Merge remote-tracking branch 'origin/3.0' into fix/TD-20052

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 7321fbb
+        GIT_TAG 16eb34f
         SOURCE_DIR "${TD_SOURCE_DIR}/tools/taos-tools"
         BINARY_DIR ""
         #BUILD_IN_SOURCE TRUE

include/common/tglobal.h

@@ -119,6 +119,7 @@ extern SDiskCfg tsDiskCfg[];
 // udf
 extern bool tsStartUdfd;
 extern char tsUdfdResFuncs[];
+extern char tsUdfdLdLibPath[];
 
 // schemaless
 extern char tsSmlChildTableName[];

include/common/tmsg.h

@@ -57,7 +57,8 @@ extern int32_t tMsgDict[];
 #define TMSG_SEG_SEQ(TYPE)  ((TYPE)&0xff)
 #define TMSG_INFO(TYPE)                                                                                                \
   ((TYPE) < TDMT_DND_MAX_MSG || (TYPE) < TDMT_MND_MAX_MSG || (TYPE) < TDMT_VND_MAX_MSG || (TYPE) < TDMT_SCH_MAX_MSG || \
-   (TYPE) < TDMT_STREAM_MAX_MSG || (TYPE) < TDMT_MON_MAX_MSG || (TYPE) < TDMT_SYNC_MAX_MSG)                            \
+   (TYPE) < TDMT_VND_TMQ_MAX_MSG || (TYPE) < TDMT_STREAM_MAX_MSG || (TYPE) < TDMT_VND_STREAM_MAX_MSG ||                \
+   (TYPE) < TDMT_MON_MAX_MSG || (TYPE) < TDMT_SYNC_MAX_MSG)                                                            \
       ? tMsgInfo[tMsgDict[TMSG_SEG_CODE(TYPE)] + TMSG_SEG_SEQ(TYPE)]                                                   \
       : 0
 #define TMSG_INDEX(TYPE) (tMsgDict[TMSG_SEG_CODE(TYPE)] + TMSG_SEG_SEQ(TYPE))

include/common/tmsgdef.h

@@ -138,17 +138,15 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_MND_GET_TABLE_INDEX, "get-table-index", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_BATCH_META, "batch-meta", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_TABLE_CFG, "table-cfg", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_MND_CREATE_TOPIC, "create-topic", SMCreateTopicReq, SMCreateTopicRsp)
+  TD_DEF_MSG_TYPE(TDMT_MND_TMQ_CREATE_TOPIC, "create-topic", SMCreateTopicReq, SMCreateTopicRsp)
-  TD_DEF_MSG_TYPE(TDMT_MND_ALTER_TOPIC, "alter-topic", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_MND_TMQ_DROP_TOPIC, "drop-topic", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_MND_DROP_TOPIC, "drop-topic", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_MND_TMQ_SUBSCRIBE, "subscribe", SCMSubscribeReq, SCMSubscribeRsp)
-  TD_DEF_MSG_TYPE(TDMT_MND_SUBSCRIBE, "subscribe", SCMSubscribeReq, SCMSubscribeRsp)
+  TD_DEF_MSG_TYPE(TDMT_MND_TMQ_ASK_EP, "ask-ep", SMqAskEpReq, SMqAskEpRsp)
-  TD_DEF_MSG_TYPE(TDMT_MND_MQ_ASK_EP, "ask-ep", SMqAskEpReq, SMqAskEpRsp)
+  TD_DEF_MSG_TYPE(TDMT_MND_TMQ_CONSUMER_LOST, "consumer-lost", SMqConsumerLostMsg, NULL)
-  TD_DEF_MSG_TYPE(TDMT_MND_MQ_CONSUMER_LOST, "consumer-lost", SMqConsumerLostMsg, NULL)
+  TD_DEF_MSG_TYPE(TDMT_MND_TMQ_CONSUMER_RECOVER, "consumer-recover", SMqConsumerRecoverMsg, NULL)
-  TD_DEF_MSG_TYPE(TDMT_MND_MQ_CONSUMER_RECOVER, "consumer-recover", SMqConsumerRecoverMsg, NULL)
+  TD_DEF_MSG_TYPE(TDMT_MND_TMQ_HB, "consumer-hb", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_MND_MQ_HB, "consumer-hb", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_MND_TMQ_DO_REBALANCE, "do-rebalance", SMqDoRebalanceMsg, NULL)
-  TD_DEF_MSG_TYPE(TDMT_MND_MQ_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_MQ_DROP_CGROUP, "drop-cgroup", SMqDropCGroupReq, SMqDropCGroupRsp)
-  TD_DEF_MSG_TYPE(TDMT_MND_MQ_COMMIT_OFFSET, "mnode-commit-offset", SMqCMCommitOffsetReq, SMqCMCommitOffsetRsp)
   TD_DEF_MSG_TYPE(TDMT_MND_MQ_TIMER, "mq-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)
@@ -186,21 +184,6 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_VND_CREATE_STB, "vnode-create-stb", SVCreateStbReq, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_ALTER_STB, "vnode-alter-stb", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_DROP_STB, "vnode-drop-stb", SVDropStbReq, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_MQ_VG_CHANGE, "vnode-mq-vg-change", SMqRebVgReq, SMqRebVgRsp)
-  TD_DEF_MSG_TYPE(TDMT_VND_MQ_VG_DELETE, "vnode-mq-vg-delete", SMqVDeleteReq, SMqVDeleteRsp)
-  TD_DEF_MSG_TYPE(TDMT_VND_MQ_COMMIT_OFFSET, "vnode-commit-offset", STqOffset, STqOffset)
-  TD_DEF_MSG_TYPE(TDMT_VND_ADD_CHECK_INFO, "vnode-add-check-info", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_DELETE_CHECK_INFO, "vnode-delete-check-info", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_CREATE_TOPIC, "vnode-create-topic", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_ALTER_TOPIC, "vnode-alter-topic", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_DROP_TOPIC, "vnode-drop-topic", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_SUBSCRIBE, "vnode-subscribe", SMVSubscribeReq, SMVSubscribeRsp)
-  TD_DEF_MSG_TYPE(TDMT_VND_CONSUME, "vnode-consume", SMqPollReq, SMqDataBlkRsp)
-  TD_DEF_MSG_TYPE(TDMT_VND_STREAM_TRIGGER, "vnode-stream-trigger", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_STREAM_DISPATCH_WRITE, "vnode-stream-task-dispatch-write", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_STREAM_RECOVER_STEP1, "vnode-stream-recover1", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_STREAM_RECOVER_STEP2, "vnode-stream-recover2", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_STREAM_RECOVER_FINISH, "vnode-stream-finish", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_CREATE_SMA, "vnode-create-sma", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_CANCEL_SMA, "vnode-cancel-sma", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_DROP_SMA, "vnode-drop-sma", NULL, NULL)
@@ -232,15 +215,30 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_SCH_LINK_BROKEN, "link-broken", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_SCH_MAX_MSG, "sch-max", NULL, NULL)
 
+  TD_NEW_MSG_SEG(TDMT_VND_TMQ_MSG)
+  TD_DEF_MSG_TYPE(TDMT_VND_TMQ_SUBSCRIBE, "vnode-tmq-subscribe", SMqRebVgReq, SMqRebVgRsp)
+  TD_DEF_MSG_TYPE(TDMT_VND_TMQ_DELETE_SUB, "vnode-tmq-delete-sub", SMqVDeleteReq, SMqVDeleteRsp)
+  TD_DEF_MSG_TYPE(TDMT_VND_TMQ_COMMIT_OFFSET, "vnode-tmq-commit-offset", STqOffset, STqOffset)
+  TD_DEF_MSG_TYPE(TDMT_VND_TMQ_ADD_CHECKINFO, "vnode-tmq-add-checkinfo", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_VND_TMQ_DEL_CHECKINFO, "vnode-del-checkinfo", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_VND_TMQ_CONSUME, "vnode-tmq-consume", SMqPollReq, SMqDataBlkRsp)
+  TD_DEF_MSG_TYPE(TDMT_VND_TMQ_MAX_MSG, "vnd-tmq-max", NULL, NULL)
+
   TD_NEW_MSG_SEG(TDMT_STREAM_MSG)
   TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_DEPLOY, "stream-task-deploy", SStreamTaskDeployReq, SStreamTaskDeployRsp)
   TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_DROP, "stream-task-drop", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_RUN, "stream-task-run", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_DISPATCH, "stream-task-dispatch", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_RECOVER, "stream-task-recover", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_STREAM_RETRIEVE, "stream-retrieve", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_STREAM_RECOVER_FINISH, "vnode-stream-finish", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_STREAM_MAX_MSG, "stream-max", NULL, NULL)
 
+  TD_NEW_MSG_SEG(TDMT_VND_STREAM_MSG)
+  TD_DEF_MSG_TYPE(TDMT_VND_STREAM_TRIGGER, "vnode-stream-trigger", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_VND_STREAM_RECOVER_STEP1, "vnode-stream-recover1", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_VND_STREAM_RECOVER_STEP2, "vnode-stream-recover2", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_VND_STREAM_MAX_MSG, "vnd-stream-max", NULL, NULL)
+
   TD_NEW_MSG_SEG(TDMT_MON_MSG)
   TD_DEF_MSG_TYPE(TDMT_MON_MAX_MSG, "monitor-max", NULL, NULL)
 

include/common/tname.h

@@ -73,11 +73,11 @@ bool tNameTbNameEqual(SName* left, SName* right);
 typedef struct {
   // input
   SArray*     tags;            // element is SSmlKv
-  const char* sTableName;     // super table name
+  const char* stbFullName;     // super table name
-  uint8_t     sTableNameLen;  // the length of super table name
+  uint8_t     stbFullNameLen;  // the length of super table name
 
   // output
-  char*    childTableName;  // must have size of TSDB_TABLE_NAME_LEN;
+  char*    ctbShortName;  // must have size of TSDB_TABLE_NAME_LEN;
   uint64_t uid;           // child table uid, may be useful
 } RandTableName;
 

include/libs/catalog/catalog.h

@@ -309,6 +309,9 @@ int32_t catalogGetUdfInfo(SCatalog* pCtg, SRequestConnInfo* pConn, const char* f
 int32_t catalogChkAuth(SCatalog* pCtg, SRequestConnInfo* pConn, const char* user, const char* dbFName, AUTH_TYPE type,
                        bool* pass);
 
+int32_t catalogChkAuthFromCache(SCatalog* pCtg, SRequestConnInfo* pConn, const char* user, const char* dbFName, AUTH_TYPE type,
+                                       bool* pass, bool* exists);
+
 int32_t catalogUpdateUserAuthInfo(SCatalog* pCtg, SGetUserAuthRsp* pAuth);
 
 int32_t catalogUpdateVgEpSet(SCatalog* pCtg, const char* dbFName, int32_t vgId, SEpSet* epSet);

include/libs/executor/executor.h

@@ -79,6 +79,7 @@ qTaskInfo_t qCreateStreamExecTaskInfo(void* msg, SReadHandle* readers);
  */
 qTaskInfo_t qCreateQueueExecTaskInfo(void* msg, SReadHandle* readers, int32_t* numOfCols, SSchemaWrapper** pSchema);
 
+int32_t qSetStreamOpOpen(qTaskInfo_t tinfo);
 /**
  * Set multiple input data blocks for the stream scan.
  * @param tinfo

include/libs/function/functionMgt.h

@@ -216,6 +216,7 @@ bool fmIsMultiRowsFunc(int32_t funcId);
 bool fmIsKeepOrderFunc(int32_t funcId);
 bool fmIsCumulativeFunc(int32_t funcId);
 bool fmIsInterpPseudoColumnFunc(int32_t funcId);
+bool fmIsGroupKeyFunc(int32_t funcId);
 
 void getLastCacheDataType(SDataType* pType);
 

include/libs/stream/tstream.h

@@ -441,6 +441,7 @@ typedef struct {
 } SStreamRetrieveRsp;
 
 typedef struct {
+  SMsgHead msgHead;
   int64_t  streamId;
   int32_t  taskId;
 } SStreamRecoverStep1Req, SStreamRecoverStep2Req;

include/os/osDef.h

@@ -244,6 +244,16 @@ void syslog(int unused, const char *format, ...);
 #define TD_CHARSET_LEN  64
 #define TD_TIMEZONE_LEN 96
 
+#ifdef WINDOWS
+#define TD_PATH_MAX 260
+#elif defined(PATH_MAX)
+#define TD_PATH_MAX PATH_MAX
+#elif defined(_XOPEN_PATH_MAX)
+#define TD_PATH_MAX _XOPEN_PATH_MAX
+#else
+#define TD_PATH_MAX _POSIX_PATH_MAX
+#endif
+
 #ifdef __cplusplus
 }
 #endif

include/util/taoserror.h

@@ -451,6 +451,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_WAL_OUT_OF_MEMORY             TAOS_DEF_ERROR_CODE(0, 0x1004)
 #define TSDB_CODE_WAL_LOG_NOT_EXIST             TAOS_DEF_ERROR_CODE(0, 0x1005)
 #define TSDB_CODE_WAL_CHKSUM_MISMATCH           TAOS_DEF_ERROR_CODE(0, 0x1006)
+#define TSDB_CODE_WAL_LOG_INCOMPLETE            TAOS_DEF_ERROR_CODE(0, 0x1007)
 
 // tfs
 #define TSDB_CODE_FS_INVLD_CFG                  TAOS_DEF_ERROR_CODE(0, 0x2201)

source/client/src/clientTmq.c

@@ -508,7 +508,7 @@ static int32_t tmqSendCommitReq(tmq_t* tmq, SMqClientVg* pVg, SMqClientTopic* pT
   pMsgSendInfo->param = pParam;
   pMsgSendInfo->paramFreeFp = taosMemoryFree;
   pMsgSendInfo->fp = tmqCommitCb;
-  pMsgSendInfo->msgType = TDMT_VND_MQ_COMMIT_OFFSET;
+  pMsgSendInfo->msgType = TDMT_VND_TMQ_COMMIT_OFFSET;
   // send msg
 
   atomic_add_fetch_32(&pParamSet->waitingRspNum, 1);
@@ -750,7 +750,7 @@ void tmqSendHbReq(void* param, void* tmrId) {
   sendInfo->requestObjRefId = 0;
   sendInfo->param = NULL;
   sendInfo->fp = tmqHbCb;
-  sendInfo->msgType = TDMT_MND_MQ_HB;
+  sendInfo->msgType = TDMT_MND_TMQ_HB;
 
   SEpSet epSet = getEpSet_s(&tmq->pTscObj->pAppInfo->mgmtEp);
 
@@ -1038,7 +1038,7 @@ int32_t tmq_subscribe(tmq_t* tmq, const tmq_list_t* topic_list) {
   sendInfo->requestObjRefId = 0;
   sendInfo->param = &param;
   sendInfo->fp = tmqSubscribeCb;
-  sendInfo->msgType = TDMT_MND_SUBSCRIBE;
+  sendInfo->msgType = TDMT_MND_TMQ_SUBSCRIBE;
 
   SEpSet epSet = getEpSet_s(&tmq->pTscObj->pAppInfo->mgmtEp);
 
@@ -1420,7 +1420,7 @@ int32_t tmqAskEp(tmq_t* tmq, bool async) {
   sendInfo->requestObjRefId = 0;
   sendInfo->param = pParam;
   sendInfo->fp = tmqAskEpCb;
-  sendInfo->msgType = TDMT_MND_MQ_ASK_EP;
+  sendInfo->msgType = TDMT_MND_TMQ_ASK_EP;
 
   SEpSet epSet = getEpSet_s(&tmq->pTscObj->pAppInfo->mgmtEp);
 
@@ -1573,7 +1573,7 @@ int32_t tmqPollImpl(tmq_t* tmq, int64_t timeout) {
       sendInfo->requestObjRefId = 0;
       sendInfo->param = pParam;
       sendInfo->fp = tmqPollCb;
-      sendInfo->msgType = TDMT_VND_CONSUME;
+      sendInfo->msgType = TDMT_VND_TMQ_CONSUME;
 
       int64_t transporterId = 0;
       /*printf("send poll\n");*/

source/common/src/tdatablock.c

@@ -1132,6 +1132,10 @@ void blockDataCleanup(SSDataBlock* pDataBlock) {
   pDataBlock->info.window.ekey = 0;
   pDataBlock->info.window.skey = 0;
 
+  if (pDataBlock->info.capacity == 0) {
+    return;
+  }
+
   size_t numOfCols = taosArrayGetSize(pDataBlock->pDataBlock);
   for (int32_t i = 0; i < numOfCols; ++i) {
     SColumnInfoData* p = taosArrayGet(pDataBlock->pDataBlock, i);
@@ -1186,6 +1190,7 @@ static int32_t doEnsureCapacity(SColumnInfoData* pColumn, const SDataBlockInfo*
 }
 
 void colInfoDataCleanup(SColumnInfoData* pColumn, uint32_t numOfRows) {
+  pColumn->hasNull = false;
   if (IS_VAR_DATA_TYPE(pColumn->info.type)) {
     pColumn->varmeta.length = 0;
     if (pColumn->varmeta.offset != NULL) {
@@ -2063,8 +2068,8 @@ int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SSDataBlock* pDataB
           default:
             if (pColInfoData->info.type < TSDB_DATA_TYPE_MAX && pColInfoData->info.type > TSDB_DATA_TYPE_NULL) {
               if (colDataIsNull_s(pColInfoData, j)) {
-                tdAppendColValToRow(&rb, PRIMARYKEY_TIMESTAMP_COL_ID + k, pCol->type, TD_VTYPE_NULL, NULL, false, offset,
+                tdAppendColValToRow(&rb, PRIMARYKEY_TIMESTAMP_COL_ID + k, pCol->type, TD_VTYPE_NULL, NULL, false,
-                                    k);
+                                    offset, k);
               } else if (pCol->type == pColInfoData->info.type) {
                 tdAppendColValToRow(&rb, PRIMARYKEY_TIMESTAMP_COL_ID + k, pCol->type, TD_VTYPE_NORM, var, true, offset,
                                     k);
@@ -2137,10 +2142,26 @@ int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SSDataBlock* pDataB
   return TSDB_CODE_SUCCESS;
 }
 
-char* buildCtbNameByGroupId(const char* stbName, uint64_t groupId) {
+char* buildCtbNameByGroupId(const char* stbFullName, uint64_t groupId) {
-  ASSERT(stbName[0] != 0);
+  ASSERT(stbFullName[0] != 0);
   SArray* tags = taosArrayInit(0, sizeof(void*));
+  if (tags == NULL) {
+    return NULL;
+  }
+
   SSmlKv* pTag = taosMemoryCalloc(1, sizeof(SSmlKv));
+  if (pTag == NULL) {
+    taosArrayDestroy(tags);
+    return NULL;
+  }
+
+  void* cname = taosMemoryCalloc(1, TSDB_TABLE_NAME_LEN + 1);
+  if (cname == NULL) {
+    taosArrayDestroy(tags);
+    taosMemoryFree(pTag);
+    return NULL;
+  }
+
   pTag->key = "group_id";
   pTag->keyLen = strlen(pTag->key);
   pTag->type = TSDB_DATA_TYPE_UBIGINT;
@@ -2148,13 +2169,11 @@ char* buildCtbNameByGroupId(const char* stbName, uint64_t groupId) {
   pTag->length = sizeof(uint64_t);
   taosArrayPush(tags, &pTag);
 
-  void* cname = taosMemoryCalloc(1, TSDB_TABLE_NAME_LEN + 1);
-
   RandTableName rname = {
       .tags = tags,
-      .sTableName = stbName,
+      .stbFullName = stbFullName,
-      .sTableNameLen = strlen(stbName),
+      .stbFullNameLen = strlen(stbFullName),
-      .childTableName = cname,
+      .ctbShortName = cname,
   };
 
   buildChildTableName(&rname);
@@ -2162,8 +2181,8 @@ char* buildCtbNameByGroupId(const char* stbName, uint64_t groupId) {
   taosMemoryFree(pTag);
   taosArrayDestroy(tags);
 
-  ASSERT(rname.childTableName && rname.childTableName[0]);
+  ASSERT(rname.ctbShortName && rname.ctbShortName[0]);
-  return rname.childTableName;
+  return rname.ctbShortName;
 }
 
 void blockEncode(const SSDataBlock* pBlock, char* data, int32_t* dataLen, int32_t numOfCols, int8_t needCompress) {

source/common/src/tglobal.c

@@ -163,7 +163,8 @@ int32_t tsTtlUnit = 86400;
 int32_t tsTtlPushInterval = 86400;
 int32_t tsGrantHBInterval = 60;
 int32_t tsUptimeInterval = 300;     // seconds
-char    tsUdfdResFuncs[1024] = "";  // udfd resident funcs that teardown when udfd exits
+char    tsUdfdResFuncs[512] = "";  // udfd resident funcs that teardown when udfd exits
+char    tsUdfdLdLibPath[512] = "";
 
 #ifndef _STORAGE
 int32_t taosSetTfsCfg(SConfig *pCfg) {
@@ -424,6 +425,7 @@ 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;
   GRANT_CFG_ADD;
   return 0;
 }
@@ -722,7 +724,7 @@ static int32_t taosSetServerCfg(SConfig *pCfg) {
 
   tsStartUdfd = cfgGetItem(pCfg, "udf")->bval;
   tstrncpy(tsUdfdResFuncs, cfgGetItem(pCfg, "udfdResFuncs")->str, sizeof(tsUdfdResFuncs));
-
+  tstrncpy(tsUdfdLdLibPath, cfgGetItem(pCfg, "udfdLdLibPath")->str, sizeof(tsUdfdLdLibPath));
   if (tsQueryBufferSize >= 0) {
     tsQueryBufferSizeBytes = tsQueryBufferSize * 1048576UL;
   }

source/common/src/tname.c

@@ -315,7 +315,7 @@ static int compareKv(const void* p1, const void* p2) {
  */
 void buildChildTableName(RandTableName* rName) {
   SStringBuilder sb = {0};
-  taosStringBuilderAppendStringLen(&sb, rName->sTableName, rName->sTableNameLen);
+  taosStringBuilderAppendStringLen(&sb, rName->stbFullName, rName->stbFullNameLen);
   taosArraySort(rName->tags, compareKv);
   for (int j = 0; j < taosArrayGetSize(rName->tags); ++j) {
     taosStringBuilderAppendChar(&sb, ',');
@@ -336,11 +336,11 @@ void buildChildTableName(RandTableName* rName) {
   tMD5Final(&context);
 
   char temp[8] = {0};
-  rName->childTableName[0] = 't';
+  rName->ctbShortName[0] = 't';
-  rName->childTableName[1] = '_';
+  rName->ctbShortName[1] = '_';
   for (int i = 0; i < 16; i++) {
     sprintf(temp, "%02x", context.digest[i]);
-    strcat(rName->childTableName, temp);
+    strcat(rName->ctbShortName, temp);
   }
   taosStringBuilderDestroy(&sb);
   rName->uid = *(uint64_t*)(context.digest);

source/dnode/mgmt/CMakeLists.txt

@@ -13,4 +13,18 @@ target_include_directories(
     taosd
     PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/node_mgmt/inc"
 )
+
+IF (TD_LINUX_64 AND JEMALLOC_ENABLED)
+  ADD_DEFINITIONS(-DTD_JEMALLOC_ENABLED -I${CMAKE_BINARY_DIR}/build/include -L${CMAKE_BINARY_DIR}/build/lib -Wl,-rpath,${CMAKE_BINARY_DIR}/build/lib -ljemalloc)
+  SET(LINK_JEMALLOC "-L${CMAKE_BINARY_DIR}/build/lib -ljemalloc")
+ELSE ()
+  SET(LINK_JEMALLOC "")
+ENDIF ()
+
+IF (TD_LINUX_64 AND JEMALLOC_ENABLED)
+    ADD_DEPENDENCIES(taosd jemalloc)
+    target_link_libraries(taosd dnode ${LINK_JEMALLOC})
+ELSE ()
     target_link_libraries(taosd dnode)
+ENDIF ()
+

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

@@ -141,15 +141,13 @@ SArray *mmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_MND_DROP_STREAM, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_MND_GET_INDEX, mmPutMsgToReadQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_MND_GET_TABLE_INDEX, mmPutMsgToReadQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_MND_CREATE_TOPIC, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_MND_TMQ_CREATE_TOPIC, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_MND_ALTER_TOPIC, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_MND_TMQ_DROP_TOPIC, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_MND_DROP_TOPIC, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_MND_TMQ_SUBSCRIBE, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_MND_SUBSCRIBE, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_MND_TMQ_ASK_EP, mmPutMsgToReadQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_MND_MQ_ASK_EP, mmPutMsgToReadQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_MND_TMQ_HB, mmPutMsgToReadQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_MND_MQ_HB, mmPutMsgToReadQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_MND_TMQ_DROP_CGROUP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_MND_MQ_DROP_CGROUP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_MND_TMQ_DROP_CGROUP_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_MND_MQ_DROP_CGROUP_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_MND_MQ_COMMIT_OFFSET, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_MND_KILL_TRANS, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_MND_KILL_QUERY, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_MND_KILL_CONN, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
@@ -171,10 +169,10 @@ SArray *mmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_VND_DROP_STB_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_CREATE_SMA_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_DROP_SMA_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_VG_CHANGE_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_SUBSCRIBE_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_VG_DELETE_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_DELETE_SUB_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_ADD_CHECK_INFO_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_ADD_CHECKINFO_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_DELETE_CHECK_INFO_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_DEL_CHECKINFO_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SCH_DROP_TASK, mmPutMsgToFetchQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DEPLOY_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DROP_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;

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

@@ -72,8 +72,6 @@ SArray *smGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RUN, smPutNodeMsgToStreamQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DISPATCH, smPutNodeMsgToStreamQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DISPATCH_RSP, smPutNodeMsgToStreamQueue, 1) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RECOVER, smPutNodeMsgToStreamQueue, 1) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RECOVER_RSP, smPutNodeMsgToStreamQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_RETRIEVE, smPutNodeMsgToStreamQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_RETRIEVE_RSP, smPutNodeMsgToStreamQueue, 1) == NULL) goto _OVER;
 

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

@@ -392,12 +392,12 @@ SArray *vmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_VND_CANCEL_SMA, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_DROP_SMA, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_SUBMIT_RSMA, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_VG_CHANGE, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_SUBSCRIBE, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_VG_DELETE, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_DELETE_SUB, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_COMMIT_OFFSET, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_COMMIT_OFFSET, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_ADD_CHECK_INFO, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_ADD_CHECKINFO, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_DELETE_CHECK_INFO, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_DEL_CHECKINFO, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_CONSUME, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_CONSUME, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_DELETE, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_BATCH_DEL, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_COMMIT, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
@@ -409,10 +409,9 @@ SArray *vmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RUN, vmPutMsgToStreamQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DISPATCH, vmPutMsgToStreamQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DISPATCH_RSP, vmPutMsgToStreamQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RECOVER, vmPutMsgToStreamQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RECOVER_RSP, vmPutMsgToStreamQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_RETRIEVE, vmPutMsgToStreamQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_RETRIEVE_RSP, vmPutMsgToStreamQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_STREAM_RECOVER_FINISH, vmPutMsgToStreamQueue, 0) == NULL) goto _OVER;
 
   if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_REPLICA, vmPutMsgToMgmtQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_CONFIG, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;

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

@@ -63,12 +63,12 @@ int32_t mndInitConsumer(SMnode *pMnode) {
       .deleteFp = (SdbDeleteFp)mndConsumerActionDelete,
   };
 
