpcMsg *pMsg) {
   pMsg->info.node = pMgmt->pMnode;
   if (mndPreProcessQueryMsg(pMsg) != 0) {
-    dError("msg:%p, failed to pre-process in mnode since %s, type:%s", pMsg, terrstr(), TMSG_INFO(pMsg->msgType));
+    const STraceId *trace = &pMsg->info.traceId;
+    dGError("msg:%p, failed to pre-process in mnode since %s, type:%s", pMsg, terrstr(), TMSG_INFO(pMsg->msgType));
     return -1;
   }
   return mmPutMsgToWorker(pMgmt, &pMgmt->queryWorker, pMsg);
 }
 
 int32_t mmPutMsgToFetchQueue(SMnodeMgmt *pMgmt, SRpcMsg *pMsg) {
-  pMsg->info.node = pMgmt->pMnode;
-
   return mmPutMsgToWorker(pMgmt, &pMgmt->fetchWorker, pMsg);
 }
 
-
 int32_t mmPutMsgToMonitorQueue(SMnodeMgmt *pMgmt, SRpcMsg *pMsg) {
   return mmPutMsgToWorker(pMgmt, &pMgmt->monitorWorker, pMsg);
 }

source/dnode/mgmt/mgmt_snode/src/smHandle.c

@@ -95,7 +95,7 @@ SArray *smGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_MON_SM_INFO, smPutNodeMsgToMonitorQueue, 0) == NULL) goto _OVER;
 
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DEPLOY, smPutNodeMsgToMgmtQueue, 1) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_STREAM_TASK_DROP, smPutNodeMsgToMgmtQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DROP, smPutNodeMsgToMgmtQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RUN, smPutNodeMsgToSharedQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DISPATCH, smPutNodeMsgToSharedQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DISPATCH_RSP, smPutNodeMsgToSharedQueue, 1) == NULL) goto _OVER;

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

@@ -351,7 +351,7 @@ SArray *vmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_VND_QUERY_HEARTBEAT, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
 
   if (dmSetMgmtHandle(pArray, TDMT_VND_STREAM_TRIGGER, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_STREAM_TASK_DROP, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DROP, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DEPLOY, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RUN, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DISPATCH, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;

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

@@ -17,9 +17,44 @@
 #include "dmMgmt.h"
 #include "qworker.h"
 
-static void dmSendRedirectRsp(SRpcMsg *pMsg, const SEpSet *pNewEpSet);
-static void dmSendRsp(SRpcMsg *pMsg);
-static void dmBuildMnodeRedirectRsp(SDnode *pDnode, SRpcMsg *pMsg);
+static inline void dmSendRsp(SRpcMsg *pMsg) {
+  SMgmtWrapper *pWrapper = pMsg->info.wrapper;
+  if (InChildProc(pWrapper)) {
+    dmPutToProcPQueue(&pWrapper->proc, pMsg, DND_FUNC_RSP);
+  } else {
+    rpcSendResponse(pMsg);
+  }
+}
+
+static inline void dmBuildMnodeRedirectRsp(SDnode *pDnode, SRpcMsg *pMsg) {
+  SEpSet epSet = {0};
+  dmGetMnodeEpSetForRedirect(&pDnode->data, pMsg, &epSet);
+
+  const int32_t contLen = tSerializeSEpSet(NULL, 0, &epSet);
+  pMsg->pCont = rpcMallocCont(contLen);
+  if (pMsg->pCont == NULL) {
+    pMsg->code = TSDB_CODE_OUT_OF_MEMORY;
+  } else {
+    tSerializeSEpSet(pMsg->pCont, contLen, &epSet);
+    pMsg->contLen = contLen;
+  }
+}
+
+static inline void dmSendRedirectRsp(SRpcMsg *pMsg, const SEpSet *pNewEpSet) {
+  SRpcMsg rsp = {.code = TSDB_CODE_RPC_REDIRECT, .info = pMsg->info};
+  int32_t contLen = tSerializeSEpSet(NULL, 0, pNewEpSet);
+
+  rsp.pCont = rpcMallocCont(contLen);
+  if (rsp.pCont == NULL) {
+    pMsg->code = TSDB_CODE_OUT_OF_MEMORY;
+  } else {
+    tSerializeSEpSet(rsp.pCont, contLen, pNewEpSet);
+    rsp.contLen = contLen;
+  }
+  dmSendRsp(&rsp);
+  rpcFreeCont(pMsg->pCont);
+  pMsg->pCont = NULL;
+}
 
 int32_t dmProcessNodeMsg(SMgmtWrapper *pWrapper, SRpcMsg *pMsg) {
   NodeMsgFp msgFp = pWrapper->msgFps[TMSG_INDEX(pMsg->msgType)];
@@ -28,31 +63,38 @@ int32_t dmProcessNodeMsg(SMgmtWrapper *pWrapper, SRpcMsg *pMsg) {
     return -1;
   }
 
-  dTrace("msg:%p, will be processed by %s", pMsg, pWrapper->name);
+  const STraceId *trace = &pMsg->info.traceId;
+  dGTrace("msg:%p, will be processed by %s", pMsg, pWrapper->name);
   pMsg->info.wrapper = pWrapper;
   return (*msgFp)(pWrapper->pMgmt, pMsg);
 }
 
 static void dmProcessRpcMsg(SDnode *pDnode, SRpcMsg *pRpc, SEpSet *pEpSet) {
-  SDnodeTrans * pTrans = &pDnode->trans;
+  SDnodeTrans  *pTrans = &pDnode->trans;
   int32_t       code = -1;
-  SRpcMsg *     pMsg = NULL;
+  SRpcMsg      *pMsg = NULL;
   SMgmtWrapper *pWrapper = NULL;
   SDnodeHandle *pHandle = &pTrans->msgHandles[TMSG_INDEX(pRpc->msgType)];
 
-  STraceId *trace = &pRpc->info.traceId;
+  const STraceId *trace = &pRpc->info.traceId;
   dGTrace("msg:%s is received, handle:%p len:%d code:0x%x app:%p refId:%" PRId64, TMSG_INFO(pRpc->msgType),
           pRpc->info.handle, pRpc->contLen, pRpc->code, pRpc->info.ahandle, pRpc->info.refId);
 
-  if (pRpc->msgType == TDMT_DND_NET_TEST) {
+  switch (pRpc->msgType) {
+    case TDMT_DND_NET_TEST:
       dmProcessNetTestReq(pDnode, pRpc);
       return;
-  } else if (pRpc->msgType == TDMT_MND_SYSTABLE_RETRIEVE_RSP || pRpc->msgType == TDMT_VND_FETCH_RSP) {
+    case TDMT_MND_SYSTABLE_RETRIEVE_RSP:
+    case TDMT_VND_FETCH_RSP:
       qWorkerProcessFetchRsp(NULL, NULL, pRpc, 0);
       return;
-  } else if (pRpc->msgType == TDMT_MND_STATUS_RSP && pEpSet != NULL) {
+    case TDMT_MND_STATUS_RSP:
+      if (pEpSet != NULL) {
         dmSetMnodeEpSet(&pDnode->data, pEpSet);
-  } else {
+      }
+      break;
+    default:
+      break;
   }
 
   if (pDnode->status != DND_STAT_RUNNING) {
@@ -73,39 +115,43 @@ static void dmProcessRpcMsg(SDnode *pDnode, SRpcMsg *pRpc, SEpSet *pEpSet) {
   if (pHandle->defaultNtype == NODE_END) {
     terrno = TSDB_CODE_MSG_NOT_PROCESSED;
     goto _OVER;
-  } else {
+  }
+
   pWrapper = &pDnode->wrappers[pHandle->defaultNtype];
   if (pHandle->needCheckVgId) {
     if (pRpc->contLen > 0) {
-        SMsgHead *pHead = pRpc->pCont;
-        int32_t   vgId = ntohl(pHead->vgId);
-        if (vgId == QNODE_HANDLE) {
+      const SMsgHead *pHead = pRpc->pCont;
+      const int32_t   vgId = ntohl(pHead->vgId);
+      switch (vgId) {
+        case QNODE_HANDLE:
           pWrapper = &pDnode->wrappers[QNODE];
-        } else if (vgId == SNODE_HANDLE) {
+          break;
+        case SNODE_HANDLE:
           pWrapper = &pDnode->wrappers[SNODE];
-        } else if (vgId == MNODE_HANDLE) {
+          break;
+        case MNODE_HANDLE:
           pWrapper = &pDnode->wrappers[MNODE];
-        } else {
+          break;
+        default:
+          break;
       }
     } else {
       terrno = TSDB_CODE_INVALID_MSG_LEN;
       goto _OVER;
     }
   }
-  }
 
   if (dmMarkWrapper(pWrapper) != 0) {
     pWrapper = NULL;
     goto _OVER;
-  } else {
-    pRpc->info.wrapper = pWrapper;
   }
 
+  pRpc->info.wrapper = pWrapper;
   pMsg = taosAllocateQitem(sizeof(SRpcMsg), RPC_QITEM);
   if (pMsg == NULL) goto _OVER;
-  memcpy(pMsg, pRpc, sizeof(SRpcMsg));
 
-  dTrace("msg:%p, is created, type:%s handle:%p", pMsg, TMSG_INFO(pRpc->msgType), pMsg->info.handle);
+  memcpy(pMsg, pRpc, sizeof(SRpcMsg));
+  dGTrace("msg:%p, is created, type:%s handle:%p", pMsg, TMSG_INFO(pRpc->msgType), pMsg->info.handle);
 
   if (InParentProc(pWrapper)) {
     code = dmPutToProcCQueue(&pWrapper->proc, pMsg, DND_FUNC_REQ);
@@ -115,13 +161,11 @@ static void dmProcessRpcMsg(SDnode *pDnode, SRpcMsg *pRpc, SEpSet *pEpSet) {
 
 _OVER:
   if (code != 0) {
-    dTrace("failed to process msg:%p since %s, handle:%p", pMsg, terrstr(), pRpc->info.handle);
-
     if (terrno != 0) code = terrno;
+    dGTrace("msg:%p, failed to process since %s", pMsg, terrstr());
 
     if (IsReq(pRpc)) {
       SRpcMsg rsp = {.code = code, .info = pRpc->info};
-
       if ((code == TSDB_CODE_NODE_NOT_DEPLOYED || code == TSDB_CODE_APP_NOT_READY) && pRpc->msgType > TDMT_MND_MSG &&
           pRpc->msgType < TDMT_VND_MSG) {
         dmBuildMnodeRedirectRsp(pDnode, &rsp);
@@ -135,7 +179,7 @@ _OVER:
     }
 
     if (pMsg != NULL) {
-      dTrace("msg:%p, is freed", pMsg);
+      dGTrace("msg:%p, is freed", pMsg);
       taosFreeQitem(pMsg);
     }
     rpcFreeCont(pRpc->pCont);
@@ -149,11 +193,11 @@ int32_t dmInitMsgHandle(SDnode *pDnode) {
 
   for (EDndNodeType ntype = DNODE; ntype < NODE_END; ++ntype) {
     SMgmtWrapper *pWrapper = &pDnode->wrappers[ntype];
-    SArray *      pArray = (*pWrapper->func.getHandlesFp)();
+    SArray       *pArray = (*pWrapper->func.getHandlesFp)();
     if (pArray == NULL) return -1;
 
     for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {
-      SMgmtHandle * pMgmt = taosArrayGet(pArray, i);
+      SMgmtHandle  *pMgmt = taosArrayGet(pArray, i);
       SDnodeHandle *pHandle = &pTrans->msgHandles[TMSG_INDEX(pMgmt->msgType)];
       if (pMgmt->needCheckVgId) {
         pHandle->needCheckVgId = pMgmt->needCheckVgId;
@@ -184,45 +228,6 @@ static inline int32_t dmSendReq(const SEpSet *pEpSet, SRpcMsg *pMsg) {
   }
 }
 
-static inline void dmSendRsp(SRpcMsg *pMsg) {
-  SMgmtWrapper *pWrapper = pMsg->info.wrapper;
-  if (InChildProc(pWrapper)) {
-    dmPutToProcPQueue(&pWrapper->proc, pMsg, DND_FUNC_RSP);
-  } else {
-    rpcSendResponse(pMsg);
-  }
-}
-
-static void dmBuildMnodeRedirectRsp(SDnode *pDnode, SRpcMsg *pMsg) {
-  SEpSet epSet = {0};
-  dmGetMnodeEpSetForRedirect(&pDnode->data, pMsg, &epSet);
-
-  int32_t contLen = tSerializeSEpSet(NULL, 0, &epSet);
-  pMsg->pCont = rpcMallocCont(contLen);
-  if (pMsg->pCont == NULL) {
-    pMsg->code = TSDB_CODE_OUT_OF_MEMORY;
-  } else {
-    tSerializeSEpSet(pMsg->pCont, contLen, &epSet);
-    pMsg->contLen = contLen;
-  }
-}
-
-static inline void dmSendRedirectRsp(SRpcMsg *pMsg, const SEpSet *pNewEpSet) {
-  SRpcMsg rsp = {.code = TSDB_CODE_RPC_REDIRECT, .info = pMsg->info};
-  int32_t contLen = tSerializeSEpSet(NULL, 0, pNewEpSet);
-
-  rsp.pCont = rpcMallocCont(contLen);
-  if (rsp.pCont == NULL) {
-    terrno = TSDB_CODE_OUT_OF_MEMORY;
-  } else {
-    tSerializeSEpSet(rsp.pCont, contLen, pNewEpSet);
-    rsp.contLen = contLen;
-  }
-  dmSendRsp(&rsp);
-  rpcFreeCont(pMsg->pCont);
-  pMsg->pCont = NULL;
-}
-
 static inline void dmRegisterBrokenLinkArg(SRpcMsg *pMsg) {
   SMgmtWrapper *pWrapper = pMsg->info.wrapper;
   if (InChildProc(pWrapper)) {

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

@@ -40,18 +40,27 @@
 #include "wal.h"
 
 #include "libs/function/function.h"
-// clang-format off
 #ifdef __cplusplus
 extern "C" {
 #endif
 
+// clang-format off
+
 #define dFatal(...) { if (dDebugFlag & DEBUG_FATAL) { taosPrintLog("DND FATAL ", DEBUG_FATAL, 255,        __VA_ARGS__); }}
 #define dError(...) { if (dDebugFlag & DEBUG_ERROR) { taosPrintLog("DND ERROR ", DEBUG_ERROR, 255,        __VA_ARGS__); }}
 #define dWarn(...)  { if (dDebugFlag & DEBUG_WARN)  { taosPrintLog("DND WARN ",  DEBUG_WARN,  255,        __VA_ARGS__); }}
 #define dInfo(...)  { if (dDebugFlag & DEBUG_INFO)  { taosPrintLog("DND ",       DEBUG_INFO,  255,        __VA_ARGS__); }}
 #define dDebug(...) { if (dDebugFlag & DEBUG_DEBUG) { taosPrintLog("DND ",       DEBUG_DEBUG, dDebugFlag, __VA_ARGS__); }}
 #define dTrace(...) { if (dDebugFlag & DEBUG_TRACE) { taosPrintLog("DND ",       DEBUG_TRACE, dDebugFlag, __VA_ARGS__); }}
-#define dGTrace(param, ...) do { char buf[40] = {0}; TRACE_TO_STR(trace, buf); dTrace(param ",GTID: %s", __VA_ARGS__, buf);} while(0)
+
+#define dGFatal(param, ...) { char buf[40] = {0}; TRACE_TO_STR(trace, buf); dFatal(param ", gtid:%s", __VA_ARGS__, buf);}
+#define dGError(param, ...) { char buf[40] = {0}; TRACE_TO_STR(trace, buf); dError(param ", gtid:%s", __VA_ARGS__, buf);}
+#define dGWarn(param, ...)  { char buf[40] = {0}; TRACE_TO_STR(trace, buf); dWarn (param ", gtid:%s", __VA_ARGS__, buf);}
+#define dGInfo(param, ...)  { char buf[40] = {0}; TRACE_TO_STR(trace, buf); dInfo (param ", gtid:%s", __VA_ARGS__, buf);}
+#define dGDebug(param, ...) { char buf[40] = {0}; TRACE_TO_STR(trace, buf); dDebug(param ", gtid:%s", __VA_ARGS__, buf);}
+#define dGTrace(param, ...) { char buf[40] = {0}; TRACE_TO_STR(trace, buf); dTrace(param ", gtid:%s", __VA_ARGS__, buf);}
+
+// clang-format on
 
 typedef enum {
   DNODE = 0,
@@ -185,4 +194,3 @@ void    dmSetMnodeEpSet(SDnodeData *pData, SEpSet *pEpSet);
 #endif
 
 #endif /*_TD_DM_INT_H_*/
-// clang-format on

source/dnode/mgmt/test/sut/src/server.cpp

@@ -28,7 +28,7 @@ bool TestServer::Start() {
   taosThreadAttrSetDetachState(&thAttr, PTHREAD_CREATE_JOINABLE);
   taosThreadCreate(&threadId, &thAttr, serverLoop, NULL);
   taosThreadAttrDestroy(&thAttr);
-  taosMsleep(2100);
+  taosMsleep(3100);
   return true;
 }
 

source/dnode/mgmt/test/sut/src/sut.cpp

@@ -102,6 +102,7 @@ int32_t Testbase::SendShowReq(int8_t showType, const char* tb, const char* db) {
   ASSERT(pRsp->pCont != nullptr);
 
   if (pRsp->contLen == 0) return -1;
+  if (pRsp->code != 0) return -1;
 
   showRsp = (SRetrieveMetaTableRsp*)pRsp->pCont;
   showRsp->handle = htobe64(showRsp->handle);  // show Id

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

@@ -421,6 +421,7 @@ typedef struct {
   int64_t        dbUid;
   int32_t        version;
   int8_t         subType;   // column, db or stable
+  int8_t         withMeta;  // TODO
   SRWLatch       lock;
   int32_t        sqlLen;
   int32_t        astLen;
@@ -487,6 +488,7 @@ typedef struct {
   int64_t   dbUid;
   int32_t   vgNum;
   int8_t    subType;
+  int8_t    withMeta;
   int64_t   stbUid;
   SHashObj* consumerHash;   // consumerId -> SMqConsumerEp
   SArray*   unassignedVgs;  // SArray<SMqVgEp*>

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

@@ -34,13 +34,20 @@ extern "C" {
 #endif
 
 // clang-format off
+
 #define mFatal(...) { if (mDebugFlag & DEBUG_FATAL) { taosPrintLog("MND FATAL ", DEBUG_FATAL, 255,        __VA_ARGS__); }}
 #define mError(...) { if (mDebugFlag & DEBUG_ERROR) { taosPrintLog("MND ERROR ", DEBUG_ERROR, 255,        __VA_ARGS__); }}
 #define mWarn(...)  { if (mDebugFlag & DEBUG_WARN)  { taosPrintLog("MND WARN ",  DEBUG_WARN,  255,        __VA_ARGS__); }}
 #define mInfo(...)  { if (mDebugFlag & DEBUG_INFO)  { taosPrintLog("MND ",       DEBUG_INFO,  255,        __VA_ARGS__); }}
 #define mDebug(...) { if (mDebugFlag & DEBUG_DEBUG) { taosPrintLog("MND ",       DEBUG_DEBUG, mDebugFlag, __VA_ARGS__); }}
 #define mTrace(...) { if (mDebugFlag & DEBUG_TRACE) { taosPrintLog("MND ",       DEBUG_TRACE, mDebugFlag, __VA_ARGS__); }}
-#define mGTrace(param, ...) do { char buf[40] = {0}; TRACE_TO_STR(trace, buf); mTrace(param ", GTID: %s", __VA_ARGS__, buf);} while(0)
+
+#define mGFatal(param, ...) { char buf[40] = {0}; TRACE_TO_STR(trace, buf); mFatal(param ", gtid:%s", __VA_ARGS__, buf);}
+#define mGError(param, ...) { char buf[40] = {0}; TRACE_TO_STR(trace, buf); mError(param ", gtid:%s", __VA_ARGS__, buf);}
+#define mGWarn(param, ...)  { char buf[40] = {0}; TRACE_TO_STR(trace, buf); mWarn (param ", gtid:%s", __VA_ARGS__, buf);}
+#define mGInfo(param, ...)  { char buf[40] = {0}; TRACE_TO_STR(trace, buf); mInfo (param ", gtid:%s", __VA_ARGS__, buf);}
+#define mGDebug(param, ...) { char buf[40] = {0}; TRACE_TO_STR(trace, buf); mDebug(param ", gtid:%s", __VA_ARGS__, buf);}
+#define mGTrace(param, ...) { char buf[40] = {0}; TRACE_TO_STR(trace, buf); mTrace(param ", gtid:%s", __VA_ARGS__, buf);}
 
