diff --git a/include/libs/qcom/query.h b/include/libs/qcom/query.h
index cb547ee6b3732e391ca943bcb9b3e9783d587e36..b6ada5a0c7c22bbbe1bd938faa1080464c98ae5a 100644
--- a/include/libs/qcom/query.h
+++ b/include/libs/qcom/query.h
@@ -33,6 +33,7 @@ typedef enum {
   JOB_TASK_STATUS_INIT,
   JOB_TASK_STATUS_EXEC,
   JOB_TASK_STATUS_PART_SUCC,
+  JOB_TASK_STATUS_FETCH,
   JOB_TASK_STATUS_SUCC,
   JOB_TASK_STATUS_FAIL,
   JOB_TASK_STATUS_DROP,
diff --git a/source/client/src/clientTmq.c b/source/client/src/clientTmq.c
index 335db79a2b63edc0aa29943e102d8c00bc72c040..59a407656d47d2016fe00afb350851b89ae4e3fb 100644
--- a/source/client/src/clientTmq.c
+++ b/source/client/src/clientTmq.c
@@ -27,6 +27,8 @@
 #define EMPTY_BLOCK_POLL_IDLE_DURATION  10
 #define DEFAULT_AUTO_COMMIT_INTERVAL    5000
 
+typedef void (*__tmq_askep_fn_t)(tmq_t* pTmq, int32_t code, SDataBuf* pBuf, void* pParam);
+
 struct SMqMgmt {
   int8_t  inited;
   tmr_h   timer;
@@ -109,6 +111,11 @@ struct tmq_t {
   tsem_t        rspSem;
 };
 
+typedef struct SAskEpInfo {
+  int32_t code;
+  tsem_t  sem;
+} SAskEpInfo;
+
 enum {
   TMQ_VG_STATUS__IDLE = 0,
   TMQ_VG_STATUS__WAIT,
@@ -169,11 +176,10 @@ typedef struct {
 } SMqSubscribeCbParam;
 
 typedef struct {
-  int64_t refId;
-  int32_t epoch;
-  int32_t code;
-  int32_t async;
-  tsem_t  rspSem;
+  int64_t          refId;
+  int32_t          epoch;
+  void*            pParam;
+  __tmq_askep_fn_t pUserFn;
 } SMqAskEpCbParam;
 
 typedef struct {
@@ -189,16 +195,13 @@ typedef struct {
 typedef struct {
   int64_t        refId;
   int32_t        epoch;
-  int8_t         automatic;
-  int8_t         async;
   int32_t        waitingRspNum;
   int32_t        totalRspNum;
-  int32_t        rspErr;
-  tmq_commit_cb* userCb;
+  int32_t        code;
+  tmq_commit_cb* callbackFn;
   /*SArray*        successfulOffsets;*/
   /*SArray*        failedOffsets;*/
-  void*  userParam;
-  tsem_t rspSem;
+  void*          userParam;
 } SMqCommitCbParamSet;
 
 typedef struct {
@@ -209,12 +212,14 @@ typedef struct {
   tmq_t*               pTmq;
 } SMqCommitCbParam;
 
-static int32_t tmqAskEp(tmq_t* tmq, bool async);
+static int32_t doAskEp(tmq_t* tmq);
 static int32_t makeTopicVgroupKey(char* dst, const char* topicName, int32_t vg);
 static int32_t tmqCommitDone(SMqCommitCbParamSet* pParamSet);
 static int32_t doSendCommitMsg(tmq_t* tmq, SMqClientVg* pVg, const char* pTopicName, SMqCommitCbParamSet* pParamSet,
                                int32_t index, int32_t totalVgroups);
-static void tmqCommitRspCountDown(SMqCommitCbParamSet* pParamSet, int64_t consumerId, const char* pTopic, int32_t vgId);
+static void commitRspCountDown(SMqCommitCbParamSet* pParamSet, int64_t consumerId, const char* pTopic, int32_t vgId);
+static void asyncAskEp(tmq_t* pTmq, __tmq_askep_fn_t askEpFn, void* param);
+static void addToQueueCallbackFn(tmq_t* pTmq, int32_t code, SDataBuf* pDataBuf, void* param);
 
 tmq_conf_t* tmq_conf_new() {
   tmq_conf_t* conf = taosMemoryCalloc(1, sizeof(tmq_conf_t));
@@ -444,7 +449,7 @@ static int32_t tmqCommitCb(void* param, SDataBuf* pBuf, int32_t code) {
 //    taosMemoryFree(pBuf->pData);
 //    taosMemoryFree(pBuf->pEpSet);
 //
-//    tmqCommitRspCountDown(pParamSet, pParam->pTmq->consumerId, pParam->topicName, pParam->vgId);
+//    commitRspCountDown(pParamSet, pParam->pTmq->consumerId, pParam->topicName, pParam->vgId);
 //    return 0;
 //  }
 //
@@ -454,7 +459,7 @@ static int32_t tmqCommitCb(void* param, SDataBuf* pBuf, int32_t code) {
   taosMemoryFree(pBuf->pData);
   taosMemoryFree(pBuf->pEpSet);
 
-  tmqCommitRspCountDown(pParamSet, pParam->pTmq->consumerId, pParam->topicName, pParam->vgId);
+  commitRspCountDown(pParamSet, pParam->pTmq->consumerId, pParam->topicName, pParam->vgId);
   return 0;
 }
 
@@ -462,8 +467,7 @@ static int32_t doSendCommitMsg(tmq_t* tmq, SMqClientVg* pVg, const char* pTopicN
                                int32_t index, int32_t totalVgroups) {
   STqOffset* pOffset = taosMemoryCalloc(1, sizeof(STqOffset));
   if (pOffset == NULL) {
-    terrno = TSDB_CODE_OUT_OF_MEMORY;
-    return -1;
+    return TSDB_CODE_OUT_OF_MEMORY;
   }
 
   pOffset->val = pVg->currentOffset;
@@ -477,13 +481,13 @@ static int32_t doSendCommitMsg(tmq_t* tmq, SMqClientVg* pVg, const char* pTopicN
   int32_t code = 0;
   tEncodeSize(tEncodeSTqOffset, pOffset, len, code);
   if (code < 0) {
-    return -1;
+    return TSDB_CODE_INVALID_PARA;
   }
 
   void* buf = taosMemoryCalloc(1, sizeof(SMsgHead) + len);
   if (buf == NULL) {
     taosMemoryFree(pOffset);
-    return -1;
+    return TSDB_CODE_OUT_OF_MEMORY;
   }
 
   ((SMsgHead*)buf)->vgId = htonl(pVg->vgId);
@@ -500,7 +504,7 @@ static int32_t doSendCommitMsg(tmq_t* tmq, SMqClientVg* pVg, const char* pTopicN
   if (pParam == NULL) {
     taosMemoryFree(pOffset);
     taosMemoryFree(buf);
-    return -1;
+    return TSDB_CODE_OUT_OF_MEMORY;
   }
 
   pParam->params = pParamSet;
@@ -516,7 +520,7 @@ static int32_t doSendCommitMsg(tmq_t* tmq, SMqClientVg* pVg, const char* pTopicN
     taosMemoryFree(pOffset);
     taosMemoryFree(buf);
     taosMemoryFree(pParam);
-    return -1;
+    return TSDB_CODE_OUT_OF_MEMORY;
   }
 
   pMsgSendInfo->msgInfo = (SDataBuf){
@@ -547,129 +551,117 @@ static int32_t doSendCommitMsg(tmq_t* tmq, SMqClientVg* pVg, const char* pTopicN
 
   int64_t transporterId = 0;
   asyncSendMsgToServer(tmq->pTscObj->pAppInfo->pTransporter, &pVg->epSet, &transporterId, pMsgSendInfo);
-  return 0;
+
+  return TSDB_CODE_SUCCESS;
 }
 
-static int32_t tmqCommitMsgImpl(tmq_t* tmq, const TAOS_RES* msg, int8_t async, tmq_commit_cb* userCb, void* userParam) {
-  char*   topic;
-  int32_t vgId;
-  if (TD_RES_TMQ(msg)) {
-    SMqRspObj* pRspObj = (SMqRspObj*)msg;
-    topic = pRspObj->topic;
+static void asyncCommitOffset(tmq_t* tmq, const TAOS_RES* pRes, tmq_commit_cb* pCommitFp, void* userParam) {
+  char*   pTopicName = NULL;
+  int32_t vgId = 0;
+  int32_t code = 0;
+
+  if (pRes == NULL || tmq == NULL) {
+    pCommitFp(tmq, TSDB_CODE_INVALID_PARA, userParam);
+    return;
+  }
+
+  if (TD_RES_TMQ(pRes)) {
+    SMqRspObj* pRspObj = (SMqRspObj*)pRes;
+    pTopicName = pRspObj->topic;
     vgId = pRspObj->vgId;
-  } else if (TD_RES_TMQ_META(msg)) {
-    SMqMetaRspObj* pMetaRspObj = (SMqMetaRspObj*)msg;
-    topic = pMetaRspObj->topic;
+  } else if (TD_RES_TMQ_META(pRes)) {
+    SMqMetaRspObj* pMetaRspObj = (SMqMetaRspObj*)pRes;
+    pTopicName = pMetaRspObj->topic;
     vgId = pMetaRspObj->vgId;
-  } else if (TD_RES_TMQ_METADATA(msg)) {
-    SMqTaosxRspObj* pRspObj = (SMqTaosxRspObj*)msg;
-    topic = pRspObj->topic;
+  } else if (TD_RES_TMQ_METADATA(pRes)) {
+    SMqTaosxRspObj* pRspObj = (SMqTaosxRspObj*)pRes;
+    pTopicName = pRspObj->topic;
     vgId = pRspObj->vgId;
   } else {
-    return TSDB_CODE_TMQ_INVALID_MSG;
+    pCommitFp(tmq, TSDB_CODE_TMQ_INVALID_MSG, userParam);
+    return;
   }
 
   SMqCommitCbParamSet* pParamSet = taosMemoryCalloc(1, sizeof(SMqCommitCbParamSet));
   if (pParamSet == NULL) {
-    terrno = TSDB_CODE_OUT_OF_MEMORY;
-    return -1;
+    pCommitFp(tmq, TSDB_CODE_OUT_OF_MEMORY, userParam);
+    return;
   }
 
   pParamSet->refId = tmq->refId;
   pParamSet->epoch = tmq->epoch;
-  pParamSet->automatic = 0;
-  pParamSet->async = async;
-  pParamSet->userCb = userCb;
+  pParamSet->callbackFn = pCommitFp;
   pParamSet->userParam = userParam;
-  tsem_init(&pParamSet->rspSem, 0, 0);
-
-  int32_t code = -1;
 
-  taosThreadMutexLock(&tmq->lock);
   int32_t numOfTopics = taosArrayGetSize(tmq->clientTopics);
 