-  mndSetMsgHandle(pMnode, TDMT_MND_SUBSCRIBE, mndProcessSubscribeReq);
+  mndSetMsgHandle(pMnode, TDMT_MND_TMQ_SUBSCRIBE, mndProcessSubscribeReq);
-  mndSetMsgHandle(pMnode, TDMT_MND_MQ_HB, mndProcessMqHbReq);
+  mndSetMsgHandle(pMnode, TDMT_MND_TMQ_HB, mndProcessMqHbReq);
-  mndSetMsgHandle(pMnode, TDMT_MND_MQ_ASK_EP, mndProcessAskEpReq);
+  mndSetMsgHandle(pMnode, TDMT_MND_TMQ_ASK_EP, mndProcessAskEpReq);
   mndSetMsgHandle(pMnode, TDMT_MND_MQ_TIMER, mndProcessMqTimerMsg);
-  mndSetMsgHandle(pMnode, TDMT_MND_MQ_CONSUMER_LOST, mndProcessConsumerLostMsg);
+  mndSetMsgHandle(pMnode, TDMT_MND_TMQ_CONSUMER_LOST, mndProcessConsumerLostMsg);
-  mndSetMsgHandle(pMnode, TDMT_MND_MQ_CONSUMER_RECOVER, mndProcessConsumerRecoverMsg);
+  mndSetMsgHandle(pMnode, TDMT_MND_TMQ_CONSUMER_RECOVER, mndProcessConsumerRecoverMsg);
 
   mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_CONSUMERS, mndRetrieveConsumer);
   mndAddShowFreeIterHandle(pMnode, TSDB_MGMT_TABLE_CONSUMERS, mndCancelGetNextConsumer);
@@ -207,7 +207,7 @@ static int32_t mndProcessMqTimerMsg(SRpcMsg *pMsg) {
 
       pLostMsg->consumerId = pConsumer->consumerId;
       SRpcMsg pRpcMsg = {
-          .msgType = TDMT_MND_MQ_CONSUMER_LOST,
+          .msgType = TDMT_MND_TMQ_CONSUMER_LOST,
           .pCont = pLostMsg,
           .contLen = sizeof(SMqConsumerLostMsg),
       };
@@ -256,7 +256,7 @@ static int32_t mndProcessMqTimerMsg(SRpcMsg *pMsg) {
   if (taosHashGetSize(pRebMsg->rebSubHash) != 0) {
     mInfo("mq rebalance will be triggered");
     SRpcMsg rpcMsg = {
-        .msgType = TDMT_MND_MQ_DO_REBALANCE,
+        .msgType = TDMT_MND_TMQ_DO_REBALANCE,
         .pCont = pRebMsg,
         .contLen = sizeof(SMqDoRebalanceMsg),
     };
@@ -292,7 +292,7 @@ static int32_t mndProcessMqHbReq(SRpcMsg *pMsg) {
 
     pRecoverMsg->consumerId = consumerId;
     SRpcMsg pRpcMsg = {
-        .msgType = TDMT_MND_MQ_CONSUMER_RECOVER,
+        .msgType = TDMT_MND_TMQ_CONSUMER_RECOVER,
         .pCont = pRecoverMsg,
         .contLen = sizeof(SMqConsumerRecoverMsg),
     };
@@ -331,7 +331,7 @@ static int32_t mndProcessAskEpReq(SRpcMsg *pMsg) {
 
     pRecoverMsg->consumerId = consumerId;
     SRpcMsg pRpcMsg = {
-        .msgType = TDMT_MND_MQ_CONSUMER_RECOVER,
+        .msgType = TDMT_MND_TMQ_CONSUMER_RECOVER,
         .pCont = pRecoverMsg,
         .contLen = sizeof(SMqConsumerRecoverMsg),
     };

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

@@ -487,6 +487,7 @@ int32_t mndScheduleStream(SMnode* pMnode, SStreamObj* pStream) {
         qDestroyQueryPlan(pPlan);
         return -1;
       }
+      pTask->fillHistory = pStream->fillHistory;
       mndAddTaskToTaskSet(taskOneLevel, pTask);
 
       // source

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

@@ -557,78 +557,6 @@ static int32_t mndPersistTaskDropReq(STrans *pTrans, SStreamTask *pTask) {
   return 0;
 }
 
-#if 0
-static int32_t mndPersistTaskRecoverReq(STrans *pTrans, SStreamTask *pTask) {
-  SMStreamTaskRecoverReq *pReq = taosMemoryCalloc(1, sizeof(SMStreamTaskRecoverReq));
-  if (pReq == NULL) {
-    terrno = TSDB_CODE_OUT_OF_MEMORY;
-    return -1;
-  }
-  pReq->streamId = pTask->streamId;
-  pReq->taskId = pTask->taskId;
-  int32_t len;
-  int32_t code;
-  tEncodeSize(tEncodeSMStreamTaskRecoverReq, pReq, len, code);
-  if (code != 0) {
-    return -1;
-  }
-  void *buf = taosMemoryCalloc(1, sizeof(SMsgHead) + len);
-  if (buf == NULL) {
-    return -1;
-  }
-  void    *abuf = POINTER_SHIFT(buf, sizeof(SMsgHead));
-  SEncoder encoder;
-  tEncoderInit(&encoder, abuf, len);
-  tEncodeSMStreamTaskRecoverReq(&encoder, pReq);
-  ((SMsgHead *)buf)->vgId = pTask->nodeId;
-
-  STransAction action = {0};
-  memcpy(&action.epSet, &pTask->epSet, sizeof(SEpSet));
-  action.pCont = buf;
-  action.contLen = sizeof(SMsgHead) + len;
-  action.msgType = TDMT_STREAM_TASK_RECOVER;
-  if (mndTransAppendRedoAction(pTrans, &action) != 0) {
-    taosMemoryFree(buf);
-    return -1;
-  }
-  return 0;
-}
-
-int32_t mndRecoverStreamTasks(SMnode *pMnode, STrans *pTrans, SStreamObj *pStream) {
-  if (pStream->isDistributed) {
-    int32_t lv = taosArrayGetSize(pStream->tasks);
-    for (int32_t i = 0; i < lv; i++) {
-      SArray      *pTasks = taosArrayGetP(pStream->tasks, i);
-      int32_t      sz = taosArrayGetSize(pTasks);
-      SStreamTask *pTask = taosArrayGetP(pTasks, 0);
-      if (pTask->taskLevel == TASK_LEVEL__AGG) {
-        ASSERT(sz == 1);
-        if (mndPersistTaskRecoverReq(pTrans, pTask) < 0) {
-          return -1;
-        }
-      } else {
-        continue;
-      }
-    }
-  } else {
-    int32_t lv = taosArrayGetSize(pStream->tasks);
-    for (int32_t i = 0; i < lv; i++) {
-      SArray *pTasks = taosArrayGetP(pStream->tasks, i);
-      int32_t sz = taosArrayGetSize(pTasks);
-      for (int32_t j = 0; j < sz; j++) {
-        SStreamTask *pTask = taosArrayGetP(pTasks, j);
-        if (pTask->taskLevel != TASK_LEVEL__SOURCE) break;
-        ASSERT(pTask->taskLevel != TASK_LEVEL__SINK);
-        if (mndPersistTaskRecoverReq(pTrans, pTask) < 0) {
-          return -1;
-        }
-      }
-    }
-  }
-  return 0;
-}
-#endif
-
 int32_t mndDropStreamTasks(SMnode *pMnode, STrans *pTrans, SStreamObj *pStream) {
   int32_t lv = taosArrayGetSize(pStream->tasks);
   for (int32_t i = 0; i < lv; i++) {

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

@@ -73,11 +73,11 @@ int32_t mndInitSubscribe(SMnode *pMnode) {
       .deleteFp = (SdbDeleteFp)mndSubActionDelete,
   };
 
-  mndSetMsgHandle(pMnode, TDMT_VND_MQ_VG_CHANGE_RSP, mndTransProcessRsp);
+  mndSetMsgHandle(pMnode, TDMT_VND_TMQ_SUBSCRIBE_RSP, mndTransProcessRsp);
-  mndSetMsgHandle(pMnode, TDMT_VND_MQ_VG_DELETE_RSP, mndTransProcessRsp);
+  mndSetMsgHandle(pMnode, TDMT_VND_TMQ_DELETE_SUB_RSP, mndTransProcessRsp);
-  mndSetMsgHandle(pMnode, TDMT_MND_MQ_DO_REBALANCE, mndProcessRebalanceReq);
+  mndSetMsgHandle(pMnode, TDMT_MND_TMQ_DO_REBALANCE, mndProcessRebalanceReq);
-  mndSetMsgHandle(pMnode, TDMT_MND_MQ_DROP_CGROUP, mndProcessDropCgroupReq);
+  mndSetMsgHandle(pMnode, TDMT_MND_TMQ_DROP_CGROUP, mndProcessDropCgroupReq);
-  mndSetMsgHandle(pMnode, TDMT_MND_MQ_DROP_CGROUP_RSP, mndTransProcessRsp);
+  mndSetMsgHandle(pMnode, TDMT_MND_TMQ_DROP_CGROUP_RSP, mndTransProcessRsp);
 
   mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_SUBSCRIPTIONS, mndRetrieveSubscribe);
   mndAddShowFreeIterHandle(pMnode, TSDB_MGMT_TABLE_TOPICS, mndCancelGetNextSubscribe);
@@ -164,7 +164,7 @@ static int32_t mndPersistSubChangeVgReq(SMnode *pMnode, STrans *pTrans, const SM
   action.epSet = mndGetVgroupEpset(pMnode, pVgObj);
   action.pCont = buf;
   action.contLen = tlen;
-  action.msgType = TDMT_VND_MQ_VG_CHANGE;
+  action.msgType = TDMT_VND_TMQ_SUBSCRIBE;
 
   mndReleaseVgroup(pMnode, pVgObj);
   if (mndTransAppendRedoAction(pTrans, &action) != 0) {
@@ -920,7 +920,7 @@ int32_t mndDropSubByTopic(SMnode *pMnode, STrans *pTrans, const char *topicName)
       action.epSet = pVgEp->epSet;
       action.pCont = pReq;
       action.contLen = sizeof(SMqVDeleteReq);
-      action.msgType = TDMT_VND_MQ_VG_DELETE;
+      action.msgType = TDMT_VND_TMQ_DELETE_SUB;
       if (mndTransAppendRedoAction(pTrans, &action) != 0) {
         taosMemoryFree(pReq);
         return -1;

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

@@ -17,11 +17,44 @@
 #include "mndSync.h"
 #include "mndTrans.h"
 
-static int32_t mndSyncEqMsg(const SMsgCb *msgcb, SRpcMsg *pMsg) {
+static int32_t mndSyncEqCtrlMsg(const SMsgCb *msgcb, SRpcMsg *pMsg) {
+  if (pMsg == NULL || pMsg->pCont == NULL) {
+    return -1;
+  }
+
   SMsgHead *pHead = pMsg->pCont;
   pHead->contLen = htonl(pHead->contLen);
   pHead->vgId = htonl(pHead->vgId);
 
+  if (msgcb == NULL || msgcb->putToQueueFp == NULL) {
+    rpcFreeCont(pMsg->pCont);
+    pMsg->pCont = NULL;
+    return -1;
+  }
+
+  int32_t code = tmsgPutToQueue(msgcb, SYNC_CTRL_QUEUE, pMsg);
+  if (code != 0) {
+    rpcFreeCont(pMsg->pCont);
+    pMsg->pCont = NULL;
+  }
+  return code;
+}
+
+static int32_t mndSyncEqMsg(const SMsgCb *msgcb, SRpcMsg *pMsg) {
+  if (pMsg == NULL || pMsg->pCont == NULL) {
+    return -1;
+  }
+
+  SMsgHead *pHead = pMsg->pCont;
+  pHead->contLen = htonl(pHead->contLen);
+  pHead->vgId = htonl(pHead->vgId);
+
+  if (msgcb == NULL || msgcb->putToQueueFp == NULL) {
+    rpcFreeCont(pMsg->pCont);
+    pMsg->pCont = NULL;
+    return -1;
+  }
+
   int32_t code = tmsgPutToQueue(msgcb, SYNC_QUEUE, pMsg);
   if (code != 0) {
     rpcFreeCont(pMsg->pCont);
@@ -204,7 +237,7 @@ int32_t mndInitSync(SMnode *pMnode) {
       .msgcb = NULL,
       .syncSendMSg = mndSyncSendMsg,
       .syncEqMsg = mndSyncEqMsg,
-      .syncEqCtrlMsg = NULL,
+      .syncEqCtrlMsg = mndSyncEqCtrlMsg,
       .pingMs = 5000,
       .electMs = 3000,
       .heartbeatMs = 500,

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

@@ -53,11 +53,10 @@ int32_t mndInitTopic(SMnode *pMnode) {
       .deleteFp = (SdbDeleteFp)mndTopicActionDelete,
   };
 
-  mndSetMsgHandle(pMnode, TDMT_MND_CREATE_TOPIC, mndProcessCreateTopicReq);
+  mndSetMsgHandle(pMnode, TDMT_MND_TMQ_CREATE_TOPIC, mndProcessCreateTopicReq);
-  mndSetMsgHandle(pMnode, TDMT_MND_DROP_TOPIC, mndProcessDropTopicReq);
+  mndSetMsgHandle(pMnode, TDMT_MND_TMQ_DROP_TOPIC, mndProcessDropTopicReq);
-  mndSetMsgHandle(pMnode, TDMT_VND_DROP_TOPIC_RSP, mndTransProcessRsp);
+  mndSetMsgHandle(pMnode, TDMT_VND_TMQ_ADD_CHECKINFO_RSP, mndTransProcessRsp);
-  mndSetMsgHandle(pMnode, TDMT_VND_ADD_CHECK_INFO_RSP, mndTransProcessRsp);
+  mndSetMsgHandle(pMnode, TDMT_VND_TMQ_DEL_CHECKINFO_RSP, mndTransProcessRsp);
-  mndSetMsgHandle(pMnode, TDMT_VND_DELETE_CHECK_INFO_RSP, mndTransProcessRsp);
 
   mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_TOPICS, mndRetrieveTopic);
   mndAddShowFreeIterHandle(pMnode, TSDB_MGMT_TABLE_TOPICS, mndCancelGetNextTopic);
@@ -506,7 +505,7 @@ static int32_t mndCreateTopic(SMnode *pMnode, SRpcMsg *pReq, SCMCreateTopicReq *
       action.epSet = mndGetVgroupEpset(pMnode, pVgroup);
       action.pCont = buf;
       action.contLen = sizeof(SMsgHead) + len;
-      action.msgType = TDMT_VND_ADD_CHECK_INFO;
+      action.msgType = TDMT_VND_TMQ_ADD_CHECKINFO;
       if (mndTransAppendRedoAction(pTrans, &action) != 0) {
         taosMemoryFree(buf);
         sdbRelease(pSdb, pVgroup);
@@ -715,7 +714,7 @@ static int32_t mndProcessDropTopicReq(SRpcMsg *pReq) {
       action.epSet = mndGetVgroupEpset(pMnode, pVgroup);
       action.pCont = buf;
       action.contLen = sizeof(SMsgHead) + TSDB_TOPIC_FNAME_LEN;
-      action.msgType = TDMT_VND_DELETE_CHECK_INFO;
+      action.msgType = TDMT_VND_TMQ_DEL_CHECKINFO;
       if (mndTransAppendRedoAction(pTrans, &action) != 0) {
         taosMemoryFree(buf);
         sdbRelease(pSdb, pVgroup);

source/dnode/qnode/src/qnode.c

@@ -95,7 +95,7 @@ int32_t qndProcessQueryMsg(SQnode *pQnode, int64_t ts, SRpcMsg *pMsg) {
     case TDMT_SCH_DROP_TASK:
       code = qWorkerProcessDropMsg(pQnode, pQnode->pQuery, pMsg, ts);
       break;
-    case TDMT_VND_CONSUME:
+    case TDMT_VND_TMQ_CONSUME:
       // code =  tqProcessConsumeReq(pQnode->pTq, pMsg);
       // break;
     case TDMT_SCH_QUERY_HEARTBEAT:

source/dnode/snode/src/snode.c

@@ -261,18 +261,52 @@ int32_t sndProcessWriteMsg(SSnode *pSnode, SRpcMsg *pMsg, SRpcMsg *pRsp) {
   return 0;
 }
 
+int32_t sndProcessTaskRecoverFinishReq(SSnode *pSnode, SRpcMsg *pMsg) {
+  char   *msg = POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead));
+  int32_t msgLen = pMsg->contLen - sizeof(SMsgHead);
+
+  // deserialize
+  SStreamRecoverFinishReq req;
+
+  SDecoder decoder;
+  tDecoderInit(&decoder, msg, msgLen);
+  tDecodeSStreamRecoverFinishReq(&decoder, &req);
+  tDecoderClear(&decoder);
+
+  // find task
+  SStreamTask *pTask = streamMetaGetTask(pSnode->pMeta, req.taskId);
+  if (pTask == NULL) {
+    return -1;
+  }
+  // do process request
+  if (streamProcessRecoverFinishReq(pTask, req.childId) < 0) {
+    return -1;
+  }
+
+  return 0;
+}
+
+int32_t sndProcessTaskRecoverFinishRsp(SSnode *pSnode, SRpcMsg *pMsg) {
+  //
+  return 0;
+}
+
 int32_t sndProcessStreamMsg(SSnode *pSnode, SRpcMsg *pMsg) {
   switch (pMsg->msgType) {
     case TDMT_STREAM_TASK_RUN:
       return sndProcessTaskRunReq(pSnode, pMsg);
     case TDMT_STREAM_TASK_DISPATCH:
       return sndProcessTaskDispatchReq(pSnode, pMsg, true);
-    case TDMT_STREAM_RETRIEVE:
-      return sndProcessTaskRetrieveReq(pSnode, pMsg);
     case TDMT_STREAM_TASK_DISPATCH_RSP:
       return sndProcessTaskDispatchRsp(pSnode, pMsg);
+    case TDMT_STREAM_RETRIEVE:
+      return sndProcessTaskRetrieveReq(pSnode, pMsg);
     case TDMT_STREAM_RETRIEVE_RSP:
       return sndProcessTaskRetrieveRsp(pSnode, pMsg);
+    case TDMT_STREAM_RECOVER_FINISH:
+      return sndProcessTaskRecoverFinishReq(pSnode, pMsg);
+    case TDMT_STREAM_RECOVER_FINISH_RSP:
+      return sndProcessTaskRecoverFinishRsp(pSnode, pMsg);
     default:
       ASSERT(0);
   }

source/dnode/vnode/inc/vnode.h

@@ -158,7 +158,7 @@ int32_t tsdbReaderOpen(SVnode *pVnode, SQueryTableDataCond *pCond, void *pTableL
 void     tsdbReaderClose(STsdbReader *pReader);
 bool     tsdbNextDataBlock(STsdbReader *pReader);
 bool     tsdbTableNextDataBlock(STsdbReader *pReader, uint64_t uid);
-void     tsdbRetrieveDataBlockInfo(STsdbReader *pReader, SDataBlockInfo *pDataBlockInfo);
+void     tsdbRetrieveDataBlockInfo(const STsdbReader* pReader, int32_t* rows, uint64_t* uid, STimeWindow* pWindow);
 int32_t  tsdbRetrieveDatablockSMA(STsdbReader *pReader, SColumnDataAgg ***pBlockStatis, bool *allHave);
 SArray  *tsdbRetrieveDataBlock(STsdbReader *pTsdbReadHandle, SArray *pColumnIdList);
 int32_t  tsdbReaderReset(STsdbReader *pReader, SQueryTableDataCond *pCond);

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

@@ -176,8 +176,8 @@ int32_t tqCheckColModifiable(STQ* pTq, int64_t tbUid, int32_t colId);
 // tq-mq
 int32_t tqProcessAddCheckInfoReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
 int32_t tqProcessDelCheckInfoReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
-int32_t tqProcessVgChangeReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
+int32_t tqProcessSubscribeReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
-int32_t tqProcessVgDeleteReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
+int32_t tqProcessDeleteSubReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
 int32_t tqProcessOffsetCommitReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
 int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg);
 // tq-stream
@@ -187,11 +187,15 @@ int32_t tqProcessSubmitReq(STQ* pTq, SSubmitReq* data, int64_t ver);
 int32_t tqProcessDelReq(STQ* pTq, void* pReq, int32_t len, int64_t ver);
 int32_t tqProcessTaskRunReq(STQ* pTq, SRpcMsg* pMsg);
 int32_t tqProcessTaskDispatchReq(STQ* pTq, SRpcMsg* pMsg, bool exec);
-int32_t tqProcessTaskRecoverReq(STQ* pTq, SRpcMsg* pMsg);
 int32_t tqProcessTaskDispatchRsp(STQ* pTq, SRpcMsg* pMsg);
-int32_t tqProcessTaskRecoverRsp(STQ* pTq, SRpcMsg* pMsg);
+// int32_t tqProcessTaskRecoverReq(STQ* pTq, SRpcMsg* pMsg);
+// int32_t tqProcessTaskRecoverRsp(STQ* pTq, SRpcMsg* pMsg);
 int32_t tqProcessTaskRetrieveReq(STQ* pTq, SRpcMsg* pMsg);
 int32_t tqProcessTaskRetrieveRsp(STQ* pTq, SRpcMsg* pMsg);
+int32_t tqProcessTaskRecover1Req(STQ* pTq, SRpcMsg* pMsg);
+int32_t tqProcessTaskRecover2Req(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
+int32_t tqProcessTaskRecoverFinishReq(STQ* pTq, SRpcMsg* pMsg);
+int32_t tqProcessTaskRecoverFinishRsp(STQ* pTq, SRpcMsg* pMsg);
 
 SSubmitReq* tqBlockToSubmit(SVnode* pVnode, const SArray* pBlocks, const STSchema* pSchema,
                             SSchemaWrapper* pTagSchemaWrapper, bool createTb, int64_t suid, const char* stbFullName,

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

@@ -709,7 +709,7 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg) {
   return 0;
 }
 
-int32_t tqProcessVgDeleteReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen) {
+int32_t tqProcessDeleteSubReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen) {
   SMqVDeleteReq* pReq = (SMqVDeleteReq*)msg;
 
   taosWLockLatch(&pTq->pushLock);
@@ -767,7 +767,7 @@ int32_t tqProcessDelCheckInfoReq(STQ* pTq, int64_t version, char* msg, int32_t m
   return 0;
 }
 
-int32_t tqProcessVgChangeReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen) {
+int32_t tqProcessSubscribeReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen) {
   SMqRebVgReq req = {0};
   tDecodeSMqRebVgReq(msg, &req);
   // todo lock
@@ -982,25 +982,33 @@ int32_t tqProcessTaskDeployReq(STQ* pTq, int64_t version, char* msg, int32_t msg
 
   // 3.go through recover steps to fill history
   if (pTask->fillHistory) {
-    streamSetParamForRecover(pTask);
     if (pTask->taskLevel == TASK_LEVEL__SOURCE) {
+      streamSetParamForRecover(pTask);
       streamSourceRecoverPrepareStep1(pTask, version);
 
       SStreamRecoverStep1Req req;
       streamBuildSourceRecover1Req(pTask, &req);
-
-      void*   serialziedReq = (void*)&req;
       int32_t len = sizeof(SStreamRecoverStep1Req);
 
+      void* serializedReq = rpcMallocCont(len);
+      if (serializedReq == NULL) {
+        return -1;
+      }
+
+      memcpy(serializedReq, &req, len);
+
       SRpcMsg rpcMsg = {
           .contLen = len,
-          .pCont = serialziedReq,
+          .pCont = serializedReq,
           .msgType = TDMT_VND_STREAM_RECOVER_STEP1,
       };
 
-      tmsgPutToQueue(&pTq->pVnode->msgCb, STREAM_QUEUE, &rpcMsg);
+      if (tmsgPutToQueue(&pTq->pVnode->msgCb, STREAM_QUEUE, &rpcMsg) < 0) {
+        /*ASSERT(0);*/
+      }
 
     } else if (pTask->taskLevel == TASK_LEVEL__AGG) {
+      streamSetParamForRecover(pTask);
       streamAggRecoverPrepare(pTask);
     } else if (pTask->taskLevel == TASK_LEVEL__SINK) {
       // do nothing
@@ -1010,8 +1018,11 @@ int32_t tqProcessTaskDeployReq(STQ* pTq, int64_t version, char* msg, int32_t msg
   return 0;
 }
 
-int32_t tqProcessTaskRecover1Req(STQ* pTq, char* msg, int32_t msgLen) {
+int32_t tqProcessTaskRecover1Req(STQ* pTq, SRpcMsg* pMsg) {
   int32_t code;
+  char*   msg = pMsg->pCont;
+  int32_t msgLen = pMsg->contLen;
+
   SStreamRecoverStep1Req* pReq = (SStreamRecoverStep1Req*)msg;
   SStreamTask*            pTask = streamMetaGetTask(pTq->pStreamMeta, pReq->taskId);
   if (pTask == NULL) {
@@ -1035,16 +1046,24 @@ int32_t tqProcessTaskRecover1Req(STQ* pTq, char* msg, int32_t msgLen) {
     return -1;
   }
 
+  ASSERT(pReq->taskId == pTask->taskId);
+
   // serialize msg
-  int32_t len = sizeof(SStreamRecoverStep2Req);
+  int32_t len = sizeof(SStreamRecoverStep1Req);
-  void*   serializedReq = (void*)&req;
+
+  void* serializedReq = rpcMallocCont(len);
+  if (serializedReq == NULL) {
+    return -1;
+  }
+
+  memcpy(serializedReq, &req, len);
 
   // dispatch msg
   SRpcMsg rpcMsg = {
       .code = 0,
       .contLen = len,
       .msgType = TDMT_VND_STREAM_RECOVER_STEP2,
-      .pCont = (void*)serializedReq,
+      .pCont = serializedReq,
   };
 
   tmsgPutToQueue(&pTq->pVnode->msgCb, WRITE_QUEUE, &rpcMsg);
@@ -1087,15 +1106,15 @@ int32_t tqProcessTaskRecover2Req(STQ* pTq, int64_t version, char* msg, int32_t m
   return 0;
 }
 
-int32_t tqProcessTaskRecoverFinishReq(STQ* pTq, char* msg, int32_t msgLen) {
+int32_t tqProcessTaskRecoverFinishReq(STQ* pTq, SRpcMsg* pMsg) {
-  int32_t code;
+  char*   msg = POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead));
+  int32_t msgLen = pMsg->contLen - sizeof(SMsgHead);
 
   // deserialize
-  int32_t                 len;
   SStreamRecoverFinishReq req;
 
   SDecoder decoder;
-  tDecoderInit(&decoder, msg, sizeof(SStreamRecoverFinishReq));
+  tDecoderInit(&decoder, msg, msgLen);
   tDecodeSStreamRecoverFinishReq(&decoder, &req);
   tDecoderClear(&decoder);
 
@@ -1112,6 +1131,11 @@ int32_t tqProcessTaskRecoverFinishReq(STQ* pTq, char* msg, int32_t msgLen) {
   return 0;
 }
 
+int32_t tqProcessTaskRecoverFinishRsp(STQ* pTq, SRpcMsg* pMsg) {
+  //
+  return 0;
+}
+
 int32_t tqProcessDelReq(STQ* pTq, void* pReq, int32_t len, int64_t ver) {
   bool        failed = false;
   SDecoder*   pCoder = &(SDecoder){0};
@@ -1235,7 +1259,7 @@ int32_t tqProcessSubmitReq(STQ* pTq, SSubmitReq* pReq, int64_t ver) {
   pSubmit = streamDataSubmitNew(pReq);
   if (pSubmit == NULL) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
-    qError("failed to create data submit for stream since out of memory");
+    tqError("failed to create data submit for stream since out of memory");
     failed = true;
   }
 
@@ -1244,18 +1268,21 @@ int32_t tqProcessSubmitReq(STQ* pTq, SSubmitReq* pReq, int64_t ver) {
     if (pIter == NULL) break;
     SStreamTask* pTask = *(SStreamTask**)pIter;
     if (pTask->taskLevel != TASK_LEVEL__SOURCE) continue;
-    if (pTask->taskStatus == TASK_STATUS__RECOVER_PREPARE || pTask->taskStatus == TASK_STATUS__RECOVER1) continue;
+    if (pTask->taskStatus == TASK_STATUS__RECOVER_PREPARE || pTask->taskStatus == TASK_STATUS__RECOVER1) {
+      tqDebug("skip push task %d, task status %d", pTask->taskId, pTask->taskStatus);
+      continue;
+    }
 
