Merge branch main to FIX/TD-21043-main

cmake/taosadapter_CMakeLists.txt.in

@@ -2,7 +2,7 @@
 # taosadapter
 ExternalProject_Add(taosadapter
         GIT_REPOSITORY https://github.com/taosdata/taosadapter.git
-        GIT_TAG f0c1753
+        GIT_TAG 5662a6d
         SOURCE_DIR "${TD_SOURCE_DIR}/tools/taosadapter"
         BINARY_DIR ""
         #BUILD_IN_SOURCE TRUE

include/libs/qcom/query.h

@@ -268,7 +268,7 @@ extern int32_t (*queryProcessMsgRsp[TDMT_MAX])(void* output, char* msg, int32_t
   ((_code) == TSDB_CODE_SYN_NOT_LEADER || (_code) == TSDB_CODE_SYN_RESTORING || (_code) == TSDB_CODE_SYN_INTERNAL_ERROR)
 #define SYNC_OTHER_LEADER_REDIRECT_ERROR(_code) ((_code) == TSDB_CODE_MNODE_NOT_FOUND)
 
-#define NO_RET_REDIRECT_ERROR(_code) ((_code) == TSDB_CODE_RPC_BROKEN_LINK || (_code) == TSDB_CODE_RPC_NETWORK_UNAVAIL)
+#define NO_RET_REDIRECT_ERROR(_code) ((_code) == TSDB_CODE_RPC_BROKEN_LINK || (_code) == TSDB_CODE_RPC_NETWORK_UNAVAIL || (_code) == TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED)
 
 #define NEED_REDIRECT_ERROR(_code)                                              \
   (NO_RET_REDIRECT_ERROR(_code) || SYNC_UNKNOWN_LEADER_REDIRECT_ERROR(_code) || \

include/util/taoserror.h

@@ -65,6 +65,8 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_RPC_PORT_EADDRINUSE           TAOS_DEF_ERROR_CODE(0, 0x0017) //
 #define TSDB_CODE_RPC_BROKEN_LINK               TAOS_DEF_ERROR_CODE(0, 0x0018) //
 #define TSDB_CODE_RPC_TIMEOUT                   TAOS_DEF_ERROR_CODE(0, 0x0019) //
+#define TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED    TAOS_DEF_ERROR_CODE(0, 0x0020) // "Vgroup could not be connected"
+#define TSDB_CODE_RPC_SOMENODE_BROKEN_LINK      TAOS_DEF_ERROR_CODE(0, 0x0021) // 
 
 //common & util
 #define TSDB_CODE_OPS_NOT_SUPPORT               TAOS_DEF_ERROR_CODE(0, 0x0100) //
@@ -518,6 +520,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_SYN_BATCH_ERROR               TAOS_DEF_ERROR_CODE(0, 0x0913)
 #define TSDB_CODE_SYN_RESTORING                 TAOS_DEF_ERROR_CODE(0, 0x0914)
 #define TSDB_CODE_SYN_INVALID_SNAPSHOT_MSG      TAOS_DEF_ERROR_CODE(0, 0x0915)  // internal
+#define TSDB_CODE_SYN_BUFFER_FULL               TAOS_DEF_ERROR_CODE(0, 0x0916)  //
 #define TSDB_CODE_SYN_INTERNAL_ERROR            TAOS_DEF_ERROR_CODE(0, 0x09FF)
 
 // tq

packaging/tools/makepkg.sh

@@ -348,7 +348,8 @@ cd ${release_dir}
 #  install_dir has been distinguishes  cluster from  edege, so comments this code
 pkg_name=${install_dir}-${osType}-${cpuType}
 
-taostools_pkg_name=${taostools_install_dir}-${osType}-${cpuType}
+versionCompFirst=$(echo ${versionComp} | awk -F '.' '{print $1}')
+taostools_pkg_name=${taostools_install_dir}-${osType}-${cpuType}-comp${versionCompFirst}
 
 # if [ "$verMode" == "cluster" ]; then
 #   pkg_name=${install_dir}-${osType}-${cpuType}

source/client/src/clientImpl.c

@@ -1430,6 +1430,21 @@ void processMsgFromServer(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet) {
     memcpy((void*)tEpSet, (void*)pEpSet, sizeof(SEpSet));
   }
 
+  // pMsg is response msg
+  if (pMsg->msgType == TDMT_MND_CONNECT + 1) {
+    // restore origin code
+    if (pMsg->code == TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED) {
+      pMsg->code = TSDB_CODE_RPC_NETWORK_UNAVAIL;
+    } else if (pMsg->code == TSDB_CODE_RPC_SOMENODE_BROKEN_LINK) {
+      pMsg->code = TSDB_CODE_RPC_BROKEN_LINK;
+    }
+  } else {
+    // uniform to one error code: TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED
+    if (pMsg->code == TSDB_CODE_RPC_SOMENODE_BROKEN_LINK) {
+      pMsg->code = TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED;
+    }
+  }
+
   AsyncArg* arg = taosMemoryCalloc(1, sizeof(AsyncArg));
   arg->msg = *pMsg;
   arg->pEpset = tEpSet;

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

@@ -233,6 +233,14 @@ int32_t vmPutMsgToMgmtQueue(SVnodeMgmt *pMgmt, SRpcMsg *pMsg) {
 }
 
 int32_t vmPutRpcMsgToQueue(SVnodeMgmt *pMgmt, EQueueType qtype, SRpcMsg *pRpc) {
+  if (pRpc->contLen < sizeof(SMsgHead)) {
+    dError("invalid rpc msg since no msg head at pCont. pRpc:%p, type:%s, len:%d", pRpc, TMSG_INFO(pRpc->msgType),
+           pRpc->contLen);
+    rpcFreeCont(pRpc->pCont);
+    pRpc->pCont = NULL;
+    return -1;
+  }
+
   SRpcMsg *pMsg = taosAllocateQitem(sizeof(SRpcMsg), RPC_QITEM, pRpc->contLen);
   if (pMsg == NULL) {
     rpcFreeCont(pRpc->pCont);

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

@@ -248,6 +248,7 @@ static inline void dmReleaseHandle(SRpcHandleInfo *pHandle, int8_t type) { rpcRe
 
 static bool rpcRfp(int32_t code, tmsg_t msgType) {
   if (code == TSDB_CODE_RPC_NETWORK_UNAVAIL || code == TSDB_CODE_RPC_BROKEN_LINK || code == TSDB_CODE_MNODE_NOT_FOUND ||
+      code == TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED ||
       code == TSDB_CODE_SYN_NOT_LEADER || code == TSDB_CODE_SYN_RESTORING || code == TSDB_CODE_VND_STOPPED ||
       code == TSDB_CODE_APP_IS_STARTING || code == TSDB_CODE_APP_IS_STOPPING) {
     if (msgType == TDMT_SCH_QUERY || msgType == TDMT_SCH_MERGE_QUERY || msgType == TDMT_SCH_FETCH ||

