Merge branch '3.0' of https://github.com/taosdata/TDengine into feat/tsdb_refact

docs/examples/.gitignre

@@ -0,0 +1,2 @@
+.vscode
+*.lock

include/common/tmsg.h

@@ -2001,16 +2001,17 @@ typedef struct {
 
 typedef struct {
   int64_t tid;
-  int32_t status;
+  char    status[TSDB_JOB_STATUS_LEN];
 } SQuerySubDesc;
 
 typedef struct {
   char     sql[TSDB_SHOW_SQL_LEN];
   uint64_t queryId;
   int64_t  useconds;
-  int64_t  stime;
+  int64_t  stime;            // timestamp precision ms
   int64_t  reqRid;
   int32_t  pid;
+  bool     stableQuery;
   char     fqdn[TSDB_FQDN_LEN];
   int32_t  subPlanNum;
   SArray*  subDesc;  // SArray<SQuerySubDesc>
@@ -2255,6 +2256,25 @@ typedef struct {
   int8_t reserved;
 } SMqVDeleteRsp;
 
+typedef struct {
+  char    name[TSDB_STREAM_FNAME_LEN];
+  int64_t streamId;
+} SMDropStreamTaskReq;
+
+typedef struct {
+  int8_t reserved;
+} SMDropStreamTaskRsp;
+
+typedef struct {
+  SMsgHead head;
+  int64_t  leftForVer;
+  int32_t  taskId;
+} SVDropStreamTaskReq;
+
+typedef struct {
+  int8_t reserved;
+} SVDropStreamTaskRsp;
+
 typedef struct {
   int64_t leftForVer;
   int32_t vgId;

include/common/tmsgdef.h

@@ -187,6 +187,7 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_VND_CONSUME, "vnode-consume", SMqPollReq, SMqDataBlkRsp)
   TD_DEF_MSG_TYPE(TDMT_VND_STREAM_TRIGGER, "vnode-stream-trigger", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_STREAM_DISPATCH_WRITE, "vnode-stream-task-dispatch-write", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_VND_STREAM_TASK_DROP, "vnode-stream-task-drop", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_CREATE_SMA, "vnode-create-sma", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_CANCEL_SMA, "vnode-cancel-sma", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_DROP_SMA, "vnode-drop-sma", NULL, NULL)

include/libs/function/function.h

@@ -132,7 +132,6 @@ typedef struct SqlFunctionCtx {
   char                  *pOutput;       // final result output buffer, point to sdata->data
   int32_t                numOfParams;
   SFunctParam           *param;         // input parameter, e.g., top(k, 20), the number of results for top query is kept in param
-  int64_t               *ptsList;       // corresponding timestamp array list, todo remove it
   SColumnInfoData       *pTsOutput;     // corresponding output buffer for timestamp of each result, e.g., top/bottom*/
   int32_t                offset;
   struct  SResultRowEntryInfo *resultInfo;

include/libs/function/functionMgt.h

@@ -134,6 +134,7 @@ typedef enum EFunctionType {
   FUNCTION_TYPE_HYPERLOGLOG_MERGE,
   FUNCTION_TYPE_ELAPSED_PARTIAL,
   FUNCTION_TYPE_ELAPSED_MERGE,
+
   FUNCTION_TYPE_TOP_PARTIAL,
   FUNCTION_TYPE_TOP_MERGE,
   FUNCTION_TYPE_BOTTOM_PARTIAL,
@@ -184,6 +185,7 @@ bool fmIsUserDefinedFunc(int32_t funcId);
 bool fmIsDistExecFunc(int32_t funcId);
 bool fmIsForbidFillFunc(int32_t funcId);
 bool fmIsForbidStreamFunc(int32_t funcId);
+bool fmIsIntervalInterpoFunc(int32_t funcId);
 
 int32_t fmGetDistMethod(const SFunctionNode* pFunc, SFunctionNode** pPartialFunc, SFunctionNode** pMergeFunc);
 

include/libs/nodes/querynodes.h

@@ -351,6 +351,7 @@ typedef struct SQuery {
   int32_t        placeholderNum;
   SArray*        pPlaceholderValues;
   SNode*         pPrepareRoot;
+  bool           stableQuery;
 } SQuery;
 
 void nodesWalkSelectStmt(SSelectStmt* pSelect, ESqlClause clause, FNodeWalker walker, void* pContext);

include/libs/planner/planner.h

@@ -36,7 +36,7 @@ typedef struct SPlanContext {
   int64_t  watermark;
   char*    pMsg;
   int32_t  msgLen;
-  double   filesFactor;
+  // double   filesFactor;
 } SPlanContext;
 
 // Create the physical plan for the query, according to the AST.

include/util/taoserror.h

@@ -438,6 +438,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_WAL_SIZE_LIMIT                TAOS_DEF_ERROR_CODE(0, 0x1002)
 #define TSDB_CODE_WAL_INVALID_VER               TAOS_DEF_ERROR_CODE(0, 0x1003)
 #define TSDB_CODE_WAL_OUT_OF_MEMORY             TAOS_DEF_ERROR_CODE(0, 0x1004)
+#define TSDB_CODE_WAL_LOG_NOT_EXIST             TAOS_DEF_ERROR_CODE(0, 0x1005)
 
 // http
 #define TSDB_CODE_HTTP_SERVER_OFFLINE           TAOS_DEF_ERROR_CODE(0, 0x1100)  //"http server is not online"

include/util/tdef.h

@@ -243,6 +243,7 @@ typedef enum ELogicConditionType {
 #define TSDB_USET_PASSWORD_LEN 129
 #define TSDB_VERSION_LEN       12
 #define TSDB_LABEL_LEN         8
+#define TSDB_JOB_STATUS_LEN    32
 
 #define TSDB_CLUSTER_ID_LEN     40
 #define TSDB_FQDN_LEN           128

source/client/inc/clientInt.h

@@ -213,6 +213,7 @@ typedef struct SRequestObj {
   SArray*              tableList;
   SQueryExecMetric     metric;
   SRequestSendRecvBody body;
+  bool                 stableQuery;
 
   uint32_t             prevCode; //previous error code: todo refactor, add update flag for catalog
   uint32_t             retry;
@@ -294,7 +295,7 @@ void* openTransporter(const char* user, const char* auth, int32_t numOfThreads);
 bool persistConnForSpecificMsg(void* parenct, tmsg_t msgType);
 void processMsgFromServer(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet);
 
-TAOS* taos_connect_internal(const char* ip, const char* user, const char* pass, const char* auth, const char* db,
+STscObj* taos_connect_internal(const char* ip, const char* user, const char* pass, const char* auth, const char* db,
                             uint16_t port, int connType);
 
 SRequestObj* launchQuery(STscObj* pTscObj, const char* sql, int sqlLen);
@@ -305,6 +306,8 @@ int32_t getPlan(SRequestObj* pRequest, SQuery* pQuery, SQueryPlan** pPlan, SArra
 
 int32_t buildRequest(STscObj* pTscObj, const char* sql, int sqlLen, SRequestObj** pRequest);
 
+void taos_close_internal(void *taos);
+
 // --- heartbeat
 // global, called by mgmt
 int  hbMgrInit();

source/client/inc/clientStmt.h

@@ -107,7 +107,7 @@ typedef struct STscStmt {
 #define STMT_RET(c) do { int32_t _code = c; if (_code != TSDB_CODE_SUCCESS) { terrno = _code; } return _code; } while (0)
 #define STMT_ERR_JRET(c) do { code = c; if (code != TSDB_CODE_SUCCESS) { terrno = code; goto _return; } } while (0)
 
-TAOS_STMT *stmtInit(TAOS *taos);
+TAOS_STMT *stmtInit(STscObj* taos);
 int stmtClose(TAOS_STMT *stmt);
 int stmtExec(TAOS_STMT *stmt);
 const char *stmtErrstr(TAOS_STMT *stmt);

source/client/src/clientHb.c

@@ -174,7 +174,7 @@ static int32_t hbQueryHbRspHandle(SAppHbMgr *pAppHbMgr, SClientHbRsp *pRsp) {
       }
 
       if (pRsp->query->killConnection) {
-        taos_close(pTscObj);
+        taos_close_internal(pTscObj);
       }
 
       if (pRsp->query->pQnodeList) {
@@ -310,11 +310,12 @@ int32_t hbBuildQueryDesc(SQueryHbReqBasic *hbBasic, STscObj *pObj) {
     }
 
     tstrncpy(desc.sql, pRequest->sqlstr, sizeof(desc.sql));
-    desc.stime = pRequest->metric.start;
+    desc.stime = pRequest->metric.start / 1000;
     desc.queryId = pRequest->requestId;
     desc.useconds = now - pRequest->metric.start;
     desc.reqRid = pRequest->self;
     desc.pid = hbBasic->pid;
+    desc.stableQuery = pRequest->stableQuery;
     taosGetFqdn(desc.fqdn);
     desc.subPlanNum = pRequest->body.pDag ? pRequest->body.pDag->numOfSubplans : 0;
 
@@ -329,6 +330,7 @@ int32_t hbBuildQueryDesc(SQueryHbReqBasic *hbBasic, STscObj *pObj) {
       if (code) {
         taosArrayDestroy(desc.subDesc);
         desc.subDesc = NULL;
+        desc.subPlanNum = 0;
       }
     } else {
       desc.subDesc = NULL;
@@ -350,19 +352,24 @@ int32_t hbGetQueryBasicInfo(SClientHbKey *connKey, SClientHbReq *req) {
     return TSDB_CODE_QRY_APP_ERROR;
   }
 
-  int32_t numOfQueries = pTscObj->pRequests ? taosHashGetSize(pTscObj->pRequests) : 0;
-  if (numOfQueries <= 0) {
-    releaseTscObj(connKey->tscRid);
-    tscDebug("no queries on connection");
-    return TSDB_CODE_QRY_APP_ERROR;
-  }
-
   SQueryHbReqBasic *hbBasic = (SQueryHbReqBasic *)taosMemoryCalloc(1, sizeof(SQueryHbReqBasic));
   if (NULL == hbBasic) {
     tscError("calloc %d failed", (int32_t)sizeof(SQueryHbReqBasic));
     releaseTscObj(connKey->tscRid);
     return TSDB_CODE_QRY_OUT_OF_MEMORY;
   }
+  
+  hbBasic->connId = pTscObj->connId;
+  hbBasic->pid = taosGetPId();
+  taosGetAppName(hbBasic->app, NULL);
+
+  int32_t numOfQueries = pTscObj->pRequests ? taosHashGetSize(pTscObj->pRequests) : 0;
+  if (numOfQueries <= 0) {
+    req->query = hbBasic;
+    releaseTscObj(connKey->tscRid);
+    tscDebug("no queries on connection");
+    return TSDB_CODE_SUCCESS;
+  }
 
   hbBasic->queryDesc = taosArrayInit(numOfQueries, sizeof(SQueryDesc));
   if (NULL == hbBasic->queryDesc) {
@@ -372,9 +379,6 @@ int32_t hbGetQueryBasicInfo(SClientHbKey *connKey, SClientHbReq *req) {
     return TSDB_CODE_QRY_OUT_OF_MEMORY;
   }
 
-  hbBasic->connId = pTscObj->connId;
-  hbBasic->pid = taosGetPId();
-  taosGetAppName(hbBasic->app, NULL);
 
   int32_t code = hbBuildQueryDesc(hbBasic, pTscObj);
   if (code) {

source/client/src/clientImpl.c

@@ -58,7 +58,7 @@ static char* getClusterKey(const char* user, const char* auth, const char* ip, i
 static STscObj* taosConnectImpl(const char* user, const char* auth, const char* db, __taos_async_fn_t fp, void* param,
                                 SAppInstInfo* pAppInfo, int connType);
 
-TAOS* taos_connect_internal(const char* ip, const char* user, const char* pass, const char* auth, const char* db,
+STscObj* taos_connect_internal(const char* ip, const char* user, const char* pass, const char* auth, const char* db,
                             uint16_t port, int connType) {
   if (taos_init() != TSDB_CODE_SUCCESS) {
     return NULL;
@@ -692,6 +692,8 @@ SRequestObj* launchQuery(STscObj* pTscObj, const char* sql, int sqlLen) {
     return pRequest;
   }
 
+  pRequest->stableQuery = pQuery->stableQuery;
+
   return launchQueryImpl(pRequest, pQuery, false, NULL);
 }
 
@@ -917,7 +919,7 @@ STscObj* taosConnectImpl(const char* user, const char* auth, const char* db, __t
 
     terrno = pRequest->code;
     destroyRequest(pRequest);
-    taos_close(pTscObj);
+    taos_close_internal(pTscObj);
     pTscObj = NULL;
   } else {
     tscDebug("0x%" PRIx64 " connection is opening, connId:%u, dnodeConn:%p, reqId:0x%" PRIx64, pTscObj->id,
@@ -952,8 +954,8 @@ static SMsgSendInfo* buildConnectMsg(SRequestObj* pRequest) {
   taosMemoryFreeClear(db);
 
   connectReq.connType = pObj->connType;
-  connectReq.pid = htonl(appInfo.pid);
+  connectReq.pid = appInfo.pid;
-  connectReq.startTime = htobe64(appInfo.startTime);
+  connectReq.startTime = appInfo.startTime;
 
   tstrncpy(connectReq.app, appInfo.appName, sizeof(connectReq.app));
   tstrncpy(connectReq.user, pObj->user, sizeof(connectReq.user));
@@ -1081,7 +1083,12 @@ TAOS* taos_connect_auth(const char* ip, const char* user, const char* auth, cons
     return NULL;
   }
 
-  return taos_connect_internal(ip, user, NULL, auth, db, port, CONN_TYPE__QUERY);
+  STscObj* pObj = taos_connect_internal(ip, user, NULL, auth, db, port, CONN_TYPE__QUERY);
+  if (pObj) {
+    return (TAOS*)pObj->id;
+  }
+  
+  return (TAOS*)0;
 }
 
 TAOS* taos_connect_l(const char* ip, int ipLen, const char* user, int userLen, const char* pass, int passLen,

source/client/src/clientMain.c

@@ -97,10 +97,15 @@ TAOS *taos_connect(const char *ip, const char *user, const char *pass, const cha
     pass = TSDB_DEFAULT_PASS;
   }
 
-  return taos_connect_internal(ip, user, pass, NULL, db, port, CONN_TYPE__QUERY);
+  STscObj* pObj = taos_connect_internal(ip, user, pass, NULL, db, port, CONN_TYPE__QUERY);
+  if (pObj) {
+    return (TAOS*)pObj->id;
+  }
+  
+  return (TAOS*)0;
 }
 
