Merge branch '3.0' into feature/tq

include/common/tmsg.h

@@ -791,19 +791,24 @@ typedef struct {
 int32_t tSerializeSQnodeListReq(void* buf, int32_t bufLen, SQnodeListReq* pReq);
 int32_t tDeserializeSQnodeListReq(void* buf, int32_t bufLen, SQnodeListReq* pReq);
 
+typedef struct SQueryNodeAddr {
+  int32_t nodeId;  // vgId or qnodeId
+  SEpSet  epSet;
+} SQueryNodeAddr;
+
 typedef struct {
-  SArray* addrsList;  // SArray<SQueryNodeAddr>
+  SQueryNodeAddr addr;
+  uint64_t       load;
+} SQueryNodeLoad;
+
+typedef struct {
+  SArray* qnodeList;  // SArray<SQueryNodeLoad>
 } SQnodeListRsp;
 
 int32_t tSerializeSQnodeListRsp(void* buf, int32_t bufLen, SQnodeListRsp* pRsp);
 int32_t tDeserializeSQnodeListRsp(void* buf, int32_t bufLen, SQnodeListRsp* pRsp);
 void    tFreeSQnodeListRsp(SQnodeListRsp* pRsp);
 
-typedef struct SQueryNodeAddr {
-  int32_t nodeId;  // vgId or qnodeId
-  SEpSet  epSet;
-} SQueryNodeAddr;
-
 typedef struct {
   SArray* pArray;  // Array of SUseDbRsp
 } SUseDbBatchRsp;
@@ -926,6 +931,21 @@ typedef struct {
   int32_t syncState;
 } SMnodeLoad;
 
+typedef struct {
+  int32_t dnodeId;
+  int64_t numOfProcessedQuery;
+  int64_t numOfProcessedCQuery;
+  int64_t numOfProcessedFetch;
+  int64_t numOfProcessedDrop;
+  int64_t numOfProcessedHb;
+  int64_t cacheDataSize;
+  int64_t numOfQueryInQueue;
+  int64_t numOfFetchInQueue;
+  int64_t timeInQueryQueue;
+  int64_t timeInFetchQueue;
+} SQnodeLoad;
+
+
 typedef struct {
   int32_t     sver;      // software version
   int64_t     dnodeVer;  // dnode table version in sdb
@@ -937,6 +957,7 @@ typedef struct {
   int32_t     numOfSupportVnodes;
   char        dnodeEp[TSDB_EP_LEN];
   SMnodeLoad  mload;
+  SQnodeLoad  qload;
   SClusterCfg clusterCfg;
   SArray*     pVloads;  // array of SVnodeLoad
 } SStatusReq;
@@ -1012,6 +1033,7 @@ typedef struct {
 
   // for tsma
   int8_t isTsma;
+  void*  pTsma;
 
 } SCreateVnodeReq;
 
@@ -1946,6 +1968,7 @@ typedef struct {
   int8_t   killConnection;
   int8_t   align[3];
   SEpSet   epSet;
+  SArray  *pQnodeList;
 } SQueryHbRspBasic;
 
 typedef struct {
@@ -2025,7 +2048,10 @@ static FORCE_INLINE void tFreeClientKv(void* pKv) {
 
 static FORCE_INLINE void tFreeClientHbRsp(void* pRsp) {
   SClientHbRsp* rsp = (SClientHbRsp*)pRsp;
+  if (rsp->query) {
+    taosArrayDestroy(rsp->query->pQnodeList);
     taosMemoryFreeClear(rsp->query);
+  }
   if (rsp->info) taosArrayDestroyEx(rsp->info, tFreeClientKv);
 }
 

include/common/tmsgdef.h

@@ -252,6 +252,7 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_MON_BM_INFO, "monitor-binfo", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MON_VM_LOAD, "monitor-vload", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MON_MM_LOAD, "monitor-mload", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_MON_QM_LOAD, "monitor-qload", NULL, NULL)
   
 #if defined(TD_MSG_NUMBER_)
   TDMT_MAX

include/dnode/qnode/qnode.h

@@ -25,20 +25,6 @@ extern "C" {
 /* ------------------------ TYPES EXPOSED ------------------------ */
 typedef struct SQnode SQnode;
 
-typedef struct {
-  int64_t numOfProcessedQuery;
-  int64_t numOfProcessedCQuery;
-  int64_t numOfProcessedFetch;
-  int64_t numOfProcessedDrop;
-  int64_t memSizeInCache;
-  int64_t dataSizeSend;
-  int64_t dataSizeRecv;
-  int64_t numOfQueryInQueue;
-  int64_t numOfFetchInQueue;
-  int64_t waitTimeInQueryQUeue;
-  int64_t waitTimeInFetchQUeue;
-} SQnodeLoad;
-
 typedef struct {
   SMsgCb msgCb;
 } SQnodeOpt;

include/libs/executor/dataSinkMgt.h

@@ -32,6 +32,10 @@ extern "C" {
 struct SDataSink;
 struct SSDataBlock;
 
+typedef struct SDataSinkStat {
+  uint64_t cachedSize;
+} SDataSinkStat;
+
 typedef struct SDataSinkMgtCfg {
   uint32_t maxDataBlockNum;           // todo: this should be numOfRows?
   uint32_t maxDataBlockNumPerQuery;
@@ -62,6 +66,8 @@ typedef struct SOutputData {
  */
 int32_t dsCreateDataSinker(const SDataSinkNode* pDataSink, DataSinkHandle* pHandle);
 
+int32_t dsDataSinkGetCacheSize(SDataSinkStat *pStat);
+
 /**
  * Put the result set returned by the executor into datasinker.
  * @param handle
@@ -88,6 +94,8 @@ void dsGetDataLength(DataSinkHandle handle, int32_t* pLen, bool* pQueryEnd);
  */
 int32_t dsGetDataBlock(DataSinkHandle handle, SOutputData* pOutput);
 
+int32_t dsGetCacheSize(DataSinkHandle handle, uint64_t *pSize);
+
 /**
  * After dsGetStatus returns DS_NEED_SCHEDULE, the caller need to put this into the work queue.
  * @param ahandle

include/libs/index/index.h

@@ -194,6 +194,7 @@ void indexInit();
 /* index filter */
 typedef struct SIndexMetaArg {
   void*    metaHandle;
+  void*    metaEx;
   uint64_t suid;
 } SIndexMetaArg;
 

include/libs/monitor/monitor.h

@@ -171,6 +171,7 @@ void    tFreeSMonVmInfo(SMonVmInfo *pInfo);
 typedef struct {
   SMonSysInfo sys;
   SMonLogs    log;
+  SQnodeLoad  load;
 } SMonQmInfo;
 
 int32_t tSerializeSMonQmInfo(void *buf, int32_t bufLen, SMonQmInfo *pInfo);
@@ -210,6 +211,10 @@ typedef struct {
 int32_t tSerializeSMonMloadInfo(void *buf, int32_t bufLen, SMonMloadInfo *pInfo);
 int32_t tDeserializeSMonMloadInfo(void *buf, int32_t bufLen, SMonMloadInfo *pInfo);
 
+int32_t tSerializeSQnodeLoad(void *buf, int32_t bufLen, SQnodeLoad *pInfo);
+int32_t tDeserializeSQnodeLoad(void *buf, int32_t bufLen, SQnodeLoad *pInfo);
+
+
 typedef struct {
   const char *server;
   uint16_t    port;

include/libs/qworker/qworker.h

@@ -22,7 +22,7 @@ extern "C" {
 
 #include "tmsgcb.h"
 #include "trpc.h"
-
+#include "executor.h"
 
 enum {
   NODE_TYPE_VNODE = 1,
@@ -40,13 +40,19 @@ typedef struct SQWorkerCfg {
 } SQWorkerCfg;
 
 typedef struct {
-  uint64_t numOfStartTask;
-  uint64_t numOfStopTask;
-  uint64_t numOfRecvedFetch;
-  uint64_t numOfSentHb;
-  uint64_t numOfSentFetch;
-  uint64_t numOfTaskInQueue;
+  uint64_t cacheDataSize;
+  
+  uint64_t queryProcessed;
+  uint64_t cqueryProcessed;
+  uint64_t fetchProcessed;
+  uint64_t dropProcessed;
+  uint64_t hbProcessed;
+  
+  uint64_t numOfQueryInQueue;
   uint64_t numOfFetchInQueue;
+  uint64_t timeInQueryQueue;
+  uint64_t timeInFetchQueue;
+  
   uint64_t numOfErrors;
 } SQWorkerStat;
 
@@ -68,7 +74,7 @@ int32_t qWorkerProcessHbMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg, int64_
 
 void qWorkerDestroy(void **qWorkerMgmt);
 
-int64_t qWorkerGetWaitTimeInQueue(void *qWorkerMgmt, EQueueType type);
+int32_t qWorkerGetStat(SReadHandle *handle, void *qWorkerMgmt, SQWorkerStat *pStat);
 
 #ifdef __cplusplus
 }

include/util/taoserror.h

@@ -182,8 +182,9 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_MND_BNODE_ALREADY_EXIST       TAOS_DEF_ERROR_CODE(0, 0x0356)
 #define TSDB_CODE_MND_BNODE_NOT_EXIST           TAOS_DEF_ERROR_CODE(0, 0x0357)
 #define TSDB_CODE_MND_TOO_FEW_MNODES            TAOS_DEF_ERROR_CODE(0, 0x0358)
-#define TSDB_CODE_MND_MNODE_DEPLOYED            TAOS_DEF_ERROR_CODE(0, 0x0359)
-#define TSDB_CODE_MND_CANT_DROP_MASTER          TAOS_DEF_ERROR_CODE(0, 0x035A)
+#define TSDB_CODE_MND_TOO_MANY_MNODES           TAOS_DEF_ERROR_CODE(0, 0x0359)
+#define TSDB_CODE_MND_MNODE_DEPLOYED            TAOS_DEF_ERROR_CODE(0, 0x035A)
+#define TSDB_CODE_MND_CANT_DROP_MASTER          TAOS_DEF_ERROR_CODE(0, 0x035B)
 
 // mnode-acct
 #define TSDB_CODE_MND_ACCT_ALREADY_EXIST        TAOS_DEF_ERROR_CODE(0, 0x0360)

include/util/tdef.h

@@ -253,8 +253,7 @@ typedef enum ELogicConditionType {
 
 #define TSDB_TRANS_STAGE_LEN 12
 #define TSDB_TRANS_TYPE_LEN  16
-#define TSDB_TRANS_ERROR_LEN 64
-#define TSDB_TRANS_DESC_LEN  128
+#define TSDB_TRANS_ERROR_LEN 512
 
 #define TSDB_STEP_NAME_LEN 32
 #define TSDB_STEP_DESC_LEN 128
@@ -343,7 +342,7 @@ typedef enum ELogicConditionType {
 #define TSDB_DEFAULT_DB_SCHEMALESS      TSDB_DB_SCHEMALESS_OFF
 
 #define TSDB_MIN_ROLLUP_FILE_FACTOR     0
-#define TSDB_MAX_ROLLUP_FILE_FACTOR     1
+#define TSDB_MAX_ROLLUP_FILE_FACTOR     10
 #define TSDB_DEFAULT_ROLLUP_FILE_FACTOR 0.1
 #define TSDB_MIN_TABLE_TTL              0
 #define TSDB_DEFAULT_TABLE_TTL          0

include/util/tencode.h

@@ -378,14 +378,16 @@ static FORCE_INLINE int32_t tDecodeDouble(SDecoder* pCoder, double* val) {
 }
 
 static FORCE_INLINE int32_t tDecodeBinary(SDecoder* pCoder, uint8_t** val, uint32_t* len) {
-  if (tDecodeU32v(pCoder, len) < 0) return -1;
+  uint32_t length = 0;
+  if (tDecodeU32v(pCoder, &length) < 0) return -1;
+  if (len) *len = length;
 
-  if (TD_CODER_CHECK_CAPACITY_FAILED(pCoder, *len)) return -1;
+  if (TD_CODER_CHECK_CAPACITY_FAILED(pCoder, length)) return -1;
   if (val) {
     *val = (uint8_t*)TD_CODER_CURRENT(pCoder);
   }
 
-  TD_CODER_MOVE_POS(pCoder, *len);
+  TD_CODER_MOVE_POS(pCoder, length);
   return 0;
 }
 
@@ -410,14 +412,16 @@ static int32_t tDecodeCStrTo(SDecoder* pCoder, char* val) {
 }
 
 static FORCE_INLINE int32_t tDecodeBinaryAlloc(SDecoder* pCoder, void** val, uint64_t* len) {
-  if (tDecodeU64v(pCoder, len) < 0) return -1;
+  uint64_t length = 0;
+  if (tDecodeU64v(pCoder, &length) < 0) return -1;
+  if (len) *len = length;
 
-  if (TD_CODER_CHECK_CAPACITY_FAILED(pCoder, *len)) return -1;
-  *val = taosMemoryMalloc(*len);
+  if (TD_CODER_CHECK_CAPACITY_FAILED(pCoder, length)) return -1;
+  *val = taosMemoryMalloc(length);
   if (*val == NULL) return -1;
-  memcpy(*val, TD_CODER_CURRENT(pCoder), *len);
+  memcpy(*val, TD_CODER_CURRENT(pCoder), length);
 
-  TD_CODER_MOVE_POS(pCoder, *len);
+  TD_CODER_MOVE_POS(pCoder, length);
   return 0;
 }
 

source/client/inc/clientInt.h

@@ -119,6 +119,8 @@ typedef struct SHeartBeatInfo {
 struct SAppInstInfo {
   int64_t          numOfConns;
   SCorEpSet        mgmtEp;
+  TdThreadMutex    qnodeMutex;
+  SArray*          pQnodeList;
   SInstanceSummary summary;
   SList*           pConnList;  // STscObj linked list
   uint64_t         clusterId;
@@ -290,7 +292,7 @@ SRequestObj* launchQuery(STscObj* pTscObj, const char* sql, int sqlLen);
 
 int32_t parseSql(SRequestObj* pRequest, bool topicQuery, SQuery** pQuery, SStmtCallback* pStmtCb);
 
-int32_t getPlan(SRequestObj* pRequest, SQuery* pQuery, SQueryPlan** pPlan, SArray* pNodeList);
+int32_t getPlan(SRequestObj* pRequest, SQuery* pQuery, SQueryPlan** pPlan, SArray** pNodeList);
 
 int32_t buildRequest(STscObj* pTscObj, const char* sql, int sqlLen, SRequestObj** pRequest);
 
@@ -317,6 +319,7 @@ SRequestObj* launchQueryImpl(SRequestObj* pRequest, SQuery* pQuery, int32_t code
 int32_t      getQueryPlan(SRequestObj* pRequest, SQuery* pQuery, SArray** pNodeList);
 int32_t      scheduleQuery(SRequestObj* pRequest, SQueryPlan* pDag, SArray* pNodeList, void** res);
 int32_t      refreshMeta(STscObj* pTscObj, SRequestObj* pRequest);
+int32_t updateQnodeList(SAppInstInfo*pInfo, SArray* pNodeList);
 
 #ifdef __cplusplus
 }

source/client/src/clientHb.c

@@ -160,6 +160,10 @@ static int32_t hbQueryHbRspHandle(SAppHbMgr *pAppHbMgr, SClientHbRsp *pRsp) {
         taos_close(pTscObj);
       }
 
+      if (pRsp->query->pQnodeList) {
+        updateQnodeList(pTscObj->pAppInfo, pRsp->query->pQnodeList);
+      }
+
       releaseTscObj(pRsp->connKey.tscRid);
     }
   }

source/client/src/clientImpl.c

@@ -118,6 +118,7 @@ TAOS* taos_connect_internal(const char* ip, const char* user, const char* pass,
   if (pInst == NULL) {
     p = taosMemoryCalloc(1, sizeof(struct SAppInstInfo));
     p->mgmtEp = epSet;    
+    taosThreadMutexInit(&p->qnodeMutex, NULL);
     p->pTransporter = openTransporter(user, secretEncrypt, tsNumOfCores);
     p->pAppHbMgr = appHbMgrInit(p, key);
     taosHashPut(appInfo.pInstMap, key, strlen(key), &p, POINTER_BYTES);
@@ -228,7 +229,61 @@ int32_t execDdlQuery(SRequestObj* pRequest, SQuery* pQuery) {
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t getPlan(SRequestObj* pRequest, SQuery* pQuery, SQueryPlan** pPlan, SArray* pNodeList) {
+int compareQueryNodeLoad(const void* elem1, const void* elem2) {
+  SQueryNodeLoad *node1 = (SQueryNodeLoad *)elem1;
+  SQueryNodeLoad *node2 = (SQueryNodeLoad *)elem2;
+
+  if (node1->load < node2->load) {
+    return -1;
+  }
+  
+  return node1->load > node2->load;
+}
+
+int32_t updateQnodeList(SAppInstInfo*pInfo, SArray* pNodeList) {
+  taosThreadMutexLock(&pInfo->qnodeMutex);
+  if (pInfo->pQnodeList) {
+    taosArrayDestroy(pInfo->pQnodeList);
+    pInfo->pQnodeList = NULL;
+  }
+  
+  if (pNodeList) {
+    pInfo->pQnodeList = taosArrayDup(pNodeList);
+    taosArraySort(pInfo->pQnodeList, compareQueryNodeLoad);
+  }
+  taosThreadMutexUnlock(&pInfo->qnodeMutex);
+
+  return TSDB_CODE_SUCCESS;
+}
+
+int32_t getQnodeList(SRequestObj* pRequest, SArray** pNodeList) {
+  SAppInstInfo*pInfo = pRequest->pTscObj->pAppInfo;
+  int32_t code = 0;
+  
+  taosThreadMutexLock(&pInfo->qnodeMutex);
+  if (pInfo->pQnodeList) {
+    *pNodeList = taosArrayDup(pInfo->pQnodeList);
+  }
+  taosThreadMutexUnlock(&pInfo->qnodeMutex);
+
+  if (NULL == *pNodeList) {
+    SEpSet       mgmtEpSet = getEpSet_s(&pRequest->pTscObj->pAppInfo->mgmtEp);
+    SCatalog*    pCatalog = NULL;
+    code = catalogGetHandle(pRequest->pTscObj->pAppInfo->clusterId, &pCatalog);
+    if (TSDB_CODE_SUCCESS == code) {
+      *pNodeList = taosArrayInit(5, sizeof(SQueryNodeLoad));
+      code = catalogGetQnodeList(pCatalog, pRequest->pTscObj->pAppInfo->pTransporter, &mgmtEpSet, *pNodeList);
+    }
+    
+    if (TSDB_CODE_SUCCESS == code && *pNodeList) {
+      code = updateQnodeList(pInfo, *pNodeList);
+    }
+  }
+
+  return code;
+}
+
+int32_t getPlan(SRequestObj* pRequest, SQuery* pQuery, SQueryPlan** pPlan, SArray** pNodeList) {
   pRequest->type = pQuery->msgType;
   SPlanContext cxt = {.queryId = pRequest->requestId,
                       .acctId = pRequest->pTscObj->acctId,
@@ -237,14 +292,10 @@ int32_t getPlan(SRequestObj* pRequest, SQuery* pQuery, SQueryPlan** pPlan, SArra
                       .showRewrite = pQuery->showRewrite,
                       .pMsg = pRequest->msgBuf,
                       .msgLen = ERROR_MSG_BUF_DEFAULT_SIZE};
-  SEpSet       mgmtEpSet = getEpSet_s(&pRequest->pTscObj->pAppInfo->mgmtEp);
-  SCatalog*    pCatalog = NULL;
-  int32_t      code = catalogGetHandle(pRequest->pTscObj->pAppInfo->clusterId, &pCatalog);
-  if (TSDB_CODE_SUCCESS == code) {
-    code = catalogGetQnodeList(pCatalog, pRequest->pTscObj->pAppInfo->pTransporter, &mgmtEpSet, pNodeList);
-  }
+
+  int32_t code = getQnodeList(pRequest, pNodeList);
   if (TSDB_CODE_SUCCESS == code) {
-    code = qCreateQueryPlan(&cxt, pPlan, pNodeList);
+    code = qCreateQueryPlan(&cxt, pPlan, *pNodeList);
   }
   return code;
 }
@@ -369,8 +420,7 @@ int32_t scheduleQuery(SRequestObj* pRequest, SQueryPlan* pDag, SArray* pNodeList
 }
 
 int32_t getQueryPlan(SRequestObj* pRequest, SQuery* pQuery, SArray** pNodeList) {
-  *pNodeList = taosArrayInit(4, sizeof(struct SQueryNodeAddr));
-  return getPlan(pRequest, pQuery, &pRequest->body.pDag, *pNodeList);
+  return getPlan(pRequest, pQuery, &pRequest->body.pDag, pNodeList);
 }
 
 int32_t validateSversion(SRequestObj* pRequest, void* res) {
@@ -456,8 +506,8 @@ SRequestObj* launchQueryImpl(SRequestObj* pRequest, SQuery* pQuery, int32_t code
         code = execDdlQuery(pRequest, pQuery);
         break;
       case QUERY_EXEC_MODE_SCHEDULE: {
-        SArray* pNodeList = taosArrayInit(4, sizeof(struct SQueryNodeAddr));
-        code = getPlan(pRequest, pQuery, &pRequest->body.pDag, pNodeList);
+        SArray* pNodeList = NULL;
+        code = getPlan(pRequest, pQuery, &pRequest->body.pDag, &pNodeList);
         if (TSDB_CODE_SUCCESS == code) {
           code = scheduleQuery(pRequest, pRequest->body.pDag, pNodeList, &pRes);
           if (NULL != pRes) {

source/common/src/systable.c

@@ -215,7 +215,6 @@ static const SSysDbTableSchema transSchema[] = {
     {.name = "create_time", .bytes = 8, .type = TSDB_DATA_TYPE_TIMESTAMP},
     {.name = "stage", .bytes = TSDB_TRANS_STAGE_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
     {.name = "db", .bytes = SYSTABLE_SCH_DB_NAME_LEN, .type = TSDB_DATA_TYPE_VARCHAR},
-    {.name = "type", .bytes = TSDB_TRANS_TYPE_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
     {.name = "failed_times", .bytes = 4, .type = TSDB_DATA_TYPE_INT},
     {.name = "last_exec_time", .bytes = 8, .type = TSDB_DATA_TYPE_TIMESTAMP},
     {.name = "last_error", .bytes = (TSDB_TRANS_ERROR_LEN - 1) + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},

source/common/src/tmsg.c

@@ -147,12 +147,25 @@ int32_t tEncodeSQueryNodeAddr(SEncoder *pEncoder, SQueryNodeAddr *pAddr) {
   return 0;
 }
 
