test: add develop=test/5-taos-tools/taosbenchmark/telnet_tcp.py

Jenkinsfile2

@@ -430,8 +430,6 @@ pipeline {
                                         rm -rf ${WKC}/debug
                                         cd ${WKC}/tests/parallel_test
                                         time ./container_build.sh -w ${WKDIR} -t 10 -e
-                                        rm -f /tmp/cases.task
-                                        ./collect_cases.sh -e
                                     '''
                                     def extra_param = ""
                                     def log_server_file = "/home/log_server.json"
@@ -462,7 +460,7 @@ pipeline {
                                         cd ${WKC}/tests/parallel_test
                                         export DEFAULT_RETRY_TIME=2
                                         date
-                                        ''' + timeout_cmd + ''' time ./run.sh -e -m /home/m.json -t /tmp/cases.task -b ${BRANCH_NAME}_${BUILD_ID} -l ${WKDIR}/log -o 480 ''' + extra_param + '''
+                                        ''' + timeout_cmd + ''' time ./run.sh -e -m /home/m.json -t cases.task -b ${BRANCH_NAME}_${BUILD_ID} -l ${WKDIR}/log -o 480 ''' + extra_param + '''
                                     '''
                                 }
                             }

cmake/cmake.platform

@@ -102,6 +102,12 @@ IF ("${CPUTYPE}" STREQUAL "")
     SET(TD_ARM_64 TRUE)
     ADD_DEFINITIONS("-D_TD_ARM_")
     ADD_DEFINITIONS("-D_TD_ARM_64")
+  ELSEIF (CMAKE_SYSTEM_PROCESSOR MATCHES "loongarch64")
+    MESSAGE(STATUS "The current platform is loongarch64")
+    SET(PLATFORM_ARCH_STR "loongarch64")
+    SET(TD_LOONGARCH_64 TRUE)
+    ADD_DEFINITIONS("-D_TD_LOONGARCH_")
+    ADD_DEFINITIONS("-D_TD_LOONGARCH_64")
   ENDIF ()
 ELSE ()
   # if generate ARM version:
@@ -118,6 +124,12 @@ ELSE ()
     ADD_DEFINITIONS("-D_TD_ARM_")
     ADD_DEFINITIONS("-D_TD_ARM_64")
     SET(TD_ARM_64 TRUE)
+  ELSEIF (${CPUTYPE} MATCHES "loongarch64")
+    SET(PLATFORM_ARCH_STR "loongarch64")
+    MESSAGE(STATUS "input cpuType: loongarch64")
+    SET(TD_LOONGARCH_64 TRUE)
+    ADD_DEFINITIONS("-D_TD_LOONGARCH_")
+    ADD_DEFINITIONS("-D_TD_LOONGARCH_64")
   ELSEIF (${CPUTYPE} MATCHES "mips64")
     SET(PLATFORM_ARCH_STR "mips")
     MESSAGE(STATUS "input cpuType: mips64")

cmake/cmake.version

@@ -65,13 +65,14 @@ ELSE ()
 ENDIF ()
 
 MESSAGE(STATUS "============= compile version parameter information start ============= ")
-MESSAGE(STATUS "ver number:" ${TD_VER_NUMBER})
+MESSAGE(STATUS "version:    " ${TD_VER_NUMBER})
-MESSAGE(STATUS "compatible ver number:" ${TD_VER_COMPATIBLE})
+MESSAGE(STATUS "compatible: " ${TD_VER_COMPATIBLE})
-MESSAGE(STATUS "communit commit id:" ${TD_VER_GIT})
+MESSAGE(STATUS "commit id:  " ${TD_VER_GIT})
-MESSAGE(STATUS "build date:" ${TD_VER_DATE})
+MESSAGE(STATUS "build date: " ${TD_VER_DATE})
-MESSAGE(STATUS "ver type:" ${TD_VER_VERTYPE})
+MESSAGE(STATUS "build type: " ${CMAKE_BUILD_TYPE})
-MESSAGE(STATUS "ver cpu:" ${TD_VER_CPUTYPE})
+MESSAGE(STATUS "type: " ${TD_VER_VERTYPE})
-MESSAGE(STATUS "os type:" ${TD_VER_OSTYPE})
+MESSAGE(STATUS "cpu:  " ${TD_VER_CPUTYPE})
+MESSAGE(STATUS "os:   " ${TD_VER_OSTYPE})
 MESSAGE(STATUS "============= compile version parameter information end  ============= ")
 
 STRING(REPLACE "." "_" TD_LIB_VER_NUMBER ${TD_VER_NUMBER})

cmake/taostools_CMakeLists.txt.in

@@ -2,7 +2,7 @@
 # taos-tools
 ExternalProject_Add(taos-tools
         GIT_REPOSITORY https://github.com/taosdata/taos-tools.git
-        GIT_TAG a921bd4
+        GIT_TAG 23e2b73
         SOURCE_DIR "${TD_SOURCE_DIR}/tools/taos-tools"
         BINARY_DIR ""
         #BUILD_IN_SOURCE TRUE

docs/en/14-reference/09-support-platform/index.md

@@ -14,7 +14,7 @@ Note: ● means officially tested and verified, ○ means unofficially tested an
 
 ## List of supported platforms for TDengine clients and connectors
 
-TDengine's connector can support a wide range of platforms, including X64/X86/ARM64/ARM32/MIPS/Alpha hardware platforms and Linux/Win64/Win32/macOS development environments.
+TDengine's connector can support a wide range of platforms, including X64/X86/ARM64/ARM32/MIPS/Alpha/LoongArch64 hardware platforms and Linux/Win64/Win32/macOS development environments.
 
 The comparison matrix is as follows.
 

docs/zh/05-get-started/index.md

@@ -3,6 +3,8 @@ title: 立即开始
 description: '快速设置 TDengine 环境并体验其高效写入和查询'
 ---
 
+import xiaot from './tdengine.webp'
+
 TDengine 完整的软件包包括服务端（taosd）、用于与第三方系统对接并提供 RESTful 接口的 taosAdapter、应用驱动（taosc）、命令行程序 (CLI，taos) 和一些工具软件。TDengine 除了提供多种语言的连接器之外，还通过 [taosAdapter](../reference/taosadapter) 提供 [RESTful 接口](../connector/rest-api)。
 
 本章主要介绍如何利用 Docker 或者安装包快速设置 TDengine 环境并体验其高效写入和查询。
@@ -12,4 +14,10 @@ import DocCardList from '@theme/DocCardList';
 import {useCurrentSidebarCategory} from '@docusaurus/theme-common';
 
 <DocCardList items={useCurrentSidebarCategory().items}/>
-```
+```
+
+### 开发者技术交流群
+
+微信扫描下面二维码，加“小 T”为好友，即可加入“物联网大数据技术前沿群”，与大家共同交流物联网大数据技术应用、TDengine 使用问题和技巧等话题。
+
+<img src={xiaot} alt="小 T 的二维码" width="200" />

docs/zh/08-connector/10-cpp.mdx

@@ -115,6 +115,7 @@ TDengine 客户端驱动的安装请参考 [安装指南](../#安装步骤)
 <summary>订阅和消费</summary>
 
 ```c
+  {{#include examples/c/tmq.c}}
 ```
 
 </details>

docs/zh/14-reference/09-support-platform/index.md

@@ -16,7 +16,7 @@ description: "TDengine 服务端、客户端和连接器支持的平台列表"
 
 ## TDengine 客户端和连接器支持的平台列表
 
-目前 TDengine 的连接器可支持的平台广泛，目前包括：X64/X86/ARM64/ARM32/MIPS/Alpha 等硬件平台，以及 Linux/Win64/Win32/macOS 等开发环境。
+目前 TDengine 的连接器可支持的平台广泛，目前包括：X64/X86/ARM64/ARM32/MIPS/LoongArch64 等硬件平台，以及 Linux/Win64/Win32/macOS 等开发环境。
 
 对照矩阵如下：
 

docs/zh/14-reference/11-docker/index.md

@@ -119,7 +119,7 @@ taos -h tdengine -P 6030
 FROM ubuntu:20.04
 RUN apt-get update && apt-get install -y wget
 ENV TDENGINE_VERSION=3.0.0.0
-RUN wget -c https://www.tdengine.com/assets-download/TDengine-client-${TDENGINE_VERSION}-Linux-x64.tar.gz \
+RUN wget -c https://www.tdengine.com/assets-download/3.0/TDengine-client-${TDENGINE_VERSION}-Linux-x64.tar.gz \
    && tar xvf TDengine-client-${TDENGINE_VERSION}-Linux-x64.tar.gz \
    && cd TDengine-client-${TDENGINE_VERSION} \
    && ./install_client.sh \

docs/zh/28-releases/01-tdengine.md

@@ -1,9 +1,11 @@
 ---
 sidebar_label: TDengine 发布历史
-title: TDengine 发布历史
+title: TDengine 发布历史及下载链接
 description: TDengine 发布历史、Release Notes 及下载链接
 ---
 
+各版本 TDengine 安装包下载链接如下：
+
 import Release from "/components/ReleaseV3";
 
 ## 3.0.1.6
@@ -33,4 +35,3 @@ import Release from "/components/ReleaseV3";
 ## 3.0.1.0
 
 <Release type="tdengine" version="3.0.1.0" />
-

docs/zh/28-releases/02-tools.md

@@ -1,9 +1,11 @@
 ---
 sidebar_label: taosTools 发布历史
-title: taosTools 发布历史
+title: taosTools 发布历史及下载链接
 description: taosTools 的发布历史、Release Notes 和下载链接
 ---
 
+各版本 taosTools 安装包下载链接如下：
+
 import Release from "/components/ReleaseV3";
 
 ## 2.2.7

include/common/tdatablock.h

@@ -244,7 +244,7 @@ int32_t      blockDataAppendColInfo(SSDataBlock* pBlock, SColumnInfoData* pColIn
 SColumnInfoData  createColumnInfoData(int16_t type, int32_t bytes, int16_t colId);
 SColumnInfoData* bdGetColumnInfoData(const SSDataBlock* pBlock, int32_t index);
 
-void blockEncode(const SSDataBlock* pBlock, char* data, int32_t* dataLen, int32_t numOfCols, int8_t needCompress);
+int32_t blockEncode(const SSDataBlock* pBlock, char* data, int32_t numOfCols);
 const char* blockDecode(SSDataBlock* pBlock, const char* pData);
 
 void blockDebugShowDataBlock(SSDataBlock* pBlock, const char* flag);

include/common/tdataformat.h

@@ -130,7 +130,7 @@ void    tColDataInit(SColData *pColData, int16_t cid, int8_t type, int8_t smaOn)
 void    tColDataClear(SColData *pColData);
 int32_t tColDataAppendValue(SColData *pColData, SColVal *pColVal);
 void    tColDataGetValue(SColData *pColData, int32_t iVal, SColVal *pColVal);
-uint8_t tColDataGetBitValue(SColData *pColData, int32_t iVal);
+uint8_t tColDataGetBitValue(const SColData *pColData, int32_t iVal);
 int32_t tColDataCopy(SColData *pColDataSrc, SColData *pColDataDest);
 extern void (*tColDataCalcSMA[])(SColData *pColData, int64_t *sum, int64_t *max, int64_t *min, int16_t *numOfNull);
 

include/common/tglobal.h

@@ -135,6 +135,9 @@ extern int32_t tsTtlPushInterval;
 extern int32_t tsGrantHBInterval;
 extern int32_t tsUptimeInterval;
 
+extern int32_t tsRpcRetryLimit;
+extern int32_t tsRpcRetryInterval;
+
 //#define NEEDTO_COMPRESSS_MSG(size) (tsCompressMsgSize != -1 && (size) > tsCompressMsgSize)
 
 int32_t taosCreateLog(const char *logname, int32_t logFileNum, const char *cfgDir, const char **envCmd,

include/common/tmsg.h

@@ -190,6 +190,7 @@ typedef struct {
   int64_t dbId;
   int32_t vgVersion;
   int32_t numOfTable;  // unit is TSDB_TABLE_NUM_UNIT
+  int64_t stateTs;
 } SBuildUseDBInput;
 
 typedef struct SField {
@@ -417,13 +418,17 @@ static FORCE_INLINE int32_t taosEncodeSSchemaWrapper(void** buf, const SSchemaWr
 static FORCE_INLINE void* taosDecodeSSchemaWrapper(const void* buf, SSchemaWrapper* pSW) {
   buf = taosDecodeVariantI32(buf, &pSW->nCols);
   buf = taosDecodeVariantI32(buf, &pSW->version);
-  pSW->pSchema = (SSchema*)taosMemoryCalloc(pSW->nCols, sizeof(SSchema));
+  if (pSW->nCols > 0) {
-  if (pSW->pSchema == NULL) {
+    pSW->pSchema = (SSchema*)taosMemoryCalloc(pSW->nCols, sizeof(SSchema));
-    return NULL;
+    if (pSW->pSchema == NULL) {
-  }
+      return NULL;
+    }
 
-  for (int32_t i = 0; i < pSW->nCols; i++) {
+    for (int32_t i = 0; i < pSW->nCols; i++) {
-    buf = taosDecodeSSchema(buf, &pSW->pSchema[i]);
+      buf = taosDecodeSSchema(buf, &pSW->pSchema[i]);
+    }
+  } else {
+    pSW->pSchema = NULL;
   }
   return (void*)buf;
 }
@@ -838,7 +843,7 @@ typedef struct {
   int64_t dbId;
   int32_t vgVersion;
   int32_t numOfTable;  // unit is TSDB_TABLE_NUM_UNIT
-  int64_t stateTs;      // ms
+  int64_t stateTs;     // ms
 } SUseDbReq;
 
 int32_t tSerializeSUseDbReq(void* buf, int32_t bufLen, SUseDbReq* pReq);
@@ -1805,7 +1810,7 @@ int32_t tDeserializeSCMCreateTopicRsp(void* buf, int32_t bufLen, SCMCreateTopicR
 
 typedef struct {
   int64_t consumerId;
-} SMqConsumerLostMsg, SMqConsumerRecoverMsg;
+} SMqConsumerLostMsg, SMqConsumerRecoverMsg, SMqConsumerClearMsg;
 
 typedef struct {
   int64_t consumerId;
@@ -2989,7 +2994,8 @@ static FORCE_INLINE void* tDecodeSMqSubTopicEp(void* buf, SMqSubTopicEp* pTopicE
 }
 
 static FORCE_INLINE void tDeleteSMqSubTopicEp(SMqSubTopicEp* pSubTopicEp) {
-  if (pSubTopicEp->schema.nCols) taosMemoryFreeClear(pSubTopicEp->schema.pSchema);
+  taosMemoryFreeClear(pSubTopicEp->schema.pSchema);
+  pSubTopicEp->schema.nCols = 0;
   taosArrayDestroy(pSubTopicEp->vgs);
 }
 

include/common/tmsgdef.h

@@ -149,7 +149,7 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_MND_TMQ_DO_REBALANCE, "do-rebalance", SMqDoRebalanceMsg, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_TMQ_DROP_CGROUP, "drop-cgroup", SMqDropCGroupReq, SMqDropCGroupRsp)
   TD_DEF_MSG_TYPE(TDMT_MND_UNUSED2, "unused2", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_MND_TMQ_TIMER, "mq-tmr", SMTimerReq, NULL)
+  TD_DEF_MSG_TYPE(TDMT_MND_TMQ_TIMER, "tmq-tmr", SMTimerReq, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_TELEM_TIMER, "telem-tmr", SMTimerReq, SMTimerReq)
   TD_DEF_MSG_TYPE(TDMT_MND_TRANS_TIMER, "trans-tmr", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_TTL_TIMER, "ttl-tmr", NULL, NULL)
@@ -171,6 +171,7 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_MND_SHOW_VARIABLES, "show-variables", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_SERVER_VERSION, "server-version", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_UPTIME_TIMER, "uptime-timer", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_MND_TMQ_LOST_CONSUMER_CLEAR, "lost-consumer-clear", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_MAX_MSG, "mnd-max", NULL, NULL)
 
   TD_NEW_MSG_SEG(TDMT_VND_MSG)
@@ -249,8 +250,8 @@ enum {
 
   TD_NEW_MSG_SEG(TDMT_SYNC_MSG)
   TD_DEF_MSG_TYPE(TDMT_SYNC_TIMEOUT, "sync-timer", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_SYNC_PING, "sync-ping", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_SYNC_PING, "sync-ping", NULL, NULL)             // no longer used
-  TD_DEF_MSG_TYPE(TDMT_SYNC_PING_REPLY, "sync-ping-reply", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_SYNC_PING_REPLY, "sync-ping-reply", NULL, NULL) // no longer used
   TD_DEF_MSG_TYPE(TDMT_SYNC_CLIENT_REQUEST, "sync-client-request", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_SYNC_CLIENT_REQUEST_BATCH, "sync-client-request-batch", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_SYNC_CLIENT_REQUEST_REPLY, "sync-client-request-reply", NULL, NULL)

include/common/ttokendef.h

@@ -174,61 +174,61 @@
 #define TK_SUBSCRIPTIONS        156
 #define TK_VNODES               157
 #define TK_LIKE                 158
-#define TK_INDEX                159
+#define TK_TBNAME               159
-#define TK_FUNCTION             160
+#define TK_QTAGS                160
-#define TK_INTERVAL             161
+#define TK_AS                   161
-#define TK_TOPIC                162
+#define TK_INDEX                162
-#define TK_AS                   163
+#define TK_FUNCTION             163
-#define TK_WITH                 164
+#define TK_INTERVAL             164
-#define TK_META                 165
+#define TK_TOPIC                165
-#define TK_CONSUMER             166
+#define TK_WITH                 166
-#define TK_GROUP                167
+#define TK_META                 167
-#define TK_DESC                 168
+#define TK_CONSUMER             168
-#define TK_DESCRIBE             169
+#define TK_GROUP                169
-#define TK_RESET                170
+#define TK_DESC                 170
-#define TK_QUERY                171
+#define TK_DESCRIBE             171
-#define TK_CACHE                172
+#define TK_RESET                172
-#define TK_EXPLAIN              173
+#define TK_QUERY                173
-#define TK_ANALYZE              174
+#define TK_CACHE                174
-#define TK_VERBOSE              175
+#define TK_EXPLAIN              175
-#define TK_NK_BOOL              176
+#define TK_ANALYZE              176
-#define TK_RATIO                177
+#define TK_VERBOSE              177
-#define TK_NK_FLOAT             178
+#define TK_NK_BOOL              178
-#define TK_OUTPUTTYPE           179
+#define TK_RATIO                179
-#define TK_AGGREGATE            180
+#define TK_NK_FLOAT             180
-#define TK_BUFSIZE              181
+#define TK_OUTPUTTYPE           181
-#define TK_STREAM               182
+#define TK_AGGREGATE            182
-#define TK_INTO                 183
+#define TK_BUFSIZE              183
-#define TK_TRIGGER              184
+#define TK_STREAM               184
-#define TK_AT_ONCE              185
+#define TK_INTO                 185
-#define TK_WINDOW_CLOSE         186
+#define TK_TRIGGER              186
-#define TK_IGNORE               187
+#define TK_AT_ONCE              187
-#define TK_EXPIRED              188
+#define TK_WINDOW_CLOSE         188
-#define TK_FILL_HISTORY         189
+#define TK_IGNORE               189
-#define TK_SUBTABLE             190
+#define TK_EXPIRED              190
-#define TK_KILL                 191
+#define TK_FILL_HISTORY         191
-#define TK_CONNECTION           192
+#define TK_SUBTABLE             192
-#define TK_TRANSACTION          193
+#define TK_KILL                 193
-#define TK_BALANCE              194
+#define TK_CONNECTION           194
-#define TK_VGROUP               195
+#define TK_TRANSACTION          195
-#define TK_MERGE                196
+#define TK_BALANCE              196
-#define TK_REDISTRIBUTE         197
+#define TK_VGROUP               197
-#define TK_SPLIT                198
+#define TK_MERGE                198
-#define TK_DELETE               199
+#define TK_REDISTRIBUTE         199
-#define TK_INSERT               200
+#define TK_SPLIT                200
-#define TK_NULL                 201
+#define TK_DELETE               201
-#define TK_NK_QUESTION          202
+#define TK_INSERT               202
-#define TK_NK_ARROW             203
+#define TK_NULL                 203
-#define TK_ROWTS                204
+#define TK_NK_QUESTION          204
-#define TK_TBNAME               205
+#define TK_NK_ARROW             205
-#define TK_QSTART               206
+#define TK_ROWTS                206
-#define TK_QEND                 207
+#define TK_QSTART               207
-#define TK_QDURATION            208
+#define TK_QEND                 208
-#define TK_WSTART               209
+#define TK_QDURATION            209
-#define TK_WEND                 210
+#define TK_WSTART               210
-#define TK_WDURATION            211
+#define TK_WEND                 211
-#define TK_IROWTS               212
+#define TK_WDURATION            212
-#define TK_QTAGS                213
+#define TK_IROWTS               213
 #define TK_CAST                 214
 #define TK_NOW                  215
 #define TK_TODAY                216

include/libs/catalog/catalog.h

@@ -56,6 +56,7 @@ typedef struct SDbInfo {
   int32_t vgVer;
   int32_t tbNum;
   int64_t dbId;
+  int64_t stateTs;
 } SDbInfo;
 
 typedef struct STablesReq {
@@ -153,7 +154,7 @@ int32_t catalogInit(SCatalogCfg* cfg);
  */
 int32_t catalogGetHandle(uint64_t clusterId, SCatalog** catalogHandle);
 
-int32_t catalogGetDBVgVersion(SCatalog* pCtg, const char* dbFName, int32_t* version, int64_t* dbId, int32_t* tableNum);
+int32_t catalogGetDBVgVersion(SCatalog* pCtg, const char* dbFName, int32_t* version, int64_t* dbId, int32_t* tableNum, int64_t* stateTs);
 
 /**
  * Get a DB's all vgroup info.

include/libs/nodes/cmdnodes.h

@@ -284,6 +284,13 @@ typedef struct SShowVnodesStmt {
   SNode*    pDnodeEndpoint;
 } SShowVnodesStmt;
 
+typedef struct SShowTableTagsStmt {
+  ENodeType  type;
+  SNode*     pDbName;  // SValueNode
+  SNode*     pTbName;  // SValueNode
+  SNodeList* pTags;
+} SShowTableTagsStmt;
+
 typedef enum EIndexType { INDEX_TYPE_SMA = 1, INDEX_TYPE_FULLTEXT } EIndexType;
 
 typedef struct SIndexOptions {

include/libs/nodes/querynodes.h

@@ -297,6 +297,7 @@ typedef struct SSelectStmt {
   bool        hasStateKey;
   bool        onlyHasKeepOrderFunc;
   bool        groupSort;
+  bool        tagScan;
 } SSelectStmt;
 
 typedef enum ESetOperatorType { SET_OP_TYPE_UNION_ALL = 1, SET_OP_TYPE_UNION } ESetOperatorType;

include/libs/qcom/query.h

@@ -127,6 +127,7 @@ typedef struct SDBVgInfo {
   int16_t   hashSuffix;
   int8_t    hashMethod;
   int32_t   numOfTable;  // DB's table num, unit is TSDB_TABLE_NUM_UNIT
+  int64_t   stateTs;
   SHashObj* vgHash;      // key:vgId, value:SVgroupInfo
 } SDBVgInfo;
 
@@ -203,6 +204,8 @@ int32_t taosAsyncExec(__async_exec_fn_t execFn, void* execParam, int32_t* code);
 
 void destroySendMsgInfo(SMsgSendInfo* pMsgBody);
 
+void destroyAhandle(void* ahandle);
+
 int32_t asyncSendMsgToServerExt(void* pTransporter, SEpSet* epSet, int64_t* pTransporterId, SMsgSendInfo* pInfo,
                                 bool persistHandle, void* ctx);
 

include/libs/sync/sync.h

@@ -58,7 +58,6 @@ typedef int64_t  SyncIndex;
 typedef uint64_t SyncTerm;
 
 typedef struct SSyncNode      SSyncNode;
-typedef struct SSyncBuffer    SSyncBuffer;
 typedef struct SWal           SWal;
 typedef struct SSyncRaftEntry SSyncRaftEntry;
 
@@ -139,6 +138,7 @@ typedef struct SSyncFSM {
   void (*FpReConfigCb)(const struct SSyncFSM* pFsm, const SRpcMsg* pMsg, const SReConfigCbMeta* pMeta);
   void (*FpLeaderTransferCb)(const struct SSyncFSM* pFsm, const SRpcMsg* pMsg, const SFsmCbMeta* pMeta);
   bool (*FpApplyQueueEmptyCb)(const struct SSyncFSM* pFsm);
+  int32_t (*FpApplyQueueItems)(const struct SSyncFSM* pFsm);
 
   void (*FpBecomeLeaderCb)(const struct SSyncFSM* pFsm);
   void (*FpBecomeFollowerCb)(const struct SSyncFSM* pFsm);

include/libs/transport/trpc.h

@@ -72,15 +72,18 @@ typedef struct SRpcMsg {
 typedef void (*RpcCfp)(void *parent, SRpcMsg *, SEpSet *epset);
 typedef bool (*RpcRfp)(int32_t code, tmsg_t msgType);
 typedef bool (*RpcTfp)(int32_t code, tmsg_t msgType);
+typedef void (*RpcDfp)(void *ahandle);
 
 typedef struct SRpcInit {
   char     localFqdn[TSDB_FQDN_LEN];
-  uint16_t localPort;     // local port
+  uint16_t localPort;      // local port
-  char    *label;         // for debug purpose
+  char    *label;          // for debug purpose
-  int32_t  numOfThreads;  // number of threads to handle connections
+  int32_t  numOfThreads;   // number of threads to handle connections
-  int32_t  sessions;      // number of sessions allowed
+  int32_t  sessions;       // number of sessions allowed
-  int8_t   connType;      // TAOS_CONN_UDP, TAOS_CONN_TCPC, TAOS_CONN_TCPS
+  int8_t   connType;       // TAOS_CONN_UDP, TAOS_CONN_TCPC, TAOS_CONN_TCPS
-  int32_t  idleTime;      // milliseconds, 0 means idle timer is disabled
+  int32_t  idleTime;       // milliseconds, 0 means idle timer is disabled
+  int32_t  retryLimit;     // retry limit
+  int32_t  retryInterval;  // retry interval ms
 
   int32_t compressSize;  // -1: no compress, 0 : all data compressed, size: compress data if larger than size
   int8_t  encryption;    // encrypt or not
@@ -97,6 +100,9 @@ typedef struct SRpcInit {
   // set up timeout for particular msg
   RpcTfp tfp;
 
+  // destroy client ahandle;
+  RpcDfp dfp;
+
   void *parent;
 } SRpcInit;
 

include/util/tdef.h

@@ -26,7 +26,7 @@ extern "C" {
 
 #define TSKEY             int64_t
 #define TSKEY_MIN         INT64_MIN
-#define TSKEY_MAX         (INT64_MAX - 1)
+#define TSKEY_MAX         INT64_MAX
 #define TSKEY_INITIAL_VAL TSKEY_MIN
 
 #define TD_VER_MAX UINT64_MAX  // TODO: use the real max version from query handle
@@ -290,7 +290,7 @@ typedef enum ELogicConditionType {
 #define TSDB_DEFAULT_VN_PER_DB          2
 #define TSDB_MIN_BUFFER_PER_VNODE       3      // unit MB
 #define TSDB_MAX_BUFFER_PER_VNODE       16384  // unit MB
-#define TSDB_DEFAULT_BUFFER_PER_VNODE   96
+#define TSDB_DEFAULT_BUFFER_PER_VNODE   256
 #define TSDB_MIN_PAGES_PER_VNODE        64
 #define TSDB_MAX_PAGES_PER_VNODE        (INT32_MAX - 1)
 #define TSDB_DEFAULT_PAGES_PER_VNODE    256

packaging/docker/dockerbuild.sh

@@ -5,7 +5,7 @@ set -e
 #set -x
 
 # dockerbuild.sh 
-#             -c [aarch32 | aarch64 | amd64 | x86 | mips64 ...]  
+#             -c [aarch32 | aarch64 | amd64 | x86 | mips64 | loongarch64...]
 #             -n [version number]
 #             -p [password for docker hub]
 #             -V [stable | beta]
@@ -57,7 +57,7 @@ do
       dockerLatest=$(echo $OPTARG)
       ;;
     h)
-      echo "Usage: `basename $0`  -c [aarch32 | aarch64 | amd64 | x86 | mips64 ...] "
+      echo "Usage: `basename $0`  -c [aarch32 | aarch64 | amd64 | x86 | mips64 | loongarch64...] "
       echo "                      -n [version number] "
       echo "                      -p [password for docker hub] "
       echo "                      -V [stable | beta] "
@@ -136,4 +136,4 @@ if  [ "$cloudBuild" != "y" ] && [ ${dockerLatest} == 'y' ]  ;then
   docker push tdengine/tdengine-${dockername}:latest
 fi
 
-rm -f ${pkgFile}
+rm -f ${pkgFile}

packaging/docker/dockerbuildi.sh

@@ -5,7 +5,7 @@ set -e
 #set -x
 
 # dockerbuild.sh 
-#             -c [aarch32 | aarch64 | amd64 | x86 | mips64 ...]  
+#             -c [aarch32 | aarch64 | amd64 | x86 | mips64 | loongarch64...]
 #             -n [version number]
 #             -p [password for docker hub]
 
@@ -30,7 +30,7 @@ do
       passWord=$(echo $OPTARG)
       ;;
     h)
-      echo "Usage: `basename $0` -c [aarch32 | aarch64 | amd64 | x86 | mips64 ...] "
+      echo "Usage: `basename $0` -c [aarch32 | aarch64 | amd64 | x86 | mips64 | loongarch64...] "
       echo "                      -n [version number] "
       echo "                      -p [password for docker hub] "
       exit 0

packaging/release.sh

@@ -6,7 +6,7 @@ set -e
 #set -x
 
 # release.sh  -v [cluster | edge]
-#             -c [aarch32 | aarch64 | x64 | x86 | mips64 ...]
+#             -c [aarch32 | aarch64 | x64 | x86 | mips64 | loongarch64...]
 #             -o [Linux | Kylin | Alpine | Raspberrypi | Darwin | Windows | Ningsi60 | Ningsi80 |...]
 #             -V [stable | beta]
 #             -l [full | lite]
@@ -19,7 +19,7 @@ set -e
 # set parameters by default value
 verMode=edge    # [cluster, edge, cloud]
 verType=stable  # [stable, beta]
-cpuType=x64     # [aarch32 | aarch64 | x64 | x86 | mips64 ...]
+cpuType=x64     # [aarch32 | aarch64 | x64 | x86 | mips64 loongarch64...]
 osType=Linux    # [Linux | Kylin | Alpine | Raspberrypi | Darwin | Windows | Ningsi60 | Ningsi80 |...]
 pagMode=full    # [full | lite]
 soMode=dynamic  # [static | dynamic]
@@ -77,7 +77,7 @@ while getopts "hv:V:c:o:l:s:d:a:n:m:H:" arg; do
     ;;
   h)
     echo "Usage: $(basename $0) -v [cluster | edge] "
-    echo "                  -c [aarch32 | aarch64 | x64 | x86 | mips64 ...] "
+    echo "                  -c [aarch32 | aarch64 | x64 | x86 | mips64 | loongarch64 ...] "
     echo "                  -o [Linux | Kylin | Alpine | Raspberrypi | Darwin | Windows | Ningsi60 | Ningsi80 |...] "
     echo "                  -V [stable | beta] "
     echo "                  -l [full | lite] "
@@ -216,7 +216,7 @@ else
 fi
 
 # check support cpu type
-if [[ "$cpuType" == "x64" ]] || [[ "$cpuType" == "aarch64" ]] || [[ "$cpuType" == "aarch32" ]] || [[ "$cpuType" == "arm64" ]] || [[ "$cpuType" == "arm32" ]] || [[ "$cpuType" == "mips64" ]]; then
+if [[ "$cpuType" == "x64" ]] || [[ "$cpuType" == "aarch64" ]] || [[ "$cpuType" == "aarch32" ]] || [[ "$cpuType" == "arm64" ]] || [[ "$cpuType" == "arm32" ]] || [[ "$cpuType" == "mips64" ]] || [[ "$cpuType" == "loongarch64" ]] ; then
   if [ "$verMode" == "edge" ]; then
     # community-version compile
     cmake ../ -DCPUTYPE=${cpuType} -DWEBSOCKET=true -DOSTYPE=${osType} -DSOMODE=${soMode} -DDBNAME=${dbName} -DVERTYPE=${verType} -DVERDATE="${build_time}" -DGITINFO=${gitinfo} -DGITINFOI=${gitinfoOfInternal} -DVERNUMBER=${verNumber} -DVERCOMPATIBLE=${verNumberComp} -DPAGMODE=${pagMode} -DBUILD_HTTP=${BUILD_HTTP} -DBUILD_TOOLS=${BUILD_TOOLS} ${allocator_macro}

source/client/src/clientEnv.c

@@ -140,12 +140,15 @@ void *openTransporter(const char *user, const char *auth, int32_t numOfThread) {
   rpcInit.numOfThreads = numOfThread;
   rpcInit.cfp = processMsgFromServer;
   rpcInit.rfp = clientRpcRfp;
-  // rpcInit.tfp = clientRpcTfp;
   rpcInit.sessions = 1024;
   rpcInit.connType = TAOS_CONN_CLIENT;
   rpcInit.user = (char *)user;
   rpcInit.idleTime = tsShellActivityTimer * 1000;
   rpcInit.compressSize = tsCompressMsgSize;
+  rpcInit.dfp = destroyAhandle;
+  rpcInit.retryLimit = tsRpcRetryLimit;
+  rpcInit.retryInterval = tsRpcRetryInterval;
+
   void *pDnodeConn = rpcOpen(&rpcInit);
   if (pDnodeConn == NULL) {
     tscError("failed to init connection to server");

source/client/src/clientHb.c

@@ -61,7 +61,8 @@ static int32_t hbProcessDBInfoRsp(void *value, int32_t valueLen, struct SCatalog
   int32_t numOfBatchs = taosArrayGetSize(batchUseRsp.pArray);
   for (int32_t i = 0; i < numOfBatchs; ++i) {
     SUseDbRsp *rsp = taosArrayGet(batchUseRsp.pArray, i);
-    tscDebug("hb db rsp, db:%s, vgVersion:%d, uid:%" PRIx64, rsp->db, rsp->vgVersion, rsp->uid);
+    tscDebug("hb db rsp, db:%s, vgVersion:%d, stateTs:%" PRId64 ", uid:%" PRIx64, rsp->db, rsp->vgVersion, rsp->stateTs,
+             rsp->uid);
 
     if (rsp->vgVersion < 0) {
       code = catalogRemoveDB(pCatalog, rsp->db, rsp->uid);
@@ -72,6 +73,7 @@ static int32_t hbProcessDBInfoRsp(void *value, int32_t valueLen, struct SCatalog
       }
 
       vgInfo->vgVersion = rsp->vgVersion;
+      vgInfo->stateTs = rsp->stateTs;
       vgInfo->hashMethod = rsp->hashMethod;
       vgInfo->hashPrefix = rsp->hashPrefix;
       vgInfo->hashSuffix = rsp->hashSuffix;
@@ -292,6 +294,7 @@ static int32_t hbAsyncCallBack(void *param, SDataBuf *pMsg, int32_t code) {
     taosThreadMutexUnlock(&appInfo.mutex);
     tscError("cluster not exist, key:%s", key);
     taosMemoryFree(pMsg->pData);
+    taosMemoryFree(pMsg->pEpSet);
     tFreeClientHbBatchRsp(&pRsp);
     return -1;
   }
@@ -321,6 +324,7 @@ static int32_t hbAsyncCallBack(void *param, SDataBuf *pMsg, int32_t code) {
 
   tFreeClientHbBatchRsp(&pRsp);
   taosMemoryFree(pMsg->pData);
+  taosMemoryFree(pMsg->pEpSet);
   return code;
 }
 
@@ -486,6 +490,7 @@ int32_t hbGetExpiredDBInfo(SClientHbKey *connKey, struct SCatalog *pCatalog, SCl
     db->dbId = htobe64(db->dbId);
     db->vgVersion = htonl(db->vgVersion);
     db->numOfTable = htonl(db->numOfTable);
+    db->stateTs = htobe64(db->stateTs);
   }
 
   SKv kv = {

source/client/src/clientImpl.c

@@ -1971,6 +1971,8 @@ TSDB_SERVER_STATUS taos_check_server_status(const char* fqdn, int port, char* de
   rpcInit.idleTime = tsShellActivityTimer * 1000;
   rpcInit.compressSize = tsCompressMsgSize;
   rpcInit.user = "_dnd";
+  rpcInit.retryLimit = tsRpcRetryLimit;
+  rpcInit.retryInterval = tsRpcRetryInterval;
 
   clientRpc = rpcOpen(&rpcInit);
   if (clientRpc == NULL) {

source/client/src/clientMain.c

@@ -990,7 +990,7 @@ void taos_fetch_rows_a(TAOS_RES *res, __taos_async_fn_t fp, void *param) {
   // all data has returned to App already, no need to try again
   if (pResultInfo->completed) {
     // it is a local executed query, no need to do async fetch
-    if (QUERY_EXEC_MODE_LOCAL == pRequest->body.execMode) {
+    if (QUERY_EXEC_MODE_SCHEDULE != pRequest->body.execMode) {
       if (pResultInfo->localResultFetched) {
         pResultInfo->numOfRows = 0;
         pResultInfo->current = 0;

source/client/src/clientMsgHandler.c

@@ -181,7 +181,7 @@ int32_t processUseDbRsp(void* param, SDataBuf* pMsg, int32_t code) {
     tDeserializeSUseDbRsp(pMsg->pData, pMsg->len, &usedbRsp);
     struct SCatalog* pCatalog = NULL;
 
-    if (usedbRsp.vgVersion >= 0) {
+    if (usedbRsp.vgVersion >= 0) { // cached in local
       uint64_t clusterId = pRequest->pTscObj->pAppInfo->clusterId;
       int32_t  code1 = catalogGetHandle(clusterId, &pCatalog);
       if (code1 != TSDB_CODE_SUCCESS) {
@@ -442,8 +442,7 @@ static int32_t buildShowVariablesRsp(SArray* pVars, SRetrieveTableRsp** pRsp) {
   (*pRsp)->numOfRows = htonl(pBlock->info.rows);
   (*pRsp)->numOfCols = htonl(SHOW_VARIABLES_RESULT_COLS);
 
-  int32_t len = 0;
+  int32_t len = blockEncode(pBlock, (*pRsp)->data, SHOW_VARIABLES_RESULT_COLS);
-  blockEncode(pBlock, (*pRsp)->data, &len, SHOW_VARIABLES_RESULT_COLS, false);
   ASSERT(len == rspSize - sizeof(SRetrieveTableRsp));
 
   blockDataDestroy(pBlock);

source/client/src/clientSml.c

@@ -2082,7 +2082,7 @@ static int32_t smlParseJSONString(SSmlHandle *info, cJSON *root, SSmlTableInfo *
 
 static int32_t smlParseInfluxLine(SSmlHandle *info, const char *sql, const int len) {
   SSmlLineInfo elements = {0};
-  uDebug("SML:0x%" PRIx64 " smlParseInfluxLine sql:%s, hello", info->id, sql);
+  uDebug("SML:0x%" PRIx64 " smlParseInfluxLine sql", info->id);
 
   int ret = smlParseInfluxString(sql, sql + len, &elements, &info->msgBuf);
   if (ret != TSDB_CODE_SUCCESS) {

source/common/src/tdatablock.c

@@ -334,10 +334,12 @@ int32_t colDataAssign(SColumnInfoData* pColumnInfoData, const SColumnInfoData* p
     }
 
     pColumnInfoData->varmeta.length = pSource->varmeta.length;
-    memcpy(pColumnInfoData->pData, pSource->pData, pSource->varmeta.length);
+    if (pColumnInfoData->pData != NULL && pSource->pData != NULL) {
+      memcpy(pColumnInfoData->pData, pSource->pData, pSource->varmeta.length);
+    }
   } else {
     memcpy(pColumnInfoData->nullbitmap, pSource->nullbitmap, BitmapLen(numOfRows));
-    if (pSource->pData) {
+    if (pSource->pData != NULL) {
       memcpy(pColumnInfoData->pData, pSource->pData, pSource->info.bytes * numOfRows);
     }
   }
@@ -2197,7 +2199,9 @@ char* buildCtbNameByGroupId(const char* stbFullName, uint64_t groupId) {
   return rname.ctbShortName;
 }
 