-void taos_close(TAOS *taos) {
+void taos_close_internal(void *taos) {
   if (taos == NULL) {
     return;
   }
@@ -111,6 +116,17 @@ void taos_close(TAOS *taos) {
   taosRemoveRef(clientConnRefPool, pTscObj->id);
 }
 
+void taos_close(TAOS *taos) {
+  STscObj* pObj = acquireTscObj((int64_t)taos);
+  if (NULL == pObj) {
+    return;
+  }
+  
+  taos_close_internal(pObj);
+  releaseTscObj((int64_t)taos);
+}
+
+
 int taos_errno(TAOS_RES *tres) {
   if (tres == NULL) {
     return terrno;
@@ -190,29 +206,36 @@ static void syncQueryFn(void *param, void *res, int32_t code) {
 }
 
 TAOS_RES *taos_query(TAOS *taos, const char *sql) {
-  if (taos == NULL || sql == NULL) {
+  STscObj* pTscObj = acquireTscObj((int64_t)taos);
+  if (pTscObj == NULL || sql == NULL) {
+    terrno = TSDB_CODE_TSC_DISCONNECTED;
     return NULL;
   }
 
-  STscObj *pTscObj = (STscObj *)taos;
-
 #if SYNC_ON_TOP_OF_ASYNC
   SSyncQueryParam *param = taosMemoryCalloc(1, sizeof(SSyncQueryParam));
   tsem_init(&param->sem, 0, 0);
 
-  taos_query_a(pTscObj, sql, syncQueryFn, param);
+  taos_query_a(taos, sql, syncQueryFn, param);
   tsem_wait(&param->sem);
 
+  releaseTscObj((int64_t)taos);
+
   return param->pRequest;
 #else
   size_t sqlLen = strlen(sql);
   if (sqlLen > (size_t)TSDB_MAX_ALLOWED_SQL_LEN) {
+    releaseTscObj((int64_t)taos);
     tscError("sql string exceeds max length:%d", TSDB_MAX_ALLOWED_SQL_LEN);
     terrno = TSDB_CODE_TSC_EXCEED_SQL_LIMIT;
     return NULL;
   }
 
-  return execQuery(pTscObj, sql, sqlLen);
+  TAOS_RES* pRes = execQuery(pTscObj, sql, sqlLen);
+
+  releaseTscObj((int64_t)taos);
+
+  return pRes;
 #endif
 }
 
@@ -429,13 +452,15 @@ int taos_result_precision(TAOS_RES *res) {
 }
 
 int taos_select_db(TAOS *taos, const char *db) {
-  STscObj *pObj = (STscObj *)taos;
+  STscObj* pObj = acquireTscObj((int64_t)taos);
   if (pObj == NULL) {
+    releaseTscObj((int64_t)taos);
     terrno = TSDB_CODE_TSC_DISCONNECTED;
     return TSDB_CODE_TSC_DISCONNECTED;
   }
 
   if (db == NULL || strlen(db) == 0) {
+    releaseTscObj((int64_t)taos);
     terrno = TSDB_CODE_TSC_INVALID_INPUT;
     return terrno;
   }
@@ -447,6 +472,7 @@ int taos_select_db(TAOS *taos, const char *db) {
   int32_t   code = taos_errno(pRequest);
 
   taos_free_result(pRequest);
+  releaseTscObj((int64_t)taos);
   return code;
 }
 
@@ -593,19 +619,26 @@ int *taos_get_column_data_offset(TAOS_RES *res, int columnIndex) {
 int taos_validate_sql(TAOS *taos, const char *sql) { return true; }
 
 void taos_reset_current_db(TAOS *taos) {
-  if (taos == NULL) {
+  STscObj* pTscObj = acquireTscObj((int64_t)taos);
+  if (pTscObj == NULL) {
+    terrno = TSDB_CODE_TSC_DISCONNECTED;
     return;
   }
 
-  resetConnectDB(taos);
+  resetConnectDB(pTscObj);
+
+  releaseTscObj((int64_t)taos);
 }
 
 const char *taos_get_server_info(TAOS *taos) {
-  if (taos == NULL) {
+  STscObj* pTscObj = acquireTscObj((int64_t)taos);
+  if (pTscObj == NULL) {
+    terrno = TSDB_CODE_TSC_DISCONNECTED;
     return NULL;
   }
 
-  STscObj *pTscObj = (STscObj *)taos;
+  releaseTscObj((int64_t)taos);
+
   return pTscObj->ver;
 }
 
@@ -637,6 +670,7 @@ void retrieveMetaCallback(SMetaData *pResultMeta, void *param, int32_t code) {
 
   if (code == TSDB_CODE_SUCCESS) {
     code = qAnalyseSqlSemantic(pWrapper->pCtx, &pWrapper->catalogReq, pResultMeta, pQuery);
+    pRequest->stableQuery = pQuery->stableQuery;
   }
 
   if (code == TSDB_CODE_SUCCESS) {
@@ -670,10 +704,14 @@ void retrieveMetaCallback(SMetaData *pResultMeta, void *param, int32_t code) {
 }
 
 void taos_query_a(TAOS *taos, const char *sql, __taos_async_fn_t fp, void *param) {
-  ASSERT(fp != NULL);
+  STscObj* pTscObj = acquireTscObj((int64_t)taos);
-
+  if (pTscObj == NULL || sql == NULL || NULL == fp) {
-  if (taos == NULL || sql == NULL) {
     terrno = TSDB_CODE_INVALID_PARA;
+    if (pTscObj) {
+      releaseTscObj((int64_t)taos);
+    } else {
+      terrno = TSDB_CODE_TSC_DISCONNECTED;
+    }
     fp(param, NULL, terrno);
     return;
   }
@@ -688,7 +726,7 @@ void taos_query_a(TAOS *taos, const char *sql, __taos_async_fn_t fp, void *param
   }
 
   SRequestObj *pRequest = NULL;
-  int32_t      code = buildRequest(taos, sql, sqlLen, &pRequest);
+  int32_t code = buildRequest(pTscObj, sql, sqlLen, &pRequest);
   if (code != TSDB_CODE_SUCCESS) {
     terrno = code;
     fp(param, NULL, terrno);
@@ -888,13 +926,18 @@ int taos_load_table_info(TAOS *taos, const char *tableNameList) {
 }
 
 TAOS_STMT *taos_stmt_init(TAOS *taos) {
-  if (taos == NULL) {
+  STscObj* pObj = acquireTscObj((int64_t)taos);
-    tscError("NULL parameter for %s", __FUNCTION__);
+  if (NULL == pObj) {
-    terrno = TSDB_CODE_INVALID_PARA;
+    tscError("invalid parameter for %s", __FUNCTION__);
+    terrno = TSDB_CODE_TSC_DISCONNECTED;
     return NULL;
   }
 
-  return stmtInit(taos);
+  TAOS_STMT* pStmt = stmtInit(pObj);
+  
+  releaseTscObj((int64_t)taos);
+
+  return pStmt;
 }
 
 int taos_stmt_prepare(TAOS_STMT *stmt, const char *sql, unsigned long length) {

source/client/src/clientSml.c

@@ -309,7 +309,7 @@ static int32_t smlApplySchemaAction(SSmlHandle *info, SSchemaAction *action) {
     case SCHEMA_ACTION_ADD_COLUMN: {
       int n = sprintf(result, "alter stable `%s` add column ", action->alterSTable.sTableName);
       smlBuildColumnDescription(action->alterSTable.field, result + n, capacity - n, &outBytes);
-      TAOS_RES *res = taos_query(info->taos, result);  // TODO async doAsyncQuery
+      TAOS_RES *res = taos_query((TAOS*)info->taos->id, result);  // TODO async doAsyncQuery
       code = taos_errno(res);
       const char *errStr = taos_errstr(res);
       if (code != TSDB_CODE_SUCCESS) {
@@ -323,7 +323,7 @@ static int32_t smlApplySchemaAction(SSmlHandle *info, SSchemaAction *action) {
     case SCHEMA_ACTION_ADD_TAG: {
       int n = sprintf(result, "alter stable `%s` add tag ", action->alterSTable.sTableName);
       smlBuildColumnDescription(action->alterSTable.field, result + n, capacity - n, &outBytes);
-      TAOS_RES *res = taos_query(info->taos, result);  // TODO async doAsyncQuery
+      TAOS_RES *res = taos_query((TAOS*)info->taos->id, result);  // TODO async doAsyncQuery
       code = taos_errno(res);
       const char *errStr = taos_errstr(res);
       if (code != TSDB_CODE_SUCCESS) {
@@ -337,7 +337,7 @@ static int32_t smlApplySchemaAction(SSmlHandle *info, SSchemaAction *action) {
     case SCHEMA_ACTION_CHANGE_COLUMN_SIZE: {
       int n = sprintf(result, "alter stable `%s` modify column ", action->alterSTable.sTableName);
       smlBuildColumnDescription(action->alterSTable.field, result + n, capacity - n, &outBytes);
-      TAOS_RES *res = taos_query(info->taos, result);  // TODO async doAsyncQuery
+      TAOS_RES *res = taos_query((TAOS*)info->taos->id, result);  // TODO async doAsyncQuery
       code = taos_errno(res);
       if (code != TSDB_CODE_SUCCESS) {
         uError("SML:0x%" PRIx64 " apply schema action. error : %s", info->id, taos_errstr(res));
@@ -350,7 +350,7 @@ static int32_t smlApplySchemaAction(SSmlHandle *info, SSchemaAction *action) {
     case SCHEMA_ACTION_CHANGE_TAG_SIZE: {
       int n = sprintf(result, "alter stable `%s` modify tag ", action->alterSTable.sTableName);
       smlBuildColumnDescription(action->alterSTable.field, result + n, capacity - n, &outBytes);
-      TAOS_RES *res = taos_query(info->taos, result);  // TODO async doAsyncQuery
+      TAOS_RES *res = taos_query((TAOS*)info->taos->id, result);  // TODO async doAsyncQuery
       code = taos_errno(res);
       if (code != TSDB_CODE_SUCCESS) {
         uError("SML:0x%" PRIx64 " apply schema action. error : %s", info->id, taos_errstr(res));
@@ -405,7 +405,7 @@ static int32_t smlApplySchemaAction(SSmlHandle *info, SSchemaAction *action) {
       pos--;
       ++freeBytes;
       outBytes = snprintf(pos, freeBytes, ")");
-      TAOS_RES *res = taos_query(info->taos, result);
+      TAOS_RES *res = taos_query((TAOS*)info->taos->id, result);
       code = taos_errno(res);
       if (code != TSDB_CODE_SUCCESS) {
         uError("SML:0x%" PRIx64 " apply schema action. error : %s", info->id, taos_errstr(res));
@@ -1453,9 +1453,9 @@ static void smlDestroyInfo(SSmlHandle *info) {
   taosMemoryFreeClear(info);
 }
 
-static SSmlHandle *smlBuildSmlInfo(TAOS *taos, SRequestObj *request, SMLProtocolType protocol, int8_t precision) {
+static SSmlHandle* smlBuildSmlInfo(STscObj* pTscObj, SRequestObj* request, SMLProtocolType protocol, int8_t precision){
-  int32_t     code = TSDB_CODE_SUCCESS;
+  int32_t code = TSDB_CODE_SUCCESS;
-  SSmlHandle *info = (SSmlHandle *)taosMemoryCalloc(1, sizeof(SSmlHandle));
+  SSmlHandle* info = (SSmlHandle*)taosMemoryCalloc(1, sizeof(SSmlHandle));
   if (NULL == info) {
     return NULL;
   }
@@ -1476,7 +1476,7 @@ static SSmlHandle *smlBuildSmlInfo(TAOS *taos, SRequestObj *request, SMLProtocol
   }
   ((SVnodeModifOpStmt *)(info->pQuery->pRoot))->payloadType = PAYLOAD_TYPE_KV;
 
-  info->taos = (STscObj *)taos;
+  info->taos        = pTscObj;
   code = catalogGetHandle(info->taos->pAppInfo->clusterId, &info->pCatalog);
   if (code != TSDB_CODE_SUCCESS) {
     uError("SML:0x%" PRIx64 " get catalog error %d", info->id, code);
@@ -2433,14 +2433,22 @@ static void smlInsertCallback(void *param, void *res, int32_t code) {
  *
  */
 
-TAOS_RES *taos_schemaless_insert(TAOS *taos, char *lines[], int numLines, int protocol, int precision) {
+TAOS_RES* taos_schemaless_insert(TAOS* taos, char* lines[], int numLines, int protocol, int precision) {
-  SRequestObj *request = (SRequestObj *)createRequest((STscObj *)taos, TSDB_SQL_INSERT);
+  STscObj* pTscObj = acquireTscObj((int64_t)taos);
-  if (!request) {
+  if (NULL == pTscObj) {
+    terrno = TSDB_CODE_TSC_DISCONNECTED;
+    uError("SML:taos_schemaless_insert invalid taos");
+    return NULL;
+  }
+  
+  SRequestObj* request = (SRequestObj*)createRequest(pTscObj, TSDB_SQL_INSERT);
+  if(!request){
+    releaseTscObj((int64_t)taos);
     uError("SML:taos_schemaless_insert error request is null");
     return NULL;
   }
 
-  ((STscObj *)taos)->schemalessType = 1;
+  pTscObj->schemalessType = 1;
   SSmlMsgBuf msg = {ERROR_MSG_BUF_DEFAULT_SIZE, request->msgBuf};
 
   int    cnt = ceil(((double)numLines) / LINE_BATCH);
@@ -2455,7 +2463,7 @@ TAOS_RES *taos_schemaless_insert(TAOS *taos, char *lines[], int numLines, int pr
     goto end;
   }
 