+int32_t tEncodeSQueryNodeLoad(SEncoder *pEncoder, SQueryNodeLoad *pLoad) {
+  if (tEncodeSQueryNodeAddr(pEncoder, &pLoad->addr) < 0) return -1;
+  if (tEncodeU64(pEncoder, pLoad->load) < 0) return -1;
+  return 0;
+}
+
 int32_t tDecodeSQueryNodeAddr(SDecoder *pDecoder, SQueryNodeAddr *pAddr) {
   if (tDecodeI32(pDecoder, &pAddr->nodeId) < 0) return -1;
   if (tDecodeSEpSet(pDecoder, &pAddr->epSet) < 0) return -1;
   return 0;
 }
 
+int32_t tDecodeSQueryNodeLoad(SDecoder *pDecoder, SQueryNodeLoad *pLoad) {
+  if (tDecodeSQueryNodeAddr(pDecoder, &pLoad->addr) < 0) return -1;
+  if (tDecodeU64(pDecoder, &pLoad->load) < 0) return -1;
+  return 0;
+}
+
+
 int32_t taosEncodeSEpSet(void **buf, const SEpSet *pEp) {
   int32_t tlen = 0;
   tlen += taosEncodeFixedI8(buf, pEp->inUse);
@@ -304,6 +317,12 @@ static int32_t tSerializeSClientHbRsp(SEncoder *pEncoder, const SClientHbRsp *pR
     if (tEncodeI32(pEncoder, pRsp->query->onlineDnodes) < 0) return -1;
     if (tEncodeI8(pEncoder, pRsp->query->killConnection) < 0) return -1;
     if (tEncodeSEpSet(pEncoder, &pRsp->query->epSet) < 0) return -1;
+    int32_t num = taosArrayGetSize(pRsp->query->pQnodeList);
+    if (tEncodeI32(pEncoder, num) < 0) return -1;
+    for (int32_t i = 0; i < num; ++i) {
+      SQueryNodeLoad *pLoad = taosArrayGet(pRsp->query->pQnodeList, i);
+      if (tEncodeSQueryNodeLoad(pEncoder, pLoad) < 0) return -1;
+    }
   } else {
     if (tEncodeI32(pEncoder, queryNum) < 0) return -1;
   }
@@ -333,6 +352,15 @@ static int32_t tDeserializeSClientHbRsp(SDecoder *pDecoder, SClientHbRsp *pRsp)
     if (tDecodeI32(pDecoder, &pRsp->query->onlineDnodes) < 0) return -1;
     if (tDecodeI8(pDecoder, &pRsp->query->killConnection) < 0) return -1;
     if (tDecodeSEpSet(pDecoder, &pRsp->query->epSet) < 0) return -1;
+    int32_t pQnodeNum = 0;
+    if (tDecodeI32(pDecoder, &pQnodeNum) < 0) return -1;
+    if (pQnodeNum > 0) {
+      pRsp->query->pQnodeList = taosArrayInit(pQnodeNum, sizeof(SQueryNodeLoad));
+      if (NULL == pRsp->query->pQnodeList) return -1;
+      SQueryNodeLoad load = {0};
+      if (tDecodeSQueryNodeLoad(pDecoder, &load) < 0) return -1;
+      taosArrayPush(pRsp->query->pQnodeList, &load);
+    }
   }
 
   int32_t kvNum = 0;
@@ -898,6 +926,18 @@ int32_t tSerializeSStatusReq(void *buf, int32_t bufLen, SStatusReq *pReq) {
   // mnode loads
   if (tEncodeI32(&encoder, pReq->mload.syncState) < 0) return -1;
 
+  if (tEncodeI32(&encoder, pReq->qload.dnodeId) < 0) return -1;
+  if (tEncodeI64(&encoder, pReq->qload.numOfProcessedQuery) < 0) return -1;
+  if (tEncodeI64(&encoder, pReq->qload.numOfProcessedCQuery) < 0) return -1;
+  if (tEncodeI64(&encoder, pReq->qload.numOfProcessedFetch) < 0) return -1;
+  if (tEncodeI64(&encoder, pReq->qload.numOfProcessedDrop) < 0) return -1;
+  if (tEncodeI64(&encoder, pReq->qload.numOfProcessedHb) < 0) return -1;
+  if (tEncodeI64(&encoder, pReq->qload.cacheDataSize) < 0) return -1;
+  if (tEncodeI64(&encoder, pReq->qload.numOfQueryInQueue) < 0) return -1;
+  if (tEncodeI64(&encoder, pReq->qload.numOfFetchInQueue) < 0) return -1;
+  if (tEncodeI64(&encoder, pReq->qload.timeInQueryQueue) < 0) return -1;
+  if (tEncodeI64(&encoder, pReq->qload.timeInFetchQueue) < 0) return -1;
+
   tEndEncode(&encoder);
 
   int32_t tlen = encoder.pos;
@@ -955,6 +995,18 @@ int32_t tDeserializeSStatusReq(void *buf, int32_t bufLen, SStatusReq *pReq) {
 
   if (tDecodeI32(&decoder, &pReq->mload.syncState) < 0) return -1;
 
+  if (tDecodeI32(&decoder, &pReq->qload.dnodeId) < 0) return -1;
+  if (tDecodeI64(&decoder, &pReq->qload.numOfProcessedQuery) < 0) return -1;
+  if (tDecodeI64(&decoder, &pReq->qload.numOfProcessedCQuery) < 0) return -1;
+  if (tDecodeI64(&decoder, &pReq->qload.numOfProcessedFetch) < 0) return -1;
+  if (tDecodeI64(&decoder, &pReq->qload.numOfProcessedDrop) < 0) return -1;
+  if (tDecodeI64(&decoder, &pReq->qload.numOfProcessedHb) < 0) return -1;
+  if (tDecodeI64(&decoder, &pReq->qload.cacheDataSize) < 0) return -1;
+  if (tDecodeI64(&decoder, &pReq->qload.numOfQueryInQueue) < 0) return -1;
+  if (tDecodeI64(&decoder, &pReq->qload.numOfFetchInQueue) < 0) return -1;
+  if (tDecodeI64(&decoder, &pReq->qload.timeInQueryQueue) < 0) return -1;
+  if (tDecodeI64(&decoder, &pReq->qload.timeInFetchQueue) < 0) return -1;
+
   tEndDecode(&decoder);
   tDecoderClear(&decoder);
   return 0;
@@ -1921,11 +1973,11 @@ int32_t tSerializeSQnodeListRsp(void *buf, int32_t bufLen, SQnodeListRsp *pRsp)
   tEncoderInit(&encoder, buf, bufLen);
 
   if (tStartEncode(&encoder) < 0) return -1;
-  int32_t num = taosArrayGetSize(pRsp->addrsList);
+  int32_t num = taosArrayGetSize(pRsp->qnodeList);
   if (tEncodeI32(&encoder, num) < 0) return -1;
   for (int32_t i = 0; i < num; ++i) {
-    SQueryNodeAddr *addr = taosArrayGet(pRsp->addrsList, i);
-    if (tEncodeSQueryNodeAddr(&encoder, addr) < 0) return -1;
+    SQueryNodeLoad *pLoad = taosArrayGet(pRsp->qnodeList, i);
+    if (tEncodeSQueryNodeLoad(&encoder, pLoad) < 0) return -1;
   }
   tEndEncode(&encoder);
 
@@ -1941,15 +1993,15 @@ int32_t tDeserializeSQnodeListRsp(void *buf, int32_t bufLen, SQnodeListRsp *pRsp
   if (tStartDecode(&decoder) < 0) return -1;
   int32_t num = 0;
   if (tDecodeI32(&decoder, &num) < 0) return -1;
-  if (NULL == pRsp->addrsList) {
-    pRsp->addrsList = taosArrayInit(num, sizeof(SQueryNodeAddr));
-    if (NULL == pRsp->addrsList) return -1;
+  if (NULL == pRsp->qnodeList) {
+    pRsp->qnodeList = taosArrayInit(num, sizeof(SQueryNodeLoad));
+    if (NULL == pRsp->qnodeList) return -1;
   }
 
   for (int32_t i = 0; i < num; ++i) {
-    SQueryNodeAddr addr = {0};
-    if (tDecodeSQueryNodeAddr(&decoder, &addr) < 0) return -1;
-    taosArrayPush(pRsp->addrsList, &addr);
+    SQueryNodeLoad load = {0};
+    if (tDecodeSQueryNodeLoad(&decoder, &load) < 0) return -1;
+    taosArrayPush(pRsp->qnodeList, &load);
   }
   tEndDecode(&decoder);
 
@@ -1957,7 +2009,7 @@ int32_t tDeserializeSQnodeListRsp(void *buf, int32_t bufLen, SQnodeListRsp *pRsp
   return 0;
 }
 
-void tFreeSQnodeListRsp(SQnodeListRsp *pRsp) { taosArrayDestroy(pRsp->addrsList); }
+void tFreeSQnodeListRsp(SQnodeListRsp *pRsp) { taosArrayDestroy(pRsp->qnodeList); }
 
 int32_t tSerializeSCompactDbReq(void *buf, int32_t bufLen, SCompactDbReq *pReq) {
   SEncoder encoder = {0};
@@ -2921,6 +2973,11 @@ int32_t tSerializeSCreateVnodeReq(void *buf, int32_t bufLen, SCreateVnodeReq *pR
   }
 
   if (tEncodeI8(&encoder, pReq->isTsma) < 0) return -1;
+  if (pReq->isTsma) {
+    uint32_t tsmaLen = (uint32_t)(htonl(((SMsgHead *)pReq->pTsma)->contLen));
+    if (tEncodeBinary(&encoder, (const uint8_t *)pReq->pTsma, tsmaLen) < 0) return -1;
+  }
+
   tEndEncode(&encoder);
 
   int32_t tlen = encoder.pos;
@@ -2984,6 +3041,9 @@ int32_t tDeserializeSCreateVnodeReq(void *buf, int32_t bufLen, SCreateVnodeReq *
   }
 
   if (tDecodeI8(&decoder, &pReq->isTsma) < 0) return -1;
+  if (pReq->isTsma) {
+    if (tDecodeBinaryAlloc(&decoder, &pReq->pTsma, NULL) < 0) return -1;
+  }
 
   tEndDecode(&decoder);
   tDecoderClear(&decoder);
@@ -2993,6 +3053,9 @@ int32_t tDeserializeSCreateVnodeReq(void *buf, int32_t bufLen, SCreateVnodeReq *
 int32_t tFreeSCreateVnodeReq(SCreateVnodeReq *pReq) {
   taosArrayDestroy(pReq->pRetensions);
   pReq->pRetensions = NULL;
+  if(pReq->isTsma) {
+    taosMemoryFreeClear(pReq->pTsma);
+  }
   return 0;
 }
 

source/dnode/mgmt/mgmt_dnode/inc/dmInt.h

@@ -35,6 +35,7 @@ typedef struct SDnodeMgmt {
   SendMonitorReportFp sendMonitorReportFp;
   GetVnodeLoadsFp     getVnodeLoadsFp;
   GetMnodeLoadsFp     getMnodeLoadsFp;
+  GetQnodeLoadsFp     getQnodeLoadsFp;
 } SDnodeMgmt;
 
 // dmHandle.c

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

@@ -79,6 +79,8 @@ void dmSendStatusReq(SDnodeMgmt *pMgmt) {
   (*pMgmt->getMnodeLoadsFp)(&minfo);
   req.mload = minfo.load;
 
+  (*pMgmt->getQnodeLoadsFp)(&req.qload);
+
   int32_t contLen = tSerializeSStatusReq(NULL, 0, &req);
   void   *pHead = rpcMallocCont(contLen);
   tSerializeSStatusReq(pHead, contLen, &req);

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

@@ -48,6 +48,7 @@ static int32_t dmOpenMgmt(SMgmtInputOpt *pInput, SMgmtOutputOpt *pOutput) {
   pMgmt->sendMonitorReportFp = pInput->sendMonitorReportFp;
   pMgmt->getVnodeLoadsFp = pInput->getVnodeLoadsFp;
   pMgmt->getMnodeLoadsFp = pInput->getMnodeLoadsFp;
+  pMgmt->getQnodeLoadsFp = pInput->getQnodeLoadsFp;
 
   if (dmStartWorker(pMgmt) != 0) {
     return -1;

source/dnode/mgmt/mgmt_qnode/src/qmHandle.c

@@ -20,6 +20,14 @@ void qmGetMonitorInfo(SQnodeMgmt *pMgmt, SMonQmInfo *qmInfo) {
   SQnodeLoad qload = {0};
   qndGetLoad(pMgmt->pQnode, &qload);
 
+  qload.dnodeId = pMgmt->pData->dnodeId;
+
+}
+
+void qmGetQnodeLoads(SQnodeMgmt *pMgmt, SQnodeLoad *pInfo) {
+  qndGetLoad(pMgmt->pQnode, pInfo);
+
+  pInfo->dnodeId = pMgmt->pData->dnodeId;
 }
 
 int32_t qmProcessGetMonitorInfoReq(SQnodeMgmt *pMgmt, SRpcMsg *pMsg) {

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

@@ -140,6 +140,7 @@ static void vmGenerateVnodeCfg(SCreateVnodeReq *pCreate, SVnodeCfg *pCfg) {
   pCfg->szCache = pCreate->pages;
   pCfg->szBuf = (uint64_t)pCreate->buffer * 1024 * 1024;
   pCfg->isWeak = true;
+  pCfg->isTsma = pCreate->isTsma;
   pCfg->tsdbCfg.compression = pCreate->compression;
   pCfg->tsdbCfg.precision = pCreate->precision;
   pCfg->tsdbCfg.days = pCreate->daysPerFile;
@@ -209,7 +210,7 @@ int32_t vmProcessCreateVnodeReq(SVnodeMgmt *pMgmt, SRpcMsg *pMsg) {
 
   SVnode *pImpl = vnodeOpen(path, pMgmt->pTfs, pMgmt->msgCb);
   if (pImpl == NULL) {
-    dError("vgId:%d, failed to create vnode since %s", createReq.vgId, terrstr());
+    dError("vgId:%d, failed to open vnode since %s", createReq.vgId, terrstr());
     code = terrno;
     goto _OVER;
   }

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

@@ -104,7 +104,7 @@ static void vmProcessWriteQueue(SQueueInfo *pInfo, STaosQall *qall, int32_t numO
 
     dTrace("msg:%p, get from vnode-write queue", pMsg);
     if (taosArrayPush(pArray, &pMsg) == NULL) {
-      dTrace("msg:%p, failed to process since %s", pMsg, terrstr());
+      dTrace("msg:%p, failed to push to array since %s", pMsg, terrstr());
       vmSendRsp(pMsg, TSDB_CODE_OUT_OF_MEMORY);
     }
   }

source/dnode/mgmt/node_mgmt/inc/dmMgmt.h

@@ -168,6 +168,7 @@ int32_t dmProcessNodeMsg(SMgmtWrapper *pWrapper, SRpcMsg *pMsg);
 void dmSendMonitorReport();
 void dmGetVnodeLoads(SMonVloadInfo *pInfo);
 void dmGetMnodeLoads(SMonMloadInfo *pInfo);
+void dmGetQnodeLoads(SQnodeLoad *pInfo);
 
 #ifdef __cplusplus
 }

source/dnode/mgmt/node_mgmt/inc/dmNodes.h

@@ -37,6 +37,7 @@ void bmGetMonitorInfo(void *pMgmt, SMonBmInfo *pInfo);
 
 void vmGetVnodeLoads(void *pMgmt, SMonVloadInfo *pInfo);
 void mmGetMnodeLoads(void *pMgmt, SMonMloadInfo *pInfo);
+void qmGetQnodeLoads(void *pMgmt, SQnodeLoad *pInfo);
 
 #ifdef __cplusplus
 }

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

@@ -178,6 +178,7 @@ SMgmtInputOpt dmBuildMgmtInputOpt(SMgmtWrapper *pWrapper) {
       .sendMonitorReportFp = dmSendMonitorReport,
       .getVnodeLoadsFp = dmGetVnodeLoads,
       .getMnodeLoadsFp = dmGetMnodeLoads,
+      .getQnodeLoadsFp = dmGetQnodeLoads,
   };
 
   opt.msgCb = dmGetMsgcb(pWrapper->pDnode);

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

@@ -170,3 +170,17 @@ void dmGetMnodeLoads(SMonMloadInfo *pInfo) {
     dmReleaseWrapper(pWrapper);
   }
 }
+
+void dmGetQnodeLoads(SQnodeLoad *pInfo) {
+  SDnode       *pDnode = dmInstance();
+  SMgmtWrapper *pWrapper = &pDnode->wrappers[QNODE];
+  if (dmMarkWrapper(pWrapper) == 0) {
+    if (tsMultiProcess) {
+      dmSendLocalRecv(pDnode, TDMT_MON_QM_LOAD, tDeserializeSQnodeLoad, pInfo);
+    } else if (pWrapper->pMgmt != NULL) {
+      qmGetQnodeLoads(pWrapper->pMgmt, pInfo);
+    }
+    dmReleaseWrapper(pWrapper);
+  }
+}
+

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

@@ -34,6 +34,7 @@
 
 #include "dnode.h"
 #include "mnode.h"
+#include "qnode.h"
 #include "monitor.h"
 #include "sync.h"
 #include "wal.h"
@@ -92,6 +93,7 @@ typedef int32_t (*ProcessDropNodeFp)(EDndNodeType ntype, SRpcMsg *pMsg);
 typedef void (*SendMonitorReportFp)();
 typedef void (*GetVnodeLoadsFp)(SMonVloadInfo *pInfo);
 typedef void (*GetMnodeLoadsFp)(SMonMloadInfo *pInfo);
+typedef void (*GetQnodeLoadsFp)(SQnodeLoad *pInfo);
 
 typedef struct {
   int32_t        dnodeId;
@@ -118,6 +120,7 @@ typedef struct {
   SendMonitorReportFp sendMonitorReportFp;
   GetVnodeLoadsFp     getVnodeLoadsFp;
   GetMnodeLoadsFp     getMnodeLoadsFp;
+  GetQnodeLoadsFp     getQnodeLoadsFp;
 } SMgmtInputOpt;
 
 typedef struct {

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

@@ -54,9 +54,11 @@ typedef enum {
 } EAuthOp;
 
 typedef enum {
-  TRN_STEP_LOG = 1,
-  TRN_STEP_ACTION = 2,
-} ETrnStep;
+  TRN_CONFLICT_NOTHING = 0,
+  TRN_CONFLICT_GLOBAL = 1,
+  TRN_CONFLICT_DB = 2,
+  TRN_CONFLICT_DB_INSIDE = 3,
+} ETrnConflct;
 
 typedef enum {
   TRN_STAGE_PREPARE = 0,
@@ -68,69 +70,15 @@ typedef enum {
   TRN_STAGE_FINISHED = 6
 } ETrnStage;
 
-typedef enum {
-  TRN_TYPE_BASIC_SCOPE = 1000,
-  TRN_TYPE_CREATE_ACCT = 1001,
-  TRN_TYPE_CREATE_CLUSTER = 1002,
-  TRN_TYPE_CREATE_USER = 1003,
-  TRN_TYPE_ALTER_USER = 1004,
-  TRN_TYPE_DROP_USER = 1005,
-  TRN_TYPE_CREATE_FUNC = 1006,
-  TRN_TYPE_DROP_FUNC = 1007,
-
-  TRN_TYPE_CREATE_SNODE = 1010,
-  TRN_TYPE_DROP_SNODE = 1011,
-  TRN_TYPE_CREATE_QNODE = 1012,
-  TRN_TYPE_DROP_QNODE = 10013,
-  TRN_TYPE_CREATE_BNODE = 1014,
-  TRN_TYPE_DROP_BNODE = 1015,
-  TRN_TYPE_CREATE_MNODE = 1016,
-  TRN_TYPE_DROP_MNODE = 1017,
-
-  TRN_TYPE_CREATE_TOPIC = 1020,
-  TRN_TYPE_DROP_TOPIC = 1021,
-  TRN_TYPE_SUBSCRIBE = 1022,
-  TRN_TYPE_REBALANCE = 1023,
-  TRN_TYPE_COMMIT_OFFSET = 1024,
-  TRN_TYPE_CREATE_STREAM = 1025,
-  TRN_TYPE_DROP_STREAM = 1026,
-  TRN_TYPE_ALTER_STREAM = 1027,
-  TRN_TYPE_CONSUMER_LOST = 1028,
-  TRN_TYPE_CONSUMER_RECOVER = 1029,
-  TRN_TYPE_DROP_CGROUP = 1030,
-  TRN_TYPE_BASIC_SCOPE_END,
-
-  TRN_TYPE_GLOBAL_SCOPE = 2000,
-  TRN_TYPE_CREATE_DNODE = 2001,
-  TRN_TYPE_DROP_DNODE = 2002,
-  TRN_TYPE_GLOBAL_SCOPE_END,
-
-  TRN_TYPE_DB_SCOPE = 3000,
-  TRN_TYPE_CREATE_DB = 3001,
-  TRN_TYPE_ALTER_DB = 3002,
-  TRN_TYPE_DROP_DB = 3003,
-  TRN_TYPE_SPLIT_VGROUP = 3004,
-  TRN_TYPE_MERGE_VGROUP = 3015,
-  TRN_TYPE_DB_SCOPE_END,
-
-  TRN_TYPE_STB_SCOPE = 4000,
-  TRN_TYPE_CREATE_STB = 4001,
-  TRN_TYPE_ALTER_STB = 4002,
-  TRN_TYPE_DROP_STB = 4003,
-  TRN_TYPE_CREATE_SMA = 4004,
-  TRN_TYPE_DROP_SMA = 4005,
-  TRN_TYPE_STB_SCOPE_END,
-} ETrnType;
-
 typedef enum {
   TRN_POLICY_ROLLBACK = 0,
   TRN_POLICY_RETRY = 1,
 } ETrnPolicy;
 
 typedef enum {
-  TRN_EXEC_PARALLEL = 0,
-  TRN_EXEC_NO_PARALLEL = 1,
-} ETrnExecType;
+  TRN_EXEC_PRARLLEL = 0,
+  TRN_EXEC_SERIAL = 1,
+} ETrnExec;
 
 typedef enum {
   DND_REASON_ONLINE = 0,
@@ -159,8 +107,8 @@ typedef struct {
   int32_t        id;
   ETrnStage      stage;
   ETrnPolicy     policy;
-  ETrnType       type;
-  ETrnExecType   parallel;
+  ETrnConflct    conflict;
+  ETrnExec       exec;
   int32_t        code;
   int32_t        failedTimes;
   SRpcHandleInfo rpcInfo;
@@ -172,10 +120,11 @@ typedef struct {
   SArray*        commitActions;
   int64_t        createdTime;
   int64_t        lastExecTime;
-  int64_t        dbUid;
+  int32_t        lastErrorAction;
+  int32_t        lastErrorNo;
+  tmsg_t         lastErrorMsgType;
+  SEpSet         lastErrorEpset;
   char           dbname[TSDB_DB_FNAME_LEN];
-  char           lastError[TSDB_TRANS_ERROR_LEN];
-  char           desc[TSDB_TRANS_DESC_LEN];
   int32_t        startFunc;
   int32_t        stopFunc;
   int32_t        paramLen;
@@ -219,6 +168,7 @@ typedef struct {
   int64_t    createdTime;
   int64_t    updateTime;
   SDnodeObj* pDnode;
+  SQnodeLoad load;  
 } SQnodeObj;
 
 typedef struct {
@@ -343,6 +293,7 @@ typedef struct {
   int8_t    isTsma;
   int8_t    replica;
   SVnodeGid vnodeGid[TSDB_MAX_REPLICA];
+  void*     pTsma;
 } SVgObj;
 
 typedef struct {

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

@@ -22,9 +22,15 @@
 extern "C" {
 #endif
 
+#define QNODE_LOAD_VALUE(pQnode) (pQnode ? (pQnode->load.numOfQueryInQueue + pQnode->load.numOfFetchInQueue) : 0)
+
 int32_t mndInitQnode(SMnode *pMnode);
 void    mndCleanupQnode(SMnode *pMnode);
 