-void blockEncode(const SSDataBlock* pBlock, char* data, int32_t* dataLen, int32_t numOfCols, int8_t needCompress) {
+int32_t blockEncode(const SSDataBlock* pBlock, char* data, int32_t numOfCols) {
+  int32_t dataLen = 0;
+
   // todo extract method
   int32_t* version = (int32_t*)data;
   *version = 1;
@@ -2238,7 +2242,7 @@ void blockEncode(const SSDataBlock* pBlock, char* data, int32_t* dataLen, int32_
   int32_t* colSizes = (int32_t*)data;
   data += numOfCols * sizeof(int32_t);
 
-  *dataLen = blockDataGetSerialMetaSize(numOfCols);
+  dataLen = blockDataGetSerialMetaSize(numOfCols);
 
   int32_t numOfRows = pBlock->info.rows;
   for (int32_t col = 0; col < numOfCols; ++col) {
@@ -2255,26 +2259,25 @@ void blockEncode(const SSDataBlock* pBlock, char* data, int32_t* dataLen, int32_
     }
 
     data += metaSize;
-    (*dataLen) += metaSize;
+    dataLen += metaSize;
 
-    if (needCompress) {
+    colSizes[col] = colDataGetLength(pColRes, numOfRows);
-      colSizes[col] = blockCompressColData(pColRes, numOfRows, data, needCompress);
+    dataLen += colSizes[col];
-      data += colSizes[col];
+    if (pColRes->pData != NULL) {
-      (*dataLen) += colSizes[col];
-    } else {
-      colSizes[col] = colDataGetLength(pColRes, numOfRows);
-      (*dataLen) += colSizes[col];
       memmove(data, pColRes->pData, colSizes[col]);
-      data += colSizes[col];
     }
+    data += colSizes[col];
 
     colSizes[col] = htonl(colSizes[col]);
   }
 
-  *actualLen = *dataLen;
+  *actualLen = dataLen;
   *groupId = pBlock->info.groupId;
-  ASSERT(*dataLen > 0);
+  ASSERT(dataLen > 0);
-  uDebug("build data block, actualLen:%d, rows:%d, cols:%d", *dataLen, *rows, *cols);
+
+  uDebug("build data block, actualLen:%d, rows:%d, cols:%d", dataLen, *rows, *cols);
+
+  return dataLen;
 }
 
 const char* blockDecode(SSDataBlock* pBlock, const char* pData) {

source/common/src/tdataformat.c

@@ -1557,7 +1557,7 @@ void tColDataGetValue(SColData *pColData, int32_t iVal, SColVal *pColVal) {
   tColDataGetValueImpl[pColData->flag](pColData, iVal, pColVal);
 }
 
-uint8_t tColDataGetBitValue(SColData *pColData, int32_t iVal) {
+uint8_t tColDataGetBitValue(const SColData *pColData, int32_t iVal) {
   uint8_t v;
   switch (pColData->flag) {
     case HAS_NONE:

source/common/src/tglobal.c

@@ -163,10 +163,12 @@ int32_t tsMqRebalanceInterval = 2;
 int32_t tsTtlUnit = 86400;
 int32_t tsTtlPushInterval = 86400;
 int32_t tsGrantHBInterval = 60;
-int32_t tsUptimeInterval = 300;     // seconds
+int32_t tsUptimeInterval = 300;    // seconds
 char    tsUdfdResFuncs[512] = "";  // udfd resident funcs that teardown when udfd exits
 char    tsUdfdLdLibPath[512] = "";
 
+int32_t tsRpcRetryLimit = 100;
+int32_t tsRpcRetryInterval = 15;
 #ifndef _STORAGE
 int32_t taosSetTfsCfg(SConfig *pCfg) {
   SConfigItem *pItem = cfgGetItem(pCfg, "dataDir");
@@ -297,6 +299,8 @@ static int32_t taosAddClientCfg(SConfig *pCfg) {
   if (cfgAddString(pCfg, "smlTagName", tsSmlTagName, 1) != 0) return -1;
   if (cfgAddBool(pCfg, "smlDataFormat", tsSmlDataFormat, 1) != 0) return -1;
   if (cfgAddInt32(pCfg, "maxMemUsedByInsert", tsMaxMemUsedByInsert, 1, INT32_MAX, true) != 0) return -1;
+  if (cfgAddInt32(pCfg, "rpcRetryLimit", tsRpcRetryLimit, 1, 100000, 0) != 0) return -1;
+  if (cfgAddInt32(pCfg, "rpcRetryInterval", tsRpcRetryInterval, 1, 100000, 0) != 0) return -1;
 
   tsNumOfTaskQueueThreads = tsNumOfCores / 2;
   tsNumOfTaskQueueThreads = TMAX(tsNumOfTaskQueueThreads, 4);
@@ -422,6 +426,10 @@ static int32_t taosAddServerCfg(SConfig *pCfg) {
   if (cfgAddBool(pCfg, "udf", tsStartUdfd, 0) != 0) return -1;
   if (cfgAddString(pCfg, "udfdResFuncs", tsUdfdResFuncs, 0) != 0) return -1;
   if (cfgAddString(pCfg, "udfdLdLibPath", tsUdfdLdLibPath, 0) != 0) return -1;
+
+  if (cfgAddInt32(pCfg, "rpcRetryLimit", tsRpcRetryLimit, 1, 100000, 0) != 0) return -1;
+  if (cfgAddInt32(pCfg, "rpcRetryInterval", tsRpcRetryInterval, 1, 100000, 0) != 0) return -1;
+
   GRANT_CFG_ADD;
   return 0;
 }
@@ -634,6 +642,9 @@ static int32_t taosSetClientCfg(SConfig *pCfg) {
   tsQueryNodeChunkSize = cfgGetItem(pCfg, "queryNodeChunkSize")->i32;
   tsQueryUseNodeAllocator = cfgGetItem(pCfg, "queryUseNodeAllocator")->bval;
   tsKeepColumnName = cfgGetItem(pCfg, "keepColumnName")->bval;
+
+  tsRpcRetryLimit = cfgGetItem(pCfg, "rpcRetryLimit")->i32;
+  tsRpcRetryInterval = cfgGetItem(pCfg, "rpcRetryInterval")->i32;
   return 0;
 }
 
@@ -708,6 +719,9 @@ static int32_t taosSetServerCfg(SConfig *pCfg) {
   if (tsQueryBufferSize >= 0) {
     tsQueryBufferSizeBytes = tsQueryBufferSize * 1048576UL;
   }
+
+  tsRpcRetryLimit = cfgGetItem(pCfg, "rpcRetryLimit")->i32;
+  tsRpcRetryInterval = cfgGetItem(pCfg, "rpcRetryInterval")->i32;
   GRANT_CFG_GET;
   return 0;
 }

source/dnode/mgmt/mgmt_dnode/src/dmHandle.c

@@ -307,8 +307,7 @@ int32_t dmProcessRetrieve(SDnodeMgmt *pMgmt, SRpcMsg *pMsg) {
     pStart += sizeof(SSysTableSchema);
   }
 
-  int32_t len = 0;
+  int32_t len = blockEncode(pBlock, pStart, numOfCols);
-  blockEncode(pBlock, pStart, &len, numOfCols, false);
 
   pRsp->numOfRows = htonl(pBlock->info.rows);
   pRsp->precision = TSDB_TIME_PRECISION_MILLI;  // millisecond time precision

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

@@ -183,8 +183,6 @@ SArray *mmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_VND_COMPACT_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
 
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_TIMEOUT, mmPutMsgToSyncQueue, 1) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_SYNC_PING, mmPutMsgToSyncQueue, 1) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_SYNC_PING_REPLY, mmPutMsgToSyncQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_CLIENT_REQUEST, mmPutMsgToSyncQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_CLIENT_REQUEST_BATCH, mmPutMsgToSyncQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_CLIENT_REQUEST_REPLY, mmPutMsgToSyncQueue, 1) == NULL) goto _OVER;

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

@@ -450,8 +450,6 @@ SArray *vmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_DND_DROP_VNODE, vmPutMsgToMgmtQueue, 0) == NULL) goto _OVER;
 
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_TIMEOUT, vmPutMsgToSyncQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_SYNC_PING, vmPutMsgToSyncQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_SYNC_PING_REPLY, vmPutMsgToSyncQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_CLIENT_REQUEST, vmPutMsgToSyncQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_CLIENT_REQUEST_BATCH, vmPutMsgToSyncQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_CLIENT_REQUEST_REPLY, vmPutMsgToSyncQueue, 0) == NULL) goto _OVER;

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

@@ -258,6 +258,8 @@ int32_t dmInitClient(SDnode *pDnode) {
   rpcInit.parent = pDnode;
   rpcInit.rfp = rpcRfp;
   rpcInit.compressSize = tsCompressMsgSize;
+  rpcInit.retryLimit = tsRpcRetryLimit;
+  rpcInit.retryInterval = tsRpcRetryInterval;
 
   pTrans->clientRpc = rpcOpen(&rpcInit);
   if (pTrans->clientRpc == NULL) {

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

@@ -44,6 +44,7 @@ SSdbRaw *mndConsumerActionEncode(SMqConsumerObj *pConsumer);
 SSdbRow *mndConsumerActionDecode(SSdbRaw *pRaw);
 
 int32_t mndSetConsumerCommitLogs(SMnode *pMnode, STrans *pTrans, SMqConsumerObj *pConsumer);
+int32_t mndSetConsumerDropLogs(SMnode *pMnode, STrans *pTrans, SMqConsumerObj *pConsumer);
 
 bool mndRebTryStart();
 void mndRebEnd();

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

@@ -32,7 +32,8 @@
 #define MND_CONSUMER_VER_NUMBER   1
 #define MND_CONSUMER_RESERVE_SIZE 64
 
-#define MND_CONSUMER_LOST_HB_CNT 3
+#define MND_CONSUMER_LOST_HB_CNT          3
+#define MND_CONSUMER_LOST_CLEAR_THRESHOLD 43200
 
 static int8_t mqRebInExecCnt = 0;
 
@@ -50,6 +51,7 @@ static int32_t mndProcessAskEpReq(SRpcMsg *pMsg);
 static int32_t mndProcessMqHbReq(SRpcMsg *pMsg);
 static int32_t mndProcessMqTimerMsg(SRpcMsg *pMsg);
 static int32_t mndProcessConsumerLostMsg(SRpcMsg *pMsg);
+static int32_t mndProcessConsumerClearMsg(SRpcMsg *pMsg);
 static int32_t mndProcessConsumerRecoverMsg(SRpcMsg *pMsg);
 
 int32_t mndInitConsumer(SMnode *pMnode) {
@@ -69,6 +71,7 @@ int32_t mndInitConsumer(SMnode *pMnode) {
   mndSetMsgHandle(pMnode, TDMT_MND_TMQ_TIMER, mndProcessMqTimerMsg);
   mndSetMsgHandle(pMnode, TDMT_MND_TMQ_CONSUMER_LOST, mndProcessConsumerLostMsg);
   mndSetMsgHandle(pMnode, TDMT_MND_TMQ_CONSUMER_RECOVER, mndProcessConsumerRecoverMsg);
+  mndSetMsgHandle(pMnode, TDMT_MND_TMQ_LOST_CONSUMER_CLEAR, mndProcessConsumerClearMsg);
 
   mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_CONSUMERS, mndRetrieveConsumer);
   mndAddShowFreeIterHandle(pMnode, TSDB_MGMT_TABLE_CONSUMERS, mndCancelGetNextConsumer);
@@ -162,6 +165,43 @@ FAIL:
   return -1;
 }
 
+static int32_t mndProcessConsumerClearMsg(SRpcMsg *pMsg) {
+  SMnode              *pMnode = pMsg->info.node;
+  SMqConsumerClearMsg *pClearMsg = pMsg->pCont;
+  SMqConsumerObj      *pConsumer = mndAcquireConsumer(pMnode, pClearMsg->consumerId);
+  if (pConsumer == NULL) {
+    return 0;
+  }
+
+  mInfo("receive consumer clear msg, consumer id %" PRId64 ", status %s", pClearMsg->consumerId,
+        mndConsumerStatusName(pConsumer->status));
+
+  if (pConsumer->status != MQ_CONSUMER_STATUS__LOST_REBD) {
+    mndReleaseConsumer(pMnode, pConsumer);
+    return -1;
+  }
+
+  SMqConsumerObj *pConsumerNew = tNewSMqConsumerObj(pConsumer->consumerId, pConsumer->cgroup);
+  pConsumerNew->updateType = CONSUMER_UPDATE__LOST;
+
+  mndReleaseConsumer(pMnode, pConsumer);
+
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_NOTHING, pMsg, "clear-csm");
+  if (pTrans == NULL) goto FAIL;
+  if (mndSetConsumerDropLogs(pMnode, pTrans, pConsumerNew) != 0) goto FAIL;
+  if (mndTransPrepare(pMnode, pTrans) != 0) goto FAIL;
+
+  tDeleteSMqConsumerObj(pConsumerNew);
+  taosMemoryFree(pConsumerNew);
+  mndTransDrop(pTrans);
+  return 0;
+FAIL:
+  tDeleteSMqConsumerObj(pConsumerNew);
+  taosMemoryFree(pConsumerNew);
+  mndTransDrop(pTrans);
+  return -1;
+}
+
 static SMqRebInfo *mndGetOrCreateRebSub(SHashObj *pHash, const char *key) {
   SMqRebInfo *pRebInfo = taosHashGet(pHash, key, strlen(key) + 1);
   if (pRebInfo == NULL) {
@@ -206,15 +246,28 @@ static int32_t mndProcessMqTimerMsg(SRpcMsg *pMsg) {
       SMqConsumerLostMsg *pLostMsg = rpcMallocCont(sizeof(SMqConsumerLostMsg));
 
       pLostMsg->consumerId = pConsumer->consumerId;
-      SRpcMsg pRpcMsg = {
+      SRpcMsg rpcMsg = {
           .msgType = TDMT_MND_TMQ_CONSUMER_LOST,
           .pCont = pLostMsg,
           .contLen = sizeof(SMqConsumerLostMsg),
       };
-      tmsgPutToQueue(&pMnode->msgCb, WRITE_QUEUE, &pRpcMsg);
+      tmsgPutToQueue(&pMnode->msgCb, WRITE_QUEUE, &rpcMsg);
     }
-    if (status == MQ_CONSUMER_STATUS__LOST_REBD || status == MQ_CONSUMER_STATUS__READY) {
+
+    if (status == MQ_CONSUMER_STATUS__READY) {
       // do nothing
+    } else if (status == MQ_CONSUMER_STATUS__LOST_REBD) {
+      if (hbStatus > MND_CONSUMER_LOST_CLEAR_THRESHOLD) {
+        SMqConsumerClearMsg *pClearMsg = rpcMallocCont(sizeof(SMqConsumerClearMsg));
+
+        pClearMsg->consumerId = pConsumer->consumerId;
+        SRpcMsg rpcMsg = {
+            .msgType = TDMT_MND_TMQ_LOST_CONSUMER_CLEAR,
+            .pCont = pClearMsg,
+            .contLen = sizeof(SMqConsumerClearMsg),
+        };
+        tmsgPutToQueue(&pMnode->msgCb, WRITE_QUEUE, &rpcMsg);
+      }
     } else if (status == MQ_CONSUMER_STATUS__LOST) {
       taosRLockLatch(&pConsumer->lock);
       int32_t topicNum = taosArrayGetSize(pConsumer->currentTopics);
@@ -444,6 +497,14 @@ FAIL:
   return -1;
 }
 
+int32_t mndSetConsumerDropLogs(SMnode *pMnode, STrans *pTrans, SMqConsumerObj *pConsumer) {
+  SSdbRaw *pCommitRaw = mndConsumerActionEncode(pConsumer);
+  if (pCommitRaw == NULL) return -1;
+  if (mndTransAppendCommitlog(pTrans, pCommitRaw) != 0) return -1;
+  if (sdbSetRawStatus(pCommitRaw, SDB_STATUS_DROPPED) != 0) return -1;
+  return 0;
+}
+
 int32_t mndSetConsumerCommitLogs(SMnode *pMnode, STrans *pTrans, SMqConsumerObj *pConsumer) {
   SSdbRaw *pCommitRaw = mndConsumerActionEncode(pConsumer);
   if (pCommitRaw == NULL) return -1;

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

@@ -606,7 +606,7 @@ static int32_t mndProcessCreateDbReq(SRpcMsg *pReq) {
   } else {
     if (terrno == TSDB_CODE_MND_DB_IN_CREATING) {
       if (mndSetRpcInfoForDbTrans(pMnode, pReq, MND_OPER_CREATE_DB, createReq.db) == 0) {
-        mInfo("db:%s, is creating and response after trans finished", createReq.db);
+        mInfo("db:%s, is creating and createdb response after trans finished", createReq.db);
         code = TSDB_CODE_ACTION_IN_PROGRESS;
         goto _OVER;
       } else {
@@ -1175,7 +1175,7 @@ int32_t mndExtractDbInfo(SMnode *pMnode, SDbObj *pDb, SUseDbRsp *pRsp, const SUs
 
   int32_t numOfTable = mndGetDBTableNum(pDb, pMnode);
 
-  if (pReq == NULL || pReq->vgVersion < pDb->vgVersion || pReq->dbId != pDb->uid || numOfTable != pReq->numOfTable) {
+  if (pReq == NULL || pReq->vgVersion < pDb->vgVersion || pReq->dbId != pDb->uid || numOfTable != pReq->numOfTable || pReq->stateTs < pDb->stateTs) {
     mndBuildDBVgroupInfo(pDb, pMnode, pRsp->pVgroupInfos);
   }
 
@@ -1225,6 +1225,14 @@ static int32_t mndProcessUseDbReq(SRpcMsg *pReq) {
       usedbRsp.vgVersion = usedbReq.vgVersion;
       usedbRsp.errCode = terrno;
 
+      if (terrno == TSDB_CODE_MND_DB_IN_CREATING) {
+        if (mndSetRpcInfoForDbTrans(pMnode, pReq, MND_OPER_CREATE_DB, usedbReq.db) == 0) {
+          mInfo("db:%s, is creating and usedb response after trans finished", usedbReq.db);
+          code = TSDB_CODE_ACTION_IN_PROGRESS;
+          goto _OVER;
+        }
+      }
+
       mError("db:%s, failed to process use db req since %s", usedbReq.db, terrstr());
     } else {
       if (mndCheckDbPrivilege(pMnode, pReq->info.conn.user, MND_OPER_USE_DB, pDb) != 0) {
@@ -1255,7 +1263,7 @@ static int32_t mndProcessUseDbReq(SRpcMsg *pReq) {
   pReq->info.rspLen = contLen;
 
 _OVER:
-  if (code != 0) {
+  if (code != 0 && code != TSDB_CODE_ACTION_IN_PROGRESS) {
     mError("db:%s, failed to process use db req since %s", usedbReq.db, terrstr());
   }
 
@@ -1275,12 +1283,31 @@ int32_t mndValidateDbInfo(SMnode *pMnode, SDbVgVersion *pDbs, int32_t numOfDbs,
 
   for (int32_t i = 0; i < numOfDbs; ++i) {
     SDbVgVersion *pDbVgVersion = &pDbs[i];
-    pDbVgVersion->dbId = htobe64(pDbVgVersion->dbId);
+    pDbVgVersion->dbId = be64toh(pDbVgVersion->dbId);
     pDbVgVersion->vgVersion = htonl(pDbVgVersion->vgVersion);
     pDbVgVersion->numOfTable = htonl(pDbVgVersion->numOfTable);
+    pDbVgVersion->stateTs = be64toh(pDbVgVersion->stateTs);
 
     SUseDbRsp usedbRsp = {0};
 
+    if ((0 == strcasecmp(pDbVgVersion->dbFName, TSDB_INFORMATION_SCHEMA_DB) || (0 == strcasecmp(pDbVgVersion->dbFName, TSDB_PERFORMANCE_SCHEMA_DB)))) {
+      memcpy(usedbRsp.db, pDbVgVersion->dbFName, TSDB_DB_FNAME_LEN);
+      int32_t vgVersion = mndGetGlobalVgroupVersion(pMnode);
+      if (pDbVgVersion->vgVersion < vgVersion) {
+        usedbRsp.pVgroupInfos = taosArrayInit(10, sizeof(SVgroupInfo));
+    
+        mndBuildDBVgroupInfo(NULL, pMnode, usedbRsp.pVgroupInfos);
+        usedbRsp.vgVersion = vgVersion++;
+      } else {
+        usedbRsp.vgVersion = pDbVgVersion->vgVersion;
+      }
+      usedbRsp.vgNum = taosArrayGetSize(usedbRsp.pVgroupInfos);
+      
+      taosArrayPush(batchUseRsp.pArray, &usedbRsp);
+      
+      continue;
+    }
+
     SDbObj *pDb = mndAcquireDb(pMnode, pDbVgVersion->dbFName);
     if (pDb == NULL) {
       mTrace("db:%s, no exist", pDbVgVersion->dbFName);
@@ -1293,8 +1320,8 @@ int32_t mndValidateDbInfo(SMnode *pMnode, SDbVgVersion *pDbs, int32_t numOfDbs,
 
     int32_t numOfTable = mndGetDBTableNum(pDb, pMnode);
 
-    if (pDbVgVersion->vgVersion >= pDb->vgVersion && numOfTable == pDbVgVersion->numOfTable /* &&
+    if (pDbVgVersion->vgVersion >= pDb->vgVersion && numOfTable == pDbVgVersion->numOfTable  &&
-        pDbVgVersion->stateTs == pDb->stateTs */) {
+        pDbVgVersion->stateTs == pDb->stateTs) {
       mTrace("db:%s, valid dbinfo, vgVersion:%d stateTs:%" PRId64
              " numOfTables:%d, not changed vgVersion:%d stateTs:%" PRId64 " numOfTables:%d",
              pDbVgVersion->dbFName, pDbVgVersion->vgVersion, pDbVgVersion->stateTs, pDbVgVersion->numOfTable,

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

@@ -303,8 +303,7 @@ static int32_t mndProcessRetrieveSysTableReq(SRpcMsg *pReq) {
       pStart += sizeof(SSysTableSchema);
     }
 
-    int32_t len = 0;
+    int32_t len = blockEncode(pBlock, pStart, pShow->pMeta->numOfColumns);
-    blockEncode(pBlock, pStart, &len, pShow->pMeta->numOfColumns, false);
   }
 
   pRsp->numOfRows = htonl(rowsRead);

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

@@ -532,6 +532,7 @@ static int32_t mndCreateSma(SMnode *pMnode, SRpcMsg *pReq, SMCreateSmaReq *pCrea
   streamObj.sql = strdup(pCreate->sql);
   streamObj.smaId = smaObj.uid;
   streamObj.watermark = pCreate->watermark;
+  streamObj.fillHistory = STREAM_FILL_HISTORY_ON;
   streamObj.trigger = STREAM_TRIGGER_WINDOW_CLOSE;
   streamObj.triggerParam = pCreate->maxDelay;
   streamObj.ast = strdup(smaObj.ast);

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

@@ -205,8 +205,23 @@ static void mndBecomeLeader(const SSyncFSM *pFsm) {
 static bool mndApplyQueueEmpty(const SSyncFSM *pFsm) {
   SMnode *pMnode = pFsm->data;
 
-  int32_t itemSize = tmsgGetQueueSize(&pMnode->msgCb, 1, APPLY_QUEUE);
+  if (pMnode != NULL && pMnode->msgCb.qsizeFp != NULL) {
-  return (itemSize == 0);
+    int32_t itemSize = tmsgGetQueueSize(&pMnode->msgCb, 1, APPLY_QUEUE);
+    return (itemSize == 0);
+  } else {
+    return true;
+  }
+}
+
+static int32_t mndApplyQueueItems(const SSyncFSM *pFsm) {
+  SMnode *pMnode = pFsm->data;
+
+  if (pMnode != NULL && pMnode->msgCb.qsizeFp != NULL) {
+    int32_t itemSize = tmsgGetQueueSize(&pMnode->msgCb, 1, APPLY_QUEUE);
+    return itemSize;
+  } else {
+    return -1;
+  }
 }
 
 SSyncFSM *mndSyncMakeFsm(SMnode *pMnode) {
@@ -218,6 +233,7 @@ SSyncFSM *mndSyncMakeFsm(SMnode *pMnode) {
   pFsm->FpRestoreFinishCb = mndRestoreFinish;
   pFsm->FpLeaderTransferCb = NULL;
   pFsm->FpApplyQueueEmptyCb = mndApplyQueueEmpty;
+  pFsm->FpApplyQueueItems = mndApplyQueueItems;
   pFsm->FpReConfigCb = NULL;
   pFsm->FpBecomeLeaderCb = mndBecomeLeader;
   pFsm->FpBecomeFollowerCb = mndBecomeFollower;
@@ -291,7 +307,7 @@ int32_t mndSyncPropose(SMnode *pMnode, SSdbRaw *pRaw, int32_t transId) {
 
   SRpcMsg req = {.msgType = TDMT_MND_APPLY_MSG, .contLen = sdbGetRawTotalSize(pRaw)};
   if (req.contLen <= 0) return -1;
-  
+
   req.pCont = rpcMallocCont(req.contLen);
   if (req.pCont == NULL) return -1;
   memcpy(req.pCont, pRaw, req.contLen);

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

@@ -637,6 +637,7 @@ static int32_t mndProcessDropTopicReq(SRpcMsg *pReq) {
     if (pIter == NULL) break;
 
     if (pConsumer->status == MQ_CONSUMER_STATUS__LOST_REBD) continue;
+
     int32_t sz = taosArrayGetSize(pConsumer->assignedTopics);
     for (int32_t i = 0; i < sz; i++) {
       char *name = taosArrayGetP(pConsumer->assignedTopics, i);
@@ -649,6 +650,33 @@ static int32_t mndProcessDropTopicReq(SRpcMsg *pReq) {
         return -1;
       }
     }
+
+    sz = taosArrayGetSize(pConsumer->rebNewTopics);
+    for (int32_t i = 0; i < sz; i++) {
+      char *name = taosArrayGetP(pConsumer->rebNewTopics, i);
+      if (strcmp(name, pTopic->name) == 0) {
+        mndReleaseConsumer(pMnode, pConsumer);
+        mndReleaseTopic(pMnode, pTopic);
+        terrno = TSDB_CODE_MND_TOPIC_SUBSCRIBED;
+        mError("topic:%s, failed to drop since subscribed by consumer:%" PRId64 ", in consumer group %s (reb new)",
+               dropReq.name, pConsumer->consumerId, pConsumer->cgroup);
+        return -1;
+      }
+    }
+
+    sz = taosArrayGetSize(pConsumer->rebRemovedTopics);
+    for (int32_t i = 0; i < sz; i++) {
+      char *name = taosArrayGetP(pConsumer->rebRemovedTopics, i);
+      if (strcmp(name, pTopic->name) == 0) {
+        mndReleaseConsumer(pMnode, pConsumer);
+        mndReleaseTopic(pMnode, pTopic);
+        terrno = TSDB_CODE_MND_TOPIC_SUBSCRIBED;
+        mError("topic:%s, failed to drop since subscribed by consumer:%" PRId64 ", in consumer group %s (reb remove)",
+               dropReq.name, pConsumer->consumerId, pConsumer->cgroup);
+        return -1;
+      }
+    }
+
     sdbRelease(pSdb, pConsumer);
   }
 
@@ -675,15 +703,6 @@ static int32_t mndProcessDropTopicReq(SRpcMsg *pReq) {
 
   mInfo("trans:%d, used to drop topic:%s", pTrans->id, pTopic->name);
 
-#if 0
-  if (mndDropOffsetByTopic(pMnode, pTrans, dropReq.name) < 0) {
-    ASSERT(0);
-    mndTransDrop(pTrans);
-    mndReleaseTopic(pMnode, pTopic);
-    return -1;
-  }
-#endif
-
   // TODO check if rebalancing
   if (mndDropSubByTopic(pMnode, pTrans, dropReq.name) < 0) {
     /*ASSERT(0);*/

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

@@ -938,11 +938,15 @@ static void mndTransSendRpcRsp(SMnode *pMnode, STrans *pTrans) {
   for (int32_t i = 0; i < size; ++i) {
     SRpcHandleInfo *pInfo = taosArrayGet(pTrans->pRpcArray, i);
     if (pInfo->handle != NULL) {
-      mInfo("trans:%d, send rsp, code:0x%x stage:%s app:%p", pTrans->id, code, mndTransStr(pTrans->stage),
-            pInfo->ahandle);
       if (code == TSDB_CODE_RPC_NETWORK_UNAVAIL) {
         code = TSDB_CODE_MND_TRANS_NETWORK_UNAVAILL;
       }
+      if (i != 0 && code == 0) {
+        code = TSDB_CODE_RPC_REDIRECT;
+      }
+      mInfo("trans:%d, client:%d send rsp, code:0x%x stage:%s app:%p", pTrans->id, i, code, mndTransStr(pTrans->stage),
+            pInfo->ahandle);
+
       SRpcMsg rspMsg = {.code = code, .info = *pInfo};
 
       if (pTrans->originRpcType == TDMT_MND_CREATE_DB) {

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

@@ -70,6 +70,7 @@ int32_t metaCacheDrop(SMeta* pMeta, int64_t uid);
 int32_t metaStatsCacheUpsert(SMeta* pMeta, SMetaStbStats* pInfo);
 int32_t metaStatsCacheDrop(SMeta* pMeta, int64_t uid);
 int32_t metaStatsCacheGet(SMeta* pMeta, int64_t uid, SMetaStbStats* pInfo);
+void metaUpdateStbStats(SMeta *pMeta, int64_t uid, int64_t delta);
 
 struct SMeta {
   TdThreadRwlock lock;

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

@@ -110,7 +110,6 @@ static FORCE_INLINE int64_t tsdbLogicToFileSize(int64_t lSize, int32_t szPage) {
 #define tsdbRowFromBlockData(BLOCKDATA, IROW) ((TSDBROW){.type = 1, .pBlockData = (BLOCKDATA), .iRow = (IROW)})
 void    tsdbRowGetColVal(TSDBROW *pRow, STSchema *pTSchema, int32_t iCol, SColVal *pColVal);
 int32_t tPutTSDBRow(uint8_t *p, TSDBROW *pRow);
-int32_t tGetTSDBRow(uint8_t *p, TSDBROW *pRow);
 int32_t tsdbRowCmprFn(const void *p1, const void *p2);
 // SRowIter
 void     tRowIterInit(SRowIter *pIter, TSDBROW *pRow, STSchema *pTSchema);
@@ -210,11 +209,10 @@ void     tsdbRefMemTable(SMemTable *pMemTable);
 void     tsdbUnrefMemTable(SMemTable *pMemTable);
 SArray  *tsdbMemTableGetTbDataArray(SMemTable *pMemTable);
 // STbDataIter
-int32_t  tsdbTbDataIterCreate(STbData *pTbData, TSDBKEY *pFrom, int8_t backward, STbDataIter **ppIter);
+int32_t tsdbTbDataIterCreate(STbData *pTbData, TSDBKEY *pFrom, int8_t backward, STbDataIter **ppIter);
-void    *tsdbTbDataIterDestroy(STbDataIter *pIter);
+void   *tsdbTbDataIterDestroy(STbDataIter *pIter);
-void     tsdbTbDataIterOpen(STbData *pTbData, TSDBKEY *pFrom, int8_t backward, STbDataIter *pIter);
+void    tsdbTbDataIterOpen(STbData *pTbData, TSDBKEY *pFrom, int8_t backward, STbDataIter *pIter);
-TSDBROW *tsdbTbDataIterGet(STbDataIter *pIter);
+bool    tsdbTbDataIterNext(STbDataIter *pIter);
-bool     tsdbTbDataIterNext(STbDataIter *pIter);
 // STbData
 int32_t tsdbGetNRowsInTbData(STbData *pTbData);
 // tsdbFile.c ==============================================================================================
@@ -772,6 +770,40 @@ static FORCE_INLINE int32_t tsdbKeyCmprFn(const void *p1, const void *p2) {
   return 0;
 }
 
+#define SL_NODE_FORWARD(n, l)  ((n)->forwards[l])
+#define SL_NODE_BACKWARD(n, l) ((n)->forwards[(n)->level + (l)])
+#define SL_NODE_DATA(n)        (&SL_NODE_BACKWARD(n, (n)->level))
+
+static FORCE_INLINE int32_t tGetTSDBRow(uint8_t *p, TSDBROW *pRow) {
+  int32_t n = tGetI64(p, &pRow->version);
+  pRow->pTSRow = (STSRow *)(p + n);
+  n += pRow->pTSRow->len;
+  return n;
+}
+
+static FORCE_INLINE TSDBROW *tsdbTbDataIterGet(STbDataIter *pIter) {
+  if (pIter == NULL) return NULL;
+
+  if (pIter->pRow) {
+    return pIter->pRow;
+  }
+
+  if (pIter->backward) {
+    if (pIter->pNode == pIter->pTbData->sl.pHead) {
+      return NULL;
+    }
+  } else {
+    if (pIter->pNode == pIter->pTbData->sl.pTail) {
+      return NULL;
+    }
+  }
+
+  tGetTSDBRow((uint8_t *)SL_NODE_DATA(pIter->pNode), &pIter->row);
+  pIter->pRow = &pIter->row;
+
+  return pIter->pRow;
+}
+
 #ifdef __cplusplus
 }
 #endif

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

@@ -61,14 +61,14 @@ struct SVBufPoolNode {
 };
 
 struct SVBufPool {
-  SVBufPool*       next;
+  SVBufPool*        next;
-  SVnode*          pVnode;
+  SVnode*           pVnode;
-  volatile int32_t nRef;
+  TdThreadSpinlock* lock;
-  TdThreadSpinlock lock;
+  volatile int32_t  nRef;
-  int64_t          size;
+  int64_t           size;
-  uint8_t*         ptr;
+  uint8_t*          ptr;
-  SVBufPoolNode*   pTail;
+  SVBufPoolNode*    pTail;
-  SVBufPoolNode    node;
+  SVBufPoolNode     node;
 };
 
 int32_t vnodeOpenBufPool(SVnode* pVnode);

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

@@ -1445,3 +1445,13 @@ int32_t metaGetStbStats(SMeta *pMeta, int64_t uid, SMetaStbStats *pInfo) {
 _exit:
   return code;
 }
+
+void metaUpdateStbStats(SMeta *pMeta, int64_t uid, int64_t delta) {
+  SMetaStbStats stats = {0};
+
+  if (metaStatsCacheGet(pMeta, uid, &stats) == TSDB_CODE_SUCCESS) {
+    stats.ctbNum += delta;
+
+    metaStatsCacheUpsert(pMeta, &stats);
+  }
+}

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

@@ -371,7 +371,7 @@ int metaAlterSTable(SMeta *pMeta, int64_t version, SVCreateStbReq *pReq) {
   // update uid index
   metaUpdateUidIdx(pMeta, &nStbEntry);
 
-  metaStatsCacheDrop(pMeta, nStbEntry.uid);
+  // metaStatsCacheDrop(pMeta, nStbEntry.uid);
 
   metaULock(pMeta);
 
@@ -450,6 +450,10 @@ int metaCreateTable(SMeta *pMeta, int64_t version, SVCreateTbReq *pReq, STableMe
 #endif
 
     ++pMeta->pVnode->config.vndStats.numOfCTables;
+
+    metaWLock(pMeta);
+    metaUpdateStbStats(pMeta, me.ctbEntry.suid, 1);
+    metaULock(pMeta);
   } else {
     me.ntbEntry.ctime = pReq->ctime;
     me.ntbEntry.ttlDays = pReq->ttl;
@@ -670,6 +674,8 @@ static int metaDropTableByUid(SMeta *pMeta, tb_uid_t uid, int *type) {
     tdbTbDelete(pMeta->pCtbIdx, &(SCtbIdxKey){.suid = e.ctbEntry.suid, .uid = uid}, sizeof(SCtbIdxKey), &pMeta->txn);
 
     --pMeta->pVnode->config.vndStats.numOfCTables;
+
+    metaUpdateStbStats(pMeta, e.ctbEntry.suid, -1);
   } else if (e.type == TSDB_NORMAL_TABLE) {
     // drop schema.db (todo)
 

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

@@ -582,10 +582,10 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg) {
       code = -1;
     }
 
-    tqDebug("tmq poll: consumer %" PRId64 ", subkey %s, vg %d, send data blockNum:%d, offset type:%d, uid:%" PRId64
+    tqDebug("tmq poll: consumer %" PRId64
-            ", version:%" PRId64 "",
+            ", subkey %s, vg %d, send data blockNum:%d, offset type:%d, uid/version:%" PRId64 ", ts:%" PRId64 "",
             consumerId, pHandle->subKey, TD_VID(pTq->pVnode), dataRsp.blockNum, dataRsp.rspOffset.type,
-            dataRsp.rspOffset.uid, dataRsp.rspOffset.version);
+            dataRsp.rspOffset.uid, dataRsp.rspOffset.ts);
 
     tDeleteSMqDataRsp(&dataRsp);
     return code;

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

@@ -27,9 +27,7 @@ int32_t tqAddBlockDataToRsp(const SSDataBlock* pBlock, SMqDataRsp* pRsp, int32_t
   pRetrieve->completed = 1;
   pRetrieve->numOfRows = htonl(pBlock->info.rows);
 
-  // TODO enable compress
+  int32_t actualLen = blockEncode(pBlock, pRetrieve->data, numOfCols);
-  int32_t actualLen = 0;
-  blockEncode(pBlock, pRetrieve->data, &actualLen, numOfCols, false);
   actualLen += sizeof(SRetrieveTableRsp);
   ASSERT(actualLen <= dataStrLen);
   taosArrayPush(pRsp->blockDataLen, &actualLen);

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

@@ -244,7 +244,7 @@ SSubmitReq* tqBlockToSubmit(SVnode* pVnode, const SArray* pBlocks, const STSchem
 
     int32_t rows = pDataBlock->info.rows;
 
-    tqDebug("tq sink, convert block %d, rows: %d", i, rows);
+    tqDebug("tq sink, convert block1 %d, rows: %d", i, rows);
 
     int32_t dataLen = 0;
     int32_t schemaLen = 0;
@@ -486,7 +486,7 @@ void tqSinkToTablePipeline(SStreamTask* pTask, void* vnode, int64_t ver, void* d
       blkHead->uid = 0;
       blkHead->schemaLen = 0;
 
-      tqDebug("tq sink, convert block %d, rows: %d", i, rows);
+      tqDebug("tq sink, convert block2 %d, rows: %d", i, rows);
 
       int32_t dataLen = 0;
       void*   blkSchema = POINTER_SHIFT(blkHead, sizeof(SSubmitBlk));
@@ -514,6 +514,9 @@ void tqSinkToTablePipeline(SStreamTask* pTask, void* vnode, int64_t ver, void* d
             tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NULL, NULL, false, pColumn->offset, k);
           } else {
             void* colData = colDataGetData(pColData, j);
+            if (k == 0) {
+              tqDebug("tq sink, row %d ts %" PRId64, j, *(int64_t*)colData);
+            }
             tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NORM, colData, true, pColumn->offset, k);
           }
         }