-    qDebug("data submit enqueue stream task: %d, ver: %" PRId64, pTask->taskId, ver);
+    tqDebug("data submit enqueue stream task: %d, ver: %" PRId64, pTask->taskId, ver);
 
     if (!failed) {
       if (streamTaskInput(pTask, (SStreamQueueItem*)pSubmit) < 0) {
-        qError("stream task input failed, task id %d", pTask->taskId);
+        tqError("stream task input failed, task id %d", pTask->taskId);
         continue;
       }
 
       if (streamSchedExec(pTask) < 0) {
-        qError("stream task launch failed, task id %d", pTask->taskId);
+        tqError("stream task launch failed, task id %d", pTask->taskId);
         continue;
       }
     } else {
@@ -1306,33 +1333,6 @@ int32_t tqProcessTaskDispatchReq(STQ* pTq, SRpcMsg* pMsg, bool exec) {
   }
 }
 
-#if 0
-int32_t tqProcessTaskRecoverReq(STQ* pTq, SRpcMsg* pMsg) {
-  SStreamTaskRecoverReq* pReq = pMsg->pCont;
-  int32_t                taskId = pReq->taskId;
-  SStreamTask*           pTask = streamMetaGetTask(pTq->pStreamMeta, taskId);
-  if (pTask) {
-    streamProcessRecoverReq(pTask, pReq, pMsg);
-    return 0;
-  } else {
-    return -1;
-  }
-}
-
-int32_t tqProcessTaskRecoverRsp(STQ* pTq, SRpcMsg* pMsg) {
-  SStreamTaskRecoverRsp* pRsp = pMsg->pCont;
-  int32_t                taskId = pRsp->rspTaskId;
-
-  SStreamTask* pTask = streamMetaGetTask(pTq->pStreamMeta, taskId);
-  if (pTask) {
-    streamProcessRecoverRsp(pTask, pRsp);
-    return 0;
-  } else {
-    return -1;
-  }
-}
-#endif
-
 int32_t tqProcessTaskDispatchRsp(STQ* pTq, SRpcMsg* pMsg) {
   SStreamDispatchRsp* pRsp = POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead));
   int32_t             taskId = pRsp->taskId;

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

@@ -48,7 +48,7 @@ int tsdbOpen(SVnode *pVnode, STsdb **ppTsdb, const char *dir, STsdbKeepCfg *pKee
   }
 
   pTsdb->path = (char *)&pTsdb[1];
-  sprintf(pTsdb->path, "%s%s%s", pVnode->path, TD_DIRSEP, dir);
+  snprintf(pTsdb->path, TD_PATH_MAX, "%s%s%s", pVnode->path, TD_DIRSEP, dir);
   taosRealPath(pTsdb->path, NULL, slen);
   pTsdb->pVnode = pVnode;
   taosThreadRwlockInit(&pTsdb->rwLock, NULL);

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

@@ -35,11 +35,17 @@ typedef struct {
   int32_t numOfLastFiles;
 } SBlockNumber;
 
+typedef struct SBlockIndex {
+  int32_t     ordinalIndex;
+  int64_t     inFileOffset;
+  STimeWindow window;
+} SBlockIndex;
+
 typedef struct STableBlockScanInfo {
   uint64_t  uid;
   TSKEY     lastKey;
   SMapData  mapData;            // block info (compressed)
-  SArray*   pBlockList;         // block data index list
+  SArray*   pBlockList;         // block data index list, SArray<SBlockIndex>
   SIterInfo iter;               // mem buffer skip list iterator
   SIterInfo iiter;              // imem buffer skip list iterator
   SArray*   delSkyline;         // delete info for this table
@@ -72,6 +78,7 @@ typedef struct SIOCostSummary {
   double  lastBlockLoadTime;
   int64_t composedBlocks;
   double  buildComposedBlockTime;
+  double  createScanInfoList;
 } SIOCostSummary;
 
 typedef struct SBlockLoadSuppInfo {
@@ -234,6 +241,8 @@ static SHashObj* createDataBlockScanInfo(STsdbReader* pTsdbReader, const STableK
     return NULL;
   }
 
+  int64_t st = taosGetTimestampUs();
+
   for (int32_t j = 0; j < numOfTables; ++j) {
     STableBlockScanInfo info = {.lastKey = 0, .uid = idList[j].uid};
     if (ASCENDING_TRAVERSE(pTsdbReader->order)) {
@@ -249,8 +258,10 @@ static SHashObj* createDataBlockScanInfo(STsdbReader* pTsdbReader, const STableK
               pTsdbReader->idStr);
   }
 
-  tsdbDebug("%p create %d tables scan-info, size:%.2f Kb, %s", pTsdbReader, numOfTables,
+  pTsdbReader->cost.createScanInfoList = (taosGetTimestampUs() - st) / 1000.0;
-            (sizeof(STableBlockScanInfo) * numOfTables) / 1024.0, pTsdbReader->idStr);
+  tsdbDebug("%p create %d tables scan-info, size:%.2f Kb, elapsed time:%.2f ms, %s", pTsdbReader, numOfTables,
+            (sizeof(STableBlockScanInfo) * numOfTables) / 1024.0, pTsdbReader->cost.createScanInfoList,
+            pTsdbReader->idStr);
 
   return pTableMap;
 }
@@ -568,7 +579,7 @@ static int32_t doLoadBlockIndex(STsdbReader* pReader, SDataFReader* pFileReader,
 
     STableBlockScanInfo* pScanInfo = p;
     if (pScanInfo->pBlockList == NULL) {
-      pScanInfo->pBlockList = taosArrayInit(4, sizeof(int32_t));
+      pScanInfo->pBlockList = taosArrayInit(4, sizeof(SBlockIndex));
     }
 
     taosArrayPush(pIndexList, pBlockIdx);
@@ -630,7 +641,10 @@ static int32_t doLoadFileBlock(STsdbReader* pReader, SArray* pIndexList, SBlockN
         continue;
       }
 
-      void* p = taosArrayPush(pScanInfo->pBlockList, &j);
+      SBlockIndex bIndex = {.ordinalIndex = j, .inFileOffset = block.aSubBlock->offset};
+      bIndex.window = (STimeWindow) {.skey = block.minKey.ts, .ekey = block.maxKey.ts};
+
+      void* p = taosArrayPush(pScanInfo->pBlockList, &bIndex);
       if (p == NULL) {
         tMapDataClear(&pScanInfo->mapData);
         return TSDB_CODE_OUT_OF_MEMORY;
@@ -821,7 +835,8 @@ int32_t getEndPosInDataBlock(STsdbReader* pReader, SBlockData* pBlockData, SData
   } else if (!asc && pReader->window.skey <= pBlock->minKey.ts) {
     endPos = 0;
   } else {
-    endPos = doBinarySearchKey(pBlockData->aTSKEY, pBlock->nRow, pos, pReader->window.ekey, pReader->order);
+    int64_t key = asc ? pReader->window.ekey : pReader->window.skey;
+    endPos = doBinarySearchKey(pBlockData->aTSKEY, pBlock->nRow, pos, key, pReader->order);
   }
 
   return endPos;
@@ -852,8 +867,9 @@ static int32_t copyBlockDataToSDataBlock(STsdbReader* pReader, STableBlockScanIn
       // pDumpInfo->rowIndex = pBlock->nRow - 1;
     } else {
       int32_t pos = asc ? pBlock->nRow - 1 : 0;
-      int32_t order = (pReader->order == TSDB_ORDER_ASC) ? TSDB_ORDER_DESC : TSDB_ORDER_ASC;
+      int32_t order = asc ? TSDB_ORDER_DESC : TSDB_ORDER_ASC;
-      pDumpInfo->rowIndex = doBinarySearchKey(pBlockData->aTSKEY, pBlock->nRow, pos, pReader->window.skey, order);
+      int64_t key = asc ? pReader->window.skey : pReader->window.ekey;
+      pDumpInfo->rowIndex = doBinarySearchKey(pBlockData->aTSKEY, pBlock->nRow, pos, key, order);
     }
   }
 
@@ -1059,8 +1075,8 @@ static int32_t doSetCurrentBlock(SDataBlockIter* pBlockIter, const char* idStr)
       return TSDB_CODE_INVALID_PARA;
     }
 
-    int32_t* mapDataIndex = taosArrayGet(pScanInfo->pBlockList, pBlockInfo->tbBlockIdx);
+    SBlockIndex* pIndex = taosArrayGet(pScanInfo->pBlockList, pBlockInfo->tbBlockIdx);
-    tMapDataGetItemByIdx(&pScanInfo->mapData, *mapDataIndex, &pBlockIter->block, tGetDataBlk);
+    tMapDataGetItemByIdx(&pScanInfo->mapData, pIndex->ordinalIndex, &pBlockIter->block, tGetDataBlk);
   }
 
 #if 0
@@ -1073,6 +1089,7 @@ static int32_t doSetCurrentBlock(SDataBlockIter* pBlockIter, const char* idStr)
 static int32_t initBlockIterator(STsdbReader* pReader, SDataBlockIter* pBlockIter, int32_t numOfBlocks) {
   bool asc = ASCENDING_TRAVERSE(pReader->order);
 
+  SBlockOrderSupporter sup = {0};
   pBlockIter->numOfBlocks = numOfBlocks;
   taosArrayClear(pBlockIter->blockList);
   pBlockIter->pTableMap = pReader->status.pTableMap;
@@ -1081,8 +1098,6 @@ static int32_t initBlockIterator(STsdbReader* pReader, SDataBlockIter* pBlockIte
   int32_t numOfTables = (int32_t)taosHashGetSize(pReader->status.pTableMap);
 
   int64_t st = taosGetTimestampUs();
-
-  SBlockOrderSupporter sup = {0};
   int32_t code = initBlockOrderSupporter(&sup, numOfTables);
   if (code != TSDB_CODE_SUCCESS) {
     return code;
@@ -1111,17 +1126,11 @@ static int32_t initBlockIterator(STsdbReader* pReader, SDataBlockIter* pBlockIte
     }
 
     sup.pDataBlockInfo[sup.numOfTables] = (SBlockOrderWrapper*)buf;
-    SDataBlk block = {0};
+
     for (int32_t k = 0; k < num; ++k) {
-      SBlockOrderWrapper wrapper = {0};
+      SBlockIndex* pIndex = taosArrayGet(pTableScanInfo->pBlockList, k);
-
+      sup.pDataBlockInfo[sup.numOfTables][k] =
-      int32_t* mapDataIndex = taosArrayGet(pTableScanInfo->pBlockList, k);
+          (SBlockOrderWrapper){.uid = pTableScanInfo->uid, .offset = pIndex->inFileOffset};
-      tMapDataGetItemByIdx(&pTableScanInfo->mapData, *mapDataIndex, &block, tGetDataBlk);
-
-      wrapper.uid = pTableScanInfo->uid;
-      wrapper.offset = block.aSubBlock[0].offset;
-
-      sup.pDataBlockInfo[sup.numOfTables][k] = wrapper;
       cnt++;
     }
 
@@ -1212,25 +1221,22 @@ static int32_t dataBlockPartiallyRequired(STimeWindow* pWindow, SVersionRange* p
          (pVerRange->maxVer < pBlock->maxVer && pVerRange->maxVer >= pBlock->minVer);
 }
 
-static SDataBlk* getNeighborBlockOfSameTable(SFileDataBlockInfo* pBlockInfo, STableBlockScanInfo* pTableBlockScanInfo,
+static bool getNeighborBlockOfSameTable(SFileDataBlockInfo* pBlockInfo, STableBlockScanInfo* pTableBlockScanInfo,
-                                             int32_t* nextIndex, int32_t order) {
+                                        int32_t* nextIndex, int32_t order, SBlockIndex* pBlockIndex) {
   bool asc = ASCENDING_TRAVERSE(order);
   if (asc && pBlockInfo->tbBlockIdx >= taosArrayGetSize(pTableBlockScanInfo->pBlockList) - 1) {
-    return NULL;
+    return false;
   }
 
   if (!asc && pBlockInfo->tbBlockIdx == 0) {
-    return NULL;
+    return false;
   }
 
   int32_t step = asc ? 1 : -1;
   *nextIndex = pBlockInfo->tbBlockIdx + step;
-
+  *pBlockIndex = *(SBlockIndex*) taosArrayGet(pTableBlockScanInfo->pBlockList, *nextIndex);
-  SDataBlk* pBlock = taosMemoryCalloc(1, sizeof(SDataBlk));
+//  tMapDataGetItemByIdx(&pTableBlockScanInfo->mapData, pIndex->ordinalIndex, pBlock, tGetDataBlk);
-  int32_t*  indexInMapdata = taosArrayGet(pTableBlockScanInfo->pBlockList, *nextIndex);
+  return true;
-
-  tMapDataGetItemByIdx(&pTableBlockScanInfo->mapData, *indexInMapdata, pBlock, tGetDataBlk);
-  return pBlock;
 }
 
 static int32_t findFileBlockInfoIndex(SDataBlockIter* pBlockIter, SFileDataBlockInfo* pFBlockInfo) {
@@ -1272,12 +1278,12 @@ static int32_t setFileBlockActiveInBlockIter(SDataBlockIter* pBlockIter, int32_t
   return TSDB_CODE_SUCCESS;
 }
 
-static bool overlapWithNeighborBlock(SDataBlk* pBlock, SDataBlk* pNeighbor, int32_t order) {
+static bool overlapWithNeighborBlock(SDataBlk* pBlock, SBlockIndex* pNeighborBlockIndex, int32_t order) {
   // it is the last block in current file, no chance to overlap with neighbor blocks.
   if (ASCENDING_TRAVERSE(order)) {
-    return pBlock->maxKey.ts == pNeighbor->minKey.ts;
+    return pBlock->maxKey.ts == pNeighborBlockIndex->window.skey;
   } else {
-    return pBlock->minKey.ts == pNeighbor->maxKey.ts;
+    return pBlock->minKey.ts == pNeighborBlockIndex->window.ekey;
   }
 }
 
@@ -1365,12 +1371,13 @@ static void getBlockToLoadInfo(SDataBlockToLoadInfo* pInfo, SFileDataBlockInfo*
                                STableBlockScanInfo* pScanInfo, TSDBKEY keyInBuf, SLastBlockReader* pLastBlockReader,
                                STsdbReader* pReader) {
   int32_t     neighborIndex = 0;
-  SDataBlk* pNeighbor = getNeighborBlockOfSameTable(pBlockInfo, pScanInfo, &neighborIndex, pReader->order);
+  SBlockIndex bIndex = {0};
+
+  bool hasNeighbor = getNeighborBlockOfSameTable(pBlockInfo, pScanInfo, &neighborIndex, pReader->order, &bIndex);
 
   // overlap with neighbor
-  if (pNeighbor) {
+  if (hasNeighbor) {
-    pInfo->overlapWithNeighborBlock = overlapWithNeighborBlock(pBlock, pNeighbor, pReader->order);
+    pInfo->overlapWithNeighborBlock = overlapWithNeighborBlock(pBlock, &bIndex, pReader->order);
-    taosMemoryFree(pNeighbor);
   }
 
   // has duplicated ts of different version in this block
@@ -3068,14 +3075,14 @@ static int32_t checkForNeighborFileBlock(STsdbReader* pReader, STableBlockScanIn
   int32_t step = ASCENDING_TRAVERSE(pReader->order) ? 1 : -1;
 
   int32_t     nextIndex = -1;
-  SDataBlk* pNeighborBlock = getNeighborBlockOfSameTable(pFBlock, pScanInfo, &nextIndex, pReader->order);
+  SBlockIndex bIndex = {0};
-  if (pNeighborBlock == NULL) {  // do nothing
+
+  bool hasNeighbor = getNeighborBlockOfSameTable(pFBlock, pScanInfo, &nextIndex, pReader->order, &bIndex);
+  if (!hasNeighbor) {  // do nothing
     return 0;
   }
 
-  bool overlap = overlapWithNeighborBlock(pBlock, pNeighborBlock, pReader->order);
+  bool overlap = overlapWithNeighborBlock(pBlock, &bIndex, pReader->order);
-  taosMemoryFree(pNeighborBlock);
-
   if (overlap) {  // load next block
     SReaderStatus*  pStatus = &pReader->status;
     SDataBlockIter* pBlockIter = &pStatus->blockIter;
@@ -3496,7 +3503,6 @@ static int32_t doOpenReaderImpl(STsdbReader* pReader) {
 
   initFilesetIterator(&pReader->status.fileIter, pReader->pReadSnap->fs.aDFileSet, pReader);
   resetDataBlockIterator(&pReader->status.blockIter, pReader->order);
-//  resetDataBlockScanInfo(pReader->status.pTableMap, pReader->window.skey);
 
   // no data in files, let's try buffer in memory
   if (pReader->status.fileIter.numOfFiles == 0) {
@@ -3697,15 +3703,16 @@ void tsdbReaderClose(STsdbReader* pReader) {
     taosMemoryFree(pLReader);
   }
 
-  tsdbDebug(
+  tsdbDebug("%p :io-cost summary: head-file:%" PRIu64 ", head-file time:%.2f ms, SMA:%" PRId64
-      "%p :io-cost summary: head-file:%" PRIu64 ", head-file time:%.2f ms, SMA:%" PRId64
             " SMA-time:%.2f ms, fileBlocks:%" PRId64
             ", fileBlocks-load-time:%.2f ms, "
-      "build in-memory-block-time:%.2f ms, lastBlocks:%" PRId64 ", lastBlocks-time:%.2f ms, composed-blocks:%" PRId64
+            "build in-memory-block-time:%.2f ms, lastBlocks:%" PRId64
-      ", composed-blocks-time:%.2fms, STableBlockScanInfo size:%.2f Kb %s",
+            ", lastBlocks-time:%.2f ms, composed-blocks:%" PRId64
-      pReader, pCost->headFileLoad, pCost->headFileLoadTime, pCost->smaDataLoad, pCost->smaLoadTime, pCost->numOfBlocks,
+            ", composed-blocks-time:%.2fms, STableBlockScanInfo size:%.2f Kb, creatTime:%.2f ms, %s",
-      pCost->blockLoadTime, pCost->buildmemBlock, pCost->lastBlockLoad, pCost->lastBlockLoadTime, pCost->composedBlocks,
+            pReader, pCost->headFileLoad, pCost->headFileLoadTime, pCost->smaDataLoad, pCost->smaLoadTime,
-      pCost->buildComposedBlockTime, numOfTables * sizeof(STableBlockScanInfo) / 1000.0, pReader->idStr);
+            pCost->numOfBlocks, pCost->blockLoadTime, pCost->buildmemBlock, pCost->lastBlockLoad,
+            pCost->lastBlockLoadTime, pCost->composedBlocks, pCost->buildComposedBlockTime,
+            numOfTables * sizeof(STableBlockScanInfo) / 1000.0, pCost->createScanInfoList, pReader->idStr);
 
   taosMemoryFree(pReader->idStr);
   taosMemoryFree(pReader->pSchema);
@@ -3799,24 +3806,24 @@ bool tsdbTableNextDataBlock(STsdbReader* pReader, uint64_t uid) {
   return true;
 }
 
-static void setBlockInfo(STsdbReader* pReader, SDataBlockInfo* pDataBlockInfo) {
+static void setBlockInfo(const STsdbReader* pReader, int32_t* rows, uint64_t* uid, STimeWindow* pWindow) {
-  ASSERT(pDataBlockInfo != NULL && pReader != NULL);
+  ASSERT(pReader != NULL);
-  pDataBlockInfo->rows = pReader->pResBlock->info.rows;
+  *rows = pReader->pResBlock->info.rows;
-  pDataBlockInfo->uid = pReader->pResBlock->info.uid;
+  *uid = pReader->pResBlock->info.uid;
-  pDataBlockInfo->window = pReader->pResBlock->info.window;
+  *pWindow = pReader->pResBlock->info.window;
 }
 
-void tsdbRetrieveDataBlockInfo(STsdbReader* pReader, SDataBlockInfo* pDataBlockInfo) {
+void tsdbRetrieveDataBlockInfo(const STsdbReader* pReader, int32_t* rows, uint64_t* uid, STimeWindow* pWindow) {
   if (pReader->type == TIMEWINDOW_RANGE_EXTERNAL) {
     if (pReader->step == EXTERNAL_ROWS_MAIN) {
-      setBlockInfo(pReader, pDataBlockInfo);
+      setBlockInfo(pReader, rows, uid, pWindow);
     } else if (pReader->step == EXTERNAL_ROWS_PREV) {
-      setBlockInfo(pReader->innerReader[0], pDataBlockInfo);
+      setBlockInfo(pReader->innerReader[0], rows, uid, pWindow);
     } else {
-      setBlockInfo(pReader->innerReader[1], pDataBlockInfo);
+      setBlockInfo(pReader->innerReader[1], rows, uid, pWindow);
     }
   } else {
-    setBlockInfo(pReader, pDataBlockInfo);
+    setBlockInfo(pReader, rows, uid, pWindow);
   }
 }
 
@@ -3838,7 +3845,7 @@ int32_t tsdbRetrieveDatablockSMA(STsdbReader* pReader, SColumnDataAgg*** pBlockS
   SFileDataBlockInfo* pFBlock = getCurrentBlockInfo(&pReader->status.blockIter);
 
   SDataBlk* pBlock = getCurrentBlock(&pReader->status.blockIter);
-  int64_t   stime = taosGetTimestampUs();
+//  int64_t   stime = taosGetTimestampUs();
 
   SBlockLoadSuppInfo* pSup = &pReader->suppInfo;
 
@@ -3869,7 +3876,9 @@ int32_t tsdbRetrieveDatablockSMA(STsdbReader* pReader, SColumnDataAgg*** pBlockS
   size_t numOfCols = blockDataGetNumOfCols(pReader->pResBlock);
 
   int32_t i = 0, j = 0;
-  while (j < numOfCols && i < taosArrayGetSize(pSup->pColAgg)) {
+  size_t size = taosArrayGetSize(pSup->pColAgg);
+
+  while (j < numOfCols && i < size) {
     SColumnDataAgg* pAgg = taosArrayGet(pSup->pColAgg, i);
     if (pAgg->colId == pSup->colIds[j]) {
       if (IS_BSMA_ON(&(pReader->pSchema->columns[i]))) {
@@ -3886,14 +3895,10 @@ int32_t tsdbRetrieveDatablockSMA(STsdbReader* pReader, SColumnDataAgg*** pBlockS
     }
   }
 
-  double elapsed = (taosGetTimestampUs() - stime) / 1000.0;
-  pReader->cost.smaLoadTime += elapsed;
   pReader->cost.smaDataLoad += 1;
-
   *pBlockStatis = pSup->plist;
 
-  tsdbDebug("vgId:%d, succeed to load block SMA for uid %" PRIu64 ", elapsed time:%.2f ms, %s", 0, pFBlock->uid,
+  tsdbDebug("vgId:%d, succeed to load block SMA for uid %" PRIu64 ", %s", 0, pFBlock->uid, pReader->idStr);
-            elapsed, pReader->idStr);
 
   return code;
 }

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

@@ -228,30 +228,30 @@ int32_t vnodeProcessWriteMsg(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRp
       if (vnodeProcessBatchDeleteReq(pVnode, version, pReq, len, pRsp) < 0) goto _err;
       break;
     /* TQ */
-    case TDMT_VND_MQ_VG_CHANGE:
+    case TDMT_VND_TMQ_SUBSCRIBE:
-      if (tqProcessVgChangeReq(pVnode->pTq, version, POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead)),
+      if (tqProcessSubscribeReq(pVnode->pTq, version, POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead)),
                                 pMsg->contLen - sizeof(SMsgHead)) < 0) {
         goto _err;
       }
       break;
-    case TDMT_VND_MQ_VG_DELETE:
+    case TDMT_VND_TMQ_DELETE_SUB:
-      if (tqProcessVgDeleteReq(pVnode->pTq, version, pMsg->pCont, pMsg->contLen) < 0) {
+      if (tqProcessDeleteSubReq(pVnode->pTq, version, pMsg->pCont, pMsg->contLen) < 0) {
         goto _err;
       }
       break;
-    case TDMT_VND_MQ_COMMIT_OFFSET:
+    case TDMT_VND_TMQ_COMMIT_OFFSET:
       if (tqProcessOffsetCommitReq(pVnode->pTq, version, POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead)),
                                    pMsg->contLen - sizeof(SMsgHead)) < 0) {
         goto _err;
       }
       break;
-    case TDMT_VND_ADD_CHECK_INFO:
+    case TDMT_VND_TMQ_ADD_CHECKINFO:
       if (tqProcessAddCheckInfoReq(pVnode->pTq, version, POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead)),
                                    pMsg->contLen - sizeof(SMsgHead)) < 0) {
         goto _err;
       }
       break;
-    case TDMT_VND_DELETE_CHECK_INFO:
+    case TDMT_VND_TMQ_DEL_CHECKINFO:
       if (tqProcessDelCheckInfoReq(pVnode->pTq, version, POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead)),
                                    pMsg->contLen - sizeof(SMsgHead)) < 0) {
         goto _err;
@@ -268,6 +268,11 @@ int32_t vnodeProcessWriteMsg(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRp
         goto _err;
       }
     } break;
+    case TDMT_VND_STREAM_RECOVER_STEP2: {
+      if (tqProcessTaskRecover2Req(pVnode->pTq, version, pMsg->pCont, pMsg->contLen) < 0) {
+        goto _err;
+      }
+    } break;
     case TDMT_VND_ALTER_CONFIRM:
       vnodeProcessAlterConfirmReq(pVnode, version, pReq, len, pRsp);
       break;
@@ -356,14 +361,11 @@ int32_t vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg, SQueueInfo *pInfo) {
     return 0;
   }
 
-  if (pMsg->msgType == TDMT_VND_CONSUME && !pVnode->restored) {
+  if (pMsg->msgType == TDMT_VND_TMQ_CONSUME && !pVnode->restored) {
     vnodeRedirectRpcMsg(pVnode, pMsg);
     return 0;
   }
 
-  char   *msgstr = POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead));
-  int32_t msgLen = pMsg->contLen - sizeof(SMsgHead);
-
   switch (pMsg->msgType) {
     case TDMT_SCH_FETCH:
     case TDMT_SCH_MERGE_FETCH:
@@ -382,7 +384,7 @@ int32_t vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg, SQueueInfo *pInfo) {
       return vnodeGetTableCfg(pVnode, pMsg, true);
     case TDMT_VND_BATCH_META:
       return vnodeGetBatchMeta(pVnode, pMsg);
-    case TDMT_VND_CONSUME:
+    case TDMT_VND_TMQ_CONSUME:
       return tqProcessPollReq(pVnode->pTq, pMsg);
     case TDMT_STREAM_TASK_RUN:
       return tqProcessTaskRunReq(pVnode->pTq, pMsg);
@@ -390,16 +392,18 @@ int32_t vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg, SQueueInfo *pInfo) {
     case TDMT_STREAM_TASK_DISPATCH:
       return tqProcessTaskDispatchReq(pVnode->pTq, pMsg, true);
 #endif
-      /*case TDMT_STREAM_TASK_RECOVER:*/
-      /*return tqProcessTaskRecoverReq(pVnode->pTq, pMsg);*/
-    case TDMT_STREAM_RETRIEVE:
-      return tqProcessTaskRetrieveReq(pVnode->pTq, pMsg);
     case TDMT_STREAM_TASK_DISPATCH_RSP:
       return tqProcessTaskDispatchRsp(pVnode->pTq, pMsg);
-      /*case TDMT_STREAM_TASK_RECOVER_RSP:*/
+    case TDMT_STREAM_RETRIEVE:
-      /*return tqProcessTaskRecoverRsp(pVnode->pTq, pMsg);*/
+      return tqProcessTaskRetrieveReq(pVnode->pTq, pMsg);
     case TDMT_STREAM_RETRIEVE_RSP:
       return tqProcessTaskRetrieveRsp(pVnode->pTq, pMsg);
+    case TDMT_VND_STREAM_RECOVER_STEP1:
+      return tqProcessTaskRecover1Req(pVnode->pTq, pMsg);
+    case TDMT_STREAM_RECOVER_FINISH:
+      return tqProcessTaskRecoverFinishReq(pVnode->pTq, pMsg);
+    case TDMT_STREAM_RECOVER_FINISH_RSP:
+      return tqProcessTaskRecoverFinishRsp(pVnode->pTq, pMsg);
     default:
       vError("unknown msg type:%d in fetch queue", pMsg->msgType);
       return TSDB_CODE_VND_APP_ERROR;