-  if (isSchemalessDb(((STscObj *)taos), request) != TSDB_CODE_SUCCESS) {
+  if(isSchemalessDb(pTscObj, request) != TSDB_CODE_SUCCESS){
     request->code = TSDB_CODE_SML_INVALID_DB_CONF;
     smlBuildInvalidDataMsg(&msg, "Cannot write data to a non schemaless database", NULL);
     goto end;
@@ -2481,14 +2489,14 @@ TAOS_RES *taos_schemaless_insert(TAOS *taos, char *lines[], int numLines, int pr
   }
 
   for (int i = 0; i < cnt; ++i) {
-    SRequestObj *req = (SRequestObj *)createRequest((STscObj *)taos, TSDB_SQL_INSERT);
+    SRequestObj* req = (SRequestObj*)createRequest(pTscObj, TSDB_SQL_INSERT);
-    if (!req) {
+    if(!req){
       request->code = TSDB_CODE_OUT_OF_MEMORY;
       uError("SML:taos_schemaless_insert error request is null");
       goto end;
     }
-    SSmlHandle *info = smlBuildSmlInfo(taos, req, (SMLProtocolType)protocol, precision);
+    SSmlHandle* info = smlBuildSmlInfo(pTscObj, req, (SMLProtocolType)protocol, precision);
-    if (!info) {
+    if(!info){
       request->code = TSDB_CODE_OUT_OF_MEMORY;
       uError("SML:taos_schemaless_insert error SSmlHandle is null");
       goto end;
@@ -2520,8 +2528,9 @@ TAOS_RES *taos_schemaless_insert(TAOS *taos, char *lines[], int numLines, int pr
 end:
   taosThreadSpinDestroy(&params.lock);
   tsem_destroy(&params.sem);
-  //  ((STscObj *)taos)->schemalessType = 0;
+//  ((STscObj *)taos)->schemalessType = 0;
-  ((STscObj *)taos)->schemalessType = 1;
+  pTscObj->schemalessType = 1;
   uDebug("resultend:%s", request->msgBuf);
-  return (TAOS_RES *)request;
+  releaseTscObj((int64_t)taos);
+  return (TAOS_RES*)request;
 }

source/client/src/clientStmt.c

@@ -478,7 +478,7 @@ int32_t stmtResetStmt(STscStmt* pStmt) {
   return TSDB_CODE_SUCCESS;
 }
 
-TAOS_STMT* stmtInit(TAOS* taos) {
+TAOS_STMT* stmtInit(STscObj* taos) {
   STscObj*  pObj = (STscObj*)taos;
   STscStmt* pStmt = NULL;
 

source/common/src/systable.c

@@ -257,8 +257,8 @@ static const SSysTableMeta infosMeta[] = {
 static const SSysDbTableSchema connectionsSchema[] = {
     {.name = "conn_id", .bytes = 4, .type = TSDB_DATA_TYPE_UINT},
     {.name = "user", .bytes = TSDB_USER_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_BINARY},
-    {.name = "program", .bytes = TSDB_APP_NAME_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_BINARY},
+    {.name = "app", .bytes = TSDB_APP_NAME_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_BINARY},
-    {.name = "pid", .bytes = 4, .type = TSDB_DATA_TYPE_INT},
+    {.name = "pid", .bytes = 4, .type = TSDB_DATA_TYPE_UINT},
     {.name = "end_point", .bytes = TSDB_EP_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_BINARY},
     {.name = "login_time", .bytes = 8, .type = TSDB_DATA_TYPE_TIMESTAMP},
     {.name = "last_access", .bytes = 8, .type = TSDB_DATA_TYPE_TIMESTAMP},
@@ -302,19 +302,18 @@ static const SSysDbTableSchema offsetSchema[] = {
 };
 
 static const SSysDbTableSchema querySchema[] = {
-    {.name = "query_id", .bytes = 4, .type = TSDB_DATA_TYPE_INT},
+    {.name = "query_id", .bytes = 26 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
-    {.name = "connId", .bytes = 4, .type = TSDB_DATA_TYPE_INT},
+    {.name = "req_id", .bytes = 8, .type = TSDB_DATA_TYPE_UBIGINT},
+    {.name = "connId", .bytes = 4, .type = TSDB_DATA_TYPE_UINT},
+    {.name = "app", .bytes = TSDB_APP_NAME_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},    
+    {.name = "pid", .bytes = 4, .type = TSDB_DATA_TYPE_INT},
     {.name = "user", .bytes = TSDB_USER_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
     {.name = "end_point", .bytes = TSDB_IPv4ADDR_LEN + 6 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
-    {.name = "qid", .bytes = 22 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
     {.name = "create_time", .bytes = 8, .type = TSDB_DATA_TYPE_TIMESTAMP},
-    {.name = "time", .bytes = 8, .type = TSDB_DATA_TYPE_BIGINT},
+    {.name = "exec_usec", .bytes = 8, .type = TSDB_DATA_TYPE_BIGINT},
-    {.name = "sql_obj_id", .bytes = QUERY_OBJ_ID_SIZE + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
-    {.name = "pid", .bytes = 4, .type = TSDB_DATA_TYPE_INT},
-    {.name = "ep", .bytes = TSDB_EP_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
     {.name = "stable_query", .bytes = 1, .type = TSDB_DATA_TYPE_BOOL},
-    {.name = "sub_queries", .bytes = 4, .type = TSDB_DATA_TYPE_INT},
+    {.name = "sub_num", .bytes = 4, .type = TSDB_DATA_TYPE_INT},
-    {.name = "sub_query_info", .bytes = TSDB_SHOW_SUBQUERY_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
+    {.name = "sub_status", .bytes = TSDB_SHOW_SUBQUERY_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
     {.name = "sql", .bytes = TSDB_SHOW_SQL_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
 };
 

source/common/src/tmsg.c

@@ -210,6 +210,7 @@ static int32_t tSerializeSClientHbReq(SEncoder *pEncoder, const SClientHbReq *pR
         if (tEncodeI64(pEncoder, desc->stime) < 0) return -1;
         if (tEncodeI64(pEncoder, desc->reqRid) < 0) return -1;
         if (tEncodeI32(pEncoder, desc->pid) < 0) return -1;
+        if (tEncodeI8(pEncoder, desc->stableQuery) < 0) return -1;
         if (tEncodeCStr(pEncoder, desc->fqdn) < 0) return -1;
         if (tEncodeI32(pEncoder, desc->subPlanNum) < 0) return -1;
 
@@ -218,7 +219,7 @@ static int32_t tSerializeSClientHbReq(SEncoder *pEncoder, const SClientHbReq *pR
         for (int32_t m = 0; m < snum; ++m) {
           SQuerySubDesc *sDesc = taosArrayGet(desc->subDesc, m);
           if (tEncodeI64(pEncoder, sDesc->tid) < 0) return -1;
-          if (tEncodeI32(pEncoder, sDesc->status) < 0) return -1;
+          if (tEncodeCStr(pEncoder, sDesc->status) < 0) return -1;
         }
       }
     } else {
@@ -265,6 +266,7 @@ static int32_t tDeserializeSClientHbReq(SDecoder *pDecoder, SClientHbReq *pReq)
           if (tDecodeI64(pDecoder, &desc.stime) < 0) return -1;
           if (tDecodeI64(pDecoder, &desc.reqRid) < 0) return -1;
           if (tDecodeI32(pDecoder, &desc.pid) < 0) return -1;
+          if (tDecodeI8(pDecoder, (int8_t*)&desc.stableQuery) < 0) return -1;
           if (tDecodeCStrTo(pDecoder, desc.fqdn) < 0) return -1;
           if (tDecodeI32(pDecoder, &desc.subPlanNum) < 0) return -1;
 
@@ -277,7 +279,7 @@ static int32_t tDeserializeSClientHbReq(SDecoder *pDecoder, SClientHbReq *pReq)
             for (int32_t m = 0; m < snum; ++m) {
               SQuerySubDesc sDesc = {0};
               if (tDecodeI64(pDecoder, &sDesc.tid) < 0) return -1;
-              if (tDecodeI32(pDecoder, &sDesc.status) < 0) return -1;
+              if (tDecodeCStrTo(pDecoder, sDesc.status) < 0) return -1;
               taosArrayPush(desc.subDesc, &sDesc);
             }
           }

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

@@ -215,6 +215,7 @@ SArray *mmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_VG_DELETE_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_DROP_TASK, mmPutMsgToQueryQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DEPLOY_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_STREAM_TASK_DROP_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_CONFIG_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_REPLICA_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_CONFIRM_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;

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

@@ -350,6 +350,7 @@ SArray *vmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_VND_QUERY_HEARTBEAT, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
 
   if (dmSetMgmtHandle(pArray, TDMT_VND_STREAM_TRIGGER, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_STREAM_TASK_DROP, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DEPLOY, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RUN, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DISPATCH, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;

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

@@ -39,6 +39,7 @@ typedef struct {
   int64_t  lastAccessTimeMs;
   uint64_t killId;
   int32_t  numOfQueries;
+  SRWLatch queryLock;
   SArray  *pQueries;  // SArray<SQueryDesc>
 } SConnObj;
 
@@ -53,8 +54,8 @@ static int32_t   mndProcessHeartBeatReq(SRpcMsg *pReq);
 static int32_t   mndProcessConnectReq(SRpcMsg *pReq);
 static int32_t   mndProcessKillQueryReq(SRpcMsg *pReq);
 static int32_t   mndProcessKillConnReq(SRpcMsg *pReq);
-static int32_t   mndRetrieveConns(SRpcMsg *pReq, SShowObj *pShow, char *data, int32_t rows);
+static int32_t   mndRetrieveConns(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBlock, int32_t rows);
-static int32_t   mndRetrieveQueries(SRpcMsg *pReq, SShowObj *pShow, char *data, int32_t rows);
+static int32_t   mndRetrieveQueries(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBlock, int32_t rows);
 static void      mndCancelGetNextQuery(SMnode *pMnode, void *pIter);
 
 int32_t mndInitProfile(SMnode *pMnode) {
@@ -74,9 +75,9 @@ int32_t mndInitProfile(SMnode *pMnode) {
   mndSetMsgHandle(pMnode, TDMT_MND_KILL_QUERY, mndProcessKillQueryReq);
   mndSetMsgHandle(pMnode, TDMT_MND_KILL_CONN, mndProcessKillConnReq);
 
-  //  mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_CONNS, mndRetrieveConns);
+  mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_CONNS, mndRetrieveConns);
   mndAddShowFreeIterHandle(pMnode, TSDB_MGMT_TABLE_CONNS, mndCancelGetNextConn);
-  //  mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_QUERIES, mndRetrieveQueries);
+  mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_QUERIES, mndRetrieveQueries);
   mndAddShowFreeIterHandle(pMnode, TSDB_MGMT_TABLE_QUERIES, mndCancelGetNextQuery);
 
   return 0;
@@ -129,7 +130,9 @@ static SConnObj *mndCreateConn(SMnode *pMnode, const char *user, int8_t connType
 }
 
 static void mndFreeConn(SConnObj *pConn) {
+  taosWLockLatch(&pConn->queryLock);
   taosArrayDestroyEx(pConn->pQueries, tFreeClientHbQueryDesc);
+  taosWUnLockLatch(&pConn->queryLock);
   
   mTrace("conn:%u, is destroyed, data:%p", pConn->id, pConn);
 }
@@ -222,8 +225,6 @@ static int32_t mndProcessConnectReq(SRpcMsg *pReq) {
     goto CONN_OVER;
   }
 
-  mndAcquireConn(pMnode, pConn->id);
-
   SConnectRsp connectRsp = {0};
   connectRsp.acctId = pUser->acctId;
   connectRsp.superUser = pUser->superUser;
@@ -259,12 +260,17 @@ CONN_OVER:
 }
 
 static int32_t mndSaveQueryList(SConnObj *pConn, SQueryHbReqBasic *pBasic) {
+  taosWLockLatch(&pConn->queryLock);
+
   taosArrayDestroyEx(pConn->pQueries, tFreeClientHbQueryDesc);
 
   pConn->pQueries = pBasic->queryDesc;
+  pConn->numOfQueries = pBasic->queryDesc ? taosArrayGetSize(pBasic->queryDesc) : 0;
   pBasic->queryDesc = NULL;
 
-  pConn->numOfQueries = pBasic->queryDesc ? taosArrayGetSize(pBasic->queryDesc) : 0;
+  mDebug("queries updated in conn %d, num:%d", pConn->id, pConn->numOfQueries);
+
+  taosWUnLockLatch(&pConn->queryLock);
 
   return TSDB_CODE_SUCCESS;
 }
@@ -354,13 +360,8 @@ static int32_t mndProcessQueryHeartBeat(SMnode *pMnode, SRpcMsg *pMsg, SClientHb
       } else {
         mDebug("user:%s, conn:%u is freed and create a new conn:%u", connInfo.user, pBasic->connId, pConn->id);
       }
-    } else if (pConn->killed) {
-      mError("user:%s, conn:%u is already killed", connInfo.user, pConn->id);
-      mndReleaseConn(pMnode, pConn);
-      terrno = TSDB_CODE_MND_INVALID_CONNECTION;
-      return -1;
     }
-
+    
     SQueryHbRspBasic *rspBasic = taosMemoryCalloc(1, sizeof(SQueryHbRspBasic));
     if (rspBasic == NULL) {
       mndReleaseConn(pMnode, pConn);
@@ -389,6 +390,8 @@ static int32_t mndProcessQueryHeartBeat(SMnode *pMnode, SRpcMsg *pMsg, SClientHb
     mndReleaseConn(pMnode, pConn);
 
     hbRsp.query = rspBasic;
+  } else {
+    mDebug("no query info in hb msg");
   }
 
   int32_t kvNum = taosHashGetSize(pHbReq->info);
@@ -559,76 +562,13 @@ static int32_t mndProcessKillConnReq(SRpcMsg *pReq) {
   }
 }
 
-static int32_t mndRetrieveConns(SRpcMsg *pReq, SShowObj *pShow, char *data, int32_t rows) {
+static int32_t mndRetrieveConns(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBlock, int32_t rows) {
-  SMnode   *pMnode = pReq->info.node;
+  SMnode    *pMnode = pReq->info.node;
-  int32_t   numOfRows = 0;
+  SSdb      *pSdb = pMnode->pSdb;
-  SConnObj *pConn = NULL;
+  int32_t    numOfRows = 0;
-  int32_t   cols = 0;
+  int32_t    cols = 0;
-  char     *pWrite;
+  SConnObj  *pConn = NULL;
-  char      ipStr[TSDB_IPv4ADDR_LEN + 6];
+  
-
-  if (pShow->pIter == NULL) {
-    SProfileMgmt *pMgmt = &pMnode->profileMgmt;
-    pShow->pIter = taosCacheCreateIter(pMgmt->cache);
-  }
-
-  while (numOfRows < rows) {
-    pConn = mndGetNextConn(pMnode, pShow->pIter);
-    if (pConn == NULL) break;
-
-    cols = 0;
-#if 0
-    pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
-    *(uint32_t *)pWrite = pConn->id;
-    cols++;
-
-    pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
-    STR_WITH_MAXSIZE_TO_VARSTR(pWrite, pConn->user, pShow->pMeta->pSchemas[cols].bytes);
-    cols++;
-
-    // app name
-    pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
-    STR_WITH_MAXSIZE_TO_VARSTR(pWrite, pConn->app, pShow->pMeta->pSchemas[cols].bytes);
-    cols++;
-
-    // app pid
-    pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
-    *(int32_t *)pWrite = pConn->pid;
-    cols++;
-
-    pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
-    taosIpPort2String(pConn->ip, pConn->port, ipStr);
-    STR_WITH_MAXSIZE_TO_VARSTR(pWrite, ipStr, pShow->pMeta->pSchemas[cols].bytes);
-    cols++;
-
-    pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
-    *(int64_t *)pWrite = pConn->loginTimeMs;
-    cols++;
-
-    pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
-    if (pConn->lastAccessTimeMs < pConn->loginTimeMs) pConn->lastAccessTimeMs = pConn->loginTimeMs;
-    *(int64_t *)pWrite = pConn->lastAccessTimeMs;
-    cols++;
-#endif
-
-    numOfRows++;
-  }
-
-  pShow->numOfRows += numOfRows;
-
-  return numOfRows;
-}
-
-static int32_t mndRetrieveQueries(SRpcMsg *pReq, SShowObj *pShow, char *data, int32_t rows) {
-  SMnode *pMnode = pReq->info.node;
-  int32_t numOfRows = 0;
-#if 0
-  SConnObj *pConn = NULL;
-  int32_t   cols = 0;
-  char     *pWrite;
-  void     *pIter;
-  char      str[TSDB_IPv4ADDR_LEN + 6] = {0};
-
   if (pShow->pIter == NULL) {
     SProfileMgmt *pMgmt = &pMnode->profileMgmt;
     pShow->pIter = taosCacheCreateIter(pMgmt->cache);
@@ -641,85 +581,142 @@ static int32_t mndRetrieveQueries(SRpcMsg *pReq, SShowObj *pShow, char *data, in
       break;
     }
 