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

@@ -957,7 +957,7 @@ static void mndTransSendRpcRsp(SMnode *pMnode, STrans *pTrans) {
   for (int32_t i = 0; i < size; ++i) {
     SRpcHandleInfo *pInfo = taosArrayGet(pTrans->pRpcArray, i);
     if (pInfo->handle != NULL) {
-      if (code == TSDB_CODE_RPC_NETWORK_UNAVAIL) {
+      if (code == TSDB_CODE_RPC_NETWORK_UNAVAIL || code == TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED) {
         code = TSDB_CODE_MND_TRANS_NETWORK_UNAVAILL;
       }
       if (code == TSDB_CODE_SYN_TIMEOUT) {

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

@@ -113,8 +113,8 @@ int32_t metaSnapRead(SMetaSnapReader* pReader, uint8_t** ppData) {
   pHdr->size = nData;
   memcpy(pHdr->data, pData, nData);
 
-  metaInfo("vgId:%d, vnode snapshot meta read data, version:%" PRId64 " uid:%" PRId64 " nData:%d",
-           TD_VID(pReader->pMeta->pVnode), key.version, key.uid, nData);
+  metaDebug("vgId:%d, vnode snapshot meta read data, version:%" PRId64 " uid:%" PRId64 " blockLen:%d",
+            TD_VID(pReader->pMeta->pVnode), key.version, key.uid, nData);
 
 _exit:
   return code;

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

@@ -257,8 +257,8 @@ _exit:
     pReader->index++;
     *nData = sizeof(SSnapDataHdr) + pHdr->size;
     pHdr->index = pReader->index;
-    vInfo("vgId:%d, vnode snapshot read data,index:%" PRId64 " type:%d nData:%d ", TD_VID(pReader->pVnode),
-          pReader->index, pHdr->type, *nData);
+    vDebug("vgId:%d, vnode snapshot read data, index:%" PRId64 " type:%d blockLen:%d ", TD_VID(pReader->pVnode),
+           pReader->index, pHdr->type, *nData);
   } else {
     vInfo("vgId:%d, vnode snapshot read data end, index:%" PRId64, TD_VID(pReader->pVnode), pReader->index);
   }

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

@@ -425,9 +425,9 @@ static int32_t vnodeSyncApplyMsg(const SSyncFSM *pFsm, SRpcMsg *pMsg, const SFsm
 
   const STraceId *trace = &pMsg->info.traceId;
   vGTrace("vgId:%d, commit-cb is excuted, fsm:%p, index:%" PRId64 ", term:%" PRIu64 ", msg-index:%" PRId64
-          ", weak:%d, code:%d, state:%d %s, type:%s",
+          ", weak:%d, code:%d, state:%d %s, type:%s code:0x%x",
           pVnode->config.vgId, pFsm, pMeta->index, pMeta->term, pMsg->info.conn.applyIndex, pMeta->isWeak, pMeta->code,
-          pMeta->state, syncStr(pMeta->state), TMSG_INFO(pMsg->msgType));
+          pMeta->state, syncStr(pMeta->state), TMSG_INFO(pMsg->msgType), pMsg->code);
 
   return tmsgPutToQueue(&pVnode->msgCb, APPLY_QUEUE, pMsg);
 }

source/libs/catalog/src/ctgAsync.c

@@ -471,17 +471,20 @@ int32_t ctgHandleForceUpdate(SCatalog* pCtg, int32_t taskNum, SCtgJob* pJob, con
 }
 
 int32_t ctgInitTask(SCtgJob* pJob, CTG_TASK_TYPE type, void* param, int32_t* taskId) {
+  int32_t code = 0;
   int32_t tid = atomic_fetch_add_32(&pJob->taskIdx, 1);
 
   CTG_LOCK(CTG_WRITE, &pJob->taskLock);
-  CTG_ERR_RET((*gCtgAsyncFps[type].initFp)(pJob, tid, param));
-  CTG_UNLOCK(CTG_WRITE, &pJob->taskLock);
+  CTG_ERR_JRET((*gCtgAsyncFps[type].initFp)(pJob, tid, param));
 
   if (taskId) {
     *taskId = tid;
   }
 
-  return TSDB_CODE_SUCCESS;
+_return:
+  CTG_UNLOCK(CTG_WRITE, &pJob->taskLock);
+  
+  return code;
 }
 
 int32_t ctgInitJob(SCatalog* pCtg, SRequestConnInfo* pConn, SCtgJob** job, const SCatalogReq* pReq, catalogCallback fp,

source/libs/catalog/src/ctgCache.c

@@ -2500,6 +2500,7 @@ int32_t ctgGetTbMetasFromCache(SCatalog *pCtg, SRequestConnInfo *pConn, SCtgTbMe
 
     CTG_LOCK(CTG_READ, &pCache->metaLock);
     if (NULL == pCache->pMeta) {
+      CTG_UNLOCK(CTG_READ, &pCache->metaLock);
       ctgDebug("tb %s meta not in cache, dbFName:%s", pName->tname, dbFName);
       ctgAddFetch(&ctx->pFetchs, dbIdx, i, fetchIdx, baseResIdx + i, flag);
       taosArraySetSize(ctx->pResList, taosArrayGetSize(ctx->pResList) + 1);

source/libs/executor/src/timewindowoperator.c

@@ -907,7 +907,7 @@ static void removeDeleteResults(SHashObj* pUpdatedMap, SArray* pDelWins) {
 }
 
 bool isOverdue(TSKEY ekey, STimeWindowAggSupp* pTwSup) {
-  ASSERT(pTwSup->maxTs == INT64_MIN || pTwSup->maxTs > 0);
+  ASSERTS(pTwSup->maxTs == INT64_MIN || pTwSup->maxTs > 0, "maxts should greater than 0");
   return pTwSup->maxTs != INT64_MIN && ekey < pTwSup->maxTs - pTwSup->waterMark;
 }
 