-  tscDebug("consumer:0x%" PRIx64 " user invoked commit offset for %d", tmq->consumerId, numOfTopics);
-  for (int32_t i = 0; i < numOfTopics; i++) {
+  tscDebug("consumer:0x%" PRIx64 " do manual commit offset for %s, vgId:%d", tmq->consumerId, pTopicName, vgId);
+
+  int32_t i = 0;
+  for (; i < numOfTopics; i++) {
     SMqClientTopic* pTopic = taosArrayGet(tmq->clientTopics, i);
-    if (strcmp(pTopic->topicName, topic) != 0) {
-      continue;
+    if (strcmp(pTopic->topicName, pTopicName) == 0) {
+      break;
     }
+  }
 
-    int32_t numOfVgroups = taosArrayGetSize(pTopic->vgs);
-    for (int32_t j = 0; j < numOfVgroups; j++) {
-      SMqClientVg* pVg = taosArrayGet(pTopic->vgs, j);
-      if (pVg->vgId != vgId) {
-        continue;
-      }
+  if (i == numOfTopics) {
+    tscWarn("consumer:0x%" PRIx64 " failed to find the specified topic:%s, total topics:%d", tmq->consumerId, pTopicName,
+            numOfTopics);
+    taosMemoryFree(pParamSet);
+    pCommitFp(tmq, TSDB_CODE_SUCCESS, userParam);
+    return;
+  }
 
-      if (pVg->currentOffset.type > 0 && !tOffsetEqual(&pVg->currentOffset, &pVg->committedOffset)) {
-        if (doSendCommitMsg(tmq, pVg, pTopic->topicName, pParamSet, j, numOfVgroups) < 0) {
-          tsem_destroy(&pParamSet->rspSem);
-          taosMemoryFree(pParamSet);
-          goto FAIL;
-        }
-        goto HANDLE_RSP;
-      }
+  SMqClientTopic* pTopic = taosArrayGet(tmq->clientTopics, i);
+
+  int32_t j = 0;
+  int32_t numOfVgroups = taosArrayGetSize(pTopic->vgs);
+  for (j = 0; j < numOfVgroups; j++) {
+    SMqClientVg* pVg = taosArrayGet(pTopic->vgs, j);
+    if (pVg->vgId == vgId) {
+      break;
     }
   }
 
-HANDLE_RSP:
-  if (pParamSet->totalRspNum == 0) {
-    tsem_destroy(&pParamSet->rspSem);
+  if (j == numOfVgroups) {
+    tscWarn("consumer:0x%" PRIx64 " failed to find the specified vgId:%d, total Vgs:%d, topic:%s", tmq->consumerId, vgId,
+            numOfVgroups, pTopicName);
     taosMemoryFree(pParamSet);
-    taosThreadMutexUnlock(&tmq->lock);
-    return 0;
+    pCommitFp(tmq, TSDB_CODE_SUCCESS, userParam);
+    return;
   }
 
-  if (!async) {
-    taosThreadMutexUnlock(&tmq->lock);
-    tsem_wait(&pParamSet->rspSem);
-    code = pParamSet->rspErr;
-    tsem_destroy(&pParamSet->rspSem);
-    taosMemoryFree(pParamSet);
-    return code;
-  } else {
-    code = 0;
-  }
+  SMqClientVg* pVg = taosArrayGet(pTopic->vgs, j);
+  if (pVg->currentOffset.type > 0 && !tOffsetEqual(&pVg->currentOffset, &pVg->committedOffset)) {
+    code = doSendCommitMsg(tmq, pVg, pTopic->topicName, pParamSet, j, numOfVgroups);
 
-FAIL:
-  taosThreadMutexUnlock(&tmq->lock);
-  if (code != 0 && async) {
-    userCb(tmq, code, userParam);
+    // failed to commit, callback user function directly.
+    if (code != TSDB_CODE_SUCCESS) {
+      taosMemoryFree(pParamSet);
+      pCommitFp(tmq, code, userParam);
+    }
+  } else { // do not perform commit, callback user function directly.
+    taosMemoryFree(pParamSet);
+    pCommitFp(tmq, code, userParam);
   }
-
-  return 0;
 }
 
-static int32_t doAutoCommit(tmq_t* tmq, int8_t automatic, int8_t async, tmq_commit_cb* userCb, void* userParam) {
-  int32_t code = -1;
-
+static void asyncCommitAllOffsets(tmq_t* tmq, tmq_commit_cb* pCommitFp, void* userParam) {
   SMqCommitCbParamSet* pParamSet = taosMemoryCalloc(1, sizeof(SMqCommitCbParamSet));
   if (pParamSet == NULL) {
-    code = TSDB_CODE_OUT_OF_MEMORY;
-    if (async) {
-      if (automatic) {
-        tmq->commitCb(tmq, code, tmq->commitCbUserParam);
-      } else {
-        userCb(tmq, code, userParam);
-      }
-    }
-    return -1;
+    pCommitFp(tmq, TSDB_CODE_OUT_OF_MEMORY, userParam);
+    return;
   }
 
   pParamSet->refId = tmq->refId;
   pParamSet->epoch = tmq->epoch;
-
-  pParamSet->automatic = automatic;
-  pParamSet->async = async;
-  pParamSet->userCb = userCb;
+  pParamSet->callbackFn = pCommitFp;
   pParamSet->userParam = userParam;
-  tsem_init(&pParamSet->rspSem, 0, 0);
 
   // init as 1 to prevent concurrency issue
   pParamSet->waitingRspNum = 1;
 
-  taosThreadMutexLock(&tmq->lock);
   int32_t numOfTopics = taosArrayGetSize(tmq->clientTopics);
   tscDebug("consumer:0x%" PRIx64 " start to commit offset for %d topics", tmq->consumerId, numOfTopics);
 
@@ -683,7 +675,7 @@ static int32_t doAutoCommit(tmq_t* tmq, int8_t automatic, int8_t async, tmq_comm
       SMqClientVg* pVg = taosArrayGet(pTopic->vgs, j);
 
       if (pVg->currentOffset.type > 0 && !tOffsetEqual(&pVg->currentOffset, &pVg->committedOffset)) {
-        code = doSendCommitMsg(tmq, pVg, pTopic->topicName, pParamSet, j, numOfVgroups);
+        int32_t code = doSendCommitMsg(tmq, pVg, pTopic->topicName, pParamSet, j, numOfVgroups);
         if (code != TSDB_CODE_SUCCESS) {
           tscError("consumer:0x%" PRIx64 " topic:%s vgId:%d offset:%" PRId64 " failed, code:%s ordinal:%d/%d",
                    tmq->consumerId, pTopic->topicName, pVg->vgId, pVg->committedOffset.version, tstrerror(terrno),
@@ -694,7 +686,7 @@ static int32_t doAutoCommit(tmq_t* tmq, int8_t automatic, int8_t async, tmq_comm
         // update the offset value.
         pVg->committedOffset = pVg->currentOffset;
       } else {
-        tscDebug("consumer:0x%" PRIx64 " topic:%s vgId:%d, no commit, current:%" PRId64 ", ordinal:%d/%d",
+        tscDebug("consumer:0x%" PRIx64 " topic:%s vgId:%d, not commit, current:%" PRId64 ", ordinal:%d/%d",
                  tmq->consumerId, pTopic->topicName, pVg->vgId, pVg->currentOffset.version, j + 1, numOfVgroups);
       }
     }
@@ -702,39 +694,16 @@ static int32_t doAutoCommit(tmq_t* tmq, int8_t automatic, int8_t async, tmq_comm
 
   tscDebug("consumer:0x%" PRIx64 " total commit:%d for %d topics", tmq->consumerId, pParamSet->waitingRspNum - 1,
            numOfTopics);
-  taosThreadMutexUnlock(&tmq->lock);
 
   // no request is sent
   if (pParamSet->totalRspNum == 0) {
-    tsem_destroy(&pParamSet->rspSem);
     taosMemoryFree(pParamSet);
-    return 0;
+    pCommitFp(tmq, TSDB_CODE_SUCCESS, userParam);
+    return;
   }
 
   // count down since waiting rsp num init as 1
-  tmqCommitRspCountDown(pParamSet, tmq->consumerId, "", 0);
-
-  if (!async) {
-    tsem_wait(&pParamSet->rspSem);
-    code = pParamSet->rspErr;
-    tsem_destroy(&pParamSet->rspSem);
-    taosMemoryFree(pParamSet);
-#if 0
-    taosArrayDestroyP(pParamSet->successfulOffsets, taosMemoryFree);
-    taosArrayDestroyP(pParamSet->failedOffsets, taosMemoryFree);
-#endif
-  }
-
-  return code;
-}
-
-static int32_t tmqCommitInner(tmq_t* tmq, const TAOS_RES* msg, int8_t automatic, int8_t async, tmq_commit_cb* userCb,
-                              void* userParam) {
-  if (msg) { // user invoked commit
-    return tmqCommitMsgImpl(tmq, msg, async, userCb, userParam);
-  } else {  // this for auto commit
-    return doAutoCommit(tmq, automatic, async, userCb, userParam);
-  }
+  commitRspCountDown(pParamSet, tmq->consumerId, "", 0);
 }
 
 static void generateTimedTask(int64_t refId, int32_t type) {
@@ -841,6 +810,12 @@ OVER:
   taosReleaseRef(tmqMgmt.rsetId, refId);
 }
 
+static void defaultCommitCbFn(tmq_t* pTmq, int32_t code, void* param) {
+  if (code != 0) {
+    tscDebug("consumer:0x%"PRIx64", failed to commit offset, code:%s", pTmq->consumerId, tstrerror(code));
+  }
+}
+
 int32_t tmqHandleAllDelayedTask(tmq_t* pTmq) {
   STaosQall* qall = taosAllocateQall();
   taosReadAllQitems(pTmq->delayedTask, qall);
@@ -856,7 +831,7 @@ int32_t tmqHandleAllDelayedTask(tmq_t* pTmq) {
 
   while (pTaskType != NULL) {
     if (*pTaskType == TMQ_DELAYED_TASK__ASK_EP) {
-      tmqAskEp(pTmq, true);
+      asyncAskEp(pTmq, addToQueueCallbackFn, NULL);
 
       int64_t* pRefId = taosMemoryMalloc(sizeof(int64_t));
       *pRefId = pTmq->refId;
@@ -864,12 +839,13 @@ int32_t tmqHandleAllDelayedTask(tmq_t* pTmq) {
       tscDebug("consumer:0x%" PRIx64 " retrieve ep from mnode in 1s", pTmq->consumerId);
       taosTmrReset(tmqAssignAskEpTask, 1000, pRefId, tmqMgmt.timer, &pTmq->epTimer);
     } else if (*pTaskType == TMQ_DELAYED_TASK__COMMIT) {
-      tmqCommitInner(pTmq, NULL, 1, 1, pTmq->commitCb, pTmq->commitCbUserParam);
+      tmq_commit_cb* pCallbackFn = pTmq->commitCb? pTmq->commitCb:defaultCommitCbFn;
 