-    if (numOfRows + pConn->numOfQueries >= rows) {
+    cols = 0;
-      taosCacheDestroyIter(pShow->pIter);
+
+    SColumnInfoData *pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+    colDataAppend(pColInfo, numOfRows, (const char *)&pConn->id, false);
+
+    char user[TSDB_USER_LEN + VARSTR_HEADER_SIZE] = {0};
+    STR_TO_VARSTR(user, pConn->user);
+    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+    colDataAppend(pColInfo, numOfRows, (const char *)user, false);
+
+    char app[TSDB_APP_NAME_LEN + VARSTR_HEADER_SIZE];
+    STR_TO_VARSTR(app, pConn->app);
+    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+    colDataAppend(pColInfo, numOfRows, (const char *)app, false);
+
+    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+    colDataAppend(pColInfo, numOfRows, (const char *)&pConn->pid, false);
+
+    char endpoint[TSDB_IPv4ADDR_LEN + 6 + VARSTR_HEADER_SIZE] = {0};
+    sprintf(&endpoint[VARSTR_HEADER_SIZE], "%s:%d", taosIpStr(pConn->ip), pConn->port);
+    varDataLen(endpoint) = strlen(&endpoint[VARSTR_HEADER_SIZE]);
+    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+    colDataAppend(pColInfo, numOfRows, (const char *)endpoint, false);
+
+    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+    colDataAppend(pColInfo, numOfRows, (const char *)&pConn->loginTimeMs, false);
+
+    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+    colDataAppend(pColInfo, numOfRows, (const char *)&pConn->lastAccessTimeMs, false);
+
+    numOfRows++;
+  }
+
+  pShow->numOfRows += numOfRows;
+  return numOfRows;
+}
+
+static int32_t mndRetrieveQueries(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBlock, int32_t rows) {
+  SMnode    *pMnode = pReq->info.node;
+  SSdb      *pSdb = pMnode->pSdb;
+  int32_t    numOfRows = 0;
+  int32_t    cols = 0;
+  SConnObj  *pConn = NULL;
+  
+  if (pShow->pIter == NULL) {
+    SProfileMgmt *pMgmt = &pMnode->profileMgmt;
+    pShow->pIter = taosCacheCreateIter(pMgmt->cache);
+  }
+
+  while (numOfRows < rows) {
+    pConn = mndGetNextConn(pMnode, pShow->pIter);
+    if (pConn == NULL) {
       pShow->pIter = NULL;
       break;
     }
 
-    for (int32_t i = 0; i < pConn->numOfQueries; ++i) {
+    taosRLockLatch(&pConn->queryLock);
-      SQueryDesc *pDesc = pConn->pQueries + i;
+    if (NULL == pConn->pQueries || taosArrayGetSize(pConn->pQueries) <= 0) {
+      taosRUnLockLatch(&pConn->queryLock);
+      continue;
+    }
+
+    int32_t numOfQueries = taosArrayGetSize(pConn->pQueries);
+    for (int32_t i = 0; i < numOfQueries; ++i) {
+      SQueryDesc* pQuery = taosArrayGet(pConn->pQueries, i);
       cols = 0;
 
-      pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
+      char queryId[26 + VARSTR_HEADER_SIZE] = {0};
-      *(int64_t *)pWrite = htobe64(pDesc->queryId);
+      sprintf(&queryId[VARSTR_HEADER_SIZE], "%x:%" PRIx64, pConn->id, pQuery->reqRid);
-      cols++;
+      varDataLen(queryId) = strlen(&queryId[VARSTR_HEADER_SIZE]);
+      SColumnInfoData *pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+      colDataAppend(pColInfo, numOfRows, (const char *)queryId, false);
 
-      pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
+      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
-      *(int64_t *)pWrite = htobe64(pConn->id);
+      colDataAppend(pColInfo, numOfRows, (const char *)&pQuery->queryId, false);
-      cols++;
 
-      pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
+      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
-      STR_WITH_MAXSIZE_TO_VARSTR(pWrite, pConn->user, pShow->pMeta->pSchemas[cols].bytes);
+      colDataAppend(pColInfo, numOfRows, (const char *)&pConn->id, false);
-      cols++;
 
-      pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
+      char app[TSDB_APP_NAME_LEN + VARSTR_HEADER_SIZE];
-      snprintf(str, tListLen(str), "%s:%u", taosIpStr(pConn->ip), pConn->port);
+      STR_TO_VARSTR(app, pConn->app);
-      STR_WITH_MAXSIZE_TO_VARSTR(pWrite, str, pShow->pMeta->pSchemas[cols].bytes);
+      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
-      cols++;
+      colDataAppend(pColInfo, numOfRows, (const char *)app, false);
 
-      char handleBuf[24] = {0};
+      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
-      snprintf(handleBuf, tListLen(handleBuf), "%" PRIu64, htobe64(pDesc->qId));
+      colDataAppend(pColInfo, numOfRows, (const char *)&pQuery->pid, false);
-      pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
 
-      STR_WITH_MAXSIZE_TO_VARSTR(pWrite, handleBuf, pShow->pMeta->pSchemas[cols].bytes);
+      char user[TSDB_USER_LEN + VARSTR_HEADER_SIZE] = {0};
-      cols++;
+      STR_TO_VARSTR(user, pConn->user);
+      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+      colDataAppend(pColInfo, numOfRows, (const char *)user, false);
 
-      pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
+      char endpoint[TSDB_IPv4ADDR_LEN + 6 + VARSTR_HEADER_SIZE] = {0};
-      *(int64_t *)pWrite = htobe64(pDesc->stime);
+      sprintf(&endpoint[VARSTR_HEADER_SIZE], "%s:%d", taosIpStr(pConn->ip), pConn->port);
-      cols++;
+      varDataLen(endpoint) = strlen(&endpoint[VARSTR_HEADER_SIZE]);
+      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+      colDataAppend(pColInfo, numOfRows, (const char *)endpoint, false);
 
-      pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
+      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
-      *(int64_t *)pWrite = htobe64(pDesc->useconds);
+      colDataAppend(pColInfo, numOfRows, (const char *)&pQuery->stime, false);
-      cols++;
 
-      snprintf(str, tListLen(str), "0x%" PRIx64, htobe64(pDesc->sqlObjId));
+      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
-      pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
+      colDataAppend(pColInfo, numOfRows, (const char *)&pQuery->useconds, false);
-      STR_WITH_MAXSIZE_TO_VARSTR(pWrite, str, pShow->pMeta->pSchemas[cols].bytes);
-      cols++;
 
-      pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
+      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
-      *(int32_t *)pWrite = htonl(pDesc->pid);
+      colDataAppend(pColInfo, numOfRows, (const char *)&pQuery->stableQuery, false);
-      cols++;
 
-      char epBuf[TSDB_EP_LEN + 1] = {0};
+      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
-      snprintf(epBuf, tListLen(epBuf), "%s:%u", pDesc->fqdn, pConn->port);
+      colDataAppend(pColInfo, numOfRows, (const char *)&pQuery->subPlanNum, false);
-      pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
-      STR_WITH_MAXSIZE_TO_VARSTR(pWrite, epBuf, pShow->pMeta->pSchemas[cols].bytes);
-      cols++;
 
-      pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
+      char subStatus[TSDB_SHOW_SUBQUERY_LEN + VARSTR_HEADER_SIZE] = {0};
-      *(bool *)pWrite = pDesc->stableQuery;
+      int32_t strSize = sizeof(subStatus);
-      cols++;
+      int32_t offset = VARSTR_HEADER_SIZE;
+      for (int32_t i = 0; i < pQuery->subPlanNum && offset < strSize; ++i) {
+        if (i) {
+          offset += snprintf(subStatus + offset, strSize - offset - 1, ",");
+        }
+        SQuerySubDesc* pDesc = taosArrayGet(pQuery->subDesc, i);
+        offset += snprintf(subStatus + offset, strSize - offset - 1, "%" PRIu64 ":%s", pDesc->tid, pDesc->status);
+      }
+      varDataLen(subStatus) = strlen(&subStatus[VARSTR_HEADER_SIZE]);
+      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+      colDataAppend(pColInfo, numOfRows, subStatus, false);
 
-      pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
+      char sql[TSDB_SHOW_SQL_LEN + VARSTR_HEADER_SIZE] = {0};
-      *(int32_t *)pWrite = htonl(pDesc->numOfSub);
+      STR_TO_VARSTR(sql, pQuery->sql);
-      cols++;
+      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
-
+      colDataAppend(pColInfo, numOfRows, (const char *)sql, false);
-      pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
-      STR_WITH_MAXSIZE_TO_VARSTR(pWrite, pDesc->subSqlInfo, pShow->pMeta->pSchemas[cols].bytes);
-      cols++;
-
-      pWrite = data + pShow->offset[cols] * rows + pShow->pMeta->pSchemas[cols].bytes * numOfRows;
-      STR_WITH_MAXSIZE_TO_VARSTR(pWrite, pDesc->sql, pShow->pMeta->pSchemas[cols].bytes);
-      cols++;
 
       numOfRows++;
     }
+    
+    taosRUnLockLatch(&pConn->queryLock);
   }
 
   pShow->numOfRows += numOfRows;
-#endif
   return numOfRows;
 }
 

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

@@ -63,10 +63,11 @@ int32_t mndConvertRsmaTask(char** pDst, int32_t* pDstLen, const char* ast, int64
       .topicQuery = false,
       .streamQuery = true,
       .rSmaQuery = true,
-      .triggerType = triggerType,
+      .triggerType = STREAM_TRIGGER_AT_ONCE,
       .watermark = watermark,
-      .filesFactor = filesFactor,
+      /*.filesFactor = filesFactor,*/
   };
+
   if (qCreateQueryPlan(&cxt, &pPlan, NULL) < 0) {
     terrno = TSDB_CODE_QRY_INVALID_INPUT;
     goto END;

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

@@ -1284,7 +1284,7 @@ static int32_t mndBuildStbSchema(SMnode *pMnode, const char *dbFName, const char
   SStbObj *pStb = mndAcquireStb(pMnode, tbFName);
   if (pStb == NULL) {
     mndReleaseDb(pMnode, pDb);
-    terrno = TSDB_CODE_MND_INVALID_STB;
+    terrno = TSDB_CODE_PAR_TABLE_NOT_EXIST;
     return -1;
   }
 

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

@@ -35,7 +35,7 @@ static int32_t mndStreamActionInsert(SSdb *pSdb, SStreamObj *pStream);
 static int32_t mndStreamActionDelete(SSdb *pSdb, SStreamObj *pStream);
 static int32_t mndStreamActionUpdate(SSdb *pSdb, SStreamObj *pStream, SStreamObj *pNewStream);
 static int32_t mndProcessCreateStreamReq(SRpcMsg *pReq);
-/*static int32_t mndProcessDropStreamReq(SRpcMsg *pReq);*/
+static int32_t mndProcessDropStreamReq(SRpcMsg *pReq);
 /*static int32_t mndProcessDropStreamInRsp(SRpcMsg *pRsp);*/
 static int32_t mndProcessStreamMetaReq(SRpcMsg *pReq);
 static int32_t mndGetStreamMeta(SRpcMsg *pReq, SShowObj *pShow, STableMetaRsp *pMeta);
@@ -55,9 +55,8 @@ int32_t mndInitStream(SMnode *pMnode) {
 
   mndSetMsgHandle(pMnode, TDMT_MND_CREATE_STREAM, mndProcessCreateStreamReq);
   mndSetMsgHandle(pMnode, TDMT_STREAM_TASK_DEPLOY_RSP, mndTransProcessRsp);
-  /*mndSetMsgHandle(pMnode, TDMT_SND_TASK_DEPLOY_RSP, mndTransProcessRsp);*/
+  mndSetMsgHandle(pMnode, TDMT_MND_DROP_STREAM, mndProcessDropStreamReq);
-  /*mndSetMsgHandle(pMnode, TDMT_MND_DROP_STREAM, mndProcessDropStreamReq);*/
+  mndSetMsgHandle(pMnode, TDMT_MND_DROP_STREAM_RSP, mndTransProcessRsp);
-  /*mndSetMsgHandle(pMnode, TDMT_MND_DROP_STREAM_RSP, mndProcessDropStreamInRsp);*/
 
   mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_STREAMS, mndRetrieveStream);
   mndAddShowFreeIterHandle(pMnode, TSDB_MGMT_TABLE_STREAMS, mndCancelGetNextStream);
@@ -196,16 +195,6 @@ void mndReleaseStream(SMnode *pMnode, SStreamObj *pStream) {
   sdbRelease(pSdb, pStream);
 }
 
-static SDbObj *mndAcquireDbByStream(SMnode *pMnode, char *streamName) {
-  SName name = {0};
-  tNameFromString(&name, streamName, T_NAME_ACCT | T_NAME_DB | T_NAME_TABLE);
-
-  char db[TSDB_STREAM_FNAME_LEN] = {0};
-  tNameGetFullDbName(&name, db);
-
-  return mndAcquireDb(pMnode, db);
-}
-
 static int32_t mndCheckCreateStreamReq(SCMCreateStreamReq *pCreate) {
   if (pCreate->name[0] == 0 || pCreate->sql == NULL || pCreate->sql[0] == 0) {
     terrno = TSDB_CODE_MND_INVALID_STREAM_OPTION;
@@ -364,6 +353,47 @@ _OVER:
   return -1;
 }
 