@@ -1396,7 +1396,6 @@ static int32_t getAllIntervalWindow(SSHashObj* pHashMap, SHashObj* resWins) {
   while ((pIte = tSimpleHashIterate(pHashMap, pIte, &iter)) != NULL) {
     void*    key = tSimpleHashGetKey(pIte, &keyLen);
     uint64_t groupId = *(uint64_t*)key;
-    ASSERT(keyLen == GET_RES_WINDOW_KEY_LEN(sizeof(TSKEY)));
     TSKEY               ts = *(int64_t*)((char*)key + sizeof(uint64_t));
     SResultRowPosition* pPos = (SResultRowPosition*)pIte;
     int32_t             code = saveWinResult(ts, pPos->pageId, pPos->offset, groupId, resWins);
@@ -1547,7 +1546,7 @@ static void closeChildIntervalWindow(SOperatorInfo* pOperator, SArray* pChildren
   for (int32_t i = 0; i < size; i++) {
     SOperatorInfo*               pChildOp = taosArrayGetP(pChildren, i);
     SStreamIntervalOperatorInfo* pChInfo = pChildOp->info;
-    ASSERT(pChInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE);
+    ASSERTS(pChInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE, "children trigger type should be at once");
     pChInfo->twAggSup.maxTs = TMAX(pChInfo->twAggSup.maxTs, maxTs);
     closeStreamIntervalWindow(pChInfo->aggSup.pResultRowHashTable, &pChInfo->twAggSup, &pChInfo->interval, NULL, NULL,
                               NULL, pOperator);
@@ -1767,8 +1766,6 @@ SOperatorInfo* createIntervalOperatorInfo(SOperatorInfo* downstream, SIntervalPh
       .maxTs = INT64_MIN,
   };
 
-  ASSERT(as.calTrigger != STREAM_TRIGGER_MAX_DELAY);
-
   pInfo->win = pTaskInfo->window;
   pInfo->inputOrder = (pPhyNode->window.inputTsOrder == ORDER_ASC) ? TSDB_ORDER_ASC : TSDB_ORDER_DESC;
   pInfo->resultTsOrder = (pPhyNode->window.outputTsOrder == ORDER_ASC) ? TSDB_ORDER_ASC : TSDB_ORDER_DESC;
@@ -2252,7 +2249,6 @@ static void doBuildPullDataBlock(SArray* array, int32_t* pIndex, SSDataBlock* pB
     return;
   }
   blockDataEnsureCapacity(pBlock, size - (*pIndex));
-  ASSERT(3 <= taosArrayGetSize(pBlock->pDataBlock));
   SColumnInfoData* pStartTs = (SColumnInfoData*)taosArrayGet(pBlock->pDataBlock, START_TS_COLUMN_INDEX);
   SColumnInfoData* pEndTs = (SColumnInfoData*)taosArrayGet(pBlock->pDataBlock, END_TS_COLUMN_INDEX);
   SColumnInfoData* pGroupId = (SColumnInfoData*)taosArrayGet(pBlock->pDataBlock, GROUPID_COLUMN_INDEX);
@@ -2359,7 +2355,6 @@ static void doStreamIntervalAggImpl(SOperatorInfo* pOperatorInfo, SSDataBlock* p
   SResultRow*     pResult = NULL;
   int32_t         forwardRows = 0;
 
-  ASSERT(pSDataBlock->pDataBlock != NULL);
   SColumnInfoData* pColDataInfo = taosArrayGet(pSDataBlock->pDataBlock, pInfo->primaryTsIndex);
   tsCols = (int64_t*)pColDataInfo->pData;
 
@@ -2482,7 +2477,6 @@ static SSDataBlock* doStreamFinalIntervalAgg(SOperatorInfo* pOperator) {
     doBuildPullDataBlock(pInfo->pPullWins, &pInfo->pullIndex, pInfo->pPullDataRes);
     if (pInfo->pPullDataRes->info.rows != 0) {
       // process the rest of the data
-      ASSERT(IS_FINAL_OP(pInfo));
       printDataBlock(pInfo->pPullDataRes, IS_FINAL_OP(pInfo) ? "interval final" : "interval semi");
       return pInfo->pPullDataRes;
     }
@@ -2543,7 +2537,6 @@ static SSDataBlock* doStreamFinalIntervalAgg(SOperatorInfo* pOperator) {
     pInfo->numOfDatapack++;
     printDataBlock(pBlock, IS_FINAL_OP(pInfo) ? "interval final recv" : "interval semi recv");
 
-    ASSERT(pBlock->info.type != STREAM_INVERT);
     if (pBlock->info.type == STREAM_NORMAL || pBlock->info.type == STREAM_PULL_DATA) {
       pInfo->binfo.pRes->info.type = pBlock->info.type;
     } else if (pBlock->info.type == STREAM_DELETE_DATA || pBlock->info.type == STREAM_DELETE_RESULT ||
@@ -2633,7 +2626,6 @@ static SSDataBlock* doStreamFinalIntervalAgg(SOperatorInfo* pOperator) {
   doBuildPullDataBlock(pInfo->pPullWins, &pInfo->pullIndex, pInfo->pPullDataRes);
   if (pInfo->pPullDataRes->info.rows != 0) {
     // process the rest of the data
-    ASSERT(IS_FINAL_OP(pInfo));
     printDataBlock(pInfo->pPullDataRes, IS_FINAL_OP(pInfo) ? "interval final" : "interval semi");
     return pInfo->pPullDataRes;
   }
@@ -2688,7 +2680,7 @@ SOperatorInfo* createStreamFinalIntervalOperatorInfo(SOperatorInfo* downstream,
       .deleteMarkSaved = 0,
       .calTriggerSaved = 0,
   };
-  ASSERT(pInfo->twAggSup.calTrigger != STREAM_TRIGGER_MAX_DELAY);
+  ASSERTS(pInfo->twAggSup.calTrigger != STREAM_TRIGGER_MAX_DELAY, "trigger type should not be max delay");
   pInfo->primaryTsIndex = ((SColumnNode*)pIntervalPhyNode->window.pTspk)->slotId;
   size_t keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
   initResultSizeInfo(&pOperator->resultInfo, 4096);
@@ -2713,7 +2705,6 @@ SOperatorInfo* createStreamFinalIntervalOperatorInfo(SOperatorInfo* downstream,
 
   initStreamFunciton(pOperator->exprSupp.pCtx, pOperator->exprSupp.numOfExprs);
 
-  ASSERT(numOfCols > 0);
   initExecTimeWindowInfo(&pInfo->twAggSup.timeWindowData, &pTaskInfo->window);
 
   pInfo->pState = taosMemoryCalloc(1, sizeof(SStreamState));
@@ -2724,6 +2715,9 @@ SOperatorInfo* createStreamFinalIntervalOperatorInfo(SOperatorInfo* downstream,
   pInfo->pChildren = NULL;
   if (numOfChild > 0) {
     pInfo->pChildren = taosArrayInit(numOfChild, sizeof(void*));
+    if (!pInfo->pChildren) {
+      goto _error;
+    }
     for (int32_t i = 0; i < numOfChild; i++) {
       SOperatorInfo* pChildOp = createStreamFinalIntervalOperatorInfo(NULL, pPhyNode, pTaskInfo, 0);
       if (pChildOp) {
@@ -2746,7 +2740,6 @@ SOperatorInfo* createStreamFinalIntervalOperatorInfo(SOperatorInfo* downstream,
     // semi interval operator does not catch result
     pInfo->isFinal = false;
     pOperator->name = "StreamSemiIntervalOperator";
-    ASSERT(pInfo->aggSup.currentPageId == -1);
   }
 
   if (!IS_FINAL_OP(pInfo) || numOfChild == 0) {
@@ -3162,15 +3155,6 @@ static void doStreamSessionAggImpl(SOperatorInfo* pOperator, SSDataBlock* pSData
   }
 }
 