 // clang-format on
 

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

@@ -381,6 +381,7 @@ SMqSubscribeObj *tCloneSubscribeObj(const SMqSubscribeObj *pSub) {
   pSubNew->dbUid = pSub->dbUid;
   pSubNew->stbUid = pSub->stbUid;
   pSubNew->subType = pSub->subType;
+  pSubNew->withMeta = pSub->withMeta;
 
   pSubNew->vgNum = pSub->vgNum;
   pSubNew->consumerHash = taosHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), false, HASH_NO_LOCK);
@@ -414,6 +415,7 @@ int32_t tEncodeSubscribeObj(void **buf, const SMqSubscribeObj *pSub) {
   tlen += taosEncodeFixedI64(buf, pSub->dbUid);
   tlen += taosEncodeFixedI32(buf, pSub->vgNum);
   tlen += taosEncodeFixedI8(buf, pSub->subType);
+  tlen += taosEncodeFixedI8(buf, pSub->withMeta);
   tlen += taosEncodeFixedI64(buf, pSub->stbUid);
 
   void   *pIter = NULL;
@@ -440,6 +442,7 @@ void *tDecodeSubscribeObj(const void *buf, SMqSubscribeObj *pSub) {
   buf = taosDecodeFixedI64(buf, &pSub->dbUid);
   buf = taosDecodeFixedI32(buf, &pSub->vgNum);
   buf = taosDecodeFixedI8(buf, &pSub->subType);
+  buf = taosDecodeFixedI8(buf, &pSub->withMeta);
   buf = taosDecodeFixedI64(buf, &pSub->stbUid);
 
   int32_t sz;

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

@@ -59,22 +59,28 @@ static void *mndBuildTimerMsg(int32_t *pContLen) {
 static void mndPullupTrans(SMnode *pMnode) {
   int32_t contLen = 0;
   void   *pReq = mndBuildTimerMsg(&contLen);
+  if (pReq != NULL) {
     SRpcMsg rpcMsg = {.msgType = TDMT_MND_TRANS_TIMER, .pCont = pReq, .contLen = contLen};
     tmsgPutToQueue(&pMnode->msgCb, WRITE_QUEUE, &rpcMsg);
+  }
 }
 
 static void mndCalMqRebalance(SMnode *pMnode) {
   int32_t contLen = 0;
   void   *pReq = mndBuildTimerMsg(&contLen);
+  if (pReq != NULL) {
     SRpcMsg rpcMsg = {.msgType = TDMT_MND_MQ_TIMER, .pCont = pReq, .contLen = contLen};
     tmsgPutToQueue(&pMnode->msgCb, READ_QUEUE, &rpcMsg);
+  }
 }
 
 static void mndPullupTelem(SMnode *pMnode) {
   int32_t contLen = 0;
   void   *pReq = mndBuildTimerMsg(&contLen);
+  if (pReq != NULL) {
     SRpcMsg rpcMsg = {.msgType = TDMT_MND_TELEM_TIMER, .pCont = pReq, .contLen = contLen};
     tmsgPutToQueue(&pMnode->msgCb, READ_QUEUE, &rpcMsg);
+  }
 }
 
 static void mndPushTtlTime(SMnode *pMnode) {
@@ -89,10 +95,11 @@ static void mndPushTtlTime(SMnode *pMnode) {
     int32_t   contLen = sizeof(SMsgHead) + sizeof(int32_t);
     SMsgHead *pHead = rpcMallocCont(contLen);
     if (pHead == NULL) {
-      mError("ttl time malloc err. contLen:%d", contLen);
+      sdbCancelFetch(pSdb, pIter);
       sdbRelease(pSdb, pVgroup);
       continue;
     }
+
     pHead->contLen = htonl(contLen);
     pHead->vgId = htonl(pVgroup->vgId);
 
@@ -100,13 +107,13 @@ static void mndPushTtlTime(SMnode *pMnode) {
     *(int32_t *)(POINTER_SHIFT(pHead, sizeof(SMsgHead))) = htonl(t);
 
     SRpcMsg rpcMsg = {.msgType = TDMT_VND_DROP_TTL_TABLE, .pCont = pHead, .contLen = contLen};
-
     SEpSet  epSet = mndGetVgroupEpset(pMnode, pVgroup);
     int32_t code = tmsgSendReq(&epSet, &rpcMsg);
     if (code != 0) {
-      mError("ttl time seed err. code:%d", code);
+      mError("failed to send ttl time seed msg, code:0x%x", code);
+    } else {
+      mInfo("send ttl time seed msg, time:%d", t);
     }
-    mError("ttl time seed succ. time:%d", t);
     sdbRelease(pSdb, pVgroup);
   }
 }
@@ -117,7 +124,7 @@ static void *mndThreadFp(void *param) {
   setThreadName("mnode-timer");
 
   while (1) {
-    if (lastTime % (864000) == 0) {  // sleep 1 day for ttl
+    if (lastTime % 864000 == 0) {
       mndPushTtlTime(pMnode);
     }
 
@@ -549,10 +556,13 @@ int32_t mndProcessSyncMsg(SRpcMsg *pMsg) {
 static int32_t mndCheckMnodeState(SRpcMsg *pMsg) {
   if (!IsReq(pMsg)) return 0;
   if (mndAcquireRpcRef(pMsg->info.node) == 0) return 0;
+  if (pMsg->msgType == TDMT_MND_MQ_TIMER || pMsg->msgType == TDMT_MND_TELEM_TIMER ||
+      pMsg->msgType == TDMT_MND_TRANS_TIMER) {
+    return -1;
+  }
 
-  if (pMsg->msgType != TDMT_MND_MQ_TIMER && pMsg->msgType != TDMT_MND_TELEM_TIMER &&
-      pMsg->msgType != TDMT_MND_TRANS_TIMER) {
-    mError("msg:%p, failed to check mnode state since %s, type:%s", pMsg, terrstr(), TMSG_INFO(pMsg->msgType));
+  const STraceId *trace = &pMsg->info.traceId;
+  mGError("msg:%p, failed to check mnode state since %s, type:%s", pMsg, terrstr(), TMSG_INFO(pMsg->msgType));
 
   SEpSet epSet = {0};
   mndGetMnodeEpSet(pMsg->info.node, &epSet);
@@ -566,7 +576,6 @@ static int32_t mndCheckMnodeState(SRpcMsg *pMsg) {
   } else {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
   }
-  }
 
   return -1;
 }
@@ -575,7 +584,8 @@ static int32_t mndCheckMsgContent(SRpcMsg *pMsg) {
   if (!IsReq(pMsg)) return 0;
   if (pMsg->contLen != 0 && pMsg->pCont != NULL) return 0;
 
-  mError("msg:%p, failed to check msg, cont:%p contLen:%d, app:%p type:%s", pMsg, pMsg->pCont, pMsg->contLen,
+  const STraceId *trace = &pMsg->info.traceId;
+  mGError("msg:%p, failed to check msg, cont:%p contLen:%d, app:%p type:%s", pMsg, pMsg->pCont, pMsg->contLen,
          pMsg->info.ahandle, TMSG_INFO(pMsg->msgType));
   terrno = TSDB_CODE_INVALID_MSG_LEN;
   return -1;
@@ -583,9 +593,11 @@ static int32_t mndCheckMsgContent(SRpcMsg *pMsg) {
 
 int32_t mndProcessRpcMsg(SRpcMsg *pMsg) {
   SMnode         *pMnode = pMsg->info.node;
+  const STraceId *trace = &pMsg->info.traceId;
+
   MndMsgFp fp = pMnode->msgFp[TMSG_INDEX(pMsg->msgType)];
   if (fp == NULL) {
-    mError("msg:%p, failed to get msg handle, app:%p type:%s", pMsg, pMsg->info.ahandle, TMSG_INFO(pMsg->msgType));
+    mGError("msg:%p, failed to get msg handle, app:%p type:%s", pMsg, pMsg->info.ahandle, TMSG_INFO(pMsg->msgType));
     terrno = TSDB_CODE_MSG_NOT_PROCESSED;
     return -1;
   }
@@ -593,17 +605,16 @@ int32_t mndProcessRpcMsg(SRpcMsg *pMsg) {
   if (mndCheckMsgContent(pMsg) != 0) return -1;
   if (mndCheckMnodeState(pMsg) != 0) return -1;
 
-  STraceId *trace = &pMsg->info.traceId;
   mGTrace("msg:%p, start to process in mnode, app:%p type:%s", pMsg, pMsg->info.ahandle, TMSG_INFO(pMsg->msgType));
   int32_t code = (*fp)(pMsg);
   mndReleaseRpcRef(pMnode);
 
   if (code == TSDB_CODE_ACTION_IN_PROGRESS) {
-    mTrace("msg:%p, won't response immediately since in progress", pMsg);
+    mGTrace("msg:%p, won't response immediately since in progress", pMsg);
   } else if (code == 0) {
-    mTrace("msg:%p, successfully processed", pMsg);
+    mGTrace("msg:%p, successfully processed", pMsg);
   } else {
-    mError("msg:%p, failed to process since %s, app:%p type:%s", pMsg, terrstr(), pMsg->info.ahandle,
+    mGError("msg:%p, failed to process since %s, app:%p type:%s", pMsg, terrstr(), pMsg->info.ahandle,
             TMSG_INFO(pMsg->msgType));
   }
 
@@ -620,7 +631,6 @@ void mndSetMsgHandle(SMnode *pMnode, tmsg_t msgType, MndMsgFp fp) {
 // Note: uid 0 is reserved
 int64_t mndGenerateUid(char *name, int32_t len) {
   int32_t hashval = MurmurHash3_32(name, len);
-
   do {
     int64_t us = taosGetTimestampUs();
     int64_t x = (us & 0x000000FFFFFFFFFF) << 24;

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

@@ -127,7 +127,8 @@ static SConnObj *mndCreateConn(SMnode *pMnode, const char *user, int8_t connType
   uint32_t connId = mndGenerateUid(connStr, len);
   if (startTime == 0) startTime = taosGetTimestampMs();
 
-  SConnObj connObj = {.id = connId,
+  SConnObj connObj = {
+      .id = connId,
       .connType = connType,
       .appStartTimeMs = startTime,
       .pid = pid,
@@ -138,14 +139,16 @@ static SConnObj *mndCreateConn(SMnode *pMnode, const char *user, int8_t connType
       .lastAccessTimeMs = 0,
       .killId = 0,
       .numOfQueries = 0,
-                      .pQueries = NULL};
+      .pQueries = NULL,
+  };
 
   connObj.lastAccessTimeMs = connObj.loginTimeMs;
   tstrncpy(connObj.user, user, TSDB_USER_LEN);
   tstrncpy(connObj.app, app, TSDB_APP_NAME_LEN);
 
   int32_t   keepTime = tsShellActivityTimer * 3;
-  SConnObj *pConn = taosCachePut(pMgmt->connCache, &connId, sizeof(uint32_t), &connObj, sizeof(connObj), keepTime * 1000);
+  SConnObj *pConn =
+      taosCachePut(pMgmt->connCache, &connId, sizeof(uint32_t), &connObj, sizeof(connObj), keepTime * 1000);
   if (pConn == NULL) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
     mError("conn:%d, failed to put into cache since %s, user:%s", connId, user, terrstr());
@@ -174,7 +177,6 @@ static SConnObj *mndAcquireConn(SMnode *pMnode, uint32_t connId) {
   }
 
   pConn->lastAccessTimeMs = taosGetTimestampMs();
-
   mTrace("conn:%u, acquired from cache, data:%p", pConn->id, pConn);
   return pConn;
 }
@@ -214,6 +216,7 @@ static int32_t mndProcessConnectReq(SRpcMsg *pReq) {
   int32_t         code = -1;
   SConnectReq     connReq = {0};
   char            ip[30] = {0};
+  const STraceId *trace = &pReq->info.traceId;
 
   if (tDeserializeSConnectReq(pReq->pCont, pReq->contLen, &connReq) != 0) {
     terrno = TSDB_CODE_INVALID_MSG;
@@ -224,11 +227,11 @@ static int32_t mndProcessConnectReq(SRpcMsg *pReq) {
 
   pUser = mndAcquireUser(pMnode, pReq->info.conn.user);
   if (pUser == NULL) {
-    mError("user:%s, failed to login while acquire user since %s", pReq->info.conn.user, terrstr());
+    mGError("user:%s, failed to login while acquire user since %s", pReq->info.conn.user, terrstr());
     goto CONN_OVER;
   }
   if (0 != strncmp(connReq.passwd, pUser->pass, TSDB_PASSWORD_LEN - 1)) {
-    mError("user:%s, failed to auth while acquire user, input:%s", pReq->info.conn.user, connReq.passwd);
+    mGError("user:%s, failed to auth while acquire user, input:%s", pReq->info.conn.user, connReq.passwd);
     code = TSDB_CODE_RPC_AUTH_FAILURE;
     goto CONN_OVER;
   }
@@ -239,7 +242,7 @@ static int32_t mndProcessConnectReq(SRpcMsg *pReq) {
     pDb = mndAcquireDb(pMnode, db);
     if (pDb == NULL) {
       terrno = TSDB_CODE_MND_INVALID_DB;
-      mError("user:%s, failed to login from %s while use db:%s since %s", pReq->info.conn.user, ip, connReq.db,
+      mGError("user:%s, failed to login from %s while use db:%s since %s", pReq->info.conn.user, ip, connReq.db,
               terrstr());
       goto CONN_OVER;
     }
@@ -248,7 +251,7 @@ static int32_t mndProcessConnectReq(SRpcMsg *pReq) {
   pConn = mndCreateConn(pMnode, pReq->info.conn.user, connReq.connType, pReq->info.conn.clientIp,
                         pReq->info.conn.clientPort, connReq.pid, connReq.app, connReq.startTime);
   if (pConn == NULL) {
-    mError("user:%s, failed to login from %s while create connection since %s", pReq->info.conn.user, ip, terrstr());
+    mGError("user:%s, failed to login from %s while create connection since %s", pReq->info.conn.user, ip, terrstr());
     goto CONN_OVER;
   }
 
@@ -273,7 +276,7 @@ static int32_t mndProcessConnectReq(SRpcMsg *pReq) {
   pReq->info.rspLen = contLen;
   pReq->info.rsp = pRsp;
 
-  mDebug("user:%s, login from %s:%d, conn:%u, app:%s", pReq->info.conn.user, ip, pConn->port, pConn->id, connReq.app);
+  mGDebug("user:%s, login from %s:%d, conn:%u, app:%s", pReq->info.conn.user, ip, pConn->port, pConn->id, connReq.app);
 
   code = 0;
 
@@ -302,7 +305,7 @@ static int32_t mndSaveQueryList(SConnObj *pConn, SQueryHbReqBasic *pBasic) {
   return TSDB_CODE_SUCCESS;
 }
 
-static SAppObj *mndCreateApp(SMnode *pMnode, uint32_t clientIp, SAppHbReq* pReq) {
+static SAppObj *mndCreateApp(SMnode *pMnode, uint32_t clientIp, SAppHbReq *pReq) {
   SProfileMgmt *pMgmt = &pMnode->profileMgmt;
 
   SAppObj app;
@@ -314,7 +317,7 @@ static SAppObj *mndCreateApp(SMnode *pMnode, uint32_t clientIp, SAppHbReq* pReq)
   memcpy(&app.summary, &pReq->summary, sizeof(pReq->summary));
   app.lastAccessTimeMs = taosGetTimestampMs();
 
-  int32_t   keepTime = tsShellActivityTimer * 3;
+  const int32_t keepTime = tsShellActivityTimer * 3;
   SAppObj *pApp = taosCachePut(pMgmt->appCache, &pReq->appId, sizeof(pReq->appId), &app, sizeof(app), keepTime * 1000);
   if (pApp == NULL) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
@@ -326,10 +329,7 @@ static SAppObj *mndCreateApp(SMnode *pMnode, uint32_t clientIp, SAppHbReq* pReq)
   return pApp;
 }
 
-static void mndFreeApp(SAppObj *pApp) {
-  mTrace("app %" PRIx64 " is destroyed", pApp->appId);
-}
-
+static void mndFreeApp(SAppObj *pApp) { mTrace("app %" PRIx64 " is destroyed", pApp->appId); }
 
 static SAppObj *mndAcquireApp(SMnode *pMnode, int64_t appId) {
   SProfileMgmt *pMgmt = &pMnode->profileMgmt;
@@ -439,7 +439,7 @@ static SClientHbRsp *mndMqHbBuildRsp(SMnode *pMnode, SClientHbReq *pReq) {
 }
 
 static int32_t mndUpdateAppInfo(SMnode *pMnode, SClientHbReq *pHbReq, SRpcConnInfo *connInfo) {
-  SAppHbReq* pReq = &pHbReq->app;
+  SAppHbReq *pReq = &pHbReq->app;
   SAppObj   *pApp = mndAcquireApp(pMnode, pReq->appId);
   if (pApp == NULL) {
     pApp = mndCreateApp(pMnode, connInfo->clientIp, pReq);
@@ -639,7 +639,7 @@ static int32_t mndProcessKillQueryReq(SRpcMsg *pReq) {
   mInfo("kill query msg is received, queryId:%s", killReq.queryStrId);
   int32_t  connId = 0;
   uint64_t queryId = 0;
-  char* p = strchr(killReq.queryStrId, ':');
+  char    *p = strchr(killReq.queryStrId, ':');
   if (NULL == p) {
     mError("invalid query id %s", killReq.queryStrId);
     terrno = TSDB_CODE_MND_INVALID_QUERY_ID;
@@ -931,7 +931,6 @@ static int32_t mndRetrieveApps(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBlo
   return numOfRows;
 }
 
-
 static void mndCancelGetNextQuery(SMnode *pMnode, void *pIter) {
   if (pIter != NULL) {
     taosCacheDestroyIter(pIter);

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

@@ -54,9 +54,10 @@ int32_t mndInitStream(SMnode *pMnode) {
   };
 
   mndSetMsgHandle(pMnode, TDMT_MND_CREATE_STREAM, mndProcessCreateStreamReq);
-  mndSetMsgHandle(pMnode, TDMT_STREAM_TASK_DEPLOY_RSP, mndTransProcessRsp);
   mndSetMsgHandle(pMnode, TDMT_MND_DROP_STREAM, mndProcessDropStreamReq);
-  mndSetMsgHandle(pMnode, TDMT_MND_DROP_STREAM_RSP, mndTransProcessRsp);
+
+  mndSetMsgHandle(pMnode, TDMT_STREAM_TASK_DEPLOY_RSP, mndTransProcessRsp);
+  mndSetMsgHandle(pMnode, TDMT_STREAM_TASK_DROP_RSP, mndTransProcessRsp);
 
   mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_STREAMS, mndRetrieveStream);
   mndAddShowFreeIterHandle(pMnode, TSDB_MGMT_TABLE_STREAMS, mndCancelGetNextStream);
@@ -477,7 +478,7 @@ static int32_t mndPersistTaskDropReq(STrans *pTrans, SStreamTask *pTask) {
   memcpy(&action.epSet, &pTask->epSet, sizeof(SEpSet));
   action.pCont = pReq;
   action.contLen = sizeof(SVDropStreamTaskReq);
-  action.msgType = TDMT_VND_STREAM_TASK_DROP;
+  action.msgType = TDMT_STREAM_TASK_DROP;
   if (mndTransAppendRedoAction(pTrans, &action) != 0) {
     taosMemoryFree(pReq);
     return -1;
@@ -670,20 +671,24 @@ _OVER:
 
 static int32_t mndProcessDropStreamReq(SRpcMsg *pReq) {
   SMnode     *pMnode = pReq->info.node;
-  int32_t     code = -1;
   SStreamObj *pStream = NULL;
   /*SDbObj     *pDb = NULL;*/
   /*SUserObj   *pUser = NULL;*/
 