+SQnodeObj *mndAcquireQnode(SMnode *pMnode, int32_t qnodeId);
+void mndReleaseQnode(SMnode *pMnode, SQnodeObj *pObj);
+int32_t mndCreateQnodeList(SMnode       *pMnode, SArray** pList, int32_t limit);
+
 #ifdef __cplusplus
 }
 #endif

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

@@ -34,7 +34,7 @@ typedef struct {
   int32_t  errCode;
   int32_t  acceptableCode;
   int8_t   stage;
-  int8_t   isRaw;
+  int8_t   actionType; // 0-msg, 1-raw
   int8_t   rawWritten;
   int8_t   msgSent;
   int8_t   msgReceived;
@@ -52,7 +52,7 @@ void    mndCleanupTrans(SMnode *pMnode);
 STrans *mndAcquireTrans(SMnode *pMnode, int32_t transId);
 void    mndReleaseTrans(SMnode *pMnode, STrans *pTrans);
 
-STrans *mndTransCreate(SMnode *pMnode, ETrnPolicy policy, ETrnType type, const SRpcMsg *pReq);
+STrans *mndTransCreate(SMnode *pMnode, ETrnPolicy policy, ETrnConflct conflict, const SRpcMsg *pReq);
 void    mndTransDrop(STrans *pTrans);
 int32_t mndTransAppendRedolog(STrans *pTrans, SSdbRaw *pRaw);
 int32_t mndTransAppendUndolog(STrans *pTrans, SSdbRaw *pRaw);
@@ -62,7 +62,7 @@ int32_t mndTransAppendUndoAction(STrans *pTrans, STransAction *pAction);
 void    mndTransSetRpcRsp(STrans *pTrans, void *pCont, int32_t contLen);
 void    mndTransSetCb(STrans *pTrans, ETrnFunc startFunc, ETrnFunc stopFunc, void *param, int32_t paramLen);
 void    mndTransSetDbInfo(STrans *pTrans, SDbObj *pDb);
-void    mndTransSetNoParallel(STrans *pTrans);
+void    mndTransSetSerial(STrans *pTrans);
 
 int32_t mndTransPrepare(SMnode *pMnode, STrans *pTrans);
 void    mndTransProcessRsp(SRpcMsg *pRsp);

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

@@ -80,7 +80,7 @@ static int32_t mndCreateDefaultAcct(SMnode *pMnode) {
 
   mDebug("acct:%s, will be created when deploying, raw:%p", acctObj.acct, pRaw);
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_CREATE_ACCT, NULL);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_NOTHING, NULL);
   if (pTrans == NULL) {
     mError("acct:%s, failed to create since %s", acctObj.acct, terrstr());
     return -1;

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

@@ -246,7 +246,7 @@ static int32_t mndCreateBnode(SMnode *pMnode, SRpcMsg *pReq, SDnodeObj *pDnode,
   bnodeObj.createdTime = taosGetTimestampMs();
   bnodeObj.updateTime = bnodeObj.createdTime;
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_CREATE_BNODE, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to create bnode:%d", pTrans->id, pCreate->dnodeId);
@@ -363,7 +363,7 @@ static int32_t mndSetDropBnodeRedoActions(STrans *pTrans, SDnodeObj *pDnode, SBn
 static int32_t mndDropBnode(SMnode *pMnode, SRpcMsg *pReq, SBnodeObj *pObj) {
   int32_t code = -1;
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_DROP_BNODE, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to drop bnode:%d", pTrans->id, pObj->id);

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

@@ -179,10 +179,8 @@ static int32_t mndCreateDefaultCluster(SMnode *pMnode) {
   sdbSetRawStatus(pRaw, SDB_STATUS_READY);
 
   mDebug("cluster:%" PRId64 ", will be created when deploying, raw:%p", clusterObj.id, pRaw);
-#if 0
-  return sdbWrite(pMnode->pSdb, pRaw);
-#else
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_CREATE_CLUSTER, NULL);
+
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_NOTHING, NULL);
   if (pTrans == NULL) {
     mError("cluster:%" PRId64 ", failed to create since %s", clusterObj.id, terrstr());
     return -1;
@@ -204,7 +202,6 @@ static int32_t mndCreateDefaultCluster(SMnode *pMnode) {
 
   mndTransDrop(pTrans);
   return 0;
-#endif
 }
 
 static int32_t mndRetrieveClusters(SRpcMsg *pMsg, SShowObj *pShow, SSDataBlock *pBlock, int32_t rows) {

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

@@ -97,7 +97,7 @@ static int32_t mndProcessConsumerLostMsg(SRpcMsg *pMsg) {
 
   mndReleaseConsumer(pMnode, pConsumer);
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_CONSUMER_LOST, pMsg);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_NOTHING, pMsg);
   if (pTrans == NULL) goto FAIL;
   if (mndSetConsumerCommitLogs(pMnode, pTrans, pConsumerNew) != 0) goto FAIL;
   if (mndTransPrepare(pMnode, pTrans) != 0) goto FAIL;
@@ -121,7 +121,7 @@ static int32_t mndProcessConsumerRecoverMsg(SRpcMsg *pMsg) {
 
   mndReleaseConsumer(pMnode, pConsumer);
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_CONSUMER_RECOVER, pMsg);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_NOTHING, pMsg);
   if (pTrans == NULL) goto FAIL;
   if (mndSetConsumerCommitLogs(pMnode, pTrans, pConsumerNew) != 0) goto FAIL;
   if (mndTransPrepare(pMnode, pTrans) != 0) goto FAIL;
@@ -403,7 +403,7 @@ static int32_t mndProcessSubscribeReq(SRpcMsg *pMsg) {
 
   int32_t newTopicNum = taosArrayGetSize(newSub);
   // check topic existance
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_SUBSCRIBE, pMsg);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_NOTHING, pMsg);
   if (pTrans == NULL) goto SUBSCRIBE_OVER;
 
   for (int32_t i = 0; i < newTopicNum; i++) {

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

@@ -545,7 +545,7 @@ static int32_t mndCreateDb(SMnode *pMnode, SRpcMsg *pReq, SCreateDbReq *pCreate,
   }
 
   int32_t code = -1;
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_CREATE_DB, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_DB, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to create db:%s", pTrans->id, pCreate->db);
@@ -775,7 +775,7 @@ static int32_t mndSetAlterDbRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *
 
 static int32_t mndAlterDb(SMnode *pMnode, SRpcMsg *pReq, SDbObj *pOld, SDbObj *pNew) {
   int32_t code = -1;
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_ALTER_DB, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_DB, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to alter db:%s", pTrans->id, pOld->name);
@@ -1036,7 +1036,7 @@ static int32_t mndBuildDropDbRsp(SDbObj *pDb, int32_t *pRspLen, void **ppRsp, bo
 
 static int32_t mndDropDb(SMnode *pMnode, SRpcMsg *pReq, SDbObj *pDb) {
   int32_t code = -1;
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_DROP_DB, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_DB, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to drop db:%s", pTrans->id, pDb->name);

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

@@ -17,6 +17,7 @@
 #include "mndDnode.h"
 #include "mndAuth.h"
 #include "mndMnode.h"
+#include "mndQnode.h"
 #include "mndShow.h"
 #include "mndTrans.h"
 #include "mndUser.h"
@@ -100,10 +101,7 @@ static int32_t mndCreateDefaultDnode(SMnode *pMnode) {
 
   mDebug("dnode:%d, will be created when deploying, raw:%p", dnodeObj.id, pRaw);
 
-#if 0
-  return sdbWrite(pMnode->pSdb, pRaw);
-#else
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_CREATE_DNODE, NULL);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_GLOBAL, NULL);
   if (pTrans == NULL) {
     mError("dnode:%s, failed to create since %s", dnodeObj.ep, terrstr());
     return -1;
@@ -125,7 +123,6 @@ static int32_t mndCreateDefaultDnode(SMnode *pMnode) {
 
   mndTransDrop(pTrans);
   return 0;
-#endif
 }
 
 static SSdbRaw *mndDnodeActionEncode(SDnodeObj *pDnode) {
@@ -388,6 +385,12 @@ static int32_t mndProcessStatusReq(SRpcMsg *pReq) {
     mndReleaseMnode(pMnode, pObj);
   }
 
+  SQnodeObj *pQnode = mndAcquireQnode(pMnode, statusReq.qload.dnodeId);
+  if (pQnode != NULL) {
+    pQnode->load = statusReq.qload;
+    mndReleaseQnode(pMnode, pQnode);
+  }
+
   int64_t dnodeVer = sdbGetTableVer(pMnode->pSdb, SDB_DNODE) + sdbGetTableVer(pMnode->pSdb, SDB_MNODE);
   int64_t curMs = taosGetTimestampMs();
   bool    online = mndIsDnodeOnline(pMnode, pDnode, curMs);
@@ -481,7 +484,7 @@ static int32_t mndCreateDnode(SMnode *pMnode, SRpcMsg *pReq, SCreateDnodeReq *pC
   memcpy(dnodeObj.fqdn, pCreate->fqdn, TSDB_FQDN_LEN);
   snprintf(dnodeObj.ep, TSDB_EP_LEN, "%s:%u", dnodeObj.fqdn, dnodeObj.port);
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_CREATE_DNODE, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_GLOBAL, pReq);
   if (pTrans == NULL) {
     mError("dnode:%s, failed to create since %s", dnodeObj.ep, terrstr());
     return -1;
@@ -557,7 +560,7 @@ CREATE_DNODE_OVER:
 }
 
 static int32_t mndDropDnode(SMnode *pMnode, SRpcMsg *pReq, SDnodeObj *pDnode) {
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_DROP_DNODE, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_GLOBAL, pReq);
   if (pTrans == NULL) {
     mError("dnode:%d, failed to drop since %s", pDnode->id, terrstr());
     return -1;

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

@@ -215,7 +215,7 @@ static int32_t mndCreateFunc(SMnode *pMnode, SRpcMsg *pReq, SCreateFuncReq *pCre
   }
   memcpy(func.pCode, pCreate->pCode, func.codeSize);
 
-  pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_CREATE_FUNC, pReq);
+  pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to create func:%s", pTrans->id, pCreate->name);
@@ -245,7 +245,7 @@ _OVER:
 
 static int32_t mndDropFunc(SMnode *pMnode, SRpcMsg *pReq, SFuncObj *pFunc) {
   int32_t code = -1;
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_DROP_FUNC, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to drop user:%s", pTrans->id, pFunc->name);

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

@@ -18,9 +18,9 @@
 #include "mndAuth.h"
 #include "mndDnode.h"
 #include "mndShow.h"
+#include "mndSync.h"
 #include "mndTrans.h"
 #include "mndUser.h"
-#include "mndSync.h"
 
 #define MNODE_VER_NUMBER   1
 #define MNODE_RESERVE_SIZE 64
@@ -92,10 +92,7 @@ static int32_t mndCreateDefaultMnode(SMnode *pMnode) {
 
   mDebug("mnode:%d, will be created when deploying, raw:%p", mnodeObj.id, pRaw);
 
-#if 0
-  return sdbWrite(pMnode->pSdb, pRaw);
-#else
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_CREATE_DNODE, NULL);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_GLOBAL, NULL);
   if (pTrans == NULL) {
     mError("mnode:%d, failed to create since %s", mnodeObj.id, terrstr());
     return -1;
@@ -117,7 +114,6 @@ static int32_t mndCreateDefaultMnode(SMnode *pMnode) {
 
   mndTransDrop(pTrans);
   return 0;
-#endif
 }
 
 static SSdbRaw *mndMnodeActionEncode(SMnodeObj *pObj) {
@@ -363,11 +359,11 @@ static int32_t mndCreateMnode(SMnode *pMnode, SRpcMsg *pReq, SDnodeObj *pDnode,
   mnodeObj.createdTime = taosGetTimestampMs();
   mnodeObj.updateTime = mnodeObj.createdTime;
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_CREATE_MNODE, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_GLOBAL, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to create mnode:%d", pTrans->id, pCreate->dnodeId);
-  mndTransSetNoParallel(pTrans);
+  mndTransSetSerial(pTrans);
   if (mndSetCreateMnodeRedoLogs(pMnode, pTrans, &mnodeObj) != 0) goto _OVER;
   if (mndSetCreateMnodeCommitLogs(pMnode, pTrans, &mnodeObj) != 0) goto _OVER;
   if (mndSetCreateMnodeRedoActions(pMnode, pTrans, pDnode, &mnodeObj) != 0) goto _OVER;
@@ -396,6 +392,11 @@ static int32_t mndProcessCreateMnodeReq(SRpcMsg *pReq) {
 
   mDebug("mnode:%d, start to create", createReq.dnodeId);
 
+  if (sdbGetSize(pMnode->pSdb, SDB_MNODE) >= 3) {
+    terrno = TSDB_CODE_MND_TOO_MANY_MNODES;
+    goto _OVER;
+  }
+
   pObj = mndAcquireMnode(pMnode, createReq.dnodeId);
   if (pObj != NULL) {
     terrno = TSDB_CODE_MND_MNODE_ALREADY_EXIST;
@@ -535,11 +536,11 @@ static int32_t mndSetDropMnodeRedoActions(SMnode *pMnode, STrans *pTrans, SDnode
 static int32_t mndDropMnode(SMnode *pMnode, SRpcMsg *pReq, SMnodeObj *pObj) {
   int32_t code = -1;
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_DROP_MNODE, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_GLOBAL, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to drop mnode:%d", pTrans->id, pObj->id);
-  mndTransSetNoParallel(pTrans);
+  mndTransSetSerial(pTrans);
   if (mndSetDropMnodeRedoLogs(pMnode, pTrans, pObj) != 0) goto _OVER;
   if (mndSetDropMnodeCommitLogs(pMnode, pTrans, pObj) != 0) goto _OVER;
   if (mndSetDropMnodeRedoActions(pMnode, pTrans, pObj->pDnode, pObj) != 0) goto _OVER;

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

@@ -179,7 +179,7 @@ static int32_t mndProcessCommitOffsetReq(SRpcMsg *pMsg) {
 
   tDecodeSMqCMCommitOffsetReq(&decoder, &commitOffsetReq);
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_COMMIT_OFFSET, pMsg);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_NOTHING, pMsg);
 
   for (int32_t i = 0; i < commitOffsetReq.num; i++) {
     SMqOffset *pOffset = &commitOffsetReq.offsets[i];

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

@@ -18,6 +18,7 @@
 #include "mndDb.h"
 #include "mndDnode.h"
 #include "mndMnode.h"
+#include "mndQnode.h"
 #include "mndShow.h"
 #include "mndStb.h"
 #include "mndUser.h"
@@ -382,6 +383,9 @@ static int32_t mndProcessQueryHeartBeat(SMnode *pMnode, SRpcMsg *pMsg, SClientHb
     rspBasic->totalDnodes = mndGetDnodeSize(pMnode);
     rspBasic->onlineDnodes = 1;  // TODO
     mndGetMnodeEpSet(pMnode, &rspBasic->epSet);
+
+    mndCreateQnodeList(pMnode, &rspBasic->pQnodeList, -1);
+    
     mndReleaseConn(pMnode, pConn);
 
     hbRsp.query = rspBasic;

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

@@ -60,7 +60,7 @@ int32_t mndInitQnode(SMnode *pMnode) {
 
 void mndCleanupQnode(SMnode *pMnode) {}
 
-static SQnodeObj *mndAcquireQnode(SMnode *pMnode, int32_t qnodeId) {
+SQnodeObj *mndAcquireQnode(SMnode *pMnode, int32_t qnodeId) {
   SQnodeObj *pObj = sdbAcquire(pMnode->pSdb, SDB_QNODE, &qnodeId);
   if (pObj == NULL && terrno == TSDB_CODE_SDB_OBJ_NOT_THERE) {
     terrno = TSDB_CODE_MND_QNODE_NOT_EXIST;
@@ -68,7 +68,7 @@ static SQnodeObj *mndAcquireQnode(SMnode *pMnode, int32_t qnodeId) {
   return pObj;
 }
 
-static void mndReleaseQnode(SMnode *pMnode, SQnodeObj *pObj) {
+void mndReleaseQnode(SMnode *pMnode, SQnodeObj *pObj) {
   SSdb *pSdb = pMnode->pSdb;
   sdbRelease(pSdb, pObj);
 }
@@ -248,7 +248,7 @@ static int32_t mndCreateQnode(SMnode *pMnode, SRpcMsg *pReq, SDnodeObj *pDnode,
   qnodeObj.createdTime = taosGetTimestampMs();
   qnodeObj.updateTime = qnodeObj.createdTime;
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_CREATE_QNODE, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to create qnode:%d", pTrans->id, pCreate->dnodeId);
@@ -365,7 +365,7 @@ static int32_t mndSetDropQnodeRedoActions(STrans *pTrans, SDnodeObj *pDnode, SQn
 static int32_t mndDropQnode(SMnode *pMnode, SRpcMsg *pReq, SQnodeObj *pObj) {
   int32_t code = -1;
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_DROP_QNODE, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to drop qnode:%d", pTrans->id, pObj->id);
@@ -429,49 +429,62 @@ _OVER:
   return code;
 }
 
-static int32_t mndProcessQnodeListReq(SRpcMsg *pReq) {
-  int32_t       code = -1;
-  int32_t       numOfRows = 0;
-  SMnode       *pMnode = pReq->info.node;
+int32_t mndCreateQnodeList(SMnode       *pMnode, SArray** pList, int32_t limit) {
   SSdb *pSdb = pMnode->pSdb;
+  void *pIter = NULL;
   SQnodeObj    *pObj = NULL;
-  SQnodeListReq qlistReq = {0};
-  SQnodeListRsp qlistRsp = {0};
-
-  if (tDeserializeSQnodeListReq(pReq->pCont, pReq->contLen, &qlistReq) != 0) {
-    mError("failed to parse qnode list req");
-    terrno = TSDB_CODE_INVALID_MSG;
-    goto _OVER;
-  }
+  int32_t       numOfRows = 0;
 
-  qlistRsp.addrsList = taosArrayInit(5, sizeof(SQueryNodeAddr));
-  if (NULL == qlistRsp.addrsList) {
+  SArray* qnodeList = taosArrayInit(5, sizeof(SQueryNodeLoad));
+  if (NULL == qnodeList) {
     mError("failed to alloc epSet while process qnode list req");
     terrno = TSDB_CODE_OUT_OF_MEMORY;
-    goto _OVER;
+    return terrno;
   }
   
-  void *pIter = NULL;
   while (1) {
     pIter = sdbFetch(pSdb, SDB_QNODE, pIter, (void **)&pObj);
     if (pIter == NULL) break;
 
-    SQueryNodeAddr nodeAddr = {0};
-    nodeAddr.nodeId = QNODE_HANDLE;
-    nodeAddr.epSet.numOfEps = 1;
-    tstrncpy(nodeAddr.epSet.eps[0].fqdn, pObj->pDnode->fqdn, TSDB_FQDN_LEN);
-    nodeAddr.epSet.eps[0].port = pObj->pDnode->port;
+    SQueryNodeLoad nodeLoad = {0};
+    nodeLoad.addr.nodeId = QNODE_HANDLE;
+    nodeLoad.addr.epSet.numOfEps = 1;
+    tstrncpy(nodeLoad.addr.epSet.eps[0].fqdn, pObj->pDnode->fqdn, TSDB_FQDN_LEN);
+    nodeLoad.addr.epSet.eps[0].port = pObj->pDnode->port;
+    nodeLoad.load = QNODE_LOAD_VALUE(pObj);
 
-    (void)taosArrayPush(qlistRsp.addrsList, &nodeAddr);
+    (void)taosArrayPush(qnodeList, &nodeLoad);
 
     numOfRows++;
     sdbRelease(pSdb, pObj);
 
-    if (qlistReq.rowNum > 0 && numOfRows >= qlistReq.rowNum) {
+    if (limit > 0 && numOfRows >= limit) {
       break;
     }
   }
 
+  *pList = qnodeList;
+
+  return TSDB_CODE_SUCCESS;
+}
+
+
+static int32_t mndProcessQnodeListReq(SRpcMsg *pReq) {
+  int32_t       code = -1;
+  SMnode       *pMnode = pReq->info.node;
+  SQnodeListReq qlistReq = {0};
+  SQnodeListRsp qlistRsp = {0};
+
+  if (tDeserializeSQnodeListReq(pReq->pCont, pReq->contLen, &qlistReq) != 0) {
+    mError("failed to parse qnode list req");
+    terrno = TSDB_CODE_INVALID_MSG;
+    goto _OVER;
+  }
+
+  if (mndCreateQnodeList(pMnode, &qlistRsp.qnodeList, qlistReq.rowNum) != 0) {
+    goto _OVER;
+  }
+
   int32_t rspLen = tSerializeSQnodeListRsp(NULL, 0, &qlistRsp);
   void   *pRsp = rpcMallocCont(rspLen);
   if (pRsp == NULL) {