-void deleteWindow(SArray* pWinInfos, int32_t index, FDelete fp) {
-  ASSERT(index >= 0 && index < taosArrayGetSize(pWinInfos));
-  if (fp) {
-    void* ptr = taosArrayGet(pWinInfos, index);
-    fp(ptr);
-  }
-  taosArrayRemove(pWinInfos, index);
-}
-
 static void doDeleteTimeWindows(SStreamAggSupporter* pAggSup, SSDataBlock* pBlock, SArray* result) {
   SColumnInfoData* pStartTsCol = taosArrayGet(pBlock->pDataBlock, START_TS_COLUMN_INDEX);
   TSKEY*           startDatas = (TSKEY*)pStartTsCol->pData;
@@ -3218,7 +3202,6 @@ static int32_t copyUpdateResult(SSHashObj* pStUpdated, SArray* pUpdated) {
   int32_t iter = 0;
   while ((pIte = tSimpleHashIterate(pStUpdated, pIte, &iter)) != NULL) {
     void* key = tSimpleHashGetKey(pIte, &keyLen);
-    ASSERT(keyLen == sizeof(SSessionKey));
     taosArrayPush(pUpdated, key);
   }
   taosArraySort(pUpdated, sessionKeyCompareAsc);
@@ -3279,7 +3262,6 @@ static void rebuildSessionWindow(SOperatorInfo* pOperator, SArray* pWinArray, SS
   SStreamAggSupporter*           pAggSup = &pInfo->streamAggSup;
   int32_t                        numOfOutput = pSup->numOfExprs;
   int32_t                        numOfChildren = taosArrayGetSize(pInfo->pChildren);
-  ASSERT(pInfo->pChildren);
 
   for (int32_t i = 0; i < size; i++) {
     SSessionKey*      pWinKey = taosArrayGet(pWinArray, i);
@@ -3380,7 +3362,6 @@ static void copyDeleteWindowInfo(SArray* pResWins, SSHashObj* pStDeleted) {
 void initGroupResInfoFromArrayList(SGroupResInfo* pGroupResInfo, SArray* pArrayList) {
   pGroupResInfo->pRows = pArrayList;
   pGroupResInfo->index = 0;
-  ASSERT(pGroupResInfo->index <= getNumOfTotalRes(pGroupResInfo));
 }
 
 void doBuildSessionResult(SOperatorInfo* pOperator, SStreamState* pState, SGroupResInfo* pGroupResInfo,
@@ -4811,7 +4792,6 @@ SOperatorInfo* createStreamIntervalOperatorInfo(SOperatorInfo* downstream, SPhys
   int32_t    code = TSDB_CODE_SUCCESS;
   int32_t    numOfCols = 0;
   SExprInfo* pExprInfo = createExprInfo(pIntervalPhyNode->window.pFuncs, NULL, &numOfCols);
-  ASSERT(numOfCols > 0);
 
   SSDataBlock* pResBlock = createDataBlockFromDescNode(pPhyNode->pOutputDataBlockDesc);
   SInterval    interval = {
@@ -4831,7 +4811,7 @@ SOperatorInfo* createStreamIntervalOperatorInfo(SOperatorInfo* downstream, SPhys
       .deleteMark = getDeleteMark(pIntervalPhyNode),
   };
 
-  ASSERT(twAggSupp.calTrigger != STREAM_TRIGGER_MAX_DELAY);
+  ASSERTS(twAggSupp.calTrigger != STREAM_TRIGGER_MAX_DELAY, "trigger type should not be max delay");
 
   pOperator->pTaskInfo = pTaskInfo;
   pInfo->interval = interval;

source/libs/function/src/udfd.c

@@ -606,6 +606,7 @@ int32_t udfdLoadUdf(char *udfName, SUdf *udf) {
 }
 static bool udfdRpcRfp(int32_t code, tmsg_t msgType) {
   if (code == TSDB_CODE_RPC_NETWORK_UNAVAIL || code == TSDB_CODE_RPC_BROKEN_LINK || code == TSDB_CODE_SYN_NOT_LEADER ||
+      code == TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED ||
       code == TSDB_CODE_SYN_RESTORING || code == TSDB_CODE_MNODE_NOT_FOUND || code == TSDB_CODE_APP_IS_STARTING ||
       code == TSDB_CODE_APP_IS_STOPPING) {
     if (msgType == TDMT_SCH_QUERY || msgType == TDMT_SCH_MERGE_QUERY || msgType == TDMT_SCH_FETCH ||

source/libs/scheduler/inc/schInt.h

@@ -375,7 +375,7 @@ extern SSchedulerMgmt schMgmt;
 #define SCH_JOB_NEED_WAIT(_job)  (!SCH_IS_QUERY_JOB(_job))
 #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)
+#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)                                                                         \

source/libs/stream/src/streamExec.c

@@ -48,6 +48,10 @@ static int32_t streamTaskExecImpl(SStreamTask* pTask, const void* data, SArray*
 
   // exec
   while (1) {
+    if (pTask->taskStatus == TASK_STATUS__DROPPING) {
+      return 0;
+    }
+
     SSDataBlock* output = NULL;
     uint64_t     ts = 0;
     if ((code = qExecTask(exec, &output, &ts)) < 0) {

source/libs/sync/inc/syncInt.h

@@ -238,7 +238,7 @@ int32_t syncNodeStopPingTimer(SSyncNode* pSyncNode);
 int32_t syncNodeStartElectTimer(SSyncNode* pSyncNode, int32_t ms);
 int32_t syncNodeStopElectTimer(SSyncNode* pSyncNode);
 int32_t syncNodeRestartElectTimer(SSyncNode* pSyncNode, int32_t ms);
-int32_t syncNodeResetElectTimer(SSyncNode* pSyncNode);
+void    syncNodeResetElectTimer(SSyncNode* pSyncNode);
 int32_t syncNodeStartHeartbeatTimer(SSyncNode* pSyncNode);
 int32_t syncNodeStopHeartbeatTimer(SSyncNode* pSyncNode);
 int32_t syncNodeRestartHeartbeatTimer(SSyncNode* pSyncNode);

source/libs/sync/inc/syncPipeline.h

@@ -61,7 +61,7 @@ typedef struct SSyncLogBuffer {
 // SSyncLogRepMgr
 SSyncLogReplMgr* syncLogReplMgrCreate();
 void             syncLogReplMgrDestroy(SSyncLogReplMgr* pMgr);
-int32_t          syncLogReplMgrReset(SSyncLogReplMgr* pMgr);
+void             syncLogReplMgrReset(SSyncLogReplMgr* pMgr);
 
 int32_t          syncNodeLogReplMgrInit(SSyncNode* pNode);
 void             syncNodeLogReplMgrDestroy(SSyncNode* pNode);
@@ -109,6 +109,8 @@ SSyncRaftEntry* syncLogBufferGetOneEntry(SSyncLogBuffer* pBuf, SSyncNode* pNode,
 int32_t         syncLogBufferValidate(SSyncLogBuffer* pBuf);
 int32_t         syncLogBufferRollback(SSyncLogBuffer* pBuf, SSyncNode* pNode, SyncIndex toIndex);
 