+static int32_t mndPersistTaskDropReq(STrans *pTrans, SStreamTask *pTask) {
+  ASSERT(pTask->nodeId != 0);
+
+  // vnode
+  if (pTask->nodeId > 0) {
+    SVDropStreamTaskReq *pReq = taosMemoryCalloc(1, sizeof(SVDropStreamTaskReq));
+    if (pReq == NULL) {
+      terrno = TSDB_CODE_OUT_OF_MEMORY;
+      return -1;
+    }
+    pReq->head.vgId = htonl(pTask->nodeId);
+    pReq->taskId = pTask->taskId;
+    STransAction action = {0};
+    memcpy(&action.epSet, &pTask->epSet, sizeof(SEpSet));
+    action.pCont = pReq;
+    action.contLen = sizeof(SVDropStreamTaskReq);
+    action.msgType = TDMT_VND_STREAM_TASK_DROP;
+    if (mndTransAppendRedoAction(pTrans, &action) != 0) {
+      taosMemoryFree(pReq);
+      return -1;
+    }
+  }
+
+  return 0;
+}
+
+static int32_t mndDropStreamTasks(SMnode *pMnode, STrans *pTrans, SStreamObj *pStream) {
+  int32_t lv = taosArrayGetSize(pStream->tasks);
+  for (int32_t i = 0; i < lv; i++) {
+    SArray *pTasks = taosArrayGetP(pStream->tasks, i);
+    int32_t sz = taosArrayGetSize(pTasks);
+    for (int32_t j = 0; j < sz; j++) {
+      SStreamTask *pTask = taosArrayGetP(pTasks, j);
+      if (mndPersistTaskDropReq(pTrans, pTask) < 0) {
+        return -1;
+      }
+    }
+  }
+  return 0;
+}
+
 static int32_t mndCreateStream(SMnode *pMnode, SRpcMsg *pReq, SCMCreateStreamReq *pCreate, SDbObj *pDb) {
   mDebug("stream:%s to create", pCreate->name);
   SStreamObj streamObj = {0};
@@ -457,6 +487,7 @@ static int32_t mndProcessCreateStreamReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
+  // TODO check auth
   pDb = mndAcquireDb(pMnode, createStreamReq.sourceDB);
   if (pDb == NULL) {
     terrno = TSDB_CODE_MND_DB_NOT_SELECTED;
@@ -482,6 +513,47 @@ _OVER:
   return code;
 }
 
+static int32_t mndProcessDropStreamReq(SRpcMsg *pReq) {
+  SMnode     *pMnode = pReq->info.node;
+  int32_t     code = -1;
+  SStreamObj *pStream = NULL;
+  /*SDbObj     *pDb = NULL;*/
+  /*SUserObj   *pUser = NULL;*/
+
+  SMDropStreamTaskReq dropStreamReq = *(SMDropStreamTaskReq *)pReq->pCont;
+
+  pStream = mndAcquireStream(pMnode, dropStreamReq.name);
+
+  if (pStream == NULL) {
+    terrno = TSDB_CODE_MND_STREAM_NOT_EXIST;
+    return -1;
+  }
+
+#if 0
+  // todo check auth
+  pUser = mndAcquireUser(pMnode, pReq->conn.user);
+  if (pUser == NULL) {
+    goto DROP_STREAM_OVER;
+  }
+#endif
+
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_NOTHING, pReq);
+  if (pTrans == NULL) {
+    mError("stream:%s, failed to drop since %s", dropStreamReq.name, terrstr());
+    return -1;
+  }
+  mDebug("trans:%d, used to drop stream:%s", pTrans->id, dropStreamReq.name);
+
+  // drop all tasks
+  if (mndDropStreamTasks(pMnode, pTrans, pStream) < 0) {
+    mError("stream:%s, failed to drop task since %s", dropStreamReq.name, terrstr());
+    return -1;
+  }
+
+DROP_STREAM_OVER:
+  return 0;
+}
+
 static int32_t mndGetNumOfStreams(SMnode *pMnode, char *dbName, int32_t *pNumOfStreams) {
   SSdb   *pSdb = pMnode->pSdb;
   SDbObj *pDb = mndAcquireDb(pMnode, dbName);

source/dnode/mnode/impl/test/profile/profile.cpp

@@ -88,7 +88,7 @@ TEST_F(MndTestProfile, 02_ConnectMsg_InvalidDB) {
 
 TEST_F(MndTestProfile, 03_ConnectMsg_Show) {
   test.SendShowReq(TSDB_MGMT_TABLE_CONNS, "connections", "");
-  EXPECT_EQ(test.GetShowRows(), 0);
+  EXPECT_EQ(test.GetShowRows(), 1);
 }
 
 TEST_F(MndTestProfile, 04_HeartBeatMsg) {

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

@@ -140,6 +140,7 @@ int32_t tqProcessVgDeleteReq(STQ* pTq, char* msg, int32_t msgLen);
 int32_t tqProcessOffsetCommitReq(STQ* pTq, char* msg, int32_t msgLen);
 int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId);
 int32_t tqProcessTaskDeploy(STQ* pTq, char* msg, int32_t msgLen);
+int32_t tqProcessTaskDropReq(STQ* pTq, char* msg, int32_t msgLen);
 int32_t tqProcessStreamTrigger(STQ* pTq, SSubmitReq* data);
 int32_t tqProcessTaskRunReq(STQ* pTq, SRpcMsg* pMsg);
 int32_t tqProcessTaskDispatchReq(STQ* pTq, SRpcMsg* pMsg);

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

@@ -510,3 +510,13 @@ int32_t tqProcessTaskRecoverRsp(STQ* pTq, SRpcMsg* pMsg) {
   streamProcessRecoverRsp(pTask, pRsp);
   return 0;
 }
+
+int32_t tqProcessTaskDropReq(STQ* pTq, char* msg, int32_t msgLen) {
+  SVDropStreamTaskReq* pReq = (SVDropStreamTaskReq*)msg;
+  int32_t              code = taosHashRemove(pTq->pStreamTasks, &pReq->taskId, sizeof(int32_t));
+  if (code == 0) {
+    // sendrsp
+  }
+  ASSERT(code == 0);
+  return code;
+}

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

@@ -168,6 +168,11 @@ int32_t vnodeProcessWriteReq(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRp
         goto _err;
       }
     } break;
+    case TDMT_VND_STREAM_TASK_DROP: {
+      if (tqProcessTaskDropReq(pVnode->pTq, pMsg->pCont, pMsg->contLen) < 0) {
+        goto _err;
+      }
+    } break;
     case TDMT_VND_ALTER_CONFIRM:
       vnodeProcessAlterConfirmReq(pVnode, version, pReq, len, pRsp);
       break;
@@ -375,8 +380,9 @@ int32_t vnodeProcessSyncReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
       ret = syncNodeOnAppendEntriesReplyCb(pSyncNode, pSyncMsg);
       syncAppendEntriesReplyDestroy(pSyncMsg);
 
-    } else if (pRpcMsg->msgType == TDMT_SYNC_SET_MNODE_STANDBY) {
+    } else if (pRpcMsg->msgType == TDMT_SYNC_SET_VNODE_STANDBY) {
       ret = syncSetStandby(pVnode->sync);
+      vInfo("vgId:%d, set standby result:0x%x rid:%" PRId64, pVnode->config.vgId, ret, pVnode->sync);
       SRpcMsg rsp = {.code = ret, .info = pMsg->info};
       tmsgSendRsp(&rsp);
     } else {

source/libs/executor/inc/executil.h

@@ -44,7 +44,6 @@
 typedef struct SGroupResInfo {
   int32_t index;
   SArray* pRows;      // SArray<SResKeyPos>
-  int32_t position;
 } SGroupResInfo;
 
 typedef struct SResultRow {
@@ -56,7 +55,7 @@ typedef struct SResultRow {
   uint32_t      numOfRows;   // number of rows of current time window
   STimeWindow   win;
   struct SResultRowEntryInfo pEntryInfo[];  // For each result column, there is a resultInfo
-//  char         *key;         // start key of current result row
+//  char         *key;       // start key of current result row
 } SResultRow;
 
 typedef struct SResultRowPosition {
@@ -71,9 +70,7 @@ typedef struct SResKeyPos {
 } SResKeyPos;
 
 typedef struct SResultRowInfo {
-  SResultRowPosition *pPosition;  // todo remove this
   int32_t      size;       // number of result set
-  int32_t      capacity;   // max capacity
   SResultRowPosition cur;
   SList*       openWindow;
 } SResultRowInfo;

source/libs/executor/inc/executorInt.h

@@ -20,6 +20,8 @@
 extern "C" {
 #endif
 
+extern int32_t exchangeObjRefPool;
+
 typedef struct {
   char*           pData;
   bool            isNull;

source/libs/executor/inc/executorimpl.h

@@ -181,7 +181,6 @@ typedef struct SExecTaskInfo {
   STaskCostInfo    cost;
   int64_t          owner;  // if it is in execution
   int32_t          code;
-//  uint64_t         totalRows;            // total number of rows
   struct {
     char          *tablename;
     char          *dbname;
@@ -222,10 +221,10 @@ typedef struct STaskRuntimeEnv {
 } STaskRuntimeEnv;
 
 enum {
-  OP_NOT_OPENED = 0x0,
+  OP_NOT_OPENED    = 0x0,
-  OP_OPENED = 0x1,
+  OP_OPENED        = 0x1,
   OP_RES_TO_RETURN = 0x5,
-  OP_EXEC_DONE = 0x9,
+  OP_EXEC_DONE     = 0x9,
 };
 
 typedef struct SOperatorFpSet {
@@ -262,12 +261,12 @@ typedef enum {
 } EX_SOURCE_STATUS;
 
 typedef struct SSourceDataInfo {
-  struct SExchangeInfo* pEx;
   int32_t               index;
   SRetrieveTableRsp*    pRsp;
   uint64_t              totalRows;
   int32_t               code;
   EX_SOURCE_STATUS      status;
+  const char*           taskId;
 } SSourceDataInfo;
 
 typedef struct SLoadRemoteDataInfo {
@@ -285,6 +284,7 @@ typedef struct SExchangeInfo {
   bool                seqLoadData;  // sequential load data or not, false by default
   int32_t             current;
   SLoadRemoteDataInfo loadInfo;
+  uint64_t            self;
 } SExchangeInfo;
 
 #define COL_MATCH_FROM_COL_ID  0x1
@@ -470,10 +470,8 @@ typedef struct SIntervalAggOperatorInfo {
   bool               timeWindowInterpo;  // interpolation needed or not
   char**             pRow;               // previous row/tuple of already processed datablock
   SArray*            pInterpCols;        // interpolation columns
-  STableQueryInfo*   pCurrent;           // current tableQueryInfo struct
   int32_t            order;              // current SSDataBlock scan order
   EOPTR_EXEC_MODEL   execModel;          // operator execution model [batch model|stream model]
-  SArray*            pUpdatedWindow;     // updated time window due to the input data block from the downstream operator.
   STimeWindowAggSupp twAggSup;
   bool               invertible;
   SArray*            pPrevValues;        //  SArray<SGroupKeys> used to keep the previous not null value for interpolation.
@@ -502,8 +500,6 @@ typedef struct SAggOperatorInfo {
   STableQueryInfo   *current;
   uint64_t           groupId;
   SGroupResInfo      groupResInfo;
-  STableQueryInfo   *pTableQueryInfo;
-
   SExprInfo         *pScalarExprInfo;
   int32_t            numOfScalarExpr;      // the number of scalar expression before the aggregate function can be applied
   SqlFunctionCtx    *pScalarCtx;                 // scalar function requried sql function struct.
@@ -639,8 +635,13 @@ typedef struct SStreamSessionAggOperatorInfo {
 
 typedef struct STimeSliceOperatorInfo {
   SOptrBasicInfo binfo;
+  STimeWindow    win;
   SInterval      interval;
-  SGroupResInfo  groupResInfo;  // multiple results build supporter
+  int64_t        current;
+  SArray*        pPrevRow;      // SArray<SGroupValue>
+  SArray*        pCols;         // SArray<SColumn>
+  int32_t        fillType;      // fill type
+  struct SFillColInfo*  pFillColInfo;  // fill column info
 } STimeSliceOperatorInfo;
 
 typedef struct SStateWindowOperatorInfo {
@@ -733,6 +734,8 @@ typedef struct SJoinOperatorInfo {
 #define OPTR_IS_OPENED(_optr)  (((_optr)->status & OP_OPENED) == OP_OPENED)
 #define OPTR_SET_OPENED(_optr) ((_optr)->status |= OP_OPENED)
 
+void doDestroyExchangeOperatorInfo(void* param);
+
 SOperatorFpSet createOperatorFpSet(__optr_open_fn_t openFn, __optr_fn_t nextFn, __optr_fn_t streamFn,
     __optr_fn_t cleanup, __optr_close_fn_t closeFn, __optr_encode_fn_t encode,
     __optr_decode_fn_t decode, __optr_explain_fn_t explain);
@@ -840,7 +843,7 @@ SOperatorInfo* createPartitionOperatorInfo(SOperatorInfo* downstream, SExprInfo*
                                            SSDataBlock* pResultBlock, SArray* pGroupColList, SExecTaskInfo* pTaskInfo);
 
 SOperatorInfo* createTimeSliceOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t numOfCols,
-                                           SSDataBlock* pResultBlock, SExecTaskInfo* pTaskInfo);
+                                           SSDataBlock* pResultBlock, const SNodeListNode* pValNode, SExecTaskInfo* pTaskInfo);
 
 SOperatorInfo* createMergeJoinOperatorInfo(SOperatorInfo** pDownstream, int32_t numOfDownstream, SExprInfo* pExprInfo, int32_t numOfCols, SSDataBlock* pResBlock, SNode* pOnCondition, SExecTaskInfo* pTaskInfo);
 
@@ -912,6 +915,15 @@ int32_t compareTimeWindow(const void* p1, const void* p2, const void* param);
 int32_t finalizeResultRowIntoResultDataBlock(SDiskbasedBuf* pBuf, SResultRowPosition* resultRowPosition,
                                        SqlFunctionCtx* pCtx, SExprInfo* pExprInfo, int32_t numOfExprs, const int32_t* rowCellOffset,
                                        SSDataBlock* pBlock, SExecTaskInfo* pTaskInfo);
+
+int32_t getTableList(void* metaHandle, int32_t tableType, uint64_t tableUid, STableListInfo* pListInfo,
+                            SNode* pTagCond);
+int32_t createMultipleDataReaders(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
+                                  STableListInfo* pTableListInfo, SArray* arrayReader, uint64_t queryId,
+                                  uint64_t taskId, SNode* pTagCond);
+SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanNode, SArray* dataReaders,
+                                                SReadHandle* readHandle, SExecTaskInfo* pTaskInfo);
+
 void copyUpdateDataBlock(SSDataBlock* pDest, SSDataBlock* pSource, int32_t tsColIndex);
 