-  SMDropStreamReq dropReq = *(SMDropStreamReq *)pReq->pCont;
+  SMDropStreamReq dropReq = {0};
+  if (tDeserializeSMDropStreamReq(pReq->pCont, pReq->contLen, &dropReq) < 0) {
+    ASSERT(0);
+    terrno = TSDB_CODE_INVALID_MSG;
+    return -1;
+  }
 
   pStream = mndAcquireStream(pMnode, dropReq.name);
 
   if (pStream == NULL) {
     if (dropReq.igNotExists) {
       mDebug("stream:%s, not exist, ignore not exist is set", dropReq.name);
-      code = 0;
-      goto DROP_STREAM_OVER;
+      sdbRelease(pMnode->pSdb, pStream);
+      return -1;
     } else {
       terrno = TSDB_CODE_MND_STREAM_NOT_EXIST;
       return -1;
@@ -701,14 +706,16 @@ static int32_t mndProcessDropStreamReq(SRpcMsg *pReq) {
   STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) {
     mError("stream:%s, failed to drop since %s", dropReq.name, terrstr());
-    return code;
+    sdbRelease(pMnode->pSdb, pStream);
+    return -1;
   }
   mDebug("trans:%d, used to drop stream:%s", pTrans->id, dropReq.name);
 
   // drop all tasks
   if (mndDropStreamTasks(pMnode, pTrans, pStream) < 0) {
     mError("stream:%s, failed to drop task since %s", dropReq.name, terrstr());
-    return code;
+    sdbRelease(pMnode->pSdb, pStream);
+    return -1;
   }
 
   // drop stream
@@ -717,8 +724,16 @@ static int32_t mndProcessDropStreamReq(SRpcMsg *pReq) {
     return -1;
   }
 
-DROP_STREAM_OVER:
-  return code;
+  if (mndTransPrepare(pMnode, pTrans) != 0) {
+    mError("trans:%d, failed to prepare drop stream trans since %s", pTrans->id, terrstr());
+    sdbRelease(pMnode->pSdb, pStream);
+    mndTransDrop(pTrans);
+    return -1;
+  }
+
+  sdbRelease(pMnode->pSdb, pStream);
+
+  return TSDB_CODE_ACTION_IN_PROGRESS;
 }
 
 int32_t mndDropStreamByDb(SMnode *pMnode, STrans *pTrans, SDbObj *pDb) {

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

@@ -96,6 +96,7 @@ static SMqSubscribeObj *mndCreateSub(SMnode *pMnode, const SMqTopicObj *pTopic,
   pSub->dbUid = pTopic->dbUid;
   pSub->stbUid = pTopic->stbUid;
   pSub->subType = pTopic->subType;
+  pSub->withMeta = pTopic->withMeta;
 
   ASSERT(pSub->unassignedVgs->size == 0);
   ASSERT(taosHashGetSize(pSub->consumerHash) == 0);
@@ -120,6 +121,7 @@ static int32_t mndBuildSubChangeReq(void **pBuf, int32_t *pLen, const SMqSubscri
   req.vgId = pRebVg->pVgEp->vgId;
   req.qmsg = pRebVg->pVgEp->qmsg;
   req.subType = pSub->subType;
+  req.withMeta = pSub->withMeta;
   req.suid = pSub->stbUid;
   strncpy(req.subKey, pSub->key, TSDB_SUBSCRIBE_KEY_LEN);
 

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

@@ -141,6 +141,7 @@ SSdbRaw *mndTopicActionEncode(SMqTopicObj *pTopic) {
   SDB_SET_INT64(pRaw, dataPos, pTopic->dbUid, TOPIC_ENCODE_OVER);
   SDB_SET_INT32(pRaw, dataPos, pTopic->version, TOPIC_ENCODE_OVER);
   SDB_SET_INT8(pRaw, dataPos, pTopic->subType, TOPIC_ENCODE_OVER);
+  SDB_SET_INT8(pRaw, dataPos, pTopic->withMeta, TOPIC_ENCODE_OVER);
 
   SDB_SET_INT64(pRaw, dataPos, pTopic->stbUid, TOPIC_ENCODE_OVER);
   SDB_SET_INT32(pRaw, dataPos, pTopic->sqlLen, TOPIC_ENCODE_OVER);
@@ -208,6 +209,7 @@ SSdbRow *mndTopicActionDecode(SSdbRaw *pRaw) {
   SDB_GET_INT64(pRaw, dataPos, &pTopic->dbUid, TOPIC_DECODE_OVER);
   SDB_GET_INT32(pRaw, dataPos, &pTopic->version, TOPIC_DECODE_OVER);
   SDB_GET_INT8(pRaw, dataPos, &pTopic->subType, TOPIC_DECODE_OVER);
+  SDB_GET_INT8(pRaw, dataPos, &pTopic->withMeta, TOPIC_DECODE_OVER);
 
   SDB_GET_INT64(pRaw, dataPos, &pTopic->stbUid, TOPIC_DECODE_OVER);
   SDB_GET_INT32(pRaw, dataPos, &pTopic->sqlLen, TOPIC_DECODE_OVER);
@@ -357,6 +359,10 @@ static int32_t mndCreateTopic(SMnode *pMnode, SRpcMsg *pReq, SCMCreateTopicReq *
   topicObj.sql = strdup(pCreate->sql);
   topicObj.sqlLen = strlen(pCreate->sql) + 1;
   topicObj.subType = pCreate->subType;
+  topicObj.withMeta = pCreate->withMeta;
+  if (topicObj.withMeta) {
+    ASSERT(topicObj.subType != TOPIC_SUB_TYPE__COLUMN);
+  }
 
   if (pCreate->subType == TOPIC_SUB_TYPE__COLUMN) {
     topicObj.ast = strdup(pCreate->ast);

source/dnode/mnode/impl/test/user/user.cpp

@@ -26,514 +26,7 @@ class MndTestUser : public ::testing::Test {
 Testbase MndTestUser::test;
 
 TEST_F(MndTestUser, 01_Show_User) {
+  uInfo("show users");
   test.SendShowReq(TSDB_MGMT_TABLE_USER, "user_users", "");
   EXPECT_EQ(test.GetShowRows(), 1);
 }
-
-TEST_F(MndTestUser, 02_Create_User) {
-  {
-    SCreateUserReq createReq = {0};
-    strcpy(createReq.user, "");
-    strcpy(createReq.pass, "p1");
-
-    int32_t contLen = tSerializeSCreateUserReq(NULL, 0, &createReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSCreateUserReq(pReq, contLen, &createReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_CREATE_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, TSDB_CODE_MND_INVALID_USER_FORMAT);
-  }
-
-  {
-    SCreateUserReq createReq = {0};
-    strcpy(createReq.user, "u1");
-    strcpy(createReq.pass, "");
-
-    int32_t contLen = tSerializeSCreateUserReq(NULL, 0, &createReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSCreateUserReq(pReq, contLen, &createReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_CREATE_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, TSDB_CODE_MND_INVALID_PASS_FORMAT);
-  }
-
-  {
-    SCreateUserReq createReq = {0};
-    strcpy(createReq.user, "root");
-    strcpy(createReq.pass, "1");
-
-    int32_t contLen = tSerializeSCreateUserReq(NULL, 0, &createReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSCreateUserReq(pReq, contLen, &createReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_CREATE_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, TSDB_CODE_MND_USER_ALREADY_EXIST);
-  }
-
-  {
-    SCreateUserReq createReq = {0};
-    strcpy(createReq.user, "u1");
-    strcpy(createReq.pass, "p1");
-
-    int32_t contLen = tSerializeSCreateUserReq(NULL, 0, &createReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSCreateUserReq(pReq, contLen, &createReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_CREATE_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-
-    test.SendShowReq(TSDB_MGMT_TABLE_USER, "user_users", "");
-    EXPECT_EQ(test.GetShowRows(), 2);
-  }
-
-  {
-    SDropUserReq dropReq = {0};
-    strcpy(dropReq.user, "u1");
-
-    int32_t contLen = tSerializeSDropUserReq(NULL, 0, &dropReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSDropUserReq(pReq, contLen, &dropReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_DROP_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-
-    test.SendShowReq(TSDB_MGMT_TABLE_USER, "user_users", "");
-    EXPECT_EQ(test.GetShowRows(), 1);
-  }
-
-  {
-    SCreateUserReq createReq = {0};
-    strcpy(createReq.user, "u2");
-    strcpy(createReq.pass, "p1");
-    createReq.superUser = 1;
-
-    int32_t contLen = tSerializeSCreateUserReq(NULL, 0, &createReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSCreateUserReq(pReq, contLen, &createReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_CREATE_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-
-    test.SendShowReq(TSDB_MGMT_TABLE_USER, "user_users", "");
-    EXPECT_EQ(test.GetShowRows(), 2);
-  }
-
-  {
-    SDropUserReq dropReq = {0};
-    strcpy(dropReq.user, "u2");
-
-    int32_t contLen = tSerializeSDropUserReq(NULL, 0, &dropReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSDropUserReq(pReq, contLen, &dropReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_DROP_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-
-    test.SendShowReq(TSDB_MGMT_TABLE_USER, "user_users", "");
-    EXPECT_EQ(test.GetShowRows(), 1);
-  }
-}
-
-TEST_F(MndTestUser, 03_Alter_User) {
-  {
-    SCreateUserReq createReq = {0};
-    strcpy(createReq.user, "u3");
-    strcpy(createReq.pass, "p1");
-    createReq.superUser = 1;
-
-    int32_t contLen = tSerializeSCreateUserReq(NULL, 0, &createReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSCreateUserReq(pReq, contLen, &createReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_CREATE_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-
-    test.SendShowReq(TSDB_MGMT_TABLE_USER, "user_users", "");
-    EXPECT_EQ(test.GetShowRows(), 2);
-  }
-
-  {
-    SAlterUserReq alterReq = {0};
-    alterReq.alterType = TSDB_ALTER_USER_PASSWD;
-    strcpy(alterReq.user, "");
-    strcpy(alterReq.pass, "p1");
-
-    int32_t contLen = tSerializeSAlterUserReq(NULL, 0, &alterReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSAlterUserReq(pReq, contLen, &alterReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_ALTER_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, TSDB_CODE_MND_INVALID_USER_FORMAT);
-  }
-
-  {
-    SAlterUserReq alterReq = {0};
-    alterReq.alterType = TSDB_ALTER_USER_PASSWD;
-    strcpy(alterReq.user, "u3");
-    strcpy(alterReq.pass, "");
-
-    int32_t contLen = tSerializeSAlterUserReq(NULL, 0, &alterReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSAlterUserReq(pReq, contLen, &alterReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_ALTER_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, TSDB_CODE_MND_INVALID_PASS_FORMAT);
-  }
-
-  {
-    SAlterUserReq alterReq = {0};
-    alterReq.alterType = TSDB_ALTER_USER_PASSWD;
-    strcpy(alterReq.user, "u4");
-    strcpy(alterReq.pass, "1");
-
-    int32_t contLen = tSerializeSAlterUserReq(NULL, 0, &alterReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSAlterUserReq(pReq, contLen, &alterReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_ALTER_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, TSDB_CODE_MND_USER_NOT_EXIST);
-  }
-
-  {
-    SAlterUserReq alterReq = {0};
-    alterReq.alterType = TSDB_ALTER_USER_PASSWD;
-    strcpy(alterReq.user, "u3");
-    strcpy(alterReq.pass, "1");
-
-    int32_t contLen = tSerializeSAlterUserReq(NULL, 0, &alterReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSAlterUserReq(pReq, contLen, &alterReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_ALTER_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-  }
-
-  {
-    SAlterUserReq alterReq = {0};
-    alterReq.alterType = TSDB_ALTER_USER_SUPERUSER;
-    strcpy(alterReq.user, "u3");
-    strcpy(alterReq.pass, "1");
-    alterReq.superUser = 1;
-
-    int32_t contLen = tSerializeSAlterUserReq(NULL, 0, &alterReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSAlterUserReq(pReq, contLen, &alterReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_ALTER_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-  }
-
-  {
-    SAlterUserReq alterReq = {0};
-    alterReq.alterType = TSDB_ALTER_USER_REMOVE_ALL_DB;
-    strcpy(alterReq.user, "u3");
-    strcpy(alterReq.pass, "1");
-    strcpy(alterReq.dbname, "1.*");
-
-    int32_t contLen = tSerializeSAlterUserReq(NULL, 0, &alterReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSAlterUserReq(pReq, contLen, &alterReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_ALTER_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-  }
-
-  {
-    SAlterUserReq alterReq = {0};
-    alterReq.alterType = TSDB_ALTER_USER_REMOVE_ALL_DB;
-    strcpy(alterReq.user, "u3");
-    strcpy(alterReq.pass, "1");
-    strcpy(alterReq.dbname, "1.*");
-
-    int32_t contLen = tSerializeSAlterUserReq(NULL, 0, &alterReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSAlterUserReq(pReq, contLen, &alterReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_ALTER_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-  }
-
-  {
-    SAlterUserReq alterReq = {0};
-    alterReq.alterType = TSDB_ALTER_USER_ADD_READ_DB;
-    strcpy(alterReq.user, "u3");
-    strcpy(alterReq.pass, "1");
-    strcpy(alterReq.dbname, "d1");
-
-    int32_t contLen = tSerializeSAlterUserReq(NULL, 0, &alterReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSAlterUserReq(pReq, contLen, &alterReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_ALTER_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, TSDB_CODE_MND_DB_NOT_EXIST);
-  }
-
-  {
-    SCreateDbReq createReq = {0};
-    strcpy(createReq.db, "1.d2");
-    createReq.numOfVgroups = 2;
-    createReq.buffer = -1;
-    createReq.pageSize = -1;
-    createReq.pages = -1;
-    createReq.daysPerFile = 10 * 1440;
-    createReq.daysToKeep0 = 3650 * 1440;
-    createReq.daysToKeep1 = 3650 * 1440;
-    createReq.daysToKeep2 = 3650 * 1440;
-    createReq.minRows = 100;
-    createReq.maxRows = 4096;
-    createReq.fsyncPeriod = 3000;
-    createReq.walLevel = 1;
-    createReq.precision = 0;
-    createReq.compression = 2;
-    createReq.replications = 1;
-    createReq.strict = 1;
-    createReq.cacheLastRow = 0;
-    createReq.ignoreExist = 1;
-
-    int32_t contLen = tSerializeSCreateDbReq(NULL, 0, &createReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSCreateDbReq(pReq, contLen, &createReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_CREATE_DB, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-  }
-
-  {
-    SAlterUserReq alterReq = {0};
-    alterReq.alterType = TSDB_ALTER_USER_ADD_READ_DB;
-    strcpy(alterReq.user, "u3");
-    strcpy(alterReq.pass, "1");
-    strcpy(alterReq.dbname, "1.d2");
-
-    int32_t contLen = tSerializeSAlterUserReq(NULL, 0, &alterReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSAlterUserReq(pReq, contLen, &alterReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_ALTER_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-  }
-
-  {
-    SAlterUserReq alterReq = {0};
-    alterReq.alterType = TSDB_ALTER_USER_ADD_READ_DB;
-    strcpy(alterReq.user, "u3");
-    strcpy(alterReq.pass, "1");
-    strcpy(alterReq.dbname, "1.d2");
-
-    int32_t contLen = tSerializeSAlterUserReq(NULL, 0, &alterReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSAlterUserReq(pReq, contLen, &alterReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_ALTER_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-  }
-
-  {
-    SGetUserAuthReq authReq = {0};
-    strcpy(authReq.user, "u3");
-    int32_t contLen = tSerializeSGetUserAuthReq(NULL, 0, &authReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSGetUserAuthReq(pReq, contLen, &authReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_GET_USER_AUTH, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-
-    SGetUserAuthRsp authRsp = {0};
-    tDeserializeSGetUserAuthRsp(pRsp->pCont, pRsp->contLen, &authRsp);
-    EXPECT_STREQ(authRsp.user, "u3");
-    EXPECT_EQ(authRsp.superAuth, 1);
-    int32_t numOfReadDbs = taosHashGetSize(authRsp.readDbs);
-    int32_t numOfWriteDbs = taosHashGetSize(authRsp.writeDbs);
-    EXPECT_EQ(numOfReadDbs, 1);
-    EXPECT_EQ(numOfWriteDbs, 0);
-
-    char* dbname = (char*)taosHashGet(authRsp.readDbs, "1.d2", 4);
-    EXPECT_TRUE(dbname != NULL);
-
-    taosHashCleanup(authRsp.readDbs);
-    taosHashCleanup(authRsp.writeDbs);
-  }
-
-  {
-    SAlterUserReq alterReq = {0};
-    alterReq.alterType = TSDB_ALTER_USER_REMOVE_READ_DB;
-    strcpy(alterReq.user, "u3");
-    strcpy(alterReq.pass, "1");
-    strcpy(alterReq.dbname, "1.d2");
-
-    int32_t contLen = tSerializeSAlterUserReq(NULL, 0, &alterReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSAlterUserReq(pReq, contLen, &alterReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_ALTER_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-  }
-
-  {
-    SDropUserReq dropReq = {0};
-    strcpy(dropReq.user, "u3");
-
-    int32_t contLen = tSerializeSDropUserReq(NULL, 0, &dropReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSDropUserReq(pReq, contLen, &dropReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_DROP_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-
-    test.SendShowReq(TSDB_MGMT_TABLE_USER, "user_users", "");
-    EXPECT_EQ(test.GetShowRows(), 1);
-  }
-}
-
-TEST_F(MndTestUser, 05_Drop_User) {
-  {
-    SDropUserReq dropReq = {0};
-    strcpy(dropReq.user, "");
-
-    int32_t contLen = tSerializeSDropUserReq(NULL, 0, &dropReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSDropUserReq(pReq, contLen, &dropReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_DROP_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, TSDB_CODE_MND_INVALID_USER_FORMAT);
-  }
-
-  {
-    SDropUserReq dropReq = {0};
-    strcpy(dropReq.user, "u4");
-
-    int32_t contLen = tSerializeSDropUserReq(NULL, 0, &dropReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSDropUserReq(pReq, contLen, &dropReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_DROP_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, TSDB_CODE_MND_USER_NOT_EXIST);
-  }
-
-  {
-    SCreateUserReq createReq = {0};
-    strcpy(createReq.user, "u1");
-    strcpy(createReq.pass, "p1");
-
-    int32_t contLen = tSerializeSCreateUserReq(NULL, 0, &createReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSCreateUserReq(pReq, contLen, &createReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_CREATE_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-  }
-
-  {
-    SDropUserReq dropReq = {0};
-    strcpy(dropReq.user, "u1");
-
-    int32_t contLen = tSerializeSDropUserReq(NULL, 0, &dropReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSDropUserReq(pReq, contLen, &dropReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_DROP_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-  }
-
-  test.SendShowReq(TSDB_MGMT_TABLE_USER, "user_users", "");
-  EXPECT_EQ(test.GetShowRows(), 1);
-}
-
-TEST_F(MndTestUser, 06_Create_Drop_Alter_User) {
-  {
-    SCreateUserReq createReq = {0};
-    strcpy(createReq.user, "u1");
-    strcpy(createReq.pass, "p1");
-
-    int32_t contLen = tSerializeSCreateUserReq(NULL, 0, &createReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSCreateUserReq(pReq, contLen, &createReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_CREATE_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-  }
-
-  {
-    SCreateUserReq createReq = {0};
-    strcpy(createReq.user, "u2");
-    strcpy(createReq.pass, "p2");
-
-    int32_t contLen = tSerializeSCreateUserReq(NULL, 0, &createReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSCreateUserReq(pReq, contLen, &createReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_CREATE_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-  }
-
-  test.SendShowReq(TSDB_MGMT_TABLE_USER, "user_users", "");
-  EXPECT_EQ(test.GetShowRows(), 3);
-
-  {
-    SAlterUserReq alterReq = {0};
-    alterReq.alterType = TSDB_ALTER_USER_PASSWD;
-    strcpy(alterReq.user, "u1");
-    strcpy(alterReq.pass, "p2");
-
-    int32_t contLen = tSerializeSAlterUserReq(NULL, 0, &alterReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSAlterUserReq(pReq, contLen, &alterReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_ALTER_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-  }
-
-  test.SendShowReq(TSDB_MGMT_TABLE_USER, "user_users", "");
-  EXPECT_EQ(test.GetShowRows(), 3);
-  {
-    SDropUserReq dropReq = {0};
-    strcpy(dropReq.user, "u1");
-
-    int32_t contLen = tSerializeSDropUserReq(NULL, 0, &dropReq);
-    void*   pReq = rpcMallocCont(contLen);
-    tSerializeSDropUserReq(pReq, contLen, &dropReq);
-
-    SRpcMsg* pRsp = test.SendReq(TDMT_MND_DROP_USER, pReq, contLen);
-    ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
-  }
-
-  test.SendShowReq(TSDB_MGMT_TABLE_USER, "user_users", "");
-  EXPECT_EQ(test.GetShowRows(), 2);
-
-  // restart
-  test.Restart();
-
-  taosMsleep(1000);
-  test.SendShowReq(TSDB_MGMT_TABLE_USER, "user_users", "");
-  EXPECT_EQ(test.GetShowRows(), 2);
-}

source/dnode/snode/src/snode.c

@@ -92,7 +92,7 @@ static int32_t sndProcessTaskDeployReq(SSnode *pNode, SRpcMsg *pMsg) {
   }
   tDecoderClear(&decoder);
 