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

@@ -475,7 +475,7 @@ _err:
   }
   return code;
 }
-
+/*
 static int32_t getTableDelIdx(SDelFReader *pDelFReader, tb_uid_t suid, tb_uid_t uid, SDelIdx *pDelIdx) {
   int32_t code = 0;
   SArray *pDelIdxArray = NULL;
@@ -499,7 +499,7 @@ _err:
   }
   return code;
 }
-
+*/
 typedef enum {
   SFSLASTNEXTROW_FS,
   SFSLASTNEXTROW_FILESET,
@@ -997,8 +997,6 @@ static int32_t nextRowIterOpen(CacheNextRowIter *pIter, tb_uid_t uid, STsdb *pTs
 
   pIter->pSkyline = taosArrayInit(32, sizeof(TSDBKEY));
 
-  SDelIdx delIdx;
-
   SDelFile *pDelFile = pReadSnap->fs.pDelFile;
   if (pDelFile) {
     SDelFReader *pDelFReader;
@@ -1006,18 +1004,20 @@ static int32_t nextRowIterOpen(CacheNextRowIter *pIter, tb_uid_t uid, STsdb *pTs
     code = tsdbDelFReaderOpen(&pDelFReader, pDelFile, pTsdb);
     if (code) goto _err;
 
-    code = getTableDelIdx(pDelFReader, suid, uid, &delIdx);
+    SArray *pDelIdxArray = taosArrayInit(32, sizeof(SDelIdx));
-    if (code) {
+
-      tsdbDelFReaderClose(&pDelFReader);
+    code = tsdbReadDelIdx(pDelFReader, pDelIdxArray);
-      goto _err;
+    if (code) goto _err;
-    }
+
-
+    SDelIdx *delIdx = taosArraySearch(pDelIdxArray, &(SDelIdx){.suid = suid, .uid = uid}, tCmprDelIdx, TD_EQ);
-    code = getTableDelSkyline(pMem, pIMem, pDelFReader, &delIdx, pIter->pSkyline);
+
+    code = getTableDelSkyline(pMem, pIMem, pDelFReader, delIdx, pIter->pSkyline);
     if (code) {
       tsdbDelFReaderClose(&pDelFReader);
       goto _err;
     }
 
+    taosArrayDestroy(pDelIdxArray);
     tsdbDelFReaderClose(&pDelFReader);
   } else {
     code = getTableDelSkyline(pMem, pIMem, NULL, NULL, pIter->pSkyline);

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

@@ -495,6 +495,10 @@ static int32_t tsdbCommitFileDataStart(SCommitter *pCommitter) {
   pCommitter->commitFid = tsdbKeyFid(pCommitter->nextKey, pCommitter->minutes, pCommitter->precision);
   tsdbFidKeyRange(pCommitter->commitFid, pCommitter->minutes, pCommitter->precision, &pCommitter->minKey,
                   &pCommitter->maxKey);
+#if 0
+  ASSERT(pCommitter->minKey <= pCommitter->nextKey && pCommitter->maxKey >= pCommitter->nextKey);
+#endif
+  
   pCommitter->nextKey = TSKEY_MAX;
 
   // Reader

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

@@ -294,31 +294,6 @@ bool tsdbTbDataIterNext(STbDataIter *pIter) {
   return true;
 }
 
-TSDBROW *tsdbTbDataIterGet(STbDataIter *pIter) {
-  // we add here for commit usage
-  if (pIter == NULL) return NULL;
-
-  if (pIter->pRow) {
-    goto _exit;
-  }
-
-  if (pIter->backward) {
-    if (pIter->pNode == pIter->pTbData->sl.pHead) {
-      goto _exit;
-    }
-  } else {
-    if (pIter->pNode == pIter->pTbData->sl.pTail) {
-      goto _exit;
-    }
-  }
-
-  tGetTSDBRow((uint8_t *)SL_NODE_DATA(pIter->pNode), &pIter->row);
-  pIter->pRow = &pIter->row;
-
-_exit:
-  return pIter->pRow;
-}
-
 static int32_t tsdbMemTableRehash(SMemTable *pMemTable) {
   int32_t code = 0;
 
@@ -483,11 +458,10 @@ static void tbDataMovePosTo(STbData *pTbData, SMemSkipListNode **pos, TSDBKEY *p
 }
 
 static FORCE_INLINE int8_t tsdbMemSkipListRandLevel(SMemSkipList *pSl) {
-  int8_t         level = 1;
+  int8_t level = 1;
-  int8_t         tlevel = TMIN(pSl->maxLevel, pSl->level + 1);
+  int8_t tlevel = TMIN(pSl->maxLevel, pSl->level + 1);
-  const uint32_t factor = 4;
 
-  while ((taosRandR(&pSl->seed) % factor) == 0 && level < tlevel) {
+  while ((taosRandR(&pSl->seed) & 0x3) == 0 && level < tlevel) {
     level++;
   }
 
@@ -593,7 +567,9 @@ static int32_t tsdbInsertTableDataImpl(SMemTable *pMemTable, STbData *pTbData, i
     do {
       key.ts = row.pTSRow->ts;
       nRow++;
-      tbDataMovePosTo(pTbData, pos, &key, SL_MOVE_FROM_POS);
+      if (SL_NODE_FORWARD(pos[0], 0) != pTbData->sl.pTail) {
+        tbDataMovePosTo(pTbData, pos, &key, SL_MOVE_FROM_POS);
+      }
       code = tbDataDoPut(pMemTable, pTbData, pos, &row, 1);
       if (code) {
         goto _err;

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

@@ -893,7 +893,7 @@ static int doBinarySearchKey(TSKEY* keyList, int num, int pos, TSKEY key, int or
   }
 }
 
-int32_t getEndPosInDataBlock(STsdbReader* pReader, SBlockData* pBlockData, SDataBlk* pBlock, int32_t pos) {
+static int32_t getEndPosInDataBlock(STsdbReader* pReader, SBlockData* pBlockData, SDataBlk* pBlock, int32_t pos) {
   // NOTE: reverse the order to find the end position in data block
   int32_t endPos = -1;
   bool    asc = ASCENDING_TRAVERSE(pReader->order);
@@ -910,6 +910,117 @@ int32_t getEndPosInDataBlock(STsdbReader* pReader, SBlockData* pBlockData, SData
   return endPos;
 }
 
+static void copyPrimaryTsCol(const SBlockData* pBlockData, SFileBlockDumpInfo* pDumpInfo, SColumnInfoData* pColData,
+                             int32_t dumpedRows, bool asc) {
+  if (asc) {
+    memcpy(pColData->pData, &pBlockData->aTSKEY[pDumpInfo->rowIndex], dumpedRows * sizeof(int64_t));
+  } else {
+    int32_t startIndex = pDumpInfo->rowIndex - dumpedRows + 1;
+    memcpy(pColData->pData, &pBlockData->aTSKEY[startIndex], dumpedRows * sizeof(int64_t));
+
+    // todo: opt perf by extract the loop
+    // reverse the array list
+    int32_t  mid = dumpedRows >> 1u;
+    int64_t* pts = (int64_t*)pColData->pData;
+    for (int32_t j = 0; j < mid; ++j) {
+      int64_t t = pts[j];
+      pts[j] = pts[dumpedRows - j - 1];
+      pts[dumpedRows - j - 1] = t;
+    }
+  }
+}
+
+// a faster version of copy procedure.
+static void copyNumericCols(const SColData* pData, SFileBlockDumpInfo* pDumpInfo, SColumnInfoData* pColData,
+                            int32_t dumpedRows, bool asc)  {
+  uint8_t* p = NULL;
+  if (asc) {
+    p = pData->pData + tDataTypes[pData->type].bytes * pDumpInfo->rowIndex;
+  } else {
+    int32_t startIndex = pDumpInfo->rowIndex - dumpedRows + 1;
+    p = pData->pData + tDataTypes[pData->type].bytes * startIndex;
+  }
+
+  int32_t step = asc? 1:-1;
+
+  // make sure it is aligned to 8bit
+  ASSERT((((uint64_t)pColData->pData) & (0x8 - 1)) == 0);
+
+  // 1. copy data in a batch model
+  memcpy(pColData->pData, p, dumpedRows * tDataTypes[pData->type].bytes);
+
+  // 2. reverse the array list in case of descending order scan data block
+  if (!asc) {
+    switch(pColData->info.type) {
+      case TSDB_DATA_TYPE_TIMESTAMP:
+      case TSDB_DATA_TYPE_DOUBLE:
+      case TSDB_DATA_TYPE_BIGINT:
+      case TSDB_DATA_TYPE_UBIGINT:
+      {
+        int32_t  mid = dumpedRows >> 1u;
+        int64_t* pts = (int64_t*)pColData->pData;
+        for (int32_t j = 0; j < mid; ++j) {
+          int64_t t = pts[j];
+          pts[j] = pts[dumpedRows - j - 1];
+          pts[dumpedRows - j - 1] = t;
+        }
+        break;
+      }
+
+      case TSDB_DATA_TYPE_BOOL:
+      case TSDB_DATA_TYPE_TINYINT:
+      case TSDB_DATA_TYPE_UTINYINT: {
+        int32_t  mid = dumpedRows >> 1u;
+        int8_t* pts = (int8_t*)pColData->pData;
+        for (int32_t j = 0; j < mid; ++j) {
+          int8_t t = pts[j];
+          pts[j] = pts[dumpedRows - j - 1];
+          pts[dumpedRows - j - 1] = t;
+        }
+        break;
+      }
+
+      case TSDB_DATA_TYPE_SMALLINT:
+      case TSDB_DATA_TYPE_USMALLINT: {
+        int32_t  mid = dumpedRows >> 1u;
+        int16_t* pts = (int16_t*)pColData->pData;
+        for (int32_t j = 0; j < mid; ++j) {
+          int64_t t = pts[j];
+          pts[j] = pts[dumpedRows - j - 1];
+          pts[dumpedRows - j - 1] = t;
+        }
+        break;
+      }
+
+      case TSDB_DATA_TYPE_FLOAT:
+      case TSDB_DATA_TYPE_INT:
+      case TSDB_DATA_TYPE_UINT: {
+        int32_t  mid = dumpedRows >> 1u;
+        int32_t* pts = (int32_t*)pColData->pData;
+        for (int32_t j = 0; j < mid; ++j) {
+          int32_t t = pts[j];
+          pts[j] = pts[dumpedRows - j - 1];
+          pts[dumpedRows - j - 1] = t;
+        }
+        break;
+      }
+    }
+  }
+
+  // 3. if the  null value exists, check items one-by-one
+  if (pData->flag != HAS_VALUE) {
+    int32_t rowIndex = 0;
+
+    for (int32_t j = pDumpInfo->rowIndex; rowIndex < dumpedRows; j += step, rowIndex++) {
+      uint8_t v = tColDataGetBitValue(pData, j);
+      if (v == 0 || v == 1) {
+        colDataSetNull_f(pColData->nullbitmap, rowIndex);
+        pColData->hasNull = true;
+      }
+    }
+  }
+}
+
 static int32_t copyBlockDataToSDataBlock(STsdbReader* pReader, STableBlockScanInfo* pBlockScanInfo) {
   SReaderStatus*  pStatus = &pReader->status;
   SDataBlockIter* pBlockIter = &pStatus->blockIter;
@@ -949,24 +1060,17 @@ static int32_t copyBlockDataToSDataBlock(STsdbReader* pReader, STableBlockScanIn
   }
 
   endIndex += step;
-  int32_t remain = asc ? (endIndex - pDumpInfo->rowIndex) : (pDumpInfo->rowIndex - endIndex);
+  int32_t dumpedRows = asc ? (endIndex - pDumpInfo->rowIndex) : (pDumpInfo->rowIndex - endIndex);
-  if (remain > pReader->capacity) {  // output buffer check
+  if (dumpedRows > pReader->capacity) {  // output buffer check
-    remain = pReader->capacity;
+    dumpedRows = pReader->capacity;
   }
 
+  int32_t i = 0;
   int32_t rowIndex = 0;
 
-  int32_t          i = 0;
   SColumnInfoData* pColData = taosArrayGet(pResBlock->pDataBlock, i);
   if (pColData->info.colId == PRIMARYKEY_TIMESTAMP_COL_ID) {
-    if (asc) {
+    copyPrimaryTsCol(pBlockData, pDumpInfo, pColData, dumpedRows, asc);
-      memcpy(pColData->pData, &pBlockData->aTSKEY[pDumpInfo->rowIndex], remain * sizeof(int64_t));
-    } else {
-      for (int32_t j = pDumpInfo->rowIndex; rowIndex < remain; j += step) {
-        colDataAppendInt64(pColData, rowIndex++, &pBlockData->aTSKEY[j]);
-      }
-    }
-
     i += 1;
   }
 
@@ -981,23 +1085,12 @@ static int32_t copyBlockDataToSDataBlock(STsdbReader* pReader, STableBlockScanIn
       colIndex += 1;
     } else if (pData->cid == pColData->info.colId) {
       if (pData->flag == HAS_NONE || pData->flag == HAS_NULL || pData->flag == (HAS_NULL | HAS_NONE)) {
-        colDataAppendNNULL(pColData, 0, remain);
+        colDataAppendNNULL(pColData, 0, dumpedRows);
       } else {
-        if (IS_NUMERIC_TYPE(pColData->info.type) && asc) {
+        if (IS_MATHABLE_TYPE(pColData->info.type)) {
-          uint8_t* p = pData->pData + tDataTypes[pData->type].bytes * pDumpInfo->rowIndex;
+          copyNumericCols(pData, pDumpInfo, pColData, dumpedRows, asc);
-          memcpy(pColData->pData, p, remain * tDataTypes[pData->type].bytes);
+        } else {  // varchar/nchar type
-
+          for (int32_t j = pDumpInfo->rowIndex; rowIndex < dumpedRows; j += step) {
-          // null value exists, check one-by-one
-          if (pData->flag != HAS_VALUE) {
-            for (int32_t j = pDumpInfo->rowIndex; rowIndex < remain; j += step, rowIndex++) {
-              uint8_t v = tColDataGetBitValue(pData, j);
-              if (v == 0 || v == 1) {
-                colDataSetNull_f(pColData->nullbitmap, rowIndex);
-              }
-            }
-          }
-        } else {
-          for (int32_t j = pDumpInfo->rowIndex; rowIndex < remain; j += step) {
             tColDataGetValue(pData, j, &cv);
             doCopyColVal(pColData, rowIndex++, i, &cv, pSupInfo);
           }
@@ -1007,7 +1100,7 @@ static int32_t copyBlockDataToSDataBlock(STsdbReader* pReader, STableBlockScanIn
       colIndex += 1;
       i += 1;
     } else {  // the specified column does not exist in file block, fill with null data
-      colDataAppendNNULL(pColData, 0, remain);
+      colDataAppendNNULL(pColData, 0, dumpedRows);
       i += 1;
     }
   }
@@ -1015,12 +1108,12 @@ static int32_t copyBlockDataToSDataBlock(STsdbReader* pReader, STableBlockScanIn
   // fill the mis-matched columns with null value
   while (i < numOfOutputCols) {
     pColData = taosArrayGet(pResBlock->pDataBlock, i);
-    colDataAppendNNULL(pColData, 0, remain);
+    colDataAppendNNULL(pColData, 0, dumpedRows);
     i += 1;
   }
 
-  pResBlock->info.rows = remain;
+  pResBlock->info.rows = dumpedRows;
-  pDumpInfo->rowIndex += step * remain;
+  pDumpInfo->rowIndex += step * dumpedRows;
 
   // check if current block are all handled
   if (pDumpInfo->rowIndex >= 0 && pDumpInfo->rowIndex < pBlock->nRow) {
@@ -1039,7 +1132,7 @@ static int32_t copyBlockDataToSDataBlock(STsdbReader* pReader, STableBlockScanIn
   int32_t unDumpedRows = asc ? pBlock->nRow - pDumpInfo->rowIndex : pDumpInfo->rowIndex + 1;
   tsdbDebug("%p copy file block to sdatablock, global index:%d, table index:%d, brange:%" PRId64 "-%" PRId64
             ", rows:%d, remain:%d, minVer:%" PRId64 ", maxVer:%" PRId64 ", elapsed time:%.2f ms, %s",
-            pReader, pBlockIter->index, pBlockInfo->tbBlockIdx, pBlock->minKey.ts, pBlock->maxKey.ts, remain,
+            pReader, pBlockIter->index, pBlockInfo->tbBlockIdx, pBlock->minKey.ts, pBlock->maxKey.ts, dumpedRows,
             unDumpedRows, pBlock->minVer, pBlock->maxVer, elapsedTime, pReader->idStr);
 
   return TSDB_CODE_SUCCESS;

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

@@ -505,8 +505,8 @@ int32_t tsdbKeyFid(TSKEY key, int32_t minutes, int8_t precision) {
 }
 
 void tsdbFidKeyRange(int32_t fid, int32_t minutes, int8_t precision, TSKEY *minKey, TSKEY *maxKey) {
-  *minKey = fid * minutes * tsTickPerMin[precision];
+  *minKey = tsTickPerMin[precision] * fid * minutes;
-  *maxKey = *minKey + minutes * tsTickPerMin[precision] - 1;
+  *maxKey = *minKey + tsTickPerMin[precision] * minutes - 1;
 }
 
 int32_t tsdbFidLevel(int32_t fid, STsdbKeepCfg *pKeepCfg, int64_t now) {
@@ -575,16 +575,6 @@ int32_t tPutTSDBRow(uint8_t *p, TSDBROW *pRow) {
   return n;
 }
 
-int32_t tGetTSDBRow(uint8_t *p, TSDBROW *pRow) {
-  int32_t n = 0;
-
-  n += tGetI64(p, &pRow->version);
-  pRow->pTSRow = (STSRow *)(p + n);
-  n += pRow->pTSRow->len;
-
-  return n;
-}
-
 int32_t tsdbRowCmprFn(const void *p1, const void *p2) {
   return tsdbKeyCmprFn(&TSDBROW_KEY((TSDBROW *)p1), &TSDBROW_KEY((TSDBROW *)p2));
 }
@@ -1025,6 +1015,191 @@ _err:
   return code;
 }
 
+static int32_t tBlockDataAppendBlockRow(SBlockData *pBlockData, SBlockData *pBlockDataFrom, int32_t iRow) {
+  int32_t code = 0;
+
+  SColVal   cv = {0};
+  int32_t   iColDataFrom = 0;
+  SColData *pColDataFrom =
+      (iColDataFrom < pBlockDataFrom->nColData) ? &((SColData *)pBlockDataFrom->aColData->pData)[iColDataFrom] : NULL;
+
+  for (int32_t iColDataTo = 0; iColDataTo < pBlockData->nColData; iColDataTo++) {
+    SColData *pColDataTo = &((SColData *)pBlockData->aColData->pData)[iColDataTo];
+
+    while (pColDataFrom && pColDataFrom->cid < pColDataTo->cid) {
+      iColDataFrom++;
+      pColDataFrom = (iColDataFrom < pBlockDataFrom->nColData)
+                         ? &((SColData *)pBlockDataFrom->aColData->pData)[iColDataFrom]
+                         : NULL;
+    }
+
+    if (pColDataFrom == NULL || pColDataFrom->cid > pColDataTo->cid) {
+      code = tColDataAppendValue(pColDataTo, &COL_VAL_NONE(pColDataTo->cid, pColDataTo->type));
+      if (code) goto _exit;
+    } else {
+      tColDataGetValue(pColDataFrom, iRow, &cv);
+
+      code = tColDataAppendValue(pColDataTo, &cv);
+      if (code) goto _exit;
+
+      iColDataFrom++;
+      pColDataFrom = (iColDataFrom < pBlockDataFrom->nColData)
+                         ? &((SColData *)pBlockDataFrom->aColData->pData)[iColDataFrom]
+                         : NULL;
+    }
+  }
+
+_exit:
+  return code;
+}
+
+static int32_t tBlockDataAppendTPRow(SBlockData *pBlockData, STSRow *pRow, STSchema *pTSchema) {
+  int32_t code = 0;
+
+  int32_t   iTColumn = 1;
+  STColumn *pTColumn = (iTColumn < pTSchema->numOfCols) ? &pTSchema->columns[iTColumn] : NULL;
+  void     *pBitmap = pRow->statis ? tdGetBitmapAddrTp(pRow, pTSchema->flen) : NULL;
+
+  for (int32_t iColData = 0; iColData < pBlockData->nColData; iColData++) {
+    SColData *pColData = &((SColData *)pBlockData->aColData->pData)[iColData];
+
+    while (pTColumn && pTColumn->colId < pColData->cid) {
+      iTColumn++;
+      pTColumn = (iTColumn < pTSchema->numOfCols) ? &pTSchema->columns[iTColumn] : NULL;
+    }
+
+    if (pTColumn == NULL || pTColumn->colId > pColData->cid) {
+      code = tColDataAppendValue(pColData, &COL_VAL_NONE(pColData->cid, pColData->type));
+      if (code) goto _exit;
+    } else {
+      ASSERT(pTColumn->type == pColData->type);
+
+      SColVal cv = {.cid = pTColumn->colId, .type = pTColumn->type};
+
+      if (pRow->statis) {
+        TDRowValT vt = TD_VTYPE_MAX;
+        tdGetBitmapValTypeII(pBitmap, iTColumn - 1, &vt);
+
+        if (vt == TD_VTYPE_NORM) {
+          cv.flag = CV_FLAG_VALUE;
+
+          if (IS_VAR_DATA_TYPE(pTColumn->type)) {
+            void *pData = (char*)pRow + *(int32_t *)(pRow->data + pTColumn->offset - sizeof(TSKEY));
+            cv.value.nData = varDataLen(pData);
+            cv.value.pData = varDataVal(pData);
+          } else {
+            memcpy(&cv.value.val, pRow->data + pTColumn->offset - sizeof(TSKEY), pTColumn->bytes);
+          }
+
+          code = tColDataAppendValue(pColData, &cv);
+          if (code) goto _exit;
+        } else if (vt == TD_VTYPE_NONE) {
+          code = tColDataAppendValue(pColData, &COL_VAL_NONE(pColData->cid, pColData->type));
+          if (code) goto _exit;
+        } else if (vt == TD_VTYPE_NULL) {
+          code = tColDataAppendValue(pColData, &COL_VAL_NULL(pColData->cid, pColData->type));
+          if (code) goto _exit;
+        } else {
+          ASSERT(0);
+        }
+      } else {
+        cv.flag = CV_FLAG_VALUE;
+
+        if (IS_VAR_DATA_TYPE(pTColumn->type)) {
+          void *pData = (char*)pRow + *(int32_t *)(pRow->data + pTColumn->offset - sizeof(TSKEY));
+          cv.value.nData = varDataLen(pData);
+          cv.value.pData = varDataVal(pData);
+        } else {
+          memcpy(&cv.value.val, pRow->data + pTColumn->offset - sizeof(TSKEY), pTColumn->bytes);
+        }
+
+        code = tColDataAppendValue(pColData, &cv);
+        if (code) goto _exit;
+      }
+
+      iTColumn++;
+      pTColumn = (iTColumn < pTSchema->numOfCols) ? &pTSchema->columns[iTColumn] : NULL;
+    }
+  }
+
+_exit:
+  return code;
+}
+
+static int32_t tBlockDataAppendKVRow(SBlockData *pBlockData, STSRow *pRow, STSchema *pTSchema) {
+  int32_t code = 0;
+
+  col_id_t  kvIter = 0;
+  col_id_t  nKvCols = tdRowGetNCols(pRow) - 1;
+  void     *pColIdx = TD_ROW_COL_IDX(pRow);
+  void     *pBitmap = tdGetBitmapAddrKv(pRow, tdRowGetNCols(pRow));
+  int32_t   iTColumn = 1;
+  STColumn *pTColumn = (iTColumn < pTSchema->numOfCols) ? &pTSchema->columns[iTColumn] : NULL;
+
+  for (int32_t iColData = 0; iColData < pBlockData->nColData; iColData++) {
+    SColData *pColData = &((SColData *)pBlockData->aColData->pData)[iColData];
+
+    while (pTColumn && pTColumn->colId < pColData->cid) {
+      iTColumn++;
+      pTColumn = (iTColumn < pTSchema->numOfCols) ? &pTSchema->columns[iTColumn] : NULL;
+    }
+
+    if (pTColumn == NULL || pTColumn->colId > pColData->cid) {
+      code = tColDataAppendValue(pColData, &COL_VAL_NONE(pColData->cid, pColData->type));
+      if (code) goto _exit;
+    } else {
+      ASSERT(pTColumn->type == pColData->type);
+
+      SColVal    cv = {.cid = pTColumn->colId, .type = pTColumn->type};
+      TDRowValT  vt = TD_VTYPE_NONE; // default is NONE
+      SKvRowIdx *pKvIdx = NULL;
+
+      while (kvIter < nKvCols) {
+        pKvIdx = (SKvRowIdx *)POINTER_SHIFT(pColIdx, kvIter * sizeof(SKvRowIdx));
+        if (pKvIdx->colId == pTColumn->colId) {
+          tdGetBitmapValTypeII(pBitmap, kvIter, &vt);
+          ++kvIter;
+          break;
+        } else if (pKvIdx->colId > pTColumn->colId) {
+          vt = TD_VTYPE_NONE;
+          break;
+        } else {
+          ++kvIter;
+        }
+      }
+
+      if (vt == TD_VTYPE_NORM) {
+        cv.flag = CV_FLAG_VALUE;
+
+        void *pData = POINTER_SHIFT(pRow, pKvIdx->offset);
+        if (IS_VAR_DATA_TYPE(pTColumn->type)) {
+          cv.value.nData = varDataLen(pData);
+          cv.value.pData = varDataVal(pData);
+        } else {
+          memcpy(&cv.value.val, pData, pTColumn->bytes);
+        }
+
+        code = tColDataAppendValue(pColData, &cv);
+        if (code) goto _exit;
+      } else if (vt == TD_VTYPE_NONE) {
+        code = tColDataAppendValue(pColData, &COL_VAL_NONE(pColData->cid, pColData->type));
+        if (code) goto _exit;
+      } else if (vt == TD_VTYPE_NULL) {
+        code = tColDataAppendValue(pColData, &COL_VAL_NULL(pColData->cid, pColData->type));
+        if (code) goto _exit;
+      } else {
+        ASSERT(0);
+      }
+
+      iTColumn++;
+      pTColumn = (iTColumn < pTSchema->numOfCols) ? &pTSchema->columns[iTColumn] : NULL;
+    }
+  }
+
+_exit:
+  return code;
+}
+
 int32_t tBlockDataAppendRow(SBlockData *pBlockData, TSDBROW *pRow, STSchema *pTSchema, int64_t uid) {
   int32_t code = 0;
 
@@ -1046,27 +1221,20 @@ int32_t tBlockDataAppendRow(SBlockData *pBlockData, TSDBROW *pRow, STSchema *pTS
   if (code) goto _err;
   pBlockData->aTSKEY[pBlockData->nRow] = TSDBROW_TS(pRow);
 
-  // OTHER
+  SColVal cv = {0};
-  SRowIter rIter = {0};
+  if (pRow->type == 0) {
-  SColVal *pColVal;
+    if (TD_IS_TP_ROW(pRow->pTSRow)) {
-
+      code = tBlockDataAppendTPRow(pBlockData, pRow->pTSRow, pTSchema);
-  tRowIterInit(&rIter, pRow, pTSchema);
+      if (code) goto _err;
-  pColVal = tRowIterNext(&rIter);
+    } else if (TD_IS_KV_ROW(pRow->pTSRow)) {
-  for (int32_t iColData = 0; iColData < pBlockData->nColData; iColData++) {
+      code = tBlockDataAppendKVRow(pBlockData, pRow->pTSRow, pTSchema);
-    SColData *pColData = tBlockDataGetColDataByIdx(pBlockData, iColData);
-
-    while (pColVal && pColVal->cid < pColData->cid) {
-      pColVal = tRowIterNext(&rIter);
-    }
-
-    if (pColVal == NULL || pColVal->cid > pColData->cid) {
-      code = tColDataAppendValue(pColData, &COL_VAL_NONE(pColData->cid, pColData->type));
       if (code) goto _err;
     } else {
-      code = tColDataAppendValue(pColData, pColVal);
+      ASSERT(0);
-      if (code) goto _err;
-      pColVal = tRowIterNext(&rIter);
     }
+  } else {
+    code = tBlockDataAppendBlockRow(pBlockData, pRow->pBlockData, pRow->iRow);
+    if (code) goto _err;
   }
   pBlockData->nRow++;
 

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

@@ -15,6 +15,15 @@
 
 #include "tsdb.h"
 
+/**
+ * @brief max key by precision
+ *  approximately calculation:
+ *  ms: 3600*1000*8765*1000         // 1970 + 1000 years
+ *  us: 3600*1000000*8765*1000      // 1970 + 1000 years
+ *  ns: 3600*1000000000*8765*292    // 1970 + 292 years
+ */
+static int64_t tsMaxKeyByPrecision[] = {31556995200000L, 31556995200000000L, 9214646400000000000L};
+
 // static int tsdbScanAndConvertSubmitMsg(STsdb *pTsdb, SSubmitReq *pMsg);
 
 int tsdbInsertData(STsdb *pTsdb, int64_t version, SSubmitReq *pMsg, SSubmitRsp *pRsp) {
@@ -97,7 +106,7 @@ int tsdbScanAndConvertSubmitMsg(STsdb *pTsdb, SSubmitReq *pMsg) {
   STsdbKeepCfg  *pCfg = &pTsdb->keepCfg;
   TSKEY          now = taosGetTimestamp(pCfg->precision);
   TSKEY          minKey = now - tsTickPerMin[pCfg->precision] * pCfg->keep2;
-  TSKEY          maxKey = now + tsTickPerMin[pCfg->precision] * pCfg->days;
+  TSKEY          maxKey = tsMaxKeyByPrecision[pCfg->precision];
 
   terrno = TSDB_CODE_SUCCESS;
   // pMsg->length = htonl(pMsg->length);

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

@@ -27,10 +27,21 @@ static int vnodeBufPoolCreate(SVnode *pVnode, int64_t size, SVBufPool **ppPool)
     return -1;
   }
 
-  if (taosThreadSpinInit(&pPool->lock, 0) != 0) {
+  if (VND_IS_RSMA(pVnode)) {
-    taosMemoryFree(pPool);
+    pPool->lock = taosMemoryMalloc(sizeof(TdThreadSpinlock));
-    terrno = TAOS_SYSTEM_ERROR(errno);
+    if (!pPool->lock) {
-    return -1;
+      taosMemoryFree(pPool);
+      terrno = TSDB_CODE_OUT_OF_MEMORY;
+      return -1;
+    }
+    if (taosThreadSpinInit(pPool->lock, 0) != 0) {
+      taosMemoryFree((void*)pPool->lock);
+      taosMemoryFree(pPool);
+      terrno = TAOS_SYSTEM_ERROR(errno);
+      return -1;
+    }
+  } else {
+    pPool->lock = NULL;
   }
 
   pPool->next = NULL;
@@ -49,7 +60,10 @@ static int vnodeBufPoolCreate(SVnode *pVnode, int64_t size, SVBufPool **ppPool)
 
 static int vnodeBufPoolDestroy(SVBufPool *pPool) {
   vnodeBufPoolReset(pPool);
-  taosThreadSpinDestroy(&pPool->lock);
+  if (pPool->lock) {
+    taosThreadSpinDestroy(pPool->lock);
+    taosMemoryFree((void*)pPool->lock);
+  }
   taosMemoryFree(pPool);
   return 0;
 }
@@ -114,7 +128,7 @@ void *vnodeBufPoolMalloc(SVBufPool *pPool, int size) {
   void          *p = NULL;
   ASSERT(pPool != NULL);
 
-  taosThreadSpinLock(&pPool->lock);
+  if (pPool->lock) taosThreadSpinLock(pPool->lock);
   if (pPool->node.size >= pPool->ptr - pPool->node.data + size) {
     // allocate from the anchor node
     p = pPool->ptr;
@@ -125,7 +139,7 @@ void *vnodeBufPoolMalloc(SVBufPool *pPool, int size) {
     pNode = taosMemoryMalloc(sizeof(*pNode) + size);
     if (pNode == NULL) {
       terrno = TSDB_CODE_OUT_OF_MEMORY;
-      taosThreadSpinUnlock(&pPool->lock);
+      if (pPool->lock) taosThreadSpinUnlock(pPool->lock);
       return NULL;
     }
 
@@ -138,7 +152,7 @@ void *vnodeBufPoolMalloc(SVBufPool *pPool, int size) {
 
     pPool->size = pPool->size + sizeof(*pNode) + size;
   }
-  taosThreadSpinUnlock(&pPool->lock);
+  if (pPool->lock) taosThreadSpinUnlock(pPool->lock);
   return p;
 }
 

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

@@ -37,6 +37,12 @@ struct SVnodeGlobal vnodeGlobal;
 
 static void* loop(void* arg);
 
+static tsem_t canCommit = {0};
+
+static void vnodeInitCommit() { tsem_init(&canCommit, 0, 4); };
+void        vnode_wait_commit() { tsem_wait(&canCommit); }
+void        vnode_done_commit() { tsem_wait(&canCommit); }
+
 int vnodeInit(int nthreads) {
   int8_t init;
   int    ret;

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

@@ -438,8 +438,24 @@ static void vnodeBecomeLeader(const SSyncFSM *pFsm) {
 
 static bool vnodeApplyQueueEmpty(const SSyncFSM *pFsm) {
   SVnode *pVnode = pFsm->data;
-  int32_t itemSize = tmsgGetQueueSize(&pVnode->msgCb, pVnode->config.vgId, APPLY_QUEUE);
+
-  return (itemSize == 0);
+  if (pVnode != NULL && pVnode->msgCb.qsizeFp != NULL) {
+    int32_t itemSize = tmsgGetQueueSize(&pVnode->msgCb, pVnode->config.vgId, APPLY_QUEUE);
+    return (itemSize == 0);
+  } else {
+    return true;
+  }
+}
+
+static int32_t vnodeApplyQueueItems(const SSyncFSM *pFsm) {
+  SVnode *pVnode = pFsm->data;
+
+  if (pVnode != NULL && pVnode->msgCb.qsizeFp != NULL) {
+    int32_t itemSize = tmsgGetQueueSize(&pVnode->msgCb, pVnode->config.vgId, APPLY_QUEUE);
+    return itemSize;
+  } else {
+    return -1;
+  }
 }
 
 static SSyncFSM *vnodeSyncMakeFsm(SVnode *pVnode) {
@@ -452,6 +468,7 @@ static SSyncFSM *vnodeSyncMakeFsm(SVnode *pVnode) {
   pFsm->FpRestoreFinishCb = vnodeRestoreFinish;
   pFsm->FpLeaderTransferCb = NULL;
   pFsm->FpApplyQueueEmptyCb = vnodeApplyQueueEmpty;
+  pFsm->FpApplyQueueItems = vnodeApplyQueueItems;
   pFsm->FpBecomeLeaderCb = vnodeBecomeLeader;
   pFsm->FpBecomeFollowerCb = vnodeBecomeFollower;
   pFsm->FpReConfigCb = NULL;

source/libs/catalog/src/catalog.c

@@ -715,10 +715,10 @@ _return:
   CTG_API_LEAVE(code);
 }
 
-int32_t catalogGetDBVgVersion(SCatalog* pCtg, const char* dbFName, int32_t* version, int64_t* dbId, int32_t* tableNum) {
+int32_t catalogGetDBVgVersion(SCatalog* pCtg, const char* dbFName, int32_t* version, int64_t* dbId, int32_t* tableNum, int64_t* pStateTs) {
   CTG_API_ENTER();
 
-  if (NULL == pCtg || NULL == dbFName || NULL == version || NULL == dbId) {
+  if (NULL == pCtg || NULL == dbFName || NULL == version || NULL == dbId || NULL == tableNum || NULL == pStateTs) {
     CTG_API_LEAVE(TSDB_CODE_CTG_INVALID_INPUT);
   }
 

source/libs/catalog/src/ctgAsync.c

@@ -1204,11 +1204,12 @@ int32_t ctgHandleGetTbMetasRsp(SCtgTaskReq* tReq, int32_t reqType, const SDataBu
           stbCtx.flag = flag;
           stbCtx.pName = &stbName;
 
-          STableMeta *stbMeta = NULL;
+          STableMeta* stbMeta = NULL;
           ctgReadTbMetaFromCache(pCtg, &stbCtx, &stbMeta);
           if (stbMeta && stbMeta->sversion >= pOut->tbMeta->sversion) {
             ctgDebug("use cached stb meta, tbName:%s", tNameGetTableName(pName));
             exist = 1;
+            taosMemoryFreeClear(stbMeta);
           } else {
             ctgDebug("need to get/update stb meta, tbName:%s", tNameGetTableName(pName));
             taosMemoryFreeClear(pOut->tbMeta);
@@ -1998,6 +1999,7 @@ int32_t ctgLaunchGetDbInfoTask(SCtgTask* pTask) {
     pInfo->vgVer = dbCache->vgCache.vgInfo->vgVersion;
     pInfo->dbId = dbCache->dbId;
     pInfo->tbNum = dbCache->vgCache.vgInfo->numOfTable;
+    pInfo->stateTs = dbCache->vgCache.vgInfo->stateTs;
 
     ctgReleaseVgInfoToCache(pCtg, dbCache);
     dbCache = NULL;

source/libs/catalog/src/ctgCache.c

@@ -72,8 +72,8 @@ void ctgRUnlockVgInfo(SCtgDBCache *dbCache) { CTG_UNLOCK(CTG_READ, &dbCache->vgC
 
 void ctgWUnlockVgInfo(SCtgDBCache *dbCache) { CTG_UNLOCK(CTG_WRITE, &dbCache->vgCache.vgLock); }
 
-void ctgReleaseDBCache(SCatalog *pCtg, SCtgDBCache *dbCache) { 
+void ctgReleaseDBCache(SCatalog *pCtg, SCtgDBCache *dbCache) {
-  CTG_UNLOCK(CTG_READ, &dbCache->dbLock); 
+  CTG_UNLOCK(CTG_READ, &dbCache->dbLock);
   taosHashRelease(pCtg->dbCache, dbCache);
 }
 
@@ -90,7 +90,7 @@ int32_t ctgAcquireDBCacheImpl(SCatalog *pCtg, const char *dbFName, SCtgDBCache *
   } else {
     dbCache = (SCtgDBCache *)taosHashGet(pCtg->dbCache, dbFName, strlen(dbFName));
   }
-  
+
   if (NULL == dbCache) {
     *pCache = NULL;
     ctgDebug("db not in cache, dbFName:%s", dbFName);
@@ -422,6 +422,7 @@ _return:
 
   ctgReleaseTbMetaToCache(pCtg, dbCache, tbCache);
   taosMemoryFreeClear(*pTableMeta);
+  *pTableMeta = NULL;
 
   CTG_RET(code);
 }
@@ -1231,14 +1232,16 @@ int32_t ctgAddNewDBCache(SCatalog *pCtg, const char *dbFName, uint64_t dbId) {
 
   CTG_CACHE_STAT_INC(numOfDb, 1);
 