+      asyncCommitAllOffsets(pTmq, pCallbackFn, pTmq->commitCbUserParam);
       int64_t* pRefId = taosMemoryMalloc(sizeof(int64_t));
       *pRefId = pTmq->refId;
 
-      tscDebug("consumer:0x%" PRIx64 " commit to vnode(s) in %.2fs", pTmq->consumerId,
+      tscDebug("consumer:0x%" PRIx64 " next commit to vnode(s) in %.2fs", pTmq->consumerId,
                pTmq->autoCommitInterval / 1000.0);
       taosTmrReset(tmqAssignDelayedCommitTask, pTmq->autoCommitInterval, pRefId, tmqMgmt.timer, &pTmq->commitTimer);
     } else if (*pTaskType == TMQ_DELAYED_TASK__REPORT) {
@@ -1061,8 +1037,6 @@ tmq_t* tmq_consumer_new(tmq_conf_t* conf, char* errstr, int32_t errstrLen) {
   pTmq->status = TMQ_CONSUMER_STATUS__INIT;
   pTmq->pollCnt = 0;
   pTmq->epoch = 0;
-  /*pTmq->epStatus = 0;*/
-  /*pTmq->epSkipCnt = 0;*/
 
   // set conf
   strcpy(pTmq->clientId, conf->clientId);
@@ -1214,7 +1188,7 @@ int32_t tmq_subscribe(tmq_t* tmq, const tmq_list_t* topic_list) {
   }
 
   int32_t retryCnt = 0;
-  while (TSDB_CODE_MND_CONSUMER_NOT_READY == tmqAskEp(tmq, false)) {
+  while (TSDB_CODE_MND_CONSUMER_NOT_READY == doAskEp(tmq)) {
     if (retryCnt++ > MAX_RETRY_COUNT) {
       goto FAIL;
     }
@@ -1348,7 +1322,7 @@ int32_t tmqPollCb(void* param, SDataBuf* pMsg, int32_t code) {
 
     char buf[80];
     tFormatOffset(buf, 80, &pRspWrapper->dataRsp.rspOffset);
-    tscDebug("consumer:0x%" PRIx64 " recv poll rsp, vgId:%d, req:%" PRId64 ", rsp:%s type %d, reqId:0x%" PRIx64,
+    tscDebug("consumer:0x%" PRIx64 " recv poll rsp, vgId:%d, req ver:%" PRId64 ", rsp:%s type %d, reqId:0x%" PRIx64,
              tmq->consumerId, vgId, pRspWrapper->dataRsp.reqOffset.version, buf, rspType, requestId);
   } else if (rspType == TMQ_MSG_TYPE__POLL_META_RSP) {
     SDecoder decoder;
@@ -1514,28 +1488,30 @@ static bool tmqUpdateEp(tmq_t* tmq, int32_t epoch, const SMqAskEpRsp* pRsp) {
   return set;
 }
 
-static int32_t tmqAskEpCb(void* param, SDataBuf* pMsg, int32_t code) {
+int32_t askEpCallbackFn(void* param, SDataBuf* pMsg, int32_t code) {
   SMqAskEpCbParam* pParam = (SMqAskEpCbParam*)param;
-  int8_t           async = pParam->async;
-  tmq_t*           tmq = taosAcquireRef(tmqMgmt.rsetId, pParam->refId);
+  tmq_t* tmq = taosAcquireRef(tmqMgmt.rsetId, pParam->refId);
 
   if (tmq == NULL) {
-    if (!async) {
-      tsem_destroy(&pParam->rspSem);
-    } else {
-      taosMemoryFree(pParam);
-    }
+    terrno = TSDB_CODE_TMQ_CONSUMER_CLOSED;
+    pParam->pUserFn(tmq, terrno, NULL, pParam->pParam);
+
     taosMemoryFree(pMsg->pData);
     taosMemoryFree(pMsg->pEpSet);
-    terrno = TSDB_CODE_TMQ_CONSUMER_CLOSED;
-    return -1;
+    taosMemoryFree(pParam);
+    return terrno;
   }
 
-  pParam->code = code;
   if (code != TSDB_CODE_SUCCESS) {
-    tscError("consumer:0x%" PRIx64 ", get topic endpoint error, async:%d, code:%s", tmq->consumerId, pParam->async,
-             tstrerror(code));
-    goto END;
+    tscError("consumer:0x%" PRIx64 ", get topic endpoint error, code:%s", tmq->consumerId, tstrerror(code));
+    pParam->pUserFn(tmq, code, NULL, pParam->pParam);
+
+    taosReleaseRef(tmqMgmt.rsetId, pParam->refId);
+
+    taosMemoryFree(pMsg->pData);
+    taosMemoryFree(pMsg->pEpSet);
+    taosMemoryFree(pParam);
+    return code;
   }
 
   // tmq's epoch is monotonically increase,
@@ -1546,6 +1522,7 @@ static int32_t tmqAskEpCb(void* param, SDataBuf* pMsg, int32_t code) {
   if (head->epoch <= epoch) {
     tscDebug("consumer:0x%" PRIx64 ", recv ep, msg epoch %d, current epoch %d, no need to update local ep",
              tmq->consumerId, head->epoch, epoch);
+
     if (tmq->status == TMQ_CONSUMER_STATUS__RECOVER) {
       SMqAskEpRsp rsp;
       tDecodeSMqAskEpRsp(POINTER_SHIFT(pMsg->pData, sizeof(SMqRspHead)), &rsp);
@@ -1554,45 +1531,17 @@ static int32_t tmqAskEpCb(void* param, SDataBuf* pMsg, int32_t code) {
       tDeleteSMqAskEpRsp(&rsp);
     }
 
-    goto END;
-  }
-
-  tscDebug("consumer:0x%" PRIx64 ", recv ep, msg epoch %d, current epoch %d, update local ep", tmq->consumerId,
-           head->epoch, epoch);
-
-  if (!async) {
-    SMqAskEpRsp rsp;
-    tDecodeSMqAskEpRsp(POINTER_SHIFT(pMsg->pData, sizeof(SMqRspHead)), &rsp);
-    tmqUpdateEp(tmq, head->epoch, &rsp);
-    tDeleteSMqAskEpRsp(&rsp);
   } else {
-    SMqAskEpRspWrapper* pWrapper = taosAllocateQitem(sizeof(SMqAskEpRspWrapper), DEF_QITEM, 0);
-    if (pWrapper == NULL) {
-      terrno = TSDB_CODE_OUT_OF_MEMORY;
-      code = -1;
-      goto END;
-    }
-
-    pWrapper->tmqRspType = TMQ_MSG_TYPE__EP_RSP;
-    pWrapper->epoch = head->epoch;
-    memcpy(&pWrapper->msg, pMsg->pData, sizeof(SMqRspHead));
-    tDecodeSMqAskEpRsp(POINTER_SHIFT(pMsg->pData, sizeof(SMqRspHead)), &pWrapper->msg);
-
-    taosWriteQitem(tmq->mqueue, pWrapper);
-    tsem_post(&tmq->rspSem);
+    tscDebug("consumer:0x%" PRIx64 ", recv ep, msg epoch %d, current epoch %d, update local ep", tmq->consumerId,
+             head->epoch, epoch);
+    pParam->pUserFn(tmq, code, pMsg, pParam->pParam);
   }
 
-END:
   taosReleaseRef(tmqMgmt.rsetId, pParam->refId);
 
-  if (!async) {
-    tsem_post(&pParam->rspSem);
-  } else {
-    taosMemoryFree(pParam);
-  }
-
   taosMemoryFree(pMsg->pEpSet);
   taosMemoryFree(pMsg->pData);
+  taosMemoryFree(pParam);
   return code;
 }
 
@@ -1981,7 +1930,7 @@ TAOS_RES* tmq_consumer_poll(tmq_t* tmq, int64_t timeout) {
 
   while (atomic_load_8(&tmq->status) == TMQ_CONSUMER_STATUS__RECOVER) {
     int32_t retryCnt = 0;
-    while (TSDB_CODE_MND_CONSUMER_NOT_READY == tmqAskEp(tmq, false)) {
+    while (TSDB_CODE_MND_CONSUMER_NOT_READY == doAskEp(tmq)) {
       if (retryCnt++ > 40) {
         return NULL;
       }
@@ -2163,96 +2112,162 @@ const char* tmq_get_table_name(TAOS_RES* res) {
   return NULL;
 }
 
-void tmq_commit_async(tmq_t* tmq, const TAOS_RES* msg, tmq_commit_cb* cb, void* param) {
-  tmqCommitInner(tmq, msg, 0, 1, cb, param);
+void tmq_commit_async(tmq_t* tmq, const TAOS_RES* pRes, tmq_commit_cb* cb, void* param) {
+  if (pRes == NULL) {  // here needs to commit all offsets.
+    asyncCommitAllOffsets(tmq, cb, param);
+  } else {  // only commit one offset
+    asyncCommitOffset(tmq, pRes, cb, param);
+  }
 }
 
-int32_t tmq_commit_sync(tmq_t* tmq, const TAOS_RES* msg) {
-  return tmqCommitInner(tmq, msg, 0, 0, NULL, NULL);
+typedef struct SSyncCommitInfo {
+  tsem_t  sem;
+  int32_t code;
+} SSyncCommitInfo;
+
+static void commitCallBackFn(tmq_t *pTmq, int32_t code, void* param) {
+  SSyncCommitInfo* pInfo = (SSyncCommitInfo*) param;
+  pInfo->code = code;
+  tsem_post(&pInfo->sem);
 }
 
-int32_t tmqAskEp(tmq_t* tmq, bool async) {
-  int32_t code = TSDB_CODE_SUCCESS;
-#if 0
-  int8_t  epStatus = atomic_val_compare_exchange_8(&tmq->epStatus, 0, 1);
-  if (epStatus == 1) {
-    int32_t epSkipCnt = atomic_add_fetch_32(&tmq->epSkipCnt, 1);
-    tscTrace("consumer:0x%" PRIx64 ", skip ask ep cnt %d", tmq->consumerId, epSkipCnt);
-    if (epSkipCnt < 5000) return 0;
+int32_t tmq_commit_sync(tmq_t* tmq, const TAOS_RES* pRes) {
+  int32_t code = 0;
+
+  SSyncCommitInfo* pInfo = taosMemoryMalloc(sizeof(SSyncCommitInfo));
+  tsem_init(&pInfo->sem, 0, 0);
+  pInfo->code = 0;
+
+  if (pRes == NULL) {
+    asyncCommitAllOffsets(tmq, commitCallBackFn, pInfo);
+  } else {
+    asyncCommitOffset(tmq, pRes, commitCallBackFn, pInfo);
   }
-  atomic_store_32(&tmq->epSkipCnt, 0);
-#endif
 