-  pTask->status = TASK_STATUS__IDLE;
+  pTask->execStatus = TASK_EXEC_STATUS__IDLE;
 
   pTask->inputQueue = streamQueueOpen();
   pTask->outputQueue = streamQueueOpen();
@@ -205,7 +205,7 @@ int32_t sndProcessUMsg(SSnode *pSnode, SRpcMsg *pMsg) {
   switch (pMsg->msgType) {
     case TDMT_STREAM_TASK_DEPLOY:
       return sndProcessTaskDeployReq(pSnode, pMsg);
-    case TDMT_VND_STREAM_TASK_DROP:
+    case TDMT_STREAM_TASK_DROP:
       return sndProcessTaskDropReq(pSnode, pMsg);
     default:
       ASSERT(0);

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

@@ -114,6 +114,7 @@ typedef struct {
   char    subKey[TSDB_SUBSCRIBE_KEY_LEN];
   int64_t consumerId;
   int32_t epoch;
+  int8_t  fetchMeta;
 
   // reader
   SWalReadHandle* pWalReader;

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

@@ -26,13 +26,21 @@ extern "C" {
 #endif
 
 // clang-format off
-#define vFatal(...) do { if (vDebugFlag & DEBUG_FATAL) { taosPrintLog("VND FATAL ", DEBUG_FATAL, 255, __VA_ARGS__); }}     while(0)
-#define vError(...) do { if (vDebugFlag & DEBUG_ERROR) { taosPrintLog("VND ERROR ", DEBUG_ERROR, 255, __VA_ARGS__); }}     while(0)
-#define vWarn(...)  do { if (vDebugFlag & DEBUG_WARN)  { taosPrintLog("VND WARN ", DEBUG_WARN, 255, __VA_ARGS__); }}       while(0)
-#define vInfo(...)  do { if (vDebugFlag & DEBUG_INFO)  { taosPrintLog("VND ", DEBUG_INFO, 255, __VA_ARGS__); }}            while(0)
-#define vDebug(...) do { if (vDebugFlag & DEBUG_DEBUG) { taosPrintLog("VND ", DEBUG_DEBUG, vDebugFlag, __VA_ARGS__); }}    while(0)
-#define vTrace(...) do { if (vDebugFlag & DEBUG_TRACE) { taosPrintLog("VND ", DEBUG_TRACE, vDebugFlag, __VA_ARGS__); }}    while(0)
-#define vGTrace(param, ...) do { char buf[40] = {0}; TRACE_TO_STR(trace, buf); vTrace(param " GTID: %s", __VA_ARGS__, buf);}      while(0)//#define vDye(...)   do
+
+#define vFatal(...) { if (vDebugFlag & DEBUG_FATAL) { taosPrintLog("VND FATAL ", DEBUG_FATAL, 255,        __VA_ARGS__); }}
+#define vError(...) { if (vDebugFlag & DEBUG_ERROR) { taosPrintLog("VND ERROR ", DEBUG_ERROR, 255,        __VA_ARGS__); }}
+#define vWarn(...)  { if (vDebugFlag & DEBUG_WARN)  { taosPrintLog("VND WARN ",  DEBUG_WARN, 255,         __VA_ARGS__); }}
+#define vInfo(...)  { if (vDebugFlag & DEBUG_INFO)  { taosPrintLog("VND ",       DEBUG_INFO, 255,         __VA_ARGS__); }}
+#define vDebug(...) { if (vDebugFlag & DEBUG_DEBUG) { taosPrintLog("VND ",       DEBUG_DEBUG, vDebugFlag, __VA_ARGS__); }}
+#define vTrace(...) { if (vDebugFlag & DEBUG_TRACE) { taosPrintLog("VND ",       DEBUG_TRACE, vDebugFlag, __VA_ARGS__); }}
+
+#define vGFatal(param, ...) { char buf[40] = {0}; TRACE_TO_STR(trace, buf); vFatal(param ", gtid:%s", __VA_ARGS__, buf);}
+#define vGError(param, ...) { char buf[40] = {0}; TRACE_TO_STR(trace, buf); vError(param ", gtid:%s", __VA_ARGS__, buf);}
+#define vGWarn(param, ...)  { char buf[40] = {0}; TRACE_TO_STR(trace, buf); vWarn (param ", gtid:%s", __VA_ARGS__, buf);}
+#define vGInfo(param, ...)  { char buf[40] = {0}; TRACE_TO_STR(trace, buf); vInfo (param ", gtid:%s", __VA_ARGS__, buf);}
+#define vGDebug(param, ...) { char buf[40] = {0}; TRACE_TO_STR(trace, buf); vDebug(param ", gtid:%s", __VA_ARGS__, buf);}
+#define vGTrace(param, ...) { char buf[40] = {0}; TRACE_TO_STR(trace, buf); vTrace(param ", gtid:%s", __VA_ARGS__, buf);}
+
 // clang-format on
 
 // vnodeCfg.c

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

@@ -85,6 +85,34 @@ void tqClose(STQ* pTq) {
   }
 }
 
+int32_t tqSendMetaPollRsp(STQ* pTq, const SRpcMsg* pMsg, const SMqPollReq* pReq, const SMqMetaRsp* pRsp) {
+  int32_t tlen = sizeof(SMqRspHead) + tEncodeSMqMetaRsp(NULL, pRsp);
+  void*   buf = rpcMallocCont(tlen);
+  if (buf == NULL) {
+    return -1;
+  }
+
+  ((SMqRspHead*)buf)->mqMsgType = TMQ_MSG_TYPE__POLL_META_RSP;
+  ((SMqRspHead*)buf)->epoch = pReq->epoch;
+  ((SMqRspHead*)buf)->consumerId = pReq->consumerId;
+
+  void* abuf = POINTER_SHIFT(buf, sizeof(SMqRspHead));
+  tEncodeSMqMetaRsp(&abuf, pRsp);
+
+  SRpcMsg resp = {
+      .info = pMsg->info,
+      .pCont = buf,
+      .contLen = tlen,
+      .code = 0,
+  };
+  tmsgSendRsp(&resp);
+
+  tqDebug("vg %d from consumer %ld (epoch %d) send rsp, res msg type %d, reqOffset: %ld, rspOffset: %ld",
+          TD_VID(pTq->pVnode), pReq->consumerId, pReq->epoch, pRsp->resMsgType, pRsp->reqOffset, pRsp->rspOffset);
+
+  return 0;
+}
+
 int32_t tqSendPollRsp(STQ* pTq, const SRpcMsg* pMsg, const SMqPollReq* pReq, const SMqDataBlkRsp* pRsp) {
   int32_t tlen = sizeof(SMqRspHead) + tEncodeSMqDataBlkRsp(NULL, pRsp);
   void*   buf = rpcMallocCont(tlen);
@@ -250,8 +278,23 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
         /*ASSERT(0);*/
       }
     } else {
-      // TODO
-      ASSERT(0);
+      ASSERT(pHandle->fetchMeta);
+      ASSERT(pHead->msgType == TDMT_VND_CREATE_STB || pHead->msgType == TDMT_VND_ALTER_STB ||
+             pHead->msgType == TDMT_VND_DROP_STB || pHead->msgType == TDMT_VND_CREATE_TABLE ||
+             pHead->msgType == TDMT_VND_ALTER_TABLE || pHead->msgType == TDMT_VND_DROP_TABLE ||
+             pHead->msgType == TDMT_VND_DROP_TTL_TABLE);
+      // return
+      SMqMetaRsp metaRsp = {0};
+      metaRsp.reqOffset = pReq->currentOffset;
+      metaRsp.rspOffset = fetchOffset;
+      metaRsp.resMsgType = pHead->msgType;
+      metaRsp.metaRspLen = pHead->bodyLen;
+      metaRsp.metaRsp = pHead->body;
+      if (tqSendMetaPollRsp(pTq, pMsg, pReq, &metaRsp) < 0) {
+        code = -1;
+      }
+      code = 0;
+      goto OVER;
     }
 
     // TODO batch optimization:
@@ -276,7 +319,7 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
   if (tqSendPollRsp(pTq, pMsg, pReq, &rsp) < 0) {
     code = -1;
   }
-
+OVER:
   // TODO wrap in destroy func
   taosArrayDestroy(rsp.blockDataLen);
   taosArrayDestroyP(rsp.blockData, (FDelete)taosMemoryFree);
@@ -384,7 +427,7 @@ int32_t tqProcessTaskDeployReq(STQ* pTq, char* msg, int32_t msgLen) {
   }
   tDecoderClear(&decoder);
 
-  pTask->status = TASK_STATUS__IDLE;
+  pTask->execStatus = TASK_EXEC_STATUS__IDLE;
 
   pTask->inputQueue = streamQueueOpen();
   pTask->outputQueue = streamQueueOpen();
@@ -459,6 +502,9 @@ int32_t tqProcessStreamTrigger(STQ* pTq, SSubmitReq* pReq) {
     pIter = taosHashIterate(pTq->pStreamTasks, pIter);
     if (pIter == NULL) break;
     SStreamTask* pTask = *(SStreamTask**)pIter;
+    if (atomic_load_8(&pTask->taskStatus) == TASK_STATUS__DROPPING) {
+      continue;
+    }
     if (pTask->inputType != STREAM_INPUT__DATA_SUBMIT) continue;
 
     if (!failed) {
@@ -487,6 +533,9 @@ int32_t tqProcessTaskRunReq(STQ* pTq, SRpcMsg* pMsg) {
   SStreamTaskRunReq* pReq = pMsg->pCont;
   int32_t            taskId = pReq->taskId;
   SStreamTask*       pTask = *(SStreamTask**)taosHashGet(pTq->pStreamTasks, &taskId, sizeof(int32_t));
+  if (atomic_load_8(&pTask->taskStatus) != TASK_STATUS__NORMAL) {
+    return 0;
+  }
   streamProcessRunReq(pTask);
   return 0;
 }
@@ -501,6 +550,9 @@ int32_t tqProcessTaskDispatchReq(STQ* pTq, SRpcMsg* pMsg) {
   tDecodeStreamDispatchReq(&decoder, &req);
   int32_t      taskId = req.taskId;
   SStreamTask* pTask = *(SStreamTask**)taosHashGet(pTq->pStreamTasks, &taskId, sizeof(int32_t));
+  if (atomic_load_8(&pTask->taskStatus) != TASK_STATUS__NORMAL) {
+    return 0;
+  }
   SRpcMsg rsp = {
       .info = pMsg->info,
       .code = 0,
@@ -513,6 +565,9 @@ int32_t tqProcessTaskRecoverReq(STQ* pTq, SRpcMsg* pMsg) {
   SStreamTaskRecoverReq* pReq = pMsg->pCont;
   int32_t                taskId = pReq->taskId;
   SStreamTask*           pTask = *(SStreamTask**)taosHashGet(pTq->pStreamTasks, &taskId, sizeof(int32_t));
+  if (atomic_load_8(&pTask->taskStatus) != TASK_STATUS__NORMAL) {
+    return 0;
+  }
   streamProcessRecoverReq(pTask, pReq, pMsg);
   return 0;
 }
@@ -521,6 +576,9 @@ int32_t tqProcessTaskDispatchRsp(STQ* pTq, SRpcMsg* pMsg) {
   SStreamDispatchRsp* pRsp = POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead));
   int32_t             taskId = pRsp->taskId;
   SStreamTask*        pTask = *(SStreamTask**)taosHashGet(pTq->pStreamTasks, &taskId, sizeof(int32_t));
+  if (atomic_load_8(&pTask->taskStatus) != TASK_STATUS__NORMAL) {
+    return 0;
+  }
   streamProcessDispatchRsp(pTask, pRsp);
   return 0;
 }
@@ -529,16 +587,32 @@ int32_t tqProcessTaskRecoverRsp(STQ* pTq, SRpcMsg* pMsg) {
   SStreamTaskRecoverRsp* pRsp = pMsg->pCont;
   int32_t                taskId = pRsp->taskId;
   SStreamTask*           pTask = *(SStreamTask**)taosHashGet(pTq->pStreamTasks, &taskId, sizeof(int32_t));
+  if (atomic_load_8(&pTask->taskStatus) != TASK_STATUS__NORMAL) {
+    return 0;
+  }
   streamProcessRecoverRsp(pTask, pRsp);
   return 0;
 }
 
 int32_t tqProcessTaskDropReq(STQ* pTq, char* msg, int32_t msgLen) {
   SVDropStreamTaskReq* pReq = (SVDropStreamTaskReq*)msg;
+
+  SStreamTask* pTask = *(SStreamTask**)taosHashGet(pTq->pStreamTasks, &pReq->taskId, sizeof(int32_t));
+  atomic_store_8(&pTask->taskStatus, TASK_STATUS__DROPPING);
+  // todo
+  // clear queue
+  // push drop req into queue
+  // launch exec to free memory
+  // remove from hash
+  return 0;
+
+#if 0
   int32_t              code = taosHashRemove(pTq->pStreamTasks, &pReq->taskId, sizeof(int32_t));
+  // set status dropping
   ASSERT(code == 0);
   if (code == 0) {
     // sendrsp
   }
   return code;
+#endif
 }

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

@@ -45,9 +45,9 @@ static int32_t tqLoopExecFromQueue(STQ* pTq, STqHandle* pHandle, SStreamDataSubm
 int32_t tqExecFromInputQ(STQ* pTq, STqHandle* pHandle) {
   SMqDataBlkRsp rsp = {0};
   // 1. guard and set status executing
-  int8_t execStatus =
-      atomic_val_compare_exchange_8(&pHandle->pushHandle.execStatus, TASK_STATUS__IDLE, TASK_STATUS__EXECUTING);
-  if (execStatus == TASK_STATUS__IDLE) {
+  int8_t execStatus = atomic_val_compare_exchange_8(&pHandle->pushHandle.execStatus, TASK_EXEC_STATUS__IDLE,
+                                                    TASK_EXEC_STATUS__EXECUTING);
+  if (execStatus == TASK_EXEC_STATUS__IDLE) {
     SStreamDataSubmit* pSubmit = NULL;
     // 2. check processedVer
     // 2.1. if not missed, get msg from queue
@@ -68,18 +68,18 @@ int32_t tqExecFromInputQ(STQ* pTq, STqHandle* pHandle) {
       goto SEND_RSP;
     }
     // set exec status closing
-    atomic_store_8(&pHandle->pushHandle.execStatus, TASK_STATUS__CLOSING);
+    atomic_store_8(&pHandle->pushHandle.execStatus, TASK_EXEC_STATUS__CLOSING);
     // second run
     if (tqLoopExecFromQueue(pTq, pHandle, &pSubmit, &rsp) == 0) {
       goto SEND_RSP;
     }
     // set exec status idle
-    atomic_store_8(&pHandle->pushHandle.execStatus, TASK_STATUS__IDLE);
+    atomic_store_8(&pHandle->pushHandle.execStatus, TASK_EXEC_STATUS__IDLE);
   }
 SEND_RSP:
   // 4. if get result
   // 4.1 set exec input status blocked and exec status idle
-  atomic_store_8(&pHandle->pushHandle.execStatus, TASK_STATUS__IDLE);
+  atomic_store_8(&pHandle->pushHandle.execStatus, TASK_EXEC_STATUS__IDLE);
   // 4.2 rpc send
   rsp.rspOffset = pHandle->pushHandle.processedVer;
   /*if (tqSendPollRsp(pTq, pMsg, pReq, &rsp) < 0) {*/
@@ -150,7 +150,7 @@ int32_t tqEnqueueAll(STQ* pTq, SSubmitReq* pReq) {
       continue;
     }
     int8_t execStatus = atomic_load_8(&pHandle->pushHandle.execStatus);
-    if (execStatus == TASK_STATUS__IDLE || execStatus == TASK_STATUS__CLOSING) {
+    if (execStatus == TASK_EXEC_STATUS__IDLE || execStatus == TASK_EXEC_STATUS__CLOSING) {
       tqSendExecReq(pTq, pHandle);
     }
   }

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

@@ -42,6 +42,25 @@ int64_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, SWalHead*
       code = 0;
       goto END;
     } else {
+      if (pHandle->fetchMeta) {
+        SWalReadHead* pHead = &((*ppHeadWithCkSum)->head);
+        if (pHead->msgType == TDMT_VND_CREATE_STB || pHead->msgType == TDMT_VND_ALTER_STB ||
+            pHead->msgType == TDMT_VND_DROP_STB || pHead->msgType == TDMT_VND_CREATE_TABLE ||
+            pHead->msgType == TDMT_VND_ALTER_TABLE || pHead->msgType == TDMT_VND_DROP_TABLE ||
+            pHead->msgType == TDMT_VND_DROP_TTL_TABLE) {
+          code = walFetchBody(pHandle->pWalReader, ppHeadWithCkSum);
+
+          if (code < 0) {
+            ASSERT(0);
+            *fetchOffset = offset;
+            code = -1;
+            goto END;
+          }
+          *fetchOffset = offset;
+          code = 0;
+          goto END;
+        }
+      }
       code = walSkipFetchBody(pHandle->pWalReader, *ppHeadWithCkSum);
       if (code < 0) {
         ASSERT(0);

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

@@ -172,7 +172,7 @@ int32_t vnodeProcessWriteReq(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRp
         goto _err;
       }
     } break;
-    case TDMT_VND_STREAM_TASK_DROP: {
+    case TDMT_STREAM_TASK_DROP: {
       if (tqProcessTaskDropReq(pVnode->pTq, pMsg->pCont, pMsg->contLen) < 0) {
         goto _err;
       }

source/libs/stream/src/stream.c

@@ -50,6 +50,10 @@ void streamCleanUp() {
 void streamTriggerByTimer(void* param, void* tmrId) {
   SStreamTask* pTask = (void*)param;
 
+  if (atomic_load_8(&pTask->taskStatus) == TASK_STATUS__DROPPING) {
+    return;
+  }
+
   if (atomic_load_8(&pTask->triggerStatus) == TASK_TRIGGER_STATUS__ACTIVE) {
     SStreamTrigger* trigger = taosAllocateQitem(sizeof(SStreamTrigger), DEF_QITEM);
     if (trigger == NULL) return;
@@ -82,8 +86,8 @@ int32_t streamSetupTrigger(SStreamTask* pTask) {
 }
 
 int32_t streamLaunchByWrite(SStreamTask* pTask, int32_t vgId) {
-  int8_t execStatus = atomic_load_8(&pTask->status);
-  if (execStatus == TASK_STATUS__IDLE || execStatus == TASK_STATUS__CLOSING) {
+  int8_t execStatus = atomic_load_8(&pTask->execStatus);
+  if (execStatus == TASK_EXEC_STATUS__IDLE || execStatus == TASK_EXEC_STATUS__CLOSING) {
     SStreamTaskRunReq* pRunReq = rpcMallocCont(sizeof(SStreamTaskRunReq));
     if (pRunReq == NULL) return -1;
 