+int32_t syncLogFsmExecute(SSyncNode* pNode, SSyncFSM* pFsm, ESyncState role, SyncTerm term, SSyncRaftEntry* pEntry,
+                          int32_t applyCode);
 #ifdef __cplusplus
 }
 #endif

source/libs/sync/inc/syncSnapshot.h

@@ -56,7 +56,7 @@ SSyncSnapshotSender *snapshotSenderCreate(SSyncNode *pSyncNode, int32_t replicaI
 void                 snapshotSenderDestroy(SSyncSnapshotSender *pSender);
 bool                 snapshotSenderIsStart(SSyncSnapshotSender *pSender);
 int32_t              snapshotSenderStart(SSyncSnapshotSender *pSender);
-int32_t              snapshotSenderStop(SSyncSnapshotSender *pSender, bool finish);
+void                 snapshotSenderStop(SSyncSnapshotSender *pSender, bool finish);
 int32_t              snapshotSend(SSyncSnapshotSender *pSender);
 int32_t              snapshotReSend(SSyncSnapshotSender *pSender);
 
@@ -79,8 +79,8 @@ typedef struct SSyncSnapshotReceiver {
 
 SSyncSnapshotReceiver *snapshotReceiverCreate(SSyncNode *pSyncNode, SRaftId fromId);
 void                   snapshotReceiverDestroy(SSyncSnapshotReceiver *pReceiver);
-int32_t                snapshotReceiverStart(SSyncSnapshotReceiver *pReceiver, SyncSnapshotSend *pBeginMsg);
-int32_t                snapshotReceiverStop(SSyncSnapshotReceiver *pReceiver);
+void                   snapshotReceiverStart(SSyncSnapshotReceiver *pReceiver, SyncSnapshotSend *pBeginMsg);
+void                   snapshotReceiverStop(SSyncSnapshotReceiver *pReceiver);
 bool                   snapshotReceiverIsStart(SSyncSnapshotReceiver *pReceiver);
 void                   snapshotReceiverForceStop(SSyncSnapshotReceiver *pReceiver);
 

source/libs/sync/src/syncMain.c

@@ -200,12 +200,15 @@ int32_t syncProcessMsg(int64_t rid, SRpcMsg* pMsg) {
       code = syncNodeOnLocalCmd(pSyncNode, pMsg);
       break;
     default:
-      sError("vgId:%d, failed to process msg:%p since invalid type:%s", pSyncNode->vgId, pMsg,
-             TMSG_INFO(pMsg->msgType));
+      terrno = TSDB_CODE_MSG_NOT_PROCESSED;
       code = -1;
   }
 
   syncNodeRelease(pSyncNode);
+  if (code != 0) {
+    sDebug("vgId:%d, failed to process sync msg:%p type:%s since 0x%x", pSyncNode->vgId, pMsg, TMSG_INFO(pMsg->msgType),
+           terrno);
+  }
   return code;
 }
 
@@ -228,8 +231,7 @@ int32_t syncSendTimeoutRsp(int64_t rid, int64_t seq) {
 
   syncNodeRelease(pNode);
   if (ret == 1) {
-    sInfo("send timeout response, seq:%" PRId64 " handle:%p ahandle:%p", seq, rpcMsg.info.handle,
-          rpcMsg.info.ahandle);
+    sInfo("send timeout response, seq:%" PRId64 " handle:%p ahandle:%p", seq, rpcMsg.info.handle, rpcMsg.info.ahandle);
     rpcSendResponse(&rpcMsg);
     return 0;
   } else {
@@ -1084,13 +1086,17 @@ SSyncNode* syncNodeOpen(SSyncInfo* pSyncInfo) {
   // snapshot senders
   for (int32_t i = 0; i < TSDB_MAX_REPLICA; ++i) {
     SSyncSnapshotSender* pSender = snapshotSenderCreate(pSyncNode, i);
-    // ASSERT(pSender != NULL);
-    (pSyncNode->senders)[i] = pSender;
-    sSDebug(pSender, "snapshot sender create new while open, data:%p", pSender);
+    if (pSender == NULL) return NULL;
+
+    pSyncNode->senders[i] = pSender;
+    sSDebug(pSender, "snapshot sender create while open sync node, data:%p", pSender);
   }
 
   // snapshot receivers
   pSyncNode->pNewNodeReceiver = snapshotReceiverCreate(pSyncNode, EMPTY_RAFT_ID);
+  if (pSyncNode->pNewNodeReceiver == NULL) return NULL;
+  sRDebug(pSyncNode->pNewNodeReceiver, "snapshot receiver create while open sync node, data:%p",
+          pSyncNode->pNewNodeReceiver);
 
   // is config changing
   pSyncNode->changing = false;
@@ -1131,10 +1137,8 @@ SSyncNode* syncNodeOpen(SSyncInfo* pSyncInfo) {
   pSyncNode->hbrSlowNum = 0;
   pSyncNode->tmrRoutineNum = 0;
 
-  sNInfo(pSyncNode, "sync open, node:%p", pSyncNode);
-  sTrace("vgId:%d, tsElectInterval:%d, tsHeartbeatInterval:%d, tsHeartbeatTimeout:%d", pSyncNode->vgId, tsElectInterval,
-         tsHeartbeatInterval, tsHeartbeatTimeout);
-
+  sNInfo(pSyncNode, "sync open, node:%p electInterval:%d heartbeatInterval:%d heartbeatTimeout:%d", pSyncNode,
+         tsElectInterval, tsHeartbeatInterval, tsHeartbeatTimeout);
   return pSyncNode;
 
 _error:
@@ -1251,6 +1255,8 @@ void syncNodePreClose(SSyncNode* pSyncNode) {
       snapshotReceiverForceStop(pSyncNode->pNewNodeReceiver);
     }
 