+  tsem_wait(&pInfo->sem);
+  code = pInfo->code;
+
+  tsem_destroy(&pInfo->sem);
+  taosMemoryFree(pInfo);
+
+  tscDebug("consumer:0x%"PRIx64" sync commit done, code:%s", tmq->consumerId, tstrerror(code));
+  return code;
+}
+
+void updateEpCallbackFn(tmq_t* pTmq, int32_t code, SDataBuf* pDataBuf, void* param) {
+  SAskEpInfo* pInfo = param;
+  pInfo->code = code;
+
+  if (code == TSDB_CODE_SUCCESS) {
+    SMqRspHead* head = pDataBuf->pData;
+
+    SMqAskEpRsp rsp;
+    tDecodeSMqAskEpRsp(POINTER_SHIFT(pDataBuf->pData, sizeof(SMqRspHead)), &rsp);
+    tmqUpdateEp(pTmq, head->epoch, &rsp);
+    tDeleteSMqAskEpRsp(&rsp);
+  }
+
+  tsem_post(&pInfo->sem);
+}
+
+void addToQueueCallbackFn(tmq_t* pTmq, int32_t code, SDataBuf* pDataBuf, void* param) {
+  if (code != TSDB_CODE_SUCCESS) {
+    terrno = code;
+    return;
+  }
+
+  SMqAskEpRspWrapper* pWrapper = taosAllocateQitem(sizeof(SMqAskEpRspWrapper), DEF_QITEM, 0);
+  if (pWrapper == NULL) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    return;
+  }
+
+  SMqRspHead* head = pDataBuf->pData;
+
+  pWrapper->tmqRspType = TMQ_MSG_TYPE__EP_RSP;
+  pWrapper->epoch = head->epoch;
+  memcpy(&pWrapper->msg, pDataBuf->pData, sizeof(SMqRspHead));
+  tDecodeSMqAskEpRsp(POINTER_SHIFT(pDataBuf->pData, sizeof(SMqRspHead)), &pWrapper->msg);
+
+  taosWriteQitem(pTmq->mqueue, pWrapper);
+}
+
+int32_t doAskEp(tmq_t* pTmq) {
+  SAskEpInfo* pInfo = taosMemoryMalloc(sizeof(SAskEpInfo));
+  tsem_init(&pInfo->sem, 0, 0);
+
+  asyncAskEp(pTmq, updateEpCallbackFn, pInfo);
+  tsem_wait(&pInfo->sem);
+
+  int32_t code = pInfo->code;
+  tsem_destroy(&pInfo->sem);
+  taosMemoryFree(pInfo);
+  return code;
+}
+
+void asyncAskEp(tmq_t* pTmq, __tmq_askep_fn_t askEpFn, void* param) {
   SMqAskEpReq req = {0};
-  req.consumerId = tmq->consumerId;
-  req.epoch = tmq->epoch;
-  strcpy(req.cgroup, tmq->groupId);
+  req.consumerId = pTmq->consumerId;
+  req.epoch = pTmq->epoch;
+  strcpy(req.cgroup, pTmq->groupId);
 
   int32_t tlen = tSerializeSMqAskEpReq(NULL, 0, &req);
   if (tlen < 0) {
-    tscError("consumer:0x%" PRIx64 ", tSerializeSMqAskEpReq failed", tmq->consumerId);
-    return -1;
+    tscError("consumer:0x%" PRIx64 ", tSerializeSMqAskEpReq failed", pTmq->consumerId);
+    askEpFn(pTmq, TSDB_CODE_INVALID_PARA, NULL, param);
+    return;
   }
 
   void* pReq = taosMemoryCalloc(1, tlen);
   if (pReq == NULL) {
-    tscError("consumer:0x%" PRIx64 ", failed to malloc askEpReq msg, size:%d", tmq->consumerId, tlen);
-    terrno = TSDB_CODE_OUT_OF_MEMORY;
-    return -1;
+    tscError("consumer:0x%" PRIx64 ", failed to malloc askEpReq msg, size:%d", pTmq->consumerId, tlen);
+    askEpFn(pTmq, TSDB_CODE_OUT_OF_MEMORY, NULL, param);
+    return;
   }
 
   if (tSerializeSMqAskEpReq(pReq, tlen, &req) < 0) {
-    tscError("consumer:0x%" PRIx64 ", tSerializeSMqAskEpReq %d failed", tmq->consumerId, tlen);
+    tscError("consumer:0x%" PRIx64 ", tSerializeSMqAskEpReq %d failed", pTmq->consumerId, tlen);
     taosMemoryFree(pReq);
-    return -1;
+
+    askEpFn(pTmq, TSDB_CODE_INVALID_PARA, NULL, param);
+    return;
   }
 
   SMqAskEpCbParam* pParam = taosMemoryCalloc(1, sizeof(SMqAskEpCbParam));
   if (pParam == NULL) {
-    tscError("consumer:0x%" PRIx64 ", failed to malloc subscribe param", tmq->consumerId);
+    tscError("consumer:0x%" PRIx64 ", failed to malloc subscribe param", pTmq->consumerId);
     taosMemoryFree(pReq);
-    return -1;
+
+    askEpFn(pTmq, TSDB_CODE_OUT_OF_MEMORY, NULL, param);
+    return;
   }
 
-  pParam->refId = tmq->refId;
-  pParam->epoch = tmq->epoch;
-  pParam->async = async;
-  tsem_init(&pParam->rspSem, 0, 0);
+  pParam->refId = pTmq->refId;
+  pParam->epoch = pTmq->epoch;
+  pParam->pUserFn = askEpFn;
+  pParam->pParam = param;
 
   SMsgSendInfo* sendInfo = taosMemoryCalloc(1, sizeof(SMsgSendInfo));
   if (sendInfo == NULL) {
-    tsem_destroy(&pParam->rspSem);
     taosMemoryFree(pParam);
     taosMemoryFree(pReq);
-    return -1;
+    askEpFn(pTmq, TSDB_CODE_OUT_OF_MEMORY, NULL, param);
+    return;
   }
 
-  sendInfo->msgInfo = (SDataBuf){
-      .pData = pReq,
-      .len = tlen,
-      .handle = NULL,
-  };
+  sendInfo->msgInfo = (SDataBuf){ .pData = pReq, .len = tlen, .handle = NULL };
 
   sendInfo->requestId = generateRequestId();
   sendInfo->requestObjRefId = 0;
   sendInfo->param = pParam;
-  sendInfo->fp = tmqAskEpCb;
+  sendInfo->fp = askEpCallbackFn;
   sendInfo->msgType = TDMT_MND_TMQ_ASK_EP;
 
-  SEpSet epSet = getEpSet_s(&tmq->pTscObj->pAppInfo->mgmtEp);
-  tscDebug("consumer:0x%" PRIx64 " ask ep from mnode, async:%d, reqId:0x%" PRIx64, tmq->consumerId, async,
-           sendInfo->requestId);
+  SEpSet epSet = getEpSet_s(&pTmq->pTscObj->pAppInfo->mgmtEp);
+  tscDebug("consumer:0x%" PRIx64 " ask ep from mnode, reqId:0x%" PRIx64, pTmq->consumerId, sendInfo->requestId);
 
   int64_t transporterId = 0;
-  asyncSendMsgToServer(tmq->pTscObj->pAppInfo->pTransporter, &epSet, &transporterId, sendInfo);
-
-  if (!async) {
-    tsem_wait(&pParam->rspSem);
-    code = pParam->code;
-    taosMemoryFree(pParam);
-  }
-
-  return code;
+  asyncSendMsgToServer(pTmq->pTscObj->pAppInfo->pTransporter, &epSet, &transporterId, sendInfo);
 }
 
 int32_t makeTopicVgroupKey(char* dst, const char* topicName, int32_t vg) {
@@ -2264,38 +2279,20 @@ int32_t tmqCommitDone(SMqCommitCbParamSet* pParamSet) {
 
   tmq_t* tmq = taosAcquireRef(tmqMgmt.rsetId, refId);
   if (tmq == NULL) {
-    if (!pParamSet->async) {
-      tsem_destroy(&pParamSet->rspSem);
-    }
     taosMemoryFree(pParamSet);
     terrno = TSDB_CODE_TMQ_CONSUMER_CLOSED;
     return -1;
   }
 
   // if no more waiting rsp
-  if (pParamSet->async) {
-    // call async cb func
-    if (pParamSet->automatic && tmq->commitCb) {
-      tmq->commitCb(tmq, pParamSet->rspErr, tmq->commitCbUserParam);
-    } else if (!pParamSet->automatic && pParamSet->userCb) { // sem post
-      pParamSet->userCb(tmq, pParamSet->rspErr, pParamSet->userParam);
-    }
-
-    taosMemoryFree(pParamSet);
-  } else {
-    tsem_post(&pParamSet->rspSem);
-  }
-
-#if 0
-  taosArrayDestroyP(pParamSet->successfulOffsets, taosMemoryFree);
-    taosArrayDestroyP(pParamSet->failedOffsets, taosMemoryFree);
-#endif
+  pParamSet->callbackFn(tmq, pParamSet->code, pParamSet->userParam);
+  taosMemoryFree(pParamSet);
 
   taosReleaseRef(tmqMgmt.rsetId, refId);
   return 0;
 }
 
-void tmqCommitRspCountDown(SMqCommitCbParamSet* pParamSet, int64_t consumerId, const char* pTopic, int32_t vgId) {
+void commitRspCountDown(SMqCommitCbParamSet* pParamSet, int64_t consumerId, const char* pTopic, int32_t vgId) {
   int32_t waitingRspNum = atomic_sub_fetch_32(&pParamSet->waitingRspNum, 1);
   if (waitingRspNum == 0) {
     tscDebug("consumer:0x%" PRIx64 " topic:%s vgId:%d all commit-rsp received, commit completed", consumerId, pTopic,
diff --git a/source/dnode/mnode/impl/src/mndTrans.c b/source/dnode/mnode/impl/src/mndTrans.c
index 39b4252618f33f4cc114eab0116e5617953ac35a..9d8a662b6b05f96a4fe687a1146a60d44368f69c 100644
--- a/source/dnode/mnode/impl/src/mndTrans.c
+++ b/source/dnode/mnode/impl/src/mndTrans.c
@@ -1091,10 +1091,13 @@ int32_t mndTransProcessRsp(SRpcMsg *pRsp) {
     pAction->msgReceived = 1;
     pAction->errCode = pRsp->code;
     pTrans->lastErrorNo = pRsp->code;
+
+    mInfo("trans:%d, %s:%d response is received, code:0x%x, accept:0x%x retry:0x%x", transId,
+          mndTransStr(pAction->stage), action, pRsp->code, pAction->acceptableCode, pAction->retryCode);
+  } else {
+    mInfo("trans:%d, invalid action, index:%d, code:0x%x", transId, action, pRsp->code);
   }
 