-  SDbVgVersion vgVersion = {.dbId = newDBCache.dbId, .vgVersion = -1};
+  SDbVgVersion vgVersion = {.dbId = newDBCache.dbId, .vgVersion = -1, .stateTs = 0};
   tstrncpy(vgVersion.dbFName, dbFName, sizeof(vgVersion.dbFName));
 
   ctgDebug("db added to cache, dbFName:%s, dbId:0x%" PRIx64, dbFName, dbId);
 
-  CTG_ERR_RET(ctgMetaRentAdd(&pCtg->dbRent, &vgVersion, dbId, sizeof(SDbVgVersion)));
+  if (!IS_SYS_DBNAME(dbFName)) {
+    CTG_ERR_RET(ctgMetaRentAdd(&pCtg->dbRent, &vgVersion, dbId, sizeof(SDbVgVersion)));
 
-  ctgDebug("db added to rent, dbFName:%s, vgVersion:%d, dbId:0x%" PRIx64, dbFName, vgVersion.vgVersion, dbId);
+    ctgDebug("db added to rent, dbFName:%s, vgVersion:%d, dbId:0x%" PRIx64, dbFName, vgVersion.vgVersion, dbId);
+  }
 
   return TSDB_CODE_SUCCESS;
 
@@ -1434,7 +1437,7 @@ int32_t ctgWriteTbMetaToCache(SCatalog *pCtg, SCtgDBCache *dbCache, char *dbFNam
   if (pCache) {
     CTG_ERR_RET(ctgUpdateRentStbVersion(pCtg, dbFName, tbName, dbId, meta->suid, pCache));
   }
-  
+
   return TSDB_CODE_SUCCESS;
 }
 
@@ -1563,7 +1566,8 @@ int32_t ctgOpUpdateVgroup(SCtgCacheOperation *operation) {
   }
 
   bool         newAdded = false;
-  SDbVgVersion vgVersion = {.dbId = msg->dbId, .vgVersion = dbInfo->vgVersion, .numOfTable = dbInfo->numOfTable};
+  SDbVgVersion vgVersion = {
+      .dbId = msg->dbId, .vgVersion = dbInfo->vgVersion, .numOfTable = dbInfo->numOfTable, .stateTs = dbInfo->stateTs};
 
   SCtgDBCache *dbCache = NULL;
   CTG_ERR_JRET(ctgGetAddDBCache(msg->pCtg, dbFName, msg->dbId, &dbCache));
@@ -1579,15 +1583,17 @@ int32_t ctgOpUpdateVgroup(SCtgCacheOperation *operation) {
     SDBVgInfo *vgInfo = vgCache->vgInfo;
 
     if (dbInfo->vgVersion < vgInfo->vgVersion) {
-      ctgDebug("db vgVer is old, dbFName:%s, vgVer:%d, curVer:%d", dbFName, dbInfo->vgVersion, vgInfo->vgVersion);
+      ctgDebug("db updateVgroup is ignored, dbFName:%s, vgVer:%d, curVer:%d", dbFName, dbInfo->vgVersion,
+               vgInfo->vgVersion);
       ctgWUnlockVgInfo(dbCache);
 
       goto _return;
     }
 
-    if (dbInfo->vgVersion == vgInfo->vgVersion && dbInfo->numOfTable == vgInfo->numOfTable) {
+    if (dbInfo->vgVersion == vgInfo->vgVersion && dbInfo->numOfTable == vgInfo->numOfTable &&
-      ctgDebug("no new db vgVer or numOfTable, dbFName:%s, vgVer:%d, numOfTable:%d", dbFName, dbInfo->vgVersion,
+        dbInfo->stateTs == vgInfo->stateTs) {
-               dbInfo->numOfTable);
+      ctgDebug("no new db vgroup update info, dbFName:%s, vgVer:%d, numOfTable:%d, stateTs:%" PRId64, dbFName,
+               dbInfo->vgVersion, dbInfo->numOfTable, dbInfo->stateTs);
       ctgWUnlockVgInfo(dbCache);
 
       goto _return;
@@ -1599,15 +1605,18 @@ int32_t ctgOpUpdateVgroup(SCtgCacheOperation *operation) {
   vgCache->vgInfo = dbInfo;
   msg->dbInfo = NULL;
 
-  ctgDebug("db vgInfo updated, dbFName:%s, vgVer:%d, dbId:0x%" PRIx64, dbFName, vgVersion.vgVersion, vgVersion.dbId);
+  ctgDebug("db vgInfo updated, dbFName:%s, vgVer:%d, stateTs:%" PRId64 ", dbId:0x%" PRIx64, dbFName,
+           vgVersion.vgVersion, vgVersion.stateTs, vgVersion.dbId);
 
   ctgWUnlockVgInfo(dbCache);
 
   dbCache = NULL;
 
-  tstrncpy(vgVersion.dbFName, dbFName, sizeof(vgVersion.dbFName));
+  if (!IS_SYS_DBNAME(dbFName)) {
-  CTG_ERR_JRET(ctgMetaRentUpdate(&msg->pCtg->dbRent, &vgVersion, vgVersion.dbId, sizeof(SDbVgVersion),
+    tstrncpy(vgVersion.dbFName, dbFName, sizeof(vgVersion.dbFName));
-                                 ctgDbVgVersionSortCompare, ctgDbVgVersionSearchCompare));
+    CTG_ERR_JRET(ctgMetaRentUpdate(&msg->pCtg->dbRent, &vgVersion, vgVersion.dbId, sizeof(SDbVgVersion),
+                                   ctgDbVgVersionSortCompare, ctgDbVgVersionSearchCompare));
+  }
 
 _return:
 
@@ -1688,7 +1697,7 @@ int32_t ctgOpUpdateTbMeta(SCtgCacheOperation *operation) {
   if (pCtg->stopUpdate) {
     goto _return;
   }
-  
+
   if ((!CTG_IS_META_CTABLE(pMeta->metaType)) && NULL == pMeta->tbMeta) {
     ctgError("no valid tbmeta got from meta rsp, dbFName:%s, tbName:%s", pMeta->dbFName, pMeta->tbName);
     CTG_ERR_JRET(TSDB_CODE_CTG_INTERNAL_ERROR);
@@ -1898,7 +1907,7 @@ int32_t ctgOpUpdateEpset(SCtgCacheOperation *operation) {
   int32_t             code = 0;
   SCtgUpdateEpsetMsg *msg = operation->data;
   SCatalog           *pCtg = msg->pCtg;
-  SCtgDBCache *dbCache = NULL;
+  SCtgDBCache        *dbCache = NULL;
 
   if (pCtg->stopUpdate) {
     goto _return;
@@ -2170,7 +2179,6 @@ void *ctgUpdateThreadFunc(void *param) {
     CTG_RT_STAT_INC(numOfOpDequeue, 1);
 
     ctgdShowCacheInfo();
-    ctgdShowClusterCache(pCtg);
   }
 
   qInfo("catalog update thread stopped");
@@ -2206,6 +2214,7 @@ int32_t ctgGetTbMetaFromCache(SCatalog *pCtg, SCtgTbMetaCtx *ctx, STableMeta **p
     }
 
     taosMemoryFreeClear(*pTableMeta);
+    *pTableMeta = NULL;
   }
 
   if (CTG_FLAG_IS_UNKNOWN_STB(ctx->flag)) {

source/libs/catalog/src/ctgDbg.c

@@ -78,7 +78,7 @@ void ctgdUserCallback(SMetaData *pResult, void *param, int32_t code) {
     num = taosArrayGetSize(pResult->pDbInfo);
     for (int32_t i = 0; i < num; ++i) {
       SDbInfo *pDb = taosArrayGet(pResult->pDbInfo, i);
-      qDebug("db %d dbInfo: vgVer:%d, tbNum:%d, dbId:0x%" PRIx64, i, pDb->vgVer, pDb->tbNum, pDb->dbId);
+      qDebug("db %d dbInfo: vgVer:%d, tbNum:%d, stateTs:%" PRId64 " dbId:0x%" PRIx64, i, pDb->vgVer, pDb->tbNum, pDb->stateTs, pDb->dbId);
     }
   } else {
     qDebug("empty db info");
@@ -462,6 +462,7 @@ void ctgdShowDBCache(SCatalog *pCtg, SHashObj *dbHash) {
     int32_t hashMethod = -1;
     int16_t hashPrefix = 0;
     int16_t hashSuffix = 0;
+    int64_t stateTs = 0;
     int32_t vgNum = 0;
 
     if (dbCache->vgCache.vgInfo) {
@@ -469,16 +470,35 @@ void ctgdShowDBCache(SCatalog *pCtg, SHashObj *dbHash) {
       hashMethod = dbCache->vgCache.vgInfo->hashMethod;
       hashPrefix = dbCache->vgCache.vgInfo->hashPrefix;
       hashSuffix = dbCache->vgCache.vgInfo->hashSuffix;
+      stateTs = dbCache->vgCache.vgInfo->stateTs;
       if (dbCache->vgCache.vgInfo->vgHash) {
         vgNum = taosHashGetSize(dbCache->vgCache.vgInfo->vgHash);
       }
     }
 
     ctgDebug("[%d] db [%.*s][0x%" PRIx64
-             "] %s: metaNum:%d, stbNum:%d, vgVersion:%d, hashMethod:%d, prefix:%d, suffix:%d, vgNum:%d",
+             "] %s: metaNum:%d, stbNum:%d, vgVersion:%d, stateTs:%" PRId64 ", hashMethod:%d, prefix:%d, suffix:%d, vgNum:%d",
-             i, (int32_t)len, dbFName, dbCache->dbId, dbCache->deleted ? "deleted" : "", metaNum, stbNum, vgVersion,
+             i, (int32_t)len, dbFName, dbCache->dbId, dbCache->deleted ? "deleted" : "", metaNum, stbNum, vgVersion, stateTs, 
              hashMethod, hashPrefix, hashSuffix, vgNum);
 
+    if (dbCache->vgCache.vgInfo) {
+      int32_t i = 0;
+      void *pVgIter = taosHashIterate(dbCache->vgCache.vgInfo->vgHash, NULL);
+      while (pVgIter) {
+        SVgroupInfo * pVg = (SVgroupInfo *)pVgIter;
+
+        ctgDebug("The %04dth VG [id:%d, hashBegin:%u, hashEnd:%u, numOfTable:%d, epNum:%d, inUse:%d]",
+                 i++, pVg->vgId, pVg->hashBegin, pVg->hashEnd, pVg->numOfTable, pVg->epSet.numOfEps, pVg->epSet.inUse);
+
+        for (int32_t n = 0; n < pVg->epSet.numOfEps; ++n) {
+          SEp *pEp = &pVg->epSet.eps[n];
+          ctgDebug("\tEp %d [fqdn:%s, port:%d]", n, pEp->fqdn, pEp->port);
+        }
+        
+        pVgIter = taosHashIterate(dbCache->vgCache.vgInfo->vgHash, pVgIter);      
+      }
+    }
+
     pIter = taosHashIterate(dbHash, pIter);
   }
 }

source/libs/catalog/test/catalogTests.cpp

@@ -2486,7 +2486,8 @@ TEST(dbVgroup, getSetDbVgroupCase) {
   int32_t dbVer = 0;
   int64_t dbId = 0;
   int32_t tbNum = 0;
-  code = catalogGetDBVgVersion(pCtg, ctgTestDbname, &dbVer, &dbId, &tbNum);
+  int64_t stateTs = 0;
+  code = catalogGetDBVgVersion(pCtg, ctgTestDbname, &dbVer, &dbId, &tbNum, &stateTs);
   ASSERT_EQ(code, 0);
   ASSERT_EQ(dbVer, ctgTestVgVersion);
   ASSERT_EQ(dbId, ctgTestDbId);

source/libs/command/src/command.c

@@ -39,8 +39,7 @@ static int32_t buildRetrieveTableRsp(SSDataBlock* pBlock, int32_t numOfCols, SRe
   (*pRsp)->numOfRows = htonl(pBlock->info.rows);
   (*pRsp)->numOfCols = htonl(numOfCols);
 
-  int32_t len = 0;
+  int32_t len = blockEncode(pBlock, (*pRsp)->data, numOfCols);
-  blockEncode(pBlock, (*pRsp)->data, &len, numOfCols, false);
   ASSERT(len == rspSize - sizeof(SRetrieveTableRsp));
 
   return TSDB_CODE_SUCCESS;

source/libs/command/src/explain.c

@@ -1610,8 +1610,7 @@ int32_t qExplainGetRspFromCtx(void *ctx, SRetrieveTableRsp **pRsp) {
   rsp->completed = 1;
   rsp->numOfRows = htonl(rowNum);
 
-  int32_t len = 0;
+  int32_t len = blockEncode(pBlock, rsp->data, taosArrayGetSize(pBlock->pDataBlock));
-  blockEncode(pBlock, rsp->data, &len, taosArrayGetSize(pBlock->pDataBlock), 0);
   ASSERT(len == rspSize - sizeof(SRetrieveTableRsp));
 
   rsp->compLen = htonl(len);

source/libs/executor/inc/executorimpl.h

@@ -264,6 +264,7 @@ typedef struct SExchangeInfo {
   SLoadRemoteDataInfo loadInfo;
   uint64_t            self;
   SLimitInfo          limitInfo;
+  int64_t             openedTs;         // start exec time stamp
 } SExchangeInfo;
 
 typedef struct SScanInfo {
@@ -841,10 +842,8 @@ typedef struct SJoinOperatorInfo {
 #define OPTR_IS_OPENED(_optr)  (((_optr)->status & OP_OPENED) == OP_OPENED)
 #define OPTR_SET_OPENED(_optr) ((_optr)->status |= OP_OPENED)
 
-void doDestroyExchangeOperatorInfo(void* param);
+SOperatorFpSet createOperatorFpSet(__optr_open_fn_t openFn, __optr_fn_t nextFn, __optr_fn_t cleanup,
-
+                                   __optr_close_fn_t closeFn, __optr_explain_fn_t explain);
-SOperatorFpSet createOperatorFpSet(__optr_open_fn_t openFn, __optr_fn_t nextFn, __optr_fn_t streamFn,
-                                   __optr_fn_t cleanup, __optr_close_fn_t closeFn, __optr_explain_fn_t explain);
 
 int32_t operatorDummyOpenFn(SOperatorInfo* pOperator);
 int32_t appendDownstream(SOperatorInfo* p, SOperatorInfo** pDownstream, int32_t num);
@@ -880,7 +879,11 @@ STimeWindow getFirstQualifiedTimeWindow(int64_t ts, STimeWindow* pWindow, SInter
 int32_t getTableScanInfo(SOperatorInfo* pOperator, int32_t* order, int32_t* scanFlag);
 int32_t getBufferPgSize(int32_t rowSize, uint32_t* defaultPgsz, uint32_t* defaultBufsz);
 
-void    doSetOperatorCompleted(SOperatorInfo* pOperator);
+void    doDestroyExchangeOperatorInfo(void* param);
+
+void    setOperatorCompleted(SOperatorInfo* pOperator);
+void    setOperatorInfo(SOperatorInfo* pOperator, const char* name, int32_t type, bool blocking, int32_t status, void* pInfo,
+                        SExecTaskInfo* pTaskInfo);
 void    doFilter(SSDataBlock* pBlock, SFilterInfo* pFilterInfo, SColMatchInfo* pColMatchInfo);
 int32_t addTagPseudoColumnData(SReadHandle* pHandle, const SExprInfo* pExpr, int32_t numOfExpr,
                                SSDataBlock* pBlock, int32_t rows, const char* idStr, STableMetaCacheInfo * pCache);
@@ -978,6 +981,7 @@ void setTaskKilled(SExecTaskInfo* pTaskInfo);
 void queryCostStatis(SExecTaskInfo* pTaskInfo);
 
 void    doDestroyTask(SExecTaskInfo* pTaskInfo);
+void destroyOperatorInfo(SOperatorInfo* pOperator);
 int32_t getMaximumIdleDurationSec();
 
 /*

source/libs/executor/src/cachescanoperator.c

@@ -93,16 +93,11 @@ SOperatorInfo* createCacherowsScanOperator(SLastRowScanPhysiNode* pScanNode, SRe
     p->pCtx = createSqlFunctionCtx(p->pExprInfo, p->numOfExprs, &p->rowEntryInfoOffset);
   }
 
-  pOperator->name = "LastrowScanOperator";
+  setOperatorInfo(pOperator, "CachedRowScanOperator", QUERY_NODE_PHYSICAL_PLAN_LAST_ROW_SCAN, false, OP_NOT_OPENED, pInfo, pTaskInfo);
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_LAST_ROW_SCAN;
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
-  pOperator->pTaskInfo = pTaskInfo;
   pOperator->exprSupp.numOfExprs = taosArrayGetSize(pInfo->pRes->pDataBlock);
 
   pOperator->fpSet =
-      createOperatorFpSet(operatorDummyOpenFn, doScanCache, NULL, NULL, destroyLastrowScanOperator, NULL);
+      createOperatorFpSet(operatorDummyOpenFn, doScanCache, NULL, destroyLastrowScanOperator, NULL);
 
   pOperator->cost.openCost = 0;
   return pOperator;
@@ -126,7 +121,7 @@ SSDataBlock* doScanCache(SOperatorInfo* pOperator) {
   uint64_t suid = tableListGetSuid(pTableList);
   int32_t  size = tableListGetSize(pTableList);
   if (size == 0) {
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
     return NULL;
   }
 
@@ -182,7 +177,7 @@ SSDataBlock* doScanCache(SOperatorInfo* pOperator) {
       pInfo->indexOfBufferedRes += 1;
       return pRes;
     } else {
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
       return NULL;
     }
   } else {
@@ -234,7 +229,7 @@ SSDataBlock* doScanCache(SOperatorInfo* pOperator) {
       }
     }
 
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
     return NULL;
   }
 }

source/libs/executor/src/dataDispatcher.c

@@ -76,7 +76,7 @@ static void toDataCacheEntry(SDataDispatchHandle* pHandle, const SInputData* pIn
   pEntry->dataLen = 0;
 
   pBuf->useSize = sizeof(SDataCacheEntry);
-  blockEncode(pInput->pData, pEntry->data, &pEntry->dataLen, numOfCols, pEntry->compressed);
+  pEntry->dataLen = blockEncode(pInput->pData, pEntry->data, numOfCols);
   ASSERT(pEntry->numOfRows == *(int32_t*)(pEntry->data + 8));
   ASSERT(pEntry->numOfCols == *(int32_t*)(pEntry->data + 8 + 4));
 

source/libs/executor/src/exchangeoperator.c

@@ -0,0 +1,638 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "filter.h"
+#include "function.h"
+#include "functionMgt.h"
+#include "os.h"
+#include "querynodes.h"
+#include "tfill.h"
+#include "tname.h"
+#include "tref.h"
+
+#include "tdatablock.h"
+#include "tglobal.h"
+#include "tmsg.h"
+#include "tsort.h"
+#include "ttime.h"
+
+#include "executorimpl.h"
+#include "index.h"
+#include "query.h"
+#include "tcompare.h"
+#include "thash.h"
+#include "ttypes.h"
+#include "vnode.h"
+
+typedef struct SFetchRspHandleWrapper {
+  uint32_t exchangeId;
+  int32_t  sourceIndex;
+} SFetchRspHandleWrapper;
+
+static void destroyExchangeOperatorInfo(void* param);
+static void freeBlock(void* pParam);
+static void freeSourceDataInfo(void* param);
+static void* setAllSourcesCompleted(SOperatorInfo* pOperator, int64_t startTs);
+
+static int32_t loadRemoteDataCallback(void* param, SDataBuf* pMsg, int32_t code);
+static int32_t doSendFetchDataRequest(SExchangeInfo* pExchangeInfo, SExecTaskInfo* pTaskInfo, int32_t sourceIndex);
+static int32_t getCompletedSources(const SArray* pArray);
+static int32_t prepareConcurrentlyLoad(SOperatorInfo* pOperator);
+static int32_t seqLoadRemoteData(SOperatorInfo* pOperator);
+static int32_t prepareLoadRemoteData(SOperatorInfo* pOperator);
+
+static void concurrentlyLoadRemoteDataImpl(SOperatorInfo* pOperator, SExchangeInfo* pExchangeInfo,
+                                           SExecTaskInfo* pTaskInfo) {
+  int32_t code = 0;
+  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSourceDataInfo);
+  int32_t completed = getCompletedSources(pExchangeInfo->pSourceDataInfo);
+  if (completed == totalSources) {
+    setAllSourcesCompleted(pOperator, pExchangeInfo->openedTs);
+    return;
+  }
+
+  while (1) {
+    tsem_wait(&pExchangeInfo->ready);
+
+    for (int32_t i = 0; i < totalSources; ++i) {
+      SSourceDataInfo* pDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, i);
+      if (pDataInfo->status == EX_SOURCE_DATA_EXHAUSTED) {
+        continue;
+      }
+
+      if (pDataInfo->status != EX_SOURCE_DATA_READY) {
+        continue;
+      }
+
+      if (pDataInfo->code != TSDB_CODE_SUCCESS) {
+        code = pDataInfo->code;
+        goto _error;
+      }
+
+      SRetrieveTableRsp*     pRsp = pDataInfo->pRsp;
+      SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, i);
+
+      // todo
+      SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
+      if (pRsp->numOfRows == 0) {
+        pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
+        qDebug("%s vgId:%d, taskId:0x%" PRIx64 " execId:%d index:%d completed, rowsOfSource:%" PRIu64
+                   ", totalRows:%" PRIu64 ", try next %d/%" PRIzu,
+               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, i, pDataInfo->totalRows,
+               pExchangeInfo->loadInfo.totalRows, i + 1, totalSources);
+        taosMemoryFreeClear(pDataInfo->pRsp);
+        break;
+      }
+
+      SRetrieveTableRsp* pRetrieveRsp = pDataInfo->pRsp;
+      int32_t            index = 0;
+      char*              pStart = pRetrieveRsp->data;
+      while (index++ < pRetrieveRsp->numOfBlocks) {
+        SSDataBlock* pb = createOneDataBlock(pExchangeInfo->pDummyBlock, false);
+        code = extractDataBlockFromFetchRsp(pb, pStart, NULL, &pStart);
+        if (code != 0) {
+          taosMemoryFreeClear(pDataInfo->pRsp);
+          goto _error;
+        }
+
+        taosArrayPush(pExchangeInfo->pResultBlockList, &pb);
+      }
+
+      updateLoadRemoteInfo(pLoadInfo, pRetrieveRsp->numOfRows, pRetrieveRsp->compLen, pExchangeInfo->openedTs, pOperator);
+
+      if (pRsp->completed == 1) {
+        pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
+        qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64
+                   " execId:%d index:%d completed, blocks:%d, numOfRows:%d, rowsOfSource:%" PRIu64 ", totalRows:%" PRIu64
+                   ", total:%.2f Kb, try next %d/%" PRIzu,
+               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, i, pRsp->numOfBlocks,
+               pRsp->numOfRows, pDataInfo->totalRows, pLoadInfo->totalRows, pLoadInfo->totalSize / 1024.0,
+               i + 1, totalSources);
+      } else {
+        qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64
+                   " execId:%d blocks:%d, numOfRows:%d, totalRows:%" PRIu64 ", total:%.2f Kb",
+               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, pRsp->numOfBlocks,
+               pRsp->numOfRows, pLoadInfo->totalRows, pLoadInfo->totalSize / 1024.0);
+      }
+
+      taosMemoryFreeClear(pDataInfo->pRsp);
+
+      if (pDataInfo->status != EX_SOURCE_DATA_EXHAUSTED) {
+        pDataInfo->status = EX_SOURCE_DATA_NOT_READY;
+        code = doSendFetchDataRequest(pExchangeInfo, pTaskInfo, i);
+        if (code != TSDB_CODE_SUCCESS) {
+          taosMemoryFreeClear(pDataInfo->pRsp);
+          goto _error;
+        }
+      }
+      return;
+    }  // end loop
+
+    int32_t complete1 = getCompletedSources(pExchangeInfo->pSourceDataInfo);
+    if (complete1 == totalSources) {
+      qDebug("all sources are completed, %s", GET_TASKID(pTaskInfo));
+      return;
+    }
+  }
+
+  _error:
+  pTaskInfo->code = code;
+}
+
+static SSDataBlock* doLoadRemoteDataImpl(SOperatorInfo* pOperator) {
+  SExchangeInfo* pExchangeInfo = pOperator->info;
+  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
+
+  pTaskInfo->code = pOperator->fpSet._openFn(pOperator);
+  if (pTaskInfo->code != TSDB_CODE_SUCCESS) {
+    return NULL;
+  }
+
+  size_t totalSources = taosArrayGetSize(pExchangeInfo->pSources);
+
+  SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
+  if (pOperator->status == OP_EXEC_DONE) {
+    qDebug("%s all %" PRIzu " source(s) are exhausted, total rows:%" PRIu64 " bytes:%" PRIu64 ", elapsed:%.2f ms",
+           GET_TASKID(pTaskInfo), totalSources, pLoadInfo->totalRows, pLoadInfo->totalSize,
+           pLoadInfo->totalElapsed / 1000.0);
+    return NULL;
+  }
+
+  size_t size = taosArrayGetSize(pExchangeInfo->pResultBlockList);
+  if (size == 0 || pExchangeInfo->rspBlockIndex >= size) {
+    pExchangeInfo->rspBlockIndex = 0;
+    taosArrayClearEx(pExchangeInfo->pResultBlockList, freeBlock);
+    if (pExchangeInfo->seqLoadData) {
+      seqLoadRemoteData(pOperator);
+    } else {
+      concurrentlyLoadRemoteDataImpl(pOperator, pExchangeInfo, pTaskInfo);
+    }
+
+    if (taosArrayGetSize(pExchangeInfo->pResultBlockList) == 0) {
+      return NULL;
+    }
+  }
+
+  // we have buffered retrieved datablock, return it directly
+  return taosArrayGetP(pExchangeInfo->pResultBlockList, pExchangeInfo->rspBlockIndex++);
+}
+
+static SSDataBlock* doLoadRemoteData(SOperatorInfo* pOperator) {
+  SExchangeInfo* pExchangeInfo = pOperator->info;
+  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
+
+  if (pOperator->status == OP_EXEC_DONE) {
+    return NULL;
+  }
+
+  while (1) {
+    SSDataBlock* pBlock = doLoadRemoteDataImpl(pOperator);
+    if (pBlock == NULL) {
+      return NULL;
+    }
+
+    SLimitInfo* pLimitInfo = &pExchangeInfo->limitInfo;
+    if (hasLimitOffsetInfo(pLimitInfo)) {
+      int32_t status = handleLimitOffset(pOperator, pLimitInfo, pBlock, false);
+      if (status == PROJECT_RETRIEVE_CONTINUE) {
+        continue;
+      } else if (status == PROJECT_RETRIEVE_DONE) {
+        size_t rows = pBlock->info.rows;
+        pExchangeInfo->limitInfo.numOfOutputRows += rows;
+
+        if (rows == 0) {
+          setOperatorCompleted(pOperator);
+          return NULL;
+        } else {
+          return pBlock;
+        }
+      }
+    } else {
+      return pBlock;
+    }
+  }
+}
+
+static int32_t initDataSource(int32_t numOfSources, SExchangeInfo* pInfo, const char* id) {
+  pInfo->pSourceDataInfo = taosArrayInit(numOfSources, sizeof(SSourceDataInfo));
+  if (pInfo->pSourceDataInfo == NULL) {
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+
+  for (int32_t i = 0; i < numOfSources; ++i) {
+    SSourceDataInfo dataInfo = {0};
+    dataInfo.status = EX_SOURCE_DATA_NOT_READY;
+    dataInfo.taskId = id;
+    dataInfo.index = i;
+    SSourceDataInfo* pDs = taosArrayPush(pInfo->pSourceDataInfo, &dataInfo);
+    if (pDs == NULL) {
+      taosArrayDestroy(pInfo->pSourceDataInfo);
+      return TSDB_CODE_OUT_OF_MEMORY;
+    }
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
+static int32_t initExchangeOperator(SExchangePhysiNode* pExNode, SExchangeInfo* pInfo, const char* id) {
+  size_t numOfSources = LIST_LENGTH(pExNode->pSrcEndPoints);
+
+  if (numOfSources == 0) {
+    qError("%s invalid number: %d of sources in exchange operator", id, (int32_t)numOfSources);
+    return TSDB_CODE_INVALID_PARA;
+  }
+
+  pInfo->pSources = taosArrayInit(numOfSources, sizeof(SDownstreamSourceNode));
+  if (pInfo->pSources == NULL) {
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+
+  for (int32_t i = 0; i < numOfSources; ++i) {
+    SDownstreamSourceNode* pNode = (SDownstreamSourceNode*)nodesListGetNode((SNodeList*)pExNode->pSrcEndPoints, i);
+    taosArrayPush(pInfo->pSources, pNode);
+  }
+
+  initLimitInfo(pExNode->node.pLimit, pExNode->node.pSlimit, &pInfo->limitInfo);
+  pInfo->self = taosAddRef(exchangeObjRefPool, pInfo);
+
+  return initDataSource(numOfSources, pInfo, id);
+}
+
+SOperatorInfo* createExchangeOperatorInfo(void* pTransporter, SExchangePhysiNode* pExNode, SExecTaskInfo* pTaskInfo) {
+  SExchangeInfo* pInfo = taosMemoryCalloc(1, sizeof(SExchangeInfo));
+  SOperatorInfo* pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
+  if (pInfo == NULL || pOperator == NULL) {
+    goto _error;
+  }
+
+  int32_t code = initExchangeOperator(pExNode, pInfo, GET_TASKID(pTaskInfo));
+  if (code != TSDB_CODE_SUCCESS) {
+    goto _error;
+  }
+
+  tsem_init(&pInfo->ready, 0, 0);
+  pInfo->pDummyBlock = createResDataBlock(pExNode->node.pOutputDataBlockDesc);
+  pInfo->pResultBlockList = taosArrayInit(1, POINTER_BYTES);
+
+  pInfo->seqLoadData = false;
+  pInfo->pTransporter = pTransporter;
+
+  setOperatorInfo(pOperator, "ExchangeOperator", QUERY_NODE_PHYSICAL_PLAN_EXCHANGE, false, OP_NOT_OPENED, pInfo, pTaskInfo);
+  pOperator->exprSupp.numOfExprs = taosArrayGetSize(pInfo->pDummyBlock->pDataBlock);
+
+  pOperator->fpSet = createOperatorFpSet(prepareLoadRemoteData, doLoadRemoteData, NULL, destroyExchangeOperatorInfo, NULL);
+  return pOperator;
+
+  _error:
+  if (pInfo != NULL) {
+    doDestroyExchangeOperatorInfo(pInfo);
+  }
+
+  taosMemoryFreeClear(pOperator);
+  pTaskInfo->code = code;
+  return NULL;
+}
+
+void destroyExchangeOperatorInfo(void* param) {
+  SExchangeInfo* pExInfo = (SExchangeInfo*)param;
+  taosRemoveRef(exchangeObjRefPool, pExInfo->self);
+}
+
+void freeBlock(void* pParam) {
+  SSDataBlock* pBlock = *(SSDataBlock**)pParam;
+  blockDataDestroy(pBlock);
+}
+
+void freeSourceDataInfo(void* p) {
+  SSourceDataInfo* pInfo = (SSourceDataInfo*)p;
+  taosMemoryFreeClear(pInfo->pRsp);
+}
+
+void doDestroyExchangeOperatorInfo(void* param) {
+  SExchangeInfo* pExInfo = (SExchangeInfo*)param;
+
+  taosArrayDestroy(pExInfo->pSources);
+  taosArrayDestroyEx(pExInfo->pSourceDataInfo, freeSourceDataInfo);
+
+  if (pExInfo->pResultBlockList != NULL) {
+    taosArrayDestroyEx(pExInfo->pResultBlockList, freeBlock);
+    pExInfo->pResultBlockList = NULL;
+  }
+
+  blockDataDestroy(pExInfo->pDummyBlock);
+
+  tsem_destroy(&pExInfo->ready);
+  taosMemoryFreeClear(param);
+}
+
+int32_t loadRemoteDataCallback(void* param, SDataBuf* pMsg, int32_t code) {
+  SFetchRspHandleWrapper* pWrapper = (SFetchRspHandleWrapper*)param;
+
+  SExchangeInfo* pExchangeInfo = taosAcquireRef(exchangeObjRefPool, pWrapper->exchangeId);
+  if (pExchangeInfo == NULL) {
+    qWarn("failed to acquire exchange operator, since it may have been released");
+    taosMemoryFree(pMsg->pData);
+    return TSDB_CODE_SUCCESS;
+  }
+
+  int32_t          index = pWrapper->sourceIndex;
+  SSourceDataInfo* pSourceDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, index);
+
+  if (code == TSDB_CODE_SUCCESS) {
+    pSourceDataInfo->pRsp = pMsg->pData;
+
+    SRetrieveTableRsp* pRsp = pSourceDataInfo->pRsp;
+    pRsp->numOfRows = htonl(pRsp->numOfRows);
+    pRsp->compLen = htonl(pRsp->compLen);
+    pRsp->numOfCols = htonl(pRsp->numOfCols);
+    pRsp->useconds = htobe64(pRsp->useconds);
+    pRsp->numOfBlocks = htonl(pRsp->numOfBlocks);
+
+    ASSERT(pRsp != NULL);
+    qDebug("%s fetch rsp received, index:%d, blocks:%d, rows:%d", pSourceDataInfo->taskId, index, pRsp->numOfBlocks,
+           pRsp->numOfRows);
+  } else {
+    taosMemoryFree(pMsg->pData);
+    pSourceDataInfo->code = code;
+    qDebug("%s fetch rsp received, index:%d, error:%s", pSourceDataInfo->taskId, index, tstrerror(code));
+  }
+
+  pSourceDataInfo->status = EX_SOURCE_DATA_READY;
+
+  tsem_post(&pExchangeInfo->ready);
+  taosReleaseRef(exchangeObjRefPool, pWrapper->exchangeId);
+
+  return TSDB_CODE_SUCCESS;
+}
+
+int32_t doSendFetchDataRequest(SExchangeInfo* pExchangeInfo, SExecTaskInfo* pTaskInfo, int32_t sourceIndex) {
+  size_t totalSources = taosArrayGetSize(pExchangeInfo->pSources);
+
+  SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, sourceIndex);
+  SSourceDataInfo*       pDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, sourceIndex);
+
+  ASSERT(pDataInfo->status == EX_SOURCE_DATA_NOT_READY);
+
+  SFetchRspHandleWrapper* pWrapper = taosMemoryCalloc(1, sizeof(SFetchRspHandleWrapper));
+  pWrapper->exchangeId = pExchangeInfo->self;
+  pWrapper->sourceIndex = sourceIndex;
+
+  if (pSource->localExec) {
+    SDataBuf pBuf = {0};
+    int32_t  code =
+        (*pTaskInfo->localFetch.fp)(pTaskInfo->localFetch.handle, pSource->schedId, pTaskInfo->id.queryId,
+                                    pSource->taskId, 0, pSource->execId, &pBuf.pData, pTaskInfo->localFetch.explainRes);
+    loadRemoteDataCallback(pWrapper, &pBuf, code);
+    taosMemoryFree(pWrapper);
+  } else {
+    SResFetchReq* pMsg = taosMemoryCalloc(1, sizeof(SResFetchReq));
+    if (NULL == pMsg) {
+      pTaskInfo->code = TSDB_CODE_QRY_OUT_OF_MEMORY;
+      taosMemoryFree(pWrapper);
+      return pTaskInfo->code;
+    }
+
+    qDebug("%s build fetch msg and send to vgId:%d, ep:%s, taskId:0x%" PRIx64 ", execId:%d, %d/%" PRIzu,
+           GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->addr.epSet.eps[0].fqdn, pSource->taskId,
+           pSource->execId, sourceIndex, totalSources);
+
+    pMsg->header.vgId = htonl(pSource->addr.nodeId);
+    pMsg->sId = htobe64(pSource->schedId);
+    pMsg->taskId = htobe64(pSource->taskId);
+    pMsg->queryId = htobe64(pTaskInfo->id.queryId);
+    pMsg->execId = htonl(pSource->execId);
+
+    // send the fetch remote task result reques
+    SMsgSendInfo* pMsgSendInfo = taosMemoryCalloc(1, sizeof(SMsgSendInfo));
+    if (NULL == pMsgSendInfo) {
+      taosMemoryFreeClear(pMsg);
+      taosMemoryFree(pWrapper);
+      qError("%s prepare message %d failed", GET_TASKID(pTaskInfo), (int32_t)sizeof(SMsgSendInfo));
+      pTaskInfo->code = TSDB_CODE_QRY_OUT_OF_MEMORY;
+      return pTaskInfo->code;
+    }
+
+    pMsgSendInfo->param = pWrapper;
+    pMsgSendInfo->paramFreeFp = taosMemoryFree;
+    pMsgSendInfo->msgInfo.pData = pMsg;
+    pMsgSendInfo->msgInfo.len = sizeof(SResFetchReq);
+    pMsgSendInfo->msgType = pSource->fetchMsgType;
+    pMsgSendInfo->fp = loadRemoteDataCallback;
+
+    int64_t transporterId = 0;
+    int32_t code =
+        asyncSendMsgToServer(pExchangeInfo->pTransporter, &pSource->addr.epSet, &transporterId, pMsgSendInfo);
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
+void updateLoadRemoteInfo(SLoadRemoteDataInfo* pInfo, int32_t numOfRows, int32_t dataLen, int64_t startTs,
+                          SOperatorInfo* pOperator) {
+  pInfo->totalRows += numOfRows;
+  pInfo->totalSize += dataLen;
+  pInfo->totalElapsed += (taosGetTimestampUs() - startTs);
+  pOperator->resultInfo.totalRows += numOfRows;
+}
+
+int32_t extractDataBlockFromFetchRsp(SSDataBlock* pRes, char* pData, SArray* pColList, char** pNextStart) {
+  if (pColList == NULL) {  // data from other sources
+    blockDataCleanup(pRes);
+    *pNextStart = (char*)blockDecode(pRes, pData);
+  } else {  // extract data according to pColList
+    char* pStart = pData;
+
+    int32_t numOfCols = htonl(*(int32_t*)pStart);
+    pStart += sizeof(int32_t);
+
+    // todo refactor:extract method
+    SSysTableSchema* pSchema = (SSysTableSchema*)pStart;
+    for (int32_t i = 0; i < numOfCols; ++i) {
+      SSysTableSchema* p = (SSysTableSchema*)pStart;
+
+      p->colId = htons(p->colId);
+      p->bytes = htonl(p->bytes);
+      pStart += sizeof(SSysTableSchema);
+    }
+
+    SSDataBlock* pBlock = createDataBlock();
+    for (int32_t i = 0; i < numOfCols; ++i) {
+      SColumnInfoData idata = createColumnInfoData(pSchema[i].type, pSchema[i].bytes, pSchema[i].colId);
+      blockDataAppendColInfo(pBlock, &idata);
+    }
+
+    blockDecode(pBlock, pStart);
+    blockDataEnsureCapacity(pRes, pBlock->info.rows);
+
+    // data from mnode
+    pRes->info.rows = pBlock->info.rows;
+    relocateColumnData(pRes, pColList, pBlock->pDataBlock, false);
+    blockDataDestroy(pBlock);
+  }
+
+  // todo move this to time window aggregator, since the primary timestamp may not be known by exchange operator.
+  blockDataUpdateTsWindow(pRes, 0);
+  return TSDB_CODE_SUCCESS;
+}
+
+void* setAllSourcesCompleted(SOperatorInfo* pOperator, int64_t startTs) {
+  SExchangeInfo* pExchangeInfo = pOperator->info;
+  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
+
+  int64_t              el = taosGetTimestampUs() - startTs;
+  SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
+
+  pLoadInfo->totalElapsed += el;
+
+  size_t totalSources = taosArrayGetSize(pExchangeInfo->pSources);
+  qDebug("%s all %" PRIzu " sources are exhausted, total rows: %" PRIu64 " bytes:%" PRIu64 ", elapsed:%.2f ms",
+         GET_TASKID(pTaskInfo), totalSources, pLoadInfo->totalRows, pLoadInfo->totalSize,
+         pLoadInfo->totalElapsed / 1000.0);
+
+  setOperatorCompleted(pOperator);
+  return NULL;
+}
+
+int32_t getCompletedSources(const SArray* pArray) {
+  size_t total = taosArrayGetSize(pArray);
+
+  int32_t completed = 0;
+  for (int32_t k = 0; k < total; ++k) {
+    SSourceDataInfo* p = taosArrayGet(pArray, k);
+    if (p->status == EX_SOURCE_DATA_EXHAUSTED) {
+      completed += 1;
+    }
+  }
+
+  return completed;
+}
+
+int32_t prepareConcurrentlyLoad(SOperatorInfo* pOperator) {
+  SExchangeInfo* pExchangeInfo = pOperator->info;
+  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
+
+  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
+  int64_t startTs = taosGetTimestampUs();
+
+  // Asynchronously send all fetch requests to all sources.
+  for (int32_t i = 0; i < totalSources; ++i) {
+    int32_t code = doSendFetchDataRequest(pExchangeInfo, pTaskInfo, i);
+    if (code != TSDB_CODE_SUCCESS) {
+      pTaskInfo->code = code;
+      return code;
+    }
+  }
+
+  int64_t endTs = taosGetTimestampUs();
+  qDebug("%s send all fetch requests to %" PRIzu " sources completed, elapsed:%.2fms", GET_TASKID(pTaskInfo),
+         totalSources, (endTs - startTs) / 1000.0);
+
+  pOperator->status = OP_RES_TO_RETURN;
+  pOperator->cost.openCost = taosGetTimestampUs() - startTs;
+
+  tsem_wait(&pExchangeInfo->ready);
+  tsem_post(&pExchangeInfo->ready);
+  return TSDB_CODE_SUCCESS;
+}
+
+int32_t seqLoadRemoteData(SOperatorInfo* pOperator) {
+  SExchangeInfo* pExchangeInfo = pOperator->info;
+  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
+
+  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
+  int64_t startTs = taosGetTimestampUs();
+
+  while (1) {
+    if (pExchangeInfo->current >= totalSources) {
+      setAllSourcesCompleted(pOperator, startTs);
+      return TSDB_CODE_SUCCESS;
+    }
+
+    doSendFetchDataRequest(pExchangeInfo, pTaskInfo, pExchangeInfo->current);
+    tsem_wait(&pExchangeInfo->ready);
+
+    SSourceDataInfo*       pDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, pExchangeInfo->current);
+    SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, pExchangeInfo->current);
+
+    if (pDataInfo->code != TSDB_CODE_SUCCESS) {
+      qError("%s vgId:%d, taskID:0x%" PRIx64 " execId:%d error happens, code:%s", GET_TASKID(pTaskInfo),
+             pSource->addr.nodeId, pSource->taskId, pSource->execId, tstrerror(pDataInfo->code));
+      pOperator->pTaskInfo->code = pDataInfo->code;
+      return pOperator->pTaskInfo->code;
+    }
+
+    SRetrieveTableRsp*   pRsp = pDataInfo->pRsp;
+    SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
+    if (pRsp->numOfRows == 0) {
+      qDebug("%s vgId:%d, taskID:0x%" PRIx64 " execId:%d %d of total completed, rowsOfSource:%" PRIu64
+                 ", totalRows:%" PRIu64 " try next",
+             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, pExchangeInfo->current + 1,
+             pDataInfo->totalRows, pLoadInfo->totalRows);
+
+      pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
+      pExchangeInfo->current += 1;
+      taosMemoryFreeClear(pDataInfo->pRsp);
+      continue;
+    }
+
+    SRetrieveTableRsp* pRetrieveRsp = pDataInfo->pRsp;
+
+    char*   pStart = pRetrieveRsp->data;
+    int32_t code = extractDataBlockFromFetchRsp(NULL, pStart, NULL, &pStart);
+
+    if (pRsp->completed == 1) {
+      qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " execId:%d numOfRows:%d, rowsOfSource:%" PRIu64
+                 ", totalRows:%" PRIu64 ", totalBytes:%" PRIu64 " try next %d/%" PRIzu,
+             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, pRetrieveRsp->numOfRows,
+             pDataInfo->totalRows, pLoadInfo->totalRows, pLoadInfo->totalSize, pExchangeInfo->current + 1,
+             totalSources);
+
+      pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
+      pExchangeInfo->current += 1;
+    } else {
+      qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " execId:%d numOfRows:%d, totalRows:%" PRIu64
+                 ", totalBytes:%" PRIu64,
+             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, pRetrieveRsp->numOfRows,
+             pLoadInfo->totalRows, pLoadInfo->totalSize);
+    }
+
+    updateLoadRemoteInfo(pLoadInfo, pRetrieveRsp->numOfRows, pRetrieveRsp->compLen, startTs, pOperator);
+    pDataInfo->totalRows += pRetrieveRsp->numOfRows;
+
+    taosMemoryFreeClear(pDataInfo->pRsp);
+    return TSDB_CODE_SUCCESS;
+  }
+}
+
+int32_t prepareLoadRemoteData(SOperatorInfo* pOperator) {
+  if (OPTR_IS_OPENED(pOperator)) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+  int64_t st = taosGetTimestampUs();
+
+  SExchangeInfo* pExchangeInfo = pOperator->info;
+  if (!pExchangeInfo->seqLoadData) {
+    int32_t code = prepareConcurrentlyLoad(pOperator);
+    if (code != TSDB_CODE_SUCCESS) {
+      return code;
+    }
+    pExchangeInfo->openedTs = taosGetTimestampUs();
+  }
+
+  OPTR_SET_OPENED(pOperator);
+  pOperator->cost.openCost = (taosGetTimestampUs() - st) / 1000.0;
+  return TSDB_CODE_SUCCESS;
+}