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

@@ -409,7 +409,8 @@ static int32_t mndSetCreateSmaRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj
   return 0;
 }
 
-static int32_t mndSetCreateSmaVgroupRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup) {
+static int32_t mndSetCreateSmaVgroupRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup,
+                                                SSmaObj *pSma) {
   SVnodeGid *pVgid = pVgroup->vnodeGid + 0;
   SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);
   if (pDnode == NULL) return -1;
@@ -419,9 +420,14 @@ static int32_t mndSetCreateSmaVgroupRedoActions(SMnode *pMnode, STrans *pTrans,
   mndReleaseDnode(pMnode, pDnode);
 
   // todo add sma info here
+  int32_t smaContLen = 0;
+  void   *pSmaReq = mndBuildVCreateSmaReq(pMnode, pVgroup, pSma, &smaContLen);
+  if (pSmaReq == NULL) return -1;
+  pVgroup->pTsma = pSmaReq;
 
   int32_t contLen = 0;
   void   *pReq = mndBuildCreateVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen);
+  taosMemoryFreeClear(pSmaReq);
   if (pReq == NULL) return -1;
 
   action.pCont = pReq;
@@ -502,19 +508,19 @@ static int32_t mndCreateSma(SMnode *pMnode, SRpcMsg *pReq, SMCreateSmaReq *pCrea
   streamObj.fixedSinkVgId = smaObj.dstVgId;
 
   int32_t code = -1;
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_CREATE_SMA, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_DB, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to create sma:%s", pTrans->id, pCreate->name);
   mndTransSetDbInfo(pTrans, pDb);
-  mndTransSetNoParallel(pTrans);
+  mndTransSetSerial(pTrans);
 
   if (mndSetCreateSmaRedoLogs(pMnode, pTrans, &smaObj) != 0) goto _OVER;
   if (mndSetCreateSmaVgroupRedoLogs(pMnode, pTrans, &streamObj.fixedSinkVg) != 0) goto _OVER;
   if (mndSetCreateSmaCommitLogs(pMnode, pTrans, &smaObj) != 0) goto _OVER;
   if (mndSetCreateSmaVgroupCommitLogs(pMnode, pTrans, &streamObj.fixedSinkVg) != 0) goto _OVER;
   if (mndSetCreateSmaRedoActions(pMnode, pTrans, pDb, &smaObj) != 0) goto _OVER;
-  if (mndSetCreateSmaVgroupRedoActions(pMnode, pTrans, pDb, &streamObj.fixedSinkVg) != 0) goto _OVER;
+  if (mndSetCreateSmaVgroupRedoActions(pMnode, pTrans, pDb, &streamObj.fixedSinkVg, &smaObj) != 0) goto _OVER;
   if (mndAddStreamToTrans(pMnode, &streamObj, pCreate->ast, STREAM_TRIGGER_AT_ONCE, 0, pTrans) != 0) goto _OVER;
   if (mndTransPrepare(pMnode, pTrans) != 0) goto _OVER;
 
@@ -747,7 +753,7 @@ static int32_t mndDropSma(SMnode *pMnode, SRpcMsg *pReq, SDbObj *pDb, SSmaObj *p
   pVgroup = mndAcquireVgroup(pMnode, pSma->dstVgId);
   if (pVgroup == NULL) goto _OVER;
 
-  pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_DROP_SMA, pReq);
+  pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_DB, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to drop sma:%s", pTrans->id, pSma->name);

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

@@ -253,7 +253,7 @@ static int32_t mndCreateSnode(SMnode *pMnode, SRpcMsg *pReq, SDnodeObj *pDnode,
   snodeObj.createdTime = taosGetTimestampMs();
   snodeObj.updateTime = snodeObj.createdTime;
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_CREATE_SNODE, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to create snode:%d", pTrans->id, pCreate->dnodeId);
@@ -372,7 +372,7 @@ static int32_t mndSetDropSnodeRedoActions(STrans *pTrans, SDnodeObj *pDnode, SSn
 static int32_t mndDropSnode(SMnode *pMnode, SRpcMsg *pReq, SSnodeObj *pObj) {
   int32_t code = -1;
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_DROP_SNODE, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to drop snode:%d", pTrans->id, pObj->id);

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

@@ -735,7 +735,7 @@ static int32_t mndCreateStb(SMnode *pMnode, SRpcMsg *pReq, SMCreateStbReq *pCrea
 
   int32_t code = -1;
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_CREATE_STB, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_DB_INSIDE, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to create stb:%s", pTrans->id, pCreate->name);
@@ -1257,7 +1257,7 @@ static int32_t mndAlterStb(SMnode *pMnode, SRpcMsg *pReq, const SMAlterStbReq *p
   if (code != 0) goto _OVER;
 
   code = -1;
-  pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_ALTER_STB, pReq);
+  pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_DB_INSIDE, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to alter stb:%s", pTrans->id, pAlter->name);
@@ -1403,7 +1403,7 @@ static int32_t mndSetDropStbRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *
 
 static int32_t mndDropStb(SMnode *pMnode, SRpcMsg *pReq, SDbObj *pDb, SStbObj *pStb) {
   int32_t code = -1;
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_DROP_STB, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_DB_INSIDE, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to drop stb:%s", pTrans->id, pStb->name);

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

@@ -402,7 +402,7 @@ static int32_t mndCreateStream(SMnode *pMnode, SRpcMsg *pReq, SCMCreateStreamReq
     tstrncpy(streamObj.targetDb, pDb->name, TSDB_DB_FNAME_LEN);
   }
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_CREATE_STREAM, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) {
     mError("stream:%s, failed to create since %s", pCreate->name, terrstr());
     return -1;

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

@@ -394,8 +394,8 @@ static int32_t mndDoRebalance(SMnode *pMnode, const SMqRebInputObj *pInput, SMqR
   mInfo("rebalance calculation completed, rebalanced vg:");
   for (int32_t i = 0; i < taosArrayGetSize(pOutput->rebVgs); i++) {
     SMqRebOutputVg *pOutputRebVg = taosArrayGet(pOutput->rebVgs, i);
-    mInfo("vg: %d moved from consumer %ld to consumer %ld", pOutputRebVg->pVgEp->vgId, pOutputRebVg->oldConsumerId,
-          pOutputRebVg->newConsumerId);
+    mInfo("vgId:%d moved from consumer %" PRId64 " to consumer %" PRId64, pOutputRebVg->pVgEp->vgId,
+          pOutputRebVg->oldConsumerId, pOutputRebVg->newConsumerId);
   }
 
   // 9. clear
@@ -405,10 +405,9 @@ static int32_t mndDoRebalance(SMnode *pMnode, const SMqRebInputObj *pInput, SMqR
 }
 
 static int32_t mndPersistRebResult(SMnode *pMnode, SRpcMsg *pMsg, const SMqRebOutputObj *pOutput) {
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_REBALANCE, pMsg);
-  if (pTrans == NULL) {
-    return -1;
-  }
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_NOTHING, pMsg);
+  if (pTrans == NULL) return -1;
+
   // make txn:
   // 1. redo action: action to all vg
   const SArray *rebVgs = pOutput->rebVgs;
@@ -625,7 +624,7 @@ static int32_t mndProcessDropCgroupReq(SRpcMsg *pReq) {
     return -1;
   }
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_DROP_CGROUP, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) {
     mError("cgroup: %s on topic:%s, failed to drop since %s", dropReq.cgroup, dropReq.topic, terrstr());
     mndReleaseSubscribe(pMnode, pSub);

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

@@ -383,7 +383,7 @@ static int32_t mndCreateTopic(SMnode *pMnode, SRpcMsg *pReq, SCMCreateTopicReq *
     /*topicObj.withSchema = 1;*/
   }
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_CREATE_TOPIC, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) {
     mError("topic:%s, failed to create since %s", pCreate->name, terrstr());
     taosMemoryFreeClear(topicObj.ast);
@@ -551,7 +551,7 @@ static int32_t mndProcessDropTopicReq(SRpcMsg *pReq) {
   }
 #endif
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_DROP_TOPIC, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) {
     mError("topic:%s, failed to drop since %s", pTopic->name, terrstr());
     return -1;

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

@@ -79,10 +79,7 @@ static int32_t mndCreateDefaultUser(SMnode *pMnode, char *acct, char *user, char
 
   mDebug("user:%s, will be created when deploying, raw:%p", userObj.user, pRaw);
 
-#if 0
-  return sdbWrite(pMnode->pSdb, pRaw);
-#else
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_CREATE_USER, NULL);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_NOTHING, NULL);
   if (pTrans == NULL) {
     mError("user:%s, failed to create since %s", userObj.user, terrstr());
     return -1;
@@ -104,7 +101,6 @@ static int32_t mndCreateDefaultUser(SMnode *pMnode, char *acct, char *user, char
 
   mndTransDrop(pTrans);
   return 0;
-#endif
 }
 
 static int32_t mndCreateDefaultUsers(SMnode *pMnode) {
@@ -291,7 +287,7 @@ static int32_t mndCreateUser(SMnode *pMnode, char *acct, SCreateUserReq *pCreate
   userObj.updateTime = userObj.createdTime;
   userObj.superUser = pCreate->superUser;
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_CREATE_USER, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) {
     mError("user:%s, failed to create since %s", pCreate->user, terrstr());
     return -1;
@@ -371,7 +367,7 @@ _OVER:
 }
 
 static int32_t mndAlterUser(SMnode *pMnode, SUserObj *pOld, SUserObj *pNew, SRpcMsg *pReq) {
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_ALTER_USER, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) {
     mError("user:%s, failed to alter since %s", pOld->user, terrstr());
     return -1;
@@ -578,7 +574,7 @@ _OVER:
 }
 
 static int32_t mndDropUser(SMnode *pMnode, SRpcMsg *pReq, SUserObj *pUser) {
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_DROP_USER, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) {
     mError("user:%s, failed to drop since %s", pUser->user, terrstr());
     return -1;

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

@@ -218,6 +218,8 @@ void *mndBuildCreateVnodeReq(SMnode *pMnode, SDnodeObj *pDnode, SDbObj *pDb, SVg
   createReq.hashMethod = pDb->cfg.hashMethod;
   createReq.numOfRetensions = pDb->cfg.numOfRetensions;
   createReq.pRetensions = pDb->cfg.pRetensions;
+  createReq.isTsma = pVgroup->isTsma;
+  createReq.pTsma = pVgroup->pTsma;
 
   for (int32_t v = 0; v < pVgroup->replica; ++v) {
     SReplica  *pReplica = &createReq.replicas[v];

source/dnode/mnode/impl/test/trans/trans2.cpp

@@ -11,6 +11,8 @@
 
 #include <gtest/gtest.h>
 
+#if 0
+
 #include "mndTrans.h"
 #include "mndUser.h"
 #include "tcache.h"
@@ -103,7 +105,7 @@ class MndTestTrans2 : public ::testing::Test {
   void SetUp() override {}
   void TearDown() override {}
 
-  int32_t CreateUserLog(const char *acct, const char *user, ETrnType type, SDbObj *pDb) {
+  int32_t CreateUserLog(const char *acct, const char *user, ETrnConflct conflict, SDbObj *pDb) {
     SUserObj userObj = {0};
     taosEncryptPass_c((uint8_t *)"taosdata", strlen("taosdata"), userObj.pass);
     tstrncpy(userObj.user, user, TSDB_USER_LEN);
@@ -113,7 +115,7 @@ class MndTestTrans2 : public ::testing::Test {
     userObj.superUser = 1;
 
     SRpcMsg  rpcMsg = {0};
-    STrans  *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, type, &rpcMsg);
+    STrans  *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, conflict, &rpcMsg);
     SSdbRaw *pRedoRaw = mndUserActionEncode(&userObj);
     mndTransAppendRedolog(pTrans, pRedoRaw);
     sdbSetRawStatus(pRedoRaw, SDB_STATUS_READY);
@@ -135,7 +137,7 @@ class MndTestTrans2 : public ::testing::Test {
     return code;
   }
 
-  int32_t CreateUserAction(const char *acct, const char *user, bool hasUndoAction, ETrnPolicy policy, ETrnType type,
+  int32_t CreateUserAction(const char *acct, const char *user, bool hasUndoAction, ETrnPolicy policy, ETrnConflct conflict,
                            SDbObj *pDb) {
     SUserObj userObj = {0};
     taosEncryptPass_c((uint8_t *)"taosdata", strlen("taosdata"), userObj.pass);
@@ -146,7 +148,7 @@ class MndTestTrans2 : public ::testing::Test {
     userObj.superUser = 1;
 
     SRpcMsg  rpcMsg = {0};
-    STrans  *pTrans = mndTransCreate(pMnode, policy, type, &rpcMsg);
+    STrans  *pTrans = mndTransCreate(pMnode, policy, conflict, &rpcMsg);
     SSdbRaw *pRedoRaw = mndUserActionEncode(&userObj);
     mndTransAppendRedolog(pTrans, pRedoRaw);
     sdbSetRawStatus(pRedoRaw, SDB_STATUS_READY);
@@ -218,7 +220,7 @@ class MndTestTrans2 : public ::testing::Test {
     userObj.superUser = 1;
 
     SRpcMsg  rpcMsg = {0};
-    STrans  *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_CREATE_USER, &rpcMsg);
+    STrans  *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_NOTHING, &rpcMsg);
     SSdbRaw *pRedoRaw = mndUserActionEncode(&userObj);
     mndTransAppendRedolog(pTrans, pRedoRaw);
     sdbSetRawStatus(pRedoRaw, SDB_STATUS_READY);
@@ -528,3 +530,5 @@ TEST_F(MndTestTrans2, 04_Conflict) {
     mndReleaseUser(pMnode, pUser);
   }
 }
+
+#endif
\ No newline at end of file

source/dnode/mnode/impl/test/user/CMakeLists.txt

@@ -5,7 +5,9 @@ target_link_libraries(
     PUBLIC sut
 )
 
-add_test(
+if(NOT TD_WINDOWS)
+    add_test(
         NAME userTest
         COMMAND userTest
-)
+    )
+endif(NOT TD_WINDOWS)

source/dnode/qnode/src/qnode.c

@@ -41,12 +41,24 @@ void qndClose(SQnode *pQnode) {
 }
 
 int32_t qndGetLoad(SQnode *pQnode, SQnodeLoad *pLoad) { 
-  SMsgCb* pCb = &pQnode->msgCb;
+  SReadHandle handle = {.pMsgCb = &pQnode->msgCb};
+  SQWorkerStat stat = {0};
+
+  int32_t code = qWorkerGetStat(&handle, pQnode->pQuery, &stat);
+  if (code) {
+    return code;
+  }
 
-  pLoad->numOfQueryInQueue = pCb->qsizeFp(pCb->mgmt, pQnode->qndId, QUERY_QUEUE);
-  pLoad->numOfFetchInQueue = pCb->qsizeFp(pCb->mgmt, pQnode->qndId, FETCH_QUEUE);
-  pLoad->waitTimeInQueryQUeue = qWorkerGetWaitTimeInQueue(pQnode->pQuery, QUERY_QUEUE);
-  pLoad->waitTimeInFetchQUeue = qWorkerGetWaitTimeInQueue(pQnode->pQuery, FETCH_QUEUE);
+  pLoad->numOfQueryInQueue = stat.numOfQueryInQueue;
+  pLoad->numOfFetchInQueue = stat.numOfFetchInQueue;
+  pLoad->timeInQueryQueue = stat.timeInQueryQueue;
+  pLoad->timeInFetchQueue = stat.timeInFetchQueue;
+  pLoad->cacheDataSize = stat.cacheDataSize;
+  pLoad->numOfProcessedQuery = stat.queryProcessed;
+  pLoad->numOfProcessedCQuery = stat.cqueryProcessed;
+  pLoad->numOfProcessedFetch = stat.fetchProcessed;
+  pLoad->numOfProcessedDrop = stat.dropProcessed;
+  pLoad->numOfProcessedHb = stat.hbProcessed;
   
   return 0; 
 }

source/dnode/vnode/inc/vnode.h

@@ -80,6 +80,18 @@ int32_t     metaGetTableEntryByUid(SMetaReader *pReader, tb_uid_t uid);
 int32_t     metaReadNext(SMetaReader *pReader);
 const void *metaGetTableTagVal(SMetaEntry *pEntry, int16_t cid);
 
+typedef struct SMetaFltParam {
+  tb_uid_t suid;
+  int16_t  cid;
+  int16_t  type;
+  char *   val;
+  bool     reverse;
+  int (*filterFunc)(void *a, void *b, int16_t type);
+
+} SMetaFltParam;
+
+int32_t metaFilteTableIds(SMeta *pMeta, SMetaFltParam *param, SArray *results);
+
 #if 1  // refact APIs below (TODO)
 typedef SVCreateTbReq   STbCfg;
 typedef SVCreateTSmaReq SSmaCfg;
@@ -159,12 +171,13 @@ struct SVnodeCfg {
   uint64_t szBuf;
   bool     isHeap;
   bool     isWeak;
+  int8_t   isTsma;
+  int8_t   hashMethod;
   STsdbCfg tsdbCfg;
   SWalCfg  walCfg;
   SSyncCfg syncCfg;
   uint32_t hashBegin;
   uint32_t hashEnd;
-  int8_t   hashMethod;
 };
 
 typedef struct {

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

@@ -16,8 +16,8 @@
 #ifndef _TD_VNODE_META_H_
 #define _TD_VNODE_META_H_
 
-#include "vnodeInt.h"
 #include "index.h"
+#include "vnodeInt.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -45,8 +45,6 @@ int32_t metaULock(SMeta* pMeta);
 int metaEncodeEntry(SEncoder* pCoder, const SMetaEntry* pME);
 int metaDecodeEntry(SDecoder* pCoder, SMetaEntry* pME);
 
-// metaTable ==================
-
 // metaQuery ==================
 int metaGetTableEntryByVersion(SMetaReader* pReader, int64_t version, tb_uid_t uid);
 
@@ -118,6 +116,10 @@ typedef struct {
   int64_t  smaUid;
 } SSmaIdxKey;
 
+// metaTable ==================
+int metaCreateTagIdxKey(tb_uid_t suid, int32_t cid, const void* pTagData, int8_t type, tb_uid_t uid,
+                        STagIdxKey** ppTagIdxKey, int32_t* nTagIdxKey);
+
 #ifndef META_REFACT
 // SMetaDB
 int  metaOpenDB(SMeta* pMeta);

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

@@ -31,7 +31,7 @@ void metaReaderClear(SMetaReader *pReader) {
 }
 
 int metaGetTableEntryByVersion(SMetaReader *pReader, int64_t version, tb_uid_t uid) {
-  SMeta   *pMeta = pReader->pMeta;
+  SMeta *  pMeta = pReader->pMeta;
   STbDbKey tbDbKey = {.version = version, .uid = uid};
 
   // query table.db
@@ -54,7 +54,7 @@ _err:
 }
 
 int metaGetTableEntryByUid(SMetaReader *pReader, tb_uid_t uid) {
-  SMeta  *pMeta = pReader->pMeta;
+  SMeta * pMeta = pReader->pMeta;
   int64_t version;
 
   // query uid.idx
@@ -68,7 +68,7 @@ int metaGetTableEntryByUid(SMetaReader *pReader, tb_uid_t uid) {
 }
 
 int metaGetTableEntryByName(SMetaReader *pReader, const char *name) {
-  SMeta   *pMeta = pReader->pMeta;
+  SMeta *  pMeta = pReader->pMeta;
   tb_uid_t uid;
 
   // query name.idx
@@ -82,7 +82,7 @@ int metaGetTableEntryByName(SMetaReader *pReader, const char *name) {
 }
 
 tb_uid_t metaGetTableEntryUidByName(SMeta *pMeta, const char *name) {
-  void    *pData = NULL;
+  void *   pData = NULL;
   int      nData = 0;
   tb_uid_t uid = 0;
 
@@ -134,7 +134,7 @@ void metaCloseTbCursor(SMTbCursor *pTbCur) {
 
 int metaTbCursorNext(SMTbCursor *pTbCur) {
   int    ret;
-  void  *pBuf;
+  void * pBuf;
   STbCfg tbCfg;
 
   for (;;) {
@@ -155,7 +155,7 @@ int metaTbCursorNext(SMTbCursor *pTbCur) {
 }
 
 SSchemaWrapper *metaGetTableSchema(SMeta *pMeta, tb_uid_t uid, int32_t sver, bool isinline) {
-  void           *pData = NULL;
+  void *          pData = NULL;
   int             nData = 0;
   int64_t         version;
   SSchemaWrapper  schema = {0};
@@ -205,11 +205,11 @@ _err:
 }
 
 struct SMCtbCursor {
-  SMeta   *pMeta;
-  TBC     *pCur;
+  SMeta *  pMeta;
+  TBC *    pCur;
   tb_uid_t suid;
-  void    *pKey;
-  void    *pVal;
+  void *   pKey;
+  void *   pVal;
   int      kLen;
   int      vLen;
 };
@@ -281,10 +281,10 @@ tb_uid_t metaCtbCursorNext(SMCtbCursor *pCtbCur) {
 STSchema *metaGetTbTSchema(SMeta *pMeta, tb_uid_t uid, int32_t sver) {
   tb_uid_t        quid;
   SMetaReader     mr = {0};
-  STSchema       *pTSchema = NULL;
+  STSchema *      pTSchema = NULL;
   SSchemaWrapper *pSW = NULL;
   STSchemaBuilder sb = {0};
-  SSchema        *pSchema;
+  SSchema *       pSchema;
 
   metaReaderInit(&mr, pMeta, 0);
   metaGetTableEntryByUid(&mr, uid);
@@ -321,11 +321,11 @@ int metaGetTbNum(SMeta *pMeta) {
 }
 
 typedef struct {
-  SMeta   *pMeta;
-  TBC     *pCur;
+  SMeta *  pMeta;
+  TBC *    pCur;
   tb_uid_t uid;
-  void    *pKey;
-  void    *pVal;
+  void *   pKey;
+  void *   pVal;
   int      kLen;
   int      vLen;
 } SMSmaCursor;
@@ -397,7 +397,7 @@ tb_uid_t metaSmaCursorNext(SMSmaCursor *pSmaCur) {
 