-  mInfo("trans:%d, %s:%d response is received, code:0x%x, accept:0x%x retry:0x%x", transId, mndTransStr(pAction->stage),
-        action, pRsp->code, pAction->acceptableCode, pAction->retryCode);
   mndTransExecute(pMnode, pTrans, true);
 
 _OVER:
diff --git a/source/libs/executor/src/executor.c b/source/libs/executor/src/executor.c
index 572acb309240d55efe268a1a47450e58fd73f8ea..fdb2a3da060f0039a1ee466d101e22c924ce2f3c 100644
--- a/source/libs/executor/src/executor.c
+++ b/source/libs/executor/src/executor.c
@@ -755,8 +755,8 @@ int32_t qKillTask(qTaskInfo_t tinfo, int32_t rspCode) {
     return TSDB_CODE_QRY_INVALID_QHANDLE;
   }
 
-  qDebug("%s execTask async killed", GET_TASKID(pTaskInfo));
-  setTaskKilled(pTaskInfo, rspCode);
+  qDebug("%s sync killed execTask", GET_TASKID(pTaskInfo));
+  setTaskKilled(pTaskInfo, TSDB_CODE_TSC_QUERY_KILLED);
 
   while(qTaskIsExecuting(pTaskInfo)) {
     taosMsleep(10);
diff --git a/source/libs/executor/src/scanoperator.c b/source/libs/executor/src/scanoperator.c
index d2b3a8c3c5e8de3f956a90ced462cf7dd5b149b7..89ca7510e9e07286175bf7a982a1f7aa5e320433 100644
--- a/source/libs/executor/src/scanoperator.c
+++ b/source/libs/executor/src/scanoperator.c
@@ -771,24 +771,32 @@ static SSDataBlock* doTableScan(SOperatorInfo* pOperator) {
 
   // scan table one by one sequentially
   if (pInfo->scanMode == TABLE_SCAN__TABLE_ORDER) {
-    int32_t numOfTables = tableListGetSize(pTaskInfo->pTableInfoList);
+    int32_t numOfTables = 0;//tableListGetSize(pTaskInfo->pTableInfoList);
+    STableKeyInfo tInfo = {0};
 
     while (1) {
       SSDataBlock* result = doGroupedTableScan(pOperator);
-      if (result || (pOperator->status == OP_EXEC_DONE)) {
+      if (result || (pOperator->status == OP_EXEC_DONE) || isTaskKilled(pTaskInfo)) {
         return result;
       }
 
       // if no data, switch to next table and continue scan
       pInfo->currentTable++;
+
+      taosRLockLatch(&pTaskInfo->lock);
+      numOfTables = tableListGetSize(pTaskInfo->pTableInfoList);
+
       if (pInfo->currentTable >= numOfTables) {
         qDebug("all table checked in table list, total:%d, return NULL, %s", numOfTables, GET_TASKID(pTaskInfo));
+        taosRUnLockLatch(&pTaskInfo->lock);
         return NULL;
       }
 
-      STableKeyInfo* pTableInfo = tableListGetInfo(pTaskInfo->pTableInfoList, pInfo->currentTable);
-      tsdbSetTableList(pInfo->base.dataReader, pTableInfo, 1);
-      qDebug("set uid:%" PRIu64 " into scanner, total tables:%d, index:%d/%d %s", pTableInfo->uid, numOfTables,
+      tInfo = *(STableKeyInfo*) tableListGetInfo(pTaskInfo->pTableInfoList, pInfo->currentTable);
+      taosRUnLockLatch(&pTaskInfo->lock);
+
+      tsdbSetTableList(pInfo->base.dataReader, &tInfo, 1);
+      qDebug("set uid:%" PRIu64 " into scanner, total tables:%d, index:%d/%d %s", tInfo.uid, numOfTables,
              pInfo->currentTable, numOfTables, GET_TASKID(pTaskInfo));
 
       tsdbReaderReset(pInfo->base.dataReader, &pInfo->base.cond);
diff --git a/source/libs/parser/src/parCalcConst.c b/source/libs/parser/src/parCalcConst.c
index a7f6d7cb3c89a480eaaa8bfe2b88a08c3ccd0c2b..c25d0e7036aef4d979069976e6f0e221c2b84615 100644
--- a/source/libs/parser/src/parCalcConst.c
+++ b/source/libs/parser/src/parCalcConst.c
@@ -337,8 +337,14 @@ static SNodeList* getChildProjection(SNode* pStmt) {
 static void eraseSetOpChildProjection(SSetOperator* pSetOp, int32_t index) {
   SNodeList* pLeftProjs = getChildProjection(pSetOp->pLeft);
   nodesListErase(pLeftProjs, nodesListGetCell(pLeftProjs, index));
+  if (QUERY_NODE_SET_OPERATOR == nodeType(pSetOp->pLeft)) {
+    eraseSetOpChildProjection((SSetOperator*)pSetOp->pLeft, index);
+  }
   SNodeList* pRightProjs = getChildProjection(pSetOp->pRight);
   nodesListErase(pRightProjs, nodesListGetCell(pRightProjs, index));
+  if (QUERY_NODE_SET_OPERATOR == nodeType(pSetOp->pRight)) {
+    eraseSetOpChildProjection((SSetOperator*)pSetOp->pRight, index);
+  }
 }
 
 typedef struct SNotRefByOrderByCxt {
diff --git a/source/libs/parser/src/parInsertSml.c b/source/libs/parser/src/parInsertSml.c
index 106ee641af9b2c9688654cd1100b80a22c40612c..0bcd777d2a4bd332a146e6cceb1bb464cdbbe573 100644
--- a/source/libs/parser/src/parInsertSml.c
+++ b/source/libs/parser/src/parInsertSml.c
@@ -25,6 +25,8 @@ static void clearColValArray(SArray* pCols) {
     if (TSDB_DATA_TYPE_NCHAR == pCol->type) {
       taosMemoryFreeClear(pCol->value.pData);
     }
+    pCol->flag = CV_FLAG_NONE;
+    pCol->value.val = 0;
   }
 }
 
diff --git a/source/libs/qcom/src/queryUtil.c b/source/libs/qcom/src/queryUtil.c
index c68a08682c57886b407dcb9f6ab7c6245ddf61a2..9d8c17000345a7aa73761c1e3f9af5ae4a4a3363 100644
--- a/source/libs/qcom/src/queryUtil.c
+++ b/source/libs/qcom/src/queryUtil.c
@@ -194,6 +194,8 @@ char* jobTaskStatusStr(int32_t status) {
       return "EXECUTING";
     case JOB_TASK_STATUS_PART_SUCC:
       return "PARTIAL_SUCCEED";
+    case JOB_TASK_STATUS_FETCH:
+      return "FETCHING";
     case JOB_TASK_STATUS_SUCC:
       return "SUCCEED";
     case JOB_TASK_STATUS_FAIL:
diff --git a/source/libs/qworker/src/qwDbg.c b/source/libs/qworker/src/qwDbg.c
index b8d5d2e6eee3d76f96b458afd6a4043a4e5e685b..59e63e9eae2636ae1676ac612083e3857e3a7510 100644
--- a/source/libs/qworker/src/qwDbg.c
+++ b/source/libs/qworker/src/qwDbg.c
@@ -259,15 +259,26 @@ void qwDbgSimulateDead(QW_FPARAMS_DEF, SQWTaskCtx *ctx, bool *rsped) {
   static int32_t ignoreTime = 0;
 
   if (++ignoreTime > 10 && 0 == taosRand() % 9) {
+    if (ctx->msgType == TDMT_SCH_FETCH) {
+      qwBuildAndSendErrorRsp(TDMT_SCH_LINK_BROKEN, &ctx->ctrlConnInfo, TSDB_CODE_RPC_BROKEN_LINK);
+      qwBuildAndSendErrorRsp(ctx->msgType + 1, &ctx->dataConnInfo, TSDB_CODE_QRY_TASK_CTX_NOT_EXIST);
+      *rsped = true;
+      
+      taosSsleep(3);
+      return;
+    }
+
+#if 0
     SRpcHandleInfo *pConn =
         ((ctx->msgType == TDMT_SCH_FETCH || ctx->msgType == TDMT_SCH_MERGE_FETCH) ? &ctx->dataConnInfo
-                                                                                  : &ctx->ctrlConnInfo);
+                                                                                  : &ctx->ctrlConnInfo);                                                                              
     qwBuildAndSendErrorRsp(ctx->msgType + 1, pConn, TSDB_CODE_RPC_BROKEN_LINK);
-
+    
     qwBuildAndSendDropMsg(QW_FPARAMS(), pConn);
     *rsped = true;
-
+    
     return;
+#endif    
   }
 }
 
diff --git a/source/libs/scheduler/inc/schInt.h b/source/libs/scheduler/inc/schInt.h
index 85b952937fbc356879adc1b53a059767b9164e5f..d002b5dfa96f182d343d85390f32263fcd4912a8 100644
--- a/source/libs/scheduler/inc/schInt.h
+++ b/source/libs/scheduler/inc/schInt.h
@@ -193,7 +193,7 @@ typedef struct SSchLevel {
   int32_t  taskSucceed;
   int32_t  taskNum;
   int32_t  taskLaunchedNum;
-  int32_t  taskDoneNum;
+  int32_t  taskExecDoneNum;
   SArray  *subTasks;  // Element is SSchTask
 } SSchLevel;
 
@@ -299,6 +299,7 @@ typedef struct SSchJob {
   SExecResult          execRes;
   void                *fetchRes;  // TODO free it or not
   bool                 fetched;
+  bool                 noMoreRetry;
   int64_t              resNumOfRows; // from int32_t to int64_t
   SSchResInfo          userRes;
   char                *sql;
@@ -333,13 +334,16 @@ extern SSchedulerMgmt schMgmt;
   ((_job)->attr.localExec && SCH_IS_QUERY_JOB(_job) && (!SCH_IS_INSERT_JOB(_job)) && \
    (!SCH_IS_DATA_BIND_QRY_TASK(_task)))
 
-#define SCH_UPDATE_REDICT_CODE(job, _code) atomic_val_compare_exchange_32(&((job)->redirectCode), 0, _code)
-#define SCH_GET_REDICT_CODE(job, _code) (((!NO_RET_REDIRECT_ERROR(_code)) || (job)->redirectCode == 0) ? (_code) : (job)->redirectCode)
+#define SCH_UPDATE_REDIRECT_CODE(job, _code) atomic_val_compare_exchange_32(&((job)->redirectCode), 0, _code)
+#define SCH_GET_REDIRECT_CODE(job, _code) (((!NO_RET_REDIRECT_ERROR(_code)) || (job)->redirectCode == 0) ? (_code) : (job)->redirectCode)
 
 #define SCH_SET_TASK_STATUS(task, st) atomic_store_8(&(task)->status, st)
 #define SCH_GET_TASK_STATUS(task)     atomic_load_8(&(task)->status)
 #define SCH_GET_TASK_STATUS_STR(task) jobTaskStatusStr(SCH_GET_TASK_STATUS(task))
 