 #ifdef __cplusplus

source/libs/executor/inc/tfill.h

@@ -28,8 +28,6 @@ struct SSDataBlock;
 
 typedef struct SFillColInfo {
   SExprInfo *pExpr;
-//  SResSchema schema;
-//  int16_t    functionId;      // sql function id
   int16_t    flag;            // column flag: TAG COLUMN|NORMAL COLUMN
   int16_t    tagIndex;        // index of current tag in SFillTagColInfo array list
   SVariant   fillVal;

source/libs/executor/src/executil.c

@@ -41,13 +41,7 @@ int32_t getOutputInterResultBufSize(STaskAttr* pQueryAttr) {
 
 int32_t initResultRowInfo(SResultRowInfo *pResultRowInfo, int32_t size) {
   pResultRowInfo->size       = 0;
-  pResultRowInfo->capacity   = size;
   pResultRowInfo->cur.pageId = -1;
-  
-  pResultRowInfo->pPosition = taosMemoryCalloc(pResultRowInfo->capacity, sizeof(SResultRowPosition));
-  if (pResultRowInfo->pPosition == NULL) {
-    return TSDB_CODE_QRY_OUT_OF_MEMORY;
-  }
   return TSDB_CODE_SUCCESS;
 }
 
@@ -56,25 +50,14 @@ void cleanupResultRowInfo(SResultRowInfo *pResultRowInfo) {
     return;
   }
 
-  if (pResultRowInfo->capacity == 0) {
-//    assert(pResultRowInfo->pResult == NULL);
-    return;
-  }
-
   for(int32_t i = 0; i < pResultRowInfo->size; ++i) {
 //    if (pResultRowInfo->pResult[i]) {
 //      taosMemoryFreeClear(pResultRowInfo->pResult[i]->key);
 //    }
   }
-  
-  taosMemoryFreeClear(pResultRowInfo->pPosition);
 }
 
 void resetResultRowInfo(STaskRuntimeEnv *pRuntimeEnv, SResultRowInfo *pResultRowInfo) {
-  if (pResultRowInfo == NULL || pResultRowInfo->capacity == 0) {
-    return;
-  }
-
   for (int32_t i = 0; i < pResultRowInfo->size; ++i) {
 //    SResultRow *pWindowRes = pResultRowInfo->pResult[i];
 //    clearResultRow(pRuntimeEnv, pWindowRes);
@@ -288,232 +271,3 @@ void orderTheResultRows(STaskRuntimeEnv* pRuntimeEnv) {
 
   taosArraySort(pRuntimeEnv->pResultRowArrayList, fn);
 }
-
-static int32_t mergeIntoGroupResultImplRv(STaskRuntimeEnv *pRuntimeEnv, SGroupResInfo* pGroupResInfo, uint64_t groupId, int32_t* rowCellInfoOffset) {
-  if (pGroupResInfo->pRows == NULL) {
-    pGroupResInfo->pRows = taosArrayInit(100, POINTER_BYTES);
-  }
-
-  size_t len = taosArrayGetSize(pRuntimeEnv->pResultRowArrayList);
-  for(; pGroupResInfo->position < len; ++pGroupResInfo->position) {
-    SResultRowCell* pResultRowCell = taosArrayGet(pRuntimeEnv->pResultRowArrayList, pGroupResInfo->position);
-    if (pResultRowCell->groupId != groupId) {
-      break;
-    }
-
-
-    int64_t num = 0;//getNumOfResultWindowRes(pRuntimeEnv, &pResultRowCell->pos, rowCellInfoOffset);
-    if (num <= 0) {
-      continue;
-    }
-
-    taosArrayPush(pGroupResInfo->pRows, &pResultRowCell->pos);
-//    pResultRowCell->pRow->numOfRows = (uint32_t) num;
-  }
-
-  return TSDB_CODE_SUCCESS;
-}
-
-static UNUSED_FUNC int32_t mergeIntoGroupResultImpl(STaskRuntimeEnv *pRuntimeEnv, SGroupResInfo* pGroupResInfo, SArray *pTableList,
-    int32_t* rowCellInfoOffset) {
-  bool ascQuery = true;
-#if 0
-  int32_t code = TSDB_CODE_SUCCESS;
-
-  int32_t *posList = NULL;
-  SMultiwayMergeTreeInfo *pTree = NULL;
-  STableQueryInfo **pTableQueryInfoList = NULL;
-
-  size_t size = taosArrayGetSize(pTableList);
-  if (pGroupResInfo->pRows == NULL) {
-    pGroupResInfo->pRows = taosArrayInit(100, POINTER_BYTES);
-  }
-
-  posList = taosMemoryCalloc(size, sizeof(int32_t));
-  pTableQueryInfoList = taosMemoryMalloc(POINTER_BYTES * size);
-
-  if (pTableQueryInfoList == NULL || posList == NULL || pGroupResInfo->pRows == NULL || pGroupResInfo->pRows == NULL) {
-//    qError("QInfo:%"PRIu64" failed alloc memory", GET_TASKID(pRuntimeEnv));
-    code = TSDB_CODE_QRY_OUT_OF_MEMORY;
-    goto _end;
-  }
-
-  int32_t numOfTables = 0;
-  for (int32_t i = 0; i < size; ++i) {
-    STableQueryInfo *item = taosArrayGetP(pTableList, i);
-//    if (item->resInfo.size > 0) {
-//      pTableQueryInfoList[numOfTables++] = item;
-//    }
-  }
-
-  // there is no data in current group
-  // no need to merge results since only one table in each group
-//  if (numOfTables == 0) {
-//    goto _end;
-//  }
-
-  int32_t order = TSDB_ORDER_ASC;
-  SCompSupporter cs = {pTableQueryInfoList, posList, order};
-
-  int32_t ret = tMergeTreeCreate(&pTree, numOfTables, &cs, tableResultComparFn);
-  if (ret != TSDB_CODE_SUCCESS) {
-    code = TSDB_CODE_QRY_OUT_OF_MEMORY;
-    goto _end;
-  }
-
-  int64_t lastTimestamp = ascQuery? INT64_MIN:INT64_MAX;
-  int64_t startt = taosGetTimestampMs();
-
-  while (1) {
-    int32_t tableIndex = tMergeTreeGetChosenIndex(pTree);
-
-    SResultRowInfo *pWindowResInfo = &pTableQueryInfoList[tableIndex]->resInfo;
-    ASSERT(0);
-    SResultRow  *pWindowRes = NULL;//getResultRow(pBuf, pWindowResInfo, cs.rowIndex[tableIndex]);
-
-    int64_t num = 0;//getNumOfResultWindowRes(pRuntimeEnv, pWindowRes, rowCellInfoOffset);
-    if (num <= 0) {
-      cs.rowIndex[tableIndex] += 1;
-
-      if (cs.rowIndex[tableIndex] >= pWindowResInfo->size) {
-        cs.rowIndex[tableIndex] = -1;
-        if (--numOfTables == 0) { // all input sources are exhausted
-          break;
-        }
-      }
-    } else {
-      assert((pWindowRes->win.skey >= lastTimestamp && ascQuery) || (pWindowRes->win.skey <= lastTimestamp && !ascQuery));
-
-      if (pWindowRes->win.skey != lastTimestamp) {
-        taosArrayPush(pGroupResInfo->pRows, &pWindowRes);
-        pWindowRes->numOfRows = (uint32_t) num;
-      }
-
-      lastTimestamp = pWindowRes->win.skey;
-
-      // move to the next row of current entry
-      if ((++cs.rowIndex[tableIndex]) >= pWindowResInfo->size) {
-        cs.rowIndex[tableIndex] = -1;
-
-        // all input sources are exhausted
-        if ((--numOfTables) == 0) {
-          break;
-        }
-      }
-    }
-
-    tMergeTreeAdjust(pTree, tMergeTreeGetAdjustIndex(pTree));
-  }
-
-  int64_t endt = taosGetTimestampMs();
-
-//  qDebug("QInfo:%"PRIx64" result merge completed for group:%d, elapsed time:%" PRId64 " ms", GET_TASKID(pRuntimeEnv),
-//         pGroupResInfo->currentGroup, endt - startt);
-
-  _end:
-  taosMemoryFreeClear(pTableQueryInfoList);
-  taosMemoryFreeClear(posList);
-  taosMemoryFreeClear(pTree);
-
-  return code;
-}
-
-int32_t mergeIntoGroupResult(SGroupResInfo* pGroupResInfo, STaskRuntimeEnv* pRuntimeEnv, int32_t* offset) {
-  int64_t st = taosGetTimestampUs();
-
-  while (pGroupResInfo->currentGroup < pGroupResInfo->totalGroup) {
-    mergeIntoGroupResultImplRv(pRuntimeEnv, pGroupResInfo, pGroupResInfo->currentGroup, offset);
-
-    // this group generates at least one result, return results
-    if (taosArrayGetSize(pGroupResInfo->pRows) > 0) {
-      break;
-    }
-
-//    qDebug("QInfo:%"PRIu64" no result in group %d, continue", GET_TASKID(pRuntimeEnv), pGroupResInfo->currentGroup);
-    cleanupGroupResInfo(pGroupResInfo);
-    incNextGroup(pGroupResInfo);
-  }
-
-//  int64_t elapsedTime = taosGetTimestampUs() - st;
-//  qDebug("QInfo:%"PRIu64" merge res data into group, index:%d, total group:%d, elapsed time:%" PRId64 "us", GET_TASKID(pRuntimeEnv),
-//         pGroupResInfo->currentGroup, pGroupResInfo->totalGroup, elapsedTime);
-#endif
-
-  return TSDB_CODE_SUCCESS;
-}
-
-//void blockDistInfoToBinary(STableBlockDist* pDist, struct SBufferWriter* bw) {
-//  tbufWriteUint32(bw, pDist->numOfTables);
-//  tbufWriteUint16(bw, pDist->numOfFiles);
-//  tbufWriteUint64(bw, pDist->totalSize);
-//  tbufWriteUint64(bw, pDist->totalRows);
-//  tbufWriteInt32(bw, pDist->maxRows);
-//  tbufWriteInt32(bw, pDist->minRows);
-//  tbufWriteUint32(bw, pDist->numOfInmemRows);
-//  tbufWriteUint32(bw, pDist->numOfSmallBlocks);
-//  tbufWriteUint64(bw, taosArrayGetSize(pDist->dataBlockInfos));
-//
-//  // compress the binary string
-//  char* p = TARRAY_GET_START(pDist->dataBlockInfos);
-//
-//  // compress extra bytes
-//  size_t x = taosArrayGetSize(pDist->dataBlockInfos) * pDist->dataBlockInfos->elemSize;
-//  char* tmp = taosMemoryMalloc(x + 2);
-//
-//  bool comp = false;
-//  int32_t len = tsCompressString(p, (int32_t)x, 1, tmp, (int32_t)x, ONE_STAGE_COMP, NULL, 0);
-//  if (len == -1 || len >= x) { // compress failed, do not compress this binary data
-//    comp = false;
-//    len = (int32_t)x;
-//  } else {
-//    comp = true;
-//  }
-//
-//  tbufWriteUint8(bw, comp);
-//  tbufWriteUint32(bw, len);
-//  if (comp) {
-//    tbufWriteBinary(bw, tmp, len);
-//  } else {
-//    tbufWriteBinary(bw, p, len);
-//  }
-//  taosMemoryFreeClear(tmp);
-//}
-
-//void blockDistInfoFromBinary(const char* data, int32_t len, STableBlockDist* pDist) {
-//  SBufferReader br = tbufInitReader(data, len, false);
-//
-//  pDist->numOfTables = tbufReadUint32(&br);
-//  pDist->numOfFiles  = tbufReadUint16(&br);
-//  pDist->totalSize   = tbufReadUint64(&br);
-//  pDist->totalRows   = tbufReadUint64(&br);
-//  pDist->maxRows     = tbufReadInt32(&br);
-//  pDist->minRows     = tbufReadInt32(&br);
-//  pDist->numOfInmemRows = tbufReadUint32(&br);
-//  pDist->numOfSmallBlocks = tbufReadUint32(&br);
-//  int64_t numSteps = tbufReadUint64(&br);
-//
-//  bool comp = tbufReadUint8(&br);
-//  uint32_t compLen = tbufReadUint32(&br);
-//
-//  size_t originalLen = (size_t) (numSteps *sizeof(SFileBlockInfo));
-//
-//  char* outputBuf = NULL;
-//  if (comp) {
-//    outputBuf = taosMemoryMalloc(originalLen);
-//
-//    size_t actualLen = compLen;
-//    const char* compStr = tbufReadBinary(&br, &actualLen);
-//
-//    int32_t orignalLen = tsDecompressString(compStr, compLen, 1, outputBuf,
-//                                            (int32_t)originalLen , ONE_STAGE_COMP, NULL, 0);
-//    assert(orignalLen == numSteps *sizeof(SFileBlockInfo));
-//  } else {
-//    outputBuf = (char*) tbufReadBinary(&br, &originalLen);
-//  }
-//
-//  pDist->dataBlockInfos = taosArrayFromList(outputBuf, (uint32_t)numSteps, sizeof(SFileBlockInfo));
-//  if (comp) {
-//    taosMemoryFreeClear(outputBuf);
-//  }
-//}
-

source/libs/executor/src/executorMain.c

@@ -13,28 +13,30 @@
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */
 
-#include <vnode.h>
-#include "dataSinkMgt.h"
-#include "texception.h"
 #include "os.h"
-#include "tarray.h"
+#include "tref.h"
-#include "tcache.h"
+#include "dataSinkMgt.h"
-#include "tglobal.h"
 #include "tmsg.h"
 #include "tudf.h"
 
 #include "executor.h"
 #include "executorimpl.h"
 #include "query.h"
-#include "thash.h"
+
-#include "tlosertree.h"
+static TdThreadOnce initPoolOnce = PTHREAD_ONCE_INIT;
-#include "ttypes.h"
+int32_t exchangeObjRefPool = -1;
+
+static void initRefPool() {
+  exchangeObjRefPool = taosOpenRef(1024, doDestroyExchangeOperatorInfo);
+}
 
 int32_t qCreateExecTask(SReadHandle* readHandle, int32_t vgId, uint64_t taskId, SSubplan* pSubplan,
-    qTaskInfo_t* pTaskInfo, DataSinkHandle* handle, EOPTR_EXEC_MODEL model) {
+                        qTaskInfo_t* pTaskInfo, DataSinkHandle* handle, EOPTR_EXEC_MODEL model) {
   assert(readHandle != NULL && pSubplan != NULL);
   SExecTaskInfo** pTask = (SExecTaskInfo**)pTaskInfo;
 