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

@@ -244,7 +244,13 @@ int32_t vnodeProcessSyncMsg(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
 }
 
 static int32_t vnodeSyncEqCtrlMsg(const SMsgCb *msgcb, SRpcMsg *pMsg) {
-  if (msgcb == NULL) {
+  if (pMsg == NULL || pMsg->pCont == NULL) {
+    return -1;
+  }
+
+  if (msgcb == NULL || msgcb->putToQueueFp == NULL) {
+    rpcFreeCont(pMsg->pCont);
+    pMsg->pCont = NULL;
     return -1;
   }
 
@@ -257,7 +263,13 @@ static int32_t vnodeSyncEqCtrlMsg(const SMsgCb *msgcb, SRpcMsg *pMsg) {
 }
 
 static int32_t vnodeSyncEqMsg(const SMsgCb *msgcb, SRpcMsg *pMsg) {
-  if (msgcb == NULL) {
+  if (pMsg == NULL || pMsg->pCont == NULL) {
+    return -1;
+  }
+
+  if (msgcb == NULL || msgcb->putToQueueFp == NULL) {
+    rpcFreeCont(pMsg->pCont);
+    pMsg->pCont = NULL;
     return -1;
   }
 

source/libs/catalog/src/catalog.c

@@ -320,7 +320,7 @@ _return:
 }
 
 int32_t ctgChkAuth(SCatalog* pCtg, SRequestConnInfo* pConn, const char* user, const char* dbFName, AUTH_TYPE type,
-                   bool* pass) {
+                   bool* pass, bool* exists) {
   bool    inCache = false;
   int32_t code = 0;
 
@@ -329,6 +329,13 @@ int32_t ctgChkAuth(SCatalog* pCtg, SRequestConnInfo* pConn, const char* user, co
   CTG_ERR_RET(ctgChkAuthFromCache(pCtg, (char*)user, (char*)dbFName, type, &inCache, pass));
 
   if (inCache) {
+    if (exists) {
+      *exists = true;
+    }
+    
+    return TSDB_CODE_SUCCESS;
+  } else if (exists) {
+    *exists = false;
     return TSDB_CODE_SUCCESS;
   }
 
@@ -1032,7 +1039,7 @@ int32_t catalogChkTbMetaVersion(SCatalog* pCtg, SRequestConnInfo* pConn, SArray*
       switch (tbType) {
         case TSDB_CHILD_TABLE: {
           SName stb = name;
-          strcpy(stb.tname, stbName);
+          tstrncpy(stb.tname, stbName, sizeof(stb.tname));
           ctgRemoveTbMeta(pCtg, &stb);
           break;
         }
@@ -1373,13 +1380,30 @@ int32_t catalogChkAuth(SCatalog* pCtg, SRequestConnInfo* pConn, const char* user
   }
 
   int32_t code = 0;
-  CTG_ERR_JRET(ctgChkAuth(pCtg, pConn, user, dbFName, type, pass));
+  CTG_ERR_JRET(ctgChkAuth(pCtg, pConn, user, dbFName, type, pass, NULL));
 
 _return:
 
   CTG_API_LEAVE(code);
 }
 
+int32_t catalogChkAuthFromCache(SCatalog* pCtg, SRequestConnInfo* pConn, const char* user, const char* dbFName, AUTH_TYPE type,
+                                        bool* pass, bool* exists) {
+  CTG_API_ENTER();
+
+  if (NULL == pCtg || NULL == pConn || NULL == user || NULL == dbFName || NULL == pass || NULL == exists) {
+    CTG_API_LEAVE(TSDB_CODE_CTG_INVALID_INPUT);
+  }
+
+  int32_t code = 0;
+  CTG_ERR_JRET(ctgChkAuth(pCtg, pConn, user, dbFName, type, pass, exists));
+
+_return:
+
+  CTG_API_LEAVE(code);
+}
+
+
 int32_t catalogGetServerVersion(SCatalog* pCtg, SRequestConnInfo* pConn, char** pVersion) {
   CTG_API_ENTER();
 

source/libs/catalog/src/ctgUtil.c

@@ -924,6 +924,11 @@ int32_t ctgGetVgInfosFromHashValue(SCatalog* pCtg, SCtgTaskReq* tReq, SDBVgInfo*
 
   if (1 == vgNum) {
     void* pIter = taosHashIterate(dbInfo->vgHash, NULL);
+    if (NULL == pIter) {
+      ctgError("empty vgHash, db:%s, vgroup number:%d", dbFName, vgNum);
+      CTG_ERR_RET(TSDB_CODE_CTG_INTERNAL_ERROR);
+    }
+    
     for (int32_t i = 0; i < tbNum; ++i) {
       vgInfo = taosMemoryMalloc(sizeof(SVgroupInfo));
       if (NULL == vgInfo) {

source/libs/catalog/test/catalogTests.cpp

@@ -2771,10 +2771,30 @@ TEST(apiTest, catalogChkAuth_test) {
   ASSERT_EQ(code, 0);
 
   bool pass = false;
+  bool exists = false;
+  code = catalogChkAuthFromCache(pCtg, mockPointer, ctgTestUsername, ctgTestDbname, AUTH_TYPE_READ, &pass, &exists);
+  ASSERT_EQ(code, 0);
+  ASSERT_EQ(exists, false);
+  
   code = catalogChkAuth(pCtg, mockPointer, ctgTestUsername, ctgTestDbname, AUTH_TYPE_READ, &pass);
   ASSERT_EQ(code, 0);
   ASSERT_EQ(pass, true);
 
+  while (true) {
+    uint64_t n = 0;
+    ctgdGetStatNum("runtime.numOfOpDequeue", (void *)&n);
+    if (n != 1) {
+      taosMsleep(50);
+    } else {
+      break;
+    }
+  }
+
+  code = catalogChkAuthFromCache(pCtg, mockPointer, ctgTestUsername, ctgTestDbname, AUTH_TYPE_READ, &pass, &exists);
+  ASSERT_EQ(code, 0);
+  ASSERT_EQ(pass, true);
+  ASSERT_EQ(exists, true);
+
   catalogDestroy();
 }
 

source/libs/executor/inc/executil.h

@@ -96,11 +96,12 @@ typedef struct SColMatchInfo {
   int32_t matchType;  // determinate the source according to col id or slot id
 } SColMatchInfo;
 
+typedef struct SExecTaskInfo SExecTaskInfo;
 typedef struct STableListInfo STableListInfo;
 struct SqlFunctionCtx;
 
 int32_t createScanTableListInfo(SScanPhysiNode* pScanNode, SNodeList* pGroupTags, bool groupSort, SReadHandle* pHandle,
-                                STableListInfo* pTableListInfo, SNode* pTagCond, SNode* pTagIndexCond, const char* id);
+                                STableListInfo* pTableListInfo, SNode* pTagCond, SNode* pTagIndexCond, SExecTaskInfo* pTaskInfo);
 
 STableListInfo* tableListCreate();
 void*          tableListDestroy(STableListInfo* pTableListInfo);

source/libs/executor/inc/executorimpl.h

@@ -89,25 +89,10 @@ typedef struct STableScanAnalyzeInfo SFileBlockLoadRecorder;
 typedef struct STaskCostInfo {
   int64_t  created;
   int64_t  start;
-  uint64_t loadStatisTime;
-  uint64_t loadFileBlockTime;
-  uint64_t loadDataInCacheTime;
-  uint64_t loadStatisSize;
-  uint64_t loadFileBlockSize;
-  uint64_t loadDataInCacheSize;
-
-  uint64_t loadDataTime;
-
-  SFileBlockLoadRecorder* pRecoder;
   uint64_t elapsedTime;
-
+  double   extractListTime;
-  uint64_t winInfoSize;
+  double   groupIdMapTime;
-  uint64_t tableInfoSize;
+  SFileBlockLoadRecorder* pRecoder;
-  uint64_t hashSize;
-  uint64_t numOfTimeWindows;
-
-  SArray*   queryProfEvents;      // SArray<SQueryProfEvent>
-  SHashObj* operatorProfResults;  // map<operator_type, SQueryProfEvent>
 } STaskCostInfo;
 
 typedef struct SOperatorCostInfo {
@@ -150,18 +135,13 @@ typedef struct {
 
   SSchemaWrapper*     schema;
   char                tbName[TSDB_TABLE_NAME_LEN];
-  SSDataBlock*        pullOverBlk;  // for streaming
-  SWalFilterCond      cond;
-  int64_t             lastScanUid;
   int8_t              recoverStep;
   SQueryTableDataCond tableCond;
-  int64_t             recoverStartVer;
-  int64_t             recoverEndVer;
   int64_t             fillHistoryVer1;
   int64_t             fillHistoryVer2;
 
-  int8_t        triggerSaved;
+  // int8_t        triggerSaved;
-  int64_t       deleteMarkSaved;
+  // int64_t       deleteMarkSaved;
   SStreamState* pState;
 } SStreamTaskInfo;
 
@@ -461,8 +441,10 @@ typedef struct SPartitionDataInfo {
 
 typedef struct STimeWindowAggSupp {
   int8_t          calTrigger;
-  int64_t         waterMark;
+  int8_t          calTriggerSaved;
   int64_t         deleteMark;
+  int64_t         deleteMarkSaved;
+  int64_t         waterMark;
   TSKEY           maxTs;
   TSKEY           minTs;
   SColumnInfoData timeWindowData;  // query time window info for scalar function execution.
@@ -921,7 +903,7 @@ int32_t getBufferPgSize(int32_t rowSize, uint32_t* defaultPgsz, uint32_t* defaul
 void    doSetOperatorCompleted(SOperatorInfo* pOperator);
 void    doFilter(const SNode* pFilterNode, SSDataBlock* pBlock, SColMatchInfo* pColMatchInfo, SFilterInfo* pFilterInfo);
 int32_t addTagPseudoColumnData(SReadHandle* pHandle, SExprInfo* pPseudoExpr, int32_t numOfPseudoExpr,
-                               SSDataBlock* pBlock, const char* idStr);
+                               SSDataBlock* pBlock, int32_t rows, const char* idStr);
 
 void cleanupAggSup(SAggSupporter* pAggSup);
 void appendOneRowToDataBlock(SSDataBlock* pBlock, STupleHandle* pTupleHandle);

source/libs/executor/src/cachescanoperator.c

@@ -152,25 +152,27 @@ SSDataBlock* doScanCache(SOperatorInfo* pOperator) {
       pInfo->indexOfBufferedRes = 0;
     }
 
+    SSDataBlock* pRes = pInfo->pRes;
+
     if (pInfo->indexOfBufferedRes < pInfo->pBufferredRes->info.rows) {
       for (int32_t i = 0; i < taosArrayGetSize(pInfo->matchInfo.pList); ++i) {
         SColMatchItem* pMatchInfo = taosArrayGet(pInfo->matchInfo.pList, i);
         int32_t        slotId = pMatchInfo->dstSlotId;
 
         SColumnInfoData* pSrc = taosArrayGet(pInfo->pBufferredRes->pDataBlock, slotId);
-        SColumnInfoData* pDst = taosArrayGet(pInfo->pRes->pDataBlock, slotId);
+        SColumnInfoData* pDst = taosArrayGet(pRes->pDataBlock, slotId);
 
         char* p = colDataGetData(pSrc, pInfo->indexOfBufferedRes);
         bool  isNull = colDataIsNull_s(pSrc, pInfo->indexOfBufferedRes);
         colDataAppend(pDst, 0, p, isNull);
       }
 
-      pInfo->pRes->info.uid = *(tb_uid_t*)taosArrayGet(pInfo->pUidList, pInfo->indexOfBufferedRes);
+      pRes->info.uid = *(tb_uid_t*)taosArrayGet(pInfo->pUidList, pInfo->indexOfBufferedRes);
-      pInfo->pRes->info.rows = 1;
+      pRes->info.rows = 1;
 
       if (pInfo->pseudoExprSup.numOfExprs > 0) {
         SExprSupp* pSup = &pInfo->pseudoExprSup;
-        int32_t    code = addTagPseudoColumnData(&pInfo->readHandle, pSup->pExprInfo, pSup->numOfExprs, pInfo->pRes,
+        int32_t    code = addTagPseudoColumnData(&pInfo->readHandle, pSup->pExprInfo, pSup->numOfExprs, pRes, pRes->info.rows,
                                                  GET_TASKID(pTaskInfo));
         if (code != TSDB_CODE_SUCCESS) {
           pTaskInfo->code = code;
@@ -178,10 +180,9 @@ SSDataBlock* doScanCache(SOperatorInfo* pOperator) {
         }
       }
 
-      pInfo->pRes->info.groupId = getTableGroupId(pTableList, pInfo->pRes->info.uid);
+      pRes->info.groupId = getTableGroupId(pTableList, pRes->info.uid);
-
       pInfo->indexOfBufferedRes += 1;
-      return pInfo->pRes;
+      return pRes;
     } else {
       doSetOperatorCompleted(pOperator);
       return NULL;
@@ -221,7 +222,7 @@ SSDataBlock* doScanCache(SOperatorInfo* pOperator) {
             ASSERT((pInfo->retrieveType & CACHESCAN_RETRIEVE_LAST_ROW) == CACHESCAN_RETRIEVE_LAST_ROW);
 
             pInfo->pRes->info.uid = *(tb_uid_t*)taosArrayGet(pInfo->pUidList, 0);
-            code = addTagPseudoColumnData(&pInfo->readHandle, pSup->pExprInfo, pSup->numOfExprs, pInfo->pRes,
+            code = addTagPseudoColumnData(&pInfo->readHandle, pSup->pExprInfo, pSup->numOfExprs, pInfo->pRes, pInfo->pRes->info.rows,
                                           GET_TASKID(pTaskInfo));
             if (code != TSDB_CODE_SUCCESS) {
               pTaskInfo->code = code;

source/libs/executor/src/executil.c

@@ -775,75 +775,6 @@ end:
   return code;
 }
 
-static int tableUidCompare(const void* a, const void* b) {
-  int64_t u1 = *(uint64_t*)a;
-  int64_t u2 = *(uint64_t*)b;
-  if (u1 == u2) {
-    return 0;
-  }
-  return u1 < u2 ? -1 : 1;
-}
-
-static int32_t optimizeTbnameInCond(void* metaHandle, int64_t suid, SArray* list, SNode* cond, SHashObj* tags) {
-  int32_t ret = -1;
-  if (nodeType(cond) == QUERY_NODE_OPERATOR) {
-    ret = optimizeTbnameInCondImpl(metaHandle, suid, list, cond);
-    if (ret != -1) {
-      metaGetTableTagsByUids(metaHandle, suid, list, tags);
-      removeInvalidTable(list, tags);
-    }
-  }
-
-  if (nodeType(cond) != QUERY_NODE_LOGIC_CONDITION || ((SLogicConditionNode*)cond)->condType != LOGIC_COND_TYPE_AND) {
-    return ret;
-  }
-
-  bool                 hasTbnameCond = false;
-  SLogicConditionNode* pNode = (SLogicConditionNode*)cond;
-  SNodeList*           pList = (SNodeList*)pNode->pParameterList;
-
-  int32_t len = LIST_LENGTH(pList);
-  if (len <= 0) return ret;
-
-  SListCell* cell = pList->pHead;
-  for (int i = 0; i < len; i++) {
-    if (cell == NULL) break;
-    if (optimizeTbnameInCondImpl(metaHandle, suid, list, cell->pNode) == 0) {
-      hasTbnameCond = true;
-      break;
-    }
-    cell = cell->pNext;
-  }
-  taosArraySort(list, tableUidCompare);
-  taosArrayRemoveDuplicate(list, tableUidCompare, NULL);
-
-  if (hasTbnameCond) {
-    ret = metaGetTableTagsByUids(metaHandle, suid, list, tags);
-    removeInvalidTable(list, tags);
-  }
-  return ret;
-}
-
-/*
- * handle invalid uid
- */
-static int32_t removeInvalidTable(SArray* uids, SHashObj* tags) {
-  if (taosArrayGetSize(uids) <= 0) return 0;
-
-  SArray* validUid = taosArrayInit(taosArrayGetSize(uids), sizeof(int64_t));
-
-  for (int32_t i = 0; i < taosArrayGetSize(uids); i++) {
-    int64_t* uid = taosArrayGet(uids, i);
-    if (taosHashGet(tags, uid, sizeof(int64_t)) != NULL) {
-      taosArrayPush(validUid, uid);
-    }
-  }
-
-  taosArraySwap(uids, validUid);
-  taosArrayDestroy(validUid);
-  return 0;
-}
-
 static int32_t nameComparFn(const void* p1, const void* p2) {
   const char* pName1 = *(const char**)p1;
   const char* pName2 = *(const char**)p2;
@@ -897,6 +828,81 @@ static SArray* getTableNameList(const SNodeListNode* pList) {
   return pNewList;
 }
 
+static int tableUidCompare(const void* a, const void* b) {
+  uint64_t u1 = *(uint64_t*)a;
+  uint64_t u2 = *(uint64_t*)b;
+
+  if (u1 == u2) {
+    return 0;
+  }
+
+  return u1 < u2 ? -1 : 1;
+}
+
+static int32_t optimizeTbnameInCond(void* metaHandle, int64_t suid, SArray* list, SNode* cond, SHashObj* tags) {
+  int32_t ret = -1;
+  if (nodeType(cond) == QUERY_NODE_OPERATOR) {
+    ret = optimizeTbnameInCondImpl(metaHandle, suid, list, cond);
+    if (ret != -1) {
+      metaGetTableTagsByUids(metaHandle, suid, list, tags);
+      removeInvalidTable(list, tags);
+    }
+  }
+
+  if (nodeType(cond) != QUERY_NODE_LOGIC_CONDITION || ((SLogicConditionNode*)cond)->condType != LOGIC_COND_TYPE_AND) {
+    return ret;
+  }
+
+  bool                 hasTbnameCond = false;
+  SLogicConditionNode* pNode = (SLogicConditionNode*)cond;
+  SNodeList*           pList = (SNodeList*)pNode->pParameterList;
+
+  int32_t len = LIST_LENGTH(pList);
+  if (len <= 0) {
+    return ret;
+  }
+
+  SListCell* cell = pList->pHead;
+  for (int i = 0; i < len; i++) {
+    if (cell == NULL) break;
+    if (optimizeTbnameInCondImpl(metaHandle, suid, list, cell->pNode) == 0) {
+      hasTbnameCond = true;
+      break;
+    }
+    cell = cell->pNext;
+  }
+
+  taosArraySort(list, tableUidCompare);
+  taosArrayRemoveDuplicate(list, tableUidCompare, NULL);
+
+  if (hasTbnameCond) {
+    ret = metaGetTableTagsByUids(metaHandle, suid, list, tags);
+    removeInvalidTable(list, tags);
+  }
+
+  return ret;
+}
+
+/*
+ * handle invalid uid
+ */
+static int32_t removeInvalidTable(SArray* uids, SHashObj* tags) {
+  if (taosArrayGetSize(uids) <= 0) return 0;
+
+  SArray* validUid = taosArrayInit(taosArrayGetSize(uids), sizeof(int64_t));
+
+  for (int32_t i = 0; i < taosArrayGetSize(uids); i++) {
+    int64_t* uid = taosArrayGet(uids, i);
+    if (taosHashGet(tags, uid, sizeof(int64_t)) != NULL) {
+      taosArrayPush(validUid, uid);
+    }
+  }
+
+  taosArraySwap(uids, validUid);
+  taosArrayDestroy(validUid);
+  return 0;
+}
+
 static int32_t optimizeTbnameInCondImpl(void* metaHandle, int64_t suid, SArray* list, SNode* pTagCond) {
   if (nodeType(pTagCond) != QUERY_NODE_OPERATOR) {
     return -1;
@@ -1757,7 +1763,7 @@ STableListInfo* tableListCreate() {
     goto _error;
   }
 
-  pListInfo->map = taosHashInit(32, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), false, HASH_ENTRY_LOCK);
+  pListInfo->map = taosHashInit(32, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), false, HASH_ENTRY_LOCK);
   if (pListInfo->map == NULL) {
     goto _error;
   }
@@ -1882,8 +1888,9 @@ int32_t buildGroupIdMapForAllTables(STableListInfo* pTableListInfo, SReadHandle*
 
 int32_t createScanTableListInfo(SScanPhysiNode* pScanNode, SNodeList* pGroupTags, bool groupSort, SReadHandle* pHandle,
                                 STableListInfo* pTableListInfo, SNode* pTagCond, SNode* pTagIndexCond,
-                                const char* idStr) {
+                                struct SExecTaskInfo* pTaskInfo) {
   int64_t st = taosGetTimestampUs();
+  const char* idStr = GET_TASKID(pTaskInfo);
 
   if (pHandle == NULL) {
     qError("invalid handle, in creating operator tree, %s", idStr);
@@ -1899,7 +1906,8 @@ int32_t createScanTableListInfo(SScanPhysiNode* pScanNode, SNodeList* pGroupTags
   ASSERT(pTableListInfo->numOfOuputGroups == 1);
 
   int64_t st1 = taosGetTimestampUs();
-  qDebug("generate queried table list completed, elapsed time:%.2f ms %s", (st1 - st) / 1000.0, idStr);
+  pTaskInfo->cost.extractListTime = (st1 - st) / 1000.0;
+  qDebug("extract queried table list completed, elapsed time:%.2f ms %s", pTaskInfo->cost.extractListTime, idStr);
 
   if (taosArrayGetSize(pTableListInfo->pTableList) == 0) {
     qDebug("no table qualified for query, %s" PRIx64, idStr);
@@ -1911,8 +1919,8 @@ int32_t createScanTableListInfo(SScanPhysiNode* pScanNode, SNodeList* pGroupTags
     return code;
   }
 
-  int64_t st2 = taosGetTimestampUs();
+  pTaskInfo->cost.groupIdMapTime = (taosGetTimestampUs() - st1)/1000.0;
-  qDebug("generate group id map completed, elapsed time:%.2f ms %s", (st2 - st1) / 1000.0, idStr);
+  qDebug("generate group id map completed, elapsed time:%.2f ms %s", pTaskInfo->cost.groupIdMapTime, idStr);
 
   return TSDB_CODE_SUCCESS;
 }

source/libs/executor/src/executor.c

@@ -70,6 +70,26 @@ static int32_t doSetSMABlock(SOperatorInfo* pOperator, void* input, size_t numOf
   }
 }
 
+static int32_t doSetStreamOpOpen(SOperatorInfo* pOperator, char* id) {
+  {
+    ASSERT(pOperator != NULL);
+    if (pOperator->operatorType != QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN) {
+      if (pOperator->numOfDownstream == 0) {
+        qError("failed to find stream scan operator to set the input data block, %s" PRIx64, id);
+        return TSDB_CODE_QRY_APP_ERROR;
+      }
+
+      if (pOperator->numOfDownstream > 1) {  // not handle this in join query
+        qError("join not supported for stream block scan, %s" PRIx64, id);
+        return TSDB_CODE_QRY_APP_ERROR;
+      }
+      pOperator->status = OP_NOT_OPENED;
+      return doSetStreamOpOpen(pOperator->pDownstream[0], id);
+    }
+  }
+  return 0;
+}
+
 static int32_t doSetStreamBlock(SOperatorInfo* pOperator, void* input, size_t numOfBlocks, int32_t type, char* id) {
   ASSERT(pOperator != NULL);
   if (pOperator->operatorType != QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN) {
@@ -117,7 +137,22 @@ static int32_t doSetStreamBlock(SOperatorInfo* pOperator, void* input, size_t nu
   }
 }
 
-static FORCE_INLINE void streamInputBlockDataDestory(void* pBlock) { blockDataDestroy((SSDataBlock*)pBlock); }
+int32_t qSetStreamOpOpen(qTaskInfo_t tinfo) {
+  if (tinfo == NULL) {
+    return TSDB_CODE_QRY_APP_ERROR;
+  }
+
+  SExecTaskInfo* pTaskInfo = (SExecTaskInfo*)tinfo;
+
+  int32_t code = doSetStreamOpOpen(pTaskInfo->pRoot, GET_TASKID(pTaskInfo));
+  if (code != TSDB_CODE_SUCCESS) {
+    qError("%s failed to set the stream block data", GET_TASKID(pTaskInfo));
+  } else {
+    qDebug("%s set the stream block successfully", GET_TASKID(pTaskInfo));
+  }
+
+  return code;
+}
 
 int32_t qSetMultiStreamInput(qTaskInfo_t tinfo, const void* pBlocks, size_t numOfBlocks, int32_t type) {
   if (tinfo == NULL) {
@@ -706,8 +741,7 @@ int32_t qStreamInput(qTaskInfo_t tinfo, void* pItem) {
 int32_t qStreamSourceRecoverStep1(qTaskInfo_t tinfo, int64_t ver) {
   SExecTaskInfo* pTaskInfo = (SExecTaskInfo*)tinfo;
   ASSERT(pTaskInfo->execModel == OPTR_EXEC_MODEL_STREAM);
-  pTaskInfo->streamInfo.recoverStartVer = 0;
+  pTaskInfo->streamInfo.fillHistoryVer1 = ver;
-  pTaskInfo->streamInfo.recoverEndVer = ver;
   pTaskInfo->streamInfo.recoverStep = STREAM_RECOVER_STEP__PREPARE1;
   return 0;
 }
@@ -715,8 +749,7 @@ int32_t qStreamSourceRecoverStep1(qTaskInfo_t tinfo, int64_t ver) {
 int32_t qStreamSourceRecoverStep2(qTaskInfo_t tinfo, int64_t ver) {
   SExecTaskInfo* pTaskInfo = (SExecTaskInfo*)tinfo;
   ASSERT(pTaskInfo->execModel == OPTR_EXEC_MODEL_STREAM);
-  pTaskInfo->streamInfo.recoverStartVer = pTaskInfo->streamInfo.recoverEndVer;
+  pTaskInfo->streamInfo.fillHistoryVer2 = ver;
-  pTaskInfo->streamInfo.recoverEndVer = ver;
   pTaskInfo->streamInfo.recoverStep = STREAM_RECOVER_STEP__PREPARE2;
   return 0;
 }
@@ -737,22 +770,44 @@ int32_t qStreamSetParamForRecover(qTaskInfo_t tinfo) {
         pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_SEMI_INTERVAL ||
         pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_INTERVAL) {
       SStreamIntervalOperatorInfo* pInfo = pOperator->info;
-      pTaskInfo->streamInfo.triggerSaved = pInfo->twAggSup.calTrigger;
+      ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE ||
-      pTaskInfo->streamInfo.deleteMarkSaved = pInfo->twAggSup.deleteMark;
+             pInfo->twAggSup.calTrigger == STREAM_TRIGGER_WINDOW_CLOSE);
+      ASSERT(pInfo->twAggSup.calTriggerSaved == 0);
+      ASSERT(pInfo->twAggSup.deleteMarkSaved == 0);
+
+      qInfo("save stream param for interval: %d,  %" PRId64, pInfo->twAggSup.calTrigger, pInfo->twAggSup.deleteMark);
+
+      pInfo->twAggSup.calTriggerSaved = pInfo->twAggSup.calTrigger;
+      pInfo->twAggSup.deleteMarkSaved = pInfo->twAggSup.deleteMark;
       pInfo->twAggSup.calTrigger = STREAM_TRIGGER_AT_ONCE;
       pInfo->twAggSup.deleteMark = INT64_MAX;
+
     } else if (pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_SESSION ||
                pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_SEMI_SESSION ||
                pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_SESSION) {
       SStreamSessionAggOperatorInfo* pInfo = pOperator->info;
-      pTaskInfo->streamInfo.triggerSaved = pInfo->twAggSup.calTrigger;
+      ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE ||
-      pTaskInfo->streamInfo.deleteMarkSaved = pInfo->twAggSup.deleteMark;
+             pInfo->twAggSup.calTrigger == STREAM_TRIGGER_WINDOW_CLOSE);
+      ASSERT(pInfo->twAggSup.calTriggerSaved == 0);
+      ASSERT(pInfo->twAggSup.deleteMarkSaved == 0);
+
+      qInfo("save stream param for session: %d,  %" PRId64, pInfo->twAggSup.calTrigger, pInfo->twAggSup.deleteMark);
+
+      pInfo->twAggSup.calTriggerSaved = pInfo->twAggSup.calTrigger;
+      pInfo->twAggSup.deleteMarkSaved = pInfo->twAggSup.deleteMark;
       pInfo->twAggSup.calTrigger = STREAM_TRIGGER_AT_ONCE;
       pInfo->twAggSup.deleteMark = INT64_MAX;
     } else if (pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_STATE) {
       SStreamStateAggOperatorInfo* pInfo = pOperator->info;
-      pTaskInfo->streamInfo.triggerSaved = pInfo->twAggSup.calTrigger;
+      ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE ||
-      pTaskInfo->streamInfo.deleteMarkSaved = pInfo->twAggSup.deleteMark;
+             pInfo->twAggSup.calTrigger == STREAM_TRIGGER_WINDOW_CLOSE);
+      ASSERT(pInfo->twAggSup.calTriggerSaved == 0);
+      ASSERT(pInfo->twAggSup.deleteMarkSaved == 0);
+
+      qInfo("save stream param for state: %d,  %" PRId64, pInfo->twAggSup.calTrigger, pInfo->twAggSup.deleteMark);
+
+      pInfo->twAggSup.calTriggerSaved = pInfo->twAggSup.calTrigger;
+      pInfo->twAggSup.deleteMarkSaved = pInfo->twAggSup.deleteMark;
       pInfo->twAggSup.calTrigger = STREAM_TRIGGER_AT_ONCE;
       pInfo->twAggSup.deleteMark = INT64_MAX;
     }
@@ -782,21 +837,36 @@ int32_t qStreamRestoreParam(qTaskInfo_t tinfo) {
         pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_SEMI_INTERVAL ||
         pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_INTERVAL) {
       SStreamIntervalOperatorInfo* pInfo = pOperator->info;
+      ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE);
+      ASSERT(pInfo->twAggSup.deleteMark == INT64_MAX);
 