+#define SCH_TASK_ALREADY_LAUNCHED(task) (SCH_GET_TASK_STATUS(task) >= JOB_TASK_STATUS_EXEC)
+#define SCH_TASK_EXEC_DONE(task) (SCH_GET_TASK_STATUS(task) >= JOB_TASK_STATUS_PART_SUCC)
+
 #define SCH_GET_TASK_HANDLE(_task)          ((_task) ? (_task)->handle : NULL)
 #define SCH_SET_TASK_HANDLE(_task, _handle) ((_task)->handle = (_handle))
 
@@ -361,6 +365,7 @@ extern SSchedulerMgmt schMgmt;
   (SCH_IS_DATA_BIND_QRY_TASK(_task) && SCH_JOB_NEED_FLOW_CTRL(_job) && SCH_IS_LEVEL_UNFINISHED((_task)->level))
 #define SCH_FETCH_TYPE(_pSrcTask)  (SCH_IS_DATA_BIND_QRY_TASK(_pSrcTask) ? TDMT_SCH_FETCH : TDMT_SCH_MERGE_FETCH)
 #define SCH_TASK_NEED_FETCH(_task) ((_task)->plan->subplanType != SUBPLAN_TYPE_MODIFY)
+#define SCH_MULTI_LEVEL_LAUNCHED(_job) ((_job)->levelIdx != ((_job)->levelNum - 1))
 
 #define SCH_SET_JOB_TYPE(_job, type)     \
   do {                                   \
@@ -377,16 +382,24 @@ extern SSchedulerMgmt schMgmt;
 #define SCH_JOB_NEED_DROP(_job)  (SCH_IS_QUERY_JOB(_job))
 #define SCH_IS_EXPLAIN_JOB(_job) (EXPLAIN_MODE_ANALYZE == (_job)->attr.explainMode)
 #define SCH_NETWORK_ERR(_code)   ((_code) == TSDB_CODE_RPC_BROKEN_LINK || (_code) == TSDB_CODE_RPC_NETWORK_UNAVAIL || (_code) == TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED)
-#define SCH_MERGE_TASK_NETWORK_ERR(_task, _code, _len) \
-  (SCH_NETWORK_ERR(_code) && (((_len) > 0) || (!SCH_IS_DATA_BIND_TASK(_task)) || (_task)->redirectCtx.inRedirect))
 #define SCH_REDIRECT_MSGTYPE(_msgType)                                                                         \
   ((_msgType) == TDMT_SCH_LINK_BROKEN || (_msgType) == TDMT_SCH_QUERY || (_msgType) == TDMT_SCH_MERGE_QUERY || \
    (_msgType) == TDMT_SCH_FETCH || (_msgType) == TDMT_SCH_MERGE_FETCH)
-#define SCH_TASK_NEED_REDIRECT(_task, _msgType, _code, _rspLen) \
-  (SCH_REDIRECT_MSGTYPE(_msgType) &&                            \
-   (NEED_SCHEDULER_REDIRECT_ERROR(_code) || SCH_MERGE_TASK_NETWORK_ERR((_task), (_code), (_rspLen))))
-#define SCH_NEED_RETRY(_msgType, _code) \
-  ((SCH_NETWORK_ERR(_code) && SCH_REDIRECT_MSGTYPE(_msgType)) || (_code) == TSDB_CODE_SCH_TIMEOUT_ERROR)
+#define SCH_LOW_LEVEL_NETWORK_ERR(_job, _task, _code) \
+    (SCH_NETWORK_ERR(_code) && ((_task)->level->level == (_job)->levelIdx))
+#define SCH_TOP_LEVEL_NETWORK_ERR(_job, _task, _code) \
+    (SCH_NETWORK_ERR(_code) && ((_task)->level->level > (_job)->levelIdx))
+#define SCH_TASK_RETRY_NETWORK_ERR(_task, _code) \
+    (SCH_NETWORK_ERR(_code) && (_task)->redirectCtx.inRedirect)
+
+#define SCH_JOB_NEED_RETRY(_job, _task, _msgType, _code)      \
+   (SCH_REDIRECT_MSGTYPE(_msgType) && SCH_TOP_LEVEL_NETWORK_ERR(_job, _task, _code))
+#define SCH_TASKSET_NEED_RETRY(_job, _task, _msgType, _code) \
+   (SCH_REDIRECT_MSGTYPE(_msgType) &&                         \
+   (NEED_SCHEDULER_REDIRECT_ERROR(_code) || SCH_LOW_LEVEL_NETWORK_ERR((_job), (_task), (_code)) || SCH_TASK_RETRY_NETWORK_ERR((_task), (_code))))
+#define SCH_TASK_NEED_RETRY(_msgType, _code) \
+   ((SCH_REDIRECT_MSGTYPE(_msgType) && SCH_NETWORK_ERR(_code)) || (_code) == TSDB_CODE_SCH_TIMEOUT_ERROR)
+
 
 #define SCH_IS_LEVEL_UNFINISHED(_level) ((_level)->taskLaunchedNum < (_level)->taskNum)
 #define SCH_GET_CUR_EP(_addr)           (&(_addr)->epSet.eps[(_addr)->epSet.inUse])
@@ -510,6 +523,11 @@ extern SSchedulerMgmt schMgmt;
     }                                                                                                 \
   } while (0)
 
+#define SCH_RESET_JOB_LEVEL_IDX(_job) do {                        \
+  (_job)->levelIdx = (_job)->levelNum - 1;                       \
+  SCH_JOB_DLOG("set job levelIdx to %d", (_job)->levelIdx);       \
+} while (0)
+
 void     schDeregisterTaskHb(SSchJob *pJob, SSchTask *pTask);
 void     schCleanClusterHb(void *pTrans);
 int32_t  schLaunchTask(SSchJob *job, SSchTask *task);
@@ -562,7 +580,7 @@ int32_t  schInitJob(int64_t *pJobId, SSchedulerReq *pReq);
 int32_t  schExecJob(SSchJob *pJob, SSchedulerReq *pReq);
 int32_t  schDumpJobExecRes(SSchJob *pJob, SExecResult *pRes);
 int32_t  schUpdateTaskCandidateAddr(SSchJob *pJob, SSchTask *pTask, SEpSet *pEpSet);
-int32_t  schHandleRedirect(SSchJob *pJob, SSchTask *pTask, SDataBuf *pData, int32_t rspCode);
+int32_t  schHandleTaskSetRetry(SSchJob *pJob, SSchTask *pTask, SDataBuf *pData, int32_t rspCode);
 void     schProcessOnOpEnd(SSchJob *pJob, SCH_OP_TYPE type, SSchedulerReq *pReq, int32_t errCode);
 int32_t  schProcessOnOpBegin(SSchJob *pJob, SCH_OP_TYPE type, SSchedulerReq *pReq);
 void     schProcessOnCbEnd(SSchJob *pJob, SSchTask *pTask, int32_t errCode);
@@ -591,6 +609,10 @@ int32_t  schHandleJobDrop(SSchJob *pJob, int32_t errCode);
 bool     schChkCurrentOp(SSchJob *pJob, int32_t op, int8_t sync);
 int32_t  schProcessFetchRsp(SSchJob *pJob, SSchTask *pTask, char *msg, int32_t rspCode);
 int32_t  schProcessExplainRsp(SSchJob *pJob, SSchTask *pTask, SExplainRsp *rsp);
+int32_t  schHandleJobRetry(SSchJob *pJob, SSchTask *pTask, SDataBuf *pMsg, int32_t rspCode);
+int32_t  schChkResetJobRetry(SSchJob *pJob, int32_t rspCode);
+void     schResetTaskForRetry(SSchJob *pJob, SSchTask *pTask);
+int32_t  schChkUpdateRedirectCtx(SSchJob *pJob, SSchTask *pTask, SEpSet *pEpSet, int32_t rspCode);
 
 extern SSchDebug gSCHDebug;
 
diff --git a/source/libs/scheduler/src/schFlowCtrl.c b/source/libs/scheduler/src/schFlowCtrl.c
index 5e4fe4b8a1f2795cfbd4dff838304fe0e36343ff..9cb95a6bbe7aad0bf9a6576dac13db3114d1559d 100644
--- a/source/libs/scheduler/src/schFlowCtrl.c
+++ b/source/libs/scheduler/src/schFlowCtrl.c
@@ -282,7 +282,6 @@ int32_t schLaunchTasksInFlowCtrlList(SSchJob *pJob, SSchTask *pTask) {
   }
 
   int32_t code = schLaunchTasksInFlowCtrlListImpl(pJob, ctrl);
-  ;
   SCH_ERR_RET(code);
 
   return code;  // to avoid compiler error
diff --git a/source/libs/scheduler/src/schJob.c b/source/libs/scheduler/src/schJob.c
index 980a8ac6a193e32c38e64f0b323c5311ddacef55..e7bfe95795f44f61ee4d1cc02873a21b80438369 100644
--- a/source/libs/scheduler/src/schJob.c
+++ b/source/libs/scheduler/src/schJob.c
@@ -83,6 +83,10 @@ int32_t schUpdateJobStatus(SSchJob *pJob, int8_t newStatus) {
     oriStatus = SCH_GET_JOB_STATUS(pJob);
 
     if (oriStatus == newStatus) {
+      if (JOB_TASK_STATUS_FETCH == newStatus) {
+        return code;
+      }
+      
       SCH_ERR_JRET(TSDB_CODE_SCH_IGNORE_ERROR);
     }
 
@@ -108,10 +112,19 @@ int32_t schUpdateJobStatus(SSchJob *pJob, int8_t newStatus) {
         break;
       case JOB_TASK_STATUS_PART_SUCC:
         if (newStatus != JOB_TASK_STATUS_FAIL && newStatus != JOB_TASK_STATUS_SUCC &&
-            newStatus != JOB_TASK_STATUS_DROP && newStatus != JOB_TASK_STATUS_EXEC) {
+            newStatus != JOB_TASK_STATUS_DROP && newStatus != JOB_TASK_STATUS_EXEC &&
+            newStatus != JOB_TASK_STATUS_FETCH) {
           SCH_ERR_JRET(TSDB_CODE_APP_ERROR);
         }
 
+        break;
+      case JOB_TASK_STATUS_FETCH:
+        if (newStatus != JOB_TASK_STATUS_FAIL && newStatus != JOB_TASK_STATUS_SUCC &&
+            newStatus != JOB_TASK_STATUS_DROP && newStatus != JOB_TASK_STATUS_EXEC &&
+            newStatus != JOB_TASK_STATUS_FETCH) {
+          SCH_ERR_JRET(TSDB_CODE_APP_ERROR);
+        }
+      
         break;
       case JOB_TASK_STATUS_SUCC:
       case JOB_TASK_STATUS_FAIL:
@@ -288,7 +301,7 @@ int32_t schValidateAndBuildJob(SQueryPlan *pDag, SSchJob *pJob) {
   }
 
   pJob->levelNum = levelNum;
-  pJob->levelIdx = levelNum - 1;
+  SCH_RESET_JOB_LEVEL_IDX(pJob);
 