 STSmaWrapper *metaGetSmaInfoByTable(SMeta *pMeta, tb_uid_t uid, bool deepCopy) {
   STSmaWrapper *pSW = NULL;
-  SArray       *pSmaIds = NULL;
+  SArray *      pSmaIds = NULL;
 
   if (!(pSmaIds = metaGetSmaIdsByTable(pMeta, uid))) {
     return NULL;
@@ -421,7 +421,7 @@ STSmaWrapper *metaGetSmaInfoByTable(SMeta *pMeta, tb_uid_t uid, bool deepCopy) {
   metaReaderInit(&mr, pMeta, 0);
   int64_t smaId;
   int     smaIdx = 0;
-  STSma  *pTSma = NULL;
+  STSma * pTSma = NULL;
   for (int i = 0; i < pSW->number; ++i) {
     smaId = *(tb_uid_t *)taosArrayGet(pSmaIds, i);
     if (metaGetTableEntryByUid(&mr, smaId) < 0) {
@@ -469,7 +469,7 @@ _err:
 }
 
 STSma *metaGetSmaInfoByIndex(SMeta *pMeta, int64_t indexUid) {
-  STSma      *pTSma = NULL;
+  STSma *     pTSma = NULL;
   SMetaReader mr = {0};
   metaReaderInit(&mr, pMeta, 0);
   if (metaGetTableEntryByUid(&mr, indexUid) < 0) {
@@ -491,7 +491,7 @@ STSma *metaGetSmaInfoByIndex(SMeta *pMeta, int64_t indexUid) {
 }
 
 SArray *metaGetSmaIdsByTable(SMeta *pMeta, tb_uid_t uid) {
-  SArray     *pUids = NULL;
+  SArray *    pUids = NULL;
   SSmaIdxKey *pSmaIdxKey = NULL;
 
   SMSmaCursor *pCur = metaOpenSmaCursor(pMeta, uid);
@@ -529,7 +529,7 @@ SArray *metaGetSmaIdsByTable(SMeta *pMeta, tb_uid_t uid) {
 }
 
 SArray *metaGetSmaTbUids(SMeta *pMeta) {
-  SArray     *pUids = NULL;
+  SArray *    pUids = NULL;
   SSmaIdxKey *pSmaIdxKey = NULL;
   tb_uid_t    lastUid = 0;
 
@@ -577,3 +577,78 @@ const void *metaGetTableTagVal(SMetaEntry *pEntry, int16_t cid) {
   ASSERT(pEntry->type == TSDB_CHILD_TABLE);
   return tdGetKVRowValOfCol((const SKVRow)pEntry->ctbEntry.pTags, cid);
 }
+typedef struct {
+  SMeta *  pMeta;
+  TBC *    pCur;
+  tb_uid_t suid;
+  int16_t  cid;
+  int16_t  type;
+  void *   pKey;
+  void *   pVal;
+  int32_t  kLen;
+  int32_t  vLen;
+} SIdxCursor;
+
+int32_t metaFilteTableIds(SMeta *pMeta, SMetaFltParam *param, SArray *pUids) {
+  SIdxCursor *pCursor = NULL;
+
+  char *tagData = param->val;
+
+  int32_t ret = 0, valid = 0;
+  pCursor = (SIdxCursor *)taosMemoryCalloc(1, sizeof(SIdxCursor));
+  pCursor->pMeta = pMeta;
+  pCursor->suid = param->suid;
+  pCursor->cid = param->cid;
+  pCursor->type = param->type;
+
+  metaRLock(pMeta);
+  ret = tdbTbcOpen(pMeta->pTagIdx, &pCursor->pCur, NULL);
+  if (ret < 0) {
+    goto END;
+  }
+  STagIdxKey *pKey = NULL;
+  int32_t     nKey = 0;
+
+  ret = metaCreateTagIdxKey(pCursor->suid, pCursor->cid, param->val, pCursor->type,
+                            param->reverse ? INT64_MAX : INT64_MIN, &pKey, &nKey);
+  if (ret != 0) {
+    goto END;
+  }
+  int cmp = 0;
+  if (tdbTbcMoveTo(pCursor->pCur, pKey, nKey, &cmp) < 0) {
+    goto END;
+  }
+  void *  entryKey = NULL, *entryVal = NULL;
+  int32_t nEntryKey, nEntryVal;
+  while (1) {
+    valid = tdbTbcGet(pCursor->pCur, (const void **)&entryKey, &nEntryKey, (const void **)&entryVal, &nEntryVal);
+    if (valid < 0) {
+      break;
+    }
+    STagIdxKey *p = entryKey;
+    if (p != NULL) {
+      int32_t cmp = (*param->filterFunc)(p->data, pKey->data, pKey->type);
+      if (cmp == 0) {
+        // match
+        tb_uid_t tuid = *(tb_uid_t *)(p->data + tDataTypes[pCursor->type].bytes);
+        taosArrayPush(pUids, &tuid);
+      } else if (cmp == 1) {
+        // not match but should continue to iter
+      } else {
+        // not match and no more result
+        break;
+      }
+    }
+    valid = param->reverse ? tdbTbcMoveToPrev(pCursor->pCur) : tdbTbcMoveToNext(pCursor->pCur);
+    if (valid < 0) {
+      break;
+    }
+  }
+END:
+  if (pCursor->pMeta) metaULock(pCursor->pMeta);
+  if (pCursor->pCur) tdbTbcClose(pCursor->pCur);
+
+  taosMemoryFree(pCursor);
+
+  return ret;
+}

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

@@ -721,7 +721,7 @@ static int metaUpdateCtbIdx(SMeta *pMeta, const SMetaEntry *pME) {
   return tdbTbInsert(pMeta->pCtbIdx, &ctbIdxKey, sizeof(ctbIdxKey), NULL, 0, &pMeta->txn);
 }
 
-static int metaCreateTagIdxKey(tb_uid_t suid, int32_t cid, const void *pTagData, int8_t type, tb_uid_t uid,
+int metaCreateTagIdxKey(tb_uid_t suid, int32_t cid, const void *pTagData, int8_t type, tb_uid_t uid,
                         STagIdxKey **ppTagIdxKey, int32_t *nTagIdxKey) {
   int32_t nTagData = 0;
 

source/dnode/vnode/src/sma/smaRollup.c

@@ -414,7 +414,7 @@ static FORCE_INLINE int32_t tdExecuteRSmaImpl(SSma *pSma, const void *pMsg, int3
     }
     taosMemoryFreeClear(pReq);
   } else {
-    smaWarn("vgId:%d no rsma % " PRIi8 " data generated since %s", SMA_VID(pSma), level, tstrerror(terrno));
+    smaDebug("vgId:%d no rsma % " PRIi8 " data generated since %s", SMA_VID(pSma), level, tstrerror(terrno));
   }
 
   taosArrayDestroy(pResult);

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

@@ -56,6 +56,7 @@ int vnodeEncodeConfig(const void *pObj, SJson *pJson) {
   if (tjsonAddIntegerToObject(pJson, "szBuf", pCfg->szBuf) < 0) return -1;
   if (tjsonAddIntegerToObject(pJson, "isHeap", pCfg->isHeap) < 0) return -1;
   if (tjsonAddIntegerToObject(pJson, "isWeak", pCfg->isWeak) < 0) return -1;
+  if (tjsonAddIntegerToObject(pJson, "isTsma", pCfg->isTsma) < 0) return -1;
   if (tjsonAddIntegerToObject(pJson, "precision", pCfg->tsdbCfg.precision) < 0) return -1;
   if (tjsonAddIntegerToObject(pJson, "update", pCfg->tsdbCfg.update) < 0) return -1;
   if (tjsonAddIntegerToObject(pJson, "compression", pCfg->tsdbCfg.compression) < 0) return -1;
@@ -130,6 +131,8 @@ int vnodeDecodeConfig(const SJson *pJson, void *pObj) {
   if(code < 0) return -1;
   tjsonGetNumberValue(pJson, "isWeak", pCfg->isWeak, code);
   if(code < 0) return -1;
+  tjsonGetNumberValue(pJson, "isTsma", pCfg->isTsma, code);
+  if(code < 0) return -1;
   tjsonGetNumberValue(pJson, "precision", pCfg->tsdbCfg.precision, code);
   if(code < 0) return -1;
   tjsonGetNumberValue(pJson, "update", pCfg->tsdbCfg.update, code);

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

@@ -97,7 +97,7 @@ SVnode *vnodeOpen(const char *path, STfs *pTfs, SMsgCb msgCb) {
   }
 
   // open tsdb
-  if (!vnodeIsRollup(pVnode) && tsdbOpen(pVnode, &VND_TSDB(pVnode), VNODE_TSDB_DIR, TSDB_TYPE_TSDB) < 0) {
+  if (!vnodeIsRollup(pVnode) && tsdbOpen(pVnode, &VND_TSDB(pVnode), VNODE_TSDB_DIR, NULL) < 0) {
     vError("vgId:%d failed to open vnode tsdb since %s", TD_VID(pVnode), tstrerror(terrno));
     goto _err;
   }

source/libs/catalog/inc/catalogInt.h

@@ -342,16 +342,16 @@ typedef struct SCtgOperation {
   ctgOpFunc func;
 } SCtgOperation;
 
-#define CTG_QUEUE_ADD() atomic_add_fetch_64(&gCtgMgmt.queue.qRemainNum, 1)
-#define CTG_QUEUE_SUB() atomic_sub_fetch_64(&gCtgMgmt.queue.qRemainNum, 1)
+#define CTG_QUEUE_INC() atomic_add_fetch_64(&gCtgMgmt.queue.qRemainNum, 1)
+#define CTG_QUEUE_DEC() atomic_sub_fetch_64(&gCtgMgmt.queue.qRemainNum, 1)
 
-#define CTG_STAT_ADD(_item, _n) atomic_add_fetch_64(&(_item), _n)
-#define CTG_STAT_SUB(_item, _n) atomic_sub_fetch_64(&(_item), _n)
+#define CTG_STAT_INC(_item, _n) atomic_add_fetch_64(&(_item), _n)
+#define CTG_STAT_DEC(_item, _n) atomic_sub_fetch_64(&(_item), _n)
 #define CTG_STAT_GET(_item) atomic_load_64(&(_item))
 
-#define CTG_RUNTIME_STAT_ADD(item, n) (CTG_STAT_ADD(gCtgMgmt.stat.runtime.item, n))
-#define CTG_CACHE_STAT_ADD(item, n) (CTG_STAT_ADD(gCtgMgmt.stat.cache.item, n))
-#define CTG_CACHE_STAT_SUB(item, n) (CTG_STAT_SUB(gCtgMgmt.stat.cache.item, n))
+#define CTG_RT_STAT_INC(item, n) (CTG_STAT_INC(gCtgMgmt.stat.runtime.item, n))
+#define CTG_CACHE_STAT_INC(item, n) (CTG_STAT_INC(gCtgMgmt.stat.cache.item, n))
+#define CTG_CACHE_STAT_DEC(item, n) (CTG_STAT_DEC(gCtgMgmt.stat.cache.item, n))
 
 #define CTG_IS_META_NULL(type) ((type) == META_TYPE_NULL_TABLE)
 #define CTG_IS_META_CTABLE(type) ((type) == META_TYPE_CTABLE)

source/libs/catalog/src/catalog.c

@@ -558,7 +558,7 @@ int32_t catalogGetHandle(uint64_t clusterId, SCatalog** catalogHandle) {
 
   *catalogHandle = clusterCtg;
 
-  CTG_CACHE_STAT_ADD(clusterNum, 1);
+  CTG_CACHE_STAT_INC(clusterNum, 1);
   
   return TSDB_CODE_SUCCESS;
 
@@ -579,7 +579,7 @@ void catalogFreeHandle(SCatalog* pCtg) {
     return;
   }
 
-  CTG_CACHE_STAT_SUB(clusterNum, 1);
+  CTG_CACHE_STAT_DEC(clusterNum, 1);
 
   uint64_t clusterId = pCtg->clusterId;
   
@@ -990,7 +990,7 @@ int32_t catalogGetAllMeta(SCatalog* pCtg, void *pTrans, const SEpSet* pMgmtEps,
   }
 
   if (pReq->qNodeRequired) {
-    pRsp->pQnodeList = taosArrayInit(10, sizeof(SQueryNodeAddr));
+    pRsp->pQnodeList = taosArrayInit(10, sizeof(SQueryNodeLoad));
     CTG_ERR_JRET(ctgGetQnodeListFromMnode(CTG_PARAMS_LIST(), pRsp->pQnodeList, NULL));
   }
 

source/libs/catalog/src/ctgCache.c

@@ -178,7 +178,7 @@ int32_t ctgAcquireVgInfoFromCache(SCatalog* pCtg, const char *dbFName, SCtgDBCac
 
   *pCache = dbCache;
 
-  CTG_CACHE_STAT_ADD(vgHitNum, 1);
+  CTG_CACHE_STAT_INC(vgHitNum, 1);
 
   ctgDebug("Got db vgInfo from cache, dbFName:%s", dbFName);
   
@@ -192,7 +192,7 @@ _return:
 
   *pCache = NULL;
 
-  CTG_CACHE_STAT_ADD(vgMissNum, 1);
+  CTG_CACHE_STAT_INC(vgMissNum, 1);
   
   return TSDB_CODE_SUCCESS;
 }
@@ -279,7 +279,7 @@ int32_t ctgReadTbMetaFromCache(SCatalog* pCtg, SCtgTbMetaCtx* ctx, STableMeta**
     ctgReleaseDBCache(pCtg, dbCache);
     ctgDebug("Got meta from cache, type:%d, dbFName:%s, tbName:%s", tbMeta->tableType, dbFName, ctx->pName->tname);
 
-    CTG_CACHE_STAT_ADD(tblHitNum, 1);
+    CTG_CACHE_STAT_INC(tblHitNum, 1);
     return TSDB_CODE_SUCCESS;
   }
   
@@ -312,7 +312,7 @@ int32_t ctgReadTbMetaFromCache(SCatalog* pCtg, SCtgTbMetaCtx* ctx, STableMeta**
 
   ctgReleaseDBCache(pCtg, dbCache);
 
-  CTG_CACHE_STAT_ADD(tblHitNum, 1);
+  CTG_CACHE_STAT_INC(tblHitNum, 1);
 
   ctgDebug("Got tbmeta from cache, dbFName:%s, tbName:%s", dbFName, ctx->pName->tname);
   
@@ -323,7 +323,7 @@ _return:
   ctgReleaseDBCache(pCtg, dbCache);
   taosMemoryFreeClear(*pTableMeta);
 
-  CTG_CACHE_STAT_ADD(tblMissNum, 1);
+  CTG_CACHE_STAT_INC(tblMissNum, 1);
   
   CTG_RET(code);
 }
@@ -462,7 +462,7 @@ int32_t ctgChkAuthFromCache(SCatalog* pCtg, const char* user, const char* dbFNam
   *inCache = true;
 
   ctgDebug("Got user from cache, user:%s", user);
-  CTG_CACHE_STAT_ADD(userHitNum, 1);
+  CTG_CACHE_STAT_INC(userHitNum, 1);
   
   if (pUser->superUser) {
     *pass = true;
@@ -491,7 +491,7 @@ int32_t ctgChkAuthFromCache(SCatalog* pCtg, const char* user, const char* dbFNam
 _return:
 
   *inCache = false;
-  CTG_CACHE_STAT_ADD(userMissNum, 1);
+  CTG_CACHE_STAT_INC(userMissNum, 1);
   
   return TSDB_CODE_SUCCESS;
 }
@@ -521,7 +521,7 @@ void ctgDequeue(SCtgCacheOperation **op) {
   SCtgQNode *node = gCtgMgmt.queue.head->next;
   gCtgMgmt.queue.head = gCtgMgmt.queue.head->next;
 
-  CTG_QUEUE_SUB();
+  CTG_QUEUE_DEC();
   
   taosMemoryFreeClear(orig);
 
@@ -545,8 +545,8 @@ int32_t ctgEnqueue(SCatalog* pCtg, SCtgCacheOperation *operation) {
   gCtgMgmt.queue.tail = node;
   CTG_UNLOCK(CTG_WRITE, &gCtgMgmt.queue.qlock);
 
-  CTG_QUEUE_ADD();
-  CTG_RUNTIME_STAT_ADD(qNum, 1);
+  CTG_QUEUE_INC();
+  CTG_RT_STAT_INC(qNum, 1);
 
   tsem_post(&gCtgMgmt.queue.reqSem);
 
@@ -988,7 +988,7 @@ int32_t ctgAddNewDBCache(SCatalog *pCtg, const char *dbFName, uint64_t dbId) {
     CTG_ERR_JRET(TSDB_CODE_CTG_MEM_ERROR);
   }
 
-  CTG_CACHE_STAT_ADD(dbNum, 1);
+  CTG_CACHE_STAT_INC(dbNum, 1);
  
   SDbVgVersion vgVersion = {.dbId = newDBCache.dbId, .vgVersion = -1};
   strncpy(vgVersion.dbFName, dbFName, sizeof(vgVersion.dbFName));
@@ -1048,7 +1048,7 @@ int32_t ctgRemoveDBFromCache(SCatalog* pCtg, SCtgDBCache *dbCache, const char* d
     CTG_ERR_RET(TSDB_CODE_CTG_DB_DROPPED);
   }
 
-  CTG_CACHE_STAT_SUB(dbNum, 1);
+  CTG_CACHE_STAT_DEC(dbNum, 1);
 
   ctgInfo("db removed from cache, dbFName:%s, dbId:%"PRIx64, dbFName, dbId);
   
@@ -1187,7 +1187,7 @@ int32_t ctgWriteTbMetaToCache(SCatalog *pCtg, SCtgDBCache *dbCache, char *dbFNam
       if (taosHashRemove(tbCache->stbCache, &orig->suid, sizeof(orig->suid))) {
         ctgError("stb not exist in stbCache, dbFName:%s, stb:%s, suid:%"PRIx64, dbFName, tbName, orig->suid);
       } else {
-        CTG_CACHE_STAT_SUB(stblNum, 1);
+        CTG_CACHE_STAT_DEC(stblNum, 1);
       }
       CTG_UNLOCK(CTG_WRITE, &tbCache->stbLock);
 
@@ -1214,7 +1214,7 @@ int32_t ctgWriteTbMetaToCache(SCatalog *pCtg, SCtgDBCache *dbCache, char *dbFNam
   }
 
   if (NULL == orig) {
-    CTG_CACHE_STAT_ADD(tblNum, 1);
+    CTG_CACHE_STAT_INC(tblNum, 1);
   }
 
   ctgDebug("tbmeta updated to cache, dbFName:%s, tbName:%s, tbType:%d", dbFName, tbName, meta->tableType);
@@ -1233,7 +1233,7 @@ int32_t ctgWriteTbMetaToCache(SCatalog *pCtg, SCtgDBCache *dbCache, char *dbFNam
     CTG_ERR_RET(TSDB_CODE_CTG_MEM_ERROR);
   }
 
-  CTG_CACHE_STAT_ADD(stblNum, 1);
+  CTG_CACHE_STAT_INC(stblNum, 1);
   
   CTG_UNLOCK(CTG_WRITE, &tbCache->stbLock);
 
@@ -1371,14 +1371,14 @@ int32_t ctgOpDropStbMeta(SCtgCacheOperation *operation) {
   if (taosHashRemove(dbCache->tbCache.stbCache, &msg->suid, sizeof(msg->suid))) {
     ctgDebug("stb not exist in stbCache, may be removed, dbFName:%s, stb:%s, suid:%"PRIx64, msg->dbFName, msg->stbName, msg->suid);
   } else {
-    CTG_CACHE_STAT_SUB(stblNum, 1);
+    CTG_CACHE_STAT_DEC(stblNum, 1);
   }
 
   CTG_LOCK(CTG_READ, &dbCache->tbCache.metaLock);
   if (taosHashRemove(dbCache->tbCache.metaCache, msg->stbName, strlen(msg->stbName))) {  
     ctgError("stb not exist in cache, dbFName:%s, stb:%s, suid:%"PRIx64, msg->dbFName, msg->stbName, msg->suid);
   } else {
-    CTG_CACHE_STAT_SUB(tblNum, 1);
+    CTG_CACHE_STAT_DEC(tblNum, 1);
   }
   CTG_UNLOCK(CTG_READ, &dbCache->tbCache.metaLock);
   
@@ -1419,7 +1419,7 @@ int32_t ctgOpDropTbMeta(SCtgCacheOperation *operation) {
     ctgError("stb not exist in cache, dbFName:%s, tbName:%s", msg->dbFName, msg->tbName);
     CTG_ERR_RET(TSDB_CODE_CTG_INTERNAL_ERROR);
   } else {
-    CTG_CACHE_STAT_SUB(tblNum, 1);
+    CTG_CACHE_STAT_DEC(tblNum, 1);
   }
   CTG_UNLOCK(CTG_READ, &dbCache->tbCache.metaLock);
 
@@ -1578,7 +1578,7 @@ void* ctgUpdateThreadFunc(void* param) {
       tsem_post(&gCtgMgmt.queue.rspSem);
     }
 
-    CTG_RUNTIME_STAT_ADD(qDoneNum, 1); 
+    CTG_RT_STAT_INC(qDoneNum, 1); 
 
     ctgdShowClusterCache(pCtg);
   }

source/libs/catalog/src/ctgRemote.c

@@ -275,7 +275,7 @@ int32_t ctgGetQnodeListFromMnode(CTG_PARAMS, SArray *out, SCtgTask* pTask) {
   }
 
   if (pTask) {
-    void* pOut = taosArrayInit(4, sizeof(struct SQueryNodeAddr));
+    void* pOut = taosArrayInit(4, sizeof(SQueryNodeLoad));
     if (NULL == pOut) {
       CTG_ERR_RET(TSDB_CODE_OUT_OF_MEMORY);
     }

source/libs/catalog/src/ctgUtil.c

@@ -85,7 +85,7 @@ void ctgFreeTbMetaCache(SCtgTbMetaCache *cache) {
     int32_t stblNum = taosHashGetSize(cache->stbCache);  
     taosHashCleanup(cache->stbCache);
     cache->stbCache = NULL;
-    CTG_CACHE_STAT_SUB(stblNum, stblNum);
+    CTG_CACHE_STAT_DEC(stblNum, stblNum);
   }
   CTG_UNLOCK(CTG_WRITE, &cache->stbLock);
 
@@ -94,7 +94,7 @@ void ctgFreeTbMetaCache(SCtgTbMetaCache *cache) {
     int32_t tblNum = taosHashGetSize(cache->metaCache);
     taosHashCleanup(cache->metaCache);
     cache->metaCache = NULL;
-    CTG_CACHE_STAT_SUB(tblNum, tblNum);
+    CTG_CACHE_STAT_DEC(tblNum, tblNum);
   }
   CTG_UNLOCK(CTG_WRITE, &cache->metaLock);
 }
@@ -145,7 +145,7 @@ void ctgFreeHandle(SCatalog* pCtg) {
 
     taosHashCleanup(pCtg->dbCache);
     