-      pInfo->twAggSup.calTrigger = pTaskInfo->streamInfo.triggerSaved;
+      pInfo->twAggSup.calTrigger = pInfo->twAggSup.calTriggerSaved;
-      pInfo->twAggSup.deleteMark = pTaskInfo->streamInfo.deleteMarkSaved;
+      pInfo->twAggSup.deleteMark = pInfo->twAggSup.deleteMarkSaved;
+      ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE ||
+             pInfo->twAggSup.calTrigger == STREAM_TRIGGER_WINDOW_CLOSE);
+      qInfo("restore stream param for interval: %d,  %" PRId64, pInfo->twAggSup.calTrigger, pInfo->twAggSup.deleteMark);
     } else if (pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_SESSION ||
                pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_SEMI_SESSION ||
                pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_SESSION) {
       SStreamSessionAggOperatorInfo* pInfo = pOperator->info;
+      ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE);
+      ASSERT(pInfo->twAggSup.deleteMark == INT64_MAX);
 
-      pInfo->twAggSup.calTrigger = pTaskInfo->streamInfo.triggerSaved;
+      pInfo->twAggSup.calTrigger = pInfo->twAggSup.calTriggerSaved;
-      pInfo->twAggSup.deleteMark = pTaskInfo->streamInfo.deleteMarkSaved;
+      pInfo->twAggSup.deleteMark = pInfo->twAggSup.deleteMarkSaved;
+      ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE ||
+             pInfo->twAggSup.calTrigger == STREAM_TRIGGER_WINDOW_CLOSE);
+      qInfo("restore stream param for session: %d,  %" PRId64, pInfo->twAggSup.calTrigger, pInfo->twAggSup.deleteMark);
     } else if (pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_STATE) {
       SStreamStateAggOperatorInfo* pInfo = pOperator->info;
+      ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE);
+      ASSERT(pInfo->twAggSup.deleteMark == INT64_MAX);
 
-      pInfo->twAggSup.calTrigger = pTaskInfo->streamInfo.triggerSaved;
+      pInfo->twAggSup.calTrigger = pInfo->twAggSup.calTriggerSaved;
-      pInfo->twAggSup.deleteMark = pTaskInfo->streamInfo.deleteMarkSaved;
+      pInfo->twAggSup.deleteMark = pInfo->twAggSup.deleteMarkSaved;
+      ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE ||
+             pInfo->twAggSup.calTrigger == STREAM_TRIGGER_WINDOW_CLOSE);
+      qInfo("restore stream param for state: %d,  %" PRId64, pInfo->twAggSup.calTrigger, pInfo->twAggSup.deleteMark);
     }
 
     // iterate operator tree

source/libs/executor/src/executorimpl.c

@@ -574,6 +574,9 @@ int32_t projectApplyFunctions(SExprInfo* pExpr, SSDataBlock* pResult, SSDataBloc
   // if the source equals to the destination, it is to create a new column as the result of scalar
   // function or some operators.
   bool createNewColModel = (pResult == pSrcBlock);
+  if (createNewColModel) {
+    blockDataEnsureCapacity(pResult, pResult->info.rows);
+  }
 
   int32_t numOfRows = 0;
 
@@ -623,6 +626,7 @@ int32_t projectApplyFunctions(SExprInfo* pExpr, SSDataBlock* pResult, SSDataBloc
 
       int32_t startOffset = createNewColModel ? 0 : pResult->info.rows;
       ASSERT(pResult->info.capacity > 0);
+
       colDataMergeCol(pResColData, startOffset, (int32_t*)&pResult->info.capacity, &idata, dest.numOfRows);
       colDataDestroy(&idata);
 
@@ -844,57 +848,6 @@ STimeWindow getAlignQueryTimeWindow(SInterval* pInterval, int32_t precision, int
   return win;
 }
 
-#if 0
-static bool overlapWithTimeWindow(STaskAttr* pQueryAttr, SDataBlockInfo* pBlockInfo) {
-  STimeWindow w = {0};
-
-  TSKEY sk = TMIN(pQueryAttr->window.skey, pQueryAttr->window.ekey);
-  TSKEY ek = TMAX(pQueryAttr->window.skey, pQueryAttr->window.ekey);
-
-  if (true) {
-    //    getAlignQueryTimeWindow(pQueryAttr, pBlockInfo->window.skey, sk, ek, &w);
-    assert(w.ekey >= pBlockInfo->window.skey);
-
-    if (w.ekey < pBlockInfo->window.ekey) {
-      return true;
-    }
-
-    while (1) {
-      //      getNextTimeWindow(pQueryAttr, &w);
-      if (w.skey > pBlockInfo->window.ekey) {
-        break;
-      }
-
-      assert(w.ekey > pBlockInfo->window.ekey);
-      if (w.skey <= pBlockInfo->window.ekey && w.skey > pBlockInfo->window.skey) {
-        return true;
-      }
-    }
-  } else {
-    //    getAlignQueryTimeWindow(pQueryAttr, pBlockInfo->window.ekey, sk, ek, &w);
-    assert(w.skey <= pBlockInfo->window.ekey);
-
-    if (w.skey > pBlockInfo->window.skey) {
-      return true;
-    }
-
-    while (1) {
-      //      getNextTimeWindow(pQueryAttr, &w);
-      if (w.ekey < pBlockInfo->window.skey) {
-        break;
-      }
-
-      assert(w.skey < pBlockInfo->window.skey);
-      if (w.ekey < pBlockInfo->window.ekey && w.ekey >= pBlockInfo->window.skey) {
-        return true;
-      }
-    }
-  }
-
-  return false;
-}
-#endif
-
 int32_t loadDataBlockOnDemand(SExecTaskInfo* pTaskInfo, STableScanInfo* pTableScanInfo, SSDataBlock* pBlock,
                               uint32_t* status) {
   *status = BLK_DATA_NOT_LOAD;
@@ -1447,40 +1400,18 @@ void doBuildResultDatablock(SOperatorInfo* pOperator, SOptrBasicInfo* pbInfo, SG
   }
 }
 
-static int32_t compressQueryColData(SColumnInfoData* pColRes, int32_t numOfRows, char* data, int8_t compressed) {
-  int32_t colSize = pColRes->info.bytes * numOfRows;
-  return (*(tDataTypes[pColRes->info.type].compFunc))(pColRes->pData, colSize, numOfRows, data,
-                                                      colSize + COMP_OVERFLOW_BYTES, compressed, NULL, 0);
-}
-
 void queryCostStatis(SExecTaskInfo* pTaskInfo) {
   STaskCostInfo* pSummary = &pTaskInfo->cost;
 
-  //  uint64_t hashSize = taosHashGetMemSize(pQInfo->runtimeEnv.pResultRowHashTable);
-  //  hashSize += taosHashGetMemSize(pRuntimeEnv->tableqinfoGroupInfo.map);
-  //  pSummary->hashSize = hashSize;
-
-  //  SResultRowPool* p = pTaskInfo->pool;
-  //  if (p != NULL) {
-  //    pSummary->winInfoSize = getResultRowPoolMemSize(p);
-  //    pSummary->numOfTimeWindows = getNumOfAllocatedResultRows(p);
-  //  } else {
-  //    pSummary->winInfoSize = 0;
-  //    pSummary->numOfTimeWindows = 0;
-  //  }
-
   SFileBlockLoadRecorder* pRecorder = pSummary->pRecoder;
   if (pSummary->pRecoder != NULL) {
     qDebug(
-        "%s :cost summary: elapsed time:%.2f ms, total blocks:%d, load block SMA:%d, load data block:%d, total "
+        "%s :cost summary: elapsed time:%.2f ms, extract tableList:%.2f ms, createGroupIdMap:%.2f ms, total blocks:%d, "
-        "rows:%" PRId64 ", check rows:%" PRId64,
+        "load block SMA:%d, load data block:%d, total rows:%" PRId64 ", check rows:%" PRId64,
-        GET_TASKID(pTaskInfo), pSummary->elapsedTime / 1000.0, pRecorder->totalBlocks, pRecorder->loadBlockStatis,
+        GET_TASKID(pTaskInfo), pSummary->elapsedTime / 1000.0, pSummary->extractListTime, pSummary->groupIdMapTime,
-        pRecorder->loadBlocks, pRecorder->totalRows, pRecorder->totalCheckedRows);
+        pRecorder->totalBlocks, pRecorder->loadBlockStatis, pRecorder->loadBlocks, pRecorder->totalRows,
+        pRecorder->totalCheckedRows);
   }
-
-  // qDebug("QInfo:0x%"PRIx64" :cost summary: winResPool size:%.2f Kb, numOfWin:%"PRId64", tableInfoSize:%.2f Kb,
-  // hashTable:%.2f Kb", pQInfo->qId, pSummary->winInfoSize/1024.0,
-  //      pSummary->numOfTimeWindows, pSummary->tableInfoSize/1024.0, pSummary->hashSize/1024.0);
 }
 
 // static void updateOffsetVal(STaskRuntimeEnv *pRuntimeEnv, SDataBlockInfo *pBlockInfo) {
@@ -3267,6 +3198,11 @@ SOperatorInfo* createFillOperatorInfo(SOperatorInfo* downstream, SFillPhysiNode*
 
 static SExecTaskInfo* createExecTaskInfo(uint64_t queryId, uint64_t taskId, EOPTR_EXEC_MODEL model, char* dbFName) {
   SExecTaskInfo* pTaskInfo = taosMemoryCalloc(1, sizeof(SExecTaskInfo));
+  if (pTaskInfo == NULL) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    return NULL;
+  }
+
   setTaskStatus(pTaskInfo, TASK_NOT_COMPLETED);
 
   pTaskInfo->schemaInfo.dbname = strdup(dbFName);
@@ -3397,7 +3333,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
 
       int32_t code =
           createScanTableListInfo(&pTableScanNode->scan, pTableScanNode->pGroupTags, pTableScanNode->groupSort, pHandle,
-                                  pTableListInfo, pTagCond, pTagIndexCond, idstr);
+                                  pTableListInfo, pTagCond, pTagIndexCond, pTaskInfo);
       if (code) {
         pTaskInfo->code = code;
         qError("failed to createScanTableListInfo, code:%s, %s", tstrerror(code), idstr);
@@ -3415,9 +3351,9 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
       pTaskInfo->cost.pRecoder = &pScanInfo->readRecorder;
     } else if (QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN == type) {
       STableMergeScanPhysiNode* pTableScanNode = (STableMergeScanPhysiNode*)pPhyNode;
-      int32_t                   code =
+
-          createScanTableListInfo(&pTableScanNode->scan, pTableScanNode->pGroupTags, /*pTableScanNode->groupSort*/true, pHandle,
+      int32_t code = createScanTableListInfo(&pTableScanNode->scan, pTableScanNode->pGroupTags, true, pHandle,
-                                  pTableListInfo, pTagCond, pTagIndexCond, idstr);
+                                             pTableListInfo, pTagCond, pTagIndexCond, pTaskInfo);
       if (code) {
         pTaskInfo->code = code;
         qError("failed to createScanTableListInfo, code: %s", tstrerror(code));
@@ -3442,7 +3378,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
       if (pHandle->vnode) {
         int32_t code =
             createScanTableListInfo(&pTableScanNode->scan, pTableScanNode->pGroupTags, pTableScanNode->groupSort,
-                                    pHandle, pTableListInfo, pTagCond, pTagIndexCond, idstr);
+                                    pHandle, pTableListInfo, pTagCond, pTagIndexCond, pTaskInfo);
         if (code) {
           pTaskInfo->code = code;
           qError("failed to createScanTableListInfo, code: %s", tstrerror(code));
@@ -3469,7 +3405,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
       STagScanPhysiNode* pScanPhyNode = (STagScanPhysiNode*)pPhyNode;
 
       int32_t code = createScanTableListInfo(pScanPhyNode, NULL, false, pHandle, pTableListInfo, pTagCond,
-                                             pTagIndexCond, idstr);
+                                             pTagIndexCond, pTaskInfo);
       if (code != TSDB_CODE_SUCCESS) {
         pTaskInfo->code = code;
         qError("failed to getTableList, code: %s", tstrerror(code));
@@ -3502,7 +3438,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
       SLastRowScanPhysiNode* pScanNode = (SLastRowScanPhysiNode*)pPhyNode;
 
       int32_t code = createScanTableListInfo(&pScanNode->scan, pScanNode->pGroupTags, true, pHandle, pTableListInfo,
-                                             pTagCond, pTagIndexCond, idstr);
+                                             pTagCond, pTagIndexCond, pTaskInfo);
       if (code != TSDB_CODE_SUCCESS) {
         pTaskInfo->code = code;
         return NULL;
@@ -3834,10 +3770,8 @@ int32_t createExecTaskInfoImpl(SSubplan* pPlan, SExecTaskInfo** pTaskInfo, SRead
                                char* sql, EOPTR_EXEC_MODEL model) {
   uint64_t queryId = pPlan->id.queryId;
 
-  int32_t code = TSDB_CODE_SUCCESS;
   *pTaskInfo = createExecTaskInfo(queryId, taskId, model, pPlan->dbFName);
   if (*pTaskInfo == NULL) {
-    code = TSDB_CODE_QRY_OUT_OF_MEMORY;
     goto _complete;
   }
 
@@ -3856,17 +3790,16 @@ int32_t createExecTaskInfoImpl(SSubplan* pPlan, SExecTaskInfo** pTaskInfo, SRead
       createOperatorTree(pPlan->pNode, *pTaskInfo, pHandle, pPlan->pTagCond, pPlan->pTagIndexCond, pPlan->user);
 
   if (NULL == (*pTaskInfo)->pRoot) {
-    code = (*pTaskInfo)->code;
+    terrno = (*pTaskInfo)->code;
     goto _complete;
   }
 
-  return code;
+  return TSDB_CODE_SUCCESS;
 
 _complete:
   taosMemoryFree(sql);
   doDestroyTask(*pTaskInfo);
-  terrno = code;
+  return terrno;
-  return code;
 }
 
 void doDestroyTask(SExecTaskInfo* pTaskInfo) {

source/libs/executor/src/groupoperator.c

@@ -442,8 +442,8 @@ SOperatorInfo* createGroupOperatorInfo(SOperatorInfo* downstream, SAggPhysiNode
   pOperator->info = pInfo;
   pOperator->pTaskInfo = pTaskInfo;
 
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, hashGroupbyAggregate, NULL, NULL,
+  pOperator->fpSet =
-                                         destroyGroupOperatorInfo, NULL);
+      createOperatorFpSet(operatorDummyOpenFn, hashGroupbyAggregate, NULL, NULL, destroyGroupOperatorInfo, NULL);
   code = appendDownstream(pOperator, &downstream, 1);
   if (code != TSDB_CODE_SUCCESS) {
     goto _error;
@@ -765,7 +765,6 @@ SOperatorInfo* createPartitionOperatorInfo(SOperatorInfo* downstream, SPartition
     goto _error;
   }
 
-
   int32_t    numOfCols = 0;
   SExprInfo* pExprInfo = createExprInfo(pPartNode->pTargets, NULL, &numOfCols);
   pInfo->pGroupCols = extractPartitionColInfo(pPartNode->pPartitionKeys);
@@ -819,8 +818,8 @@ SOperatorInfo* createPartitionOperatorInfo(SOperatorInfo* downstream, SPartition
   pOperator->info = pInfo;
   pOperator->pTaskInfo = pTaskInfo;
 
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, hashPartition, NULL, NULL, destroyPartitionOperatorInfo,
+  pOperator->fpSet =
-                                         NULL);
+      createOperatorFpSet(operatorDummyOpenFn, hashPartition, NULL, NULL, destroyPartitionOperatorInfo, NULL);
 
   code = appendDownstream(pOperator, &downstream, 1);
   return pOperator;
@@ -965,6 +964,7 @@ static SSDataBlock* doStreamHashPartition(SOperatorInfo* pOperator) {
       case STREAM_DELETE_DATA: {
         copyDataBlock(pInfo->pDelRes, pBlock);
         pInfo->pDelRes->info.type = STREAM_DELETE_RESULT;
+        printDataBlock(pInfo->pDelRes, "stream partitionby delete");
         return pInfo->pDelRes;
       } break;
       default:
@@ -1014,6 +1014,9 @@ void initParDownStream(SOperatorInfo* downstream, SPartitionBySupporter* pParSup
   SStreamScanInfo* pScanInfo = downstream->info;
   pScanInfo->partitionSup = *pParSup;
   pScanInfo->pPartScalarSup = pExpr;
+  if (!pScanInfo->pUpdateInfo) {
+    pScanInfo->pUpdateInfo = updateInfoInit(60000, TSDB_TIME_PRECISION_MILLI, 0);
+  }
 }
 
 SOperatorInfo* createStreamPartitionOperatorInfo(SOperatorInfo* downstream, SStreamPartitionPhysiNode* pPartNode,
@@ -1108,7 +1111,6 @@ _error:
   return NULL;
 }
 
-
 SArray* extractColumnInfo(SNodeList* pNodeList) {
   size_t  numOfCols = LIST_LENGTH(pNodeList);
   SArray* pList = taosArrayInit(numOfCols, sizeof(SColumn));

source/libs/executor/src/scanoperator.c

@@ -331,7 +331,8 @@ static bool doLoadBlockSMA(STableScanInfo* pTableScanInfo, SSDataBlock* pBlock,
     }
   }
 
-  for (int32_t i = 0; i < taosArrayGetSize(pTableScanInfo->matchInfo.pList); ++i) {
+  size_t num = taosArrayGetSize(pTableScanInfo->matchInfo.pList);
+  for (int32_t i = 0; i < num; ++i) {
     SColMatchItem* pColMatchInfo = taosArrayGet(pTableScanInfo->matchInfo.pList, i);
     if (!pColMatchInfo->needOutput) {
       continue;
@@ -343,11 +344,11 @@ static bool doLoadBlockSMA(STableScanInfo* pTableScanInfo, SSDataBlock* pBlock,
   return true;
 }
 
-static void doSetTagColumnData(STableScanInfo* pTableScanInfo, SSDataBlock* pBlock, SExecTaskInfo* pTaskInfo) {
+static void doSetTagColumnData(STableScanInfo* pTableScanInfo, SSDataBlock* pBlock, SExecTaskInfo* pTaskInfo, int32_t rows) {
   if (pTableScanInfo->pseudoSup.numOfExprs > 0) {
     SExprSupp* pSup = &pTableScanInfo->pseudoSup;
 
-    int32_t code = addTagPseudoColumnData(&pTableScanInfo->readHandle, pSup->pExprInfo, pSup->numOfExprs, pBlock,
+    int32_t code = addTagPseudoColumnData(&pTableScanInfo->readHandle, pSup->pExprInfo, pSup->numOfExprs, pBlock, rows,
                                           GET_TASKID(pTaskInfo));
     if (code != TSDB_CODE_SUCCESS) {
       T_LONG_JMP(pTaskInfo->env, code);
@@ -363,7 +364,8 @@ void applyLimitOffset(SLimitInfo* pLimitInfo, SSDataBlock* pBlock, SExecTaskInfo
     if (pLimitInfo->remainOffset >= pBlock->info.rows) {
       pLimitInfo->remainOffset -= pBlock->info.rows;
       pBlock->info.rows = 0;
-      qDebug("current block ignore due to offset, current:%"PRId64", %s", pLimitInfo->remainOffset, GET_TASKID(pTaskInfo));
+      qDebug("current block ignore due to offset, current:%" PRId64 ", %s", pLimitInfo->remainOffset,
+             GET_TASKID(pTaskInfo));
     } else {
       blockDataTrimFirstNRows(pBlock, pLimitInfo->remainOffset);
       pLimitInfo->remainOffset = 0;
@@ -381,6 +383,17 @@ void applyLimitOffset(SLimitInfo* pLimitInfo, SSDataBlock* pBlock, SExecTaskInfo
   }
 }
 
+static void ensureBlockCapacity(SSDataBlock* pBlock, int32_t capacity) {
+  // keep the value of rows temporarily
+  int32_t rows = pBlock->info.rows;
+
+  pBlock->info.rows = 0;
+  blockDataEnsureCapacity(pBlock, capacity);
+
+  // restore the rows number
+  pBlock->info.rows = rows;
+}
+
 static int32_t loadDataBlock(SOperatorInfo* pOperator, STableScanInfo* pTableScanInfo, SSDataBlock* pBlock,
                              uint32_t* status) {
   SExecTaskInfo*  pTaskInfo = pOperator->pTaskInfo;
@@ -412,7 +425,10 @@ static int32_t loadDataBlock(SOperatorInfo* pOperator, STableScanInfo* pTableSca
     qDebug("%s data block skipped, brange:%" PRId64 "-%" PRId64 ", rows:%d", GET_TASKID(pTaskInfo),
            pBlockInfo->window.skey, pBlockInfo->window.ekey, pBlockInfo->rows);
 
-    doSetTagColumnData(pTableScanInfo, pBlock, pTaskInfo);
+    if (pTableScanInfo->pseudoSup.numOfExprs > 0) {
+      ensureBlockCapacity(pBlock, pBlock->info.rows);
+    }
+    doSetTagColumnData(pTableScanInfo, pBlock, pTaskInfo, 1);
     pCost->skipBlocks += 1;
     return TSDB_CODE_SUCCESS;
   } else if (*status == FUNC_DATA_REQUIRED_STATIS_LOAD) {
@@ -422,7 +438,11 @@ static int32_t loadDataBlock(SOperatorInfo* pOperator, STableScanInfo* pTableSca
     if (success) {  // failed to load the block sma data, data block statistics does not exist, load data block instead
       qDebug("%s data block SMA loaded, brange:%" PRId64 "-%" PRId64 ", rows:%d", GET_TASKID(pTaskInfo),
              pBlockInfo->window.skey, pBlockInfo->window.ekey, pBlockInfo->rows);
-      doSetTagColumnData(pTableScanInfo, pBlock, pTaskInfo);
+      if (pTableScanInfo->pseudoSup.numOfExprs > 0) {
+        ensureBlockCapacity(pBlock, pBlock->info.rows);
+      }
+
+      doSetTagColumnData(pTableScanInfo, pBlock, pTaskInfo, 1);
       return TSDB_CODE_SUCCESS;
     } else {
       qDebug("%s failed to load SMA, since not all columns have SMA", GET_TASKID(pTaskInfo));
@@ -471,8 +491,9 @@ static int32_t loadDataBlock(SOperatorInfo* pOperator, STableScanInfo* pTableSca
     return terrno;
   }
 
+  ensureBlockCapacity(pBlock, pBlock->info.rows);
   relocateColumnData(pBlock, pTableScanInfo->matchInfo.pList, pCols, true);
-  doSetTagColumnData(pTableScanInfo, pBlock, pTaskInfo);
+  doSetTagColumnData(pTableScanInfo, pBlock, pTaskInfo, pBlock->info.rows);
 
   // restore the previous value
   pCost->totalRows -= pBlock->info.rows;
@@ -512,26 +533,29 @@ static void prepareForDescendingScan(STableScanInfo* pTableScanInfo, SqlFunction
 }
 
 int32_t addTagPseudoColumnData(SReadHandle* pHandle, SExprInfo* pPseudoExpr, int32_t numOfPseudoExpr,
-                               SSDataBlock* pBlock, const char* idStr) {
+                               SSDataBlock* pBlock, int32_t rows, const char* idStr) {
   // currently only the tbname pseudo column
   if (numOfPseudoExpr == 0) {
     return TSDB_CODE_SUCCESS;
   }
 
+  // backup the rows
+  int32_t backupRows = pBlock->info.rows;
+  pBlock->info.rows = rows;
+
   SMetaReader mr = {0};
   metaReaderInit(&mr, pHandle->meta, 0);
   int32_t code = metaGetTableEntryByUid(&mr, pBlock->info.uid);
+  metaReaderReleaseLock(&mr);
+
   if (code != TSDB_CODE_SUCCESS) {
     qError("failed to get table meta, uid:0x%" PRIx64 ", code:%s, %s", pBlock->info.uid, tstrerror(terrno), idStr);
     metaReaderClear(&mr);
     return terrno;
   }
 
-  metaReaderReleaseLock(&mr);
-
   for (int32_t j = 0; j < numOfPseudoExpr; ++j) {
     SExprInfo* pExpr = &pPseudoExpr[j];
-
     int32_t    dstSlotId = pExpr->base.resSchema.slotId;
 
     SColumnInfoData* pColInfoData = taosArrayGet(pBlock->pDataBlock, dstSlotId);
@@ -571,6 +595,9 @@ int32_t addTagPseudoColumnData(SReadHandle* pHandle, SExprInfo* pPseudoExpr, int
   }
 
   metaReaderClear(&mr);
+
+  // restore the rows
+  pBlock->info.rows = backupRows;
   return TSDB_CODE_SUCCESS;
 }
 
@@ -615,14 +642,10 @@ static SSDataBlock* doTableScanImpl(SOperatorInfo* pOperator) {
 
     blockDataCleanup(pBlock);
 
-    SDataBlockInfo binfo = pBlock->info;
+    SDataBlockInfo* pBInfo = &pBlock->info;
-    tsdbRetrieveDataBlockInfo(pTableScanInfo->dataReader, &binfo);
+    tsdbRetrieveDataBlockInfo(pTableScanInfo->dataReader, &pBInfo->rows, &pBInfo->uid, &pBInfo->window);
 
-    binfo.capacity = binfo.rows;
+    ASSERT(pBInfo->uid != 0);
-    blockDataEnsureCapacity(pBlock, binfo.rows);
-    pBlock->info = binfo;
-
-    ASSERT(binfo.uid != 0);
     pBlock->info.groupId = getTableGroupId(pTaskInfo->pTableInfoList, pBlock->info.uid);
 
     uint32_t status = 0;
@@ -753,8 +776,8 @@ static SSDataBlock* doTableScan(SOperatorInfo* pOperator) {
       tableListGetGroupList(pTaskInfo->pTableInfoList, pInfo->currentGroupId, &pList, &num);
       ASSERT(pInfo->dataReader == NULL);
 
-      int32_t code = tsdbReaderOpen(pInfo->readHandle.vnode, &pInfo->cond, pList, num, (STsdbReader**)&pInfo->dataReader,
+      int32_t code = tsdbReaderOpen(pInfo->readHandle.vnode, &pInfo->cond, pList, num,
-                                    GET_TASKID(pTaskInfo));
+                                    (STsdbReader**)&pInfo->dataReader, GET_TASKID(pTaskInfo));
       if (code != TSDB_CODE_SUCCESS) {
         T_LONG_JMP(pTaskInfo->env, code);
       }
@@ -1024,7 +1047,8 @@ static int32_t initTableblockDistQueryCond(uint64_t uid, SQueryTableDataCond* pC
   return TSDB_CODE_SUCCESS;
 }
 