   SSchLevel      level = {0};
   SNodeListNode *plans = NULL;
@@ -550,9 +563,9 @@ int32_t schLaunchJobLowerLevel(SSchJob *pJob, SSchTask *pTask) {
   }
 
   SSchLevel *pLevel = pTask->level;
-  int32_t    doneNum = atomic_add_fetch_32(&pLevel->taskDoneNum, 1);
+  int32_t    doneNum = atomic_add_fetch_32(&pLevel->taskExecDoneNum, 1);
   if (doneNum == pLevel->taskNum) {
-    pJob->levelIdx--;
+    atomic_sub_fetch_32(&pJob->levelIdx, 1);
 
     pLevel = taosArrayGet(pJob->levels, pJob->levelIdx);
     for (int32_t i = 0; i < pLevel->taskNum; ++i) {
@@ -562,6 +575,10 @@ int32_t schLaunchJobLowerLevel(SSchJob *pJob, SSchTask *pTask) {
         continue;
       }
 
+      if (SCH_TASK_ALREADY_LAUNCHED(pTask)) {
+        continue;
+      }
+
       SCH_ERR_RET(schLaunchTask(pJob, pTask));
     }
   }
@@ -811,6 +828,75 @@ void schDirectPostJobRes(SSchedulerReq *pReq, int32_t errCode) {
   }
 }
 
+int32_t schChkResetJobRetry(SSchJob *pJob, int32_t rspCode) {
+  if (pJob->status >= JOB_TASK_STATUS_PART_SUCC) {
+    SCH_LOCK(SCH_WRITE, &pJob->resLock);
+    if (pJob->fetched) {
+      SCH_UNLOCK(SCH_WRITE, &pJob->resLock);
+      pJob->noMoreRetry = true;
+      SCH_JOB_ELOG("already fetched while got error %s", tstrerror(rspCode));
+      SCH_ERR_RET(rspCode);
+    }
+    SCH_UNLOCK(SCH_WRITE, &pJob->resLock);
+
+    schUpdateJobStatus(pJob, JOB_TASK_STATUS_EXEC);
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
+int32_t schResetJobForRetry(SSchJob *pJob, int32_t rspCode) {
+  SCH_ERR_RET(schChkResetJobRetry(pJob, rspCode));
+
+  int32_t numOfLevels = taosArrayGetSize(pJob->levels);
+  for (int32_t i = 0; i < numOfLevels; ++i) {
+    SSchLevel *pLevel = taosArrayGet(pJob->levels, i);
+
+    pLevel->taskExecDoneNum = 0;
+    pLevel->taskLaunchedNum = 0;
+
+    int32_t numOfTasks = taosArrayGetSize(pLevel->subTasks);
+    for (int32_t j = 0; j < numOfTasks; ++j) {
+      SSchTask *pTask = taosArrayGet(pLevel->subTasks, j);
+      SCH_LOCK_TASK(pTask);
+      SCH_ERR_RET(schChkUpdateRedirectCtx(pJob, pTask, NULL, rspCode));
+      qClearSubplanExecutionNode(pTask->plan);
+      schResetTaskForRetry(pJob, pTask);
+      SCH_UNLOCK_TASK(pTask);
+    }
+  }
+
+  SCH_RESET_JOB_LEVEL_IDX(pJob);
+
+  return TSDB_CODE_SUCCESS;
+}
+
+
+int32_t schHandleJobRetry(SSchJob *pJob, SSchTask *pTask, SDataBuf *pMsg, int32_t rspCode) {
+  int32_t code = 0;
+
+  taosMemoryFreeClear(pMsg->pData);
+  taosMemoryFreeClear(pMsg->pEpSet);
+
+  SCH_UNLOCK_TASK(pTask);
+
+  SCH_TASK_DLOG("start to redirect all job tasks cause of error: %s", tstrerror(rspCode));
+
+  SCH_ERR_JRET(schResetJobForRetry(pJob, rspCode));
+
+  SCH_ERR_JRET(schLaunchJob(pJob));
+
+  SCH_LOCK_TASK(pTask);
+
+  SCH_RET(code);
+
+_return:
+
+  SCH_LOCK_TASK(pTask);
+
+  SCH_RET(schProcessOnTaskFailure(pJob, pTask, code));
+}
+
 bool schChkCurrentOp(SSchJob *pJob, int32_t op, int8_t sync) {
   bool r = false;
   SCH_LOCK(SCH_READ, &pJob->opStatus.lock);
@@ -907,7 +993,7 @@ int32_t schProcessOnOpBegin(SSchJob *pJob, SCH_OP_TYPE type, SSchedulerReq *pReq
         SCH_ERR_RET(TSDB_CODE_APP_ERROR);
       }
 
-      if (status != JOB_TASK_STATUS_PART_SUCC) {
+      if (status != JOB_TASK_STATUS_PART_SUCC && status != JOB_TASK_STATUS_FETCH) {
         SCH_JOB_ELOG("job status error for fetch, status:%s", jobTaskStatusStr(status));
         SCH_ERR_RET(TSDB_CODE_SCH_STATUS_ERROR);
       }
diff --git a/source/libs/scheduler/src/schRemote.c b/source/libs/scheduler/src/schRemote.c
index 6f4130fd9ffe2814f0dd30655a58aefc92801a13..80fdc7594cc63b8e893de9e4e7cf193592deb885 100644
--- a/source/libs/scheduler/src/schRemote.c
+++ b/source/libs/scheduler/src/schRemote.c
@@ -36,7 +36,7 @@ int32_t schValidateRspMsgType(SSchJob *pJob, SSchTask *pTask, int32_t msgType) {
                       TMSG_INFO(msgType));
         SCH_ERR_RET(TSDB_CODE_QW_MSG_ERROR);
       }
-      if (taskStatus != JOB_TASK_STATUS_PART_SUCC) {
+      if (taskStatus != JOB_TASK_STATUS_FETCH) {
         SCH_TASK_ELOG("rsp msg conflicted with task status, status:%s, rspType:%s", jobTaskStatusStr(taskStatus),
                       TMSG_INFO(msgType));
         SCH_ERR_RET(TSDB_CODE_QW_MSG_ERROR);
@@ -137,25 +137,12 @@ int32_t schProcessExplainRsp(SSchJob *pJob, SSchTask *pTask, SExplainRsp *rsp) {
   return TSDB_CODE_SUCCESS;
 }
 
-// Note: no more task error processing, handled in function internal
-int32_t schHandleResponseMsg(SSchJob *pJob, SSchTask *pTask, int32_t execId, SDataBuf *pMsg, int32_t rspCode) {
+int32_t schProcessResponseMsg(SSchJob *pJob, SSchTask *pTask, int32_t execId, SDataBuf *pMsg, int32_t rspCode) {
   int32_t code = 0;
   char   *msg = pMsg->pData;
   int32_t msgSize = pMsg->len;
   int32_t msgType = pMsg->msgType;
 
-  bool dropExecNode = (msgType == TDMT_SCH_LINK_BROKEN || SCH_NETWORK_ERR(rspCode));
-  if (SCH_IS_QUERY_JOB(pJob)) {
-    SCH_ERR_JRET(schUpdateTaskHandle(pJob, pTask, dropExecNode, pMsg->handle, execId));
-  }
-  
-  SCH_ERR_JRET(schValidateRspMsgType(pJob, pTask, msgType));
-
-  int32_t reqType = IsReq(pMsg) ? pMsg->msgType : (pMsg->msgType - 1);
-  if (SCH_TASK_NEED_REDIRECT(pTask, reqType, rspCode, pMsg->len)) {
-    SCH_RET(schHandleRedirect(pJob, pTask, (SDataBuf *)pMsg, rspCode));
-  }
-
   pTask->redirectCtx.inRedirect = false;
 
   switch (msgType) {
@@ -423,6 +410,38 @@ _return:
 
   SCH_RET(schProcessOnTaskFailure(pJob, pTask, code));
 } 
+
+
+// Note: no more task error processing, handled in function internal
+int32_t schHandleResponseMsg(SSchJob *pJob, SSchTask *pTask, int32_t execId, SDataBuf *pMsg, int32_t rspCode) {
+  int32_t code = 0;
+  int32_t msgType = pMsg->msgType;
+  char   *msg = pMsg->pData;
+
+  bool dropExecNode = (msgType == TDMT_SCH_LINK_BROKEN || SCH_NETWORK_ERR(rspCode));
+  if (SCH_IS_QUERY_JOB(pJob)) {
+    SCH_ERR_JRET(schUpdateTaskHandle(pJob, pTask, dropExecNode, pMsg->handle, execId));
+  }
+  
+  SCH_ERR_JRET(schValidateRspMsgType(pJob, pTask, msgType));
+
+  int32_t reqType = IsReq(pMsg) ? pMsg->msgType : (pMsg->msgType - 1);
+  if (SCH_JOB_NEED_RETRY(pJob, pTask, reqType, rspCode)) {
+    SCH_RET(schHandleJobRetry(pJob, pTask, (SDataBuf *)pMsg, rspCode));
+  } else if (SCH_TASKSET_NEED_RETRY(pJob, pTask, reqType, rspCode)) {
+    SCH_RET(schHandleTaskSetRetry(pJob, pTask, (SDataBuf *)pMsg, rspCode));
+  }
+
+  pTask->redirectCtx.inRedirect = false;
+
+  SCH_RET(schProcessResponseMsg(pJob, pTask, execId, pMsg, rspCode));
+
+_return:
+
+  taosMemoryFreeClear(msg);
+
+  SCH_RET(schProcessOnTaskFailure(pJob, pTask, code));
+} 
 int32_t schHandleCallback(void *param, SDataBuf *pMsg, int32_t rspCode) {
   int32_t                code = 0;
   SSchTaskCallbackParam *pParam = (SSchTaskCallbackParam *)param;
diff --git a/source/libs/scheduler/src/schStatus.c b/source/libs/scheduler/src/schStatus.c
index 4c16a81a05003166978dd9ee0f96b2c876f6fefd..9d0ad30e2a47de92136ec1f25bffa6f761ea7725 100644
--- a/source/libs/scheduler/src/schStatus.c
+++ b/source/libs/scheduler/src/schStatus.c
@@ -34,6 +34,9 @@ int32_t schSwitchJobStatus(SSchJob* pJob, int32_t status, void* param) {
     case JOB_TASK_STATUS_PART_SUCC:
       SCH_ERR_JRET(schProcessOnJobPartialSuccess(pJob));
       break;
+    case JOB_TASK_STATUS_FETCH:
+      SCH_ERR_JRET(schJobFetchRows(pJob));
+      break;
     case JOB_TASK_STATUS_SUCC:
       break;
     case JOB_TASK_STATUS_FAIL:
diff --git a/source/libs/scheduler/src/schTask.c b/source/libs/scheduler/src/schTask.c
index bdab7393277617afa8dbd183700d54efa2caf4a7..207753ae2571fe292ca009cb6af51030bf0667f0 100644
--- a/source/libs/scheduler/src/schTask.c
+++ b/source/libs/scheduler/src/schTask.c
@@ -378,7 +378,8 @@ int32_t schChkUpdateRedirectCtx(SSchJob *pJob, SSchTask *pTask, SEpSet *pEpSet,
     if (lastTime > tsMaxRetryWaitTime) {
       SCH_TASK_DLOG("task no more redirect retry since timeout, now:%" PRId64 ", start:%" PRId64 ", max:%d, total:%d",
                     nowTs, pCtx->startTs, tsMaxRetryWaitTime, pCtx->totalTimes);
-      SCH_ERR_RET(SCH_GET_REDICT_CODE(pJob, rspCode));
+      pJob->noMoreRetry = true;                    
+      SCH_ERR_RET(SCH_GET_REDIRECT_CODE(pJob, rspCode));
     }
 