+    sDebug("vgId:%d, snapshot receiver destroy while preclose sync node, data:%p", pSyncNode->vgId,
+           pSyncNode->pNewNodeReceiver);
     snapshotReceiverDestroy(pSyncNode->pNewNodeReceiver);
     pSyncNode->pNewNodeReceiver = NULL;
   }
@@ -1295,15 +1301,15 @@ void syncNodeClose(SSyncNode* pSyncNode) {
   syncNodeStopHeartbeatTimer(pSyncNode);
 
   for (int32_t i = 0; i < TSDB_MAX_REPLICA; ++i) {
-    if ((pSyncNode->senders)[i] != NULL) {
-      sSTrace((pSyncNode->senders)[i], "snapshot sender destroy while close, data:%p", (pSyncNode->senders)[i]);
+    if (pSyncNode->senders[i] != NULL) {
+      sDebug("vgId:%d, snapshot sender destroy while close, data:%p", pSyncNode->vgId, pSyncNode->senders[i]);
 
-      if (snapshotSenderIsStart((pSyncNode->senders)[i])) {
-        snapshotSenderStop((pSyncNode->senders)[i], false);
+      if (snapshotSenderIsStart(pSyncNode->senders[i])) {
+        snapshotSenderStop(pSyncNode->senders[i], false);
       }
 
-      snapshotSenderDestroy((pSyncNode->senders)[i]);
-      (pSyncNode->senders)[i] = NULL;
+      snapshotSenderDestroy(pSyncNode->senders[i]);
+      pSyncNode->senders[i] = NULL;
     }
   }
 
@@ -1312,6 +1318,7 @@ void syncNodeClose(SSyncNode* pSyncNode) {
       snapshotReceiverForceStop(pSyncNode->pNewNodeReceiver);
     }
 
+    sDebug("vgId:%d, snapshot receiver destroy while close, data:%p", pSyncNode->vgId, pSyncNode->pNewNodeReceiver);
     snapshotReceiverDestroy(pSyncNode->pNewNodeReceiver);
     pSyncNode->pNewNodeReceiver = NULL;
   }
@@ -1382,8 +1389,7 @@ int32_t syncNodeRestartElectTimer(SSyncNode* pSyncNode, int32_t ms) {
   return ret;
 }
 
-int32_t syncNodeResetElectTimer(SSyncNode* pSyncNode) {
-  int32_t ret = 0;
+void syncNodeResetElectTimer(SSyncNode* pSyncNode) {
   int32_t electMS;
 
   if (pSyncNode->pRaftCfg->isStandBy) {
@@ -1391,11 +1397,11 @@ int32_t syncNodeResetElectTimer(SSyncNode* pSyncNode) {
   } else {
     electMS = syncUtilElectRandomMS(pSyncNode->electBaseLine, 2 * pSyncNode->electBaseLine);
   }
-  ret = syncNodeRestartElectTimer(pSyncNode, electMS);
+
+  (void)syncNodeRestartElectTimer(pSyncNode, electMS);
 
   sNTrace(pSyncNode, "reset elect timer, min:%d, max:%d, ms:%d", pSyncNode->electBaseLine, 2 * pSyncNode->electBaseLine,
           electMS);
-  return ret;
 }
 
 static int32_t syncNodeDoStartHeartbeatTimer(SSyncNode* pSyncNode) {
@@ -1455,23 +1461,20 @@ int32_t syncNodeRestartHeartbeatTimer(SSyncNode* pSyncNode) {
   return 0;
 }
 
-// utils --------------
 int32_t syncNodeSendMsgById(const SRaftId* destRaftId, SSyncNode* pSyncNode, SRpcMsg* pMsg) {
   SEpSet epSet;
   syncUtilRaftId2EpSet(destRaftId, &epSet);
+
   if (pSyncNode->syncSendMSg != NULL) {
-    // htonl
     syncUtilMsgHtoN(pMsg->pCont);
-
     pMsg->info.noResp = 1;
-    pSyncNode->syncSendMSg(&epSet, pMsg);
+    return pSyncNode->syncSendMSg(&epSet, pMsg);
   } else {
     sError("vgId:%d, sync send msg by id error, fp-send-msg is null", pSyncNode->vgId);
     rpcFreeCont(pMsg->pCont);
+    terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
     return -1;
   }
-
-  return 0;
 }
 
 int32_t syncNodeSendMsgByInfo(const SNodeInfo* nodeInfo, SSyncNode* pSyncNode, SRpcMsg* pMsg) {
@@ -1586,7 +1589,7 @@ void syncNodeDoConfigChange(SSyncNode* pSyncNode, SSyncCfg* pNewConfig, SyncInde
     memcpy(oldReplicasId, pSyncNode->replicasId, sizeof(oldReplicasId));
     SSyncSnapshotSender* oldSenders[TSDB_MAX_REPLICA];
     for (int32_t i = 0; i < TSDB_MAX_REPLICA; ++i) {
-      oldSenders[i] = (pSyncNode->senders)[i];
+      oldSenders[i] = pSyncNode->senders[i];
       sSTrace(oldSenders[i], "snapshot sender save old");
     }
 
@@ -1625,7 +1628,7 @@ void syncNodeDoConfigChange(SSyncNode* pSyncNode, SSyncCfg* pNewConfig, SyncInde
 
     // clear new
     for (int32_t i = 0; i < TSDB_MAX_REPLICA; ++i) {
-      (pSyncNode->senders)[i] = NULL;
+      pSyncNode->senders[i] = NULL;
     }
 
     // reset new
@@ -1640,16 +1643,16 @@ void syncNodeDoConfigChange(SSyncNode* pSyncNode, SSyncCfg* pNewConfig, SyncInde
           sNTrace(pSyncNode, "snapshot sender reset for: %" PRId64 ", newIndex:%d, %s:%d, %p",
                   (pSyncNode->replicasId)[i].addr, i, host, port, oldSenders[j]);
 
-          (pSyncNode->senders)[i] = oldSenders[j];
+          pSyncNode->senders[i] = oldSenders[j];
           oldSenders[j] = NULL;
           reset = true;
 
           // reset replicaIndex
-          int32_t oldreplicaIndex = (pSyncNode->senders)[i]->replicaIndex;
-          (pSyncNode->senders)[i]->replicaIndex = i;
+          int32_t oldreplicaIndex = pSyncNode->senders[i]->replicaIndex;
+          pSyncNode->senders[i]->replicaIndex = i;
 
           sNTrace(pSyncNode, "snapshot sender udpate replicaIndex from %d to %d, %s:%d, %p, reset:%d", oldreplicaIndex,
-                  i, host, port, (pSyncNode->senders)[i], reset);
+                  i, host, port, pSyncNode->senders[i], reset);
 
           break;
         }
@@ -1658,18 +1661,23 @@ void syncNodeDoConfigChange(SSyncNode* pSyncNode, SSyncCfg* pNewConfig, SyncInde
 