-    CTG_CACHE_STAT_SUB(dbNum, dbNum);
+    CTG_CACHE_STAT_DEC(dbNum, dbNum);
   }
 
   if (pCtg->userCache) {
@@ -162,7 +162,7 @@ void ctgFreeHandle(SCatalog* pCtg) {
 
     taosHashCleanup(pCtg->userCache);
 
-    CTG_CACHE_STAT_SUB(userNum, userNum);
+    CTG_CACHE_STAT_DEC(userNum, userNum);
   }
 
   taosMemoryFree(pCtg);

source/libs/executor/inc/dataSinkInt.h

@@ -37,6 +37,7 @@ typedef void (*FEndPut)(struct SDataSinkHandle* pHandle, uint64_t useconds);
 typedef void (*FGetDataLength)(struct SDataSinkHandle* pHandle, int32_t* pLen, bool* pQueryEnd);
 typedef int32_t (*FGetDataBlock)(struct SDataSinkHandle* pHandle, SOutputData* pOutput);
 typedef int32_t (*FDestroyDataSinker)(struct SDataSinkHandle* pHandle);
+typedef int32_t (*FGetCacheSize)(struct SDataSinkHandle* pHandle, uint64_t* size);
 
 typedef struct SDataSinkHandle {
   FPutDataBlock fPut;
@@ -44,6 +45,7 @@ typedef struct SDataSinkHandle {
   FGetDataLength fGetLen;
   FGetDataBlock fGetData;
   FDestroyDataSinker fDestroy;
+  FGetCacheSize fGetCacheSize;
 } SDataSinkHandle;
 
 int32_t createDataDispatcher(SDataSinkManager* pManager, const SDataSinkNode* pDataSink, DataSinkHandle* pHandle);

source/libs/executor/src/dataDispatcher.c

@@ -22,6 +22,8 @@
 #include "tglobal.h"
 #include "tqueue.h"
 
+extern SDataSinkStat gDataSinkStat;
+
 typedef struct SDataDispatchBuf {
   int32_t useSize;
   int32_t allocSize;
@@ -45,6 +47,7 @@ typedef struct SDataDispatchHandle {
   int32_t             status;
   bool                queryEnd;
   uint64_t            useconds;
+  uint64_t            cachedSize;
   TdThreadMutex       mutex;
 } SDataDispatchHandle;
 
@@ -71,7 +74,7 @@ static bool needCompress(const SSDataBlock* pData, int32_t numOfCols) {
 // +----------------+--------------+----------+--------------------------------------+-------------+-----------+-------------+-----------+
 // The length of bitmap is decided by number of rows of this data block, and the length of each column data is
 // recorded in the first segment, next to the struct header
-static void toDataCacheEntry(const SDataDispatchHandle* pHandle, const SInputData* pInput, SDataDispatchBuf* pBuf) {
+static void toDataCacheEntry(SDataDispatchHandle* pHandle, const SInputData* pInput, SDataDispatchBuf* pBuf) {
   int32_t numOfCols = LIST_LENGTH(pHandle->pSchema->pSlots);
 
   SDataCacheEntry* pEntry = (SDataCacheEntry*)pBuf->pData;
@@ -84,6 +87,9 @@ static void toDataCacheEntry(const SDataDispatchHandle* pHandle, const SInputDat
   blockCompressEncode(pInput->pData, pEntry->data, &pEntry->dataLen, numOfCols, pEntry->compressed);
 
   pBuf->useSize += pEntry->dataLen;
+  
+  atomic_add_fetch_64(&pHandle->cachedSize, pEntry->dataLen); 
+  atomic_add_fetch_64(&gDataSinkStat.cachedSize, pEntry->dataLen); 
 }
 
 static bool allocBuf(SDataDispatchHandle* pDispatcher, const SInputData* pInput, SDataDispatchBuf* pBuf) {
@@ -156,6 +162,7 @@ static void getDataLength(SDataSinkHandle* pHandle, int32_t* pLen, bool* pQueryE
   taosFreeQitem(pBuf);
   *pLen = ((SDataCacheEntry*)(pDispatcher->nextOutput.pData))->dataLen;
   *pQueryEnd = pDispatcher->queryEnd;
+  qDebug("got data len %d, row num %d in sink", *pLen, ((SDataCacheEntry*)(pDispatcher->nextOutput.pData))->numOfRows);
 }
 
 static int32_t getDataBlock(SDataSinkHandle* pHandle, SOutputData* pOutput) {
@@ -173,6 +180,10 @@ static int32_t getDataBlock(SDataSinkHandle* pHandle, SOutputData* pOutput) {
   pOutput->numOfRows = pEntry->numOfRows;
   pOutput->numOfCols = pEntry->numOfCols;
   pOutput->compressed = pEntry->compressed;
+
+  atomic_sub_fetch_64(&pDispatcher->cachedSize, pEntry->dataLen);  
+  atomic_sub_fetch_64(&gDataSinkStat.cachedSize, pEntry->dataLen); 
+
   taosMemoryFreeClear(pDispatcher->nextOutput.pData);  // todo persistent
   pOutput->bufStatus = updateStatus(pDispatcher);
   taosThreadMutexLock(&pDispatcher->mutex);
@@ -180,11 +191,14 @@ static int32_t getDataBlock(SDataSinkHandle* pHandle, SOutputData* pOutput) {
   pOutput->useconds = pDispatcher->useconds;
   pOutput->precision = pDispatcher->pSchema->precision;
   taosThreadMutexUnlock(&pDispatcher->mutex);
+
+  
   return TSDB_CODE_SUCCESS;
 }
 
 static int32_t destroyDataSinker(SDataSinkHandle* pHandle) {
   SDataDispatchHandle* pDispatcher = (SDataDispatchHandle*)pHandle;
+  atomic_sub_fetch_64(&gDataSinkStat.cachedSize, pDispatcher->cachedSize);
   taosMemoryFreeClear(pDispatcher->nextOutput.pData);
   while (!taosQueueEmpty(pDispatcher->pDataBlocks)) {
     SDataDispatchBuf* pBuf = NULL;
@@ -197,6 +211,13 @@ static int32_t destroyDataSinker(SDataSinkHandle* pHandle) {
   return TSDB_CODE_SUCCESS;
 }
 
+int32_t getCacheSize(struct SDataSinkHandle* pHandle, uint64_t* size) {
+  SDataDispatchHandle* pDispatcher = (SDataDispatchHandle*)pHandle;
+
+  *size = atomic_load_64(&pDispatcher->cachedSize);
+  return TSDB_CODE_SUCCESS;
+}
+
 int32_t createDataDispatcher(SDataSinkManager* pManager, const SDataSinkNode* pDataSink, DataSinkHandle* pHandle) {
   SDataDispatchHandle* dispatcher = taosMemoryCalloc(1, sizeof(SDataDispatchHandle));
   if (NULL == dispatcher) {
@@ -208,6 +229,7 @@ int32_t createDataDispatcher(SDataSinkManager* pManager, const SDataSinkNode* pD
   dispatcher->sink.fGetLen = getDataLength;
   dispatcher->sink.fGetData = getDataBlock;
   dispatcher->sink.fDestroy = destroyDataSinker;
+  dispatcher->sink.fGetCacheSize = getCacheSize;
   dispatcher->pManager = pManager;
   dispatcher->pSchema = pDataSink->pInputDataBlockDesc;
   dispatcher->status = DS_BUF_EMPTY;

source/libs/executor/src/dataSinkMgt.c

@@ -19,6 +19,7 @@
 #include "planner.h"
 
 static SDataSinkManager gDataSinkManager = {0};
+SDataSinkStat gDataSinkStat = {0};
 
 int32_t dsDataSinkMgtInit(SDataSinkMgtCfg *cfg) {
   gDataSinkManager.cfg = *cfg;
@@ -26,6 +27,13 @@ int32_t dsDataSinkMgtInit(SDataSinkMgtCfg *cfg) {
   return 0; // to avoid compiler eror
 }
 
+int32_t dsDataSinkGetCacheSize(SDataSinkStat *pStat) {
+  pStat->cachedSize = atomic_load_64(&gDataSinkStat.cachedSize);
+
+  return 0;
+}
+
+
 int32_t dsCreateDataSinker(const SDataSinkNode *pDataSink, DataSinkHandle* pHandle) {
   if (QUERY_NODE_PHYSICAL_PLAN_DISPATCH == nodeType(pDataSink)) {
     return createDataDispatcher(&gDataSinkManager, pDataSink, pHandle);
@@ -53,6 +61,12 @@ int32_t dsGetDataBlock(DataSinkHandle handle, SOutputData* pOutput) {
   return pHandleImpl->fGetData(pHandleImpl, pOutput);
 }
 
+int32_t dsGetCacheSize(DataSinkHandle handle, uint64_t *pSize) {
+  SDataSinkHandle* pHandleImpl = (SDataSinkHandle*)handle;
+  return pHandleImpl->fGetCacheSize(pHandleImpl, pSize);
+}
+
+
 void dsScheduleProcess(void* ahandle, void* pItem) {
   // todo
 }

source/libs/executor/src/executorimpl.c

@@ -4529,6 +4529,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
         qDebug("%s pDataReader is not NULL", GET_TASKID(pTaskInfo));
       }
       SArray*        tableIdList = extractTableIdList(pTableListInfo);
+
       SOperatorInfo* pOperator = createStreamScanOperatorInfo(pDataReader, pHandle,
           tableIdList, pTableScanNode, pTaskInfo, &twSup);
 
@@ -4942,7 +4943,7 @@ int32_t getTableList(void* metaHandle, int32_t tableType, uint64_t tableUid, STa
 
   if (tableType == TSDB_SUPER_TABLE) {
     if (pTagCond) {
-      SIndexMetaArg metaArg = {.metaHandle = tsdbGetIdx(metaHandle), .suid = tableUid};
+      SIndexMetaArg metaArg = {.metaEx = metaHandle, .metaHandle = tsdbGetIdx(metaHandle), .suid = tableUid};
 
       SArray* res = taosArrayInit(8, sizeof(uint64_t));
       code = doFilterTag(pTagCond, &metaArg, res);

source/libs/executor/src/scanoperator.c

@@ -303,7 +303,9 @@ void addTagPseudoColumnData(SReadHandle *pHandle, SExprInfo* pPseudoExpr, int32_
     int32_t dstSlotId = pExpr->base.resSchema.slotId;
 
     SColumnInfoData* pColInfoData = taosArrayGet(pBlock->pDataBlock, dstSlotId);
+    
     colInfoDataEnsureCapacity(pColInfoData, 0, pBlock->info.rows);
+    colInfoDataCleanup(pColInfoData, pBlock->info.rows);
 
     int32_t functionId = pExpr->pExpr->_function.functionId;
 

source/libs/function/src/builtins.c

@@ -183,7 +183,7 @@ static int32_t translatePercentile(SFunctionNode* pFunc, char* pErrBuf, int32_t
   return TSDB_CODE_SUCCESS;
 }
 
-static bool validAperventileAlgo(const SValueNode* pVal) {
+static bool validateApercentileAlgo(const SValueNode* pVal) {
   if (TSDB_DATA_TYPE_BINARY != pVal->node.resType.type) {
     return false;
   }
@@ -231,7 +231,7 @@ static int32_t translateApercentile(SFunctionNode* pFunc, char* pErrBuf, int32_t
     }
 
     SNode* pParamNode2 = nodesListGetNode(pFunc->pParameterList, 2);
-    if (QUERY_NODE_VALUE != nodeType(pParamNode2) || !validAperventileAlgo((SValueNode*)pParamNode2)) {
+    if (QUERY_NODE_VALUE != nodeType(pParamNode2) || !validateApercentileAlgo((SValueNode*)pParamNode2)) {
       return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
                              "Third parameter algorithm of apercentile must be 'default' or 't-digest'");
     }
@@ -438,6 +438,18 @@ static int32_t translateHLL(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
   return TSDB_CODE_SUCCESS;
 }
 
+static bool validateStateOper(const SValueNode* pVal) {
+  if (TSDB_DATA_TYPE_BINARY != pVal->node.resType.type) {
+    return false;
+  }
+  return (0 == strcasecmp(varDataVal(pVal->datum.p), "GT") ||
+          0 == strcasecmp(varDataVal(pVal->datum.p), "GE") ||
+          0 == strcasecmp(varDataVal(pVal->datum.p), "LT") ||
+          0 == strcasecmp(varDataVal(pVal->datum.p), "LE") ||
+          0 == strcasecmp(varDataVal(pVal->datum.p), "EQ") ||
+          0 == strcasecmp(varDataVal(pVal->datum.p), "NE"));
+}
+
 static int32_t translateStateCount(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
   int32_t numOfParams = LIST_LENGTH(pFunc->pParameterList);
   if (3 != numOfParams) {
@@ -464,6 +476,12 @@ static int32_t translateStateCount(SFunctionNode* pFunc, char* pErrBuf, int32_t
 
     SValueNode* pValue = (SValueNode*)pParamNode;
 
+    if (i == 1 && !validateStateOper(pValue)) {
+      return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
+                             "Second parameter of STATECOUNT function"
+                             "must be one of the following: 'GE', 'GT', 'LE', 'LT', 'EQ', 'NE'");
+    }
+
     pValue->notReserved = true;
   }
 
@@ -504,6 +522,16 @@ static int32_t translateStateDuration(SFunctionNode* pFunc, char* pErrBuf, int32
 
     SValueNode* pValue = (SValueNode*)pParamNode;
 
+    if (i == 1 && !validateStateOper(pValue)) {
+      return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
+                             "Second parameter of STATEDURATION function"
+                             "must be one of the following: 'GE', 'GT', 'LE', 'LT', 'EQ', 'NE'");
+    } else if (i == 3 && pValue->datum.i == 0) {
+      return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
+                             "STATEDURATION function time unit parameter should be greater than db precision");
+    }
+
+
     pValue->notReserved = true;
   }
 

source/libs/function/src/builtinsimpl.c

@@ -1754,6 +1754,11 @@ int32_t leastSQRFunction(SqlFunctionCtx* pCtx) {
       }
       break;
     }
+    case TSDB_DATA_TYPE_NULL: {
+      GET_RES_INFO(pCtx)->isNullRes = 1;
+      numOfElem = 1;
+      break;
+    }
 
     default:
       break;
@@ -1797,7 +1802,7 @@ int32_t leastSQRFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
   size_t len = snprintf(varDataVal(buf), sizeof(buf) - VARSTR_HEADER_SIZE, "{slop:%.6lf, intercept:%.6lf}", param[0][2], param[1][2]);
   varDataSetLen(buf, len);
 
-  colDataAppend(pCol, currentRow, buf, false);
+  colDataAppend(pCol, currentRow, buf, pResInfo->isNullRes);
 
   return pResInfo->numOfRes;
 }

source/libs/index/CMakeLists.txt

@@ -12,6 +12,7 @@ target_link_libraries(
     PUBLIC os
     PUBLIC util
     PUBLIC common
+    PUBLIC vnode
     PUBLIC nodes
     PUBLIC scalar 
     PUBLIC function 

source/libs/index/inc/indexCache.h

@@ -36,6 +36,7 @@ typedef struct MemTable {
 typedef struct IndexCache {
   T_REF_DECLARE()
   MemTable *mem, *imm;
+  int32_t   merging;
   SIndex*   index;
   char*     colName;
   int64_t   version;

source/libs/index/inc/indexComm.h

@@ -33,6 +33,7 @@ typedef enum { MATCH, CONTINUE, BREAK } TExeCond;
 
 typedef TExeCond (*_cache_range_compare)(void* a, void* b, int8_t type);
 
+__compar_fn_t indexGetCompar(int8_t type);
 TExeCond      tCompare(__compar_fn_t func, int8_t cmpType, void* a, void* b, int8_t dType);
 TExeCond      tDoCompare(__compar_fn_t func, int8_t cmpType, void* a, void* b);
 

source/libs/index/src/index.c

@@ -80,7 +80,7 @@ static TdThreadOnce isInit = PTHREAD_ONCE_INIT;
 static int indexTermSearch(SIndex* sIdx, SIndexTermQuery* term, SArray** result);
 
 static void indexInterResultsDestroy(SArray* results);
-static int  indexMergeFinalResults(SArray* interResults, EIndexOperatorType oType, SArray* finalResult);
+static int  indexMergeFinalResults(SArray* in, EIndexOperatorType oType, SArray* out);
 
 static int indexGenTFile(SIndex* index, IndexCache* cache, SArray* batch);
 
@@ -386,21 +386,21 @@ static void indexInterResultsDestroy(SArray* results) {
   taosArrayDestroy(results);
 }
 
-static int indexMergeFinalResults(SArray* interResults, EIndexOperatorType oType, SArray* fResults) {
+static int indexMergeFinalResults(SArray* in, EIndexOperatorType oType, SArray* out) {
   // refactor, merge interResults into fResults by oType
-  for (int i = 0; i < taosArrayGetSize(interResults); i--) {
-    SArray* t = taosArrayGetP(interResults, i);
+  for (int i = 0; i < taosArrayGetSize(in); i--) {
+    SArray* t = taosArrayGetP(in, i);
     taosArraySort(t, uidCompare);
     taosArrayRemoveDuplicate(t, uidCompare, NULL);
   }
 
   if (oType == MUST) {
-    iIntersection(interResults, fResults);
+    iIntersection(in, out);
   } else if (oType == SHOULD) {
-    iUnion(interResults, fResults);
+    iUnion(in, out);
   } else if (oType == NOT) {
     // just one column index, enhance later
-    taosArrayAddAll(fResults, interResults);
+    // taosArrayAddAll(fResults, interResults);
     // not use currently
   }
   return 0;
@@ -463,6 +463,9 @@ int indexFlushCacheToTFile(SIndex* sIdx, void* cache) {
   int64_t st = taosGetTimestampUs();
 
   IndexCache* pCache = (IndexCache*)cache;
+
+  while (sIdx->quit && atomic_load_32(&pCache->merging) == 1) {
+  }
   TFileReader* pReader = tfileGetReaderByCol(sIdx->tindex, pCache->suid, pCache->colName);
   if (pReader == NULL) {
     indexWarn("empty tfile reader found");
@@ -475,9 +478,9 @@ int indexFlushCacheToTFile(SIndex* sIdx, void* cache) {
     tfileReaderUnRef(pReader);
     if (sIdx->quit) {
       indexPost(sIdx);
-      // indexCacheBroadcast(pCache);
     }
     indexReleaseRef(sIdx->refId);
+    atomic_store_32(&pCache->merging, 0);
     return 0;
   }
 
@@ -539,6 +542,7 @@ int indexFlushCacheToTFile(SIndex* sIdx, void* cache) {
   if (sIdx->quit) {
     indexPost(sIdx);
   }
+  atomic_store_32(&pCache->merging, 0);
   indexReleaseRef(sIdx->refId);
 
   return ret;
@@ -605,6 +609,7 @@ static int indexGenTFile(SIndex* sIdx, IndexCache* cache, SArray* batch) {
   taosThreadMutexLock(&tf->mtx);
   tfileCachePut(tf->cache, &key, reader);
   taosThreadMutexUnlock(&tf->mtx);
+
   return ret;
 END:
   if (tw != NULL) {

source/libs/index/src/indexCache.c

@@ -494,16 +494,19 @@ static void indexCacheMakeRoomForWrite(IndexCache* cache) {
       // TODO: wake up by condition variable
       indexCacheWait(cache);
     } else {
-      bool notifyQuit = cache->occupiedMem >= MEM_SIGNAL_QUIT ? true : false;
+      bool quit = cache->occupiedMem >= MEM_SIGNAL_QUIT ? true : false;
 
       indexCacheRef(cache);
       cache->imm = cache->mem;
       cache->mem = indexInternalCacheCreate(cache->type);
       cache->mem->pCache = cache;
       cache->occupiedMem = 0;
+      if (quit == false) {
+        atomic_store_32(&cache->merging, 1);
+      }
       // sched to merge
       // unref cache in bgwork
-      indexCacheSchedToMerge(cache, notifyQuit);
+      indexCacheSchedToMerge(cache, quit);
     }
   }
 }

source/libs/index/src/indexComm.c

@@ -75,7 +75,7 @@ char* indexInt2str(int64_t val, char* dst, int radix) {
     ;
   return dst - 1;
 }
-static __compar_fn_t indexGetCompar(int8_t type) {
+__compar_fn_t indexGetCompar(int8_t type) {
   if (type == TSDB_DATA_TYPE_BINARY || type == TSDB_DATA_TYPE_NCHAR) {
     return (__compar_fn_t)strcmp;
   }
@@ -182,6 +182,9 @@ TExeCond tDoCompare(__compar_fn_t func, int8_t comparType, void* a, void* b) {
     case QUERY_GREATER_EQUAL: {
       if (ret >= 0) return MATCH;
     }
+    case QUERY_TERM: {
+      if (ret == 0) return MATCH;
+    }
   }
   return CONTINUE;
 }

source/libs/index/src/indexFilter.c

@@ -14,11 +14,13 @@
  */
 
 #include "index.h"
+#include "indexComm.h"
 #include "indexInt.h"
 #include "nodes.h"
 #include "querynodes.h"
 #include "scalar.h"
 #include "tdatablock.h"
+#include "vnode.h"
 