-SOperatorInfo* createDataBlockInfoScanOperator(SReadHandle* readHandle, SBlockDistScanPhysiNode* pBlockScanNode, SExecTaskInfo* pTaskInfo) {
+SOperatorInfo* createDataBlockInfoScanOperator(SReadHandle* readHandle, SBlockDistScanPhysiNode* pBlockScanNode,
+                                               SExecTaskInfo* pTaskInfo) {
   SBlockDistInfo* pInfo = taosMemoryCalloc(1, sizeof(SBlockDistInfo));
   SOperatorInfo*  pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
   if (pInfo == NULL || pOperator == NULL) {
@@ -1144,25 +1168,25 @@ static SSDataBlock* readPreVersionData(SOperatorInfo* pTableScanOp, uint64_t tbU
 
   bool hasBlock = tsdbNextDataBlock(pReader);
   if (hasBlock) {
-    SDataBlockInfo binfo = {0};
+    SDataBlockInfo* pBInfo = &pBlock->info;
-    tsdbRetrieveDataBlockInfo(pReader, &binfo);
+
+    int32_t rows = 0;
+    tsdbRetrieveDataBlockInfo(pReader, &rows, &pBInfo->uid, &pBInfo->window);
 
     SArray* pCols = tsdbRetrieveDataBlock(pReader, NULL);
-    blockDataEnsureCapacity(pBlock, binfo.rows);
+    blockDataEnsureCapacity(pBlock, rows);
-
+    pBlock->info.rows = rows;
-    pBlock->info.window = binfo.window;
-    pBlock->info.uid = binfo.uid;
-    pBlock->info.rows = binfo.rows;
 
     relocateColumnData(pBlock, pTableScanInfo->matchInfo.pList, pCols, true);
-    doSetTagColumnData(pTableScanInfo, pBlock, pTaskInfo);
+    doSetTagColumnData(pTableScanInfo, pBlock, pTaskInfo, rows);
 
-    pBlock->info.groupId = getTableGroupId(pTaskInfo->pTableInfoList, binfo.uid);
+    pBlock->info.groupId = getTableGroupId(pTaskInfo->pTableInfoList, pBInfo->uid);
   }
 
   tsdbReaderClose(pReader);
   qDebug("retrieve prev rows:%d, skey:%" PRId64 ", ekey:%" PRId64 " uid:%" PRIu64 ", max ver:%" PRId64
-             ", suid:%" PRIu64, pBlock->info.rows, startTs, endTs, tbUid, maxVersion, cond.suid);
+         ", suid:%" PRIu64,
+         pBlock->info.rows, startTs, endTs, tbUid, maxVersion, cond.suid);
 
   return pBlock->info.rows > 0 ? pBlock : NULL;
 }
@@ -1178,12 +1202,6 @@ static uint64_t getGroupIdByCol(SStreamScanInfo* pInfo, uint64_t uid, TSKEY ts,
 
 static uint64_t getGroupIdByUid(SStreamScanInfo* pInfo, uint64_t uid) {
   return getTableGroupId(pInfo->pTableScanOp->pTaskInfo->pTableInfoList, uid);
-//  SHashObj* map = pInfo->pTableScanOp->pTaskInfo->pTableInfoList.map;
-//  uint64_t* groupId = taosHashGet(map, &uid, sizeof(int64_t));
-//  if (groupId) {
-//    return *groupId;
-//  }
-//  return 0;
 }
 
 static uint64_t getGroupIdByData(SStreamScanInfo* pInfo, uint64_t uid, TSKEY ts, int64_t maxVersion) {
@@ -1380,7 +1398,7 @@ static int32_t generateIntervalScanRange(SStreamScanInfo* pInfo, SSDataBlock* pS
   if (rows == 0) {
     return TSDB_CODE_SUCCESS;
   }
-  int32_t code = blockDataEnsureCapacity(pDestBlock, rows * 2);
+  int32_t code = blockDataEnsureCapacity(pDestBlock, rows);
   if (code != TSDB_CODE_SUCCESS) {
     return code;
   }
@@ -1423,39 +1441,33 @@ static int32_t generateIntervalScanRange(SStreamScanInfo* pInfo, SSDataBlock* pS
 }
 
 static int32_t generateDeleteResultBlock(SStreamScanInfo* pInfo, SSDataBlock* pSrcBlock, SSDataBlock* pDestBlock) {
-  if (pSrcBlock->info.rows == 0) {
+  blockDataCleanup(pDestBlock);
+  int32_t rows = pSrcBlock->info.rows;
+  if (rows == 0) {
     return TSDB_CODE_SUCCESS;
   }
-  blockDataCleanup(pDestBlock);
+  int32_t code = blockDataEnsureCapacity(pDestBlock, rows);
-  int32_t code = blockDataEnsureCapacity(pDestBlock, pSrcBlock->info.rows);
   if (code != TSDB_CODE_SUCCESS) {
     return code;
   }
-  ASSERT(taosArrayGetSize(pSrcBlock->pDataBlock) >= 3);
-  SColumnInfoData* pStartTsCol = taosArrayGet(pSrcBlock->pDataBlock, START_TS_COLUMN_INDEX);
-  TSKEY*           startData = (TSKEY*)pStartTsCol->pData;
-  SColumnInfoData* pEndTsCol = taosArrayGet(pSrcBlock->pDataBlock, END_TS_COLUMN_INDEX);
-  TSKEY*           endData = (TSKEY*)pEndTsCol->pData;
-  SColumnInfoData* pUidCol = taosArrayGet(pSrcBlock->pDataBlock, UID_COLUMN_INDEX);
-  uint64_t*        uidCol = (uint64_t*)pUidCol->pData;
 
-  SColumnInfoData* pDestStartCol = taosArrayGet(pDestBlock->pDataBlock, START_TS_COLUMN_INDEX);
+  SColumnInfoData* pSrcStartTsCol = (SColumnInfoData*)taosArrayGet(pSrcBlock->pDataBlock, START_TS_COLUMN_INDEX);
-  SColumnInfoData* pDestEndCol = taosArrayGet(pDestBlock->pDataBlock, END_TS_COLUMN_INDEX);
+  SColumnInfoData* pSrcEndTsCol = (SColumnInfoData*)taosArrayGet(pSrcBlock->pDataBlock, END_TS_COLUMN_INDEX);
-  SColumnInfoData* pDestUidCol = taosArrayGet(pDestBlock->pDataBlock, UID_COLUMN_INDEX);
+  SColumnInfoData* pSrcUidCol = taosArrayGet(pSrcBlock->pDataBlock, UID_COLUMN_INDEX);
-  SColumnInfoData* pDestGpCol = taosArrayGet(pDestBlock->pDataBlock, GROUPID_COLUMN_INDEX);
+  uint64_t*        srcUidData = (uint64_t*)pSrcUidCol->pData;
-  SColumnInfoData* pDestCalStartTsCol = taosArrayGet(pDestBlock->pDataBlock, CALCULATE_START_TS_COLUMN_INDEX);
+  SColumnInfoData* pSrcGpCol = taosArrayGet(pSrcBlock->pDataBlock, GROUPID_COLUMN_INDEX);
-  SColumnInfoData* pDestCalEndTsCol = taosArrayGet(pDestBlock->pDataBlock, CALCULATE_END_TS_COLUMN_INDEX);
+  uint64_t*        srcGp = (uint64_t*)pSrcGpCol->pData;
-  int32_t          dummy = 0;
+  ASSERT(pSrcStartTsCol->info.type == TSDB_DATA_TYPE_TIMESTAMP);
+  TSKEY*  srcStartTsCol = (TSKEY*)pSrcStartTsCol->pData;
+  TSKEY*  srcEndTsCol = (TSKEY*)pSrcEndTsCol->pData;
   int64_t version = pSrcBlock->info.version - 1;
   for (int32_t i = 0; i < pSrcBlock->info.rows; i++) {
-    uint64_t groupId = getGroupIdByData(pInfo, uidCol[i], startData[i], version);
+    uint64_t srcUid = srcUidData[i];
-    colDataAppend(pDestStartCol, i, (const char*)(startData + i), false);
+    uint64_t groupId = srcGp[i];
-    colDataAppend(pDestEndCol, i, (const char*)(endData + i), false);
+    if (groupId == 0) {
-    colDataAppendNULL(pDestUidCol, i);
+      groupId = getGroupIdByData(pInfo, srcUid, srcStartTsCol[i], version);
-    colDataAppend(pDestGpCol, i, (const char*)&groupId, false);
+    }
-    colDataAppendNULL(pDestCalStartTsCol, i);
+    appendOneRowToStreamSpecialBlock(pDestBlock, srcStartTsCol + i, srcEndTsCol + i, srcUidData + i, &groupId, NULL);
-    colDataAppendNULL(pDestCalEndTsCol, i);
-    pDestBlock->info.rows++;
   }
   return TSDB_CODE_SUCCESS;
 }
@@ -1466,6 +1478,8 @@ static int32_t generateScanRange(SStreamScanInfo* pInfo, SSDataBlock* pSrcBlock,
     code = generateIntervalScanRange(pInfo, pSrcBlock, pDestBlock);
   } else if (isSessionWindow(pInfo) || isStateWindow(pInfo)) {
     code = generateSessionScanRange(pInfo, pSrcBlock, pDestBlock);
+  } else {
+    code = generateDeleteResultBlock(pInfo, pSrcBlock, pDestBlock);
   }
   pDestBlock->info.type = STREAM_CLEAR;
   pDestBlock->info.version = pSrcBlock->info.version;
@@ -1632,7 +1646,7 @@ static int32_t setBlockIntoRes(SStreamScanInfo* pInfo, const SSDataBlock* pBlock
   // currently only the tbname pseudo column
   if (pInfo->numOfPseudoExpr > 0) {
     int32_t code = addTagPseudoColumnData(&pInfo->readHandle, pInfo->pPseudoExpr, pInfo->numOfPseudoExpr, pInfo->pRes,
-                                          GET_TASKID(pTaskInfo));
+                                          pInfo->pRes->info.rows, GET_TASKID(pTaskInfo));
     if (code != TSDB_CODE_SUCCESS) {
       blockDataFreeRes((SSDataBlock*)pBlock);
       T_LONG_JMP(pTaskInfo->env, code);
@@ -1642,11 +1656,11 @@ static int32_t setBlockIntoRes(SStreamScanInfo* pInfo, const SSDataBlock* pBlock
   if (filter) {
     doFilter(pInfo->pCondition, pInfo->pRes, NULL, NULL);
   }
+
   blockDataUpdateTsWindow(pInfo->pRes, pInfo->primaryTsIndex);
   blockDataFreeRes((SSDataBlock*)pBlock);
 
   calBlockTbName(&pInfo->tbnameCalSup, pInfo->pRes);
-
   return 0;
 }
 
@@ -1876,6 +1890,11 @@ static SSDataBlock* doStreamScan(SOperatorInfo* pOperator) {
       pTSInfo->cond.startVersion = pTaskInfo->streamInfo.fillHistoryVer1 + 1;
       pTSInfo->cond.endVersion = pTaskInfo->streamInfo.fillHistoryVer2;
     }
+
+    /*resetTableScanInfo(pTSInfo, pWin);*/
+    tsdbReaderClose(pTSInfo->dataReader);
+    pTSInfo->dataReader = NULL;
+
     pTSInfo->scanTimes = 0;
     pTSInfo->currentGroupId = -1;
     pTaskInfo->streamInfo.recoverStep = STREAM_RECOVER_STEP__SCAN;
@@ -1889,6 +1908,10 @@ static SSDataBlock* doStreamScan(SOperatorInfo* pOperator) {
       return pBlock;
     }
     pTaskInfo->streamInfo.recoverStep = STREAM_RECOVER_STEP__NONE;
+    STableScanInfo* pTSInfo = pInfo->pTableScanOp->info;
+    pTSInfo->cond.startVersion = 0;
+    pTSInfo->cond.endVersion = -1;
+
     return NULL;
   }
 #endif
@@ -2143,7 +2166,9 @@ static SSDataBlock* doRawScan(SOperatorInfo* pOperator) {
         longjmp(pTaskInfo->env, TSDB_CODE_TSC_QUERY_CANCELLED);
       }
 
-      tsdbRetrieveDataBlockInfo(pInfo->dataReader, &pBlock->info);
+      int32_t rows = 0;
+      tsdbRetrieveDataBlockInfo(pInfo->dataReader, &rows, &pBlock->info.uid, &pBlock->info.window);
+      pBlock->info.rows = rows;
 
       SArray* pCols = tsdbRetrieveDataBlock(pInfo->dataReader, NULL);
       pBlock->pDataBlock = pCols;
@@ -4360,7 +4385,7 @@ static int32_t loadDataBlockFromOneTable2(SOperatorInfo* pOperator, STableMergeS
   // currently only the tbname pseudo column
   if (pTableScanInfo->pseudoSup.numOfExprs > 0) {
     int32_t code = addTagPseudoColumnData(&pTableScanInfo->readHandle, pTableScanInfo->pseudoSup.pExprInfo,
-                                          pTableScanInfo->pseudoSup.numOfExprs, pBlock, GET_TASKID(pTaskInfo));
+                                          pTableScanInfo->pseudoSup.numOfExprs, pBlock, pBlock->info.rows, GET_TASKID(pTaskInfo));
     if (code != TSDB_CODE_SUCCESS) {
       T_LONG_JMP(pTaskInfo->env, code);
     }
@@ -4477,7 +4502,7 @@ static int32_t loadDataBlockFromOneTable(SOperatorInfo* pOperator, STableMergeSc
   // currently only the tbname pseudo column
   if (pTableScanInfo->pseudoSup.numOfExprs > 0) {
     int32_t code = addTagPseudoColumnData(&pTableScanInfo->readHandle, pTableScanInfo->pseudoSup.pExprInfo,
-                                          pTableScanInfo->pseudoSup.numOfExprs, pBlock, GET_TASKID(pTaskInfo));
+                                          pTableScanInfo->pseudoSup.numOfExprs, pBlock, pBlock->info.rows, GET_TASKID(pTaskInfo));
     if (code != TSDB_CODE_SUCCESS) {
       T_LONG_JMP(pTaskInfo->env, code);
     }
@@ -4541,14 +4566,11 @@ static SSDataBlock* getTableDataBlockTemp(void* param) {
     }
 
     blockDataCleanup(pBlock);
-    SDataBlockInfo binfo = pBlock->info;
-    tsdbRetrieveDataBlockInfo(reader, &binfo);
 
-    blockDataEnsureCapacity(pBlock, binfo.rows);
+    int32_t rows = 0;
-    pBlock->info.type = binfo.type;
+    tsdbRetrieveDataBlockInfo(reader, &rows, &pBlock->info.uid, &pBlock->info.window);
-    pBlock->info.uid = binfo.uid;
+    blockDataEnsureCapacity(pBlock, rows);
-    pBlock->info.window = binfo.window;
+    pBlock->info.rows = rows;
-    pBlock->info.rows = binfo.rows;
 
     if (pQueryCond->order == TSDB_ORDER_ASC) {
       pQueryCond->twindows.skey = pBlock->info.window.ekey + 1;
@@ -4604,14 +4626,11 @@ static SSDataBlock* getTableDataBlock2(void* param) {
     }
 
     blockDataCleanup(pBlock);
-    SDataBlockInfo binfo = pBlock->info;
-    tsdbRetrieveDataBlockInfo(reader, &binfo);
 
-    blockDataEnsureCapacity(pBlock, binfo.rows);
+    int32_t rows = 0;
-    pBlock->info.type = binfo.type;
+    tsdbRetrieveDataBlockInfo(reader, &rows, &pBlock->info.uid, &pBlock->info.window);
-    pBlock->info.uid = binfo.uid;
+    blockDataEnsureCapacity(pBlock, rows);
-    pBlock->info.window = binfo.window;
+    pBlock->info.rows = rows;
-    pBlock->info.rows = binfo.rows;
 
     uint32_t status = 0;
     int32_t  code = loadDataBlockFromOneTable2(pOperator, pTableScanInfo, pBlock, &status);
@@ -4657,14 +4676,11 @@ static SSDataBlock* getTableDataBlock(void* param) {
     }
 
     blockDataCleanup(pBlock);
-    SDataBlockInfo binfo = pBlock->info;
-    tsdbRetrieveDataBlockInfo(reader, &binfo);
 
-    blockDataEnsureCapacity(pBlock, binfo.rows);
+    int32_t rows = 0;
-    pBlock->info.type = binfo.type;
+    tsdbRetrieveDataBlockInfo(reader, &rows, &pBlock->info.uid, &pBlock->info.window);
-    pBlock->info.uid = binfo.uid;
+    blockDataEnsureCapacity(pBlock, rows);
-    pBlock->info.window = binfo.window;
+    pBlock->info.rows = rows;
-    pBlock->info.rows = binfo.rows;
 
     uint32_t status = 0;
     int32_t  code = loadDataBlockFromOneTable(pOperator, pTableScanInfo, readerIdx, pBlock, &status);

source/libs/executor/src/timewindowoperator.c

@@ -1687,7 +1687,9 @@ void initIntervalDownStream(SOperatorInfo* downstream, uint16_t type, SAggSuppor
   SStreamScanInfo* pScanInfo = downstream->info;
   pScanInfo->windowSup.parentType = type;
   pScanInfo->windowSup.pIntervalAggSup = pSup;
+  if (!pScanInfo->pUpdateInfo) {
     pScanInfo->pUpdateInfo = updateInfoInitP(pInterval, pTwSup->waterMark);
+  }
   pScanInfo->interval = *pInterval;
   pScanInfo->twAggSup = *pTwSup;
 }
@@ -2453,7 +2455,6 @@ static SSDataBlock* doTimeslice(SOperatorInfo* pOperator) {
           } else {  // non-linear interpolation
             pSliceInfo->current =
                 taosTimeAdd(pSliceInfo->current, pInterval->interval, pInterval->intervalUnit, pInterval->precision);
-
           }
         }
 
@@ -3302,6 +3303,8 @@ SOperatorInfo* createStreamFinalIntervalOperatorInfo(SOperatorInfo* downstream,
       // for test 315360000000
       .deleteMark = 1000LL * 60LL * 60LL * 24LL * 365LL * 10LL,
       // .deleteMark = INT64_MAX,
+      .deleteMarkSaved = 0,
+      .calTriggerSaved = 0,
   };
   ASSERT(pInfo->twAggSup.calTrigger != STREAM_TRIGGER_MAX_DELAY);
   pInfo->primaryTsIndex = ((SColumnNode*)pIntervalPhyNode->window.pTspk)->slotId;

source/libs/function/CMakeLists.txt

@@ -12,6 +12,17 @@ target_include_directories(
     PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
 )
 
+IF (TD_LINUX_64 AND JEMALLOC_ENABLED)
+  ADD_DEFINITIONS(-DTD_JEMALLOC_ENABLED -I${CMAKE_BINARY_DIR}/build/include -L${CMAKE_BINARY_DIR}/build/lib -Wl,-rpath,${CMAKE_BINARY_DIR}/build/lib -ljemalloc)
+  SET(LINK_JEMALLOC "-L${CMAKE_BINARY_DIR}/build/lib -ljemalloc")
+ELSE ()
+  SET(LINK_JEMALLOC "")
+ENDIF ()
+
+IF (TD_LINUX_64 AND JEMALLOC_ENABLED)
+    ADD_DEPENDENCIES(function jemalloc)
+ENDIF ()
+
 target_link_libraries(
     function
     PRIVATE os 
@@ -21,7 +32,7 @@ target_link_libraries(
     PRIVATE qcom 
     PRIVATE scalar 
     PRIVATE transport 
-    PRIVATE stream
+    PRIVATE stream ${LINK_JEMALLOC}
     PUBLIC uv_a
 )
 
@@ -37,10 +48,15 @@ target_include_directories(
             "${TD_SOURCE_DIR}/include/os"
         PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
 )
+
+IF (TD_LINUX_64 AND JEMALLOC_ENABLED)
+    ADD_DEPENDENCIES(runUdf jemalloc)
+ENDIF ()
+
 target_link_libraries(
         runUdf
         PUBLIC uv_a
-        PRIVATE os util common nodes function
+        PRIVATE os util common nodes function ${LINK_JEMALLOC}
 )
 
 add_library(udf1 STATIC MODULE test/udf1.c)
@@ -54,8 +70,13 @@ target_include_directories(
             "${TD_SOURCE_DIR}/include/os"
         PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
 )
+
+IF (TD_LINUX_64 AND JEMALLOC_ENABLED)
+    ADD_DEPENDENCIES(udf1 jemalloc)
+ENDIF ()
+
 target_link_libraries(
-    udf1 PUBLIC os)
+    udf1 PUBLIC os ${LINK_JEMALLOC})
 
 add_library(udf2 STATIC MODULE test/udf2.c)
 target_include_directories(
@@ -68,8 +89,13 @@ target_include_directories(
             "${TD_SOURCE_DIR}/include/os"
         PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
 )
+
+IF (TD_LINUX_64 AND JEMALLOC_ENABLED)
+    ADD_DEPENDENCIES(udf2 jemalloc)
+ENDIF ()
+
 target_link_libraries(
-    udf2 PUBLIC os
+    udf2 PUBLIC os ${LINK_JEMALLOC}
 )
 
 #SET(EXECUTABLE_OUTPUT_PATH ${CMAKE_BINARY_DIR}/build/bin)
@@ -86,9 +112,13 @@ target_include_directories(
         PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
 )
 
+IF (TD_LINUX_64 AND JEMALLOC_ENABLED)
+    ADD_DEPENDENCIES(udfd jemalloc)
+ENDIF ()
+
 target_link_libraries(
     udfd
     PUBLIC uv_a
-        PRIVATE os util common nodes function
+    PRIVATE os util common nodes function ${LINK_JEMALLOC}
     )
 

source/libs/function/src/builtinsimpl.c

@@ -497,7 +497,7 @@ bool getCountFuncEnv(SFunctionNode* UNUSED_PARAM(pFunc), SFuncExecEnv* pEnv) {
   return true;
 }
 
-static FORCE_INLINE int32_t getNumOfElems(SqlFunctionCtx* pCtx) {
+static int32_t getNumOfElems(SqlFunctionCtx* pCtx) {
   int32_t numOfElem = 0;
 
   /*
@@ -2194,6 +2194,56 @@ int32_t leastSQRFunction(SqlFunctionCtx* pCtx) {
       break;
     }
 
+    case TSDB_DATA_TYPE_UTINYINT: {
+      uint8_t* plist = (uint8_t*)pCol->pData;
+      for (int32_t i = start; i < numOfRows + pInput->startRowIndex; ++i) {
+        if (pCol->hasNull && colDataIsNull_f(pCol->nullbitmap, i)) {
+          continue;
+        }
+        numOfElem++;
+        LEASTSQR_CAL(param, x, plist, i, pInfo->stepVal);
+      }
+      break;
+    }
+    case TSDB_DATA_TYPE_USMALLINT: {
+      uint16_t* plist = (uint16_t*)pCol->pData;
+      for (int32_t i = start; i < numOfRows + pInput->startRowIndex; ++i) {
+        if (pCol->hasNull && colDataIsNull_f(pCol->nullbitmap, i)) {
+          continue;
+        }
+
+        numOfElem++;
+        LEASTSQR_CAL(param, x, plist, i, pInfo->stepVal);
+      }
+      break;
+    }
+
+    case TSDB_DATA_TYPE_UINT: {
+      uint32_t* plist = (uint32_t*)pCol->pData;
+      for (int32_t i = start; i < numOfRows + pInput->startRowIndex; ++i) {
+        if (pCol->hasNull && colDataIsNull_f(pCol->nullbitmap, i)) {
+          continue;
+        }
+
+        numOfElem++;
+        LEASTSQR_CAL(param, x, plist, i, pInfo->stepVal);
+      }
+      break;
+    }
+
+    case TSDB_DATA_TYPE_UBIGINT: {
+      uint64_t* plist = (uint64_t*)pCol->pData;
+      for (int32_t i = start; i < numOfRows + pInput->startRowIndex; ++i) {
+        if (pCol->hasNull && colDataIsNull_f(pCol->nullbitmap, i)) {
+          continue;
+        }
+
+        numOfElem++;
+        LEASTSQR_CAL(param, x, plist, i, pInfo->stepVal);
+      }
+      break;
+    }
+
     case TSDB_DATA_TYPE_FLOAT: {
       float* plist = (float*)pCol->pData;
       for (int32_t i = start; i < numOfRows + pInput->startRowIndex; ++i) {

source/libs/function/src/functionMgt.c

@@ -251,6 +251,13 @@ bool fmIsSelectValueFunc(int32_t funcId) {
   return FUNCTION_TYPE_SELECT_VALUE == funcMgtBuiltins[funcId].type;
 }
 
+bool fmIsGroupKeyFunc(int32_t funcId) {
+  if (funcId < 0 || funcId >= funcMgtBuiltinsNum) {
+    return false;
+  }
+  return FUNCTION_TYPE_GROUP_KEY == funcMgtBuiltins[funcId].type;
+}
+
 void fmFuncMgtDestroy() {
   void* m = gFunMgtService.pFuncNameHashTable;
   if (m != NULL && atomic_val_compare_exchange_ptr((void**)&gFunMgtService.pFuncNameHashTable, m, 0) == m) {

source/libs/function/src/tudf.c

@@ -117,10 +117,29 @@ static int32_t udfSpawnUdfd(SUdfdData *pData) {
   char dnodeIdEnvItem[32] = {0};
   char thrdPoolSizeEnvItem[32] = {0};
   snprintf(dnodeIdEnvItem, 32, "%s=%d", "DNODE_ID", pData->dnodeId);
+
   float numCpuCores = 4;
   taosGetCpuCores(&numCpuCores);
   snprintf(thrdPoolSizeEnvItem, 32, "%s=%d", "UV_THREADPOOL_SIZE", (int)numCpuCores * 2);
-  char *envUdfd[] = {dnodeIdEnvItem, thrdPoolSizeEnvItem, NULL};
+
+  char pathTaosdLdLib[512] = {0};
+  size_t taosdLdLibPathLen = sizeof(pathTaosdLdLib);
+  uv_os_getenv("LD_LIBRARY_PATH", pathTaosdLdLib, &taosdLdLibPathLen);
+
+  char udfdPathLdLib[1024] = {0};
+  size_t udfdLdLibPathLen = strlen(tsUdfdLdLibPath);
+  strncpy(udfdPathLdLib, tsUdfdLdLibPath, udfdLdLibPathLen);
+  udfdPathLdLib[udfdLdLibPathLen] = ':';
+  strncpy(udfdPathLdLib + udfdLdLibPathLen + 1, pathTaosdLdLib, sizeof(udfdPathLdLib) - udfdLdLibPathLen);
+  if (udfdLdLibPathLen + taosdLdLibPathLen < 1024) {
+    fnInfo("udfd LD_LIBRARY_PATH: %s", udfdPathLdLib);
+  } else {
+    fnError("can not set correct udfd LD_LIBRARY_PATH");
+  }
+  char ldLibPathEnvItem[1024 + 32] = {0};
+  snprintf(ldLibPathEnvItem, 1024 + 32, "%s=%s", "LD_LIBRARY_PATH", udfdPathLdLib);
+
+  char *envUdfd[] = {dnodeIdEnvItem, thrdPoolSizeEnvItem, ldLibPathEnvItem, NULL};
   options.env = envUdfd;
 
   int err = uv_spawn(&pData->loop, &pData->process, &options);

source/libs/parser/src/parTranslater.c

@@ -5160,7 +5160,7 @@ static int32_t translateCreateIndex(STranslateContext* pCxt, SCreateIndexStmt* p
 static int32_t translateDropIndex(STranslateContext* pCxt, SDropIndexStmt* pStmt) {
   SMDropSmaReq dropSmaReq = {0};
   SName        name;
-  tNameExtractFullName(toName(pCxt->pParseCxt->acctId, pCxt->pParseCxt->db, pStmt->indexName, &name), dropSmaReq.name);
+  tNameExtractFullName(toName(pCxt->pParseCxt->acctId, pStmt->indexDbName, pStmt->indexName, &name), dropSmaReq.name);
   dropSmaReq.igNotExists = pStmt->ignoreNotExists;
   return buildCmdMsg(pCxt, TDMT_MND_DROP_SMA, (FSerializeFunc)tSerializeSMDropSmaReq, &dropSmaReq);
 }
@@ -5281,7 +5281,7 @@ static int32_t translateCreateTopic(STranslateContext* pCxt, SCreateTopicStmt* p
     code = buildCreateTopicReq(pCxt, pStmt, &createReq);
   }
   if (TSDB_CODE_SUCCESS == code) {
-    code = buildCmdMsg(pCxt, TDMT_MND_CREATE_TOPIC, (FSerializeFunc)tSerializeSCMCreateTopicReq, &createReq);
+    code = buildCmdMsg(pCxt, TDMT_MND_TMQ_CREATE_TOPIC, (FSerializeFunc)tSerializeSCMCreateTopicReq, &createReq);
   }
   tFreeSCMCreateTopicReq(&createReq);
   return code;
@@ -5295,7 +5295,7 @@ static int32_t translateDropTopic(STranslateContext* pCxt, SDropTopicStmt* pStmt
   tNameGetFullDbName(&name, dropReq.name);
   dropReq.igNotExists = pStmt->ignoreNotExists;
 