@@ -188,6 +192,7 @@ int32_t streamProcessDispatchRsp(SStreamTask* pTask, SStreamDispatchRsp* pRsp) {
 
 int32_t streamProcessRunReq(SStreamTask* pTask) {
   streamExec(pTask, pTask->pMsgCb);
+
   if (pTask->dispatchType != TASK_DISPATCH__NONE) {
     streamDispatch(pTask, pTask->pMsgCb);
   }

source/libs/stream/src/streamExec.c

@@ -33,6 +33,9 @@ static int32_t streamTaskExecImpl(SStreamTask* pTask, void* data, SArray* pRes)
     ASSERT(pTask->inputType == STREAM_INPUT__DATA_BLOCK);
     SArray* blocks = pBlock->blocks;
     qSetMultiStreamInput(exec, blocks->pData, blocks->size, STREAM_DATA_TYPE_SSDATA_BLOCK, false);
+  } else if (pItem->type == STREAM_INPUT__DROP) {
+    // TODO exec drop
+    return 0;
   }
 
   // exec
@@ -58,6 +61,10 @@ static SArray* streamExecForQall(SStreamTask* pTask, SArray* pRes) {
 
     streamTaskExecImpl(pTask, data, pRes);
 
+    if (pTask->taskStatus == TASK_STATUS__DROPPING) {
+      return NULL;
+    }
+
     if (taosArrayGetSize(pRes) != 0) {
       SStreamDataBlock* qRes = taosAllocateQitem(sizeof(SStreamDataBlock), DEF_QITEM);
       if (qRes == NULL) {
@@ -75,6 +82,10 @@ static SArray* streamExecForQall(SStreamTask* pTask, SArray* pRes) {
         return NULL;
       }
 
+      if (((SStreamQueueItem*)data)->type == STREAM_INPUT__TRIGGER) {
+        blockDataDestroy(((SStreamTrigger*)data)->pBlock);
+        taosFreeQitem(data);
+      } else {
         if (pTask->inputType == STREAM_INPUT__DATA_SUBMIT) {
           streamDataSubmitRefDec((SStreamDataSubmit*)data);
           taosFreeQitem(data);
@@ -82,6 +93,7 @@ static SArray* streamExecForQall(SStreamTask* pTask, SArray* pRes) {
           taosArrayDestroyEx(((SStreamDataBlock*)data)->blocks, (FDelete)tDeleteSSDataBlock);
           taosFreeQitem(data);
         }
+      }
       streamQueueProcessSuccess(pTask->inputQueue);
       return taosArrayInit(0, sizeof(SSDataBlock));
     }
@@ -94,25 +106,26 @@ int32_t streamExec(SStreamTask* pTask, SMsgCb* pMsgCb) {
   SArray* pRes = taosArrayInit(0, sizeof(SSDataBlock));
   if (pRes == NULL) return -1;
   while (1) {
-    int8_t execStatus = atomic_val_compare_exchange_8(&pTask->status, TASK_STATUS__IDLE, TASK_STATUS__EXECUTING);
-    if (execStatus == TASK_STATUS__IDLE) {
+    int8_t execStatus =
+        atomic_val_compare_exchange_8(&pTask->execStatus, TASK_EXEC_STATUS__IDLE, TASK_EXEC_STATUS__EXECUTING);
+    if (execStatus == TASK_EXEC_STATUS__IDLE) {
       // first run
       pRes = streamExecForQall(pTask, pRes);
       if (pRes == NULL) goto FAIL;
 
       // set status closing
-      atomic_store_8(&pTask->status, TASK_STATUS__CLOSING);
+      atomic_store_8(&pTask->execStatus, TASK_EXEC_STATUS__CLOSING);
 
       // second run, make sure inputQ and qall are cleared
       pRes = streamExecForQall(pTask, pRes);
       if (pRes == NULL) goto FAIL;
 
       taosArrayDestroy(pRes);
-      atomic_store_8(&pTask->status, TASK_STATUS__IDLE);
+      atomic_store_8(&pTask->execStatus, TASK_EXEC_STATUS__IDLE);
       return 0;
-    } else if (execStatus == TASK_STATUS__CLOSING) {
+    } else if (execStatus == TASK_EXEC_STATUS__CLOSING) {
       continue;
-    } else if (execStatus == TASK_STATUS__EXECUTING) {
+    } else if (execStatus == TASK_EXEC_STATUS__EXECUTING) {
       ASSERT(taosArrayGetSize(pRes) == 0);
       taosArrayDestroy(pRes);
       return 0;
@@ -122,7 +135,7 @@ int32_t streamExec(SStreamTask* pTask, SMsgCb* pMsgCb) {
   }
 FAIL:
   if (pRes) taosArrayDestroy(pRes);
-  atomic_store_8(&pTask->status, TASK_STATUS__IDLE);
+  atomic_store_8(&pTask->execStatus, TASK_EXEC_STATUS__IDLE);
   return -1;
 }
 

source/libs/stream/src/streamTask.c

@@ -23,7 +23,7 @@ SStreamTask* tNewSStreamTask(int64_t streamId) {
   }
   pTask->taskId = tGenIdPI32();
   pTask->streamId = streamId;
-  pTask->status = TASK_STATUS__IDLE;
+  pTask->execStatus = TASK_EXEC_STATUS__IDLE;
   pTask->inputStatus = TASK_INPUT_STATUS__NORMAL;
   pTask->outputStatus = TASK_OUTPUT_STATUS__NORMAL;
 
@@ -35,7 +35,8 @@ int32_t tEncodeSStreamTask(SEncoder* pEncoder, const SStreamTask* pTask) {
   if (tEncodeI64(pEncoder, pTask->streamId) < 0) return -1;
   if (tEncodeI32(pEncoder, pTask->taskId) < 0) return -1;
   if (tEncodeI8(pEncoder, pTask->inputType) < 0) return -1;
-  if (tEncodeI8(pEncoder, pTask->status) < 0) return -1;
+  if (tEncodeI8(pEncoder, pTask->taskStatus) < 0) return -1;
+  if (tEncodeI8(pEncoder, pTask->execStatus) < 0) return -1;
   if (tEncodeI8(pEncoder, pTask->execType) < 0) return -1;
   if (tEncodeI8(pEncoder, pTask->sinkType) < 0) return -1;
   if (tEncodeI8(pEncoder, pTask->dispatchType) < 0) return -1;
@@ -83,7 +84,8 @@ int32_t tDecodeSStreamTask(SDecoder* pDecoder, SStreamTask* pTask) {
   if (tDecodeI64(pDecoder, &pTask->streamId) < 0) return -1;
   if (tDecodeI32(pDecoder, &pTask->taskId) < 0) return -1;
   if (tDecodeI8(pDecoder, &pTask->inputType) < 0) return -1;
-  if (tDecodeI8(pDecoder, &pTask->status) < 0) return -1;
+  if (tDecodeI8(pDecoder, &pTask->taskStatus) < 0) return -1;
+  if (tDecodeI8(pDecoder, &pTask->execStatus) < 0) return -1;
   if (tDecodeI8(pDecoder, &pTask->execType) < 0) return -1;
   if (tDecodeI8(pDecoder, &pTask->sinkType) < 0) return -1;
   if (tDecodeI8(pDecoder, &pTask->dispatchType) < 0) return -1;

source/libs/transport/inc/transLog.h

@@ -33,7 +33,7 @@ extern "C" {
 #define tDump(x, y) do {if (rpcDebugFlag & DEBUG_DUMP) { taosDumpData((unsigned char *)x, y); } } while(0)
 
 //#define tTR(param, ...) do { char buf[40] = {0};TRACE_TO_STR(trace, buf);tTrace("TRID: %s "param, buf, __VA_ARGS__);} while(0)
-#define tGTrace(param, ...) do { char buf[40] = {0}; TRACE_TO_STR(trace, buf); tTrace(param ", GTID: %s", __VA_ARGS__, buf);} while(0)
+#define tGTrace(param, ...) do { char buf[40] = {0}; TRACE_TO_STR(trace, buf); tTrace(param ", gtid:%s", __VA_ARGS__, buf);} while(0)
 
 // clang-format on 
 #ifdef __cplusplus

tests/pytest/util/constant.py

+# -*- coding: utf-8 -*-
+
+# basic data type boundary
+TINYINT_MAX             = 127
+TINYINT_MIN             = -128
+
+TINYINT_UN_MAX          = 255
+TINYINT_UN_MIN          = 0
+
+SMALLINT_MAX            = 32767
+SMALLINT_MIN            = -32768
+
+SMALLINT_UN_MAX         = 65535
+MALLINT_UN_MIN          = 0
+
+INT_MAX                 = 2147483647
+INT_MIN                 = -2147483648
+
+INT_UN_MAX              = 4294967295
+INT_UN_MIN              = 0
+
+BIGINT_MAX              = 9223372036854775807
+BIGINT_MIN              = -9223372036854775808
+
+BIGINT_UN_MAX           = 18446744073709551615
+BIGINT_UN_MIN           = 0
+
+FLOAT_MAX               = 3.40E+38
+FLOAT_MIN               = -3.40E+38
+
+DOUBLE_MAX              = 1.7E+308
+DOUBLE_MIN              = -1.7E+308
+
+# schema boundary
+BINARY_LENGTH_MAX       = 16374
+NCAHR_LENGTH_MAX_       = 4093
+DBNAME_LENGTH_MAX_      = 64
+
+STBNAME_LENGTH_MAX      = 192
+STBNAME_LENGTH_MIN      = 1
+
+TBNAME_LENGTH_MAX       = 192
+TBNAME_LENGTH_MIN       = 1
+
+CHILD_TBNAME_LENGTH_MAX = 192
+CHILD_TBNAME_LENGTH_MIN = 1
+
+TAG_NAME_LENGTH_MAX     = 64
+TAG_NAME_LENGTH_MIN     = 1
+
+COL_NAME_LENGTH_MAX     = 64
+COL_NAME_LENGTH_MIN     = 1
+
+TAG_COUNT_MAX           = 128
+TAG_COUNT_MIN           = 1
+
+COL_COUNT_MAX           = 4096
+COL_COUNT_MIN           = 2
+
+TAG_COL_COUNT_MAX       = 4096
+TAG_COL_COUNT_MIN       = 3
+
+MNODE_SHM_SIZE_MAX      = 2147483647
+MNODE_SHM_SIZE_MIN      = 6292480
+MNODE_SHM_SIZE_DEFAULT  = 6292480
+
+VNODE_SHM_SIZE_MAX      = 2147483647
+VNODE_SHM_SIZE_MIN      = 6292480
+VNODE_SHM_SIZE_DEFAULT  = 31458304
\ No newline at end of file

tests/system-test/1-insert/create_retentions.py

 import datetime
 
+from dataclasses import dataclass
+from typing import List
 from util.log import *
 from util.sql import *
 from util.cases import *
@@ -28,6 +30,28 @@ CHAR_COL    = [ BINARY_COL, NCHAR_COL, ]
 BOOLEAN_COL = [ BOOL_COL, ]
 TS_TYPE_COL = [ TS_COL, ]
 
+## insert data args：
+TIME_STEP = 10000
+NOW = int(datetime.datetime.timestamp(datetime.datetime.now()) * 1000)
+
+@dataclass
+class DataSet:
+    ts_data         : List[int]     = None
+    int_data        : List[int]     = None
+    bint_data       : List[int]     = None
+    sint_data       : List[int]     = None
+    tint_data       : List[int]     = None
+    int_un_data     : List[int]     = None
+    bint_un_data    : List[int]     = None
+    sint_un_data    : List[int]     = None
+    tint_un_data    : List[int]     = None
+    float_data      : List[float]   = None
+    double_data     : List[float]   = None
+    bool_data       : List[int]     = None
+    binary_data     : List[str]     = None
+    nchar_data      : List[str]     = None
+
+
 class TDTestCase:
 
     def init(self, conn, logSql):
@@ -37,10 +61,13 @@ class TDTestCase:
     @property
     def create_databases_sql_err(self):
         return [
-            "create database if not exists  db1 retentions 0s:1d",
-            "create database if not exists  db1 retentions 1s:1y",
-            "create database if not exists  db1 retentions 1s:1n",
-            "create database if not exists  db1 retentions 1s:1n,2s:2d,3s:3d,4s:4d",
+            "create database db1 retentions 0s:1d",
+            "create database db3 retentions 1s:0d",
+            "create database db1 retentions 1s:1y",
+            "create database db1 retentions 1s:1n",
+            "create database db2 retentions 1w:1d ;",
+            "create database db5 retentions 1s:1d,3s:3d,2s:2d",
+            "create database db1 retentions 1s:1n,2s:2d,3s:3d,4s:4d",
         ]
 
     @property
@@ -60,16 +87,41 @@ class TDTestCase:
     @property
     def create_stable_sql_err(self):
         return [
-            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(ceil) delay 1",
-            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(count) delay 1",
-            f"create stable stb2 ({PRIMARY_COL} timestamp, {INT_COL} int, {BINARY_COL} binary(16)) tags (tag1 int) rollup(avg) delay 1",
-            f"create stable stb2 ({PRIMARY_COL} timestamp, {INT_COL} int, {BINARY_COL} nchar(16)) tags (tag1 int) rollup(avg) delay 1",
+            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(ceil) watermark 1s maxdelay 1m",
+            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(count) watermark  1min",
+            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(min) maxdelay -1s",
+            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(min) watermark -1m",
+            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) watermark 1m ",
+            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) maxdelay 1m ",
+            f"create stable stb2 ({PRIMARY_COL} timestamp, {INT_COL} int, {BINARY_COL} binary(16)) tags (tag1 int) rollup(avg) watermark 1s",
+            f"create stable stb2 ({PRIMARY_COL} timestamp, {INT_COL} int, {BINARY_COL} nchar(16)) tags (tag1 int) rollup(avg) maxdelay 1m",
+            # f"create table ntb_1 ({PRIMARY_COL} timestamp, {INT_COL} int, {BINARY_COL} nchar(16)) rollup(avg) watermark 1s maxdelay 1s",
+            # f"create stable stb2 ({PRIMARY_COL} timestamp, {INT_COL} int, {BINARY_COL} nchar(16)) tags (tag1 int) " ,
+            # f"create stable stb2 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) " ,
+            # f"create stable stb2 ({PRIMARY_COL} timestamp, {INT_COL} int) " ,
+            # f"create stable stb2 ({PRIMARY_COL} timestamp, {INT_COL} int, {BINARY_COL} nchar(16)) " ,
+
+            # watermark, maxdelay: [0, 900000], [ms, s, m, ?]
+            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(min) maxdelay 1u",
+            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(min) watermark 1b",
+            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(min) watermark 900001ms",
+            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(min) maxdelay 16m",
+            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(min) maxdelay 901s",
+            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(min) maxdelay 1h",
+            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(min) maxdelay 0.2h",
+            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(min) watermark 0.002d",
+
         ]
 
     @property
     def create_stable_sql_current(self):
         return [
-            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(avg) delay 5",
+            f"create stable stb1 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(avg)",
+            f"create stable stb2 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(min) watermark 5s maxdelay 1m",
+            f"create stable stb3 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(max) watermark 5s maxdelay 1m",
+            f"create stable stb4 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(sum) watermark 5s maxdelay 1m",
+            # f"create stable stb5 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(last) watermark 5s maxdelay 1m",
+            # f"create stable stb6 ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) rollup(first) watermark 5s maxdelay 1m",
         ]
 
     def test_create_stb(self):
@@ -79,14 +131,19 @@ class TDTestCase:
         for cur_sql in self.create_stable_sql_current:
             tdSql.execute(cur_sql)
         tdSql.query("show stables")
+        # assert "rollup" in tdSql.description
         tdSql.checkRows(len(self.create_stable_sql_current))
 
+        # tdSql.execute("use db")  # because db is a noraml database, not a rollup database, should not be able to create a rollup database
+        # tdSql.error(f"create stable nor_db_rollup_stb ({PRIMARY_COL} timestamp, {INT_COL} int) tags (tag1 int) file_factor 5.0")
+
+
     def test_create_databases(self):
         for err_sql in self.create_databases_sql_err:
             tdSql.error(err_sql)
         for cur_sql in self.create_databases_sql_current:
             tdSql.execute(cur_sql)
-            tdSql.query("show databases")
+            # tdSql.query("show databases")
         for alter_sql in self.alter_database_sql:
             tdSql.error(alter_sql)
 
@@ -95,9 +152,7 @@ class TDTestCase:
         self.test_create_stb()
 
     def __create_tb(self):
-        tdSql.prepare()
-
-        tdLog.printNoPrefix("==========step1:create table")
+        tdLog.printNoPrefix("==========step: create table")
         create_stb_sql  =  f'''create table stb1(
                 ts timestamp, {INT_COL} int, {BINT_COL} bigint, {SINT_COL} smallint, {TINT_COL} tinyint,
                 {FLOAT_COL} float, {DOUBLE_COL} double, {BOOL_COL} bool,
@@ -120,129 +175,118 @@ class TDTestCase:
         for i in range(4):
             tdSql.execute(f'create table ct{i+1} using stb1 tags ( {i+1} )')
 
-    def __create_data_set(self, rows):
-        now_time = int(datetime.datetime.timestamp(datetime.datetime.now()) * 1000)
-        pos_data = []
-        neg_data = []
-        spec_data = []
-        for i in range(rows):
-            pos_data.append(
-                (
-                    now_time - i * 1000, i, 11111 * i, 111 * i % 32767 , 11 * i % 127, 1.11 * i, 1100.0011 * i,
-                    i % 2, f'binary{i}', f'nchar_测试_{i}', now_time + 1 * i, 11 * i % 127, 111 * i % 32767, i, 11111 * i
-                )
-            )
-            neg_data.append(
-                (
-                    now_time - i * 7776000000, -i, -11111 * i, -111 * i % 32767, -11 * i % 127, -1.11 * i, -1100.0011 * i,
-                    i % 2, f'binary{i}', f'nchar_测试_{i}', now_time + 1 * i, 11 * i % 127, 111 * i % 32767, i, 11111 * i
-                )
-            )
+    def __data_set(self, rows):
+        data_set = DataSet()
+        # neg_data_set = DataSet()
+        data_set.ts_data = []
+        data_set.int_data = []
+        data_set.bint_data = []
+        data_set.sint_data = []
+        data_set.tint_data = []
+        data_set.int_un_data = []
+        data_set.bint_un_data = []
+        data_set.sint_un_data = []
+        data_set.tint_un_data = []
+        data_set.float_data = []
+        data_set.double_data = []
+        data_set.bool_data = []
+        data_set.binary_data = []
+        data_set.nchar_data = []
 