     // create new
     for (int32_t i = 0; i < TSDB_MAX_REPLICA; ++i) {
-      if ((pSyncNode->senders)[i] == NULL) {
-        (pSyncNode->senders)[i] = snapshotSenderCreate(pSyncNode, i);
-        sSTrace((pSyncNode->senders)[i], "snapshot sender create new while reconfig, data:%p", (pSyncNode->senders)[i]);
+      if (pSyncNode->senders[i] == NULL) {
+        pSyncNode->senders[i] = snapshotSenderCreate(pSyncNode, i);
+        if (pSyncNode->senders[i] == NULL) {
+          // will be created later while send snapshot
+          sSError(pSyncNode->senders[i], "snapshot sender create failed while reconfig");
+        } else {
+          sSDebug(pSyncNode->senders[i], "snapshot sender create while reconfig, data:%p", pSyncNode->senders[i]);
+        }
       } else {
-        sSTrace((pSyncNode->senders)[i], "snapshot sender already exist, data:%p", (pSyncNode->senders)[i]);
+        sSDebug(pSyncNode->senders[i], "snapshot sender already exist, data:%p", pSyncNode->senders[i]);
       }
     }
 
     // free old
     for (int32_t i = 0; i < TSDB_MAX_REPLICA; ++i) {
       if (oldSenders[i] != NULL) {
-        sNTrace(pSyncNode, "snapshot sender destroy old, data:%p replica-index:%d", oldSenders[i], i);
+        sSDebug(oldSenders[i], "snapshot sender destroy old, data:%p replica-index:%d", oldSenders[i], i);
         snapshotSenderDestroy(oldSenders[i]);
         oldSenders[i] = NULL;
       }
@@ -1844,8 +1852,8 @@ void syncNodeBecomeLeader(SSyncNode* pSyncNode, const char* debugStr) {
   SSyncSnapshotSender* pMySender = syncNodeGetSnapshotSender(pSyncNode, &(pSyncNode->myRaftId));
   if (pMySender != NULL) {
     for (int32_t i = 0; i < pSyncNode->pMatchIndex->replicaNum; ++i) {
-      if ((pSyncNode->senders)[i]->privateTerm > pMySender->privateTerm) {
-        pMySender->privateTerm = (pSyncNode->senders)[i]->privateTerm;
+      if (pSyncNode->senders[i]->privateTerm > pMySender->privateTerm) {
+        pMySender->privateTerm = pSyncNode->senders[i]->privateTerm;
       }
     }
     (pMySender->privateTerm) += 100;
@@ -2376,9 +2384,20 @@ int32_t syncCacheEntry(SSyncLogStore* pLogStore, SSyncRaftEntry* pEntry, LRUHand
 }
 
 int32_t syncNodeAppend(SSyncNode* ths, SSyncRaftEntry* pEntry) {
+  if (pEntry->dataLen < sizeof(SMsgHead)) {
+    sError("vgId:%d, cannot append an invalid client request with no msg head. type:%s, dataLen:%d", ths->vgId,
+           TMSG_INFO(pEntry->originalRpcType), pEntry->dataLen);
+    syncEntryDestroy(pEntry);
+    return -1;
+  }
+
   // append to log buffer
   if (syncLogBufferAppend(ths->pLogBuf, ths, pEntry) < 0) {
-    sError("vgId:%d, failed to enqueue sync log buffer. index:%" PRId64 "", ths->vgId, pEntry->index);
+    sError("vgId:%d, failed to enqueue sync log buffer, index:%" PRId64, ths->vgId, pEntry->index);
+    terrno = TSDB_CODE_SYN_BUFFER_FULL;
+    (void)syncLogFsmExecute(ths, ths->pFsm, ths->state, ths->pRaftStore->currentTerm, pEntry,
+                            TSDB_CODE_SYN_BUFFER_FULL);
+    syncEntryDestroy(pEntry);
     return -1;
   }
 
@@ -2671,16 +2690,24 @@ int32_t syncNodeOnClientRequest(SSyncNode* ths, SRpcMsg* pMsg, SyncIndex* pRetIn
     pEntry = syncEntryBuildFromRpcMsg(pMsg, term, index);
   }
 
+  if (pEntry == NULL) {
+    sError("vgId:%d, failed to process client request since %s.", ths->vgId, terrstr());
+    return -1;
+  }
+
   if (ths->state == TAOS_SYNC_STATE_LEADER) {
     if (pRetIndex) {
       (*pRetIndex) = index;
     }
 
     int32_t code = syncNodeAppend(ths, pEntry);
-    if (code < 0 && ths->vgId != 1 && vnodeIsMsgBlock(pEntry->originalRpcType)) {
-      ASSERTS(false, "failed to append blocking msg");
+    if (code < 0) {
+      sNError(ths, "failed to append blocking msg");
     }
     return code;
+  } else {
+    syncEntryDestroy(pEntry);
+    pEntry = NULL;
   }
 
   return -1;

source/libs/sync/src/syncPipeline.c

@@ -26,6 +26,11 @@
 #include "syncSnapshot.h"
 #include "syncUtil.h"
 
+static bool syncIsMsgBlock(tmsg_t type) {
+  return (type == TDMT_VND_CREATE_TABLE) || (type == TDMT_VND_ALTER_TABLE) || (type == TDMT_VND_DROP_TABLE) ||
+         (type == TDMT_VND_UPDATE_TAG_VAL) || (type == TDMT_VND_ALTER_CONFIRM);
+}
+
 int64_t syncLogBufferGetEndIndex(SSyncLogBuffer* pBuf) {
   taosThreadMutexLock(&pBuf->mutex);
   int64_t index = pBuf->endIndex;
@@ -443,16 +448,15 @@ _out:
   return matchIndex;
 }
 
-int32_t syncLogFsmExecute(SSyncNode* pNode, SSyncFSM* pFsm, ESyncState role, SyncTerm term, SSyncRaftEntry* pEntry) {
-  ASSERTS(pFsm->FpCommitCb != NULL, "No commit cb registered for the FSM");
-
+int32_t syncLogFsmExecute(SSyncNode* pNode, SSyncFSM* pFsm, ESyncState role, SyncTerm term, SSyncRaftEntry* pEntry,
+                          int32_t applyCode) {
   if ((pNode->replicaNum == 1) && pNode->restoreFinish && pNode->vgId != 1) {
     return 0;
   }
 