-  return buildCmdMsg(pCxt, TDMT_MND_DROP_TOPIC, (FSerializeFunc)tSerializeSMDropTopicReq, &dropReq);
+  return buildCmdMsg(pCxt, TDMT_MND_TMQ_DROP_TOPIC, (FSerializeFunc)tSerializeSMDropTopicReq, &dropReq);
 }
 
 static int32_t translateDropCGroup(STranslateContext* pCxt, SDropCGroupStmt* pStmt) {
@@ -5307,7 +5307,7 @@ static int32_t translateDropCGroup(STranslateContext* pCxt, SDropCGroupStmt* pSt
   dropReq.igNotExists = pStmt->ignoreNotExists;
   strcpy(dropReq.cgroup, pStmt->cgroup);
 
-  return buildCmdMsg(pCxt, TDMT_MND_MQ_DROP_CGROUP, (FSerializeFunc)tSerializeSMDropCgroupReq, &dropReq);
+  return buildCmdMsg(pCxt, TDMT_MND_TMQ_DROP_CGROUP, (FSerializeFunc)tSerializeSMDropCgroupReq, &dropReq);
 }
 
 static int32_t translateAlterLocal(STranslateContext* pCxt, SAlterLocalStmt* pStmt) {

source/libs/planner/src/planLogicCreater.c

@@ -587,6 +587,10 @@ static int32_t createIndefRowsFuncLogicNode(SLogicPlanContext* pCxt, SSelectStmt
   return code;
 }
 
+static bool isInterpFunc(int32_t funcId) {
+  return fmIsInterpFunc(funcId) || fmIsInterpPseudoColumnFunc(funcId) || fmIsGroupKeyFunc(funcId);
+}
+
 static int32_t createInterpFuncLogicNode(SLogicPlanContext* pCxt, SSelectStmt* pSelect, SLogicNode** pLogicNode) {
   if (!pSelect->hasInterpFunc) {
     return TSDB_CODE_SUCCESS;
@@ -602,7 +606,7 @@ static int32_t createInterpFuncLogicNode(SLogicPlanContext* pCxt, SSelectStmt* p
   pInterpFunc->node.resultDataOrder = pInterpFunc->node.requireDataOrder;
 
   // interp functions and _group_key functions
-  int32_t code = nodesCollectFuncs(pSelect, SQL_CLAUSE_SELECT, fmIsVectorFunc, &pInterpFunc->pFuncs);
+  int32_t code = nodesCollectFuncs(pSelect, SQL_CLAUSE_SELECT, isInterpFunc, &pInterpFunc->pFuncs);
   if (TSDB_CODE_SUCCESS == code) {
     code = rewriteExprsForSelect(pInterpFunc->pFuncs, pSelect, SQL_CLAUSE_SELECT);
   }

source/libs/qcom/src/queryUtil.c

@@ -456,6 +456,7 @@ int32_t cloneDbVgInfo(SDBVgInfo* pSrc, SDBVgInfo** pDst) {
     (*pDst)->vgHash = taosHashInit(taosHashGetSize(pSrc->vgHash), taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true,
                                    HASH_ENTRY_LOCK);
     if (NULL == (*pDst)->vgHash) {
+      taosMemoryFreeClear(*pDst);
       return TSDB_CODE_TSC_OUT_OF_MEMORY;
     }
 

source/libs/scalar/src/filter.c

@@ -1087,7 +1087,7 @@ int32_t filterAddUnitImpl(SFilterInfo *info, uint8_t optr, SFilterFieldId *left,
     if (tmp == NULL) {
       return TSDB_CODE_OUT_OF_MEMORY;
     }
-    info->units = tmp;
+    info->units = (SFilterUnit*)tmp;
     memset(info->units + psize, 0, sizeof(*info->units) * FILTER_DEFAULT_UNIT_SIZE);
   }
 
@@ -1633,12 +1633,12 @@ void filterDumpInfoToString(SFilterInfo *info, const char *msg, int32_t options)
 
           SValueNode *var = (SValueNode *)field->desc;
           SDataType  *dType = &var->node.resType;
-          if (dType->type == TSDB_DATA_TYPE_VALUE_ARRAY) {
+          //if (dType->type == TSDB_DATA_TYPE_VALUE_ARRAY) {
-            qDebug("VAL%d => [type:TS][val:[%" PRIi64 "] - [%" PRId64 "]]", i, *(int64_t *)field->data,
+          //  qDebug("VAL%d => [type:TS][val:[%" PRIi64 "] - [%" PRId64 "]]", i, *(int64_t *)field->data,
-                   *(((int64_t *)field->data) + 1));
+          //         *(((int64_t *)field->data) + 1));
-          } else {
+          //} else {
             qDebug("VAL%d => [type:%d][val:%" PRIx64 "]", i, dType->type, var->datum.i);  // TODO
-          }
+          //}
         } else if (field->data) {
           qDebug("VAL%d => [type:NIL][val:NIL]", i);  // TODO
         }
@@ -4059,11 +4059,13 @@ bool filterExecute(SFilterInfo *info, SSDataBlock *pSrc, SColumnInfoData **p, SC
     SArray *pList = taosArrayInit(1, POINTER_BYTES);
     taosArrayPush(pList, &pSrc);
 
-    FLT_ERR_RET(scalarCalculate(info->sclCtx.node, pList, &output));
+    int32_t code = scalarCalculate(info->sclCtx.node, pList, &output);
-    *p = output.columnData;
-
     taosArrayDestroy(pList);
 
+    FLT_ERR_RET(code);
+
+    *p = output.columnData;
+
     if (output.numOfQualified == output.numOfRows) {
       *pResultStatus = FILTER_RESULT_ALL_QUALIFIED;
     } else if (output.numOfQualified == 0) {

source/libs/scalar/src/scalar.c

@@ -896,6 +896,10 @@ int32_t sclExecCaseWhen(SCaseWhenNode *node, SScalarCtx *ctx, SScalarParam *outp
 
   SCL_ERR_JRET(sclGetNodeRes(pWhenThen->pWhen, ctx, &pWhen));
   SCL_ERR_JRET(sclGetNodeRes(pWhenThen->pThen, ctx, &pThen));
+  if (NULL == pWhen || NULL == pThen) {
+    sclError("invalid when/then in whenThen list");
+    SCL_ERR_JRET(TSDB_CODE_INVALID_PARA);
+  }
 
   if (pCase) {
     vectorCompare(pCase, pWhen, &comp, TSDB_ORDER_ASC, OP_TYPE_EQUAL);

source/libs/stream/inc/streamInc.h

@@ -43,6 +43,9 @@ int32_t streamBroadcastToChildren(SStreamTask* pTask, const SSDataBlock* pBlock)
 
 int32_t tEncodeStreamRetrieveReq(SEncoder* pEncoder, const SStreamRetrieveReq* pReq);
 
+int32_t streamDispatchOneRecoverFinishReq(SStreamTask* pTask, const SStreamRecoverFinishReq* pReq, int32_t vgId,
+                                          SEpSet* pEpSet);
+
 SStreamQueueItem* streamMergeQueueItem(SStreamQueueItem* dst, SStreamQueueItem* elem);
 void              streamFreeQitem(SStreamQueueItem* data);
 

source/libs/stream/src/streamDispatch.c

@@ -239,7 +239,7 @@ int32_t streamDispatchOneRecoverFinishReq(SStreamTask* pTask, const SStreamRecov
 
   msg.contLen = tlen + sizeof(SMsgHead);
   msg.pCont = buf;
-  msg.msgType = TDMT_VND_STREAM_RECOVER_FINISH;
+  msg.msgType = TDMT_STREAM_RECOVER_FINISH;
 
   tmsgSendReq(pEpSet, &msg);
 
@@ -250,7 +250,7 @@ FAIL:
   return code;
 }
 
-int32_t streamDispatchOneReq(SStreamTask* pTask, const SStreamDispatchReq* pReq, int32_t vgId, SEpSet* pEpSet) {
+int32_t streamDispatchOneDataReq(SStreamTask* pTask, const SStreamDispatchReq* pReq, int32_t vgId, SEpSet* pEpSet) {
   void*   buf = NULL;
   int32_t code = -1;
   SRpcMsg msg = {0};
@@ -292,13 +292,19 @@ FAIL:
 
 int32_t streamSearchAndAddBlock(SStreamTask* pTask, SStreamDispatchReq* pReqs, SSDataBlock* pDataBlock, int32_t vgSz,
                                 int64_t groupId) {
-  char* ctbName;
+  char* ctbName = taosMemoryCalloc(1, TSDB_TABLE_FNAME_LEN);
+  if (ctbName == NULL) {
+    return -1;
+  }
+
   if (pDataBlock->info.parTbName[0]) {
-    ctbName = taosMemoryCalloc(1, TSDB_TABLE_NAME_LEN);
     snprintf(ctbName, TSDB_TABLE_NAME_LEN, "%s.%s", pTask->shuffleDispatcher.dbInfo.db, pDataBlock->info.parTbName);
   } else {
-    ctbName = buildCtbNameByGroupId(pTask->shuffleDispatcher.stbFullName, groupId);
+    char* ctbShortName = buildCtbNameByGroupId(pTask->shuffleDispatcher.stbFullName, groupId);
+    snprintf(ctbName, TSDB_TABLE_NAME_LEN, "%s.%s", pTask->shuffleDispatcher.dbInfo.db, ctbShortName);
+    taosMemoryFree(ctbShortName);
   }
+
   SArray* vgInfo = pTask->shuffleDispatcher.dbInfo.pVgroupInfos;
 
   /*uint32_t hashValue = MurmurHash3_32(ctbName, strlen(ctbName));*/
@@ -365,7 +371,7 @@ int32_t streamDispatchAllBlocks(SStreamTask* pTask, const SStreamDataBlock* pDat
     qDebug("dispatch from task %d (child id %d) to down stream task %d in vnode %d", pTask->taskId, pTask->selfChildId,
            downstreamTaskId, vgId);
 
-    if (streamDispatchOneReq(pTask, &req, vgId, pEpSet) < 0) {
+    if (streamDispatchOneDataReq(pTask, &req, vgId, pEpSet) < 0) {
       goto FAIL_FIXED_DISPATCH;
     }
     code = 0;
@@ -427,7 +433,7 @@ int32_t streamDispatchAllBlocks(SStreamTask* pTask, const SStreamDataBlock* pDat
       if (pReqs[i].blockNum > 0) {
         // send
         SVgroupInfo* pVgInfo = taosArrayGet(vgInfo, i);
-        if (streamDispatchOneReq(pTask, &pReqs[i], pVgInfo->vgId, &pVgInfo->epSet) < 0) {
+        if (streamDispatchOneDataReq(pTask, &pReqs[i], pVgInfo->vgId, &pVgInfo->epSet) < 0) {
           goto FAIL_SHUFFLE_DISPATCH;
         }
       }

source/libs/stream/src/streamExec.c

@@ -90,6 +90,8 @@ int32_t streamScanExec(SStreamTask* pTask, int32_t batchSz) {
 
   void* exec = pTask->exec.executor;
 
+  qSetStreamOpOpen(exec);
+
   while (1) {
     SArray* pRes = taosArrayInit(0, sizeof(SSDataBlock));
     if (pRes == NULL) {
@@ -127,7 +129,10 @@ int32_t streamScanExec(SStreamTask* pTask, int32_t batchSz) {
     qRes->type = STREAM_INPUT__DATA_BLOCK;
     qRes->blocks = pRes;
     streamTaskOutput(pTask, qRes);
-    // TODO stream sched dispatch
+
+    if (pTask->outputType == TASK_OUTPUT__FIXED_DISPATCH || pTask->outputType == TASK_OUTPUT__SHUFFLE_DISPATCH) {
+      streamDispatch(pTask);
+    }
   }
   return 0;
 }

source/libs/stream/src/streamRecover.c

@@ -36,6 +36,7 @@ int32_t streamSourceRecoverPrepareStep1(SStreamTask* pTask, int64_t ver) {
 }
 
 int32_t streamBuildSourceRecover1Req(SStreamTask* pTask, SStreamRecoverStep1Req* pReq) {
+  pReq->msgHead.vgId = pTask->nodeId;
   pReq->streamId = pTask->streamId;
   pReq->taskId = pTask->taskId;
   return 0;
@@ -44,10 +45,10 @@ int32_t streamBuildSourceRecover1Req(SStreamTask* pTask, SStreamRecoverStep1Req*
 int32_t streamSourceRecoverScanStep1(SStreamTask* pTask) {
   //
   return streamScanExec(pTask, 100);
-  // TODO next: dispatch msg to launch scan step2
 }
 
 int32_t streamBuildSourceRecover2Req(SStreamTask* pTask, SStreamRecoverStep2Req* pReq) {
+  pReq->msgHead.vgId = pTask->nodeId;
   pReq->streamId = pTask->streamId;
   pReq->taskId = pTask->taskId;
   return 0;
@@ -64,11 +65,20 @@ int32_t streamSourceRecoverScanStep2(SStreamTask* pTask, int64_t ver) {
 int32_t streamDispatchRecoverFinishReq(SStreamTask* pTask) {
   SStreamRecoverFinishReq req = {
       .streamId = pTask->streamId,
-      .taskId = pTask->taskId,
       .childId = pTask->selfChildId,
   };
+  // serialize
   if (pTask->outputType == TASK_OUTPUT__FIXED_DISPATCH) {
+    req.taskId = pTask->fixedEpDispatcher.taskId;
+    streamDispatchOneRecoverFinishReq(pTask, &req, pTask->fixedEpDispatcher.nodeId, &pTask->fixedEpDispatcher.epSet);
   } else if (pTask->outputType == TASK_OUTPUT__SHUFFLE_DISPATCH) {
+    SArray* vgInfo = pTask->shuffleDispatcher.dbInfo.pVgroupInfos;
+    int32_t vgSz = taosArrayGetSize(vgInfo);
+    for (int32_t i = 0; i < vgSz; i++) {
+      SVgroupInfo* pVgInfo = taosArrayGet(vgInfo, i);
+      req.taskId = pVgInfo->taskId;
+      streamDispatchOneRecoverFinishReq(pTask, &req, pVgInfo->vgId, &pVgInfo->epSet);
+    }
   }
   return 0;
 }
@@ -76,9 +86,9 @@ int32_t streamDispatchRecoverFinishReq(SStreamTask* pTask) {
 // agg
 int32_t streamAggRecoverPrepare(SStreamTask* pTask) {
   void* exec = pTask->exec.executor;
-  if (qStreamSetParamForRecover(exec) < 0) {
+  /*if (qStreamSetParamForRecover(exec) < 0) {*/
-    return -1;
+  /*return -1;*/
-  }
+  /*}*/
   pTask->recoverWaitingChild = taosArrayGetSize(pTask->childEpInfo);
   return 0;
 }
@@ -96,11 +106,13 @@ int32_t streamAggChildrenRecoverFinish(SStreamTask* pTask) {
 }
 
 int32_t streamProcessRecoverFinishReq(SStreamTask* pTask, int32_t childId) {
+  if (pTask->taskLevel == TASK_LEVEL__AGG) {
     int32_t left = atomic_sub_fetch_32(&pTask->recoverWaitingChild, 1);
     ASSERT(left >= 0);
     if (left == 0) {
       streamAggChildrenRecoverFinish(pTask);
     }
+  }
   return 0;
 }
 

source/libs/sync/src/syncMain.c

@@ -394,7 +394,7 @@ int32_t syncEndSnapshot(int64_t rid) {
     SSyncLogStoreData* pData = pSyncNode->pLogStore->data;
     code = walEndSnapshot(pData->pWal);
     if (code != 0) {
-      sError("vgId:%d, wal snapshot end error since:%s", pSyncNode->vgId, terrstr(terrno));
+      sError("vgId:%d, wal snapshot end error since:%s", pSyncNode->vgId, terrstr());
 
       syncNodeRelease(pSyncNode);
       return -1;
@@ -2610,7 +2610,7 @@ static void syncNodeEqElectTimer(void* param, void* tmrId) {
                                             pSyncNode->vgId, pSyncNode);
   SRpcMsg      rpcMsg;
   syncTimeout2RpcMsg(pSyncMsg, &rpcMsg);
-  if (pSyncNode->syncEqMsg != NULL) {
+  if (pSyncNode->syncEqMsg != NULL && pSyncNode->msgcb != NULL && pSyncNode->msgcb->putToQueueFp != NULL) {
     int32_t code = pSyncNode->syncEqMsg(pSyncNode->msgcb, &rpcMsg);
     if (code != 0) {
       sError("vgId:%d, sync enqueue elect msg error, code:%d", pSyncNode->vgId, code);
@@ -3227,6 +3227,7 @@ int32_t syncNodeDoCommit(SSyncNode* ths, SyncIndex beginIndex, SyncIndex endInde
           // ASSERT(pEntry != NULL);
           if (code != 0 || pEntry == NULL) {
             syncNodeErrorLog(ths, "get log entry error");
+            sFatal("vgId:%d, get log entry %" PRId64 " error when commit since %s", ths->vgId, i, terrstr());
             continue;
           }
         }

source/libs/sync/src/syncTimeout.c

@@ -76,7 +76,7 @@ int32_t syncNodeTimerRoutine(SSyncNode* ths) {
     SSyncLogStoreData* pData = ths->pLogStore->data;
     int32_t            code = walEndSnapshot(pData->pWal);
     if (code != 0) {
-      sError("vgId:%d, wal snapshot end error since:%s", ths->vgId, terrstr(terrno));
+      sError("vgId:%d, timer wal snapshot end error since:%s", ths->vgId, terrstr());
       return -1;
     } else {
       do {

source/libs/transport/inc/transComm.h

@@ -428,6 +428,7 @@ void    transDestoryExHandle(void* handle);
 int32_t transGetRefMgt();
 int32_t transGetInstMgt();
 
+void transHttpEnvDestroy();
 #ifdef __cplusplus
 }
 #endif

source/libs/transport/src/thttp.c

@@ -20,11 +20,30 @@
 #include "thttp.h"
 #include "taoserror.h"
 #include "tlog.h"
+#include "transComm.h"
 
 // clang-format on
 
 #define HTTP_RECV_BUF_SIZE 1024
 
+typedef struct SHttpModule {
+  uv_loop_t*  loop;
+  SAsyncPool* asyncPool;
+  TdThread    thread;
+} SHttpModule;
+
+typedef struct SHttpMsg {
+  queue         q;
+  char*         server;
+  int32_t       port;
+  char*         cont;
+  int32_t       len;
+  EHttpCompFlag flag;
+  int8_t        quit;
+  SHttpModule*  http;
+
+} SHttpMsg;
+
 typedef struct SHttpClient {
   uv_connect_t       conn;
   uv_tcp_t           tcp;
@@ -33,8 +52,20 @@ typedef struct SHttpClient {
   char*              rbuf;
   char*              addr;
   uint16_t           port;
+  struct sockaddr_in dest;
 } SHttpClient;
 
+static TdThreadOnce transHttpInit = PTHREAD_ONCE_INIT;
+static SHttpModule* thttp = NULL;
+static void         transHttpEnvInit();
+
+static void    httpHandleReq(SHttpMsg* msg);
+static void    httpHandleQuit(SHttpMsg* msg);
+static int32_t httpSendQuit();
+
+static int32_t taosSendHttpReportImpl(const char* server, uint16_t port, char* pCont, int32_t contLen,
+                                      EHttpCompFlag flag);
+
 static int32_t taosBuildHttpHeader(const char* server, int32_t contLen, char* pHead, int32_t headLen,
                                    EHttpCompFlag flag) {
   if (flag == HTTP_FLAT) {
@@ -53,6 +84,7 @@ static int32_t taosBuildHttpHeader(const char* server, int32_t contLen, char* pH
                     "Content-Length: %d\n\n",
                     server, contLen);
   } else {
+    terrno = TSDB_CODE_INVALID_CFG;
     return -1;
   }
 }
@@ -126,88 +158,10 @@ _OVER:
   return code;
 }
 
-static FORCE_INLINE void destroyHttpClient(SHttpClient* cli) {
-  taosMemoryFree(cli->wbuf);
-  taosMemoryFree(cli->rbuf);
-  taosMemoryFree(cli->addr);
-  taosMemoryFree(cli);
-}
-static FORCE_INLINE void clientCloseCb(uv_handle_t* handle) {
-  SHttpClient* cli = handle->data;
-  destroyHttpClient(cli);
-}
-static FORCE_INLINE void clientAllocBuffCb(uv_handle_t* handle, size_t suggested_size, uv_buf_t* buf) {
-  SHttpClient* cli = handle->data;
-  buf->base = cli->rbuf;
-  buf->len = HTTP_RECV_BUF_SIZE;
-}
-static FORCE_INLINE void clientRecvCb(uv_stream_t* handle, ssize_t nread, const uv_buf_t* buf) {
-  SHttpClient* cli = handle->data;
-  if (nread < 0) {
-    uError("http-report recv error:%s", uv_err_name(nread));
-  } else {
-    uTrace("http-report succ to recv %d bytes", (int32_t)nread);
-  }
-  if (!uv_is_closing((uv_handle_t*)&cli->tcp)) {
-    uv_close((uv_handle_t*)&cli->tcp, clientCloseCb);
-  } else {
-    destroyHttpClient(cli);
-  }
-}
-static void clientSentCb(uv_write_t* req, int32_t status) {
-  SHttpClient* cli = req->data;
-  if (status != 0) {
-    terrno = TAOS_SYSTEM_ERROR(status);
-    uError("http-report failed to send data %s", uv_strerror(status));
-    if (!uv_is_closing((uv_handle_t*)&cli->tcp)) {
-      uv_close((uv_handle_t*)&cli->tcp, clientCloseCb);
-    } else {
-      destroyHttpClient(cli);
-    }
-    return;
-  } else {
-    uTrace("http-report succ to send data");
-  }
-  status = uv_read_start((uv_stream_t*)&cli->tcp, clientAllocBuffCb, clientRecvCb);
-  if (status != 0) {
-    terrno = TAOS_SYSTEM_ERROR(status);
-    uError("http-report failed to recv data,reason:%s, dst:%s:%d", uv_strerror(status), cli->addr, cli->port);
-    if (!uv_is_closing((uv_handle_t*)&cli->tcp)) {
-      uv_close((uv_handle_t*)&cli->tcp, clientCloseCb);
-    } else {
-      destroyHttpClient(cli);
-    }
-  }
-}
-static void clientConnCb(uv_connect_t* req, int32_t status) {
-  SHttpClient* cli = req->data;
-  if (status != 0) {
-    terrno = TAOS_SYSTEM_ERROR(status);
-    uError("http-report failed to conn to server, reason:%s, dst:%s:%d", uv_strerror(status), cli->addr, cli->port);
-    if (!uv_is_closing((uv_handle_t*)&cli->tcp)) {
-      uv_close((uv_handle_t*)&cli->tcp, clientCloseCb);
-    } else {
-      destroyHttpClient(cli);
-    }
-    return;
-  }
-  status = uv_write(&cli->req, (uv_stream_t*)&cli->tcp, cli->wbuf, 2, clientSentCb);
-  if (0 != status) {
-    terrno = TAOS_SYSTEM_ERROR(status);
-    uError("http-report failed to send data,reason:%s, dst:%s:%d", uv_strerror(status), cli->addr, cli->port);
-    if (!uv_is_closing((uv_handle_t*)&cli->tcp)) {
-      uv_close((uv_handle_t*)&cli->tcp, clientCloseCb);
-    } else {
-      destroyHttpClient(cli);
-    }
-  }
-}
-
 static FORCE_INLINE int32_t taosBuildDstAddr(const char* server, uint16_t port, struct sockaddr_in* dest) {
   uint32_t ip = taosGetIpv4FromFqdn(server);
   if (ip == 0xffffffff) {
-    terrno = TAOS_SYSTEM_ERROR(errno);
+    tError("http-report failed to get http server:%s since %s", server, errno == 0 ? "invalid http server" : terrstr());
-    uError("http-report failed to get http server:%s since %s", server, errno == 0 ? "invalid http server" : terrstr());
     return -1;
   }
   char buf[128] = {0};
@@ -215,27 +169,206 @@ static FORCE_INLINE int32_t taosBuildDstAddr(const char* server, uint16_t port,
   uv_ip4_addr(buf, port, dest);
   return 0;
 }
-int32_t taosSendHttpReport(const char* server, uint16_t port, char* pCont, int32_t contLen, EHttpCompFlag flag) {
-  struct sockaddr_in dest = {0};
-  if (taosBuildDstAddr(server, port, &dest) < 0) {
-    return -1;
-  }
-  if (flag == HTTP_GZIP) {
-    int32_t dstLen = taosCompressHttpRport(pCont, contLen);
-    if (dstLen > 0) {
-      contLen = dstLen;
-    } else {
-      flag = HTTP_FLAT;
-    }
-  }
-  terrno = 0;
 