-    def __insert_data(self, rows):
-        now_time = int(datetime.datetime.timestamp(datetime.datetime.now()) * 1000)
         for i in range(rows):
-            tdSql.execute(
-                f'''insert into ct1 values (
-                { now_time - i * 1000 }, {i}, {11111 * i}, {111 * i % 32767 }, {11 * i % 127}, {1.11*i}, {1100.0011*i},
-                {i%2}, 'binary{i}', 'nchar_测试_{i}', { now_time + 1 * i }, {11 * i % 127}, {111 * i % 32767}, {i}, {11111 * i} )'''
-            )
-            tdSql.execute(
-                f'''insert into ct4 values (
-                { now_time - i * 7776000000 }, {i}, {11111 * i}, {111 * i % 32767 }, {11 * i % 127}, {1.11*i}, {1100.0011*i},
-                {i%2}, 'binary{i}', 'nchar_测试_{i}', { now_time + 1 * i }, {11 * i % 127}, {111 * i % 32767}, {i}, {11111 * i} )'''
-            )
-            tdSql.execute(
-                f'''insert into ct2 values (
-                { now_time - i * 7776000000 }, {-i},  {-11111 * i}, {-111 * i % 32767 }, {-11 * i % 127}, {-1.11*i}, {-1100.0011*i},
-                {i%2}, 'binary{i}', 'nchar_测试_{i}', { now_time + 1 * i }, {11 * i % 127}, {111 * i % 32767}, {i}, {11111 * i} )'''
-            )
-        tdSql.execute(
-            f'''insert into ct1 values
-            ( { now_time - rows * 5 }, 0, 0, 0, 0, 0, 0, 0, 'binary0', 'nchar_测试_0', { now_time + 8 }, 0, 0, 0, 0)
-            ( { now_time + 10000 }, { rows }, -99999, -999, -99, -9.99, -99.99, 1, 'binary9', 'nchar_测试_9', { now_time + 9 }, 0, 0, 0, 0 )
+            data_set.ts_data.append(NOW + 1 * (rows - i))
+            data_set.int_data.append(rows - i)
+            data_set.bint_data.append(11111 * (rows - i))
+            data_set.sint_data.append(111 * (rows - i) % 32767)
+            data_set.tint_data.append(11 * (rows - i) % 127)
+            data_set.int_un_data.append(rows - i)
+            data_set.bint_un_data.append(11111 * (rows - i))
+            data_set.sint_un_data.append(111 * (rows - i) % 32767)
+            data_set.tint_un_data.append(11 * (rows - i) % 127)
+            data_set.float_data.append(1.11 * (rows - i))
+            data_set.double_data.append(1100.0011 * (rows - i))
+            data_set.bool_data.append((rows - i) % 2)
+            data_set.binary_data.append(f'binary{(rows - i)}')
+            data_set.nchar_data.append(f'nchar_测试_{(rows - i)}')
+
+            # neg_data_set.ts_data.append(-1 * i)
+            # neg_data_set.int_data.append(-i)
+            # neg_data_set.bint_data.append(-11111 * i)
+            # neg_data_set.sint_data.append(-111 * i % 32767)
+            # neg_data_set.tint_data.append(-11 * i % 127)
+            # neg_data_set.int_un_data.append(-i)
+            # neg_data_set.bint_un_data.append(-11111 * i)
+            # neg_data_set.sint_un_data.append(-111 * i % 32767)
+            # neg_data_set.tint_un_data.append(-11 * i % 127)
+            # neg_data_set.float_data.append(-1.11 * i)
+            # neg_data_set.double_data.append(-1100.0011 * i)
+            # neg_data_set.binary_data.append(f'binary{i}')
+            # neg_data_set.nchar_data.append(f'nchar_测试_{i}')
+
+        return data_set
+
+    def __insert_data(self):
+        data = self.__data_set(rows=self.rows)
+
+        # now_time = int(datetime.datetime.timestamp(datetime.datetime.now()) * 1000)
+        null_data = '''null, null, null, null, null, null, null, null, null, null, null, null, null, null'''
+        zero_data = "0, 0, 0, 0, 0, 0, 0, 'binary_0', 'nchar_0', 0, 0, 0, 0, 0"
+
+        for i in range(self.rows):
+            row_data = f'''
+                {data.int_data[i]}, {data.bint_data[i]}, {data.sint_data[i]}, {data.tint_data[i]}, {data.float_data[i]}, {data.double_data[i]},
+                {data.bool_data[i]}, '{data.binary_data[i]}', '{data.nchar_data[i]}', {data.ts_data[i]}, {data.tint_un_data[i]},
+                {data.sint_un_data[i]}, {data.int_un_data[i]}, {data.bint_un_data[i]}
             '''
-        )
-
-        tdSql.execute(
-            f'''insert into ct4 values
-            ( { now_time - rows * 7776000000 }, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL )
-            ( { now_time - rows * 3888000000 + 10800000 }, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL )
-            ( { now_time +  7776000000 }, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL )
-            (
-                { now_time + 5184000000}, {pow(2,31)-pow(2,15)}, {pow(2,63)-pow(2,30)}, 32767, 127,
-                { 3.3 * pow(10,38) }, { 1.3 * pow(10,308) }, { rows % 2 }, "binary_limit-1", "nchar_测试_limit-1", { now_time - 86400000},
-                254, 65534, {pow(2,32)-pow(2,16)}, {pow(2,64)-pow(2,31)}
-                )
-            (
-                { now_time + 2592000000 }, {pow(2,31)-pow(2,16)}, {pow(2,63)-pow(2,31)}, 32766, 126,
-                { 3.2 * pow(10,38) }, { 1.2 * pow(10,308) }, { (rows-1) % 2 }, "binary_limit-2", "nchar_测试_limit-2", { now_time - 172800000},
-                255, 65535, {pow(2,32)-pow(2,15)}, {pow(2,64)-pow(2,30)}
-                )
+            neg_row_data = f'''
+                {-1 * data.int_data[i]}, {-1 * data.bint_data[i]}, {-1 * data.sint_data[i]}, {-1 * data.tint_data[i]}, {-1 * data.float_data[i]}, {-1 * data.double_data[i]},
+                {data.bool_data[i]}, '{data.binary_data[i]}', '{data.nchar_data[i]}', {data.ts_data[i]}, {1 * data.tint_un_data[i]},
+                {1 * data.sint_un_data[i]}, {1 * data.int_un_data[i]}, {1 * data.bint_un_data[i]}
             '''
-        )
 
-        tdSql.execute(
-            f'''insert into ct2 values
-            ( { now_time - rows * 7776000000 }, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL )
-            ( { now_time - rows * 3888000000 + 10800000 }, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL )
-            ( { now_time + 7776000000 }, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL )
-            (
-                { now_time + 5184000000 }, { -1 * pow(2,31) + pow(2,15) }, { -1 * pow(2,63) + pow(2,30) }, -32766, -126, { -1 * 3.2 * pow(10,38) },
-                { -1.2 * pow(10,308) }, { rows % 2 }, "binary_limit-1", "nchar_测试_limit-1", { now_time - 86400000 }, 1, 1, 1, 1
-                )
-            (
-                { now_time + 2592000000 }, { -1 * pow(2,31) + pow(2,16) }, { -1 * pow(2,63) + pow(2,31) }, -32767, -127, { - 3.3 * pow(10,38) },
-                { -1.3 * pow(10,308) }, { (rows-1) % 2 }, "binary_limit-2", "nchar_测试_limit-2", { now_time - 172800000 }, 1, 1, 1, 1
-                )
-            '''
-        )
+            tdSql.execute( f"insert into ct1 values ( {NOW - i * TIME_STEP}, {row_data} )" )
+            tdSql.execute( f"insert into ct2 values ( {NOW - i * int(TIME_STEP * 0.6)}, {neg_row_data} )" )
+            tdSql.execute( f"insert into ct4 values ( {NOW - i * int(TIME_STEP * 0.8) }, {row_data} )" )
+            tdSql.execute( f"insert into t1 values ( {NOW - i * int(TIME_STEP * 1.2)}, {row_data} )" )
 
-        for i in range(rows):
-            insert_data = f'''insert into t1 values
-                ( { now_time - i * 3600000 }, {i}, {i * 11111}, { i % 32767 }, { i % 127}, { i * 1.11111 }, { i * 1000.1111 }, { i % 2},
-                "binary_{i}", "nchar_测试_{i}", { now_time - 1000 * i }, {i % 127}, {i % 32767}, {i}, {i * 11111})
-                '''
-            tdSql.execute(insert_data)
-        tdSql.execute(
-            f'''insert into t1 values
-            ( { now_time + 10800000 }, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL )
-            ( { now_time - (( rows // 2 ) * 60 + 30) * 60000 }, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL)
-            ( { now_time - rows * 3600000 }, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL )
-            (
-                { now_time + 7200000 }, { pow(2,31) - pow(2,15) }, { pow(2,63) - pow(2,30) }, 32767, 127, { 3.3 * pow(10,38) },
-                { 1.3 * pow(10,308) }, { rows % 2 }, "binary_limit-1", "nchar_测试_limit-1", { now_time - 86400000 },
-                254, 65534, {pow(2,32)-pow(2,16)}, {pow(2,64)-pow(2,31)}
-                )
-            (
-                { now_time + 3600000 } , { pow(2,31) - pow(2,16) }, { pow(2,63) - pow(2,31) }, 32766, 126, { 3.2 * pow(10,38) },
-                { 1.2 * pow(10,308) }, { (rows-1) % 2 }, "binary_limit-2", "nchar_测试_limit-2", { now_time - 172800000 },
-                255, 65535, {pow(2,32)-pow(2,15)}, {pow(2,64)-pow(2,30)}
-                )
-            '''
-        )
+        tdSql.execute( f"insert into ct2 values ( {NOW + int(TIME_STEP * 0.6)}, {null_data} )" )
+        tdSql.execute( f"insert into ct2 values ( {NOW - (self.rows + 1) * int(TIME_STEP * 0.6)}, {null_data} )" )
+        tdSql.execute( f"insert into ct2 values ( {NOW - self.rows * int(TIME_STEP * 0.29) }, {null_data} )" )
+
+        tdSql.execute( f"insert into ct4 values ( {NOW + int(TIME_STEP * 0.8)}, {null_data} )" )
+        tdSql.execute( f"insert into ct4 values ( {NOW - (self.rows + 1) * int(TIME_STEP * 0.8)}, {null_data} )" )
+        tdSql.execute( f"insert into ct4 values ( {NOW - self.rows * int(TIME_STEP * 0.39)}, {null_data} )" )
+
+        tdSql.execute( f"insert into t1 values ( {NOW + int(TIME_STEP * 1.2)}, {null_data} )" )
+        tdSql.execute( f"insert into t1 values ( {NOW - (self.rows + 1) * int(TIME_STEP * 1.2)}, {null_data} )" )
+        tdSql.execute( f"insert into t1 values ( {NOW - self.rows * int(TIME_STEP * 0.59)}, {null_data} )" )
 
 
     def run(self):
-        tdSql.prepare()
+        self.rows = 10
+
+
+        tdLog.printNoPrefix("==========step0:all check")
+        # self.all_test()
 
-        tdLog.printNoPrefix("==========step1:create table")
+        tdLog.printNoPrefix("==========step1:create table in normal database")
+        tdSql.prepare()
         self.__create_tb()
+        self.__insert_data()
+        # return
 
-        tdLog.printNoPrefix("==========step2:insert data")
-        self.rows = 10
-        self.__insert_data(self.rows)
+        tdLog.printNoPrefix("==========step2:create table in rollup database")
+        tdSql.execute("create database db3 retentions 1s:4m,2s:8m,3s:12m")
+        tdSql.execute("use db3")
+        self.__create_tb()
+        self.__insert_data()
 
-        tdLog.printNoPrefix("==========step3:all check")
-        self.all_test()
         tdSql.execute("drop database if exists db1 ")
         tdSql.execute("drop database if exists db2 ")
 
         tdDnodes.stop(1)
         tdDnodes.start(1)
 
-        tdSql.execute("use db")
-
         tdLog.printNoPrefix("==========step4:after wal, all check again ")
         self.all_test()
 

tests/system-test/1-insert/time_range_wise.py

+import datetime
+
+from dataclasses import dataclass
+from typing import List, Any, Tuple
+from util.log import *
+from util.sql import *
+from util.cases import *
+from util.dnodes import *
+from util.constant import *
+
+PRIMARY_COL = "ts"
+
+INT_COL = "c_int"
+BINT_COL = "c_bint"
+SINT_COL = "c_sint"
+TINT_COL = "c_tint"
+FLOAT_COL = "c_float"
+DOUBLE_COL = "c_double"
+BOOL_COL = "c_bool"
+TINT_UN_COL = "c_tint_un"
+SINT_UN_COL = "c_sint_un"
+BINT_UN_COL = "c_bint_un"
+INT_UN_COL = "c_int_un"
+BINARY_COL = "c_binary"
+NCHAR_COL = "c_nchar"
+TS_COL = "c_ts"
+
+
+
+NUM_COL = [INT_COL, BINT_COL, SINT_COL, TINT_COL, FLOAT_COL, DOUBLE_COL, ]
+CHAR_COL = [BINARY_COL, NCHAR_COL, ]
+BOOLEAN_COL = [BOOL_COL, ]
+TS_TYPE_COL = [TS_COL, ]
+
+# insert data args：
+TIME_STEP = 10000
+NOW = int(datetime.datetime.timestamp(datetime.datetime.now()) * 1000)
+
+
+@dataclass
+class DataSet:
+    ts_data     : List[int]     = None
+    int_data    : List[int]     = None
+    bint_data   : List[int]     = None
+    sint_data   : List[int]     = None
+    tint_data   : List[int]     = None
+    int_un_data : List[int]     = None
+    bint_un_data: List[int]     = None
+    sint_un_data: List[int]     = None
+    tint_un_data: List[int]     = None
+    float_data  : List[float]   = None
+    double_data : List[float]   = None
+    bool_data   : List[int]     = None
+    binary_data : List[str]     = None
+    nchar_data  : List[str]     = None
+
+    def __post_init__(self):
+        self.ts_data        = []
+        self.int_data       = []
+        self.bint_data      = []
+        self.sint_data      = []
+        self.tint_data      = []
+        self.int_un_data    = []
+        self.bint_un_data   = []
+        self.sint_un_data   = []
+        self.tint_un_data   = []
+        self.float_data     = []
+        self.double_data    = []
+        self.bool_data      = []
+        self.binary_data    = []
+        self.nchar_data     = []
+
+
+@dataclass
+class SMAschema:
+    creation    : str           = "CREATE"
+    index_name  : str           = "sma_index_1"
+    index_flag  : str           = "SMA INDEX"
+    operator    : str           = "ON"
+    tbname      : str           = None
+    watermark   : str           = None
+    maxdelay    : str           = None
+    func        : Tuple[str]    = None
+    interval    : Tuple[str]    = None
+    sliding     : str           = None
+    other       : Any           = None
+    drop        : str           = "DROP"
+    drop_flag   : str           = "INDEX"
+
+    def __post_init__(self):
+        if isinstance(self.other, dict):
+            for k,v in self.other.items():
+
+                if k.lower() == "index_name" and isinstance(v, str) and not self.index_name:
+                    self.index_name = v
+                    del self.other[k]
+
+                if k.lower() == "index_flag" and isinstance(v, str) and not self.index_flag:
+                    self.index_flag = v
+                    del self.other[k]
+
+                if k.lower() == "operator" and isinstance(v, str) and not self.operator:
+                    self.operator = v
+                    del self.other[k]
+
+                if k.lower() == "tbname" and isinstance(v, str) and not self.tbname:
+                    self.tbname = v
+                    del self.other[k]
+
+                if k.lower() == "watermark" and isinstance(v, str) and not self.watermark:
+                    self.watermark = v
+                    del self.other[k]
+
+                if k.lower() == "maxdelay" and isinstance(v, str) and not self.maxdelay:
+                    self.maxdelay = v
+                    del self.other[k]
+
+                if k.lower() == "functions" and isinstance(v, tuple) and not self.func:
+                    self.func = v
+                    del self.other[k]
+
+                if k.lower() == "interval" and isinstance(v, tuple) and not self.interval:
+                    self.interval = v
+                    del self.other[k]
+
+                if k.lower() == "sliding" and isinstance(v, str) and not self.sliding:
+                    self.sliding = v
+                    del self.other[k]
+
+                if k.lower() == "drop_flag" and isinstance(v, str) and not self.drop_flag:
+                    self.drop_flag = v
+                    del self.other[k]
+
+
+class TDTestCase:
+
+    def init(self, conn, logSql):
+        tdLog.debug(f"start to excute {__file__}")
+        tdSql.init(conn.cursor(), False)
+
+    """
+        create sma index :
+            1. only create on stable, err_type: [child-table, normal-table]
+            2. one taosd, one sma index , err_type: [
+                one stb --> multi sma index,
+                multi stb in one db--> multi sma index,
+                multi stb in multi db --> multi sma index
+            ]
+            3. arg of (interval/sliding) in query sql is equal to this arg in sma index
+            4. client timezone is equal to timezone of sma index
+            5. does not take effect unless querySmaOptimize flag is turned on,
+    """
+    def __create_sma_index(self, sma:SMAschema):
+        sql = f"{sma.creation} {sma.index_flag} {sma.index_name} {sma.operator} {sma.tbname}"
+        if sma.func:
+            sql += f" function({', '.join(sma.func)})"
+        if sma.interval:
+            sql += f" interval({', '.join(sma.interval)})"
+        if sma.sliding:
+            sql += f" sliding({sma.sliding})"
+        if sma.watermark:
+            sql += f" watermark {sma.watermark}"
+        if sma.maxdelay:
+            sql += f" maxdelay {sma.maxdelay}"
+        if isinstance(sma.other, dict):
+            for k,v in sma.other.items():
+                if isinstance(v,tuple) or isinstance(v, list):
+                    sql += f" {k} ({' '.join(v)})"
+                else:
+                    sql += f" {k} {v}"
+        if isinstance(sma.other, tuple) or isinstance(sma.other, list):
+            sql += " ".join(sma.other)
+        if isinstance(sma.other, int) or isinstance(sma.other, float) or isinstance(sma.other, str):
+            sql += sma.other
+
+        return sql
+
+    def sma_create_check(self, sma:SMAschema):
+        tdSql.query("show stables")
+        stb_in_list = False
+        for row in tdSql.queryResult:
+            if sma.tbname == row[0]:
+                stb_in_list = True
+                break
+        if not stb_in_list:
+            tdSql.error(self.__create_sma_index(sma))
+        if not sma.creation:
+            tdSql.error(self.__create_sma_index(sma))
+        if not sma.index_flag:
+            tdSql.error(self.__create_sma_index(sma))
+        if not sma.index_name:
+            tdSql.error(self.__create_sma_index(sma))
+        if not sma.operator:
+            tdSql.error(self.__create_sma_index(sma))
+        if not sma.tbname:
+            tdSql.error(self.__create_sma_index(sma))
+        if not sma.func:
+            tdSql.error(self.__create_sma_index(sma))
+        if not sma.interval:
+            tdSql.error(self.__create_sma_index(sma))
+        if not sma.sliding:
+            tdSql.error(self.__create_sma_index(sma))
+        if sma.other:
+            tdSql.error(self.__create_sma_index(sma))
+
+
+    def all_test(self):
+        pass
+
+    def __create_tb(self):
+        tdLog.printNoPrefix("==========step: create table")
+        create_stb_sql = f'''create table stb1(
+                ts timestamp, {INT_COL} int, {BINT_COL} bigint, {SINT_COL} smallint, {TINT_COL} tinyint,
+                {FLOAT_COL} float, {DOUBLE_COL} double, {BOOL_COL} bool,
+                {BINARY_COL} binary(16), {NCHAR_COL} nchar(32), {TS_COL} timestamp,
+                {TINT_UN_COL} tinyint unsigned, {SINT_UN_COL} smallint unsigned,
+                {INT_UN_COL} int unsigned, {BINT_UN_COL} bigint unsigned
+            ) tags (tag1 int)
+            '''
+        create_ntb_sql = f'''create table t1(
+                ts timestamp, {INT_COL} int, {BINT_COL} bigint, {SINT_COL} smallint, {TINT_COL} tinyint,
+                {FLOAT_COL} float, {DOUBLE_COL} double, {BOOL_COL} bool,
+                {BINARY_COL} binary(16), {NCHAR_COL} nchar(32), {TS_COL} timestamp,
+                {TINT_UN_COL} tinyint unsigned, {SINT_UN_COL} smallint unsigned,
+                {INT_UN_COL} int unsigned, {BINT_UN_COL} bigint unsigned
+            )
+            '''
+        tdSql.execute(create_stb_sql)
+        tdSql.execute(create_ntb_sql)
+
+        for i in range(4):
+            tdSql.execute(f'create table ct{i+1} using stb1 tags ( {i+1} )')
+
+    def __data_set(self, rows):
+        data_set = DataSet()
+
+        for i in range(rows):
+            data_set.ts_data.append(NOW + 1 * (rows - i))
+            data_set.int_data.append(rows - i)
+            data_set.bint_data.append(11111 * (rows - i))
+            data_set.sint_data.append(111 * (rows - i) % 32767)
+            data_set.tint_data.append(11 * (rows - i) % 127)
+            data_set.int_un_data.append(rows - i)
+            data_set.bint_un_data.append(11111 * (rows - i))
+            data_set.sint_un_data.append(111 * (rows - i) % 32767)
+            data_set.tint_un_data.append(11 * (rows - i) % 127)
+            data_set.float_data.append(1.11 * (rows - i))
+            data_set.double_data.append(1100.0011 * (rows - i))
+            data_set.bool_data.append((rows - i) % 2)
+            data_set.binary_data.append(f'binary{(rows - i)}')
+            data_set.nchar_data.append(f'nchar_测试_{(rows - i)}')
+
+        return data_set
+
+    def __insert_data(self):
+        data = self.__data_set(rows=self.rows)
+
+        # now_time = int(datetime.datetime.timestamp(datetime.datetime.now()) * 1000)
+        null_data = '''null, null, null, null, null, null, null, null, null, null, null, null, null, null'''
+        zero_data = "0, 0, 0, 0, 0, 0, 0, 'binary_0', 'nchar_0', 0, 0, 0, 0, 0"
+
+        for i in range(self.rows):
+            row_data = f'''
+                {data.int_data[i]}, {data.bint_data[i]}, {data.sint_data[i]}, {data.tint_data[i]}, {data.float_data[i]}, {data.double_data[i]},
+                {data.bool_data[i]}, '{data.binary_data[i]}', '{data.nchar_data[i]}', {data.ts_data[i]}, {data.tint_un_data[i]},
+                {data.sint_un_data[i]}, {data.int_un_data[i]}, {data.bint_un_data[i]}
+            '''
+            neg_row_data = f'''
+                {-1 * data.int_data[i]}, {-1 * data.bint_data[i]}, {-1 * data.sint_data[i]}, {-1 * data.tint_data[i]}, {-1 * data.float_data[i]}, {-1 * data.double_data[i]},
+                {data.bool_data[i]}, '{data.binary_data[i]}', '{data.nchar_data[i]}', {data.ts_data[i]}, {1 * data.tint_un_data[i]},
+                {1 * data.sint_un_data[i]}, {1 * data.int_un_data[i]}, {1 * data.bint_un_data[i]}
+            '''
+
+            tdSql.execute(
+                f"insert into ct1 values ( {NOW - i * TIME_STEP}, {row_data} )")
+            tdSql.execute(
+                f"insert into ct2 values ( {NOW - i * int(TIME_STEP * 0.6)}, {neg_row_data} )")
+            tdSql.execute(
+                f"insert into ct4 values ( {NOW - i * int(TIME_STEP * 0.8) }, {row_data} )")
+            tdSql.execute(
+                f"insert into t1 values ( {NOW - i * int(TIME_STEP * 1.2)}, {row_data} )")
+
+        tdSql.execute(
+            f"insert into ct2 values ( {NOW + int(TIME_STEP * 0.6)}, {null_data} )")
+        tdSql.execute(
+            f"insert into ct2 values ( {NOW - (self.rows + 1) * int(TIME_STEP * 0.6)}, {null_data} )")
+        tdSql.execute(
+            f"insert into ct2 values ( {NOW - self.rows * int(TIME_STEP * 0.29) }, {null_data} )")
+
+        tdSql.execute(
+            f"insert into ct4 values ( {NOW + int(TIME_STEP * 0.8)}, {null_data} )")
+        tdSql.execute(
+            f"insert into ct4 values ( {NOW - (self.rows + 1) * int(TIME_STEP * 0.8)}, {null_data} )")
+        tdSql.execute(
+            f"insert into ct4 values ( {NOW - self.rows * int(TIME_STEP * 0.39)}, {null_data} )")
+
+        tdSql.execute(
+            f"insert into t1 values ( {NOW + int(TIME_STEP * 1.2)}, {null_data} )")
+        tdSql.execute(
+            f"insert into t1 values ( {NOW - (self.rows + 1) * int(TIME_STEP * 1.2)}, {null_data} )")
+        tdSql.execute(
+            f"insert into t1 values ( {NOW - self.rows * int(TIME_STEP * 0.59)}, {null_data} )")
+
+    def run(self):
+        sma1 = SMAschema(func=("min(c1)","max(c2)"))
+        sql1 = self.__create_sma_index(sma1)
+        print("================")
+        print(sql1)
+        # a = DataSet()
+        # return
+        self.rows = 10
+
+        tdLog.printNoPrefix("==========step0:all check")
+        # self.all_test()
+
+        tdLog.printNoPrefix("==========step1:create table in normal database")
+        tdSql.prepare()
+        self.__create_tb()
+        self.__insert_data()
+        return
+
+        tdLog.printNoPrefix("==========step2:create table in rollup database")
+        tdSql.execute("create database db3 retentions 1s:4m,2s:8m,3s:12m")
+        tdSql.execute("use db3")
+        self.__create_tb()
+        self.__insert_data()
+
+        tdSql.execute("drop database if exists db1 ")
+        tdSql.execute("drop database if exists db2 ")
+
+        tdDnodes.stop(1)
+        tdDnodes.start(1)
+
+        tdLog.printNoPrefix("==========step4:after wal, all check again ")
+        self.all_test()
+
+    def stop(self):
+        tdSql.close()
+        tdLog.success(f"{__file__} successfully executed")
+
+
+tdCases.addLinux(__file__, TDTestCase())
+tdCases.addWindows(__file__, TDTestCase())