 // clang-format off
 #define SIF_ERR_RET(c) do { int32_t _code = c; if (_code != TSDB_CODE_SUCCESS) { terrno = _code; return _code; } } while (0)
@@ -259,10 +261,52 @@ static int32_t sifExecFunction(SFunctionNode *node, SIFCtx *ctx, SIFParam *outpu
   indexError("index-filter not support buildin function");
   return TSDB_CODE_QRY_INVALID_INPUT;
 }
+
+typedef int (*Filter)(void *a, void *b, int16_t dtype);
+
+int sifGreaterThan(void *a, void *b, int16_t dtype) {
+  __compar_fn_t func = indexGetCompar(dtype);
+  return tDoCompare(func, QUERY_GREATER_THAN, a, b);
+}
+int sifGreaterEqual(void *a, void *b, int16_t dtype) {
+  __compar_fn_t func = indexGetCompar(dtype);
+  return tDoCompare(func, QUERY_GREATER_EQUAL, a, b);
+}
+int sifLessEqual(void *a, void *b, int16_t dtype) {
+  __compar_fn_t func = indexGetCompar(dtype);
+  return tDoCompare(func, QUERY_LESS_EQUAL, a, b);
+}
+int sifLessThan(void *a, void *b, int16_t dtype) {
+  __compar_fn_t func = indexGetCompar(dtype);
+  return (int)tDoCompare(func, QUERY_LESS_THAN, a, b);
+}
+int sifEqual(void *a, void *b, int16_t dtype) {
+  __compar_fn_t func = indexGetCompar(dtype);
+  return (int)tDoCompare(func, QUERY_TERM, a, b);
+}
+static Filter sifGetFilterFunc(EIndexQueryType type, bool *reverse) {
+  if (type == QUERY_LESS_EQUAL || type == QUERY_LESS_THAN) {
+    *reverse = true;
+  } else {
+    *reverse = false;
+  }
+  if (type == QUERY_LESS_EQUAL)
+    return sifLessEqual;
+  else if (type == QUERY_LESS_THAN)
+    return sifLessThan;
+  else if (type == QUERY_GREATER_EQUAL)
+    return sifGreaterEqual;
+  else if (type == QUERY_GREATER_THAN)
+    return sifGreaterThan;
+  else if (type == QUERY_TERM) {
+    return sifEqual;
+  }
+  return NULL;
+}
 static int32_t sifDoIndex(SIFParam *left, SIFParam *right, int8_t operType, SIFParam *output) {
-#ifdef USE_INVERTED_INDEX
   SIndexMetaArg *arg = &output->arg;
-  SIndexTerm *   tm = indexTermCreate(arg->suid, DEFAULT, left->colValType, left->colName, strlen(left->colName),
+#ifdef USE_INVERTED_INDEX
+  SIndexTerm *tm = indexTermCreate(arg->suid, DEFAULT, left->colValType, left->colName, strlen(left->colName),
                                    right->condValue, strlen(right->condValue));
   if (tm == NULL) {
     return TSDB_CODE_QRY_OUT_OF_MEMORY;
@@ -278,8 +322,22 @@ static int32_t sifDoIndex(SIFParam *left, SIFParam *right, int8_t operType, SIFP
   indexMultiTermQueryDestroy(mtm);
   return ret;
 #else
-  return 0;
+  EIndexQueryType qtype = 0;
+  SIF_ERR_RET(sifGetFuncFromSql(operType, &qtype));
+  bool   reverse;
+  Filter filterFunc = sifGetFilterFunc(qtype, &reverse);
+
+  SMetaFltParam param = {.suid = arg->suid,
+                         .cid = left->colId,
+                         .type = left->colValType,
+                         .val = right->condValue,
+                         .reverse = reverse,
+                         .filterFunc = filterFunc};
+
+  int ret = metaFilteTableIds(arg->metaEx, &param, output->result);
+  return ret;
 #endif
+  return 0;
 }
 
 static int32_t sifLessThanFunc(SIFParam *left, SIFParam *right, SIFParam *output) {

source/libs/index/src/indexJson.c

@@ -24,8 +24,8 @@ int tIndexJsonPut(SIndexJson *index, SIndexJsonMultiTerm *terms, uint64_t uid) {
     SIndexJsonTerm *p = taosArrayGetP(terms, i);
     INDEX_TYPE_ADD_EXTERN_TYPE(p->colType, TSDB_DATA_TYPE_JSON);
   }
-  return indexPut(index, terms, uid);
   // handle put
+  return indexPut(index, terms, uid);
 }
 
 int tIndexJsonSearch(SIndexJson *index, SIndexJsonMultiTermQuery *tq, SArray *result) {
@@ -34,11 +34,11 @@ int tIndexJsonSearch(SIndexJson *index, SIndexJsonMultiTermQuery *tq, SArray *re
     SIndexJsonTerm *p = taosArrayGetP(terms, i);
     INDEX_TYPE_ADD_EXTERN_TYPE(p->colType, TSDB_DATA_TYPE_JSON);
   }
-  return indexSearch(index, tq, result);
   // handle search
+  return indexSearch(index, tq, result);
 }
 
 void tIndexJsonClose(SIndexJson *index) {
-  return indexClose(index);
   // handle close
+  return indexClose(index);
 }

source/libs/monitor/src/monMsg.c

@@ -557,3 +557,49 @@ int32_t tDeserializeSMonMloadInfo(void *buf, int32_t bufLen, SMonMloadInfo *pInf
   tDecoderClear(&decoder);
   return 0;
 }
+
+
+int32_t tSerializeSQnodeLoad(void *buf, int32_t bufLen, SQnodeLoad *pInfo) {
+  SEncoder encoder = {0};
+  tEncoderInit(&encoder, buf, bufLen);
+
+  if (tStartEncode(&encoder) < 0) return -1;
+  if (tEncodeI64(&encoder, pInfo->numOfProcessedQuery) < 0) return -1;
+  if (tEncodeI64(&encoder, pInfo->numOfProcessedCQuery) < 0) return -1;
+  if (tEncodeI64(&encoder, pInfo->numOfProcessedFetch) < 0) return -1;
+  if (tEncodeI64(&encoder, pInfo->numOfProcessedDrop) < 0) return -1;
+  if (tEncodeI64(&encoder, pInfo->numOfProcessedHb) < 0) return -1;
+  if (tEncodeI64(&encoder, pInfo->cacheDataSize) < 0) return -1;
+  if (tEncodeI64(&encoder, pInfo->numOfQueryInQueue) < 0) return -1;
+  if (tEncodeI64(&encoder, pInfo->numOfFetchInQueue) < 0) return -1;
+  if (tEncodeI64(&encoder, pInfo->timeInQueryQueue) < 0) return -1;
+  if (tEncodeI64(&encoder, pInfo->timeInFetchQueue) < 0) return -1;
+  tEndEncode(&encoder);
+
+  int32_t tlen = encoder.pos;
+  tEncoderClear(&encoder);
+  return tlen;
+}
+
+int32_t tDeserializeSQnodeLoad(void *buf, int32_t bufLen, SQnodeLoad *pInfo) {
+  SDecoder decoder = {0};
+  tDecoderInit(&decoder, buf, bufLen);
+
+  if (tStartDecode(&decoder) < 0) return -1;
+  if (tDecodeI64(&decoder, &pInfo->numOfProcessedQuery) < 0) return -1;
+  if (tDecodeI64(&decoder, &pInfo->numOfProcessedCQuery) < 0) return -1;
+  if (tDecodeI64(&decoder, &pInfo->numOfProcessedFetch) < 0) return -1;
+  if (tDecodeI64(&decoder, &pInfo->numOfProcessedDrop) < 0) return -1;
+  if (tDecodeI64(&decoder, &pInfo->numOfProcessedHb) < 0) return -1;
+  if (tDecodeI64(&decoder, &pInfo->cacheDataSize) < 0) return -1;
+  if (tDecodeI64(&decoder, &pInfo->numOfQueryInQueue) < 0) return -1;
+  if (tDecodeI64(&decoder, &pInfo->numOfFetchInQueue) < 0) return -1;
+  if (tDecodeI64(&decoder, &pInfo->timeInQueryQueue) < 0) return -1;
+  if (tDecodeI64(&decoder, &pInfo->timeInFetchQueue) < 0) return -1;
+  tEndDecode(&decoder);
+
+  tDecoderClear(&decoder);
+  return 0;
+}
+
+

source/libs/parser/src/parTranslater.c

@@ -3310,6 +3310,7 @@ static int32_t buildCreateTopicReq(STranslateContext* pCxt, SCreateTopicStmt* pS
   if ('\0' != pStmt->subSTbName[0]) {
     pReq->subType = TOPIC_SUB_TYPE__TABLE;
     toName(pCxt->pParseCxt->acctId, pStmt->subDbName, pStmt->subSTbName, &name);
+    tNameGetFullDbName(&name, pReq->subDbName);
     tNameExtractFullName(&name, pReq->subStbName);
   } else if ('\0' != pStmt->subDbName[0]) {
     pReq->subType = TOPIC_SUB_TYPE__DB;

source/libs/planner/src/planPhysiCreater.c

@@ -468,6 +468,7 @@ static int32_t createTagScanPhysiNode(SPhysiPlanContext* pCxt, SSubplan* pSubpla
     return TSDB_CODE_OUT_OF_MEMORY;
   }
   vgroupInfoToNodeAddr(pScanLogicNode->pVgroupList->vgroups, &pSubplan->execNode);
+  SQueryNodeLoad node = {.addr = pSubplan->execNode, .load = 0};
   taosArrayPush(pCxt->pExecNodeList, &pSubplan->execNode);
   return createScanPhysiNodeFinalize(pCxt, pSubplan, pScanLogicNode, (SScanPhysiNode*)pTagScan, pPhyNode);
 }
@@ -489,7 +490,8 @@ static int32_t createTableScanPhysiNode(SPhysiPlanContext* pCxt, SSubplan* pSubp
     pSubplan->execNodeStat.tableNum = pScanLogicNode->pVgroupList->vgroups[0].numOfTable;
   }
   if (pCxt->pExecNodeList) {
-    taosArrayPush(pCxt->pExecNodeList, &pSubplan->execNode);
+    SQueryNodeLoad node = {.addr = pSubplan->execNode, .load = 0};
+    taosArrayPush(pCxt->pExecNodeList, &node);
   }
   tNameGetFullDbName(&pScanLogicNode->tableName, pSubplan->dbFName);
   pTableScan->dataRequired = pScanLogicNode->dataRequired;
@@ -524,10 +526,11 @@ static int32_t createSystemTableScanPhysiNode(SPhysiPlanContext* pCxt, SSubplan*
   pScan->accountId = pCxt->pPlanCxt->acctId;
   if (0 == strcmp(pScanLogicNode->tableName.tname, TSDB_INS_TABLE_USER_TABLES)) {
     vgroupInfoToNodeAddr(pScanLogicNode->pVgroupList->vgroups, &pSubplan->execNode);
+    SQueryNodeLoad node = { .addr = pSubplan->execNode, .load = 0};
     taosArrayPush(pCxt->pExecNodeList, &pSubplan->execNode);
   } else {
-    SQueryNodeAddr addr = {.nodeId = MNODE_HANDLE, .epSet = pCxt->pPlanCxt->mgmtEpSet};
-    taosArrayPush(pCxt->pExecNodeList, &addr);
+    SQueryNodeLoad node = { .addr = {.nodeId = MNODE_HANDLE, .epSet = pCxt->pPlanCxt->mgmtEpSet}, .load = 0};
+    taosArrayPush(pCxt->pExecNodeList, &node);
   }
   pScan->mgmtEpSet = pCxt->pPlanCxt->mgmtEpSet;
   tNameGetFullDbName(&pScanLogicNode->tableName, pSubplan->dbFName);
@@ -1248,7 +1251,8 @@ static int32_t createPhysiSubplan(SPhysiPlanContext* pCxt, SLogicSubplan* pLogic
     SVnodeModifLogicNode* pModif = (SVnodeModifLogicNode*)pLogicSubplan->pNode;
     pSubplan->msgType = pModif->msgType;
     pSubplan->execNode.epSet = pModif->pVgDataBlocks->vg.epSet;
-    taosArrayPush(pCxt->pExecNodeList, &pSubplan->execNode);
+    SQueryNodeLoad node = {.addr = pSubplan->execNode, .load = 0};
+    taosArrayPush(pCxt->pExecNodeList, &node);
     code = createDataInserter(pCxt, pModif->pVgDataBlocks, &pSubplan->pDataSink);
   } else {
     pSubplan->msgType = TDMT_VND_QUERY;

source/libs/qcom/src/querymsg.c

@@ -373,7 +373,7 @@ int32_t queryProcessQnodeListRsp(void *output, char *msg, int32_t msgSize) {
     return code;
   }
 
-  out.addrsList = (SArray *)output;
+  out.qnodeList = (SArray *)output;
   if (tDeserializeSQnodeListRsp(msg, msgSize, &out) != 0) {
     qError("invalid qnode list rsp msg, msgSize:%d", msgSize);
     code = TSDB_CODE_INVALID_MSG;

source/libs/qworker/inc/qwInt.h

@@ -145,13 +145,30 @@ typedef struct SQWSchStatus {
   SHashObj      *tasksHash;  // key:queryId+taskId, value: SQWTaskStatus
 } SQWSchStatus;
 
-typedef struct SQWWaitTimeStat {
+typedef struct SQWTimeInQ {
   uint64_t num;
   uint64_t total;
-} SQWWaitTimeStat;
+} SQWTimeInQ;
+
+typedef struct SQWMsgStat {
+  SQWTimeInQ waitTime[2];
+  uint64_t queryProcessed;
+  uint64_t cqueryProcessed;
+  uint64_t fetchProcessed;
+  uint64_t fetchRspProcessed;
+  uint64_t cancelProcessed;
+  uint64_t dropProcessed;
+  uint64_t hbProcessed;
+} SQWMsgStat;
+
+typedef struct SQWRTStat {
+  uint64_t startTaskNum;
+  uint64_t stopTaskNum;
+} SQWRTStat;
 
 typedef struct SQWStat {
-  SQWWaitTimeStat msgWait[2];
+  SQWMsgStat   msgStat;
+  SQWRTStat    rtStat;
 } SQWStat;
 
 // Qnode/Vnode level task management
@@ -182,10 +199,13 @@ typedef struct SQWorkerMgmt {
 #define QW_IDS()       sId, qId, tId, rId
 #define QW_FPARAMS()   mgmt, QW_IDS()
 
-#define QW_GET_EVENT_VALUE(ctx, event) atomic_load_8(&(ctx)->events[event])
+#define QW_STAT_INC(_item, _n) atomic_add_fetch_64(&(_item), _n)
+#define QW_STAT_DEC(_item, _n) atomic_sub_fetch_64(&(_item), _n)
+#define QW_STAT_GET(_item) atomic_load_64(&(_item))
 
-#define QW_IS_EVENT_RECEIVED(ctx, event)   (atomic_load_8(&(ctx)->events[event]) == QW_EVENT_RECEIVED)
-#define QW_IS_EVENT_PROCESSED(ctx, event)  (atomic_load_8(&(ctx)->events[event]) == QW_EVENT_PROCESSED)
+#define QW_GET_EVENT(ctx, event) atomic_load_8(&(ctx)->events[event])
+#define QW_IS_EVENT_RECEIVED(ctx, event)   (QW_GET_EVENT(ctx, event) == QW_EVENT_RECEIVED)
+#define QW_IS_EVENT_PROCESSED(ctx, event)  (QW_GET_EVENT(ctx, event) == QW_EVENT_PROCESSED)
 #define QW_SET_EVENT_RECEIVED(ctx, event)  atomic_store_8(&(ctx)->events[event], QW_EVENT_RECEIVED)
 #define QW_SET_EVENT_PROCESSED(ctx, event) atomic_store_8(&(ctx)->events[event], QW_EVENT_PROCESSED)
 
@@ -332,8 +352,8 @@ int32_t qwDropTask(QW_FPARAMS_DEF);
 void qwSaveTbVersionInfo(qTaskInfo_t       pTaskInfo, SQWTaskCtx *ctx);
 int32_t qwOpenRef(void);
 void qwSetHbParam(int64_t refId, SQWHbParam **pParam);
-int32_t qwUpdateWaitTimeInQueue(SQWorker *mgmt, int64_t ts, EQueueType type);
-int64_t qwGetWaitTimeInQueue(SQWorker *mgmt, EQueueType type);
+int32_t qwUpdateTimeInQueue(SQWorker *mgmt, int64_t ts, EQueueType type);
+int64_t qwGetTimeInQueue(SQWorker *mgmt, EQueueType type);
 
 void qwDbgDumpMgmtInfo(SQWorker *mgmt);
 int32_t qwDbgValidateStatus(QW_FPARAMS_DEF, int8_t oriStatus, int8_t newStatus, bool *ignore);

source/libs/qworker/src/qwMsg.c

@@ -257,7 +257,8 @@ int32_t qWorkerProcessQueryMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg, int
   SSubQueryMsg *msg = pMsg->pCont;
   SQWorker *    mgmt = (SQWorker *)qWorkerMgmt;
 
-  qwUpdateWaitTimeInQueue(mgmt, ts, QUERY_QUEUE);
+  qwUpdateTimeInQueue(mgmt, ts, QUERY_QUEUE);
+  QW_STAT_INC(mgmt->stat.msgStat.queryProcessed, 1);
 
   if (NULL == msg || pMsg->contLen <= sizeof(*msg)) {
     QW_ELOG("invalid query msg, msg:%p, msgLen:%d", msg, pMsg->contLen);
@@ -297,7 +298,8 @@ int32_t qWorkerProcessCQueryMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg, in
   SQWTaskCtx *       handles = NULL;
   SQWorker *         mgmt = (SQWorker *)qWorkerMgmt;
 
-  qwUpdateWaitTimeInQueue(mgmt, ts, QUERY_QUEUE);
+  qwUpdateTimeInQueue(mgmt, ts, QUERY_QUEUE);
+  QW_STAT_INC(mgmt->stat.msgStat.cqueryProcessed, 1);
 
   if (NULL == msg || pMsg->contLen < sizeof(*msg)) {
     QW_ELOG("invalid cquery msg, msg:%p, msgLen:%d", msg, pMsg->contLen);
@@ -328,7 +330,8 @@ int32_t qWorkerProcessFetchMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg, int
   SResFetchReq *msg = pMsg->pCont;
   SQWorker *    mgmt = (SQWorker *)qWorkerMgmt;
 
-  qwUpdateWaitTimeInQueue(mgmt, ts, FETCH_QUEUE);
+  qwUpdateTimeInQueue(mgmt, ts, FETCH_QUEUE);
+  QW_STAT_INC(mgmt->stat.msgStat.fetchProcessed, 1);
 
   if (NULL == msg || pMsg->contLen < sizeof(*msg)) {
     QW_ELOG("invalid fetch msg, msg:%p, msgLen:%d", msg, pMsg->contLen);
@@ -357,7 +360,10 @@ int32_t qWorkerProcessFetchMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg, int
 
 int32_t qWorkerProcessFetchRsp(void *node, void *qWorkerMgmt, SRpcMsg *pMsg, int64_t ts) {
   SQWorker *      mgmt = (SQWorker *)qWorkerMgmt;
-  qwUpdateWaitTimeInQueue(mgmt, ts, FETCH_QUEUE);
+  if (mgmt) {
+    qwUpdateTimeInQueue(mgmt, ts, FETCH_QUEUE);
+    QW_STAT_INC(mgmt->stat.msgStat.fetchRspProcessed, 1);
+  }
 
   qProcessFetchRsp(NULL, pMsg, NULL);
   pMsg->pCont = NULL;
@@ -373,7 +379,8 @@ int32_t qWorkerProcessCancelMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg, in
   int32_t         code = 0;
   STaskCancelReq *msg = pMsg->pCont;
 
-  qwUpdateWaitTimeInQueue(mgmt, ts, FETCH_QUEUE);
+  qwUpdateTimeInQueue(mgmt, ts, FETCH_QUEUE);
+  QW_STAT_INC(mgmt->stat.msgStat.cancelProcessed, 1);
 
   if (NULL == msg || pMsg->contLen < sizeof(*msg)) {
     qError("invalid task cancel msg");
@@ -411,7 +418,8 @@ int32_t qWorkerProcessDropMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg, int6
   STaskDropReq *msg = pMsg->pCont;
   SQWorker *    mgmt = (SQWorker *)qWorkerMgmt;
 
-  qwUpdateWaitTimeInQueue(mgmt, ts, FETCH_QUEUE);
+  qwUpdateTimeInQueue(mgmt, ts, FETCH_QUEUE);
+  QW_STAT_INC(mgmt->stat.msgStat.dropProcessed, 1);
 
   if (NULL == msg || pMsg->contLen < sizeof(*msg)) {
     QW_ELOG("invalid task drop msg, msg:%p, msgLen:%d", msg, pMsg->contLen);
@@ -452,7 +460,8 @@ int32_t qWorkerProcessHbMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg, int64_
   SSchedulerHbReq req = {0};
   SQWorker *      mgmt = (SQWorker *)qWorkerMgmt;
 
-  qwUpdateWaitTimeInQueue(mgmt, ts, FETCH_QUEUE);
+  qwUpdateTimeInQueue(mgmt, ts, FETCH_QUEUE);
+  QW_STAT_INC(mgmt->stat.msgStat.hbProcessed, 1);
 
   if (NULL == pMsg->pCont) {
     QW_ELOG("invalid hb msg, msg:%p, msgLen:%d", pMsg->pCont, pMsg->contLen);

source/libs/qworker/src/qwUtil.c

@@ -499,7 +499,7 @@ int32_t qwOpenRef(void) {
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t qwUpdateWaitTimeInQueue(SQWorker *mgmt, int64_t ts, EQueueType type) {
+int32_t qwUpdateTimeInQueue(SQWorker *mgmt, int64_t ts, EQueueType type) {
   if (ts <= 0) {
     return TSDB_CODE_SUCCESS;
   }
@@ -507,12 +507,12 @@ int32_t qwUpdateWaitTimeInQueue(SQWorker *mgmt, int64_t ts, EQueueType type) {
   int64_t duration = taosGetTimestampUs() - ts;
   switch (type) {
     case QUERY_QUEUE:
-      ++mgmt->stat.msgWait[0].num;
-      mgmt->stat.msgWait[0].total += duration;
+      ++mgmt->stat.msgStat.waitTime[0].num;
+      mgmt->stat.msgStat.waitTime[0].total += duration;
       break;
     case FETCH_QUEUE:
-      ++mgmt->stat.msgWait[1].num;
-      mgmt->stat.msgWait[1].total += duration;
+      ++mgmt->stat.msgStat.waitTime[1].num;
+      mgmt->stat.msgStat.waitTime[1].total += duration;
       break;
     default:
       qError("unsupported queue type %d", type);
@@ -522,19 +522,20 @@ int32_t qwUpdateWaitTimeInQueue(SQWorker *mgmt, int64_t ts, EQueueType type) {
   return TSDB_CODE_SUCCESS;
 }
 
-int64_t qwGetWaitTimeInQueue(SQWorker *mgmt, EQueueType type) {
-  SQWWaitTimeStat *pStat = NULL;
+int64_t qwGetTimeInQueue(SQWorker *mgmt, EQueueType type) {
+  SQWTimeInQ *pStat = NULL;
   switch (type) {
     case QUERY_QUEUE:
-      pStat = &mgmt->stat.msgWait[0];
+      pStat = &mgmt->stat.msgStat.waitTime[0];
       return pStat->num ? (pStat->total/pStat->num) : 0;
     case FETCH_QUEUE:
-      pStat = &mgmt->stat.msgWait[1];
+      pStat = &mgmt->stat.msgStat.waitTime[1];
       return pStat->num ? (pStat->total/pStat->num) : 0;
     default:
       qError("unsupported queue type %d", type);
-      return -1;
   }
+
+  return -1;
 }
 