-  char    header[2048] = {0};
+static void* httpThread(void* arg) {
-  int32_t headLen = taosBuildHttpHeader(server, contLen, header, sizeof(header), flag);
+  SHttpModule* http = (SHttpModule*)arg;
+  setThreadName("http-cli-send-thread");
+  uv_run(http->loop, UV_RUN_DEFAULT);
+  return NULL;
+}
+
+static void httpDestroyMsg(SHttpMsg* msg) {
+  if (msg == NULL) return;
+
+  taosMemoryFree(msg->server);
+  taosMemoryFree(msg->cont);
+  taosMemoryFree(msg);
+}
+static void httpAsyncCb(uv_async_t* handle) {
+  SAsyncItem*  item = handle->data;
+  SHttpModule* http = item->pThrd;
+
+  SHttpMsg *msg = NULL, *quitMsg = NULL;
+
+  queue wq;
+  taosThreadMutexLock(&item->mtx);
+  QUEUE_MOVE(&item->qmsg, &wq);
+  taosThreadMutexUnlock(&item->mtx);
+
+  int count = 0;
+  while (!QUEUE_IS_EMPTY(&wq)) {
+    queue* h = QUEUE_HEAD(&wq);
+    QUEUE_REMOVE(h);
+    msg = QUEUE_DATA(h, SHttpMsg, q);
+    if (msg->quit) {
+      quitMsg = msg;
+    } else {
+      httpHandleReq(msg);
+    }
+  }
+  if (quitMsg) httpHandleQuit(quitMsg);
+}
+
+static FORCE_INLINE void destroyHttpClient(SHttpClient* cli) {
+  taosMemoryFree(cli->wbuf[0].base);
+  taosMemoryFree(cli->wbuf[1].base);
+  taosMemoryFree(cli->wbuf);
+  taosMemoryFree(cli->rbuf);
+  taosMemoryFree(cli->addr);
+  taosMemoryFree(cli);
+}
+
+static FORCE_INLINE void clientCloseCb(uv_handle_t* handle) {
+  SHttpClient* cli = handle->data;
+  destroyHttpClient(cli);
+}
+
+static FORCE_INLINE void clientAllocBuffCb(uv_handle_t* handle, size_t suggested_size, uv_buf_t* buf) {
+  SHttpClient* cli = handle->data;
+  buf->base = cli->rbuf;
+  buf->len = HTTP_RECV_BUF_SIZE;
+}
+
+static FORCE_INLINE void clientRecvCb(uv_stream_t* handle, ssize_t nread, const uv_buf_t* buf) {
+  SHttpClient* cli = handle->data;
+  if (nread < 0) {
+    tError("http-report recv error:%s", uv_err_name(nread));
+  } else {
+    tTrace("http-report succ to recv %d bytes", (int32_t)nread);
+  }
+  if (!uv_is_closing((uv_handle_t*)&cli->tcp)) {
+    uv_close((uv_handle_t*)&cli->tcp, clientCloseCb);
+  }
+}
+static void clientSentCb(uv_write_t* req, int32_t status) {
+  SHttpClient* cli = req->data;
+  if (status != 0) {
+    tError("http-report failed to send data, reason: %s, dst:%s:%d", uv_strerror(status), cli->addr, cli->port);
+    if (!uv_is_closing((uv_handle_t*)&cli->tcp)) {
+      uv_close((uv_handle_t*)&cli->tcp, clientCloseCb);
+    }
+    return;
+  } else {
+    tTrace("http-report succ to send data");
+  }
+  status = uv_read_start((uv_stream_t*)&cli->tcp, clientAllocBuffCb, clientRecvCb);
+  if (status != 0) {
+    tError("http-report failed to recv data,reason:%s, dst:%s:%d", uv_strerror(status), cli->addr, cli->port);
+    if (!uv_is_closing((uv_handle_t*)&cli->tcp)) {
+      uv_close((uv_handle_t*)&cli->tcp, clientCloseCb);
+    }
+  }
+}
+static void clientConnCb(uv_connect_t* req, int32_t status) {
+  SHttpClient* cli = req->data;
+  if (status != 0) {
+    tError("http-report failed to conn to server, reason:%s, dst:%s:%d", uv_strerror(status), cli->addr, cli->port);
+    if (!uv_is_closing((uv_handle_t*)&cli->tcp)) {
+      uv_close((uv_handle_t*)&cli->tcp, clientCloseCb);
+    }
+    return;
+  }
+  status = uv_write(&cli->req, (uv_stream_t*)&cli->tcp, cli->wbuf, 2, clientSentCb);
+  if (0 != status) {
+    tError("http-report failed to send data,reason:%s, dst:%s:%d", uv_strerror(status), cli->addr, cli->port);
+    if (!uv_is_closing((uv_handle_t*)&cli->tcp)) {
+      uv_close((uv_handle_t*)&cli->tcp, clientCloseCb);
+    }
+  }
+}
+
+int32_t httpSendQuit() {
+  SHttpMsg* msg = taosMemoryCalloc(1, sizeof(SHttpMsg));
+  msg->quit = 1;
+
+  SHttpModule* load = atomic_load_ptr(&thttp);
+  if (load == NULL) {
+    httpDestroyMsg(msg);
+    tError("http-report already released");
+    return -1;
+  } else {
+    msg->http = load;
+  }
+  transAsyncSend(load->asyncPool, &(msg->q));
+  return 0;
+}
+
+static int32_t taosSendHttpReportImpl(const char* server, uint16_t port, char* pCont, int32_t contLen,
+                                      EHttpCompFlag flag) {
+  SHttpMsg* msg = taosMemoryMalloc(sizeof(SHttpMsg));
+  msg->server = strdup(server);
+  msg->port = port;
+  msg->cont = taosMemoryMalloc(contLen);
+  memcpy(msg->cont, pCont, contLen);
+  msg->len = contLen;
+  msg->flag = flag;
+  msg->quit = 0;
+
+  SHttpModule* load = atomic_load_ptr(&thttp);
+  if (load == NULL) {
+    httpDestroyMsg(msg);
+    tError("http-report already released");
+    return -1;
+  } else {
+    msg->http = load;
+    transAsyncSend(load->asyncPool, &(msg->q));
+  }
+
+  return 0;
+}
+
+static void httpDestroyClientCb(uv_handle_t* handle) {
+  SHttpClient* http = handle->data;
+  destroyHttpClient(http);
+}
+static void httpWalkCb(uv_handle_t* handle, void* arg) {
+  // impl later
+  if (!uv_is_closing(handle)) {
+    uv_handle_type type = uv_handle_get_type(handle);
+    if (uv_handle_get_type(handle) == UV_TCP) {
+      uv_close(handle, httpDestroyClientCb);
+    } else {
+      uv_close(handle, NULL);
+    }
+  }
+  return;
+}
+static void httpHandleQuit(SHttpMsg* msg) {
+  SHttpModule* http = msg->http;
+  taosMemoryFree(msg);
+
+  uv_walk(http->loop, httpWalkCb, NULL);
+}
+static void httpHandleReq(SHttpMsg* msg) {
+  SHttpModule* http = msg->http;
+
+  struct sockaddr_in dest = {0};
+  if (taosBuildDstAddr(msg->server, msg->port, &dest) < 0) {
+    goto END;
+  }
+  if (msg->flag == HTTP_GZIP) {
+    int32_t dstLen = taosCompressHttpRport(msg->cont, msg->len);
+    if (dstLen > 0) {
+      msg->len = dstLen;
+    } else {
+      msg->flag = HTTP_FLAT;
+    }
+    if (dstLen < 0) {
+      goto END;
+    }
+  }
+
+  int32_t len = 2048;
+  char*   header = taosMemoryCalloc(1, len);
+  int32_t headLen = taosBuildHttpHeader(msg->server, msg->len, header, len, msg->flag);
+  if (headLen < 0) {
+    taosMemoryFree(header);
+    goto END;
+  }
 
   uv_buf_t* wb = taosMemoryCalloc(2, sizeof(uv_buf_t));
-  wb[0] = uv_buf_init((char*)header, headLen);  // stack var
+  wb[0] = uv_buf_init((char*)header, strlen(header));  //  heap var
-  wb[1] = uv_buf_init((char*)pCont, contLen);   //  heap var
+  wb[1] = uv_buf_init((char*)msg->cont, msg->len);     //  heap var
 
   SHttpClient* cli = taosMemoryCalloc(1, sizeof(SHttpClient));
   cli->conn.data = cli;
@@ -243,41 +376,71 @@ int32_t taosSendHttpReport(const char* server, uint16_t port, char* pCont, int32
   cli->req.data = cli;
   cli->wbuf = wb;
   cli->rbuf = taosMemoryCalloc(1, HTTP_RECV_BUF_SIZE);
-  cli->addr = tstrdup(server);
+  cli->addr = msg->server;
-  cli->port = port;
+  cli->port = msg->port;
+  cli->dest = dest;
+
+  taosMemoryFree(msg);
+
+  uv_tcp_init(http->loop, &cli->tcp);
 
-  uv_loop_t* loop = taosMemoryMalloc(sizeof(uv_loop_t));
-  int        err = uv_loop_init(loop);
-  if (err != 0) {
-    uError("http-report failed to init uv_loop, reason: %s", uv_strerror(err));
-    taosMemoryFree(loop);
-    terrno = TAOS_SYSTEM_ERROR(err);
-    destroyHttpClient(cli);
-    return terrno;
-  }
-  uv_tcp_init(loop, &cli->tcp);
   // set up timeout to avoid stuck;
   int32_t fd = taosCreateSocketWithTimeout(5);
-
   int     ret = uv_tcp_open((uv_tcp_t*)&cli->tcp, fd);
   if (ret != 0) {
-    uError("http-report failed to open socket, reason:%s, dst:%s:%d", uv_strerror(ret), cli->addr, cli->port);
+    tError("http-report failed to open socket, reason:%s, dst:%s:%d", uv_strerror(ret), cli->addr, cli->port);
     destroyHttpClient(cli);
-    uv_stop(loop);
+    return;
-    terrno = TAOS_SYSTEM_ERROR(ret);
-  } else {
-    ret = uv_tcp_connect(&cli->conn, &cli->tcp, (const struct sockaddr*)&dest, clientConnCb);
-    if (ret != 0) {
-      uError("http-report failed to connect to http-server, reason:%s, dst:%s:%d", uv_strerror(ret), cli->addr,
-             cli->port);
-      destroyHttpClient(cli);
-      uv_stop(loop);
-      terrno = TAOS_SYSTEM_ERROR(ret);
-    }
   }
 
-  uv_run(loop, UV_RUN_DEFAULT);
+  ret = uv_tcp_connect(&cli->conn, &cli->tcp, (const struct sockaddr*)&cli->dest, clientConnCb);
-  uv_loop_close(loop);
+  if (ret != 0) {
-  taosMemoryFree(loop);
+    tError("http-report failed to connect to http-server, reason:%s, dst:%s:%d", uv_strerror(ret), cli->addr,
-  return terrno;
+           cli->port);
+    destroyHttpClient(cli);
+  }
+  return;
+
+END:
+  tError("http-report failed to report, reason: %s, addr: %s:%d", terrstr(), msg->server, msg->port);
+  httpDestroyMsg(msg);
+}
+
+int32_t taosSendHttpReport(const char* server, uint16_t port, char* pCont, int32_t contLen, EHttpCompFlag flag) {
+  taosThreadOnce(&transHttpInit, transHttpEnvInit);
+  return taosSendHttpReportImpl(server, port, pCont, contLen, flag);
+}
+
+static void transHttpEnvInit() {
+  SHttpModule* http = taosMemoryMalloc(sizeof(SHttpModule));
+
+  http->loop = taosMemoryMalloc(sizeof(uv_loop_t));
+  uv_loop_init(http->loop);
+
+  http->asyncPool = transAsyncPoolCreate(http->loop, 1, http, httpAsyncCb);
+
+  int err = taosThreadCreate(&http->thread, NULL, httpThread, (void*)http);
+  if (err != 0) {
+    taosMemoryFree(http->loop);
+    taosMemoryFree(http);
+    http = NULL;
+  }
+  atomic_store_ptr(&thttp, http);
+}
+
+void transHttpEnvDestroy() {
+  SHttpModule* load = atomic_load_ptr(&thttp);
+  if (load == NULL) {
+    return;
+  }
+  httpSendQuit();
+  taosThreadJoin(load->thread, NULL);
+
+  TRANS_DESTROY_ASYNC_POOL_MSG(load->asyncPool, SHttpMsg, httpDestroyMsg);
+  transAsyncPoolDestroy(load->asyncPool);
+  uv_loop_close(load->loop);
+  taosMemoryFree(load->loop);
+  taosMemoryFree(load);
+
+  atomic_store_ptr(&thttp, NULL);
 }

source/libs/transport/src/trans.c

@@ -172,6 +172,8 @@ int32_t rpcInit() {
 }
 void rpcCleanup(void) {
   transCleanup();
+  transHttpEnvDestroy();
+
   return;
 }
 

source/libs/transport/src/transCli.c

@@ -187,17 +187,7 @@ static void cliReleaseUnfinishedMsg(SCliConn* conn) {
     snprintf(key, sizeof(key), "%s:%d", ip, (int)port); \
   } while (0)
 
-#define CONN_HOST_THREAD_IDX1(idx, exh, refId, pThrd) \
-  do {                                                \
-    if (exh == NULL) {                                \
-      idx = -1;                                       \
-    } else {                                          \
-      ASYNC_CHECK_HANDLE((exh), refId);               \
-      pThrd = (SCliThrd*)(exh)->pThrd;                \
-    }                                                 \
-  } while (0)
 #define CONN_PERSIST_TIME(para)   ((para) <= 90000 ? 90000 : (para))
-#define CONN_GET_HOST_THREAD(conn) (conn ? ((SCliConn*)conn)->hostThrd : NULL)
 #define CONN_GET_INST_LABEL(conn) (((STrans*)(((SCliThrd*)(conn)->hostThrd)->pTransInst))->label)
 
 #define CONN_GET_MSGCTX_BY_AHANDLE(conn, ahandle)                         \
@@ -217,6 +207,7 @@ static void cliReleaseUnfinishedMsg(SCliConn* conn) {
       tDebug("msg found, %" PRIu64 "", ahandle);                          \
     }                                                                     \
   } while (0)
+
 #define CONN_GET_NEXT_SENDMSG(conn)                 \
   do {                                              \
     int i = 0;                                      \
@@ -231,21 +222,6 @@ static void cliReleaseUnfinishedMsg(SCliConn* conn) {
     }                                               \
   } while (0)
 
-#define CONN_HANDLE_THREAD_QUIT(thrd) \
-  do {                                \
-    if (thrd->quit) {                 \
-      return;                         \
-    }                                 \
-  } while (0)
-
-#define CONN_HANDLE_BROKEN(conn) \
-  do {                           \
-    if (conn->broken) {          \
-      cliHandleExcept(conn);     \
-      return;                    \
-    }                            \
-  } while (0)
-
 #define CONN_SET_PERSIST_BY_APP(conn) \
   do {                                \
     if (conn->status == ConnNormal) { \

source/libs/wal/src/walMeta.c

@@ -288,6 +288,33 @@ void walAlignVersions(SWal* pWal) {
   pWal->vers.appliedVer = TMIN(pWal->vers.commitVer, pWal->vers.appliedVer);
 }
 
+bool walLogEntriesComplete(const SWal* pWal) {
+  int32_t sz = taosArrayGetSize(pWal->fileInfoSet);
+  bool    complete = true;
+  int32_t fileIdx = -1;
+  int64_t index = pWal->vers.firstVer;
+
+  while (++fileIdx < sz) {
+    SWalFileInfo* pFileInfo = taosArrayGet(pWal->fileInfoSet, fileIdx);
+    if (pFileInfo->firstVer != index) {
+      break;
+    }
+    index = pFileInfo->lastVer + ((fileIdx + 1 < sz) ? 1 : 0);
+  }
+  // empty is regarded as complete
+  if (sz != 0) {
+    complete = (index == pWal->vers.lastVer);
+  }
+
+  if (!complete) {
+    wError("vgId:%d, WAL log entries incomplete in range [%" PRId64 ", %" PRId64 "], aligned with snaphotVer:%" PRId64,
+           pWal->cfg.vgId, pWal->vers.firstVer, pWal->vers.lastVer, pWal->vers.snapshotVer);
+    terrno = TSDB_CODE_WAL_LOG_INCOMPLETE;
+  }
+
+  return complete;
+}
+
 int walCheckAndRepairMeta(SWal* pWal) {
   // load log files, get first/snapshot/last version info
   const char* logPattern = "^[0-9]+.log$";
@@ -401,6 +428,11 @@ int walCheckAndRepairMeta(SWal* pWal) {
   if (updateMeta) {
     (void)walSaveMeta(pWal);
   }
+
+  if (!walLogEntriesComplete(pWal)) {
+    return -1;
+  }
+
   return 0;
 }
 

source/libs/wal/src/walWrite.c

@@ -101,7 +101,7 @@ int32_t walRollback(SWal *pWal, int64_t ver) {
   taosThreadMutexLock(&pWal->mutex);
   int64_t code;
   char    fnameStr[WAL_FILE_LEN];
-  if (ver > pWal->vers.lastVer || ver < pWal->vers.commitVer) {
+  if (ver > pWal->vers.lastVer || ver < pWal->vers.commitVer || ver <= pWal->vers.snapshotVer) {
     terrno = TSDB_CODE_WAL_INVALID_VER;
     taosThreadMutexUnlock(&pWal->mutex);
     return -1;

source/libs/wal/test/walMetaTest.cpp

@@ -274,17 +274,17 @@ TEST_F(WalCleanEnv, rollbackMultiFile) {
   code = walRollback(pWal, 9);
   ASSERT_EQ(code, 0);
   ASSERT_EQ(pWal->vers.lastVer, 8);
+  code = walRollback(pWal, 6);
+  ASSERT_EQ(code, 0);
+  ASSERT_EQ(pWal->vers.lastVer, 5);
   code = walRollback(pWal, 5);
-  ASSERT_EQ(code, 0);
+  ASSERT_EQ(code, -1);
-  ASSERT_EQ(pWal->vers.lastVer, 4);
-  code = walRollback(pWal, 3);
-  ASSERT_EQ(code, 0);
 
-  ASSERT_EQ(pWal->vers.lastVer, 2);
+  ASSERT_EQ(pWal->vers.lastVer, 5);
 
-  code = walWrite(pWal, 3, 3, (void*)ranStr, ranStrLen);
+  code = walWrite(pWal, 6, 6, (void*)ranStr, ranStrLen);
   ASSERT_EQ(code, 0);
-  ASSERT_EQ(pWal->vers.lastVer, 3);
+  ASSERT_EQ(pWal->vers.lastVer, 6);
 
   code = walSaveMeta(pWal);
   ASSERT_EQ(code, 0);

source/util/src/terror.c

@@ -452,6 +452,7 @@ TAOS_DEFINE_ERROR(TSDB_CODE_WAL_INVALID_VER,              "WAL use invalid versi
 TAOS_DEFINE_ERROR(TSDB_CODE_WAL_OUT_OF_MEMORY,            "WAL out of memory")
 TAOS_DEFINE_ERROR(TSDB_CODE_WAL_LOG_NOT_EXIST,            "WAL log not exist")
 TAOS_DEFINE_ERROR(TSDB_CODE_WAL_CHKSUM_MISMATCH,          "WAL checksum mismatch")
+TAOS_DEFINE_ERROR(TSDB_CODE_WAL_LOG_INCOMPLETE,           "WAL log incomplete")
 
 // tfs
 TAOS_DEFINE_ERROR(TSDB_CODE_FS_INVLD_CFG,                 "tfs invalid mount config")

tests/develop-test/5-taos-tools/taosbenchmark/auto_create_table_json.py

@@ -25,7 +25,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor())
 

tests/develop-test/5-taos-tools/taosbenchmark/commandline.py

@@ -27,7 +27,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
 

tests/develop-test/5-taos-tools/taosbenchmark/custom_col_tag.py

@@ -24,7 +24,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
 

tests/develop-test/5-taos-tools/taosbenchmark/default_json.py

@@ -23,7 +23,7 @@ class TDTestCase:
         [TD-11510] taosBenchmark test cases
         """
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
 

tests/develop-test/5-taos-tools/taosbenchmark/demo.py

@@ -27,7 +27,7 @@ class TDTestCase:
         """
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
 

tests/develop-test/5-taos-tools/taosbenchmark/insert_alltypes_json.py

@@ -24,7 +24,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
 

tests/develop-test/5-taos-tools/taosbenchmark/invalid_commandline.py

@@ -24,7 +24,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
 

tests/develop-test/5-taos-tools/taosbenchmark/json_tag.py

@@ -24,7 +24,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
 

tests/develop-test/5-taos-tools/taosbenchmark/limit_offset_json.py

@@ -24,7 +24,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
 

tests/develop-test/5-taos-tools/taosbenchmark/query_json.py

@@ -29,7 +29,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
 

tests/develop-test/5-taos-tools/taosbenchmark/sample_csv_json.py

@@ -24,7 +24,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
 

tests/develop-test/5-taos-tools/taosbenchmark/sml_interlace.py

@@ -24,7 +24,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
 

tests/develop-test/5-taos-tools/taosbenchmark/sml_json_alltypes.py

@@ -24,7 +24,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
 

tests/develop-test/5-taos-tools/taosbenchmark/sml_telnet_alltypes.py

@@ -24,7 +24,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
 

tests/develop-test/5-taos-tools/taosbenchmark/taosadapter_json.py

@@ -25,7 +25,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
 

tests/develop-test/5-taos-tools/taosbenchmark/telnet_tcp.py

@@ -25,7 +25,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
 

tests/develop-test/5-taos-tools/taosdump/taosdumpTestInspect.py

@@ -27,7 +27,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
         self.tmpdir = "tmp"

tests/develop-test/5-taos-tools/taosdump/taosdumpTestTypeBigInt.py

@@ -27,7 +27,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
         self.tmpdir = "tmp"

tests/develop-test/5-taos-tools/taosdump/taosdumpTestTypeBinary.py

@@ -27,7 +27,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
         self.tmpdir = "tmp"

tests/develop-test/5-taos-tools/taosdump/taosdumpTestTypeBool.py

@@ -27,7 +27,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
         self.tmpdir = "tmp"

tests/develop-test/5-taos-tools/taosdump/taosdumpTestTypeDouble.py

@@ -28,7 +28,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
         self.tmpdir = "tmp"

tests/develop-test/5-taos-tools/taosdump/taosdumpTestTypeFloat.py

@@ -28,7 +28,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
         self.tmpdir = "tmp"

tests/develop-test/5-taos-tools/taosdump/taosdumpTestTypeInt.py

@@ -27,7 +27,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
         self.tmpdir = "tmp"

tests/develop-test/5-taos-tools/taosdump/taosdumpTestTypeJson.py

@@ -27,7 +27,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
         self.tmpdir = "tmp"

tests/develop-test/5-taos-tools/taosdump/taosdumpTestTypeSmallInt.py

@@ -27,7 +27,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
         self.tmpdir = "tmp"

tests/develop-test/5-taos-tools/taosdump/taosdumpTestTypeTinyInt.py

@@ -27,7 +27,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
         self.tmpdir = "tmp"

tests/develop-test/5-taos-tools/taosdump/taosdumpTestTypeUnsignedBigInt.py

@@ -27,7 +27,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
         self.tmpdir = "tmp"

tests/develop-test/5-taos-tools/taosdump/taosdumpTestTypeUnsignedInt.py

@@ -27,7 +27,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
         self.tmpdir = "tmp"

tests/develop-test/5-taos-tools/taosdump/taosdumpTestTypeUnsignedSmallInt.py

@@ -27,7 +27,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
         self.tmpdir = "tmp"

tests/develop-test/5-taos-tools/taosdump/taosdumpTestTypeUnsignedTinyInt.py

@@ -27,7 +27,7 @@ class TDTestCase:
         '''
         return
 
-    def init(self, conn, logSql):
+    def init(self, conn, logSql, replicaVarl=1):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
         self.tmpdir = "tmp"

tests/develop-test/test.py

@@ -72,8 +72,9 @@ if __name__ == "__main__":
     queryPolicy = 1
     createDnodeNums = 1
     restful = False
-    opts, args = getopt.gnu_getopt(sys.argv[1:], 'f:p:m:l:scghrd:k:e:N:M:Q:C:RD:', [
+    replicaVar = 1
-        'file=', 'path=', 'master', 'logSql', 'stop', 'cluster', 'valgrind', 'help', 'restart', 'updateCfgDict', 'killv', 'execCmd','dnodeNums','mnodeNums','queryPolicy','createDnodeNums','restful','adaptercfgupdate'])
+    opts, args = getopt.gnu_getopt(sys.argv[1:], 'f:p:m:l:scghrd:k:e:N:M:Q:C:RD:n:', [
+        'file=', 'path=', 'master', 'logSql', 'stop', 'cluster', 'valgrind', 'help', 'restart', 'updateCfgDict', 'killv', 'execCmd','dnodeNums','mnodeNums','queryPolicy','createDnodeNums','restful','adaptercfgupdate','replicaVar'])
     for key, value in opts:
         if key in ['-h', '--help']:
             tdLog.printNoPrefix(
@@ -95,8 +96,7 @@ if __name__ == "__main__":
             tdLog.printNoPrefix('-C create Dnode Numbers in one cluster')
             tdLog.printNoPrefix('-R restful realization form')
             tdLog.printNoPrefix('-D taosadapter update cfg dict ')
-
+            tdLog.printNoPrefix('-n the number of replicas')
-
             sys.exit(0)
 
         if key in ['-r', '--restart']:
@@ -168,6 +168,9 @@ if __name__ == "__main__":
                 print('adapter cfg update convert fail.')
                 sys.exit(0)
 
+        if key in ['-n', '--replicaVar']:
+            replicaVar = value
+
     if not execCmd == "":
         if restful:
             tAdapter.init(deployPath)
@@ -194,7 +197,7 @@ if __name__ == "__main__":
             processID = subprocess.check_output(psCmd, shell=True)
 
         for port in range(6030, 6041):
-            usePortPID = "lsof -i tcp:%d | grep LISTEn | awk '{print $2}'" % port
+            usePortPID = "lsof -i tcp:%d | grep LISTEN | awk '{print $2}'" % port
             processID = subprocess.check_output(usePortPID, shell=True)
 
             if processID:
@@ -206,11 +209,13 @@ if __name__ == "__main__":
             time.sleep(2)
 
         if restful:
-            toBeKilled = "taosadapter"
+            toBeKilled = "taosadapt"
 
-            killCmd = "ps -ef|grep -w %s| grep -v grep | awk '{print $2}' | xargs kill -TERM > /dev/null 2>&1" % toBeKilled
+            # killCmd = "ps -ef|grep -w %s| grep -v grep | awk '{print $2}' | xargs kill -TERM > /dev/null 2>&1" % toBeKilled
+            killCmd = f"pkill {toBeKilled}"
 
             psCmd = "ps -ef|grep -w %s| grep -v grep | awk '{print $2}'" % toBeKilled
+            # psCmd = f"pgrep {toBeKilled}"
             processID = subprocess.check_output(psCmd, shell=True)
 
             while(processID):
@@ -218,14 +223,14 @@ if __name__ == "__main__":
                 time.sleep(1)
                 processID = subprocess.check_output(psCmd, shell=True)
 
-            for port in range(6030, 6041):
+            port = 6041
-                usePortPID = "lsof -i tcp:%d | grep LISTEn | awk '{print $2}'" % port
+            usePortPID = f"lsof -i tcp:{port} | grep LISTEN | awk '{{print $2}}'"
             processID = subprocess.check_output(usePortPID, shell=True)
 
             if processID:
-                    killCmd = "kill -TERM %s" % processID
+                killCmd = f"kill -TERM {processID}"
                 os.system(killCmd)
-                fuserCmd = "fuser -k -n tcp %d" % port
+            fuserCmd = f"fuser -k -n tcp {port}"
             os.system(fuserCmd)
 
             tdLog.info('stop taosadapter')
@@ -339,6 +344,26 @@ if __name__ == "__main__":
                     print("check dnode ready")
             except Exception as r:
                 print(r)
+            if queryPolicy != 1:
+                queryPolicy=int(queryPolicy)
+                if restful:
+                    conn = taosrest.connect(url=f"http://{host}:6041")
+                else:
+                    conn = taos.connect(host,config=tdDnodes.getSimCfgPath())
+
+                cursor = conn.cursor()
+                cursor.execute("create qnode on dnode 1")
+                cursor.execute(f'alter local "queryPolicy" "{queryPolicy}"')
+                cursor.execute("show local variables")
+                res = cursor.fetchall()
+                for i in range(cursor.rowcount):
+                    if res[i][0] == "queryPolicy" :
+                        if int(res[i][1]) == int(queryPolicy):
+                            tdLog.success(f'alter queryPolicy to {queryPolicy} successfully')
+                        else:
+                            tdLog.debug(res)
+                            tdLog.exit(f"alter queryPolicy to  {queryPolicy} failed")
+                            
         if ucase is not None and hasattr(ucase, 'noConn') and ucase.noConn == True:
             conn = None
         else:
@@ -453,6 +478,26 @@ if __name__ == "__main__":
             except Exception as r:
                 print(r)
 
+            if queryPolicy != 1:
+                queryPolicy=int(queryPolicy)
+                if restful:
+                    conn = taosrest.connect(url=f"http://{host}:6041")
+                else:
+                    conn = taos.connect(host,config=tdDnodes.getSimCfgPath())
+
+                cursor = conn.cursor()
+                cursor.execute("create qnode on dnode 1")
+                cursor.execute(f'alter local "queryPolicy" "{queryPolicy}"')
+                cursor.execute("show local variables")
+                res = cursor.fetchall()
+                for i in range(cursor.rowcount):
+                    if res[i][0] == "queryPolicy" :
+                        if int(res[i][1]) == int(queryPolicy):
+                            tdLog.success(f'alter queryPolicy to {queryPolicy} successfully')
+                        else:
+                            tdLog.debug(res)
+                            tdLog.exit(f"alter queryPolicy to  {queryPolicy} failed")
+                            
 
         if testCluster:
             tdLog.info("Procedures for testing cluster")
@@ -470,7 +515,7 @@ if __name__ == "__main__":
             if fileName == "all":
                 tdCases.runAllLinux(conn)
             else:
-                tdCases.runOneLinux(conn, fileName)
+                tdCases.runOneLinux(conn, fileName, replicaVar)
 
         if restart:
             if fileName == "all":
@@ -487,7 +532,7 @@ if __name__ == "__main__":
                         conn = taosrest.connect(url=f"http://{host}:6041")
                     tdLog.info("Procedures for tdengine deployed in %s" % (host))
                     tdLog.info("query test after taosd restart")
-                    tdCases.runOneLinux(conn, sp[0] + "_" + "restart.py")
+                    tdCases.runOneLinux(conn, sp[0] + "_" + "restart.py", replicaVar)
                 else:
                     tdLog.info("not need to query")