+  taosThreadOnce(&initPoolOnce, initRefPool);
+
   int32_t code = createExecTaskInfoImpl(pSubplan, pTask, readHandle, taskId, model);
   if (code != TSDB_CODE_SUCCESS) {
     goto _error;

source/libs/executor/src/executorimpl.c

@@ -13,6 +13,7 @@
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */
 
+#include "tref.h"
 #include "filter.h"
 #include "function.h"
 #include "functionMgt.h"
@@ -37,9 +38,6 @@
 #include "vnode.h"
 
 #define IS_MAIN_SCAN(runtime)          ((runtime)->scanFlag == MAIN_SCAN)
-#define IS_REVERSE_SCAN(runtime)       ((runtime)->scanFlag == REVERSE_SCAN)
-#define IS_REPEAT_SCAN(runtime)        ((runtime)->scanFlag == REPEAT_SCAN)
-#define SET_MAIN_SCAN_FLAG(runtime)    ((runtime)->scanFlag = MAIN_SCAN)
 #define SET_REVERSE_SCAN_FLAG(runtime) ((runtime)->scanFlag = REVERSE_SCAN)
 
 #define SDATA_BLOCK_INITIALIZER \
@@ -47,12 +45,6 @@
 
 #define GET_FORWARD_DIRECTION_FACTOR(ord) (((ord) == TSDB_ORDER_ASC) ? QUERY_ASC_FORWARD_STEP : QUERY_DESC_FORWARD_STEP)
 
-enum {
-  TS_JOIN_TS_EQUAL = 0,
-  TS_JOIN_TS_NOT_EQUALS = 1,
-  TS_JOIN_TAG_NOT_EQUALS = 2,
-};
-
 #if 0
 static UNUSED_FUNC void *u_malloc (size_t __size) {
   uint32_t v = taosRand();
@@ -88,7 +80,6 @@ static UNUSED_FUNC void* u_realloc(void* p, size_t __size) {
 #endif
 
 #define CLEAR_QUERY_STATUS(q, st) ((q)->status &= (~(st)))
-//#define GET_NUM_OF_TABLEGROUP(q)    taosArrayGetSize((q)->tableqinfoGroupInfo.pGroupList)
 #define QUERY_IS_INTERVAL_QUERY(_q) ((_q)->interval.interval > 0)
 
 int32_t getMaximumIdleDurationSec() { return tsShellActivityTimer * 2; }
@@ -110,7 +101,6 @@ static void releaseQueryBuf(size_t numOfTables);
 
 static void destroySFillOperatorInfo(void* param, int32_t numOfOutput);
 static void destroyProjectOperatorInfo(void* param, int32_t numOfOutput);
-static void destroyTagScanOperatorInfo(void* param, int32_t numOfOutput);
 static void destroyOrderOperatorInfo(void* param, int32_t numOfOutput);
 static void destroyAggOperatorInfo(void* param, int32_t numOfOutput);
 
@@ -118,7 +108,6 @@ static void destroyIntervalOperatorInfo(void* param, int32_t numOfOutput);
 static void destroyExchangeOperatorInfo(void* param, int32_t numOfOutput);
 
 static void destroyOperatorInfo(SOperatorInfo* pOperator);
-static void destroySysTableScannerOperatorInfo(void* param, int32_t numOfOutput);
 
 void doSetOperatorCompleted(SOperatorInfo* pOperator) {
   pOperator->status = OP_EXEC_DONE;
@@ -562,10 +551,6 @@ void doApplyFunctions(SExecTaskInfo* taskInfo, SqlFunctionCtx* pCtx, STimeWindow
     pCtx[k].input.startRowIndex = offset;
     pCtx[k].input.numOfRows = forwardStep;
 
-    if (tsCol != NULL) {
-      pCtx[k].ptsList = tsCol;
-    }
-
     // not a whole block involved in query processing, statistics data can not be used
     // NOTE: the original value of isSet have been changed here
     if (pCtx[k].input.colDataAggIsSet && forwardStep < numOfTotal) {
@@ -1133,40 +1118,6 @@ SqlFunctionCtx* createSqlFunctionCtx(SExprInfo* pExprInfo, int32_t numOfOutput,
     pCtx->increase = false;
 
     pCtx->param = pFunct->pParam;
-    //    for (int32_t j = 0; j < pCtx->numOfParams; ++j) {
-    //      // set the order information for top/bottom query
-    //      int32_t functionId = pCtx->functionId;
-    //      if (functionId == FUNCTION_TOP || functionId == FUNCTION_BOTTOM || functionId == FUNCTION_DIFF) {
-    //        int32_t f = getExprFunctionId(&pExpr[0]);
-    //        assert(f == FUNCTION_TS || f == FUNCTION_TS_DUMMY);
-    //
-    //        //      pCtx->param[2].i = pQueryAttr->order.order;
-    //        //      pCtx->param[2].nType = TSDB_DATA_TYPE_BIGINT;
-    //        //      pCtx->param[3].i = functionId;
-    //        //      pCtx->param[3].nType = TSDB_DATA_TYPE_BIGINT;
-    //
-    //        //      pCtx->param[1].i = pQueryAttr->order.col.info.colId;
-    //      } else if (functionId == FUNCTION_INTERP) {
-    //        //      pCtx->param[2].i = (int8_t)pQueryAttr->fillType;
-    //        //      if (pQueryAttr->fillVal != NULL) {
-    //        //        if (isNull((const char *)&pQueryAttr->fillVal[i], pCtx->inputType)) {
-    //        //          pCtx->param[1].nType = TSDB_DATA_TYPE_NULL;
-    //        //        } else {  // todo refactor, taosVariantCreateFromBinary should handle the NULL value
-    //        //          if (pCtx->inputType != TSDB_DATA_TYPE_BINARY && pCtx->inputType != TSDB_DATA_TYPE_NCHAR) {
-    //        //            taosVariantCreateFromBinary(&pCtx->param[1], (char *)&pQueryAttr->fillVal[i],
-    //        pCtx->inputBytes, pCtx->inputType);
-    //        //          }
-    //        //        }
-    //        //      }
-    //      } else if (functionId == FUNCTION_TWA) {
-    //        //      pCtx->param[1].i = pQueryAttr->window.skey;
-    //        //      pCtx->param[1].nType = TSDB_DATA_TYPE_BIGINT;
-    //        //      pCtx->param[2].i = pQueryAttr->window.ekey;
-    //        //      pCtx->param[2].nType = TSDB_DATA_TYPE_BIGINT;
-    //      } else if (functionId == FUNCTION_ARITHM) {
-    //        //      pCtx->param[1].pz = (char*) getScalarFuncSupport(pRuntimeEnv->scalarSup, i);
-    //      }
-    //    }
   }
 
   for (int32_t i = 1; i < numOfOutput; ++i) {
@@ -2438,44 +2389,44 @@ static void doTableQueryInfoTimeWindowCheck(SExecTaskInfo* pTaskInfo, STableQuer
 #endif
 }
 
-// static void updateTableIdInfo(STableQueryInfo* pTableQueryInfo, SSDataBlock* pBlock, SHashObj* pTableIdInfo, int32_t
+typedef struct SFetchRspHandleWrapper {
-// order) {
+  uint32_t exchangeId;
-//   int32_t step = GET_FORWARD_DIRECTION_FACTOR(order);
+  int32_t  sourceIndex;
-//   pTableQueryInfo->lastKey = ((order == TSDB_ORDER_ASC)? pBlock->info.window.ekey:pBlock->info.window.skey) + step;
+} SFetchRspHandleWrapper;
-//
-//   if (pTableQueryInfo->pTable == NULL) {
-//     return;
-//   }
-//
-//   STableIdInfo tidInfo = createTableIdInfo(pTableQueryInfo);
-//   STableIdInfo *idinfo = taosHashGet(pTableIdInfo, &tidInfo.tid, sizeof(tidInfo.tid));
-//   if (idinfo != NULL) {
-//     assert(idinfo->tid == tidInfo.tid && idinfo->uid == tidInfo.uid);
-//     idinfo->key = tidInfo.key;
-//   } else {
-//     taosHashPut(pTableIdInfo, &tidInfo.tid, sizeof(tidInfo.tid), &tidInfo, sizeof(STableIdInfo));
-//   }
-// }
 
 int32_t loadRemoteDataCallback(void* param, const SDataBuf* pMsg, int32_t code) {
-  SSourceDataInfo* pSourceDataInfo = (SSourceDataInfo*)param;
+  SFetchRspHandleWrapper* pWrapper = (SFetchRspHandleWrapper*) param;
+
+  SExchangeInfo* pExchangeInfo = taosAcquireRef(exchangeObjRefPool, pWrapper->exchangeId);
+  if (pExchangeInfo == NULL) {
+    qWarn("failed to acquire exchange operator, since it may have been released");
+    return TSDB_CODE_SUCCESS;
+  }
+
+  int32_t index = pWrapper->sourceIndex;
+  SSourceDataInfo* pSourceDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, index);
+
   if (code == TSDB_CODE_SUCCESS) {
     pSourceDataInfo->pRsp = pMsg->pData;
 
     SRetrieveTableRsp* pRsp = pSourceDataInfo->pRsp;
     pRsp->numOfRows = htonl(pRsp->numOfRows);
-    pRsp->compLen = htonl(pRsp->compLen);
+    pRsp->compLen   = htonl(pRsp->compLen);
     pRsp->numOfCols = htonl(pRsp->numOfCols);
-    pRsp->useconds = htobe64(pRsp->useconds);
+    pRsp->useconds  = htobe64(pRsp->useconds);
 
-    ASSERT(pSourceDataInfo->pRsp != NULL);
+    ASSERT(pRsp != NULL);
-    qDebug("fetch rsp received, index:%d, rows:%d", pSourceDataInfo->index, pRsp->numOfRows);
+    qDebug("%s fetch rsp received, index:%d, rows:%d", pSourceDataInfo->taskId, index, pRsp->numOfRows);
   } else {
     pSourceDataInfo->code = code;
   }
 
   pSourceDataInfo->status = EX_SOURCE_DATA_READY;
-  tsem_post(&pSourceDataInfo->pEx->ready);
+
+  tsem_post(&pExchangeInfo->ready);
+  taosReleaseRef(exchangeObjRefPool, pWrapper->exchangeId);
+
+  taosMemoryFree(pWrapper);
   return TSDB_CODE_SUCCESS;
 }
 
@@ -2524,9 +2475,9 @@ static int32_t doSendFetchDataRequest(SExchangeInfo* pExchangeInfo, SExecTaskInf
          pSource->addr.nodeId, pSource->addr.epSet.eps[0].fqdn, pSource->taskId, sourceIndex, totalSources);
 
   pMsg->header.vgId = htonl(pSource->addr.nodeId);
-  pMsg->sId = htobe64(pSource->schedId);
+  pMsg->sId         = htobe64(pSource->schedId);
-  pMsg->taskId = htobe64(pSource->taskId);
+  pMsg->taskId      = htobe64(pSource->taskId);
-  pMsg->queryId = htobe64(pTaskInfo->id.queryId);
+  pMsg->queryId     = htobe64(pTaskInfo->id.queryId);
 
   // send the fetch remote task result reques
   SMsgSendInfo* pMsgSendInfo = taosMemoryCalloc(1, sizeof(SMsgSendInfo));
@@ -2537,11 +2488,15 @@ static int32_t doSendFetchDataRequest(SExchangeInfo* pExchangeInfo, SExecTaskInf
     return pTaskInfo->code;
   }
 
-  pMsgSendInfo->param = pDataInfo;
+  SFetchRspHandleWrapper* pWrapper = taosMemoryCalloc(1, sizeof(SFetchRspHandleWrapper));
+  pWrapper->exchangeId  = pExchangeInfo->self;
+  pWrapper->sourceIndex = sourceIndex;
+
+  pMsgSendInfo->param = pWrapper;
   pMsgSendInfo->msgInfo.pData = pMsg;
   pMsgSendInfo->msgInfo.len = sizeof(SResFetchReq);
   pMsgSendInfo->msgType = TDMT_VND_FETCH;
-  pMsgSendInfo->fp = loadRemoteDataCallback;
+  pMsgSendInfo->fp    = loadRemoteDataCallback;
 
   int64_t transporterId = 0;
   int32_t code = asyncSendMsgToServer(pExchangeInfo->pTransporter, &pSource->addr.epSet, &transporterId, pMsgSendInfo);
@@ -2689,10 +2644,10 @@ static SSDataBlock* concurrentlyLoadRemoteDataImpl(SOperatorInfo* pOperator, SEx
       SSDataBlock*         pRes = pExchangeInfo->pResult;
       SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
       if (pRsp->numOfRows == 0) {
-        qDebug("%s vgId:%d, taskID:0x%" PRIx64 " index:%d completed, rowsOfSource:%" PRIu64 ", totalRows:%" PRIu64
+        qDebug("%s vgId:%d, taskId:0x%" PRIx64 " index:%d completed, rowsOfSource:%" PRIu64 ", totalRows:%" PRIu64
-               " try next",
+               ", completed:%d try next %d/%"PRIzu,
-               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, i + 1, pDataInfo->totalRows,
+               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, i, pDataInfo->totalRows,
-               pExchangeInfo->loadInfo.totalRows);
+               pExchangeInfo->loadInfo.totalRows, completed + 1, i + 1, totalSources);
         pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
         completed += 1;
         taosMemoryFreeClear(pDataInfo->pRsp);
@@ -2708,10 +2663,11 @@ static SSDataBlock* concurrentlyLoadRemoteDataImpl(SOperatorInfo* pOperator, SEx
       }
 
       if (pRsp->completed == 1) {
-        qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, rowsOfSource:%" PRIu64
+        qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " index:%d completed, numOfRows:%d, rowsOfSource:%" PRIu64
-               ", totalRows:%" PRIu64 ", totalBytes:%" PRIu64 " try next %d/%" PRIzu,
+               ", totalRows:%" PRIu64 ", totalBytes:%" PRIu64 ", completed:%d try next %d/%" PRIzu,
-               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pRes->info.rows, pDataInfo->totalRows,
+               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, i, pRes->info.rows, pDataInfo->totalRows,
-               pLoadInfo->totalRows, pLoadInfo->totalSize, i + 1, totalSources);
+               pLoadInfo->totalRows, pLoadInfo->totalSize, completed + 1, i + 1, totalSources);
+        completed += 1;
         pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
       } else {
         qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, totalRows:%" PRIu64
@@ -2761,13 +2717,13 @@ static int32_t prepareConcurrentlyLoad(SOperatorInfo* pOperator) {
   }
 
   int64_t endTs = taosGetTimestampUs();
-  qDebug("%s send all fetch requests to %" PRIzu " sources completed, elapsed:%" PRId64, GET_TASKID(pTaskInfo),
+  qDebug("%s send all fetch requests to %" PRIzu " sources completed, elapsed:%.2fms", GET_TASKID(pTaskInfo),
-         totalSources, endTs - startTs);
+         totalSources, (endTs - startTs)/1000.0);
 