source/libs/executor/src/executor.c

@@ -1106,3 +1106,24 @@ int32_t qStreamPrepareScan(qTaskInfo_t tinfo, STqOffsetVal* pOffset, int8_t subT
   }
   return 0;
 }
+
+void qProcessRspMsg(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet) {
+  SMsgSendInfo* pSendInfo = (SMsgSendInfo*)pMsg->info.ahandle;
+  assert(pMsg->info.ahandle != NULL);
+
+  SDataBuf buf = {.len = pMsg->contLen, .pData = NULL};
+
+  if (pMsg->contLen > 0) {
+    buf.pData = taosMemoryCalloc(1, pMsg->contLen);
+    if (buf.pData == NULL) {
+      terrno = TSDB_CODE_OUT_OF_MEMORY;
+      pMsg->code = TSDB_CODE_OUT_OF_MEMORY;
+    } else {
+      memcpy(buf.pData, pMsg->pCont, pMsg->contLen);
+    }
+  }
+
+  pSendInfo->fp(pSendInfo->param, &buf, pMsg->code);
+  rpcFreeCont(pMsg->pCont);
+  destroySendMsgInfo(pSendInfo);
+}

source/libs/executor/src/executorimpl.c

@@ -32,7 +32,6 @@
 #include "index.h"
 #include "query.h"
 #include "tcompare.h"
-#include "tcompression.h"
 #include "thash.h"
 #include "ttypes.h"
 #include "vnode.h"
@@ -91,11 +90,8 @@ static void destroySortOperatorInfo(void* param);
 static void destroyAggOperatorInfo(void* param);
 
 static void destroyIntervalOperatorInfo(void* param);
-static void destroyExchangeOperatorInfo(void* param);
 
-static void destroyOperatorInfo(SOperatorInfo* pOperator);
+void setOperatorCompleted(SOperatorInfo* pOperator) {
-
-void doSetOperatorCompleted(SOperatorInfo* pOperator) {
   pOperator->status = OP_EXEC_DONE;
   ASSERT(pOperator->pTaskInfo != NULL);
 
@@ -103,14 +99,24 @@ void doSetOperatorCompleted(SOperatorInfo* pOperator) {
   setTaskStatus(pOperator->pTaskInfo, TASK_COMPLETED);
 }
 
+void setOperatorInfo(SOperatorInfo* pOperator, const char* name, int32_t type, bool blocking, int32_t status,
+                     void* pInfo, SExecTaskInfo* pTaskInfo) {
+  pOperator->name = (char*)name;
+  pOperator->operatorType = type;
+  pOperator->blocking = blocking;
+  pOperator->status = status;
+  pOperator->info = pInfo;
+  pOperator->pTaskInfo = pTaskInfo;
+}
+
 int32_t operatorDummyOpenFn(SOperatorInfo* pOperator) {
   OPTR_SET_OPENED(pOperator);
   pOperator->cost.openCost = 0;
   return TSDB_CODE_SUCCESS;
 }
 
-SOperatorFpSet createOperatorFpSet(__optr_open_fn_t openFn, __optr_fn_t nextFn, __optr_fn_t streamFn,
+SOperatorFpSet createOperatorFpSet(__optr_open_fn_t openFn, __optr_fn_t nextFn, __optr_fn_t cleanup,
-                                   __optr_fn_t cleanup, __optr_close_fn_t closeFn, __optr_explain_fn_t explain) {
+                                   __optr_close_fn_t closeFn, __optr_explain_fn_t explain) {
   SOperatorFpSet fpSet = {
       ._openFn = openFn,
       .getNextFn = nextFn,
@@ -1652,623 +1658,6 @@ int32_t appendDownstream(SOperatorInfo* p, SOperatorInfo** pDownstream, int32_t
   return TSDB_CODE_SUCCESS;
 }
 
-typedef struct SFetchRspHandleWrapper {
-  uint32_t exchangeId;
-  int32_t  sourceIndex;
-} SFetchRspHandleWrapper;
-
-int32_t loadRemoteDataCallback(void* param, SDataBuf* pMsg, int32_t code) {
-  SFetchRspHandleWrapper* pWrapper = (SFetchRspHandleWrapper*)param;
-
-  SExchangeInfo* pExchangeInfo = taosAcquireRef(exchangeObjRefPool, pWrapper->exchangeId);
-  if (pExchangeInfo == NULL) {
-    qWarn("failed to acquire exchange operator, since it may have been released");
-    taosMemoryFree(pMsg->pData);
-    return TSDB_CODE_SUCCESS;
-  }
-
-  int32_t          index = pWrapper->sourceIndex;
-  SSourceDataInfo* pSourceDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, index);
-
-  if (code == TSDB_CODE_SUCCESS) {
-    pSourceDataInfo->pRsp = pMsg->pData;
-
-    SRetrieveTableRsp* pRsp = pSourceDataInfo->pRsp;
-    pRsp->numOfRows = htonl(pRsp->numOfRows);
-    pRsp->compLen = htonl(pRsp->compLen);
-    pRsp->numOfCols = htonl(pRsp->numOfCols);
-    pRsp->useconds = htobe64(pRsp->useconds);
-    pRsp->numOfBlocks = htonl(pRsp->numOfBlocks);
-
-    ASSERT(pRsp != NULL);
-    qDebug("%s fetch rsp received, index:%d, blocks:%d, rows:%d", pSourceDataInfo->taskId, index, pRsp->numOfBlocks,
-           pRsp->numOfRows);
-  } else {
-    taosMemoryFree(pMsg->pData);
-    pSourceDataInfo->code = code;
-    qDebug("%s fetch rsp received, index:%d, error:%s", pSourceDataInfo->taskId, index, tstrerror(code));
-  }
-
-  pSourceDataInfo->status = EX_SOURCE_DATA_READY;
-
-  tsem_post(&pExchangeInfo->ready);
-  taosReleaseRef(exchangeObjRefPool, pWrapper->exchangeId);
-
-  return TSDB_CODE_SUCCESS;
-}
-
-void qProcessRspMsg(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet) {
-  SMsgSendInfo* pSendInfo = (SMsgSendInfo*)pMsg->info.ahandle;
-  assert(pMsg->info.ahandle != NULL);
-
-  SDataBuf buf = {.len = pMsg->contLen, .pData = NULL};
-
-  if (pMsg->contLen > 0) {
-    buf.pData = taosMemoryCalloc(1, pMsg->contLen);
-    if (buf.pData == NULL) {
-      terrno = TSDB_CODE_OUT_OF_MEMORY;
-      pMsg->code = TSDB_CODE_OUT_OF_MEMORY;
-    } else {
-      memcpy(buf.pData, pMsg->pCont, pMsg->contLen);
-    }
-  }
-
-  pSendInfo->fp(pSendInfo->param, &buf, pMsg->code);
-  rpcFreeCont(pMsg->pCont);
-  destroySendMsgInfo(pSendInfo);
-}
-
-static int32_t doSendFetchDataRequest(SExchangeInfo* pExchangeInfo, SExecTaskInfo* pTaskInfo, int32_t sourceIndex) {
-  size_t totalSources = taosArrayGetSize(pExchangeInfo->pSources);
-
-  SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, sourceIndex);
-  SSourceDataInfo*       pDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, sourceIndex);
-
-  ASSERT(pDataInfo->status == EX_SOURCE_DATA_NOT_READY);
-
-  SFetchRspHandleWrapper* pWrapper = taosMemoryCalloc(1, sizeof(SFetchRspHandleWrapper));
-  pWrapper->exchangeId = pExchangeInfo->self;
-  pWrapper->sourceIndex = sourceIndex;
-
-  if (pSource->localExec) {
-    SDataBuf pBuf = {0};
-    int32_t  code =
-        (*pTaskInfo->localFetch.fp)(pTaskInfo->localFetch.handle, pSource->schedId, pTaskInfo->id.queryId,
-                                    pSource->taskId, 0, pSource->execId, &pBuf.pData, pTaskInfo->localFetch.explainRes);
-    loadRemoteDataCallback(pWrapper, &pBuf, code);
-    taosMemoryFree(pWrapper);
-  } else {
-    SResFetchReq* pMsg = taosMemoryCalloc(1, sizeof(SResFetchReq));
-    if (NULL == pMsg) {
-      pTaskInfo->code = TSDB_CODE_QRY_OUT_OF_MEMORY;
-      taosMemoryFree(pWrapper);
-      return pTaskInfo->code;
-    }
-
-    qDebug("%s build fetch msg and send to vgId:%d, ep:%s, taskId:0x%" PRIx64 ", execId:%d, %d/%" PRIzu,
-           GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->addr.epSet.eps[0].fqdn, pSource->taskId,
-           pSource->execId, sourceIndex, totalSources);
-
-    pMsg->header.vgId = htonl(pSource->addr.nodeId);
-    pMsg->sId = htobe64(pSource->schedId);
-    pMsg->taskId = htobe64(pSource->taskId);
-    pMsg->queryId = htobe64(pTaskInfo->id.queryId);
-    pMsg->execId = htonl(pSource->execId);
-
-    // send the fetch remote task result reques
-    SMsgSendInfo* pMsgSendInfo = taosMemoryCalloc(1, sizeof(SMsgSendInfo));
-    if (NULL == pMsgSendInfo) {
-      taosMemoryFreeClear(pMsg);
-      taosMemoryFree(pWrapper);
-      qError("%s prepare message %d failed", GET_TASKID(pTaskInfo), (int32_t)sizeof(SMsgSendInfo));
-      pTaskInfo->code = TSDB_CODE_QRY_OUT_OF_MEMORY;
-      return pTaskInfo->code;
-    }
-
-    pMsgSendInfo->param = pWrapper;
-    pMsgSendInfo->paramFreeFp = taosMemoryFree;
-    pMsgSendInfo->msgInfo.pData = pMsg;
-    pMsgSendInfo->msgInfo.len = sizeof(SResFetchReq);
-    pMsgSendInfo->msgType = pSource->fetchMsgType;
-    pMsgSendInfo->fp = loadRemoteDataCallback;
-
-    int64_t transporterId = 0;
-    int32_t code =
-        asyncSendMsgToServer(pExchangeInfo->pTransporter, &pSource->addr.epSet, &transporterId, pMsgSendInfo);
-  }
-
-  return TSDB_CODE_SUCCESS;
-}
-
-void updateLoadRemoteInfo(SLoadRemoteDataInfo* pInfo, int32_t numOfRows, int32_t dataLen, int64_t startTs,
-                          SOperatorInfo* pOperator) {
-  pInfo->totalRows += numOfRows;
-  pInfo->totalSize += dataLen;
-  pInfo->totalElapsed += (taosGetTimestampUs() - startTs);
-  pOperator->resultInfo.totalRows += numOfRows;
-}
-
-int32_t extractDataBlockFromFetchRsp(SSDataBlock* pRes, char* pData, SArray* pColList, char** pNextStart) {
-  if (pColList == NULL) {  // data from other sources
-    blockDataCleanup(pRes);
-    *pNextStart = (char*)blockDecode(pRes, pData);
-  } else {  // extract data according to pColList
-    char* pStart = pData;
-
-    int32_t numOfCols = htonl(*(int32_t*)pStart);
-    pStart += sizeof(int32_t);
-
-    // todo refactor:extract method
-    SSysTableSchema* pSchema = (SSysTableSchema*)pStart;
-    for (int32_t i = 0; i < numOfCols; ++i) {
-      SSysTableSchema* p = (SSysTableSchema*)pStart;
-
-      p->colId = htons(p->colId);
-      p->bytes = htonl(p->bytes);
-      pStart += sizeof(SSysTableSchema);
-    }
-
-    SSDataBlock* pBlock = createDataBlock();
-    for (int32_t i = 0; i < numOfCols; ++i) {
-      SColumnInfoData idata = createColumnInfoData(pSchema[i].type, pSchema[i].bytes, pSchema[i].colId);
-      blockDataAppendColInfo(pBlock, &idata);
-    }
-
-    blockDecode(pBlock, pStart);
-    blockDataEnsureCapacity(pRes, pBlock->info.rows);
-
-    // data from mnode
-    pRes->info.rows = pBlock->info.rows;
-    relocateColumnData(pRes, pColList, pBlock->pDataBlock, false);
-    blockDataDestroy(pBlock);
-  }
-
-  // todo move this to time window aggregator, since the primary timestamp may not be known by exchange operator.
-  blockDataUpdateTsWindow(pRes, 0);
-  return TSDB_CODE_SUCCESS;
-}
-
-static void* setAllSourcesCompleted(SOperatorInfo* pOperator, int64_t startTs) {
-  SExchangeInfo* pExchangeInfo = pOperator->info;
-  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
-
-  int64_t              el = taosGetTimestampUs() - startTs;
-  SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
-
-  pLoadInfo->totalElapsed += el;
-
-  size_t totalSources = taosArrayGetSize(pExchangeInfo->pSources);
-  qDebug("%s all %" PRIzu " sources are exhausted, total rows: %" PRIu64 " bytes:%" PRIu64 ", elapsed:%.2f ms",
-         GET_TASKID(pTaskInfo), totalSources, pLoadInfo->totalRows, pLoadInfo->totalSize,
-         pLoadInfo->totalElapsed / 1000.0);
-
-  doSetOperatorCompleted(pOperator);
-  return NULL;
-}
-
-static void concurrentlyLoadRemoteDataImpl(SOperatorInfo* pOperator, SExchangeInfo* pExchangeInfo,
-                                           SExecTaskInfo* pTaskInfo) {
-  int32_t code = 0;
-  int64_t startTs = taosGetTimestampUs();
-  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
-
-  int32_t completed = 0;
-  for (int32_t k = 0; k < totalSources; ++k) {
-    SSourceDataInfo* p = taosArrayGet(pExchangeInfo->pSourceDataInfo, k);
-    if (p->status == EX_SOURCE_DATA_EXHAUSTED) {
-      completed += 1;
-    }
-  }
-
-  if (completed == totalSources) {
-    setAllSourcesCompleted(pOperator, startTs);
-    return;
-  }
-
-  while (1) {
-//    printf("1\n");
-    tsem_wait(&pExchangeInfo->ready);
-//    printf("2\n");
-
-    for (int32_t i = 0; i < totalSources; ++i) {
-      SSourceDataInfo* pDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, i);
-      if (pDataInfo->status == EX_SOURCE_DATA_EXHAUSTED) {
-//        printf("========:%d is completed\n", i);
-        continue;
-      }
-
-//      printf("index:%d -  status:%d\n", i, pDataInfo->status);
-      if (pDataInfo->status != EX_SOURCE_DATA_READY) {
-//        printf("-----------%d, status:%d, continue\n", i, pDataInfo->status);
-        continue;
-      }
-
-      if (pDataInfo->code != TSDB_CODE_SUCCESS) {
-        code = pDataInfo->code;
-        goto _error;
-      }
-
-      SRetrieveTableRsp*     pRsp = pDataInfo->pRsp;
-      SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, i);
-
-      // todo
-      SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
-      if (pRsp->numOfRows == 0) {
-        pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
-//        printf("%d completed, try next\n", i);
-
-        qDebug("%s vgId:%d, taskId:0x%" PRIx64 " execId:%d index:%d completed, rowsOfSource:%" PRIu64
-               ", totalRows:%" PRIu64 ", completed:%d try next %d/%" PRIzu,
-               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, i, pDataInfo->totalRows,
-               pExchangeInfo->loadInfo.totalRows, completed, i + 1, totalSources);
-        taosMemoryFreeClear(pDataInfo->pRsp);
-
-//        if (completed == totalSources) {
-//          return;
-//        } else {
-//          break;
-//        }
-         break;
-      }
-
-      SRetrieveTableRsp* pRetrieveRsp = pDataInfo->pRsp;
-      int32_t            index = 0;
-      char*              pStart = pRetrieveRsp->data;
-      while (index++ < pRetrieveRsp->numOfBlocks) {
-        printf("results, numOfBLock: %d\n", pRetrieveRsp->numOfBlocks);
-        SSDataBlock* pb = createOneDataBlock(pExchangeInfo->pDummyBlock, false);
-        code = extractDataBlockFromFetchRsp(pb, pStart, NULL, &pStart);
-        if (code != 0) {
-          taosMemoryFreeClear(pDataInfo->pRsp);
-          goto _error;
-        }
-
-        taosArrayPush(pExchangeInfo->pResultBlockList, &pb);
-      }
-
-      updateLoadRemoteInfo(pLoadInfo, pRetrieveRsp->numOfRows, pRetrieveRsp->compLen, startTs, pOperator);
-
-//      int32_t completed = 0;
-      if (pRsp->completed == 1) {
-        pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
-
-//        for (int32_t k = 0; k < totalSources; ++k) {
-//          SSourceDataInfo* p = taosArrayGet(pExchangeInfo->pSourceDataInfo, k);
-//          if (p->status == EX_SOURCE_DATA_EXHAUSTED) {
-//            completed += 1;
-//          }
-//        }
-
-        qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64
-               " execId:%d index:%d completed, blocks:%d, numOfRows:%d, rowsOfSource:%" PRIu64 ", totalRows:%" PRIu64
-               ", total:%.2f Kb, completed:%d try next %d/%" PRIzu,
-               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, i, pRsp->numOfBlocks,
-               pRsp->numOfRows, pDataInfo->totalRows, pLoadInfo->totalRows, pLoadInfo->totalSize / 1024.0,
-               completed, i + 1, totalSources);
-      } else {
-        qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64
-               " execId:%d blocks:%d, numOfRows:%d, totalRows:%" PRIu64 ", total:%.2f Kb",
-               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, pRsp->numOfBlocks,
-               pRsp->numOfRows, pLoadInfo->totalRows, pLoadInfo->totalSize / 1024.0);
-      }
-
-      taosMemoryFreeClear(pDataInfo->pRsp);
-
-      if (pDataInfo->status != EX_SOURCE_DATA_EXHAUSTED) {
-        pDataInfo->status = EX_SOURCE_DATA_NOT_READY;
-        code = doSendFetchDataRequest(pExchangeInfo, pTaskInfo, i);
-        if (code != TSDB_CODE_SUCCESS) {
-          taosMemoryFreeClear(pDataInfo->pRsp);
-          goto _error;
-        }
-      }
-
-//      if (completed == totalSources) {
-//        setAllSourcesCompleted(pOperator, startTs);
-//      }
-
-      return;
-    }
-
-    int32_t completed = 0;
-    for (int32_t k = 0; k < totalSources; ++k) {
-      SSourceDataInfo* p = taosArrayGet(pExchangeInfo->pSourceDataInfo, k);
-      if (p->status == EX_SOURCE_DATA_EXHAUSTED) {
-        completed += 1;
-      }
-    }
-
-    if (completed == totalSources) {
-      return;
-    }
-  }
-
-_error:
-  pTaskInfo->code = code;
-}
-
-static int32_t prepareConcurrentlyLoad(SOperatorInfo* pOperator) {
-  SExchangeInfo* pExchangeInfo = pOperator->info;
-  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
-
-  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
-  int64_t startTs = taosGetTimestampUs();
-
-  // Asynchronously send all fetch requests to all sources.
-  for (int32_t i = 0; i < totalSources; ++i) {
-    int32_t code = doSendFetchDataRequest(pExchangeInfo, pTaskInfo, i);
-    if (code != TSDB_CODE_SUCCESS) {
-      pTaskInfo->code = code;
-      return code;
-    }
-  }
-
-  int64_t endTs = taosGetTimestampUs();
-  qDebug("%s send all fetch requests to %" PRIzu " sources completed, elapsed:%.2fms", GET_TASKID(pTaskInfo),
-         totalSources, (endTs - startTs) / 1000.0);
-
-  pOperator->status = OP_RES_TO_RETURN;
-  pOperator->cost.openCost = taosGetTimestampUs() - startTs;
-
-  tsem_wait(&pExchangeInfo->ready);
-  tsem_post(&pExchangeInfo->ready);
-  return TSDB_CODE_SUCCESS;
-}
-
-static int32_t seqLoadRemoteData(SOperatorInfo* pOperator) {
-  SExchangeInfo* pExchangeInfo = pOperator->info;
-  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
-
-  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
-  int64_t startTs = taosGetTimestampUs();
-
-  while (1) {
-    if (pExchangeInfo->current >= totalSources) {
-      setAllSourcesCompleted(pOperator, startTs);
-      return TSDB_CODE_SUCCESS;
-    }
-
-    doSendFetchDataRequest(pExchangeInfo, pTaskInfo, pExchangeInfo->current);
-    tsem_wait(&pExchangeInfo->ready);
-
-    SSourceDataInfo*       pDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, pExchangeInfo->current);
-    SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, pExchangeInfo->current);
-
-    if (pDataInfo->code != TSDB_CODE_SUCCESS) {
-      qError("%s vgId:%d, taskID:0x%" PRIx64 " execId:%d error happens, code:%s", GET_TASKID(pTaskInfo),
-             pSource->addr.nodeId, pSource->taskId, pSource->execId, tstrerror(pDataInfo->code));
-      pOperator->pTaskInfo->code = pDataInfo->code;
-      return pOperator->pTaskInfo->code;
-    }
-
-    SRetrieveTableRsp*   pRsp = pDataInfo->pRsp;
-    SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
-    if (pRsp->numOfRows == 0) {
-      qDebug("%s vgId:%d, taskID:0x%" PRIx64 " execId:%d %d of total completed, rowsOfSource:%" PRIu64
-             ", totalRows:%" PRIu64 " try next",
-             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, pExchangeInfo->current + 1,
-             pDataInfo->totalRows, pLoadInfo->totalRows);
-
-      pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
-      pExchangeInfo->current += 1;
-      taosMemoryFreeClear(pDataInfo->pRsp);
-      continue;
-    }
-
-    SRetrieveTableRsp* pRetrieveRsp = pDataInfo->pRsp;
-
-    char*   pStart = pRetrieveRsp->data;
-    int32_t code = extractDataBlockFromFetchRsp(NULL, pStart, NULL, &pStart);
-
-    if (pRsp->completed == 1) {
-      qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " execId:%d numOfRows:%d, rowsOfSource:%" PRIu64
-             ", totalRows:%" PRIu64 ", totalBytes:%" PRIu64 " try next %d/%" PRIzu,
-             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, pRetrieveRsp->numOfRows,
-             pDataInfo->totalRows, pLoadInfo->totalRows, pLoadInfo->totalSize, pExchangeInfo->current + 1,
-             totalSources);
-
-      pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
-      pExchangeInfo->current += 1;
-    } else {
-      qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " execId:%d numOfRows:%d, totalRows:%" PRIu64
-             ", totalBytes:%" PRIu64,
-             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, pRetrieveRsp->numOfRows,
-             pLoadInfo->totalRows, pLoadInfo->totalSize);
-    }
-
-    updateLoadRemoteInfo(pLoadInfo, pRetrieveRsp->numOfRows, pRetrieveRsp->compLen, startTs, pOperator);
-    pDataInfo->totalRows += pRetrieveRsp->numOfRows;
-
-    taosMemoryFreeClear(pDataInfo->pRsp);
-    return TSDB_CODE_SUCCESS;
-  }
-}
-
-static int32_t prepareLoadRemoteData(SOperatorInfo* pOperator) {
-  if (OPTR_IS_OPENED(pOperator)) {
-    return TSDB_CODE_SUCCESS;
-  }
-
-  int64_t st = taosGetTimestampUs();
-
-  SExchangeInfo* pExchangeInfo = pOperator->info;
-  if (!pExchangeInfo->seqLoadData) {
-    int32_t code = prepareConcurrentlyLoad(pOperator);
-    if (code != TSDB_CODE_SUCCESS) {
-      return code;
-    }
-  }
-
-  OPTR_SET_OPENED(pOperator);
-  pOperator->cost.openCost = (taosGetTimestampUs() - st) / 1000.0;
-  return TSDB_CODE_SUCCESS;
-}
-
-static void freeBlock(void* pParam) {
-  SSDataBlock* pBlock = *(SSDataBlock**)pParam;
-  blockDataDestroy(pBlock);
-}
-
-static SSDataBlock* doLoadRemoteDataImpl(SOperatorInfo* pOperator) {
-  SExchangeInfo* pExchangeInfo = pOperator->info;
-  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
-
-  pTaskInfo->code = pOperator->fpSet._openFn(pOperator);
-  if (pTaskInfo->code != TSDB_CODE_SUCCESS) {
-    return NULL;
-  }
-
-  size_t totalSources = taosArrayGetSize(pExchangeInfo->pSources);
-
-  SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
-  if (pOperator->status == OP_EXEC_DONE) {
-    qDebug("%s all %" PRIzu " source(s) are exhausted, total rows:%" PRIu64 " bytes:%" PRIu64 ", elapsed:%.2f ms",
-           GET_TASKID(pTaskInfo), totalSources, pLoadInfo->totalRows, pLoadInfo->totalSize,
-           pLoadInfo->totalElapsed / 1000.0);
-    return NULL;
-  }
-
-  size_t size = taosArrayGetSize(pExchangeInfo->pResultBlockList);
-  if (size == 0 || pExchangeInfo->rspBlockIndex >= size) {
-    pExchangeInfo->rspBlockIndex = 0;
-    taosArrayClearEx(pExchangeInfo->pResultBlockList, freeBlock);
-    if (pExchangeInfo->seqLoadData) {
-      seqLoadRemoteData(pOperator);
-    } else {
-      concurrentlyLoadRemoteDataImpl(pOperator, pExchangeInfo, pTaskInfo);
-    }
-
-    if (taosArrayGetSize(pExchangeInfo->pResultBlockList) == 0) {
-      return NULL;
-    }
-  }
-
-  // we have buffered retrieved datablock, return it directly
-  return taosArrayGetP(pExchangeInfo->pResultBlockList, pExchangeInfo->rspBlockIndex++);
-}
-
-static SSDataBlock* doLoadRemoteData(SOperatorInfo* pOperator) {
-  SExchangeInfo* pExchangeInfo = pOperator->info;
-  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
-
-  if (pOperator->status == OP_EXEC_DONE) {
-    return NULL;
-  }
-
-  while (1) {
-    SSDataBlock* pBlock = doLoadRemoteDataImpl(pOperator);
-    if (pBlock == NULL) {
-      return NULL;
-    }
-
-    SLimitInfo* pLimitInfo = &pExchangeInfo->limitInfo;
-    if (hasLimitOffsetInfo(pLimitInfo)) {
-      int32_t status = handleLimitOffset(pOperator, pLimitInfo, pBlock, false);
-      if (status == PROJECT_RETRIEVE_CONTINUE) {
-        continue;
-      } else if (status == PROJECT_RETRIEVE_DONE) {
-        size_t rows = pBlock->info.rows;
-        pExchangeInfo->limitInfo.numOfOutputRows += rows;
-
-        if (rows == 0) {
-          doSetOperatorCompleted(pOperator);
-          return NULL;
-        } else {
-          return pBlock;
-        }
-      }
-    } else {
-      return pBlock;
-    }
-  }
-}
-
-static int32_t initDataSource(int32_t numOfSources, SExchangeInfo* pInfo, const char* id) {
-  pInfo->pSourceDataInfo = taosArrayInit(numOfSources, sizeof(SSourceDataInfo));
-  if (pInfo->pSourceDataInfo == NULL) {
-    return TSDB_CODE_OUT_OF_MEMORY;
-  }
-
-  for (int32_t i = 0; i < numOfSources; ++i) {
-    SSourceDataInfo dataInfo = {0};
-    dataInfo.status = EX_SOURCE_DATA_NOT_READY;
-    dataInfo.taskId = id;
-    dataInfo.index = i;
-    SSourceDataInfo* pDs = taosArrayPush(pInfo->pSourceDataInfo, &dataInfo);
-    if (pDs == NULL) {
-      taosArrayDestroy(pInfo->pSourceDataInfo);
-      return TSDB_CODE_OUT_OF_MEMORY;
-    }
-  }
-
-  return TSDB_CODE_SUCCESS;
-}
-
-static int32_t initExchangeOperator(SExchangePhysiNode* pExNode, SExchangeInfo* pInfo, const char* id) {
-  size_t numOfSources = LIST_LENGTH(pExNode->pSrcEndPoints);
-
-  if (numOfSources == 0) {
-    qError("%s invalid number: %d of sources in exchange operator", id, (int32_t)numOfSources);
-    return TSDB_CODE_INVALID_PARA;
-  }
-
-  pInfo->pSources = taosArrayInit(numOfSources, sizeof(SDownstreamSourceNode));
-  if (pInfo->pSources == NULL) {
-    return TSDB_CODE_OUT_OF_MEMORY;
-  }
-
-  for (int32_t i = 0; i < numOfSources; ++i) {
-    SDownstreamSourceNode* pNode = (SDownstreamSourceNode*)nodesListGetNode((SNodeList*)pExNode->pSrcEndPoints, i);
-    taosArrayPush(pInfo->pSources, pNode);
-  }
-
-  initLimitInfo(pExNode->node.pLimit, pExNode->node.pSlimit, &pInfo->limitInfo);
-  pInfo->self = taosAddRef(exchangeObjRefPool, pInfo);
-
-  return initDataSource(numOfSources, pInfo, id);
-}
-
-SOperatorInfo* createExchangeOperatorInfo(void* pTransporter, SExchangePhysiNode* pExNode, SExecTaskInfo* pTaskInfo) {
-  SExchangeInfo* pInfo = taosMemoryCalloc(1, sizeof(SExchangeInfo));
-  SOperatorInfo* pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
-  if (pInfo == NULL || pOperator == NULL) {
-    goto _error;
-  }
-
-  int32_t code = initExchangeOperator(pExNode, pInfo, GET_TASKID(pTaskInfo));
-  if (code != TSDB_CODE_SUCCESS) {
-    goto _error;
-  }
-
-  tsem_init(&pInfo->ready, 0, 0);
-  pInfo->pDummyBlock = createResDataBlock(pExNode->node.pOutputDataBlockDesc);
-  pInfo->pResultBlockList = taosArrayInit(1, POINTER_BYTES);
-
-  pInfo->seqLoadData = false;
-  pInfo->pTransporter = pTransporter;
-
-  pOperator->name = "ExchangeOperator";
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_EXCHANGE;
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
-  pOperator->exprSupp.numOfExprs = taosArrayGetSize(pInfo->pDummyBlock->pDataBlock);
-  pOperator->pTaskInfo = pTaskInfo;
-
-  pOperator->fpSet =
-      createOperatorFpSet(prepareLoadRemoteData, doLoadRemoteData, NULL, NULL, destroyExchangeOperatorInfo, NULL);
-  return pOperator;
-
-_error:
-  if (pInfo != NULL) {
-    doDestroyExchangeOperatorInfo(pInfo);
-  }
-
-  taosMemoryFreeClear(pOperator);
-  pTaskInfo->code = code;
-  return NULL;
-}
-
 static int32_t doInitAggInfoSup(SAggSupporter* pAggSup, SqlFunctionCtx* pCtx, int32_t numOfOutput, size_t keyBufSize,
                                 const char* pKey);
 
@@ -2413,7 +1802,7 @@ static SSDataBlock* getAggregateResult(SOperatorInfo* pOperator) {
   SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
   pTaskInfo->code = pOperator->fpSet._openFn(pOperator);
   if (pTaskInfo->code != TSDB_CODE_SUCCESS) {
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
     return NULL;
   }
 
@@ -2423,7 +1812,7 @@ static SSDataBlock* getAggregateResult(SOperatorInfo* pOperator) {
     doFilter(pInfo->pRes, pOperator->exprSupp.pFilterInfo, NULL);
 
     if (!hasRemainResults(&pAggInfo->groupResInfo)) {
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
       break;
     }
 
@@ -2672,7 +2061,7 @@ static SSDataBlock* doFillImpl(SOperatorInfo* pOperator) {
     SSDataBlock* pBlock = pDownstream->fpSet.getNextFn(pDownstream);
     if (pBlock == NULL) {
       if (pInfo->totalInputRows == 0) {
-        doSetOperatorCompleted(pOperator);
+        setOperatorCompleted(pOperator);
         return NULL;
       }
 
@@ -2749,7 +2138,7 @@ static SSDataBlock* doFill(SOperatorInfo* pOperator) {
   while (true) {
     fillResult = doFillImpl(pOperator);
     if (fillResult == NULL) {
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
       break;
     }
 
@@ -2780,7 +2169,7 @@ void destroyExprInfo(SExprInfo* pExpr, int32_t numOfExprs) {
   }
 }
 
-static void destroyOperatorInfo(SOperatorInfo* pOperator) {
+void destroyOperatorInfo(SOperatorInfo* pOperator) {
   if (pOperator == NULL) {
     return;
   }
@@ -2979,15 +2368,10 @@ SOperatorInfo* createAggregateOperatorInfo(SOperatorInfo* downstream, SAggPhysiN
 
   pInfo->binfo.mergeResultBlock = pAggNode->mergeDataBlock;
   pInfo->groupId = UINT64_MAX;
-  pOperator->name = "TableAggregate";
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_HASH_AGG;
-  pOperator->blocking = true;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
-  pOperator->pTaskInfo = pTaskInfo;
 
+  setOperatorInfo(pOperator, "TableAggregate", QUERY_NODE_PHYSICAL_PLAN_HASH_AGG, true, OP_NOT_OPENED, pInfo, pTaskInfo);
   pOperator->fpSet =
-      createOperatorFpSet(doOpenAggregateOptr, getAggregateResult, NULL, NULL, destroyAggOperatorInfo, NULL);
+      createOperatorFpSet(doOpenAggregateOptr, getAggregateResult, NULL, destroyAggOperatorInfo, NULL);
 
   if (downstream->operatorType == QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN) {
     STableScanInfo* pTableScanInfo = downstream->info;
@@ -3051,33 +2435,6 @@ void destroyFillOperatorInfo(void* param) {
   taosMemoryFreeClear(param);
 }
 