     pCtx->periodMs *= tsRedirectFactor;
@@ -404,32 +405,35 @@ _return:
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t schDoTaskRedirect(SSchJob *pJob, SSchTask *pTask, SDataBuf *pData, int32_t rspCode) {
-  int32_t code = 0;
-
-  SCH_TASK_DLOG("task will be redirected now, status:%s, code:%s", SCH_GET_TASK_STATUS_STR(pTask), tstrerror(rspCode));
-
-  if (NULL == pData) {
-    pTask->retryTimes = 0;
-  }
-
-  if (!NO_RET_REDIRECT_ERROR(rspCode)) {
-    SCH_UPDATE_REDICT_CODE(pJob, rspCode);
-  }
-
-  SCH_ERR_JRET(schChkUpdateRedirectCtx(pJob, pTask, pData ? pData->pEpSet : NULL, rspCode));
-
+void schResetTaskForRetry(SSchJob *pJob, SSchTask *pTask) {
   pTask->waitRetry = true;
 
   schDropTaskOnExecNode(pJob, pTask);
+  if (pTask->delayTimer) {
+    taosTmrStopA(&pTask->delayTimer);
+  }
   taosHashClear(pTask->execNodes);
   schRemoveTaskFromExecList(pJob, pTask);
   schDeregisterTaskHb(pJob, pTask);
-  atomic_sub_fetch_32(&pTask->level->taskLaunchedNum, 1);
   taosMemoryFreeClear(pTask->msg);
   pTask->msgLen = 0;
   pTask->lastMsgType = 0;
+  pTask->childReady = 0;      
   memset(&pTask->succeedAddr, 0, sizeof(pTask->succeedAddr));
+}
+
+int32_t schDoTaskRedirect(SSchJob *pJob, SSchTask *pTask, SDataBuf *pData, int32_t rspCode) {
+  int32_t code = 0;
+
+  SCH_TASK_DLOG("task will be redirected now, status:%s, code:%s", SCH_GET_TASK_STATUS_STR(pTask), tstrerror(rspCode));
+
+  if (!NO_RET_REDIRECT_ERROR(rspCode)) {
+    SCH_UPDATE_REDIRECT_CODE(pJob, rspCode);
+  }
+
+  SCH_ERR_JRET(schChkUpdateRedirectCtx(pJob, pTask, pData ? pData->pEpSet : NULL, rspCode));
+
+  schResetTaskForRetry(pJob, pTask);
 
   if (SCH_IS_DATA_BIND_TASK(pTask)) {
     if (pData && pData->pEpSet) {
@@ -445,12 +449,6 @@ int32_t schDoTaskRedirect(SSchJob *pJob, SSchTask *pTask, SDataBuf *pData, int32
       SCH_TASK_DLOG("switch task target node %d epset to %d/%d", addr->nodeId, addr->epSet.inUse, addr->epSet.numOfEps);
     }
 
-    if (SCH_TASK_NEED_FLOW_CTRL(pJob, pTask)) {
-      if (JOB_TASK_STATUS_EXEC == SCH_GET_TASK_STATUS(pTask)) {
-        SCH_ERR_JRET(schLaunchTasksInFlowCtrlList(pJob, pTask));
-      }
-    }
-
     SCH_SET_TASK_STATUS(pTask, JOB_TASK_STATUS_INIT);
 
     SCH_ERR_JRET(schDelayLaunchTask(pJob, pTask));
@@ -486,20 +484,10 @@ _return:
   SCH_RET(schProcessOnTaskFailure(pJob, pTask, code));
 }
 
-int32_t schHandleRedirect(SSchJob *pJob, SSchTask *pTask, SDataBuf *pData, int32_t rspCode) {
+int32_t schHandleTaskSetRetry(SSchJob *pJob, SSchTask *pTask, SDataBuf *pData, int32_t rspCode) {
   int32_t code = 0;
 
-  if (JOB_TASK_STATUS_PART_SUCC == pJob->status) {
-    SCH_LOCK(SCH_WRITE, &pJob->resLock);
-    if (pJob->fetched) {
-      SCH_UNLOCK(SCH_WRITE, &pJob->resLock);
-      SCH_TASK_ELOG("already fetched while got error %s", tstrerror(rspCode));
-      SCH_ERR_JRET(rspCode);
-    }
-    SCH_UNLOCK(SCH_WRITE, &pJob->resLock);
-
-    schUpdateJobStatus(pJob, JOB_TASK_STATUS_EXEC);
-  }
+  SCH_ERR_JRET(schChkResetJobRetry(pJob, rspCode));
 
   if (SYNC_OTHER_LEADER_REDIRECT_ERROR(rspCode)) {
     if (NULL == pData->pEpSet) {
@@ -509,7 +497,19 @@ int32_t schHandleRedirect(SSchJob *pJob, SSchTask *pTask, SDataBuf *pData, int32
     }
   }
 
+  SCH_TASK_DLOG("start to redirect current task set cause of error: %s", tstrerror(rspCode));
+
+  for (int32_t i = 0; i < pJob->levelNum; ++i) {
+    SSchLevel *pLevel = taosArrayGet(pJob->levels, i);
+
+    pLevel->taskExecDoneNum = 0;
+    pLevel->taskLaunchedNum = 0;
+  }
+  
+  SCH_RESET_JOB_LEVEL_IDX(pJob);
+  
   code = schDoTaskRedirect(pJob, pTask, pData, rspCode);
+  
   taosMemoryFreeClear(pData->pData);
   taosMemoryFreeClear(pData->pEpSet);
 
@@ -621,6 +621,13 @@ int32_t schMoveTaskToExecList(SSchJob *pJob, SSchTask *pTask, bool *moved) {
 */
 
 int32_t schTaskCheckSetRetry(SSchJob *pJob, SSchTask *pTask, int32_t errCode, bool *needRetry) {
+  if (pJob->noMoreRetry) {
+    *needRetry = false;
+    SCH_TASK_DLOG("task no more retry since job no more retry, retryTimes:%d/%d", pTask->retryTimes,
+                  pTask->maxRetryTimes);
+    return TSDB_CODE_SUCCESS;
+  }
+  
   if (TSDB_CODE_SCH_TIMEOUT_ERROR == errCode) {
     pTask->maxExecTimes++;
     pTask->maxRetryTimes++;
@@ -645,7 +652,7 @@ int32_t schTaskCheckSetRetry(SSchJob *pJob, SSchTask *pTask, int32_t errCode, bo
     return TSDB_CODE_SUCCESS;
   }
 
-  if (!SCH_NEED_RETRY(pTask->lastMsgType, errCode)) {
+  if (!SCH_TASK_NEED_RETRY(pTask->lastMsgType, errCode)) {
     *needRetry = false;
     SCH_TASK_DLOG("task no more retry cause of errCode, errCode:%x - %s", errCode, tstrerror(errCode));
     return TSDB_CODE_SUCCESS;
@@ -1067,7 +1074,6 @@ int32_t schLaunchTaskImpl(void *param) {
     SCH_ERR_JRET(TSDB_CODE_SCH_IGNORE_ERROR);
   }
 
-  // NOTE: race condition: the task should be put into the hash table before send msg to server
   if (SCH_GET_TASK_STATUS(pTask) != JOB_TASK_STATUS_EXEC) {
     SCH_ERR_JRET(schPushTaskToExecList(pJob, pTask));
     SCH_SET_TASK_STATUS(pTask, JOB_TASK_STATUS_EXEC);
@@ -1272,6 +1278,8 @@ int32_t schLaunchFetchTask(SSchJob *pJob) {
     return TSDB_CODE_SUCCESS;
   }
 
+  SCH_SET_TASK_STATUS(pJob->fetchTask, JOB_TASK_STATUS_FETCH);
+
   if (SCH_IS_LOCAL_EXEC_TASK(pJob, pJob->fetchTask)) {
     SCH_ERR_JRET(schExecLocalFetch(pJob, pJob->fetchTask));
   } else {
diff --git a/source/libs/scheduler/src/scheduler.c b/source/libs/scheduler/src/scheduler.c
index 7cd5e957b6b8d47871274f80632a65ec11485753..2b46a4710e24215e9b32009f4fb97c8204963fe6 100644
--- a/source/libs/scheduler/src/scheduler.c
+++ b/source/libs/scheduler/src/scheduler.c
@@ -91,7 +91,7 @@ int32_t schedulerFetchRows(int64_t jobId, SSchedulerReq *pReq) {
 
   SCH_ERR_JRET(schHandleOpBeginEvent(jobId, &pJob, SCH_OP_FETCH, pReq));
 
-  SCH_ERR_JRET(schJobFetchRows(pJob));
+  SCH_ERR_JRET(schSwitchJobStatus(pJob, JOB_TASK_STATUS_FETCH, pReq));
 
 _return:
 
diff --git a/utils/test/c/sml_test.c b/utils/test/c/sml_test.c
index 44fbd060bee9aaec14d6467630815837cc32b6e9..b0cc6f749cb703c601235a3aaf444ce169149057 100644
--- a/utils/test/c/sml_test.c
+++ b/utils/test/c/sml_test.c
@@ -939,6 +939,9 @@ int sml_ts2164_Test() {
       //      "meters,location=la,groupid=ca current=11.8,voltage=221,phase=0.27",
       "meters,location=la,groupid=ca current=11.8,voltage=221",
       "meters,location=la,groupid=ca current=11.8,voltage=221,phase=0.27",
+      "ts3038,location=l2a,groupid=ca current=L\"11.8\"",
+      "ts3038,location=l2a,groupid=ca voltage=L\"221\"",
+      "ts3038,location=l2a,groupid=ca phase=L\"221\"",
       //      "meters,location=la,groupid=cb current=11.8,voltage=221,phase=0.27",
   };