tests/system-test/2-query/distribute_agg_avg.py

+from util.log import *
+from util.cases import *
+from util.sql import *
+import numpy as np
+import random ,os ,sys
+import platform
+
+
+class TDTestCase:
+    updatecfgDict = {'debugFlag': 143 ,"cDebugFlag":143,"uDebugFlag":143 ,"rpcDebugFlag":143 , "tmrDebugFlag":143 , 
+    "jniDebugFlag":143 ,"simDebugFlag":143,"dDebugFlag":143, "dDebugFlag":143,"vDebugFlag":143,"mDebugFlag":143,"qDebugFlag":143,
+    "wDebugFlag":143,"sDebugFlag":143,"tsdbDebugFlag":143,"tqDebugFlag":143 ,"fsDebugFlag":143 ,"fnDebugFlag":143,
+    "maxTablesPerVnode":2 ,"minTablesPerVnode":2,"tableIncStepPerVnode":2 }
+
+    def init(self, conn, logSql):
+        tdLog.debug("start to execute %s" % __file__)
+        tdSql.init(conn.cursor())
+        self.vnode_disbutes = None
+        self.ts = 1537146000000
+
+
+    def check_avg_functions(self, tbname , col_name):
+
+        avg_sql = f"select avg({col_name}) from {tbname};"
+
+        same_sql = f"select {col_name} from {tbname} where {col_name} is not null "
+   
+        tdSql.query(same_sql)
+        pre_data = np.array(tdSql.queryResult)[np.array(tdSql.queryResult) != None]
+        if (platform.system().lower() == 'windows' and pre_data.dtype == 'int32'):
+            pre_data = np.array(pre_data, dtype = 'int64')
+        pre_avg = np.sum(pre_data)/len(pre_data)
+
+        tdSql.query(avg_sql)
+        tdSql.checkData(0,0,pre_avg)
+
+    def prepare_datas_of_distribute(self):
+        
+        # prepate datas for  20 tables distributed at different vgroups
+        tdSql.execute("create database if not exists testdb keep 3650 duration 1000 vgroups 5")
+        tdSql.execute(" use testdb ")
+        tdSql.execute(
+            '''create table stb1
+            (ts timestamp, c1 int, c2 bigint, c3 smallint, c4 tinyint, c5 float, c6 double, c7 bool, c8 binary(16),c9 nchar(32), c10 timestamp)
+            tags (t0 timestamp, t1 int, t2 bigint, t3 smallint, t4 tinyint, t5 float, t6 double, t7 bool, t8 binary(16),t9 nchar(32))
+            '''
+        )
+
+        tdSql.execute(
+            '''
+            create table t1
+            (ts timestamp, c1 int, c2 bigint, c3 smallint, c4 tinyint, c5 float, c6 double, c7 bool, c8 binary(16),c9 nchar(32), c10 timestamp)
+            '''
+        )
+        for i in range(20):
+            tdSql.execute(f'create table ct{i+1} using stb1 tags ( now(), {1*i}, {11111*i}, {111*i}, {11*i}, {1.11*i}, {11.11*i}, {i%2}, "binary{i}", "nchar{i}" )')
+
+        for i in range(9):
+            tdSql.execute(
+                f"insert into ct1 values ( now()-{i*10}s, {1*i}, {11111*i}, {111*i}, {11*i}, {1.11*i}, {11.11*i}, {i%2}, 'binary{i}', 'nchar{i}', now()+{1*i}a )"
+            )
+            tdSql.execute(
+                f"insert into ct4 values ( now()-{i*90}d, {1*i}, {11111*i}, {111*i}, {11*i}, {1.11*i}, {11.11*i}, {i%2}, 'binary{i}', 'nchar{i}', now()+{1*i}a )"
+            )
+
+        for i in range(1,21):
+            if i ==1 or i == 4:
+                continue
+            else:
+                tbname = "ct"+f'{i}'
+                for j in range(9):
+                    tdSql.execute(
+                f"insert into {tbname} values ( now()-{(i+j)*10}s, {1*(j+i)}, {11111*(j+i)}, {111*(j+i)}, {11*(j)}, {1.11*(j+i)}, {11.11*(j+i)}, {(j+i)%2}, 'binary{j}', 'nchar{j}', now()+{1*j}a )"
+            )
+        tdSql.execute("insert into ct1 values (now()-45s, 0, 0, 0, 0, 0, 0, 0, 'binary0', 'nchar0', now()+8a )")
+        tdSql.execute("insert into ct1 values (now()+10s, 9, -99999, -999, -99, -9.99, -99.99, 1, 'binary9', 'nchar9', now()+9a )")
+        tdSql.execute("insert into ct1 values (now()+15s, 9, -99999, -999, -99, -9.99, NULL, 1, 'binary9', 'nchar9', now()+9a )")
+        tdSql.execute("insert into ct1 values (now()+20s, 9, -99999, -999, NULL, -9.99, -99.99, 1, 'binary9', 'nchar9', now()+9a )")
+
+        tdSql.execute("insert into ct4 values (now()-810d, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL ) ")
+        tdSql.execute("insert into ct4 values (now()-400d, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL ) ")
+        tdSql.execute("insert into ct4 values (now()+90d, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL  ) ")
+
+        tdSql.execute(
+            f'''insert into t1 values
+            ( '2020-04-21 01:01:01.000', NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL )
+            ( '2020-10-21 01:01:01.000', 1, 11111, 111, 11, 1.11, 11.11, 1, "binary1", "nchar1", now()+1a )
+            ( '2020-12-31 01:01:01.000', 2, 22222, 222, 22, 2.22, 22.22, 0, "binary2", "nchar2", now()+2a )
+            ( '2021-01-01 01:01:06.000', 3, 33333, 333, 33, 3.33, 33.33, 0, "binary3", "nchar3", now()+3a )
+            ( '2021-05-07 01:01:10.000', 4, 44444, 444, 44, 4.44, 44.44, 1, "binary4", "nchar4", now()+4a )
+            ( '2021-07-21 01:01:01.000', NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL )
+            ( '2021-09-30 01:01:16.000', 5, 55555, 555, 55, 5.55, 55.55, 0, "binary5", "nchar5", now()+5a )
+            ( '2022-02-01 01:01:20.000', 6, 66666, 666, 66, 6.66, 66.66, 1, "binary6", "nchar6", now()+6a )
+            ( '2022-10-28 01:01:26.000', 7, 00000, 000, 00, 0.00, 00.00, 1, "binary7", "nchar7", "1970-01-01 08:00:00.000" )
+            ( '2022-12-01 01:01:30.000', 8, -88888, -888, -88, -8.88, -88.88, 0, "binary8", "nchar8", "1969-01-01 01:00:00.000" )
+            ( '2022-12-31 01:01:36.000', 9, -99999999999999999, -999, -99, -9.99, -999999999999999999999.99, 1, "binary9", "nchar9", "1900-01-01 00:00:00.000" )
+            ( '2023-02-21 01:01:01.000', NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL )
+            '''
+        )
+
+        tdLog.info(" prepare data for distributed_aggregate done! ")
+
+    def check_distribute_datas(self):
+        # get vgroup_ids of all
+        tdSql.query("show vgroups ")
+        vgroups = tdSql.queryResult
+
+        vnode_tables={}
+        
+        for vgroup_id in vgroups:
+            vnode_tables[vgroup_id[0]]=[]
+        
+
+        # check sub_table of per vnode ,make sure sub_table has been distributed
+        tdSql.query("show tables like 'ct%'")
+        table_names = tdSql.queryResult
+        tablenames = []
+        for table_name in table_names:
+            vnode_tables[table_name[6]].append(table_name[0]) 
+        self.vnode_disbutes = vnode_tables
+
+        count = 0
+        for k ,v in vnode_tables.items():
+            if len(v)>=2:
+                count+=1
+        if count < 2:
+            tdLog.exit(" the datas of all not satisfy sub_table has been distributed ")
+
+    def check_avg_distribute_diff_vnode(self,col_name):
+    
+        vgroup_ids = []
+        for k ,v in self.vnode_disbutes.items():
+            if len(v)>=2:
+                vgroup_ids.append(k)
+        
+        distribute_tbnames = []
+        
+        for vgroup_id in vgroup_ids:
+            vnode_tables = self.vnode_disbutes[vgroup_id]
+            distribute_tbnames.append(random.sample(vnode_tables,1)[0])
+        tbname_ins = ""
+        for tbname in distribute_tbnames:
+            tbname_ins += "'%s' ,"%tbname
+
+        tbname_filters = tbname_ins[:-1]
+        
+        avg_sql = f"select avg({col_name}) from stb1 where tbname in ({tbname_filters});"
+
+        same_sql = f"select {col_name}  from stb1 where tbname in ({tbname_filters}) and {col_name} is not null "
+
+        tdSql.query(same_sql)
+        pre_data = np.array(tdSql.queryResult)[np.array(tdSql.queryResult) != None]
+        if (platform.system().lower() == 'windows' and pre_data.dtype == 'int32'):
+            pre_data = np.array(pre_data, dtype = 'int64')
+        pre_avg = np.sum(pre_data)/len(pre_data)
+
+        tdSql.query(avg_sql)
+        tdSql.checkData(0,0,pre_avg)
+
+    def check_avg_status(self):
+        # check max function work status 
+        
+        tdSql.query("show tables like 'ct%'")
+        table_names = tdSql.queryResult
+        tablenames = []
+        for table_name in table_names:
+            tablenames.append(table_name[0])
+
+        tdSql.query("desc stb1")
+        col_names = tdSql.queryResult
+        
+        colnames = []
+        for col_name in col_names:
+            if col_name[1] in ["INT" ,"BIGINT" ,"SMALLINT" ,"TINYINT" , "FLOAT" ,"DOUBLE"]:
+                colnames.append(col_name[0])
+        
+        for tablename in tablenames:
+            for colname in colnames:
+                self.check_avg_functions(tablename,colname)
+
+        # check max function for different vnode 
+
+        for colname in colnames:
+            if colname.startswith("c"):
+                self.check_avg_distribute_diff_vnode(colname)
+            else:
+                # self.check_avg_distribute_diff_vnode(colname) # bug for tag 
+                pass
+
+        
+    def distribute_agg_query(self):
+        # basic filter
+        tdSql.query(" select avg(c1) from stb1 ")
+        tdSql.checkData(0,0,14.086956522)
+
+        tdSql.query(" select avg(a) from (select avg(c1) a  from stb1 partition by tbname) ")
+        tdSql.checkData(0,0,14.292307692)
+
+        tdSql.query(" select avg(c1) from stb1 where t1=1")
+        tdSql.checkData(0,0,6.000000000)
+
+        tdSql.query("select avg(c1+c2) from stb1 where c1 =1 ")
+        tdSql.checkData(0,0,11112.000000000)
+
+        tdSql.query("select avg(c1) from stb1 where tbname=\"ct2\"")
+        tdSql.checkData(0,0,6.000000000)
+
+        tdSql.query("select avg(c1) from stb1 partition by tbname")
+        tdSql.checkRows(20)
+
+        tdSql.query("select avg(c1) from stb1 where t1> 4  partition by tbname")
+        tdSql.checkRows(15)
+
+        # union all 
+        tdSql.query("select avg(c1) from stb1 union all select avg(c1) from stb1 ")
+        tdSql.checkRows(2)
+        tdSql.checkData(0,0,14.086956522)
+
+        tdSql.query("select avg(a) from (select avg(c1) a from stb1 union all select avg(c1) a  from stb1)")
+        tdSql.checkRows(1)
+        tdSql.checkData(0,0,14.086956522)
+
+        # join 
+
+        tdSql.execute(" create database if not exists db ")
+        tdSql.execute(" use db ")
+        tdSql.execute(" create stable st (ts timestamp , c1 int ,c2 float) tags(t1 int) ")
+        tdSql.execute(" create table tb1 using st tags(1) ")
+        tdSql.execute(" create table tb2 using st tags(2) ")
+
+        
+        for i in range(10):
+            ts = i*10 + self.ts
+            tdSql.execute(f" insert into tb1 values({ts},{i},{i}.0)")
+            tdSql.execute(f" insert into tb2 values({ts},{i},{i}.0)")
+
+        tdSql.query("select avg(tb1.c1), avg(tb2.c2) from tb1, tb2 where tb1.ts=tb2.ts")
+        tdSql.checkRows(1)
+        tdSql.checkData(0,0,4.500000000)
+        tdSql.checkData(0,1,4.500000000)
+
+        # group by 
+        tdSql.execute(" use testdb ")
+
+        # partition by tbname or partition by tag
+        tdSql.query("select avg(c1) from stb1 partition by tbname")
+        tdSql.checkRows(20)
+
+        # nest query for support max
+        tdSql.query("select avg(c2+2)+1 from (select avg(c1) c2  from stb1)")
+        tdSql.checkData(0,0,17.086956522)
+        tdSql.query("select avg(c1+2)  as c2 from (select ts ,c1 ,c2  from stb1)")
+        tdSql.checkData(0,0,16.086956522)
+        tdSql.query("select avg(a+2)  as c2 from (select ts ,abs(c1) a ,c2  from stb1)")
+        tdSql.checkData(0,0,16.086956522)
+
+        # mixup with other functions
+        tdSql.query("select max(c1),count(c1),last(c2,c3),sum(c1+c2),avg(c1) from stb1")
+        tdSql.checkData(0,0,28)
+        tdSql.checkData(0,1,184)
+        tdSql.checkData(0,2,-99999)
+        tdSql.checkData(0,3,-999)
+        tdSql.checkData(0,4,28202310.000000000)
+        tdSql.checkData(0,5,14.086956522)
+
+    def run(self):
+
+        self.prepare_datas_of_distribute()
+        self.check_distribute_datas()
+        self.check_avg_status()
+        self.distribute_agg_query()
+
+    
+    def stop(self):
+        tdSql.close()
+        tdLog.success("%s successfully executed" % __file__)
+
+tdCases.addWindows(__file__, TDTestCase())
+tdCases.addLinux(__file__, TDTestCase())

tests/system-test/2-query/distribute_agg_max.py

@@ -245,6 +245,27 @@ class TDTestCase:
         tdSql.query(" select max(c1),c2  from stb1 group by c2 ")
         tdSql.checkRows(31)
 
+        # selective common cols of datas
+        tdSql.query("select max(c1),c2,c3,c5 from stb1")
+        tdSql.checkRows(1)
+        tdSql.checkData(0,0,28)
+        tdSql.checkData(0,1,311108)
+        tdSql.checkData(0,2,3108)
+        tdSql.checkData(0,3,31.08000)
+
+        tdSql.query("select max(c1),t1,c2,t3 from stb1")
+        tdSql.checkRows(1)
+        tdSql.checkData(0,0,28)
+        tdSql.checkData(0,1,19)
+        tdSql.checkData(0,2,311108)
+
+        tdSql.query("select max(c1),ceil(t1),pow(c2,1)+2,abs(t3) from stb1")
+        tdSql.checkRows(1)
+        tdSql.checkData(0,0,28)
+        tdSql.checkData(0,1,19)
+        tdSql.checkData(0,2,311110.000000000)   
+        tdSql.checkData(0,3,2109)  
+
         # partition by tbname or partition by tag
         tdSql.query("select max(c1),tbname from stb1 partition by tbname")
         query_data = tdSql.queryResult

tests/system-test/2-query/distribute_agg_min.py

@@ -245,6 +245,24 @@ class TDTestCase:
         tdSql.query(" select min(c1),c2  from stb1 group by c2 ")
         tdSql.checkRows(31)
 