-void destroyExchangeOperatorInfo(void* param) {
-  SExchangeInfo* pExInfo = (SExchangeInfo*)param;
-  taosRemoveRef(exchangeObjRefPool, pExInfo->self);
-}
-
-void freeSourceDataInfo(void* p) {
-  SSourceDataInfo* pInfo = (SSourceDataInfo*)p;
-  taosMemoryFreeClear(pInfo->pRsp);
-}
-
-void doDestroyExchangeOperatorInfo(void* param) {
-  SExchangeInfo* pExInfo = (SExchangeInfo*)param;
-
-  taosArrayDestroy(pExInfo->pSources);
-  taosArrayDestroyEx(pExInfo->pSourceDataInfo, freeSourceDataInfo);
-
-  if (pExInfo->pResultBlockList != NULL) {
-    taosArrayDestroyEx(pExInfo->pResultBlockList, freeBlock);
-    pExInfo->pResultBlockList = NULL;
-  }
-
-  blockDataDestroy(pExInfo->pDummyBlock);
-
-  tsem_destroy(&pExInfo->ready);
-  taosMemoryFreeClear(param);
-}
-
 static int32_t initFillInfo(SFillOperatorInfo* pInfo, SExprInfo* pExpr, int32_t numOfCols, SExprInfo* pNotFillExpr,
                             int32_t numOfNotFillCols, SNodeListNode* pValNode, STimeWindow win, int32_t capacity,
                             const char* id, SInterval* pInterval, int32_t fillType, int32_t order) {
@@ -3209,15 +2566,9 @@ SOperatorInfo* createFillOperatorInfo(SOperatorInfo* downstream, SFillPhysiNode*
     goto _error;
   }
 
-  pOperator->name = "FillOperator";
+  setOperatorInfo(pOperator, "FillOperator", QUERY_NODE_PHYSICAL_PLAN_FILL, false, OP_NOT_OPENED, pInfo, pTaskInfo);
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_FILL;
   pOperator->exprSupp.numOfExprs = pInfo->numOfExpr;
-  pOperator->info = pInfo;
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doFill, NULL, destroyFillOperatorInfo, NULL);
-  pOperator->pTaskInfo = pTaskInfo;
-
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doFill, NULL, NULL, destroyFillOperatorInfo, NULL);
 
   code = appendDownstream(pOperator, &downstream, 1);
   return pOperator;
@@ -4072,6 +3423,7 @@ int32_t buildSessionResultDataBlock(SOperatorInfo* pOperator, SStreamState* pSta
     ASSERT(code == 0);
     if (code == -1) {
       // coverity scan
+      pGroupResInfo->index += 1;
       continue;
     }
     SResultRow* pRow = (SResultRow*)pVal;

source/libs/executor/src/groupoperator.c

@@ -316,7 +316,7 @@ static SSDataBlock* buildGroupResultDataBlock(SOperatorInfo* pOperator) {
     doFilter(pRes, pOperator->exprSupp.pFilterInfo, NULL);
 
     if (!hasRemainResults(&pInfo->groupResInfo)) {
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
       break;
     }
 
@@ -438,15 +438,10 @@ SOperatorInfo* createGroupOperatorInfo(SOperatorInfo* downstream, SAggPhysiNode*
   }
 
   initResultRowInfo(&pInfo->binfo.resultRowInfo);
-
+  setOperatorInfo(pOperator, "GroupbyAggOperator", 0, true, OP_NOT_OPENED, pInfo, pTaskInfo);
-  pOperator->name = "GroupbyAggOperator";
-  pOperator->blocking = true;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
-  pOperator->pTaskInfo = pTaskInfo;
 
   pOperator->fpSet =
-      createOperatorFpSet(operatorDummyOpenFn, hashGroupbyAggregate, NULL, NULL, destroyGroupOperatorInfo, NULL);
+      createOperatorFpSet(operatorDummyOpenFn, hashGroupbyAggregate, NULL, destroyGroupOperatorInfo, NULL);
   code = appendDownstream(pOperator, &downstream, 1);
   if (code != TSDB_CODE_SUCCESS) {
     goto _error;
@@ -654,7 +649,7 @@ static SSDataBlock* buildPartitionResult(SOperatorInfo* pOperator) {
     // try next group data
     ++pInfo->groupIndex;
     if (pInfo->groupIndex >= taosArrayGetSize(pInfo->sortedGroupArray)) {
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
       clearPartitionOperator(pInfo);
       return NULL;
     }
@@ -821,17 +816,12 @@ SOperatorInfo* createPartitionOperatorInfo(SOperatorInfo* downstream, SPartition
     goto _error;
   }
 
-  pOperator->name = "PartitionOperator";
+  setOperatorInfo(pOperator, "PartitionOperator", QUERY_NODE_PHYSICAL_PLAN_PARTITION, false, OP_NOT_OPENED, pInfo,
-  pOperator->blocking = true;
+                  pTaskInfo);
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_PARTITION;
   pOperator->exprSupp.numOfExprs = numOfCols;
   pOperator->exprSupp.pExprInfo = pExprInfo;
-  pOperator->info = pInfo;
-  pOperator->pTaskInfo = pTaskInfo;
 
-  pOperator->fpSet =
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, hashPartition, NULL, destroyPartitionOperatorInfo, NULL);
-      createOperatorFpSet(operatorDummyOpenFn, hashPartition, NULL, NULL, destroyPartitionOperatorInfo, NULL);
 
   code = appendDownstream(pOperator, &downstream, 1);
   return pOperator;
@@ -909,7 +899,10 @@ static SSDataBlock* buildStreamPartitionResult(SOperatorInfo* pOperator) {
       void* pData = colDataGetVarData(pCol, 0);
       // TODO check tbname validity
       if (pData != (void*)-1) {
-        memcpy(pDest->info.parTbName, varDataVal(pData), varDataLen(pData));
+        memset(pDest->info.parTbName, 0, TSDB_TABLE_NAME_LEN);
+        int32_t len = TMIN(varDataLen(pData), TSDB_TABLE_NAME_LEN - 1);
+        memcpy(pDest->info.parTbName, varDataVal(pData), len);
+        /*pDest->info.parTbName[len + 1] = 0;*/
       } else {
         pDest->info.parTbName[0] = 0;
       }
@@ -918,6 +911,8 @@ static SSDataBlock* buildStreamPartitionResult(SOperatorInfo* pOperator) {
       blockDataDestroy(pResBlock);
     }
   }
+  taosArrayDestroy(pParInfo->rowIds);
+  pParInfo->rowIds = NULL;
   blockDataUpdateTsWindow(pDest, pInfo->tsColIndex);
   pDest->info.groupId = pParInfo->groupId;
   pOperator->resultInfo.totalRows += pDest->info.rows;
@@ -963,7 +958,7 @@ static SSDataBlock* doStreamHashPartition(SOperatorInfo* pOperator) {
     pInfo->pInputDataBlock = NULL;
     SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
     if (pBlock == NULL) {
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
       return NULL;
     }
     printDataBlock(pBlock, "stream partitionby recv");
@@ -1016,6 +1011,7 @@ static void destroyStreamPartitionOperatorInfo(void* param) {
   cleanupExprSupp(&pInfo->tbnameCalSup);
   cleanupExprSupp(&pInfo->tagCalSup);
   blockDataDestroy(pInfo->pDelRes);
+  taosHashCleanup(pInfo->pPartitions);
   taosMemoryFreeClear(param);
 }
 
@@ -1103,16 +1099,12 @@ SOperatorInfo* createStreamPartitionOperatorInfo(SOperatorInfo* downstream, SStr
   int32_t    numOfCols = 0;
   SExprInfo* pExprInfo = createExprInfo(pPartNode->part.pTargets, NULL, &numOfCols);
 
-  pOperator->name = "StreamPartitionOperator";
+  setOperatorInfo(pOperator, "StreamPartitionOperator", QUERY_NODE_PHYSICAL_PLAN_STREAM_PARTITION, false, OP_NOT_OPENED,
-  pOperator->blocking = false;
+                  pInfo, pTaskInfo);
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_STREAM_PARTITION;
   pOperator->exprSupp.numOfExprs = numOfCols;
   pOperator->exprSupp.pExprInfo = pExprInfo;
-  pOperator->info = pInfo;
+  pOperator->fpSet =
-  pOperator->pTaskInfo = pTaskInfo;
+      createOperatorFpSet(operatorDummyOpenFn, doStreamHashPartition, NULL, destroyStreamPartitionOperatorInfo, NULL);
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doStreamHashPartition, NULL, NULL,
-                                         destroyStreamPartitionOperatorInfo, NULL);
 
   initParDownStream(downstream, &pInfo->partitionSup, &pInfo->scalarSup);
   code = appendDownstream(pOperator, &downstream, 1);

source/libs/executor/src/joinoperator.c

@@ -73,14 +73,10 @@ SOperatorInfo* createMergeJoinOperatorInfo(SOperatorInfo** pDownstream, int32_t
   initResultSizeInfo(&pOperator->resultInfo, 4096);
 
   pInfo->pRes = pResBlock;
-  pOperator->name = "MergeJoinOperator";
+
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN;
+  setOperatorInfo(pOperator, "MergeJoinOperator", QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN, false, OP_NOT_OPENED, pInfo, pTaskInfo);
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
   pOperator->exprSupp.pExprInfo = pExprInfo;
   pOperator->exprSupp.numOfExprs = numOfCols;
-  pOperator->info = pInfo;
-  pOperator->pTaskInfo = pTaskInfo;
 
   extractTimeCondition(pInfo, pDownstream, numOfDownstream, pJoinNode);
 
@@ -121,8 +117,7 @@ SOperatorInfo* createMergeJoinOperatorInfo(SOperatorInfo** pDownstream, int32_t
     pInfo->inputOrder = TSDB_ORDER_DESC;
   }
 
-  pOperator->fpSet =
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doMergeJoin, NULL, destroyMergeJoinOperator, NULL);
-      createOperatorFpSet(operatorDummyOpenFn, doMergeJoin, NULL, NULL, destroyMergeJoinOperator, NULL);
   code = appendDownstream(pOperator, pDownstream, numOfDownstream);
   if (code != TSDB_CODE_SUCCESS) {
     goto _error;
@@ -372,13 +367,13 @@ static void doMergeJoinImpl(struct SOperatorInfo* pOperator, SSDataBlock* pRes)
 
     if (leftTs == rightTs) {
       mergeJoinJoinDownstreamTsRanges(pOperator, leftTs, pRes, &nrows);
-    } else if (asc && leftTs < rightTs || !asc && leftTs > rightTs) {
+    } else if ((asc && leftTs < rightTs) || (!asc && leftTs > rightTs)) {
       pJoinInfo->leftPos += 1;
 
       if (pJoinInfo->leftPos >= pJoinInfo->pLeft->info.rows) {
         continue;
       }
-    } else if (asc && leftTs > rightTs || !asc && leftTs < rightTs) {
+    } else if ((asc && leftTs > rightTs) || (!asc && leftTs < rightTs)) {
       pJoinInfo->rightPos += 1;
       if (pJoinInfo->rightPos >= pJoinInfo->pRight->info.rows) {
         continue;

source/libs/executor/src/projectoperator.c

@@ -98,13 +98,9 @@ SOperatorInfo* createProjectOperatorInfo(SOperatorInfo* downstream, SProjectPhys
   }
 
   pInfo->pPseudoColInfo = setRowTsColumnOutputInfo(pOperator->exprSupp.pCtx, numOfCols);
-  pOperator->name = "ProjectOperator";
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_PROJECT;
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
 
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doProjectOperation, NULL, NULL,
+  setOperatorInfo(pOperator, "ProjectOperator", QUERY_NODE_PHYSICAL_PLAN_PROJECT, false, OP_NOT_OPENED, pInfo, pTaskInfo);
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doProjectOperation, NULL,
                                          destroyProjectOperatorInfo, NULL);
 
   code = appendDownstream(pOperator, &downstream, 1);
@@ -153,7 +149,7 @@ static int32_t setInfoForNewGroup(SSDataBlock* pBlock, SLimitInfo* pLimitInfo, S
   if (pLimitInfo->currentGroupId != 0 && pLimitInfo->currentGroupId != pBlock->info.groupId) {
     pLimitInfo->numOfOutputGroups += 1;
     if ((pLimitInfo->slimit.limit > 0) && (pLimitInfo->slimit.limit <= pLimitInfo->numOfOutputGroups)) {
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
       return PROJECT_RETRIEVE_DONE;
     }
 
@@ -187,7 +183,7 @@ static int32_t doIngroupLimitOffset(SLimitInfo* pLimitInfo, uint64_t groupId, SS
     // TODO: optimize it later when partition by + limit
     if ((pLimitInfo->slimit.limit == -1 && pLimitInfo->currentGroupId == 0) ||
         (pLimitInfo->slimit.limit > 0 && pLimitInfo->slimit.limit <= pLimitInfo->numOfOutputGroups)) {
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
     }
   }
 
@@ -252,7 +248,7 @@ SSDataBlock* doProjectOperation(SOperatorInfo* pOperator) {
         }
         qDebug("set op close, exec %d, status %d rows %d", pTaskInfo->execModel, pOperator->status,
                pFinalRes->info.rows);
-        doSetOperatorCompleted(pOperator);
+        setOperatorCompleted(pOperator);
         break;
       }
       if (pTaskInfo->execModel == OPTR_EXEC_MODEL_QUEUE) {
@@ -400,14 +396,8 @@ SOperatorInfo* createIndefinitOutputOperatorInfo(SOperatorInfo* downstream, SPhy
   pInfo->binfo.pRes = pResBlock;
   pInfo->pPseudoColInfo = setRowTsColumnOutputInfo(pSup->pCtx, numOfExpr);
 
-  pOperator->name = "IndefinitOperator";
+  setOperatorInfo(pOperator, "IndefinitOperator", QUERY_NODE_PHYSICAL_PLAN_INDEF_ROWS_FUNC, false, OP_NOT_OPENED, pInfo, pTaskInfo);
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_INDEF_ROWS_FUNC;
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doApplyIndefinitFunction, NULL, destroyIndefinitOperatorInfo, NULL);
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
-
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doApplyIndefinitFunction, NULL, NULL,
-                                         destroyIndefinitOperatorInfo, NULL);
 
   code = appendDownstream(pOperator, &downstream, 1);
   if (code != TSDB_CODE_SUCCESS) {
@@ -499,7 +489,7 @@ SSDataBlock* doApplyIndefinitFunction(SOperatorInfo* pOperator) {
         // The downstream exec may change the value of the newgroup, so use a local variable instead.
         SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
         if (pBlock == NULL) {
-          doSetOperatorCompleted(pOperator);
+          setOperatorCompleted(pOperator);
           break;
         }
 
@@ -628,7 +618,7 @@ SSDataBlock* doGenerateSourceData(SOperatorInfo* pOperator) {
 
   pOperator->resultInfo.totalRows += pRes->info.rows;
 
-  doSetOperatorCompleted(pOperator);
+  setOperatorCompleted(pOperator);
   if (pOperator->cost.openCost == 0) {
     pOperator->cost.openCost = (taosGetTimestampUs() - st) / 1000.0;
   }

source/libs/executor/src/scanoperator.c

@@ -284,7 +284,7 @@ static int32_t doDynamicPruneDataBlock(SOperatorInfo* pOperator, SDataBlockInfo*
 }
 
 static bool doFilterByBlockSMA(SFilterInfo* pFilterInfo, SColumnDataAgg** pColsAgg, int32_t numOfCols,
-                                            int32_t numOfRows) {
+                               int32_t numOfRows) {
   if (pColsAgg == NULL || pFilterInfo == NULL) {
     return true;
   }
@@ -345,7 +345,7 @@ static void doSetTagColumnData(STableScanInfo* pTableScanInfo, SSDataBlock* pBlo
 
 // todo handle the slimit info
 void applyLimitOffset(SLimitInfo* pLimitInfo, SSDataBlock* pBlock, SExecTaskInfo* pTaskInfo, SOperatorInfo* pOperator) {
-  SLimit* pLimit = &pLimitInfo->limit;
+  SLimit*     pLimit = &pLimitInfo->limit;
   const char* id = GET_TASKID(pTaskInfo);
 
   if (pLimit->offset > 0 && pLimitInfo->remainOffset > 0) {
@@ -499,7 +499,7 @@ static void prepareForDescendingScan(STableScanInfo* pTableScanInfo, SqlFunction
 
 typedef struct STableCachedVal {
   const char* pName;
-  STag* pTags;
+  STag*       pTags;
 } STableCachedVal;
 
 static void freeTableCachedVal(void* param) {
@@ -513,13 +513,11 @@ static void freeTableCachedVal(void* param) {
   taosMemoryFree(pVal);
 }
 
-//const void *key, size_t keyLen, void *value
+// const void *key, size_t keyLen, void *value
-static void freeCachedMetaItem(const void *key, size_t keyLen, void *value) {
+static void freeCachedMetaItem(const void* key, size_t keyLen, void* value) { freeTableCachedVal(value); }
-  freeTableCachedVal(value);
-}
 
-int32_t addTagPseudoColumnData(SReadHandle* pHandle, const SExprInfo* pExpr, int32_t numOfExpr,
+int32_t addTagPseudoColumnData(SReadHandle* pHandle, const SExprInfo* pExpr, int32_t numOfExpr, SSDataBlock* pBlock,
-                               SSDataBlock* pBlock, int32_t rows, const char* idStr, STableMetaCacheInfo* pCache) {
+                               int32_t rows, const char* idStr, STableMetaCacheInfo* pCache) {
   // currently only the tbname pseudo column
   if (numOfExpr <= 0) {
     return TSDB_CODE_SUCCESS;
@@ -531,11 +529,11 @@ int32_t addTagPseudoColumnData(SReadHandle* pHandle, const SExprInfo* pExpr, int
   int32_t backupRows = pBlock->info.rows;
   pBlock->info.rows = rows;
 
-  bool freeReader = false;
+  bool            freeReader = false;
   STableCachedVal val = {0};
 
   SMetaReader mr = {0};
-  LRUHandle* h = NULL;
+  LRUHandle*  h = NULL;
 
   // 1. check if it is existed in meta cache
   if (pCache == NULL) {
@@ -582,7 +580,8 @@ int32_t addTagPseudoColumnData(SReadHandle* pHandle, const SExprInfo* pExpr, int
       val = *pVal;
       freeReader = true;
 
-      int32_t ret = taosLRUCacheInsert(pCache->pTableMetaEntryCache, &pBlock->info.uid, sizeof(uint64_t), pVal, sizeof(STableCachedVal), freeCachedMetaItem, NULL, TAOS_LRU_PRIORITY_LOW);
+      int32_t ret = taosLRUCacheInsert(pCache->pTableMetaEntryCache, &pBlock->info.uid, sizeof(uint64_t), pVal,
+                                       sizeof(STableCachedVal), freeCachedMetaItem, NULL, TAOS_LRU_PRIORITY_LOW);
       if (ret != TAOS_LRU_STATUS_OK) {
         qError("failed to put meta into lru cache, code:%d, %s", ret, idStr);
         freeTableCachedVal(pVal);
@@ -594,13 +593,13 @@ int32_t addTagPseudoColumnData(SReadHandle* pHandle, const SExprInfo* pExpr, int
       taosLRUCacheRelease(pCache->pTableMetaEntryCache, h, false);
     }
 
-    qDebug("retrieve table meta from cache:%"PRIu64", hit:%"PRIu64 " miss:%"PRIu64", %s", pCache->metaFetch, pCache->cacheHit,
+    qDebug("retrieve table meta from cache:%" PRIu64 ", hit:%" PRIu64 " miss:%" PRIu64 ", %s", pCache->metaFetch,
-        (pCache->metaFetch - pCache->cacheHit), idStr);
+           pCache->cacheHit, (pCache->metaFetch - pCache->cacheHit), idStr);
   }
 
   for (int32_t j = 0; j < numOfExpr; ++j) {
     const SExprInfo* pExpr1 = &pExpr[j];
-    int32_t    dstSlotId = pExpr1->base.resSchema.slotId;
+    int32_t          dstSlotId = pExpr1->base.resSchema.slotId;
 
     SColumnInfoData* pColInfoData = taosArrayGet(pBlock->pDataBlock, dstSlotId);
     colInfoDataCleanup(pColInfoData, pBlock->info.rows);
@@ -652,7 +651,7 @@ void setTbNameColData(const SSDataBlock* pBlock, SColumnInfoData* pColInfoData,
   fmGetScalarFuncExecFuncs(functionId, &fpSet);
 
   size_t len = TSDB_TABLE_FNAME_LEN + VARSTR_HEADER_SIZE;
-  char buf[TSDB_TABLE_FNAME_LEN + VARSTR_HEADER_SIZE] = {0};
+  char   buf[TSDB_TABLE_FNAME_LEN + VARSTR_HEADER_SIZE] = {0};
   STR_TO_VARSTR(buf, name)
 
   SColumnInfoData infoData = createColumnInfoData(TSDB_DATA_TYPE_VARCHAR, len, 1);
@@ -820,7 +819,7 @@ static SSDataBlock* doTableScan(SOperatorInfo* pOperator) {
   } else {  // scan table group by group sequentially
     if (pInfo->currentGroupId == -1) {
       if ((++pInfo->currentGroupId) >= tableListGetOutputGroups(pTaskInfo->pTableInfoList)) {
-        doSetOperatorCompleted(pOperator);
+        setOperatorCompleted(pOperator);
         return NULL;
       }
 
@@ -843,7 +842,7 @@ static SSDataBlock* doTableScan(SOperatorInfo* pOperator) {
     }
 
     if ((++pInfo->currentGroupId) >= tableListGetOutputGroups(pTaskInfo->pTableInfoList)) {
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
       return NULL;
     }
 
@@ -865,7 +864,7 @@ static SSDataBlock* doTableScan(SOperatorInfo* pOperator) {
       return result;
     }
 
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
     return NULL;
   }
 }
@@ -904,12 +903,12 @@ SOperatorInfo* createTableScanOperatorInfo(STableScanPhysiNode* pTableScanNode,
     goto _error;
   }
 
-  SScanPhysiNode* pScanNode = &pTableScanNode->scan;
+  SScanPhysiNode*     pScanNode = &pTableScanNode->scan;
   SDataBlockDescNode* pDescNode = pScanNode->node.pOutputDataBlockDesc;
 
   int32_t numOfCols = 0;
-  int32_t code = extractColMatchInfo(pScanNode->pScanCols, pDescNode, &numOfCols, COL_MATCH_FROM_COL_ID,
+  int32_t code =
-                                     &pInfo->matchInfo);
+      extractColMatchInfo(pScanNode->pScanCols, pDescNode, &numOfCols, COL_MATCH_FROM_COL_ID, &pInfo->matchInfo);
   if (code != TSDB_CODE_SUCCESS) {
     goto _error;
   }
@@ -947,22 +946,18 @@ SOperatorInfo* createTableScanOperatorInfo(STableScanPhysiNode* pTableScanNode,
   pInfo->currentGroupId = -1;
   pInfo->assignBlockUid = pTableScanNode->assignBlockUid;
 
-  pOperator->name = "TableScanOperator";  // for debug purpose
+  setOperatorInfo(pOperator, "TableScanOperator", QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN, false, OP_NOT_OPENED, pInfo,
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN;
+                  pTaskInfo);
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
   pOperator->exprSupp.numOfExprs = numOfCols;
-  pOperator->pTaskInfo = pTaskInfo;
 
-  pInfo->metaCache.pTableMetaEntryCache = taosLRUCacheInit(1024*128, -1, .5);
+  pInfo->metaCache.pTableMetaEntryCache = taosLRUCacheInit(1024 * 128, -1, .5);
   if (pInfo->metaCache.pTableMetaEntryCache == NULL) {
     code = terrno;
     goto _error;
   }
 
   taosLRUCacheSetStrictCapacity(pInfo->metaCache.pTableMetaEntryCache, false);
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doTableScan, NULL, NULL, destroyTableScanOperatorInfo,
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doTableScan, NULL, destroyTableScanOperatorInfo,
                                          getTableScannerExecInfo);
 
   // for non-blocking operator, the open cost is always 0
@@ -986,14 +981,9 @@ SOperatorInfo* createTableSeqScanOperatorInfo(void* pReadHandle, SExecTaskInfo*
   pInfo->dataReader = pReadHandle;
   //  pInfo->prevGroupId       = -1;
 
-  pOperator->name = "TableSeqScanOperator";
+  setOperatorInfo(pOperator, "TableSeqScanOperator", QUERY_NODE_PHYSICAL_PLAN_TABLE_SEQ_SCAN, false, OP_NOT_OPENED,
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_TABLE_SEQ_SCAN;
+                  pInfo, pTaskInfo);
-  pOperator->blocking = false;
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doTableScanImpl, NULL, NULL, NULL);
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
-  pOperator->pTaskInfo = pTaskInfo;
-
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doTableScanImpl, NULL, NULL, NULL, NULL);
   return pOperator;
 }
 
@@ -1049,8 +1039,8 @@ static SSDataBlock* doBlockInfoScan(SOperatorInfo* pOperator) {
   SExecTaskInfo*  pTaskInfo = pOperator->pTaskInfo;
 
   STableBlockDistInfo blockDistInfo = {.minRows = INT_MAX, .maxRows = INT_MIN};
-  int32_t code = doGetTableRowSize(pBlockScanInfo->readHandle.meta, pBlockScanInfo->uid, (int32_t*)&blockDistInfo.rowSize,
+  int32_t             code = doGetTableRowSize(pBlockScanInfo->readHandle.meta, pBlockScanInfo->uid,
-                                   GET_TASKID(pTaskInfo));
+                                               (int32_t*)&blockDistInfo.rowSize, GET_TASKID(pTaskInfo));
   if (code != TSDB_CODE_SUCCESS) {
     T_LONG_JMP(pTaskInfo->env, code);
   }
@@ -1148,15 +1138,10 @@ SOperatorInfo* createDataBlockInfoScanOperator(SReadHandle* readHandle, SBlockDi
     goto _error;
   }
 
-  pOperator->name = "DataBlockDistScanOperator";
+  setOperatorInfo(pOperator, "DataBlockDistScanOperator", QUERY_NODE_PHYSICAL_PLAN_BLOCK_DIST_SCAN, false,
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_BLOCK_DIST_SCAN;
+                  OP_NOT_OPENED, pInfo, pTaskInfo);
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
-  pOperator->pTaskInfo = pTaskInfo;
-
   pOperator->fpSet =
-      createOperatorFpSet(operatorDummyOpenFn, doBlockInfoScan, NULL, NULL, destroyBlockDistScanOperatorInfo, NULL);
+      createOperatorFpSet(operatorDummyOpenFn, doBlockInfoScan, NULL, destroyBlockDistScanOperatorInfo, NULL);
   return pOperator;
 
 _error:
@@ -1599,8 +1584,10 @@ static void calBlockTbName(SExprSupp* pTbNameCalSup, SSDataBlock* pBlock) {
   void* pData = colDataGetData(pCol, 0);
   // TODO check tbname validation
   if (pData != (void*)-1 && pData != NULL) {
-    memcpy(pBlock->info.parTbName, varDataVal(pData), TMIN(varDataLen(pData), TSDB_TABLE_NAME_LEN));
+    memset(pBlock->info.parTbName, 0, TSDB_TABLE_NAME_LEN);
-    pBlock->info.parTbName[TSDB_TABLE_NAME_LEN - 1] = 0;
+    int32_t len = TMIN(varDataLen(pData), TSDB_TABLE_NAME_LEN - 1);
+    memcpy(pBlock->info.parTbName, varDataVal(pData), len);
+    /*pBlock->info.parTbName[len + 1] = 0;*/
   } else {
     pBlock->info.parTbName[0] = 0;
   }
@@ -1786,6 +1773,7 @@ static SSDataBlock* doQueueScan(SOperatorInfo* pOperator) {
         tqOffsetResetToLog(&pTaskInfo->streamInfo.prepareStatus, pTaskInfo->streamInfo.snapshotVer);
         qDebug("queue scan tsdb over, switch to wal ver %" PRId64 "", pTaskInfo->streamInfo.snapshotVer + 1);
         if (tqSeekVer(pInfo->tqReader, pTaskInfo->streamInfo.snapshotVer + 1) < 0) {
+          tqOffsetResetToLog(&pTaskInfo->streamInfo.lastStatus, pTaskInfo->streamInfo.snapshotVer);
           return NULL;
         }
         ASSERT(pInfo->tqReader->pWalReader->curVersion == pTaskInfo->streamInfo.snapshotVer + 1);
@@ -2368,11 +2356,10 @@ SOperatorInfo* createRawScanOperatorInfo(SReadHandle* pHandle, SExecTaskInfo* pT
   pInfo->vnode = pHandle->vnode;
 
   pInfo->sContext = pHandle->sContext;
-  pOperator->name = "RawScanOperator";
+  setOperatorInfo(pOperator, "RawScanOperator", QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN, false, OP_NOT_OPENED, pInfo,
-  pOperator->info = pInfo;
+                  pTaskInfo);
-  pOperator->pTaskInfo = pTaskInfo;
 
-  pOperator->fpSet = createOperatorFpSet(NULL, doRawScan, NULL, NULL, destroyRawScanOperatorInfo, NULL);
+  pOperator->fpSet = createOperatorFpSet(NULL, doRawScan, NULL, destroyRawScanOperatorInfo, NULL);
   return pOperator;
 
 _end:
@@ -2385,9 +2372,7 @@ _end:
 static void destroyStreamScanOperatorInfo(void* param) {
   SStreamScanInfo* pStreamScan = (SStreamScanInfo*)param;
   if (pStreamScan->pTableScanOp && pStreamScan->pTableScanOp->info) {
-    STableScanInfo* pTableScanInfo = pStreamScan->pTableScanOp->info;
+    destroyOperatorInfo(pStreamScan->pTableScanOp);
-    destroyTableScanOperatorInfo(pTableScanInfo);
-    taosMemoryFreeClear(pStreamScan->pTableScanOp);
   }
   if (pStreamScan->tqReader) {
     tqCloseReader(pStreamScan->tqReader);
@@ -2556,16 +2541,12 @@ SOperatorInfo* createStreamScanOperatorInfo(SReadHandle* pHandle, STableScanPhys
   pInfo->assignBlockUid = pTableScanNode->assignBlockUid;
   pInfo->partitionSup.needCalc = false;
 
-  pOperator->name = "StreamScanOperator";
+  setOperatorInfo(pOperator, "StreamScanOperator", QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN, false, OP_NOT_OPENED, pInfo,
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN;
+                  pTaskInfo);
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
   pOperator->exprSupp.numOfExprs = taosArrayGetSize(pInfo->pRes->pDataBlock);
-  pOperator->pTaskInfo = pTaskInfo;
 
   __optr_fn_t nextFn = pTaskInfo->execModel == OPTR_EXEC_MODEL_STREAM ? doStreamScan : doQueueScan;
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, nextFn, NULL, NULL, destroyStreamScanOperatorInfo, NULL);
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, nextFn, NULL, destroyStreamScanOperatorInfo, NULL);
 
   return pOperator;
 
@@ -2900,7 +2881,7 @@ static SSDataBlock* sysTableScanUserTags(SOperatorInfo* pOperator) {
     }
     blockDataDestroy(dataBlock);
     pInfo->loadInfo.totalRows += pInfo->pRes->info.rows;
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
     return (pInfo->pRes->info.rows == 0) ? NULL : pInfo->pRes;
   }
 
@@ -2953,7 +2934,7 @@ static SSDataBlock* sysTableScanUserTags(SOperatorInfo* pOperator) {
   if (ret != 0) {
     metaCloseTbCursor(pInfo->pCur);
     pInfo->pCur = NULL;
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
   }
 
   pInfo->loadInfo.totalRows += pInfo->pRes->info.rows;
@@ -3743,7 +3724,7 @@ static SSDataBlock* sysTableBuildUserTablesByUids(SOperatorInfo* pOperator) {
   }
 
   if (i >= taosArrayGetSize(pIdx->uids)) {
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
   } else {
     pIdx->lastIdx = i + 1;
   }
@@ -3925,7 +3906,7 @@ static SSDataBlock* sysTableBuildUserTables(SOperatorInfo* pOperator) {
   if (ret != 0) {
     metaCloseTbCursor(pInfo->pCur);
     pInfo->pCur = NULL;
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
   }
 
   pInfo->loadInfo.totalRows += pInfo->pRes->info.rows;
@@ -3947,7 +3928,7 @@ static SSDataBlock* sysTableScanUserTables(SOperatorInfo* pOperator) {
     doFilterResult(pInfo->pRes, pOperator->exprSupp.pFilterInfo);
     pInfo->loadInfo.totalRows += pInfo->pRes->info.rows;
 
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
     return (pInfo->pRes->info.rows == 0) ? NULL : pInfo->pRes;
   } else {
     if (pInfo->showRewrite == false) {
@@ -4199,15 +4180,10 @@ SOperatorInfo* createSysTableScanOperatorInfo(void* readHandle, SSystemTableScan
     pInfo->readHandle = *(SReadHandle*)readHandle;
   }
 
-  pOperator->name = "SysTableScanOperator";
+  setOperatorInfo(pOperator, "SysTableScanOperator", QUERY_NODE_PHYSICAL_PLAN_SYSTABLE_SCAN, false, OP_NOT_OPENED,
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_SYSTABLE_SCAN;
+                  pInfo, pTaskInfo);
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
   pOperator->exprSupp.numOfExprs = taosArrayGetSize(pInfo->pRes->pDataBlock);
-  pOperator->pTaskInfo = pTaskInfo;
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doSysTableScan, NULL, destroySysScanOperator, NULL);
-
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doSysTableScan, NULL, NULL, destroySysScanOperator, NULL);
   return pOperator;
 
 _error:
@@ -4283,7 +4259,7 @@ static SSDataBlock* doTagScan(SOperatorInfo* pOperator) {
 
     count += 1;
     if (++pInfo->curPos >= size) {
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
     }
   }
 
@@ -4319,7 +4295,7 @@ SOperatorInfo* createTagScanOperatorInfo(SReadHandle* pReadHandle, STagScanPhysi
 
   int32_t    numOfExprs = 0;
   SExprInfo* pExprInfo = createExprInfo(pPhyNode->pScanPseudoCols, NULL, &numOfExprs);
-  int32_t code = initExprSupp(&pOperator->exprSupp, pExprInfo, numOfExprs);
+  int32_t    code = initExprSupp(&pOperator->exprSupp, pExprInfo, numOfExprs);
   if (code != TSDB_CODE_SUCCESS) {
     goto _error;
   }
@@ -4335,18 +4311,12 @@ SOperatorInfo* createTagScanOperatorInfo(SReadHandle* pReadHandle, STagScanPhysi
   pInfo->readHandle = *pReadHandle;
   pInfo->curPos = 0;
 
-  pOperator->name = "TagScanOperator";
+  setOperatorInfo(pOperator, "TagScanOperator", QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN, false, OP_NOT_OPENED, pInfo,
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN;
+                  pTaskInfo);
-
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
-  pOperator->pTaskInfo = pTaskInfo;
-
   initResultSizeInfo(&pOperator->resultInfo, 4096);
   blockDataEnsureCapacity(pInfo->pRes, pOperator->resultInfo.capacity);
 
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doTagScan, NULL, NULL, destroyTagScanOperatorInfo, NULL);
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doTagScan, NULL, destroyTagScanOperatorInfo, NULL);
 
   return pOperator;
 
@@ -4455,7 +4425,7 @@ static int32_t loadDataBlockFromOneTable(SOperatorInfo* pOperator, STableMergeSc
     T_LONG_JMP(pTaskInfo->env, code);
   }
 
-  if (pOperator->exprSupp.pFilterInfo!= NULL) {
+  if (pOperator->exprSupp.pFilterInfo != NULL) {
     int64_t st = taosGetTimestampMs();
     doFilter(pBlock, pOperator->exprSupp.pFilterInfo, &pTableScanInfo->matchInfo);
 
@@ -4713,7 +4683,7 @@ SSDataBlock* doTableMergeScan(SOperatorInfo* pOperator) {
     pInfo->hasGroupId = true;
 
     if (tableListSize == 0) {
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
       return NULL;
     }
     pInfo->tableStartIndex = 0;
@@ -4732,7 +4702,7 @@ SSDataBlock* doTableMergeScan(SOperatorInfo* pOperator) {
     } else {
       stopGroupTableMergeScan(pOperator);
       if (pInfo->tableEndIndex >= tableListSize - 1) {
-        doSetOperatorCompleted(pOperator);
+        setOperatorCompleted(pOperator);
         break;
       }
       pInfo->tableStartIndex = pInfo->tableEndIndex + 1;
@@ -4831,7 +4801,6 @@ SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanN
   pInfo->sample.seed = taosGetTimestampSec();
   pInfo->dataBlockLoadFlag = pTableScanNode->dataRequired;
 
-
   code = filterInitFromNode((SNode*)pTableScanNode->scan.node.pConditions, &pOperator->exprSupp.pFilterInfo, 0);
   if (code != TSDB_CODE_SUCCESS) {
     goto _error;
@@ -4853,16 +4822,12 @@ SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanN
   int32_t rowSize = pInfo->pResBlock->info.rowSize;
   pInfo->bufPageSize = getProperSortPageSize(rowSize);
 
-  pOperator->name = "TableMergeScanOperator";
+  setOperatorInfo(pOperator, "TableMergeScanOperator", QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN, false, OP_NOT_OPENED,
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN;
+                  pInfo, pTaskInfo);
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
   pOperator->exprSupp.numOfExprs = numOfCols;
-  pOperator->pTaskInfo = pTaskInfo;
 
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doTableMergeScan, NULL, NULL,
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doTableMergeScan, NULL, destroyTableMergeScanOperatorInfo,
-                                         destroyTableMergeScanOperatorInfo, getTableMergeScanExplainExecInfo);
+                                         getTableMergeScanExplainExecInfo);
   pOperator->cost.openCost = 0;
   return pOperator;
 

source/libs/executor/src/sortoperator.c

@@ -53,11 +53,7 @@ SOperatorInfo* createSortOperatorInfo(SOperatorInfo* downstream, SSortPhysiNode*
   pInfo->pSortInfo = createSortInfo(pSortNode->pSortKeys);
   initLimitInfo(pSortNode->node.pLimit, pSortNode->node.pSlimit, &pInfo->limitInfo);
 
-  pOperator->name = "SortOperator";
+  setOperatorInfo(pOperator, "SortOperator", QUERY_NODE_PHYSICAL_PLAN_SORT, true, OP_NOT_OPENED, pInfo, pTaskInfo);
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_SORT;
-  pOperator->blocking = true;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
   pOperator->exprSupp.pExprInfo = pExprInfo;
   pOperator->exprSupp.numOfExprs = numOfCols;
 
@@ -67,7 +63,7 @@ SOperatorInfo* createSortOperatorInfo(SOperatorInfo* downstream, SSortPhysiNode*
   // TODO dynamic set the available sort buffer
 
   pOperator->fpSet =
-      createOperatorFpSet(doOpenSortOperator, doSort, NULL, NULL, destroySortOperatorInfo, getExplainExecInfo);
+      createOperatorFpSet(doOpenSortOperator, doSort, NULL, destroySortOperatorInfo, getExplainExecInfo);
 
   code = appendDownstream(pOperator, &downstream, 1);
   if (code != TSDB_CODE_SUCCESS) {
@@ -214,7 +210,7 @@ SSDataBlock* doSort(SOperatorInfo* pOperator) {
     pBlock = getSortedBlockData(pInfo->pSortHandle, pInfo->binfo.pRes, pOperator->resultInfo.capacity,
                                 pInfo->matchInfo.pList, pInfo);
     if (pBlock == NULL) {
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
       return NULL;
     }
 
@@ -428,7 +424,7 @@ SSDataBlock* doGroupSort(SOperatorInfo* pOperator) {
 
     pInfo->prefetchedSortInput = pOperator->pDownstream[0]->fpSet.getNextFn(pOperator->pDownstream[0]);
     if (pInfo->prefetchedSortInput == NULL) {
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
       return NULL;
     }
     pInfo->currGroupId = pInfo->prefetchedSortInput->info.groupId;
@@ -453,7 +449,7 @@ SSDataBlock* doGroupSort(SOperatorInfo* pOperator) {
         beginSortGroup(pOperator);
       } else if (pInfo->childOpStatus == CHILD_OP_FINISHED) {
         finishSortGroup(pOperator);
-        doSetOperatorCompleted(pOperator);
+        setOperatorCompleted(pOperator);
         return NULL;
       }
     }
@@ -509,15 +505,8 @@ SOperatorInfo* createGroupSortOperatorInfo(SOperatorInfo* downstream, SGroupSort
   }
 
   pInfo->pSortInfo = createSortInfo(pSortPhyNode->pSortKeys);
-
+  setOperatorInfo(pOperator, "GroupSortOperator", QUERY_NODE_PHYSICAL_PLAN_GROUP_SORT, false, OP_NOT_OPENED, pInfo, pTaskInfo);
-  pOperator->name = "GroupSortOperator";
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doGroupSort, NULL, destroyGroupSortOperatorInfo,
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_GROUP_SORT;
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
-  pOperator->pTaskInfo = pTaskInfo;
-
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doGroupSort, NULL, NULL, destroyGroupSortOperatorInfo,
                                          getGroupSortExplainExecInfo);
 
   code = appendDownstream(pOperator, &downstream, 1);
@@ -705,7 +694,7 @@ SSDataBlock* doMultiwayMerge(SOperatorInfo* pOperator) {
   if (pBlock != NULL) {
     pOperator->resultInfo.totalRows += pBlock->info.rows;
   } else {
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
   }
 
   return pBlock;
@@ -774,14 +763,8 @@ SOperatorInfo* createMultiwayMergeOperatorInfo(SOperatorInfo** downStreams, size
   pInfo->bufPageSize = getProperSortPageSize(rowSize);
   pInfo->sortBufSize = pInfo->bufPageSize * (numStreams + 1);  // one additional is reserved for merged result.
 