 

source/libs/qworker/src/qworker.c

-#include "qworker.h"
 #include "dataSinkMgt.h"
 #include "executor.h"
 #include "planner.h"
@@ -8,6 +7,7 @@
 #include "tcommon.h"
 #include "tmsg.h"
 #include "tname.h"
+#include "qworker.h"
 
 SQWorkerMgmt gQwMgmt = {
     .lock = 0,
@@ -954,8 +954,29 @@ void qWorkerDestroy(void **qWorkerMgmt) {
   }
 }
 
-int64_t qWorkerGetWaitTimeInQueue(void *qWorkerMgmt, EQueueType type) {
-  return qwGetWaitTimeInQueue((SQWorker *)qWorkerMgmt, type);
+int32_t qWorkerGetStat(SReadHandle *handle, void *qWorkerMgmt, SQWorkerStat *pStat) {
+  if (NULL == handle || NULL == qWorkerMgmt || NULL == pStat) {
+    QW_RET(TSDB_CODE_QRY_INVALID_INPUT);
+  }
+
+  SQWorker *mgmt = (SQWorker *)qWorkerMgmt;
+  SDataSinkStat sinkStat = {0};
+  
+  dsDataSinkGetCacheSize(&sinkStat);
+  pStat->cacheDataSize = sinkStat.cachedSize;
+  
+  pStat->queryProcessed = QW_STAT_GET(mgmt->stat.msgStat.queryProcessed);
+  pStat->cqueryProcessed = QW_STAT_GET(mgmt->stat.msgStat.cqueryProcessed);
+  pStat->fetchProcessed = QW_STAT_GET(mgmt->stat.msgStat.fetchProcessed);
+  pStat->dropProcessed = QW_STAT_GET(mgmt->stat.msgStat.dropProcessed);
+  pStat->hbProcessed = QW_STAT_GET(mgmt->stat.msgStat.hbProcessed);
+
+  pStat->numOfQueryInQueue = handle->pMsgCb->qsizeFp(handle->pMsgCb->mgmt, mgmt->nodeId, QUERY_QUEUE);
+  pStat->numOfFetchInQueue = handle->pMsgCb->qsizeFp(handle->pMsgCb->mgmt, mgmt->nodeId, FETCH_QUEUE);
+  pStat->timeInQueryQueue = qwGetTimeInQueue((SQWorker *)qWorkerMgmt, QUERY_QUEUE);
+  pStat->timeInFetchQueue = qwGetTimeInQueue((SQWorker *)qWorkerMgmt, FETCH_QUEUE);
+
+  return TSDB_CODE_SUCCESS;
 }
 
 

source/libs/scalar/test/scalar/CMakeLists.txt

@@ -17,7 +17,9 @@ TARGET_INCLUDE_DIRECTORIES(
         PUBLIC "${TD_SOURCE_DIR}/source/libs/parser/inc"
         PRIVATE "${TD_SOURCE_DIR}/source/libs/scalar/inc"
 )
-add_test(
+if(NOT TD_WINDOWS)
+        add_test(
                 NAME scalarTest
                 COMMAND scalarTest
-)
+        )
+endif(NOT TD_WINDOWS)

source/libs/scheduler/inc/schedulerInt.h

@@ -300,7 +300,7 @@ int32_t schProcessOnTaskSuccess(SSchJob *pJob, SSchTask *pTask);
 int32_t schSaveJobQueryRes(SSchJob *pJob, SQueryTableRsp *rsp);
 int32_t schProcessOnExplainDone(SSchJob *pJob, SSchTask *pTask, SRetrieveTableRsp *pRsp);
 void schProcessOnDataFetched(SSchJob *job);
-int32_t schGetTaskFromTaskList(SHashObj *pTaskList, uint64_t taskId, SSchTask **pTask);
+int32_t schGetTaskInJob(SSchJob *pJob, uint64_t taskId, SSchTask **pTask);
 int32_t schUpdateTaskExecNodeHandle(SSchTask *pTask, void *handle, int32_t rspCode);
 void schFreeRpcCtxVal(const void *arg);
 int32_t schMakeBrokenLinkVal(SSchJob *pJob, SSchTask *pTask, SRpcBrokenlinkVal *brokenVal, bool isHb);
@@ -314,12 +314,11 @@ int32_t schCancelJob(SSchJob *pJob);
 int32_t schProcessOnJobDropped(SSchJob *pJob, int32_t errCode);
 uint64_t schGenTaskId(void);
 void schCloseJobRef(void);
-int32_t schExecJob(void *pTrans, SArray *pNodeList, SQueryPlan *pDag, int64_t *pJob, const char *sql,
-                         int64_t startTs, SSchResInfo *pRes);
-int32_t schAsyncExecJob(void *pTrans, SArray *pNodeList, SQueryPlan *pDag, int64_t *pJob, const char *sql,
-                                int64_t startTs, SSchResInfo *pRes);
+int32_t schExecJob(void *pTrans, SArray *pNodeList, SQueryPlan *pDag, int64_t *pJob, const char *sql, int64_t startTs, SSchResInfo *pRes);
+int32_t schAsyncExecJob(void *pTrans, SArray *pNodeList, SQueryPlan *pDag, int64_t *pJob, const char *sql, int64_t startTs, SSchResInfo *pRes);
 int32_t schFetchRows(SSchJob *pJob);
 int32_t schAsyncFetchRows(SSchJob *pJob);
+int32_t schUpdateTaskHandle(SSchJob *pJob, SSchTask *pTask, int32_t msgType, void *handle, int32_t rspCode);
 
 
 #ifdef __cplusplus

source/libs/scheduler/src/schJob.c

@@ -342,6 +342,36 @@ int32_t schRecordTaskExecNode(SSchJob *pJob, SSchTask *pTask, SQueryNodeAddr *ad
   return TSDB_CODE_SUCCESS;
 }
 
+int32_t schDropTaskExecNode(SSchJob *pJob, SSchTask *pTask, void *handle) {
+  if (NULL == pTask->execNodes) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+  int32_t num = taosArrayGetSize(pTask->execNodes);
+  for (int32_t i = 0; i < num; ++i) {
+    SSchNodeInfo* pNode = taosArrayGet(pTask->execNodes, i);
+    if (pNode->handle == handle) {
+      taosArrayRemove(pTask->execNodes, i);
+      break;
+    }
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
+int32_t schUpdateTaskHandle(SSchJob *pJob, SSchTask *pTask, int32_t msgType, void *handle, int32_t rspCode) {
+  SCH_SET_TASK_HANDLE(pTask, handle);
+
+  schUpdateTaskExecNodeHandle(pTask, handle, rspCode);
+
+  if (msgType == TDMT_SCH_LINK_BROKEN) {
+    schDropTaskExecNode(pJob, pTask, handle);
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
+
 int32_t schRecordQueryDataSrc(SSchJob *pJob, SSchTask *pTask) {
   if (!SCH_IS_DATA_SRC_QRY_TASK(pTask)) {
     return TSDB_CODE_SUCCESS;
@@ -469,6 +499,34 @@ _return:
   SCH_RET(code);
 }
 
+int32_t schSetAddrsFromNodeList(SSchJob *pJob, SSchTask *pTask) {
+  int32_t addNum = 0;
+  int32_t nodeNum = 0;
+  
+  if (pJob->nodeList) {
+    nodeNum = taosArrayGetSize(pJob->nodeList);
+
+    for (int32_t i = 0; i < nodeNum && addNum < SCH_MAX_CANDIDATE_EP_NUM; ++i) {
+      SQueryNodeAddr *naddr = taosArrayGet(pJob->nodeList, i);
+
+      if (NULL == taosArrayPush(pTask->candidateAddrs, naddr)) {
+        SCH_TASK_ELOG("taosArrayPush execNode to candidate addrs failed, addNum:%d, errno:%d", addNum, errno);
+        SCH_ERR_RET(TSDB_CODE_QRY_OUT_OF_MEMORY);
+      }
+
+      ++addNum;
+    }
+  }
+
+  if (addNum <= 0) {
+    SCH_TASK_ELOG("no available execNode as candidates, nodeNum:%d", nodeNum);
+    SCH_ERR_RET(TSDB_CODE_QRY_INVALID_INPUT);
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
+
 int32_t schSetTaskCandidateAddrs(SSchJob *pJob, SSchTask *pTask) {
   if (NULL != pTask->candidateAddrs) {
     return TSDB_CODE_SUCCESS;
@@ -492,27 +550,7 @@ int32_t schSetTaskCandidateAddrs(SSchJob *pJob, SSchTask *pTask) {
     return TSDB_CODE_SUCCESS;
   }
 
-  int32_t addNum = 0;
-  int32_t nodeNum = 0;
-  if (pJob->nodeList) {
-    nodeNum = taosArrayGetSize(pJob->nodeList);
-
-    for (int32_t i = 0; i < nodeNum && addNum < SCH_MAX_CANDIDATE_EP_NUM; ++i) {
-      SQueryNodeAddr *naddr = taosArrayGet(pJob->nodeList, i);
-
-      if (NULL == taosArrayPush(pTask->candidateAddrs, naddr)) {
-        SCH_TASK_ELOG("taosArrayPush execNode to candidate addrs failed, addNum:%d, errno:%d", addNum, errno);
-        SCH_ERR_RET(TSDB_CODE_QRY_OUT_OF_MEMORY);
-      }
-
-      ++addNum;
-    }
-  }
-
-  if (addNum <= 0) {
-    SCH_TASK_ELOG("no available execNode as candidates, nodeNum:%d", nodeNum);
-    SCH_ERR_RET(TSDB_CODE_QRY_INVALID_INPUT);
-  }
+  SCH_ERR_RET(schSetAddrsFromNodeList(pJob, pTask));
 
   /*
     for (int32_t i = 0; i < job->dataSrcEps.numOfEps && addNum < SCH_MAX_CANDIDATE_EP_NUM; ++i) {
@@ -1001,19 +1039,19 @@ int32_t schProcessOnTaskSuccess(SSchJob *pJob, SSchTask *pTask) {
   */
 
   for (int32_t i = 0; i < parentNum; ++i) {
-    SSchTask *par = *(SSchTask **)taosArrayGet(pTask->parents, i);
-    int32_t   readyNum = atomic_add_fetch_32(&par->childReady, 1);
+    SSchTask *parent = *(SSchTask **)taosArrayGet(pTask->parents, i);
+    int32_t   readyNum = atomic_add_fetch_32(&parent->childReady, 1);
 
-    SCH_LOCK(SCH_WRITE, &par->lock);
+    SCH_LOCK(SCH_WRITE, &parent->lock);
     SDownstreamSourceNode source = {.type = QUERY_NODE_DOWNSTREAM_SOURCE,
                                     .taskId = pTask->taskId,
                                     .schedId = schMgmt.sId,
                                     .addr = pTask->succeedAddr};
-    qSetSubplanExecutionNode(par->plan, pTask->plan->id.groupId, &source);
-    SCH_UNLOCK(SCH_WRITE, &par->lock);
+    qSetSubplanExecutionNode(parent->plan, pTask->plan->id.groupId, &source);
+    SCH_UNLOCK(SCH_WRITE, &parent->lock);
 
-    if (SCH_TASK_READY_FOR_LAUNCH(readyNum, par)) {
-      SCH_ERR_RET(schLaunchTask(pJob, par));
+    if (SCH_TASK_READY_FOR_LAUNCH(readyNum, parent)) {
+      SCH_ERR_RET(schLaunchTask(pJob, parent));
     }
   }
 
@@ -1087,7 +1125,7 @@ int32_t schSaveJobQueryRes(SSchJob *pJob, SQueryTableRsp *rsp) {
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t schGetTaskFromTaskList(SHashObj *pTaskList, uint64_t taskId, SSchTask **pTask) {
+int32_t schGetTaskFromList(SHashObj *pTaskList, uint64_t taskId, SSchTask **pTask) {
   int32_t s = taosHashGetSize(pTaskList);
   if (s <= 0) {
     return TSDB_CODE_SUCCESS;
@@ -1103,6 +1141,21 @@ int32_t schGetTaskFromTaskList(SHashObj *pTaskList, uint64_t taskId, SSchTask **
   return TSDB_CODE_SUCCESS;
 }
 
+int32_t schGetTaskInJob(SSchJob *pJob, uint64_t taskId, SSchTask **pTask) {
+  schGetTaskFromList(pJob->execTasks, taskId, pTask);
+  if (NULL == *pTask) {
+    schGetTaskFromList(pJob->succTasks, taskId, pTask);
+
+    if (NULL == *pTask) {
+      SCH_JOB_ELOG("task not found in execList & succList, taskId:%" PRIx64, taskId);
+      SCH_ERR_RET(TSDB_CODE_SCH_INTERNAL_ERROR);
+    }
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
+
 int32_t schUpdateTaskExecNodeHandle(SSchTask *pTask, void *handle, int32_t rspCode) {
   if (rspCode || NULL == pTask->execNodes || taosArrayGetSize(pTask->execNodes) > 1 ||
       taosArrayGetSize(pTask->execNodes) <= 0) {

source/libs/scheduler/src/schRemote.c

@@ -358,27 +358,11 @@ int32_t schHandleCallback(void *param, const SDataBuf *pMsg, int32_t msgType, in
     SCH_ERR_JRET(TSDB_CODE_QRY_JOB_FREED);
   }
 
-  schGetTaskFromTaskList(pJob->execTasks, pParam->taskId, &pTask);
-  if (NULL == pTask) {
-    if (TDMT_VND_EXPLAIN_RSP == msgType) {
-      schGetTaskFromTaskList(pJob->succTasks, pParam->taskId, &pTask);
-    } else {
-      SCH_JOB_ELOG("task not found in execTask list, refId:%" PRIx64 ", taskId:%" PRIx64, pParam->refId,
-                   pParam->taskId);
-      SCH_ERR_JRET(TSDB_CODE_SCH_INTERNAL_ERROR);
-    }
-  }
-
-  if (NULL == pTask) {
-    SCH_JOB_ELOG("task not found in execList & succList, refId:%" PRIx64 ", taskId:%" PRIx64, pParam->refId,
-                 pParam->taskId);
-    SCH_ERR_JRET(TSDB_CODE_SCH_INTERNAL_ERROR);
-  }
+  SCH_ERR_JRET(schGetTaskInJob(pJob, pParam->taskId, &pTask));
 
   SCH_TASK_DLOG("rsp msg received, type:%s, handle:%p, code:%s", TMSG_INFO(msgType), pMsg->handle, tstrerror(rspCode));
 
-  SCH_SET_TASK_HANDLE(pTask, pMsg->handle);
-  schUpdateTaskExecNodeHandle(pTask, pMsg->handle, rspCode);
+  SCH_ERR_JRET(schUpdateTaskHandle(pJob, pTask, msgType, pMsg->handle, rspCode));
   
   SCH_ERR_JRET(schHandleResponseMsg(pJob, pTask, msgType, pMsg->pData, pMsg->len, rspCode));
 

source/libs/scheduler/src/scheduler.c

@@ -141,7 +141,7 @@ int32_t schedulerGetTasksStatus(int64_t job, SArray *pSub) {
 
   if (pJob->status < JOB_TASK_STATUS_NOT_START || pJob->levelNum <= 0 || NULL == pJob->levels) {
     qDebug("job not initialized or not executable job, refId:%" PRIx64, job);
-    SCH_ERR_RET(TSDB_CODE_SCH_STATUS_ERROR);
+    SCH_ERR_JRET(TSDB_CODE_SCH_STATUS_ERROR);
   }
 
   for (int32_t i = pJob->levelNum - 1; i >= 0; --i) {
@@ -155,7 +155,11 @@ int32_t schedulerGetTasksStatus(int64_t job, SArray *pSub) {
     }
   }
 
-  return TSDB_CODE_SUCCESS;
+_return:
+
+  schReleaseJob(job);
+
+  SCH_RET(code);
 }
 
 int32_t scheduleCancelJob(int64_t job) {

source/libs/tdb/src/db/tdbTable.c

@@ -16,7 +16,7 @@
 #include "tdbInt.h"
 
 struct STTB {
-  TDB    *pEnv;
+  TDB *   pEnv;
   SBTree *pBt;
 };
 
@@ -25,11 +25,11 @@ struct STBC {
 };
 
 int tdbTbOpen(const char *tbname, int keyLen, int valLen, tdb_cmpr_fn_t keyCmprFn, TDB *pEnv, TTB **ppTb) {
-  TTB    *pTb;
+  TTB *   pTb;
   SPager *pPager;
   int     ret;
   char    fFullName[TDB_FILENAME_LEN];
-  SPage  *pPage;
+  SPage * pPage;
   SPgno   pgno;
 
   *ppTb = NULL;

source/util/src/terror.c

@@ -187,6 +187,7 @@ TAOS_DEFINE_ERROR(TSDB_CODE_MND_SNODE_NOT_EXIST,          "Snode not there")
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_BNODE_ALREADY_EXIST,      "Bnode already exists")
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_BNODE_NOT_EXIST,          "Bnode not there")
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_TOO_FEW_MNODES,           "Too few mnodes")
+TAOS_DEFINE_ERROR(TSDB_CODE_MND_TOO_MANY_MNODES,          "Too many mnodes")
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_MNODE_DEPLOYED,           "Mnode deployed in this dnode")
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_CANT_DROP_MASTER,         "Can't drop mnode which is master")
 

tests/pytest/util/common.py

@@ -95,8 +95,12 @@ class TDCom:
         stb_list = map(lambda x: x[0], res_row_list)
         for stb in stb_list:
             if type == "taosc":
+                tdSql.execute(f'drop table if exists `{stb}`')
+                if not stb[0].isdigit():
                     tdSql.execute(f'drop table if exists {stb}')
             elif type == "restful":
+                self.restApiPost(f"drop table if exists `{stb}`")
+                if not stb[0].isdigit():
                     self.restApiPost(f"drop table if exists {stb}")
 
     def dateToTs(self, datetime_input):

tests/script/jenkins/basic.txt

@@ -57,7 +57,6 @@
 # ---- mnode
 ./test.sh -f tsim/mnode/basic1.sim
 ./test.sh -f tsim/mnode/basic2.sim
-./test.sh -f tsim/mnode/basic3.sim
 
 # ---- show
 ./test.sh -f tsim/show/basic.sim
@@ -108,6 +107,7 @@
 ./test.sh -f tsim/stable/alter_insert1.sim
 ./test.sh -f tsim/stable/alter_insert2.sim
 ./test.sh -f tsim/stable/alter_import.sim
+./test.sh -f tsim/stable/tag_filter.sim
 
 # --- for multi process mode
 ./test.sh -f tsim/user/basic1.sim -m

tests/script/tsim/mnode/basic3.sim

@@ -2,14 +2,17 @@ system sh/stop_dnodes.sh
 system sh/deploy.sh -n dnode1 -i 1
 system sh/deploy.sh -n dnode2 -i 2
 system sh/deploy.sh -n dnode3 -i 3
+system sh/deploy.sh -n dnode4 -i 4
 system sh/exec.sh -n dnode1 -s start
 system sh/exec.sh -n dnode2 -s start
 system sh/exec.sh -n dnode3 -s start
+system sh/exec.sh -n dnode4 -s start
 sql connect
 
 print =============== step1: create dnodes
 sql create dnode $hostname port 7200
 sql create dnode $hostname port 7300
+sql create dnode $hostname port 7400
 
 $x = 0
 step1: 
@@ -32,6 +35,7 @@ endi
 print =============== step2: create mnode 2
 sql create mnode on dnode 2
 sql create mnode on dnode 3
+sql_error create mnode on dnode 4
 
 $x = 0
 step2: 
@@ -135,3 +139,4 @@ endi
 system sh/exec.sh -n dnode1 -s stop
 system sh/exec.sh -n dnode2 -s stop
 system sh/exec.sh -n dnode3 -s stop
+system sh/exec.sh -n dnode4 -s stop
\ No newline at end of file

tests/script/tsim/stable/tag_filter.sim

+system sh/stop_dnodes.sh
+system sh/deploy.sh -n dnode1 -i 1
+system sh/exec.sh -n dnode1 -s start
+sql connect
+
+print ========== prepare stb and ctb
+sql create database db vgroups 1
+sql create table db.stb (ts timestamp, c1 int, c2 binary(4)) tags(t1 int, t2 binary(16)) comment "abd"
+
+
+sql create table db.ctb1 using db.stb tags(1, "102")
+sql insert into db.ctb1 values(now, 1, "2")
+
+sql create table db.ctb2 using db.stb tags(2, "102")
+sql insert into db.ctb2 values(now, 2, "2")
+
+sql create table db.ctb3 using db.stb tags(3, "102")
+sql insert into db.ctb3 values(now, 3, "2")
+
+sql create table db.ctb4 using db.stb tags(4, "102")
+sql insert into db.ctb4 values(now, 4, "2")
+
+sql create table db.ctb5 using db.stb tags(5, "102")
+sql insert into db.ctb5 values(now, 5, "2")
+
+sql create table db.ctb6 using db.stb tags(6, "102")
+sql insert into db.ctb6 values(now, 6, "2")
+
+sql select * from db.stb where t1 = 1
+if $rows != 1 then 
+  return -1
+endi
+
+sql select * from db.stb where t1 < 1 
+if $rows != 0 then 
+  return -=1
+endi
+
+sql select * from db.stb where t1 < 2 
+if $rows != 1 then 
+  return -1
+endi
+
+sql select * from db.stb where t1 <= 2 
+if $rows != 2 then 
+  return -1
+endi
+
+sql select * from db.stb where t1 >= 1 
+if $rows != 6 then 
+  return -1
+endi
+
+sql select * from db.stb where t1 > 1 
+if $rows != 5 then 
+  return -1
+endi
+
+system sh/exec.sh -n dnode1 -s stop -x SIGINT

tests/script/tsim/trans/create_db.sim

@@ -76,14 +76,6 @@ if $data[0][3] != d1 then
   return -1
 endi
 
-if $data[0][4] != create-db then
-  return -1
-endi
-
-if $data[0][7] != @Unable to establish connection@ then
-  return -1
-endi
-
 sql_error create database d1 vgroups 2;
 
 print =============== start dnode2
@@ -125,15 +117,7 @@ endi
 if $data[0][3] != d2 then
   return -1
 endi
-
-if $data[0][4] != create-db then
-  return -1
-endi
-
-if $data[0][7] != @Unable to establish connection@ then
-  return -1
-endi
-
+return
 sql_error create database d2 vgroups 2;
 
 print =============== kill transaction