-  if (pNode->vgId != 1 && vnodeIsMsgBlock(pEntry->originalRpcType)) {
-    sTrace("vgId:%d, blocking msg ready to execute. index:%" PRId64 ", term: %" PRId64 ", type: %s", pNode->vgId,
-           pEntry->index, pEntry->term, TMSG_INFO(pEntry->originalRpcType));
+  if (pNode->vgId != 1 && syncIsMsgBlock(pEntry->originalRpcType)) {
+    sTrace("vgId:%d, blocking msg ready to execute, index:%" PRId64 ", term:%" PRId64 ", type:%s code:0x%x",
+           pNode->vgId, pEntry->index, pEntry->term, TMSG_INFO(pEntry->originalRpcType), applyCode);
   }
 
   if (pEntry->originalRpcType == TDMT_VND_COMMIT) {
@@ -460,14 +464,14 @@ int32_t syncLogFsmExecute(SSyncNode* pNode, SSyncFSM* pFsm, ESyncState role, Syn
           pEntry->term);
   }
 
-  SRpcMsg rpcMsg = {0};
+  SRpcMsg rpcMsg = {.code = applyCode};
   syncEntry2OriginalRpc(pEntry, &rpcMsg);
 
   SFsmCbMeta cbMeta = {0};
   cbMeta.index = pEntry->index;
   cbMeta.lastConfigIndex = syncNodeGetSnapshotConfigIndex(pNode, pEntry->index);
   cbMeta.isWeak = pEntry->isWeak;
-  cbMeta.code = 0;
+  cbMeta.code = applyCode;
   cbMeta.state = role;
   cbMeta.seqNum = pEntry->seqNum;
   cbMeta.term = pEntry->term;
@@ -476,7 +480,6 @@ int32_t syncLogFsmExecute(SSyncNode* pNode, SSyncFSM* pFsm, ESyncState role, Syn
 
   (void)syncRespMgrGetAndDel(pNode->pSyncRespMgr, cbMeta.seqNum, &rpcMsg.info);
   int32_t code = pFsm->FpCommitCb(pFsm, &rpcMsg, &cbMeta);
-  ASSERT(rpcMsg.pCont == NULL);
   return code;
 }
 
@@ -527,7 +530,7 @@ int32_t syncLogBufferCommit(SSyncLogBuffer* pBuf, SSyncNode* pNode, int64_t comm
             pEntry->term, TMSG_INFO(pEntry->originalRpcType));
     }
 
-    if (syncLogFsmExecute(pNode, pFsm, role, term, pEntry) != 0) {
+    if (syncLogFsmExecute(pNode, pFsm, role, term, pEntry, 0) != 0) {
       sError("vgId:%d, failed to execute sync log entry. index:%" PRId64 ", term:%" PRId64
              ", role: %d, current term: %" PRId64,
              vgId, pEntry->index, pEntry->term, role, term);
@@ -574,7 +577,9 @@ _out:
   return ret;
 }
 
-int32_t syncLogReplMgrReset(SSyncLogReplMgr* pMgr) {
+void syncLogReplMgrReset(SSyncLogReplMgr* pMgr) {
+  if (pMgr == NULL) return;
+
   ASSERT(pMgr->startIndex >= 0);
   for (SyncIndex index = pMgr->startIndex; index < pMgr->endIndex; index++) {
     memset(&pMgr->states[index % pMgr->size], 0, sizeof(pMgr->states[0]));
@@ -584,7 +589,6 @@ int32_t syncLogReplMgrReset(SSyncLogReplMgr* pMgr) {
   pMgr->endIndex = 0;
   pMgr->restored = false;
   pMgr->retryBackoff = 0;
-  return 0;
 }
 
 int32_t syncLogReplMgrRetryOnNeed(SSyncLogReplMgr* pMgr, SSyncNode* pNode) {

source/libs/transport/src/transCli.c

@@ -1665,11 +1665,20 @@ int cliAppCb(SCliConn* pConn, STransMsg* pResp, SCliMsg* pMsg) {
   if (pCtx->retryCode != TSDB_CODE_SUCCESS) {
     int32_t code = pResp->code;
     // return internal code app
-    if (code == TSDB_CODE_RPC_NETWORK_UNAVAIL || code == TSDB_CODE_RPC_BROKEN_LINK) {
+    if (code == TSDB_CODE_RPC_NETWORK_UNAVAIL || code == TSDB_CODE_RPC_BROKEN_LINK || code == TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED) {
       pResp->code = pCtx->retryCode;
     }
   }
 
+  // check whole vnodes is offline on this vgroup
+  if (pCtx->epsetRetryCnt >= pCtx->epSet.numOfEps || pCtx->retryStep > 0) {
+    if (pResp->code == TSDB_CODE_RPC_NETWORK_UNAVAIL) {
+      pResp->code = TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED;
+    } else if (pResp->code == TSDB_CODE_RPC_BROKEN_LINK) {
+      pResp->code = TSDB_CODE_RPC_SOMENODE_BROKEN_LINK;
+    }
+  }
+
   STraceId* trace = &pResp->info.traceId;
   bool      hasEpSet = cliTryExtractEpSet(pResp, &pCtx->epSet);
   if (hasEpSet) {

source/util/src/terror.c

@@ -51,6 +51,7 @@ TAOS_DEFINE_ERROR(TSDB_CODE_RPC_FQDN_ERROR,               "Unable to resolve FQD
 TAOS_DEFINE_ERROR(TSDB_CODE_RPC_PORT_EADDRINUSE,          "Port already in use")
 TAOS_DEFINE_ERROR(TSDB_CODE_RPC_BROKEN_LINK,              "Conn is broken")
 TAOS_DEFINE_ERROR(TSDB_CODE_RPC_TIMEOUT,                  "Conn read timeout")
+TAOS_DEFINE_ERROR(TSDB_CODE_RPC_SOMENODE_NOT_CONNECTED,   "some vnode/qnode/mnode(s) out of service")
 
 //common & util
 TAOS_DEFINE_ERROR(TSDB_CODE_TIME_UNSYNCED,                "Client and server's time is not synchronized")
@@ -406,6 +407,7 @@ TAOS_DEFINE_ERROR(TSDB_CODE_SYN_STANDBY_NOT_READY,        "Sync not ready for st
 TAOS_DEFINE_ERROR(TSDB_CODE_SYN_BATCH_ERROR,              "Sync batch error")
 TAOS_DEFINE_ERROR(TSDB_CODE_SYN_RESTORING,                "Sync is restoring")
 TAOS_DEFINE_ERROR(TSDB_CODE_SYN_INVALID_SNAPSHOT_MSG,     "Sync invalid snapshot msg")
+TAOS_DEFINE_ERROR(TSDB_CODE_SYN_BUFFER_FULL,              "Sync buffer is full")
 TAOS_DEFINE_ERROR(TSDB_CODE_SYN_INTERNAL_ERROR,           "Sync internal error")
 
 //tq