-  pOperator->name = "MultiwayMerge";
+  setOperatorInfo(pOperator, "MultiwayMergeOperator", QUERY_NODE_PHYSICAL_PLAN_MERGE, false, OP_NOT_OPENED, pInfo, pTaskInfo);
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_MERGE;
+  pOperator->fpSet = createOperatorFpSet(doOpenMultiwayMergeOperator, doMultiwayMerge, NULL,
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
-  pOperator->pTaskInfo = pTaskInfo;
-
-  pOperator->fpSet = createOperatorFpSet(doOpenMultiwayMergeOperator, doMultiwayMerge, NULL, NULL,
                                          destroyMultiwayMergeOperatorInfo, getMultiwayMergeExplainExecInfo);
 
   code = appendDownstream(pOperator, downStreams, numStreams);

source/libs/executor/src/tfill.c

@@ -680,9 +680,9 @@ SResultCellData* getResultCell(SResultRowData* pRaw, int32_t index) {
 void* destroyFillColumnInfo(SFillColInfo* pFillCol, int32_t start, int32_t end) {
   for (int32_t i = start; i < end; i++) {
     destroyExprInfo(pFillCol[i].pExpr, 1);
-    taosMemoryFreeClear(pFillCol[i].pExpr);
     taosVariantDestroy(&pFillCol[i].fillVal);
   }
+  taosMemoryFreeClear(pFillCol[start].pExpr);
   taosMemoryFree(pFillCol);
   return NULL;
 }
@@ -1443,7 +1443,7 @@ static SSDataBlock* doStreamFill(SOperatorInfo* pOperator) {
       printDataBlock(pInfo->pRes, "stream fill");
       return pInfo->pRes;
     }
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
     resetStreamFillInfo(pInfo);
     return NULL;
   }
@@ -1512,7 +1512,7 @@ static SSDataBlock* doStreamFill(SOperatorInfo* pOperator) {
   }
 
   if (pInfo->pRes->info.rows == 0) {
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
     resetStreamFillInfo(pInfo);
     return NULL;
   }
@@ -1690,15 +1690,9 @@ SOperatorInfo* createStreamFillOperatorInfo(SOperatorInfo* downstream, SStreamFi
   }
 
   pInfo->srcRowIndex = 0;
-
+  setOperatorInfo(pOperator, "StreamFillOperator", QUERY_NODE_PHYSICAL_PLAN_STREAM_FILL, false, OP_NOT_OPENED, pInfo, pTaskInfo);
-  pOperator->name = "StreamFillOperator";
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_STREAM_FILL;
-  pOperator->info = pInfo;
-  pOperator->pTaskInfo = pTaskInfo;
   pOperator->fpSet =
-      createOperatorFpSet(operatorDummyOpenFn, doStreamFill, NULL, NULL, destroyStreamFillOperatorInfo, NULL);
+      createOperatorFpSet(operatorDummyOpenFn, doStreamFill, NULL, destroyStreamFillOperatorInfo, NULL);
 
   code = appendDownstream(pOperator, &downstream, 1);
   if (code != TSDB_CODE_SUCCESS) {

source/libs/executor/src/timewindowoperator.c

@@ -1221,7 +1221,7 @@ static SSDataBlock* doStateWindowAgg(SOperatorInfo* pOperator) {
 
   pTaskInfo->code = pOperator->fpSet._openFn(pOperator);
   if (pTaskInfo->code != TSDB_CODE_SUCCESS) {
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
     return NULL;
   }
 
@@ -1232,7 +1232,7 @@ static SSDataBlock* doStateWindowAgg(SOperatorInfo* pOperator) {
 
     bool hasRemain = hasRemainResults(&pInfo->groupResInfo);
     if (!hasRemain) {
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
       break;
     }
 
@@ -1269,7 +1269,7 @@ static SSDataBlock* doBuildIntervalResult(SOperatorInfo* pOperator) {
 
     bool hasRemain = hasRemainResults(&pInfo->groupResInfo);
     if (!hasRemain) {
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
       break;
     }
 
@@ -1739,7 +1739,6 @@ SOperatorInfo* createIntervalOperatorInfo(SOperatorInfo* downstream, SIntervalPh
 
   ASSERT(as.calTrigger != STREAM_TRIGGER_MAX_DELAY);
 
-  pOperator->pTaskInfo = pTaskInfo;
   pInfo->win = pTaskInfo->window;
   pInfo->inputOrder = (pPhyNode->window.inputTsOrder == ORDER_ASC) ? TSDB_ORDER_ASC : TSDB_ORDER_DESC;
   pInfo->resultTsOrder = (pPhyNode->window.outputTsOrder == ORDER_ASC) ? TSDB_ORDER_ASC : TSDB_ORDER_DESC;
@@ -1777,15 +1776,11 @@ SOperatorInfo* createIntervalOperatorInfo(SOperatorInfo* downstream, SIntervalPh
   }
 
   initResultRowInfo(&pInfo->binfo.resultRowInfo);
-
+  setOperatorInfo(pOperator, "TimeIntervalAggOperator", QUERY_NODE_PHYSICAL_PLAN_HASH_INTERVAL, true, OP_NOT_OPENED,
-  pOperator->name = "TimeIntervalAggOperator";
+                  pInfo, pTaskInfo);
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_HASH_INTERVAL;
-  pOperator->blocking = true;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
 
   pOperator->fpSet =
-      createOperatorFpSet(doOpenIntervalAgg, doBuildIntervalResult, NULL, NULL, destroyIntervalOperatorInfo, NULL);
+      createOperatorFpSet(doOpenIntervalAgg, doBuildIntervalResult, NULL, destroyIntervalOperatorInfo, NULL);
 
   code = appendDownstream(pOperator, &downstream, 1);
   if (code != TSDB_CODE_SUCCESS) {
@@ -1890,7 +1885,7 @@ static SSDataBlock* doSessionWindowAgg(SOperatorInfo* pOperator) {
 
       bool hasRemain = hasRemainResults(&pInfo->groupResInfo);
       if (!hasRemain) {
-        doSetOperatorCompleted(pOperator);
+        setOperatorCompleted(pOperator);
         break;
       }
 
@@ -1933,7 +1928,7 @@ static SSDataBlock* doSessionWindowAgg(SOperatorInfo* pOperator) {
 
     bool hasRemain = hasRemainResults(&pInfo->groupResInfo);
     if (!hasRemain) {
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
       break;
     }
 
@@ -2281,7 +2276,7 @@ static SSDataBlock* doTimeslice(SOperatorInfo* pOperator) {
       }
 
       if (pSliceInfo->current > pSliceInfo->win.ekey) {
-        doSetOperatorCompleted(pOperator);
+        setOperatorCompleted(pOperator);
         break;
       }
 
@@ -2330,7 +2325,7 @@ static SSDataBlock* doTimeslice(SOperatorInfo* pOperator) {
               }
 
               if (pSliceInfo->current > pSliceInfo->win.ekey) {
-                doSetOperatorCompleted(pOperator);
+                setOperatorCompleted(pOperator);
                 break;
               }
             }
@@ -2342,7 +2337,7 @@ static SSDataBlock* doTimeslice(SOperatorInfo* pOperator) {
           pSliceInfo->current =
               taosTimeAdd(pSliceInfo->current, pInterval->interval, pInterval->intervalUnit, pInterval->precision);
           if (pSliceInfo->current > pSliceInfo->win.ekey) {
-            doSetOperatorCompleted(pOperator);
+            setOperatorCompleted(pOperator);
             break;
           }
         }
@@ -2365,7 +2360,7 @@ static SSDataBlock* doTimeslice(SOperatorInfo* pOperator) {
               }
 
               if (pSliceInfo->current > pSliceInfo->win.ekey) {
-                doSetOperatorCompleted(pOperator);
+                setOperatorCompleted(pOperator);
                 break;
               }
             }
@@ -2386,7 +2381,7 @@ static SSDataBlock* doTimeslice(SOperatorInfo* pOperator) {
               }
 
               if (pSliceInfo->current > pSliceInfo->win.ekey) {
-                doSetOperatorCompleted(pOperator);
+                setOperatorCompleted(pOperator);
                 break;
               }
             } else {
@@ -2448,7 +2443,7 @@ static SSDataBlock* doTimeslice(SOperatorInfo* pOperator) {
                 }
 
                 if (pSliceInfo->current > pSliceInfo->win.ekey) {
-                  doSetOperatorCompleted(pOperator);
+                  setOperatorCompleted(pOperator);
                   break;
                 }
               }
@@ -2463,7 +2458,7 @@ static SSDataBlock* doTimeslice(SOperatorInfo* pOperator) {
         }
 
         if (pSliceInfo->current > pSliceInfo->win.ekey) {
-          doSetOperatorCompleted(pOperator);
+          setOperatorCompleted(pOperator);
           break;
         }
       }
@@ -2557,15 +2552,9 @@ SOperatorInfo* createTimeSliceOperatorInfo(SOperatorInfo* downstream, SPhysiNode
   pScanInfo->cond.twindows = pInfo->win;
   pScanInfo->cond.type = TIMEWINDOW_RANGE_EXTERNAL;
 
-  pOperator->name = "TimeSliceOperator";
+  setOperatorInfo(pOperator, "TimeSliceOperator", QUERY_NODE_PHYSICAL_PLAN_INTERP_FUNC, false, OP_NOT_OPENED, pInfo,
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_INTERP_FUNC;
+                  pTaskInfo);
-  pOperator->blocking = false;
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doTimeslice, NULL, destroyTimeSliceOperatorInfo, NULL);
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
-  pOperator->pTaskInfo = pTaskInfo;
-
-  pOperator->fpSet =
-      createOperatorFpSet(operatorDummyOpenFn, doTimeslice, NULL, NULL, destroyTimeSliceOperatorInfo, NULL);
 
   blockDataEnsureCapacity(pInfo->pRes, pOperator->resultInfo.capacity);
 
@@ -2633,15 +2622,11 @@ SOperatorInfo* createStatewindowOperatorInfo(SOperatorInfo* downstream, SStateWi
   initExecTimeWindowInfo(&pInfo->twAggSup.timeWindowData, &pTaskInfo->window);
 
   pInfo->tsSlotId = tsSlotId;
-  pOperator->name = "StateWindowOperator";
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_MERGE_STATE;
-  pOperator->blocking = true;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->pTaskInfo = pTaskInfo;
-  pOperator->info = pInfo;
 
+  setOperatorInfo(pOperator, "StateWindowOperator", QUERY_NODE_PHYSICAL_PLAN_MERGE_STATE, true, OP_NOT_OPENED, pInfo,
+                  pTaskInfo);
   pOperator->fpSet =
-      createOperatorFpSet(openStateWindowAggOptr, doStateWindowAgg, NULL, NULL, destroyStateWindowOperatorInfo, NULL);
+      createOperatorFpSet(openStateWindowAggOptr, doStateWindowAgg, NULL, destroyStateWindowOperatorInfo, NULL);
 
   code = appendDownstream(pOperator, &downstream, 1);
   if (code != TSDB_CODE_SUCCESS) {
@@ -2711,14 +2696,10 @@ SOperatorInfo* createSessionAggOperatorInfo(SOperatorInfo* downstream, SSessionW
     goto _error;
   }
 
-  pOperator->name = "SessionWindowAggOperator";
+  setOperatorInfo(pOperator, "SessionWindowAggOperator", QUERY_NODE_PHYSICAL_PLAN_MERGE_SESSION, true, OP_NOT_OPENED,
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_MERGE_SESSION;
+                  pInfo, pTaskInfo);
-  pOperator->blocking = true;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
-
   pOperator->fpSet =
-      createOperatorFpSet(operatorDummyOpenFn, doSessionWindowAgg, NULL, NULL, destroySWindowOperatorInfo, NULL);
+      createOperatorFpSet(operatorDummyOpenFn, doSessionWindowAgg, NULL, destroySWindowOperatorInfo, NULL);
   pOperator->pTaskInfo = pTaskInfo;
   code = appendDownstream(pOperator, &downstream, 1);
   if (code != TSDB_CODE_SUCCESS) {
@@ -3134,7 +3115,7 @@ static SSDataBlock* doStreamFinalIntervalAgg(SOperatorInfo* pOperator) {
       return pInfo->binfo.pRes;
     }
 
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
     if (!IS_FINAL_OP(pInfo)) {
       clearFunctionContext(&pOperator->exprSupp);
       // semi interval operator clear disk buffer
@@ -3403,7 +3384,7 @@ SOperatorInfo* createStreamFinalIntervalOperatorInfo(SOperatorInfo* downstream,
   pOperator->info = pInfo;
 
   pOperator->fpSet =
-      createOperatorFpSet(NULL, doStreamFinalIntervalAgg, NULL, NULL, destroyStreamFinalIntervalOperatorInfo, NULL);
+      createOperatorFpSet(NULL, doStreamFinalIntervalAgg, NULL, destroyStreamFinalIntervalOperatorInfo, NULL);
   if (pPhyNode->type == QUERY_NODE_PHYSICAL_PLAN_STREAM_SEMI_INTERVAL) {
     initIntervalDownStream(downstream, pPhyNode->type, &pInfo->aggSup, &pInfo->interval, &pInfo->twAggSup);
   }
@@ -3599,6 +3580,11 @@ static void removeSessionResult(SSHashObj* pHashMap, SSHashObj* pResMap, SSessio
   tSimpleHashRemove(pResMap, &key, sizeof(SSessionKey));
 }
 
+static void getSessionHashKey(const SSessionKey* pKey, SSessionKey* pHashKey) {
+  *pHashKey = *pKey;
+  pHashKey->win.ekey = pKey->win.skey;
+}
+
 static void removeSessionResults(SSHashObj* pHashMap, SArray* pWins) {
   if (tSimpleHashGetSize(pHashMap) == 0) {
     return;
@@ -3607,8 +3593,8 @@ static void removeSessionResults(SSHashObj* pHashMap, SArray* pWins) {
   for (int32_t i = 0; i < size; i++) {
     SSessionKey* pWin = taosArrayGet(pWins, i);
     if (!pWin) continue;
-    SSessionKey key = *pWin;
+    SSessionKey key = {0};
-    key.win.ekey = key.win.skey;
+    getSessionHashKey(pWin, &key);
     tSimpleHashRemove(pHashMap, &key, sizeof(SSessionKey));
   }
 }
@@ -3661,7 +3647,9 @@ static int32_t doOneWindowAggImpl(SColumnInfoData* pTimeWindowData, SResultWindo
 
 static bool doDeleteSessionWindow(SStreamAggSupporter* pAggSup, SSessionKey* pKey) {
   streamStateSessionDel(pAggSup->pState, pKey);
-  tSimpleHashRemove(pAggSup->pResultRows, pKey, sizeof(SSessionKey));
+  SSessionKey hashKey = {0};
+  getSessionHashKey(pKey, &hashKey);
+  tSimpleHashRemove(pAggSup->pResultRows, &hashKey, sizeof(SSessionKey));
   return true;
 }
 
@@ -3772,8 +3760,8 @@ static void doStreamSessionAggImpl(SOperatorInfo* pOperator, SSDataBlock* pSData
       }
     }
     if (pInfo->twAggSup.calTrigger == STREAM_TRIGGER_WINDOW_CLOSE) {
-      SSessionKey key = winInfo.sessionWin;
+      SSessionKey key = {0};
-      key.win.ekey = key.win.skey;
+      getSessionHashKey(&winInfo.sessionWin, &key);
       tSimpleHashPut(pAggSup->pResultRows, &key, sizeof(SSessionKey), &winInfo, sizeof(SResultWindowInfo));
     }
 
@@ -3844,7 +3832,7 @@ static int32_t copyUpdateResult(SSHashObj* pStUpdated, SArray* pUpdated) {
   return TSDB_CODE_SUCCESS;
 }
 
-void doBuildDeleteDataBlock(SSHashObj* pStDeleted, SSDataBlock* pBlock, void** Ite) {
+void doBuildDeleteDataBlock(SOperatorInfo* pOp, SSHashObj* pStDeleted, SSDataBlock* pBlock, void** Ite) {
   blockDataCleanup(pBlock);
   int32_t size = tSimpleHashGetSize(pStDeleted);
   if (size == 0) {
@@ -3870,6 +3858,31 @@ void doBuildDeleteDataBlock(SSHashObj* pStDeleted, SSDataBlock* pBlock, void** I
     colDataAppendNULL(pCalStCol, pBlock->info.rows);
     SColumnInfoData* pCalEdCol = taosArrayGet(pBlock->pDataBlock, CALCULATE_END_TS_COLUMN_INDEX);
     colDataAppendNULL(pCalEdCol, pBlock->info.rows);
+
+    SHashObj* pGroupIdTbNameMap = NULL;
+    if (pOp->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_SESSION ||
+        pOp->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_SEMI_SESSION ||
+        pOp->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_SESSION) {
+      SStreamSessionAggOperatorInfo* pInfo = pOp->info;
+      pGroupIdTbNameMap = pInfo->pGroupIdTbNameMap;
+    } else if (pOp->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_STATE) {
+      SStreamStateAggOperatorInfo* pInfo = pOp->info;
+      pGroupIdTbNameMap = pInfo->pGroupIdTbNameMap;
+    } else {
+      ASSERT(0);
+    }
+
+    char*            tbname = taosHashGet(pGroupIdTbNameMap, &res->groupId, sizeof(int64_t));
+    SColumnInfoData* pTableCol = taosArrayGet(pBlock->pDataBlock, TABLE_NAME_COLUMN_INDEX);
+    if (tbname == NULL) {
+      /*printf("\n\n no tbname for group id %" PRId64 "%p %p\n\n", res->groupId, pOp->info, pGroupIdTbNameMap);*/
+      colDataAppendNULL(pTableCol, pBlock->info.rows);
+    } else {
+      char parTbName[VARSTR_HEADER_SIZE + TSDB_TABLE_NAME_LEN];
+      STR_WITH_MAXSIZE_TO_VARSTR(parTbName, tbname, sizeof(parTbName));
+      colDataAppend(pTableCol, pBlock->info.rows, (const char*)parTbName, false);
+      /*printf("\n\n get tbname %s group id %" PRId64 "\n\n", tbname, res->groupId);*/
+    }
     pBlock->info.rows += 1;
   }
   if ((*Ite) == NULL) {
@@ -3895,8 +3908,8 @@ static void rebuildSessionWindow(SOperatorInfo* pOperator, SArray* pWinArray, SS
       SOperatorInfo*                 pChild = taosArrayGetP(pInfo->pChildren, j);
       SStreamSessionAggOperatorInfo* pChInfo = pChild->info;
       SStreamAggSupporter*           pChAggSup = &pChInfo->streamAggSup;
-      SSessionKey                    chWinKey = *pWinKey;
+      SSessionKey                    chWinKey = {0};
-      chWinKey.win.ekey = chWinKey.win.skey;
+      getSessionHashKey(pWinKey, &chWinKey);
       SStreamStateCur* pCur = streamStateSessionSeekKeyCurrentNext(pChAggSup->pState, &chWinKey);
       SResultRow*      pResult = NULL;
       SResultRow*      pChResult = NULL;
@@ -3977,8 +3990,8 @@ static void copyDeleteWindowInfo(SArray* pResWins, SSHashObj* pStDeleted) {
   for (int32_t i = 0; i < size; i++) {
     SSessionKey* pWinKey = taosArrayGet(pResWins, i);
     if (!pWinKey) continue;
-    SSessionKey winInfo = *pWinKey;
+    SSessionKey winInfo = {0};
-    winInfo.win.ekey = winInfo.win.skey;
+    getSessionHashKey(pWinKey, &winInfo);
     tSimpleHashPut(pStDeleted, &winInfo, sizeof(SSessionKey), NULL, 0);
   }
 }
@@ -4016,7 +4029,7 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
   if (pOperator->status == OP_EXEC_DONE) {
     return NULL;
   } else if (pOperator->status == OP_RES_TO_RETURN) {
-    doBuildDeleteDataBlock(pInfo->pStDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
+    doBuildDeleteDataBlock(pOperator, pInfo->pStDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
     if (pInfo->pDelRes->info.rows > 0) {
       printDataBlock(pInfo->pDelRes, IS_FINAL_OP(pInfo) ? "final session" : "single session");
       return pInfo->pDelRes;
@@ -4027,7 +4040,7 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
       return pBInfo->pRes;
     }
 
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
     return NULL;
   }
 
@@ -4045,7 +4058,8 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
     if (pBlock->info.parTbName[0]) {
       taosHashPut(pInfo->pGroupIdTbNameMap, &pBlock->info.groupId, sizeof(int64_t), &pBlock->info.parTbName,
                   TSDB_TABLE_NAME_LEN);
-      /*printf("\n\n put tbname %s\n\n", pBlock->info.parTbName);*/
+      /*printf("\n\n put tbname %s group id %" PRId64 "\n\n into %p %p", pBlock->info.parTbName, pBlock->info.groupId,*/
+      /*pInfo, pInfo->pGroupIdTbNameMap);*/
     }
 
     if (pBlock->info.parTbName[0]) {
@@ -4121,7 +4135,7 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
   taosMemoryFree(pBuf);
 #endif
 
-  doBuildDeleteDataBlock(pInfo->pStDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
+  doBuildDeleteDataBlock(pOperator, pInfo->pStDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
   if (pInfo->pDelRes->info.rows > 0) {
     printDataBlock(pInfo->pDelRes, IS_FINAL_OP(pInfo) ? "final session" : "single session");
     return pInfo->pDelRes;
@@ -4133,7 +4147,7 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
     return pBInfo->pRes;
   }
 
-  doSetOperatorCompleted(pOperator);
+  setOperatorCompleted(pOperator);
   return NULL;
 }
 
@@ -4200,13 +4214,11 @@ SOperatorInfo* createStreamSessionAggOperatorInfo(SOperatorInfo* downstream, SPh
   pInfo->pGroupIdTbNameMap =
       taosHashInit(1024, taosGetDefaultHashFunction(TSDB_DATA_TYPE_UBIGINT), false, HASH_NO_LOCK);
 
-  pOperator->name = "StreamSessionWindowAggOperator";
+  setOperatorInfo(pOperator, "StreamSessionWindowAggOperator", QUERY_NODE_PHYSICAL_PLAN_STREAM_SESSION, true,
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_STREAM_SESSION;
+                  OP_NOT_OPENED, pInfo, pTaskInfo);
-  pOperator->blocking = true;
+  pOperator->fpSet =
-  pOperator->status = OP_NOT_OPENED;
+      createOperatorFpSet(operatorDummyOpenFn, doStreamSessionAgg, NULL, destroyStreamSessionAggOperatorInfo, NULL);
-  pOperator->info = pInfo;
+
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doStreamSessionAgg, NULL, NULL,
-                                         destroyStreamSessionAggOperatorInfo, NULL);
   if (downstream) {
     initDownStream(downstream, &pInfo->streamAggSup, pInfo->twAggSup.waterMark, pOperator->operatorType,
                    pInfo->primaryTsIndex);
@@ -4247,7 +4259,7 @@ static SSDataBlock* doStreamSessionSemiAgg(SOperatorInfo* pOperator) {
       return pBInfo->pRes;
     }
 
-    doBuildDeleteDataBlock(pInfo->pStDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
+    doBuildDeleteDataBlock(pOperator, pInfo->pStDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
     if (pInfo->pDelRes->info.rows > 0) {
       printDataBlock(pInfo->pDelRes, "semi session delete");
       return pInfo->pDelRes;
@@ -4257,7 +4269,7 @@ static SSDataBlock* doStreamSessionSemiAgg(SOperatorInfo* pOperator) {
       clearFunctionContext(&pOperator->exprSupp);
       // semi interval operator clear disk buffer
       clearStreamSessionOperator(pInfo);
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
       return NULL;
     }
   }
@@ -4327,7 +4339,7 @@ static SSDataBlock* doStreamSessionSemiAgg(SOperatorInfo* pOperator) {
     return pBInfo->pRes;
   }
 
-  doBuildDeleteDataBlock(pInfo->pStDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
+  doBuildDeleteDataBlock(pOperator, pInfo->pStDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
   if (pInfo->pDelRes->info.rows > 0) {
     printDataBlock(pInfo->pDelRes, "semi session delete");
     return pInfo->pDelRes;
@@ -4336,7 +4348,7 @@ static SSDataBlock* doStreamSessionSemiAgg(SOperatorInfo* pOperator) {
   clearFunctionContext(&pOperator->exprSupp);
   // semi interval operator clear disk buffer
   clearStreamSessionOperator(pInfo);
-  doSetOperatorCompleted(pOperator);
+  setOperatorCompleted(pOperator);
   return NULL;
 }
 
@@ -4347,20 +4359,21 @@ SOperatorInfo* createStreamFinalSessionAggOperatorInfo(SOperatorInfo* downstream
   if (pOperator == NULL) {
     goto _error;
   }
+
   SStreamSessionAggOperatorInfo* pInfo = pOperator->info;
 
-  if (pPhyNode->type == QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_SESSION) {
+  pInfo->isFinal = (pPhyNode->type == QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_SESSION);
-    pInfo->isFinal = true;
+  char* name = (pInfo->isFinal) ? "StreamSessionFinalAggOperator" : "StreamSessionSemiAggOperator";
-    pOperator->name = "StreamSessionFinalAggOperator";
+
-  } else {
+  if (pPhyNode->type != QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_SESSION) {
-    pInfo->isFinal = false;
     pInfo->pUpdateRes = createSpecialDataBlock(STREAM_CLEAR);
     blockDataEnsureCapacity(pInfo->pUpdateRes, 128);
-    pOperator->name = "StreamSessionSemiAggOperator";
+    pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doStreamSessionSemiAgg, NULL,
-    pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doStreamSessionSemiAgg, NULL, NULL,
                                            destroyStreamSessionAggOperatorInfo, NULL);
   }
 
+  setOperatorInfo(pOperator, name, pPhyNode->type, false, OP_NOT_OPENED, pInfo, pTaskInfo);
+
   pInfo->pGroupIdTbNameMap =
       taosHashInit(1024, taosGetDefaultHashFunction(TSDB_DATA_TYPE_UBIGINT), false, HASH_NO_LOCK);
 
@@ -4561,8 +4574,8 @@ static void doStreamStateAggImpl(SOperatorInfo* pOperator, SSDataBlock* pSDataBl
     }
 
     if (pInfo->twAggSup.calTrigger == STREAM_TRIGGER_WINDOW_CLOSE) {
-      SSessionKey key = curWin.winInfo.sessionWin;
+      SSessionKey key = {0};
-      key.win.ekey = key.win.skey;
+      getSessionHashKey(&curWin.winInfo.sessionWin, &key);
       tSimpleHashPut(pAggSup->pResultRows, &key, sizeof(SSessionKey), &curWin.winInfo, sizeof(SResultWindowInfo));
     }
   }
@@ -4578,7 +4591,7 @@ static SSDataBlock* doStreamStateAgg(SOperatorInfo* pOperator) {
   SOptrBasicInfo*              pBInfo = &pInfo->binfo;
   int64_t                      maxTs = INT64_MIN;
   if (pOperator->status == OP_RES_TO_RETURN) {
-    doBuildDeleteDataBlock(pInfo->pSeDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
+    doBuildDeleteDataBlock(pOperator, pInfo->pSeDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
     if (pInfo->pDelRes->info.rows > 0) {
       printDataBlock(pInfo->pDelRes, "single state delete");
       return pInfo->pDelRes;
@@ -4590,7 +4603,7 @@ static SSDataBlock* doStreamStateAgg(SOperatorInfo* pOperator) {
       return pBInfo->pRes;
     }
 
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
     return NULL;
   }
 
@@ -4645,7 +4658,13 @@ static SSDataBlock* doStreamStateAgg(SOperatorInfo* pOperator) {
   initGroupResInfoFromArrayList(&pInfo->groupResInfo, pUpdated);
   blockDataEnsureCapacity(pInfo->binfo.pRes, pOperator->resultInfo.capacity);
 
-  doBuildDeleteDataBlock(pInfo->pSeDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
+#if 0
+  char* pBuf = streamStateSessionDump(pInfo->streamAggSup.pState);
+  qDebug("===stream===final session%s", pBuf);
+  taosMemoryFree(pBuf);
+#endif
+
+  doBuildDeleteDataBlock(pOperator, pInfo->pSeDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
   if (pInfo->pDelRes->info.rows > 0) {
     printDataBlock(pInfo->pDelRes, "single state delete");
     return pInfo->pDelRes;
@@ -4656,7 +4675,7 @@ static SSDataBlock* doStreamStateAgg(SOperatorInfo* pOperator) {
     printDataBlock(pBInfo->pRes, "single state");
     return pBInfo->pRes;
   }
-  doSetOperatorCompleted(pOperator);
+  setOperatorCompleted(pOperator);
   return NULL;
 }
 
@@ -4721,14 +4740,10 @@ SOperatorInfo* createStreamStateAggOperatorInfo(SOperatorInfo* downstream, SPhys
   pInfo->pGroupIdTbNameMap =
       taosHashInit(1024, taosGetDefaultHashFunction(TSDB_DATA_TYPE_UBIGINT), false, HASH_NO_LOCK);
 
-  pOperator->name = "StreamStateAggOperator";
+  setOperatorInfo(pOperator, "StreamStateAggOperator", QUERY_NODE_PHYSICAL_PLAN_STREAM_STATE, true, OP_NOT_OPENED,
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_STREAM_STATE;
+                  pInfo, pTaskInfo);
-  pOperator->blocking = true;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->pTaskInfo = pTaskInfo;
-  pOperator->info = pInfo;
   pOperator->fpSet =
-      createOperatorFpSet(operatorDummyOpenFn, doStreamStateAgg, NULL, NULL, destroyStreamStateOperatorInfo, NULL);
+      createOperatorFpSet(operatorDummyOpenFn, doStreamStateAgg, NULL, destroyStreamStateOperatorInfo, NULL);
   initDownStream(downstream, &pInfo->streamAggSup, pInfo->twAggSup.waterMark, pOperator->operatorType,
                  pInfo->primaryTsIndex);
   code = appendDownstream(pOperator, &downstream, 1);
@@ -4876,7 +4891,7 @@ static void doMergeAlignedIntervalAgg(SOperatorInfo* pOperator) {
         cleanupAfterGroupResultGen(pMiaInfo, pRes);
       }
 
-      doSetOperatorCompleted(pOperator);
+      setOperatorCompleted(pOperator);
       break;
     }
 
@@ -5001,16 +5016,11 @@ SOperatorInfo* createMergeAlignedIntervalOperatorInfo(SOperatorInfo* downstream,
 
   initResultRowInfo(&iaInfo->binfo.resultRowInfo);
   blockDataEnsureCapacity(iaInfo->binfo.pRes, pOperator->resultInfo.capacity);
-
+  setOperatorInfo(pOperator, "TimeMergeAlignedIntervalAggOperator", QUERY_NODE_PHYSICAL_PLAN_MERGE_ALIGNED_INTERVAL,
-  pOperator->name = "TimeMergeAlignedIntervalAggOperator";
+                  false, OP_NOT_OPENED, miaInfo, pTaskInfo);
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_MERGE_ALIGNED_INTERVAL;
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->pTaskInfo = pTaskInfo;
-  pOperator->info = miaInfo;
 
   pOperator->fpSet =
-      createOperatorFpSet(operatorDummyOpenFn, mergeAlignedIntervalAgg, NULL, NULL, destroyMAIOperatorInfo, NULL);
+      createOperatorFpSet(operatorDummyOpenFn, mergeAlignedIntervalAgg, NULL, destroyMAIOperatorInfo, NULL);
 
   code = appendDownstream(pOperator, &downstream, 1);
   if (code != TSDB_CODE_SUCCESS) {
@@ -5254,7 +5264,7 @@ static SSDataBlock* doMergeIntervalAgg(SOperatorInfo* pOperator) {
   }
 
   if (pRes->info.rows == 0) {
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
   }
 
   size_t rows = pRes->info.rows;
@@ -5313,16 +5323,10 @@ SOperatorInfo* createMergeIntervalOperatorInfo(SOperatorInfo* downstream, SMerge
   }
 
   initResultRowInfo(&pIntervalInfo->binfo.resultRowInfo);
-
+  setOperatorInfo(pOperator, "TimeMergeIntervalAggOperator", QUERY_NODE_PHYSICAL_PLAN_MERGE_INTERVAL, false,
-  pOperator->name = "TimeMergeIntervalAggOperator";
+                  OP_NOT_OPENED, pMergeIntervalInfo, pTaskInfo);
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_MERGE_INTERVAL;
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->pTaskInfo = pTaskInfo;
-  pOperator->info = pMergeIntervalInfo;
-
   pOperator->fpSet =
-      createOperatorFpSet(operatorDummyOpenFn, doMergeIntervalAgg, NULL, NULL, destroyMergeIntervalOperatorInfo, NULL);
+      createOperatorFpSet(operatorDummyOpenFn, doMergeIntervalAgg, NULL, destroyMergeIntervalOperatorInfo, NULL);
 
   code = appendDownstream(pOperator, &downstream, 1);
   if (code != TSDB_CODE_SUCCESS) {
@@ -5366,7 +5370,7 @@ static SSDataBlock* doStreamIntervalAgg(SOperatorInfo* pOperator) {
     }
     deleteIntervalDiscBuf(pInfo->pState, NULL, pInfo->twAggSup.maxTs - pInfo->twAggSup.deleteMark, &pInfo->interval,
                           &pInfo->delKey);
-    doSetOperatorCompleted(pOperator);
+    setOperatorCompleted(pOperator);
     streamStateCommit(pTaskInfo->streamInfo.pState);
     return NULL;
   }
@@ -5550,13 +5554,10 @@ SOperatorInfo* createStreamIntervalOperatorInfo(SOperatorInfo* downstream, SPhys
   pInfo->pGroupIdTbNameMap =
       taosHashInit(1024, taosGetDefaultHashFunction(TSDB_DATA_TYPE_UBIGINT), false, HASH_NO_LOCK);
 
-  pOperator->name = "StreamIntervalOperator";
+  setOperatorInfo(pOperator, "StreamIntervalOperator", QUERY_NODE_PHYSICAL_PLAN_STREAM_INTERVAL, true, OP_NOT_OPENED,
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_STREAM_INTERVAL;
+                  pInfo, pTaskInfo);
-  pOperator->blocking = true;
+  pOperator->fpSet =
-  pOperator->status = OP_NOT_OPENED;
+      createOperatorFpSet(operatorDummyOpenFn, doStreamIntervalAgg, NULL, destroyStreamFinalIntervalOperatorInfo, NULL);
-  pOperator->info = pInfo;
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doStreamIntervalAgg, NULL, NULL,
-                                         destroyStreamFinalIntervalOperatorInfo, NULL);
 
   initIntervalDownStream(downstream, pPhyNode->type, &pInfo->aggSup, &pInfo->interval, &pInfo->twAggSup);
   code = appendDownstream(pOperator, &downstream, 1);

source/libs/function/src/builtinsimpl.c

@@ -911,6 +911,7 @@ int32_t avgFunction(SqlFunctionCtx* pCtx) {
 
       case TSDB_DATA_TYPE_FLOAT: {
         float* plist = (float*)pCol->pData;
+//        float val = 0;
         for (int32_t i = start; i < numOfRows + pInput->startRowIndex; ++i) {
           if (pCol->hasNull && colDataIsNull_f(pCol->nullbitmap, i)) {
             continue;
@@ -918,8 +919,9 @@ int32_t avgFunction(SqlFunctionCtx* pCtx) {
 
           numOfElem += 1;
           pAvgRes->count += 1;
-          pAvgRes->sum.dsum += plist[i];
+          pAvgRes->sum.dsum  += plist[i];
         }
+//        pAvgRes->sum.dsum = val;
         break;
       }
 