+        # selective common cols of datas
+        tdSql.query("select min(c1),c2,c3,c5 from stb1")
+        tdSql.checkRows(1)
+        tdSql.checkData(0,0,0)
+        tdSql.checkData(0,1,0)
+        tdSql.checkData(0,2,0)
+        tdSql.checkData(0,3,0)
+
+        tdSql.query("select min(c1),t1,c2,t3 from stb1 where c1 >5")
+        tdSql.checkRows(1)
+        tdSql.checkData(0,0,6)
+        tdSql.checkData(0,2,66666)
+
+        tdSql.query("select min(c1),ceil(t1),pow(c2,1)+2,abs(t3) from stb1 where c1>12")
+        tdSql.checkRows(1)
+        tdSql.checkData(0,0,13)
+        tdSql.checkData(0,2,144445.000000000)   
+   
         # partition by tbname or partition by tag
         tdSql.query("select min(c1),tbname from stb1 partition by tbname")
         query_data = tdSql.queryResult

tests/system-test/2-query/distribute_agg_stddev.py

+from util.log import *
+from util.cases import *
+from util.sql import *
+import numpy as np
+import random ,os ,sys
+import platform
+import math
+
+class TDTestCase:
+    updatecfgDict = {'debugFlag': 143 ,"cDebugFlag":143,"uDebugFlag":143 ,"rpcDebugFlag":143 , "tmrDebugFlag":143 , 
+    "jniDebugFlag":143 ,"simDebugFlag":143,"dDebugFlag":143, "dDebugFlag":143,"vDebugFlag":143,"mDebugFlag":143,"qDebugFlag":143,
+    "wDebugFlag":143,"sDebugFlag":143,"tsdbDebugFlag":143,"tqDebugFlag":143 ,"fsDebugFlag":143 ,"fnDebugFlag":143,
+    "maxTablesPerVnode":2 ,"minTablesPerVnode":2,"tableIncStepPerVnode":2 }
+
+    def init(self, conn, logSql):
+        tdLog.debug("start to execute %s" % __file__)
+        tdSql.init(conn.cursor())
+        self.vnode_disbutes = None
+        self.ts = 1537146000000
+
+
+    def check_stddev_functions(self, tbname , col_name):
+
+        stddev_sql = f"select stddev({col_name}) from {tbname};"
+
+        same_sql = f"select {col_name} from {tbname} where {col_name} is not null "
+   
+        tdSql.query(same_sql)
+        pre_data = np.array(tdSql.queryResult)[np.array(tdSql.queryResult) != None]
+        if (platform.system().lower() == 'windows' and pre_data.dtype == 'int32'):
+            pre_data = np.array(pre_data, dtype = 'int64')
+        pre_avg = np.sum(pre_data)/len(pre_data)
+
+        # Calculate variance
+        stddev_result = 0 
+        for num in tdSql.queryResult:
+            stddev_result += (num-pre_avg)*(num-pre_avg)/len(tdSql.queryResult)
+
+        stddev_result = math.sqrt(stddev_result)
+
+        tdSql.query(stddev_sql)
+        
+        if -0.0001 < tdSql.queryResult[0][0]-stddev_result < 0.0001:
+            tdLog.info(" sql:%s; row:0 col:0 data:%d , expect:%d"%(stddev_sql,tdSql.queryResult[0][0],stddev_result))
+        else:
+            tdLog.exit(" sql:%s; row:0 col:0 data:%d , expect:%d"%(stddev_sql,tdSql.queryResult[0][0],stddev_result))
+
+    def prepare_datas_of_distribute(self):
+        
+        # prepate datas for  20 tables distributed at different vgroups
+        tdSql.execute("create database if not exists testdb keep 3650 duration 1000 vgroups 5")
+        tdSql.execute(" use testdb ")
+        tdSql.execute(
+            '''create table stb1
+            (ts timestamp, c1 int, c2 bigint, c3 smallint, c4 tinyint, c5 float, c6 double, c7 bool, c8 binary(16),c9 nchar(32), c10 timestamp)
+            tags (t0 timestamp, t1 int, t2 bigint, t3 smallint, t4 tinyint, t5 float, t6 double, t7 bool, t8 binary(16),t9 nchar(32))
+            '''
+        )
+
+        tdSql.execute(
+            '''
+            create table t1
+            (ts timestamp, c1 int, c2 bigint, c3 smallint, c4 tinyint, c5 float, c6 double, c7 bool, c8 binary(16),c9 nchar(32), c10 timestamp)
+            '''
+        )
+        for i in range(20):
+            tdSql.execute(f'create table ct{i+1} using stb1 tags ( now(), {1*i}, {11111*i}, {111*i}, {11*i}, {1.11*i}, {11.11*i}, {i%2}, "binary{i}", "nchar{i}" )')
+
+        for i in range(9):
+            tdSql.execute(
+                f"insert into ct1 values ( now()-{i*10}s, {1*i}, {11111*i}, {111*i}, {11*i}, {1.11*i}, {11.11*i}, {i%2}, 'binary{i}', 'nchar{i}', now()+{1*i}a )"
+            )
+            tdSql.execute(
+                f"insert into ct4 values ( now()-{i*90}d, {1*i}, {11111*i}, {111*i}, {11*i}, {1.11*i}, {11.11*i}, {i%2}, 'binary{i}', 'nchar{i}', now()+{1*i}a )"
+            )
+
+        for i in range(1,21):
+            if i ==1 or i == 4:
+                continue
+            else:
+                tbname = "ct"+f'{i}'
+                for j in range(9):
+                    tdSql.execute(
+                f"insert into {tbname} values ( now()-{(i+j)*10}s, {1*(j+i)}, {11111*(j+i)}, {111*(j+i)}, {11*(j)}, {1.11*(j+i)}, {11.11*(j+i)}, {(j+i)%2}, 'binary{j}', 'nchar{j}', now()+{1*j}a )"
+            )
+        tdSql.execute("insert into ct1 values (now()-45s, 0, 0, 0, 0, 0, 0, 0, 'binary0', 'nchar0', now()+8a )")
+        tdSql.execute("insert into ct1 values (now()+10s, 9, -99999, -999, -99, -9.99, -99.99, 1, 'binary9', 'nchar9', now()+9a )")
+        tdSql.execute("insert into ct1 values (now()+15s, 9, -99999, -999, -99, -9.99, NULL, 1, 'binary9', 'nchar9', now()+9a )")
+        tdSql.execute("insert into ct1 values (now()+20s, 9, -99999, -999, NULL, -9.99, -99.99, 1, 'binary9', 'nchar9', now()+9a )")
+
+        tdSql.execute("insert into ct4 values (now()-810d, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL ) ")
+        tdSql.execute("insert into ct4 values (now()-400d, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL ) ")
+        tdSql.execute("insert into ct4 values (now()+90d, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL  ) ")
+
+        tdSql.execute(
+            f'''insert into t1 values
+            ( '2020-04-21 01:01:01.000', NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL )
+            ( '2020-10-21 01:01:01.000', 1, 11111, 111, 11, 1.11, 11.11, 1, "binary1", "nchar1", now()+1a )
+            ( '2020-12-31 01:01:01.000', 2, 22222, 222, 22, 2.22, 22.22, 0, "binary2", "nchar2", now()+2a )
+            ( '2021-01-01 01:01:06.000', 3, 33333, 333, 33, 3.33, 33.33, 0, "binary3", "nchar3", now()+3a )
+            ( '2021-05-07 01:01:10.000', 4, 44444, 444, 44, 4.44, 44.44, 1, "binary4", "nchar4", now()+4a )
+            ( '2021-07-21 01:01:01.000', NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL )
+            ( '2021-09-30 01:01:16.000', 5, 55555, 555, 55, 5.55, 55.55, 0, "binary5", "nchar5", now()+5a )
+            ( '2022-02-01 01:01:20.000', 6, 66666, 666, 66, 6.66, 66.66, 1, "binary6", "nchar6", now()+6a )
+            ( '2022-10-28 01:01:26.000', 7, 00000, 000, 00, 0.00, 00.00, 1, "binary7", "nchar7", "1970-01-01 08:00:00.000" )
+            ( '2022-12-01 01:01:30.000', 8, -88888, -888, -88, -8.88, -88.88, 0, "binary8", "nchar8", "1969-01-01 01:00:00.000" )
+            ( '2022-12-31 01:01:36.000', 9, -99999999999999999, -999, -99, -9.99, -999999999999999999999.99, 1, "binary9", "nchar9", "1900-01-01 00:00:00.000" )
+            ( '2023-02-21 01:01:01.000', NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL )
+            '''
+        )
+
+        tdLog.info(" prepare data for distributed_aggregate done! ")
+
+    def check_distribute_datas(self):
+        # get vgroup_ids of all
+        tdSql.query("show vgroups ")
+        vgroups = tdSql.queryResult
+
+        vnode_tables={}
+        
+        for vgroup_id in vgroups:
+            vnode_tables[vgroup_id[0]]=[]
+        
+
+        # check sub_table of per vnode ,make sure sub_table has been distributed
+        tdSql.query("show tables like 'ct%'")
+        table_names = tdSql.queryResult
+        tablenames = []
+        for table_name in table_names:
+            vnode_tables[table_name[6]].append(table_name[0]) 
+        self.vnode_disbutes = vnode_tables
+
+        count = 0
+        for k ,v in vnode_tables.items():
+            if len(v)>=2:
+                count+=1
+        if count < 2:
+            tdLog.exit(" the datas of all not satisfy sub_table has been distributed ")
+
+    def check_stddev_distribute_diff_vnode(self,col_name):
+    
+        vgroup_ids = []
+        for k ,v in self.vnode_disbutes.items():
+            if len(v)>=2:
+                vgroup_ids.append(k)
+        
+        distribute_tbnames = []
+        
+        for vgroup_id in vgroup_ids:
+            vnode_tables = self.vnode_disbutes[vgroup_id]
+            distribute_tbnames.append(random.sample(vnode_tables,1)[0])
+        tbname_ins = ""
+        for tbname in distribute_tbnames:
+            tbname_ins += "'%s' ,"%tbname
+
+        tbname_filters = tbname_ins[:-1]
+        
+        stddev_sql = f"select stddev({col_name}) from stb1 where tbname in ({tbname_filters});"
+
+        same_sql = f"select {col_name}  from stb1 where tbname in ({tbname_filters}) and {col_name} is not null "
+
+        tdSql.query(same_sql)
+        pre_data = np.array(tdSql.queryResult)[np.array(tdSql.queryResult) != None]
+        if (platform.system().lower() == 'windows' and pre_data.dtype == 'int32'):
+            pre_data = np.array(pre_data, dtype = 'int64')
+        pre_avg = np.sum(pre_data)/len(pre_data)
+
+        # Calculate variance
+        stddev_result = 0 
+        for num in tdSql.queryResult:
+            stddev_result += (num-pre_avg)*(num-pre_avg)/len(tdSql.queryResult)
+
+        stddev_result = math.sqrt(stddev_result)
+
+        tdSql.query(stddev_sql)
+        tdSql.checkData(0,0,stddev_result)
+
+
+    def check_stddev_status(self):
+        # check max function work status 
+        
+        tdSql.query("show tables like 'ct%'")
+        table_names = tdSql.queryResult
+        tablenames = []
+        for table_name in table_names:
+            tablenames.append(table_name[0])
+
+        tdSql.query("desc stb1")
+        col_names = tdSql.queryResult
+        
+        colnames = []
+        for col_name in col_names:
+            if col_name[1] in ["INT" ,"BIGINT" ,"SMALLINT" ,"TINYINT" , "FLOAT" ,"DOUBLE"]:
+                colnames.append(col_name[0])
+        
+        for tablename in tablenames:
+            for colname in colnames:
+                if colname.startswith("c"):
+                    self.check_stddev_functions(tablename,colname)
+                else:
+                    # self.check_stddev_functions(tablename,colname) 
+                    pass
+
+
+        # check max function for different vnode 
+
+        for colname in colnames:
+            if colname.startswith("c"):
+                self.check_stddev_distribute_diff_vnode(colname)
+            else:
+                # self.check_stddev_distribute_diff_vnode(colname) # bug for tag 
+                pass
+
+        
+    def distribute_agg_query(self):
+        # basic filter
+        tdSql.query(" select stddev(c1) from stb1 ")
+        tdSql.checkData(0,0,6.694663959)
+
+        tdSql.query(" select stddev(a) from (select stddev(c1) a  from stb1 partition by tbname) ")
+        tdSql.checkData(0,0,0.156797505)
+
+        tdSql.query(" select stddev(c1) from stb1 where t1=1")
+        tdSql.checkData(0,0,2.581988897)
+
+        tdSql.query("select stddev(c1+c2) from stb1 where c1 =1 ")
+        tdSql.checkData(0,0,0.000000000)
+
+        tdSql.query("select stddev(c1) from stb1 where tbname=\"ct2\"")
+        tdSql.checkData(0,0,2.581988897)
+
+        tdSql.query("select stddev(c1) from stb1 partition by tbname")
+        tdSql.checkRows(20)
+
+        tdSql.query("select stddev(c1) from stb1 where t1> 4  partition by tbname")
+        tdSql.checkRows(15)
+
+        # union all 
+        tdSql.query("select stddev(c1) from stb1 union all select stddev(c1) from stb1 ")
+        tdSql.checkRows(2)
+        tdSql.checkData(0,0,6.694663959)
+
+        tdSql.query("select stddev(a) from (select stddev(c1) a from stb1 union all select stddev(c1) a  from stb1)")
+        tdSql.checkRows(1)
+        tdSql.checkData(0,0,0.000000000)
+
+        # join 
+
+        tdSql.execute(" create database if not exists db ")
+        tdSql.execute(" use db ")
+        tdSql.execute(" create stable st (ts timestamp , c1 int ,c2 float) tags(t1 int) ")
+        tdSql.execute(" create table tb1 using st tags(1) ")
+        tdSql.execute(" create table tb2 using st tags(2) ")
+
+        
+        for i in range(10):
+            ts = i*10 + self.ts
+            tdSql.execute(f" insert into tb1 values({ts},{i},{i}.0)")
+            tdSql.execute(f" insert into tb2 values({ts},{i},{i}.0)")
+
+        tdSql.query("select stddev(tb1.c1), stddev(tb2.c2) from tb1, tb2 where tb1.ts=tb2.ts")
+        tdSql.checkRows(1)
+        tdSql.checkData(0,0,2.872281323)
+        tdSql.checkData(0,1,2.872281323)
+
+        # group by 
+        tdSql.execute(" use testdb ")
+
+        # partition by tbname or partition by tag
+        tdSql.query("select stddev(c1) from stb1 partition by tbname")
+        tdSql.checkRows(20)
+
+        # nest query for support max
+        tdSql.query("select stddev(c2+2)+1 from (select stddev(c1) c2  from stb1)")
+        tdSql.checkData(0,0,1.000000000)
+        tdSql.query("select stddev(c1+2)  as c2 from (select ts ,c1 ,c2  from stb1)")
+        tdSql.checkData(0,0,6.694663959)
+        tdSql.query("select stddev(a+2)  as c2 from (select ts ,abs(c1) a ,c2  from stb1)")
+        tdSql.checkData(0,0,6.694663959)
+
+        # mixup with other functions
+        tdSql.query("select max(c1),count(c1),last(c2,c3),sum(c1+c2),avg(c1),stddev(c1) from stb1")
+        tdSql.checkData(0,0,28)
+        tdSql.checkData(0,1,184)
+        tdSql.checkData(0,2,-99999)
+        tdSql.checkData(0,3,-999)
+        tdSql.checkData(0,4,28202310.000000000)
+        tdSql.checkData(0,5,14.086956522)
+        tdSql.checkData(0,6,6.694663959)
+
+    def run(self):
+
+        self.prepare_datas_of_distribute()
+        self.check_distribute_datas()
+        self.check_stddev_status()
+        self.distribute_agg_query()
+
+    
+    def stop(self):
+        tdSql.close()
+        tdLog.success("%s successfully executed" % __file__)
+
+tdCases.addWindows(__file__, TDTestCase())
+tdCases.addLinux(__file__, TDTestCase())

tests/system-test/2-query/join.py

@@ -238,7 +238,7 @@ class TDTestCase:
                 ts timestamp, {INT_COL} int, {BINT_COL} bigint, {SINT_COL} smallint, {TINT_COL} tinyint,
                  {FLOAT_COL} float, {DOUBLE_COL} double, {BOOL_COL} bool,
                  {BINARY_COL} binary(16), {NCHAR_COL} nchar(32), {TS_COL} timestamp
-            ) tags (t1 int)
+            ) tags (tag1 int)
             '''
         create_ntb_sql = f'''create table t1(
                 ts timestamp, {INT_COL} int, {BINT_COL} bigint, {SINT_COL} smallint, {TINT_COL} tinyint,

tests/system-test/2-query/substr.py

@@ -23,6 +23,9 @@ CHAR_COL    = [ BINARY_COL, NCHAR_COL, ]
 BOOLEAN_COL = [ BOOL_COL, ]
 TS_TYPE_COL = [ TS_COL, ]
 
+ERR_POS     = 0
+CURRENT_POS = 1
+LENS        = 6
 
 class TDTestCase:
 
@@ -76,6 +79,7 @@ class TDTestCase:
 
             if  pos < 1:
                 tdSql.error(f"select substr( {condition}, {pos}, {lens}) , {condition} from {tbname} ")
+                break
 
             tdSql.query(f"select substr( {condition}, {pos}, {lens}) , {condition} from {tbname} ")
             for j in range(tdSql.queryRows):
@@ -127,7 +131,7 @@ class TDTestCase:
         tdLog.printNoPrefix("==========current sql condition check , must return query ok==========")
         tbname = ["ct1", "ct2", "ct4", "t1", "stb1"]
         for tb in tbname:
-            self.__substr_check(tb, 1, 6)
+            self.__substr_check(tb, CURRENT_POS, LENS)
             tdLog.printNoPrefix(f"==========current sql condition check in {tb} over==========")
 
     def __test_error(self):
@@ -137,7 +141,7 @@ class TDTestCase:
         for tb in tbname:
             for errsql in self.__substr_err_check(tb):
                 tdSql.error(sql=errsql)
-            self.__substr_check(tb, 0, 6)
+            self.__substr_check(tb, ERR_POS, LENS)
             tdLog.printNoPrefix(f"==========err sql condition check in {tb} over==========")
 
 
@@ -154,7 +158,7 @@ class TDTestCase:
                 ts timestamp, {INT_COL} int, {BINT_COL} bigint, {SINT_COL} smallint, {TINT_COL} tinyint,
                  {FLOAT_COL} float, {DOUBLE_COL} double, {BOOL_COL} bool,
                  {BINARY_COL} binary(16), {NCHAR_COL} nchar(32), {TS_COL} timestamp
-            ) tags (t1 int)
+            ) tags (tag1 int)
             '''
         create_ntb_sql = f'''create table t1(
                 ts timestamp, {INT_COL} int, {BINT_COL} bigint, {SINT_COL} smallint, {TINT_COL} tinyint,

tests/system-test/2-query/sum.py

@@ -73,7 +73,7 @@ class TDTestCase:
                     f"select sum(ceil( {un_num_col} )) from {tbanme} ",
                 )
             )
-            sqls.extend( f"select sum( {un_num_col} + {un_num_col_2} ) from {tbanme} " for un_num_col_2 in UN_NUM_COL )
+            # sqls.extend( f"select sum( {un_num_col} + {un_num_col_2} ) from {tbanme} " for un_num_col_2 in UN_NUM_COL )
 
         sqls.extend( f"select sum( {num_col} + {ts_col} ) from {tbanme} " for num_col in NUM_COL for ts_col in TS_TYPE_COL)
         sqls.extend(

tests/system-test/fulltest.sh

@@ -35,6 +35,7 @@ python3 ./test.py -f 2-query/lower.py
 python3 ./test.py -f 2-query/join.py
 python3 ./test.py -f 2-query/join2.py
 python3 ./test.py -f 2-query/cast.py
+python3 ./test.py -f 2-query/substr.py
 python3 ./test.py -f 2-query/union.py
 python3 ./test.py -f 2-query/union1.py
 python3 ./test.py -f 2-query/concat.py
@@ -105,6 +106,8 @@ python3 ./test.py -f 2-query/distribute_agg_min.py
 python3 ./test.py -f 2-query/distribute_agg_sum.py
 python3 ./test.py -f 2-query/distribute_agg_spread.py
 python3 ./test.py -f 2-query/distribute_agg_apercentile.py
+python3 ./test.py -f 2-query/distribute_agg_avg.py
+python3 ./test.py -f 2-query/distribute_agg_stddev.py
 
 python3 ./test.py -f 6-cluster/5dnode1mnode.py
 python3 ./test.py -f 6-cluster/5dnode2mnode.py

taosadapter @ c1dc11f3

-Subproject commit 29926478edd87533a043f91c1a9ed0e27671e626
+Subproject commit c1dc11f3cae64adf31dbd9a954ef8372d1e8f671