@@ -1278,14 +1280,20 @@ int32_t doMinMaxHelper(SqlFunctionCtx* pCtx, int32_t isMinFunc) {
           pBuf->assign = true;
         } else {
           // ignore the equivalent data value
-          if ((*val) == pData[i]) {
+          // NOTE: An faster version to avoid one additional comparison with FPU.
-            continue;
+          if (isMinFunc) {  // min
-          }
+            if (*val > pData[i]) {
-
+              *val = pData[i];
-          if ((*val < pData[i]) ^ isMinFunc) {
+              if (pCtx->subsidiaries.num > 0) {
-            *val = pData[i];
+                updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
-            if (pCtx->subsidiaries.num > 0) {
+              }
-              updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+            }
+          } else {  // max
+            if (*val < pData[i]) {
+              *val = pData[i];
+              if (pCtx->subsidiaries.num > 0) {
+                updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+              }
             }
           }
         }
@@ -1309,14 +1317,20 @@ int32_t doMinMaxHelper(SqlFunctionCtx* pCtx, int32_t isMinFunc) {
           pBuf->assign = true;
         } else {
           // ignore the equivalent data value
-          if ((*val) == pData[i]) {
+          // NOTE: An faster version to avoid one additional comparison with FPU.
-            continue;
+          if (isMinFunc) {  // min
-          }
+            if (*val > pData[i]) {
-
+              *val = pData[i];
-          if ((*val < pData[i]) ^ isMinFunc) {
+              if (pCtx->subsidiaries.num > 0) {
-            *val = pData[i];
+                updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
-            if (pCtx->subsidiaries.num > 0) {
+              }
-              updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+            }
+          } else {  // max
+            if (*val < pData[i]) {
+              *val = pData[i];
+              if (pCtx->subsidiaries.num > 0) {
+                updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+              }
             }
           }
         }
@@ -1340,14 +1354,20 @@ int32_t doMinMaxHelper(SqlFunctionCtx* pCtx, int32_t isMinFunc) {
           pBuf->assign = true;
         } else {
           // ignore the equivalent data value
-          if ((*val) == pData[i]) {
+          // NOTE: An faster version to avoid one additional comparison with FPU.
-            continue;
+          if (isMinFunc) {  // min
-          }
+            if (*val > pData[i]) {
-
+              *val = pData[i];
-          if ((*val < pData[i]) ^ isMinFunc) {
+              if (pCtx->subsidiaries.num > 0) {
-            *val = pData[i];
+                updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
-            if (pCtx->subsidiaries.num > 0) {
+              }
-              updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+            }
+          } else {  // max
+            if (*val < pData[i]) {
+              *val = pData[i];
+              if (pCtx->subsidiaries.num > 0) {
+                updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+              }
             }
           }
         }
@@ -1371,14 +1391,20 @@ int32_t doMinMaxHelper(SqlFunctionCtx* pCtx, int32_t isMinFunc) {
           pBuf->assign = true;
         } else {
           // ignore the equivalent data value
-          if ((*val) == pData[i]) {
+          // NOTE: An faster version to avoid one additional comparison with FPU.
-            continue;
+          if (isMinFunc) {  // min
-          }
+            if (*val > pData[i]) {
-
+              *val = pData[i];
-          if ((*val < pData[i]) ^ isMinFunc) {
+              if (pCtx->subsidiaries.num > 0) {
-            *val = pData[i];
+                updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
-            if (pCtx->subsidiaries.num > 0) {
+              }
-              updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+            }
+          } else {  // max
+            if (*val < pData[i]) {
+              *val = pData[i];
+              if (pCtx->subsidiaries.num > 0) {
+                updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+              }
             }
           }
         }
@@ -1404,14 +1430,20 @@ int32_t doMinMaxHelper(SqlFunctionCtx* pCtx, int32_t isMinFunc) {
           pBuf->assign = true;
         } else {
           // ignore the equivalent data value
-          if ((*val) == pData[i]) {
+          // NOTE: An faster version to avoid one additional comparison with FPU.
-            continue;
+          if (isMinFunc) {  // min
-          }
+            if (*val > pData[i]) {
-
+              *val = pData[i];
-          if ((*val < pData[i]) ^ isMinFunc) {
+              if (pCtx->subsidiaries.num > 0) {
-            *val = pData[i];
+                updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
-            if (pCtx->subsidiaries.num > 0) {
+              }
-              updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+            }
+          } else {  // max
+            if (*val < pData[i]) {
+              *val = pData[i];
+              if (pCtx->subsidiaries.num > 0) {
+                updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+              }
             }
           }
         }
@@ -1435,14 +1467,20 @@ int32_t doMinMaxHelper(SqlFunctionCtx* pCtx, int32_t isMinFunc) {
           pBuf->assign = true;
         } else {
           // ignore the equivalent data value
-          if ((*val) == pData[i]) {
+          // NOTE: An faster version to avoid one additional comparison with FPU.
-            continue;
+          if (isMinFunc) {  // min
-          }
+            if (*val > pData[i]) {
-
+              *val = pData[i];
-          if ((*val < pData[i]) ^ isMinFunc) {
+              if (pCtx->subsidiaries.num > 0) {
-            *val = pData[i];
+                updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
-            if (pCtx->subsidiaries.num > 0) {
+              }
-              updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+            }
+          } else {  // max
+            if (*val < pData[i]) {
+              *val = pData[i];
+              if (pCtx->subsidiaries.num > 0) {
+                updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+              }
             }
           }
         }
@@ -1466,14 +1504,20 @@ int32_t doMinMaxHelper(SqlFunctionCtx* pCtx, int32_t isMinFunc) {
           pBuf->assign = true;
         } else {
           // ignore the equivalent data value
-          if ((*val) == pData[i]) {
+          // NOTE: An faster version to avoid one additional comparison with FPU.
-            continue;
+          if (isMinFunc) {  // min
-          }
+            if (*val > pData[i]) {
-
+              *val = pData[i];
-          if ((*val < pData[i]) ^ isMinFunc) {
+              if (pCtx->subsidiaries.num > 0) {
-            *val = pData[i];
+                updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
-            if (pCtx->subsidiaries.num > 0) {
+              }
-              updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+            }
+          } else {  // max
+            if (*val < pData[i]) {
+              *val = pData[i];
+              if (pCtx->subsidiaries.num > 0) {
+                updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+              }
             }
           }
         }
@@ -1497,14 +1541,20 @@ int32_t doMinMaxHelper(SqlFunctionCtx* pCtx, int32_t isMinFunc) {
           pBuf->assign = true;
         } else {
           // ignore the equivalent data value
-          if ((*val) == pData[i]) {
+          // NOTE: An faster version to avoid one additional comparison with FPU.
-            continue;
+          if (isMinFunc) {  // min
-          }
+            if (*val > pData[i]) {
-
+              *val = pData[i];
-          if ((*val < pData[i]) ^ isMinFunc) {
+              if (pCtx->subsidiaries.num > 0) {
-            *val = pData[i];
+                updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
-            if (pCtx->subsidiaries.num > 0) {
+              }
-              updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+            }
+          } else {  // max
+            if (*val < pData[i]) {
+              *val = pData[i];
+              if (pCtx->subsidiaries.num > 0) {
+                updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+              }
             }
           }
         }
@@ -1529,14 +1579,20 @@ int32_t doMinMaxHelper(SqlFunctionCtx* pCtx, int32_t isMinFunc) {
         pBuf->assign = true;
       } else {
         // ignore the equivalent data value
-        if ((*val) == pData[i]) {
+        // NOTE: An faster version to avoid one additional comparison with FPU.
-          continue;
+        if (isMinFunc) {  // min
-        }
+          if (*val > pData[i]) {
-
+            *val = pData[i];
-        if ((*val < pData[i]) ^ isMinFunc) {
+            if (pCtx->subsidiaries.num > 0) {
-          *val = pData[i];
+              updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
-          if (pCtx->subsidiaries.num > 0) {
+            }
-            updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+          }
+        } else {  // max
+          if (*val < pData[i]) {
+            *val = pData[i];
+            if (pCtx->subsidiaries.num > 0) {
+              updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+            }
           }
         }
       }
@@ -1559,7 +1615,7 @@ int32_t doMinMaxHelper(SqlFunctionCtx* pCtx, int32_t isMinFunc) {
         }
         pBuf->assign = true;
       } else {
-        // ignore the equivalent data value
+#if 0
         if ((*val) == pData[i]) {
           continue;
         }
@@ -1570,6 +1626,23 @@ int32_t doMinMaxHelper(SqlFunctionCtx* pCtx, int32_t isMinFunc) {
             updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
           }
         }
+#endif
+        // NOTE: An faster version to avoid one additional comparison with FPU.
+        if (isMinFunc) {  // min
+          if (*val > pData[i]) {
+            *val = pData[i];
+            if (pCtx->subsidiaries.num > 0) {
+              updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+            }
+          }
+        } else {  // max
+          if (*val < pData[i]) {
+            *val = pData[i];
+            if (pCtx->subsidiaries.num > 0) {
+              updateTupleData(pCtx, i, pCtx->pSrcBlock, &pBuf->tuplePos);
+            }
+          }
+        }
       }
 
       numOfElems += 1;
@@ -2934,6 +3007,7 @@ int32_t firstFunction(SqlFunctionCtx* pCtx) {
     }
   }
 #else
+  int64_t* pts = (int64_t*) pInput->pPTS->pData;
   for (int32_t i = pInput->startRowIndex; i < pInput->startRowIndex + pInput->numOfRows; ++i) {
     if (pInputCol->hasNull && colDataIsNull(pInputCol, pInput->totalRows, i, pColAgg)) {
       continue;
@@ -2942,13 +3016,14 @@ int32_t firstFunction(SqlFunctionCtx* pCtx) {
     numOfElems++;
 
     char* data = colDataGetData(pInputCol, i);
-    TSKEY cts = getRowPTs(pInput->pPTS, i);
+    TSKEY cts = pts[i];
     if (pResInfo->numOfRes == 0 || pInfo->ts > cts) {
       doSaveCurrentVal(pCtx, i, cts, pInputCol->info.type, data);
       pResInfo->numOfRes = 1;
     }
   }
 #endif
+
   if (numOfElems == 0) {
     // save selectivity value for column consisted of all null values
     firstlastSaveTupleData(pCtx->pSrcBlock, pInput->startRowIndex, pCtx, pInfo);
@@ -3020,26 +3095,87 @@ int32_t lastFunction(SqlFunctionCtx* pCtx) {
     }
   }
 #else
-  for (int32_t i = pInput->startRowIndex; i < pInput->numOfRows + pInput->startRowIndex; ++i) {
+  int64_t* pts = (int64_t*)pInput->pPTS->pData;
-    if (pInputCol->hasNull && colDataIsNull(pInputCol, pInput->totalRows, i, pColAgg)) {
-      continue;
-    }
 
-    numOfElems++;
+#if 0
+    for (int32_t i = pInput->startRowIndex; i < pInput->numOfRows + pInput->startRowIndex; ++i) {
+      if (pInputCol->hasNull && colDataIsNull(pInputCol, pInput->totalRows, i, pColAgg)) {
+        continue;
+      }
 
-    char* data = colDataGetData(pInputCol, i);
+      numOfElems++;
-    TSKEY cts = getRowPTs(pInput->pPTS, i);
+      if (pResInfo->numOfRes == 0 || pInfo->ts < pts[i]) {
-    if (pResInfo->numOfRes == 0 || pInfo->ts < cts) {
+        char* data = colDataGetData(pInputCol, i);
-      doSaveCurrentVal(pCtx, i, cts, type, data);
+        doSaveCurrentVal(pCtx, i, pts[i], type, data);
-      pResInfo->numOfRes = 1;
+        pResInfo->numOfRes = 1;
+      }
+    }
+#else
+    if (!pInputCol->hasNull) {
+      numOfElems = 1;
+
+      int32_t round = pInput->numOfRows >> 2;
+      int32_t reminder = pInput->numOfRows & 0x03;
+
+      int32_t tick = 0;
+      for (int32_t i = pInput->startRowIndex; tick < round; i += 4, tick += 1) {
+        int64_t cts = pts[i];
+        int32_t chosen = i;
+
+        if (cts < pts[i + 1]) {
+          cts = pts[i + 1];
+          chosen = i + 1;
+        }
+
+        if (cts < pts[i + 2]) {
+          cts = pts[i + 2];
+          chosen = i + 2;
+        }
+
+        if (cts < pts[i + 3]) {
+          cts = pts[i + 3];
+          chosen = i + 3;
+        }
+
+        if (pResInfo->numOfRes == 0 || pInfo->ts < cts) {
+          char* data = colDataGetData(pInputCol, chosen);
+          doSaveCurrentVal(pCtx, i, cts, type, data);
+          pResInfo->numOfRes = 1;
+        }
+      }
+
+      for (int32_t i = pInput->startRowIndex + round * 4; i < pInput->startRowIndex + pInput->numOfRows; ++i) {
+        if (pResInfo->numOfRes == 0 || pInfo->ts < pts[i]) {
+          char* data = colDataGetData(pInputCol, i);
+          doSaveCurrentVal(pCtx, i, pts[i], type, data);
+          pResInfo->numOfRes = 1;
+        }
+      }
+    } else {
+      for (int32_t i = pInput->startRowIndex; i < pInput->startRowIndex + pInput->numOfRows; ++i) {
+        if (pInputCol->hasNull && colDataIsNull(pInputCol, pInput->totalRows, i, pColAgg)) {
+          continue;
+        }
+
+        numOfElems++;
+
+        if (pResInfo->numOfRes == 0 || pInfo->ts < pts[i]) {
+          char* data = colDataGetData(pInputCol, i);
+          doSaveCurrentVal(pCtx, i, pts[i], type, data);
+          pResInfo->numOfRes = 1;
+        }
+      }
     }
-  }
 #endif
+
+#endif
+
+  // save selectivity value for column consisted of all null values
   if (numOfElems == 0) {
-    // save selectivity value for column consisted of all null values
     firstlastSaveTupleData(pCtx->pSrcBlock, pInput->startRowIndex, pCtx, pInfo);
   }
-  SET_VAL(pResInfo, numOfElems, 1);
+
+//  SET_VAL(pResInfo, numOfElems, 1);
   return TSDB_CODE_SUCCESS;
 }
 
@@ -3189,8 +3325,8 @@ int32_t lastRowFunction(SqlFunctionCtx* pCtx) {
 #if 0
   int32_t blockDataOrder = (startKey <= endKey) ? TSDB_ORDER_ASC : TSDB_ORDER_DESC;
 
-  // the optimized version only function if all tuples in one block are monotonious increasing or descreasing.
+  // the optimized version only valid if all tuples in one block are monotonious increasing or descreasing.
-  // this is NOT always works if project operator exists in downstream.
+  // this assumption is NOT always works if project operator exists in downstream.
   if (blockDataOrder == TSDB_ORDER_ASC) {
     for (int32_t i = pInput->numOfRows + pInput->startRowIndex - 1; i >= pInput->startRowIndex; --i) {
       char* data = colDataGetData(pInputCol, i);
@@ -3216,11 +3352,13 @@ int32_t lastRowFunction(SqlFunctionCtx* pCtx) {
     }
   }
 #else
+
+  int64_t* pts = (int64_t*)pInput->pPTS->pData;
   for (int32_t i = pInput->startRowIndex; i < pInput->numOfRows + pInput->startRowIndex; ++i) {
     char* data = colDataGetData(pInputCol, i);
-    TSKEY cts = getRowPTs(pInput->pPTS, i);
+    TSKEY cts = pts[i];
-    numOfElems++;
 
+    numOfElems++;
     if (pResInfo->numOfRes == 0 || pInfo->ts < cts) {
       doSaveLastrow(pCtx, data, i, cts, pInfo);
       pResInfo->numOfRes = 1;

source/libs/function/src/tudf.c

@@ -131,7 +131,8 @@ static int32_t udfSpawnUdfd(SUdfdData *pData) {
 
   char   udfdPathLdLib[1024] = {0};
   size_t udfdLdLibPathLen = strlen(tsUdfdLdLibPath);
-  strncpy(udfdPathLdLib, tsUdfdLdLibPath, udfdLdLibPathLen);
+  strncpy(udfdPathLdLib, tsUdfdLdLibPath, tListLen(udfdPathLdLib));
+
   udfdPathLdLib[udfdLdLibPathLen] = ':';
   strncpy(udfdPathLdLib + udfdLdLibPathLen + 1, pathTaosdLdLib, sizeof(udfdPathLdLib) - udfdLdLibPathLen - 1);
   if (udfdLdLibPathLen + taosdLdLibPathLen < 1024) {

source/libs/function/src/udfd.c

@@ -864,6 +864,7 @@ void udfdCtrlAllocBufCb(uv_handle_t *handle, size_t suggested_size, uv_buf_t *bu
 void udfdCtrlReadCb(uv_stream_t *q, ssize_t nread, const uv_buf_t *buf) {
   if (nread < 0) {
     fnError("udfd ctrl pipe read error. %s", uv_err_name(nread));
+    taosMemoryFree(buf->base);
     uv_close((uv_handle_t *)q, NULL);
     uv_stop(global.loop);
     return;

source/libs/nodes/src/nodesUtilFuncs.c

@@ -424,8 +424,9 @@ SNode* nodesMakeNode(ENodeType type) {
     case QUERY_NODE_SHOW_TRANSACTIONS_STMT:
     case QUERY_NODE_SHOW_SUBSCRIPTIONS_STMT:
     case QUERY_NODE_SHOW_TAGS_STMT:
-    case QUERY_NODE_SHOW_TABLE_TAGS_STMT:
       return makeNode(type, sizeof(SShowStmt));
+    case QUERY_NODE_SHOW_TABLE_TAGS_STMT:
+      return makeNode(type, sizeof(SShowTableTagsStmt));
     case QUERY_NODE_SHOW_DNODE_VARIABLES_STMT:
       return makeNode(type, sizeof(SShowDnodeVariablesStmt));
     case QUERY_NODE_SHOW_CREATE_DATABASE_STMT:
@@ -942,13 +943,19 @@ void nodesDestroyNode(SNode* pNode) {
     case QUERY_NODE_SHOW_LOCAL_VARIABLES_STMT:
     case QUERY_NODE_SHOW_TRANSACTIONS_STMT:
     case QUERY_NODE_SHOW_SUBSCRIPTIONS_STMT:
-    case QUERY_NODE_SHOW_TAGS_STMT:
+    case QUERY_NODE_SHOW_TAGS_STMT: {
-    case QUERY_NODE_SHOW_TABLE_TAGS_STMT: {
       SShowStmt* pStmt = (SShowStmt*)pNode;
       nodesDestroyNode(pStmt->pDbName);
       nodesDestroyNode(pStmt->pTbName);
       break;
     }
+    case QUERY_NODE_SHOW_TABLE_TAGS_STMT: {
+      SShowTableTagsStmt* pStmt = (SShowTableTagsStmt*)pNode;
+      nodesDestroyNode(pStmt->pDbName);
+      nodesDestroyNode(pStmt->pTbName);
+      nodesDestroyList(pStmt->pTags);
+      break;
+    }
     case QUERY_NODE_SHOW_DNODE_VARIABLES_STMT:
       nodesDestroyNode(((SShowDnodeVariablesStmt*)pNode)->pDnodeId);
       nodesDestroyNode(((SShowDnodeVariablesStmt*)pNode)->pLikePattern);

source/libs/parser/inc/parAst.h

@@ -178,6 +178,7 @@ SNode* createShowCreateTableStmt(SAstCreateContext* pCxt, ENodeType type, SNode*
 SNode* createShowTableDistributedStmt(SAstCreateContext* pCxt, SNode* pRealTable);
 SNode* createShowDnodeVariablesStmt(SAstCreateContext* pCxt, SNode* pDnodeId, SNode* pLikePattern);
 SNode* createShowVnodesStmt(SAstCreateContext* pCxt, SNode* pDnodeId, SNode* pDnodeEndpoint);
+SNode* createShowTableTagsStmt(SAstCreateContext* pCxt, SNode* pTbName, SNode* pDbName, SNodeList* pTags);
 SNode* createCreateUserStmt(SAstCreateContext* pCxt, SToken* pUserName, const SToken* pPassword, int8_t sysinfo);
 SNode* createAlterUserStmt(SAstCreateContext* pCxt, SToken* pUserName, int8_t alterType, const SToken* pVal);
 SNode* createDropUserStmt(SAstCreateContext* pCxt, SToken* pUserName);

source/libs/parser/inc/parUtil.h

@@ -108,7 +108,7 @@ int32_t getTableMetaFromCache(SParseMetaCache* pMetaCache, const SName* pName, S
 int32_t getDbVgInfoFromCache(SParseMetaCache* pMetaCache, const char* pDbFName, SArray** pVgInfo);
 int32_t getTableVgroupFromCache(SParseMetaCache* pMetaCache, const SName* pName, SVgroupInfo* pVgroup);
 int32_t getDbVgVersionFromCache(SParseMetaCache* pMetaCache, const char* pDbFName, int32_t* pVersion, int64_t* pDbId,
-                                int32_t* pTableNum);
+                                int32_t* pTableNum, int64_t* pStateTs);
 int32_t getDbCfgFromCache(SParseMetaCache* pMetaCache, const char* pDbFName, SDbCfgInfo* pInfo);
 int32_t getUserAuthFromCache(SParseMetaCache* pMetaCache, const char* pUser, const char* pDbFName, AUTH_TYPE type,
                              bool* pPass);

source/libs/parser/inc/sql.y

@@ -426,7 +426,7 @@ cmd ::= SHOW TABLE DISTRIBUTED full_table_name(A).
 cmd ::= SHOW CONSUMERS.                                                           { pCxt->pRootNode = createShowStmt(pCxt, QUERY_NODE_SHOW_CONSUMERS_STMT); }
 cmd ::= SHOW SUBSCRIPTIONS.                                                       { pCxt->pRootNode = createShowStmt(pCxt, QUERY_NODE_SHOW_SUBSCRIPTIONS_STMT); }
 cmd ::= SHOW TAGS FROM table_name_cond(A) from_db_opt(B).                         { pCxt->pRootNode = createShowStmtWithCond(pCxt, QUERY_NODE_SHOW_TAGS_STMT, B, A, OP_TYPE_EQUAL); }
-cmd ::= SHOW TABLE TAGS FROM table_name_cond(A) from_db_opt(B).                   { pCxt->pRootNode = createShowStmtWithCond(pCxt, QUERY_NODE_SHOW_TABLE_TAGS_STMT, B, A, OP_TYPE_EQUAL); }
+cmd ::= SHOW TABLE TAGS tag_list_opt(C) FROM table_name_cond(A) from_db_opt(B).   { pCxt->pRootNode = createShowTableTagsStmt(pCxt, A, B, C); }
 cmd ::= SHOW VNODES NK_INTEGER(A).                                                { pCxt->pRootNode = createShowVnodesStmt(pCxt, createValueNode(pCxt, TSDB_DATA_TYPE_BIGINT, &A), NULL); }
 cmd ::= SHOW VNODES NK_STRING(A).                                                 { pCxt->pRootNode = createShowVnodesStmt(pCxt, NULL, createValueNode(pCxt, TSDB_DATA_TYPE_VARCHAR, &A)); }
 
@@ -441,6 +441,18 @@ table_name_cond(A) ::= table_name(B).
 from_db_opt(A) ::= .                                                              { A = createDefaultDatabaseCondValue(pCxt); }
 from_db_opt(A) ::= FROM db_name(B).                                               { A = createIdentifierValueNode(pCxt, &B); }
 
+%type tag_list_opt                                                                { SNodeList* }
+%destructor tag_list_opt                                                          { nodesDestroyList($$); }
+tag_list_opt(A) ::= .                                                             { A = NULL; }
+tag_list_opt(A) ::= tag_item(B).                                                  { A = createNodeList(pCxt, B); }
+tag_list_opt(A) ::= tag_list_opt(B) NK_COMMA tag_item(C).                         { A = addNodeToList(pCxt, B, C); }
+
+tag_item(A) ::= TBNAME(B).                                                        { A = setProjectionAlias(pCxt, createFunctionNode(pCxt, &B, NULL), &B); }
+tag_item(A) ::= QTAGS(B).                                                         { A = createFunctionNode(pCxt, &B, NULL); }
+tag_item(A) ::= column_name(B).                                                   { A = createColumnNode(pCxt, NULL, &B); }
+tag_item(A) ::= column_name(B) column_alias(C).                                   { A = setProjectionAlias(pCxt, createColumnNode(pCxt, NULL, &B), &C); }
+tag_item(A) ::= column_name(B) AS column_alias(C).                                { A = setProjectionAlias(pCxt, createColumnNode(pCxt, NULL, &B), &C); }
+
 /************************************************ create index ********************************************************/
 cmd ::= CREATE SMA INDEX not_exists_opt(D) 
   full_table_name(A) ON full_table_name(B) index_options(C).                      { pCxt->pRootNode = createCreateIndexStmt(pCxt, INDEX_TYPE_SMA, D, A, B, NULL, C); }

source/libs/parser/src/parAstCreater.c

@@ -1390,6 +1390,21 @@ SNode* createShowVnodesStmt(SAstCreateContext* pCxt, SNode* pDnodeId, SNode* pDn
   return (SNode*)pStmt;
 }
 
+SNode* createShowTableTagsStmt(SAstCreateContext* pCxt, SNode* pTbName, SNode* pDbName, SNodeList* pTags) {
+  CHECK_PARSER_STATUS(pCxt);
+  if (NULL == pDbName) {
+    snprintf(pCxt->pQueryCxt->pMsg, pCxt->pQueryCxt->msgLen, "database not specified");
+    pCxt->errCode = TSDB_CODE_PAR_SYNTAX_ERROR;
+    return NULL;
+  }
+  SShowTableTagsStmt* pStmt = (SShowTableTagsStmt*)nodesMakeNode(QUERY_NODE_SHOW_TABLE_TAGS_STMT);
+  CHECK_OUT_OF_MEM(pStmt);
+  pStmt->pDbName = pDbName;
+  pStmt->pTbName = pTbName;
+  pStmt->pTags = pTags;
+  return (SNode*)pStmt;
+}
+
 SNode* createCreateUserStmt(SAstCreateContext* pCxt, SToken* pUserName, const SToken* pPassword, int8_t sysinfo) {
   CHECK_PARSER_STATUS(pCxt);
   char password[TSDB_USET_PASSWORD_LEN] = {0};

source/libs/parser/src/parAstParser.c

@@ -440,7 +440,7 @@ static int32_t collectMetaKeyFromShowTags(SCollectMetaKeyCxt* pCxt, SShowStmt* p
   return code;
 }
 
-static int32_t collectMetaKeyFromShowStableTags(SCollectMetaKeyCxt* pCxt, SShowStmt* pStmt) {
+static int32_t collectMetaKeyFromShowStableTags(SCollectMetaKeyCxt* pCxt, SShowTableTagsStmt* pStmt) {
   return collectMetaKeyFromRealTableImpl(pCxt, ((SValueNode*)pStmt->pDbName)->literal,
                                          ((SValueNode*)pStmt->pTbName)->literal, AUTH_TYPE_READ);
 }
@@ -627,7 +627,7 @@ static int32_t collectMetaKeyFromQuery(SCollectMetaKeyCxt* pCxt, SNode* pStmt) {
     case QUERY_NODE_SHOW_TAGS_STMT:
       return collectMetaKeyFromShowTags(pCxt, (SShowStmt*)pStmt);
     case QUERY_NODE_SHOW_TABLE_TAGS_STMT:
-      return collectMetaKeyFromShowStableTags(pCxt, (SShowStmt*)pStmt);
+      return collectMetaKeyFromShowStableTags(pCxt, (SShowTableTagsStmt*)pStmt);
     case QUERY_NODE_SHOW_USERS_STMT:
       return collectMetaKeyFromShowUsers(pCxt, (SShowStmt*)pStmt);
     case QUERY_NODE_SHOW_LICENCES_STMT:

source/libs/parser/src/parTranslater.c

@@ -466,14 +466,14 @@ static int32_t getTableHashVgroup(STranslateContext* pCxt, const char* pDbName,
 }
 
 static int32_t getDBVgVersion(STranslateContext* pCxt, const char* pDbFName, int32_t* pVersion, int64_t* pDbId,
-                              int32_t* pTableNum) {
+                              int32_t* pTableNum, int64_t* pStateTs) {
   SParseContext* pParCxt = pCxt->pParseCxt;
   int32_t        code = collectUseDatabaseImpl(pDbFName, pCxt->pDbs);
   if (TSDB_CODE_SUCCESS == code) {
     if (pParCxt->async) {
-      code = getDbVgVersionFromCache(pCxt->pMetaCache, pDbFName, pVersion, pDbId, pTableNum);
+      code = getDbVgVersionFromCache(pCxt->pMetaCache, pDbFName, pVersion, pDbId, pTableNum, pStateTs);
     } else {
-      code = catalogGetDBVgVersion(pParCxt->pCatalog, pDbFName, pVersion, pDbId, pTableNum);
+      code = catalogGetDBVgVersion(pParCxt->pCatalog, pDbFName, pVersion, pDbId, pTableNum, pStateTs);
     }
   }
   if (TSDB_CODE_SUCCESS != code) {
@@ -1252,6 +1252,19 @@ static bool dataTypeEqual(const SDataType* l, const SDataType* r) {
   return (l->type == r->type && l->bytes == r->bytes && l->precision == r->precision && l->scale == r->scale);
 }
 
+// 0 means equal, 1 means the left shall prevail, -1 means the right shall prevail
+static int32_t dataTypeComp(const SDataType* l, const SDataType* r) {
+  if (l->type != r->type) {
+    return 1;
+  }
+
+  if (l->bytes != r->bytes) {
+    return l->bytes > r->bytes ? 1 : -1;
+  }
+
+  return (l->precision == r->precision && l->scale == r->scale) ? 0 : 1;
+}
+
 static EDealRes translateOperator(STranslateContext* pCxt, SOperatorNode* pOp) {
   if (isMultiResFunc(pOp->pLeft)) {
     return generateDealNodeErrMsg(pCxt, TSDB_CODE_PAR_WRONG_VALUE_TYPE, ((SExprNode*)(pOp->pLeft))->aliasName);
@@ -1865,6 +1878,7 @@ static int32_t rewriteIsTrue(SNode* pSrc, SNode** pIsTrue) {
 
 static EDealRes translateCaseWhen(STranslateContext* pCxt, SCaseWhenNode* pCaseWhen) {
   bool   first = true;
+  bool   allNullThen = true;
   SNode* pNode = NULL;
   FOREACH(pNode, pCaseWhen->pWhenThenList) {
     SWhenThenNode* pWhenThen = (SWhenThenNode*)pNode;
@@ -1876,10 +1890,31 @@ static EDealRes translateCaseWhen(STranslateContext* pCxt, SCaseWhenNode* pCaseW
       }
       pWhenThen->pWhen = pIsTrue;
     }
-    if (first) {
+
-      first = false;
+    SExprNode* pThenExpr = (SExprNode*)pNode;
-      pCaseWhen->node.resType = ((SExprNode*)pNode)->resType;
+    if (TSDB_DATA_TYPE_NULL == pThenExpr->resType.type) {
-    } else if (!dataTypeEqual(&pCaseWhen->node.resType, &((SExprNode*)pNode)->resType)) {
+      continue;
+    }
+    allNullThen = false;
+    if (first || dataTypeComp(&pCaseWhen->node.resType, &pThenExpr->resType) < 0) {
+      pCaseWhen->node.resType = pThenExpr->resType;
+    }
+    first = false;
+  }
+
+  if (allNullThen) {
+    if (NULL != pCaseWhen->pElse) {
+      pCaseWhen->node.resType = ((SExprNode*)pCaseWhen->pElse)->resType;
+    } else {
+      pCaseWhen->node.resType.type = TSDB_DATA_TYPE_NULL;
+      pCaseWhen->node.resType.bytes = tDataTypes[TSDB_DATA_TYPE_NULL].bytes;
+      return DEAL_RES_CONTINUE;
+    }
+  }
+
+  FOREACH(pNode, pCaseWhen->pWhenThenList) {
+    SWhenThenNode* pWhenThen = (SWhenThenNode*)pNode;
+    if (!dataTypeEqual(&pCaseWhen->node.resType, &((SExprNode*)pNode)->resType)) {
       SNode* pCastFunc = NULL;
       pCxt->errCode = createCastFunc(pCxt, pWhenThen->pThen, pCaseWhen->node.resType, &pCastFunc);
       if (TSDB_CODE_SUCCESS != pCxt->errCode) {
@@ -1889,6 +1924,7 @@ static EDealRes translateCaseWhen(STranslateContext* pCxt, SCaseWhenNode* pCaseW
       pWhenThen->node.resType = pCaseWhen->node.resType;
     }
   }
+
   if (NULL != pCaseWhen->pElse && !dataTypeEqual(&pCaseWhen->node.resType, &((SExprNode*)pCaseWhen->pElse)->resType)) {
     SNode* pCastFunc = NULL;
     pCxt->errCode = createCastFunc(pCxt, pCaseWhen->pElse, pCaseWhen->node.resType, &pCastFunc);
@@ -2393,6 +2429,9 @@ static int32_t translateTable(STranslateContext* pCxt, SNode* pTable) {
         if (TSDB_SUPER_TABLE == pRealTable->pMeta->tableType) {
           pCxt->stableQuery = true;
         }
+        if (TSDB_SYSTEM_TABLE == pRealTable->pMeta->tableType && isSelectStmt(pCxt->pCurrStmt)) {
+          ((SSelectStmt*)pCxt->pCurrStmt)->isTimeLineResult = false;
+        }
         code = addNamespace(pCxt, pRealTable);
       }
       break;
@@ -3440,7 +3479,8 @@ static int32_t translateSetOperProject(STranslateContext* pCxt, SSetOperator* pS
   FORBOTH(pLeft, pLeftProjections, pRight, pRightProjections) {
     SExprNode* pLeftExpr = (SExprNode*)pLeft;
     SExprNode* pRightExpr = (SExprNode*)pRight;
-    if (!dataTypeEqual(&pLeftExpr->resType, &pRightExpr->resType)) {
+    int32_t    comp = dataTypeComp(&pLeftExpr->resType, &pRightExpr->resType);
+    if (comp > 0) {
       SNode*  pRightFunc = NULL;
       int32_t code = createCastFunc(pCxt, pRight, pLeftExpr->resType, &pRightFunc);
       if (TSDB_CODE_SUCCESS != code) {
@@ -3448,9 +3488,20 @@ static int32_t translateSetOperProject(STranslateContext* pCxt, SSetOperator* pS
       }
       REPLACE_LIST2_NODE(pRightFunc);
       pRightExpr = (SExprNode*)pRightFunc;
+    } else if (comp < 0) {
+      SNode*  pLeftFunc = NULL;
+      int32_t code = createCastFunc(pCxt, pLeft, pRightExpr->resType, &pLeftFunc);
+      if (TSDB_CODE_SUCCESS != code) {
+        return code;
+      }
+      REPLACE_LIST1_NODE(pLeftFunc);
+      SExprNode* pLeftFuncExpr = (SExprNode*)pLeftFunc;
+      snprintf(pLeftFuncExpr->aliasName, sizeof(pLeftFuncExpr->aliasName), "%s", pLeftExpr->aliasName);
+      snprintf(pLeftFuncExpr->userAlias, sizeof(pLeftFuncExpr->userAlias), "%s", pLeftExpr->userAlias);
+      pLeft = pLeftFunc;
+      pLeftExpr = pLeftFuncExpr;
     }
-    strcpy(pRightExpr->aliasName, pLeftExpr->aliasName);
+    snprintf(pRightExpr->aliasName, sizeof(pRightExpr->aliasName), "%s", pLeftExpr->aliasName);
-    pRightExpr->aliasName[strlen(pLeftExpr->aliasName)] = '\0';
     if (TSDB_CODE_SUCCESS != nodesListMakeStrictAppend(&pSetOperator->pProjectionList,
                                                        createSetOperProject(pSetOperator->stmtName, pLeft))) {
       return TSDB_CODE_OUT_OF_MEMORY;
@@ -4941,7 +4992,8 @@ static int32_t translateUseDatabase(STranslateContext* pCxt, SUseDatabaseStmt* p
   SName     name = {0};
   tNameSetDbName(&name, pCxt->pParseCxt->acctId, pStmt->dbName, strlen(pStmt->dbName));
   tNameExtractFullName(&name, usedbReq.db);
-  int32_t code = getDBVgVersion(pCxt, usedbReq.db, &usedbReq.vgVersion, &usedbReq.dbId, &usedbReq.numOfTable);
+  int32_t code =
+      getDBVgVersion(pCxt, usedbReq.db, &usedbReq.vgVersion, &usedbReq.dbId, &usedbReq.numOfTable, &usedbReq.stateTs);
   if (TSDB_CODE_SUCCESS == code) {
     code = buildCmdMsg(pCxt, TDMT_MND_USE_DB, (FSerializeFunc)tSerializeSUseDbReq, &usedbReq);
   }
@@ -6253,21 +6305,29 @@ static SNode* createTagsFunction() {
   return (SNode*)pFunc;
 }
 
+static int32_t createShowTableTagsProjections(SNodeList** pProjections, SNodeList** pTags) {
+  if (NULL != *pTags) {
+    TSWAP(*pProjections, *pTags);
+    return TSDB_CODE_SUCCESS;
+  }
+  int32_t code = nodesListMakeStrictAppend(pProjections, createTbnameFunction());
+  if (TSDB_CODE_SUCCESS == code) {
+    code = nodesListStrictAppend(*pProjections, createTagsFunction());
+  }
+  return code;
+}
+
 static int32_t rewriteShowStableTags(STranslateContext* pCxt, SQuery* pQuery) {
-  const char*  cols[] = {"tbname", "_tags"};
+  SShowTableTagsStmt* pShow = (SShowTableTagsStmt*)pQuery->pRoot;
-  SShowStmt*   pShow = (SShowStmt*)pQuery->pRoot;
+  SSelectStmt*        pSelect = NULL;
-  SSelectStmt* pSelect = NULL;
   int32_t code = createSimpleSelectStmt(((SValueNode*)pShow->pDbName)->literal, ((SValueNode*)pShow->pTbName)->literal,
                                         -1, NULL, &pSelect);
   if (TSDB_CODE_SUCCESS == code) {
-    code = nodesListMakeStrictAppend(&pSelect->pProjectionList, createTbnameFunction());
+    code = createShowTableTagsProjections(&pSelect->pProjectionList, &pShow->pTags);
   }
   if (TSDB_CODE_SUCCESS == code) {
-    code = nodesListMakeStrictAppend(&pSelect->pProjectionList, createTagsFunction());
-  }
-  if (TSDB_CODE_SUCCESS == code) {
-    pSelect->isDistinct = true;
     pQuery->showRewrite = true;
+    pSelect->tagScan = true;
     nodesDestroyNode(pQuery->pRoot);
     pQuery->pRoot = (SNode*)pSelect;
   } else {

source/libs/parser/src/parUtil.c

@@ -138,7 +138,7 @@ static char* getSyntaxErrFormat(int32_t errCode) {
     case TSDB_CODE_PAR_CANNOT_DROP_PRIMARY_KEY:
       return "Primary timestamp column cannot be dropped";
     case TSDB_CODE_PAR_INVALID_MODIFY_COL:
-      return "Only binary/nchar column length could be modified";
+      return "Only binary/nchar column length could be modified, and the length can only be increased, not decreased";
     case TSDB_CODE_PAR_INVALID_TBNAME:
       return "Invalid tbname pseudo column";
     case TSDB_CODE_PAR_INVALID_FUNCTION_NAME:
@@ -876,13 +876,14 @@ int32_t reserveDbVgVersionInCache(int32_t acctId, const char* pDb, SParseMetaCac
 }
 
 int32_t getDbVgVersionFromCache(SParseMetaCache* pMetaCache, const char* pDbFName, int32_t* pVersion, int64_t* pDbId,
-                                int32_t* pTableNum) {
+                                int32_t* pTableNum, int64_t* pStateTs) {
   SDbInfo* pDbInfo = NULL;
   int32_t  code = getMetaDataFromHash(pDbFName, strlen(pDbFName), pMetaCache->pDbInfo, (void**)&pDbInfo);
   if (TSDB_CODE_SUCCESS == code) {
     *pVersion = pDbInfo->vgVer;
     *pDbId = pDbInfo->dbId;
     *pTableNum = pDbInfo->tbNum;
+    *pStateTs = pDbInfo->stateTs;
   }
   return code;
 }

source/libs/parser/test/mockCatalog.cpp

@@ -249,7 +249,7 @@ int32_t __catalogGetTableDistVgInfo(SCatalog* pCtg, SRequestConnInfo* pConn, con
 }
 
 int32_t __catalogGetDBVgVersion(SCatalog* pCtg, const char* dbFName, int32_t* version, int64_t* dbId,
-                                int32_t* tableNum) {
+                                int32_t* tableNum, int64_t* stateTs) {
   return 0;
 }
 

source/libs/parser/test/parSelectTest.cpp

@@ -425,6 +425,8 @@ TEST_F(ParserSelectTest, informationSchema) {
   run("SELECT * FROM ins_databases WHERE name = 'information_schema'");
 
   run("SELECT * FROM ins_tags WHERE db_name = 'test' and table_name = 'st1'");
+
+  run("SELECT * FROM (SELECT table_name FROM ins_tables) t WHERE table_name = 'a'");
 }
 
 TEST_F(ParserSelectTest, withoutFrom) {