-  tsem_wait(&pExchangeInfo->ready);
   pOperator->status = OP_RES_TO_RETURN;
   pOperator->cost.openCost = taosGetTimestampUs() - startTs;
 
+  tsem_wait(&pExchangeInfo->ready);
   return TSDB_CODE_SUCCESS;
 }
 
@@ -2883,7 +2839,7 @@ static SSDataBlock* doLoadRemoteData(SOperatorInfo* pOperator) {
   }
 }
 
-static int32_t initDataSource(int32_t numOfSources, SExchangeInfo* pInfo) {
+static int32_t initDataSource(int32_t numOfSources, SExchangeInfo* pInfo, const char* id) {
   pInfo->pSourceDataInfo = taosArrayInit(numOfSources, sizeof(SSourceDataInfo));
   if (pInfo->pSourceDataInfo == NULL) {
     return TSDB_CODE_OUT_OF_MEMORY;
@@ -2892,11 +2848,10 @@ static int32_t initDataSource(int32_t numOfSources, SExchangeInfo* pInfo) {
   for (int32_t i = 0; i < numOfSources; ++i) {
     SSourceDataInfo dataInfo = {0};
     dataInfo.status = EX_SOURCE_DATA_NOT_READY;
-    dataInfo.pEx = pInfo;
+    dataInfo.taskId = id;
-    dataInfo.index = i;
+    dataInfo.index  = i;
-
+    SSourceDataInfo *pDs = taosArrayPush(pInfo->pSourceDataInfo, &dataInfo);
-    void* ret = taosArrayPush(pInfo->pSourceDataInfo, &dataInfo);
+    if (pDs == NULL) {
-    if (ret == NULL) {
       taosArrayDestroy(pInfo->pSourceDataInfo);
       return TSDB_CODE_OUT_OF_MEMORY;
     }
@@ -2924,7 +2879,9 @@ static int32_t initExchangeOperator(SExchangePhysiNode* pExNode, SExchangeInfo*
     taosArrayPush(pInfo->pSources, pNode);
   }
 
-  return initDataSource(numOfSources, pInfo);
+  pInfo->self = taosAddRef(exchangeObjRefPool, pInfo);
+
+  return initDataSource(numOfSources, pInfo, id);
 }
 
 SOperatorInfo* createExchangeOperatorInfo(void* pTransporter, SExchangePhysiNode* pExNode, SExecTaskInfo* pTaskInfo) {
@@ -2939,18 +2896,18 @@ SOperatorInfo* createExchangeOperatorInfo(void* pTransporter, SExchangePhysiNode
     goto _error;
   }
 
-  pInfo->seqLoadData = false;
-  pInfo->pTransporter = pTransporter;
-  pInfo->pResult = createResDataBlock(pExNode->node.pOutputDataBlockDesc);
   tsem_init(&pInfo->ready, 0, 0);
 
-  pOperator->name = "ExchangeOperator";
+  pInfo->seqLoadData    = false;
+  pInfo->pTransporter   = pTransporter;
+  pInfo->pResult        = createResDataBlock(pExNode->node.pOutputDataBlockDesc);
+  pOperator->name       = "ExchangeOperator";
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_EXCHANGE;
-  pOperator->blocking = false;
+  pOperator->blocking   = false;
-  pOperator->status = OP_NOT_OPENED;
+  pOperator->status     = OP_NOT_OPENED;
-  pOperator->info = pInfo;
+  pOperator->info       = pInfo;
   pOperator->numOfExprs = pInfo->pResult->info.numOfCols;
-  pOperator->pTaskInfo = pTaskInfo;
+  pOperator->pTaskInfo  = pTaskInfo;
 
   pOperator->fpSet = createOperatorFpSet(prepareLoadRemoteData, doLoadRemoteData, NULL, NULL,
                                          destroyExchangeOperatorInfo, NULL, NULL, NULL);
@@ -2958,7 +2915,7 @@ SOperatorInfo* createExchangeOperatorInfo(void* pTransporter, SExchangePhysiNode
 
 _error:
   if (pInfo != NULL) {
-    destroyExchangeOperatorInfo(pInfo, LIST_LENGTH(pExNode->pSrcEndPoints));
+    doDestroyExchangeOperatorInfo(pInfo);
   }
 
   taosMemoryFreeClear(pInfo);
@@ -4105,6 +4062,12 @@ static void destroyIndefinitOperatorInfo(void* param, int32_t numOfOutput) {
 
 void destroyExchangeOperatorInfo(void* param, int32_t numOfOutput) {
   SExchangeInfo* pExInfo = (SExchangeInfo*)param;
+  taosRemoveRef(exchangeObjRefPool, pExInfo->self);
+}
+
+void doDestroyExchangeOperatorInfo(void* param) {
+  SExchangeInfo* pExInfo = (SExchangeInfo*) param;
+
   taosArrayDestroy(pExInfo->pSources);
   taosArrayDestroy(pExInfo->pSourceDataInfo);
   if (pExInfo->pResult != NULL) {
@@ -4322,6 +4285,8 @@ static int32_t initFillInfo(SFillOperatorInfo* pInfo, SExprInfo* pExpr, int32_t
 
   pInfo->p = taosMemoryCalloc(numOfCols, POINTER_BYTES);
   if (pInfo->pFillInfo == NULL || pInfo->p == NULL) {
+    taosMemoryFree(pInfo->pFillInfo);
+    taosMemoryFree(pInfo->p);
     return TSDB_CODE_OUT_OF_MEMORY;
   } else {
     return TSDB_CODE_SUCCESS;
@@ -4556,8 +4521,6 @@ static tsdbReaderT doCreateDataReader(STableScanPhysiNode* pTableScanNode, SRead
                                       STableListInfo* pTableListInfo, uint64_t queryId, uint64_t taskId,
                                       SNode* pTagCond);
 
-static int32_t getTableList(void* metaHandle, int32_t tableType, uint64_t tableUid, STableListInfo* pListInfo,
-                            SNode* pTagCond);
 static SArray* extractColumnInfo(SNodeList* pNodeList);
 
 static SArray* createSortInfo(SNodeList* pNodeList);
@@ -4691,19 +4654,30 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
       }
 
       SArray* groupKeys = extractPartitionColInfo(pTableScanNode->pPartitionKeys);
-      code = generateGroupIdMap(pTableListInfo, pHandle, groupKeys); //todo for json
+      code = generateGroupIdMap(pTableListInfo, pHandle, groupKeys);  // todo for json
       taosArrayDestroy(groupKeys);
-      if (code){
+      if (code) {
         tsdbCleanupReadHandle(pDataReader);
         return NULL;
       }
 
-      SOperatorInfo* pOperator =
+      SOperatorInfo* pOperator = createTableScanOperatorInfo(pTableScanNode, pDataReader, pHandle, pTaskInfo);
-          createTableScanOperatorInfo(pTableScanNode, pDataReader, pHandle, pTaskInfo);
-
       STableScanInfo* pScanInfo = pOperator->info;
       pTaskInfo->cost.pRecoder = &pScanInfo->readRecorder;
+      return pOperator;
+    } else if (QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN == type) {
+      STableMergeScanPhysiNode* pTableScanNode = (STableMergeScanPhysiNode*)pPhyNode;
 
+      SArray* dataReaders = taosArrayInit(8, POINTER_BYTES);
+      createMultipleDataReaders(pTableScanNode, pHandle, pTableListInfo, dataReaders, queryId, taskId, pTagCond);
+      extractTableSchemaVersion(pHandle, pTableScanNode->scan.uid, pTaskInfo);
+      SArray* groupKeys = extractPartitionColInfo(pTableScanNode->pPartitionKeys);
+      generateGroupIdMap(pTableListInfo, pHandle, groupKeys); //todo for json
+      taosArrayDestroy(groupKeys);
+      SOperatorInfo* pOperator =
+        createTableMergeScanOperatorInfo(pTableScanNode, dataReaders, pHandle, pTaskInfo);
+      STableScanInfo* pScanInfo = pOperator->info;
+      pTaskInfo->cost.pRecoder = &pScanInfo->readRecorder;
       return pOperator;
     } else if (QUERY_NODE_PHYSICAL_PLAN_EXCHANGE == type) {
       return createExchangeOperatorInfo(pHandle->pMsgCb->clientRpc, (SExchangePhysiNode*)pPhyNode, pTaskInfo);
@@ -4837,11 +4811,9 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
     int32_t tsSlotId = ((SColumnNode*)pIntervalPhyNode->window.pTspk)->slotId;
     pOptr = createMergeIntervalOperatorInfo(ops[0], pExprInfo, num, pResBlock, &interval, tsSlotId, pTaskInfo);
   } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_SEMI_INTERVAL == type) {
-    qDebug("[******]create Semi");
     int32_t children = 0;
     pOptr = createStreamFinalIntervalOperatorInfo(ops[0], pPhyNode, pTaskInfo, children);
   } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_INTERVAL == type) {
-    qDebug("[******]create Final");
     int32_t children = 1;
     pOptr = createStreamFinalIntervalOperatorInfo(ops[0], pPhyNode, pTaskInfo, children);
   } else if (QUERY_NODE_PHYSICAL_PLAN_SORT == type) {

source/libs/executor/src/groupoperator.c

@@ -332,12 +332,7 @@ static SSDataBlock* hashGroupbyAggregate(SOperatorInfo* pOperator) {
 
   pOperator->status = OP_RES_TO_RETURN;
   closeAllResultRows(&pInfo->binfo.resultRowInfo);
-  //  if (!stableQuery) { // finalize include the update of result rows
+
-  //    finalizeQueryResult(pInfo->binfo.pCtx, pOperator->numOfExprs);
-  //  } else {
-  //    updateNumOfRowsInResultRows(pInfo->binfo.pCtx, pOperator->numOfExprs, &pInfo->binfo.resultRowInfo,
-  //    pInfo->binfo.rowCellInfoOffset);
-  //  }
 #if 0
   if(pOperator->fpSet.encodeResultRow){
     char *result = NULL;
@@ -378,8 +373,9 @@ static SSDataBlock* hashGroupbyAggregate(SOperatorInfo* pOperator) {
   return (rows == 0)? NULL:pRes;
 }
 
-SOperatorInfo* createGroupOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t numOfCols, SSDataBlock* pResultBlock, SArray* pGroupColList,
+SOperatorInfo* createGroupOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t numOfCols,
-    SNode* pCondition, SExprInfo* pScalarExprInfo, int32_t numOfScalarExpr, SExecTaskInfo* pTaskInfo) {
+                                       SSDataBlock* pResultBlock, SArray* pGroupColList, SNode* pCondition,
+                                       SExprInfo* pScalarExprInfo, int32_t numOfScalarExpr, SExecTaskInfo* pTaskInfo) {
   SGroupbyOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SGroupbyOperatorInfo));
   SOperatorInfo*        pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
   if (pInfo == NULL || pOperator == NULL) {
@@ -407,7 +403,7 @@ SOperatorInfo* createGroupOperatorInfo(SOperatorInfo* downstream, SExprInfo* pEx
   pOperator->status       = OP_NOT_OPENED;
   // pOperator->operatorType = OP_Groupby;
   pOperator->pExpr        = pExprInfo;
-  pOperator->numOfExprs  = numOfCols;
+  pOperator->numOfExprs   = numOfCols;
   pOperator->info         = pInfo;
   pOperator->pTaskInfo    = pTaskInfo;
 
@@ -669,8 +665,8 @@ static void destroyPartitionOperatorInfo(void* param, int32_t numOfOutput) {
   taosMemoryFree(pInfo->columnOffset);
 }
 
-SOperatorInfo* createPartitionOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t numOfCols, SSDataBlock* pResultBlock, SArray* pGroupColList,
+SOperatorInfo* createPartitionOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t numOfCols,
-                                           SExecTaskInfo* pTaskInfo) {
+                                           SSDataBlock* pResultBlock, SArray* pGroupColList, SExecTaskInfo* pTaskInfo) {
   SPartitionOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SPartitionOperatorInfo));
   SOperatorInfo*        pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
   if (pInfo == NULL || pOperator == NULL) {

source/libs/executor/src/scanoperator.c

@@ -630,7 +630,7 @@ static SSDataBlock* doBlockInfoScan(SOperatorInfo* pOperator) {
   SColumnInfoData* pColInfo = taosArrayGet(pBlock->pDataBlock, 0);
 
   int32_t len = tSerializeBlockDistInfo(NULL, 0, &blockDistInfo);
-  char* p = taosMemoryCalloc(1, len + VARSTR_HEADER_SIZE);
+  char*   p = taosMemoryCalloc(1, len + VARSTR_HEADER_SIZE);
   tSerializeBlockDistInfo(varDataVal(p), len, &blockDistInfo);
   varDataSetLen(p, len);
 
@@ -642,7 +642,7 @@ static SSDataBlock* doBlockInfoScan(SOperatorInfo* pOperator) {
 }
 
 static void destroyBlockDistScanOperatorInfo(void* param, int32_t numOfOutput) {
-  SBlockDistInfo* pDistInfo = (SBlockDistInfo*) param;
+  SBlockDistInfo* pDistInfo = (SBlockDistInfo*)param;
   blockDataDestroy(pDistInfo->pResBlock);
 }
 
@@ -654,24 +654,25 @@ SOperatorInfo* createDataBlockInfoScanOperator(void* dataReader, SExecTaskInfo*
     goto _error;
   }
 
-  pInfo->pHandle   = dataReader;
+  pInfo->pHandle = dataReader;
 
   pInfo->pResBlock = taosMemoryCalloc(1, sizeof(SSDataBlock));
 
   SColumnInfoData infoData = {0};
-  infoData.info.type       = TSDB_DATA_TYPE_VARCHAR;
+  infoData.info.type = TSDB_DATA_TYPE_VARCHAR;
-  infoData.info.bytes      = 1024;
+  infoData.info.bytes = 1024;
 
   taosArrayPush(pInfo->pResBlock->pDataBlock, &infoData);
 
-  pOperator->name          = "DataBlockInfoScanOperator";
+  pOperator->name = "DataBlockInfoScanOperator";
   //  pOperator->operatorType = OP_TableBlockInfoScan;
-  pOperator->blocking      = false;
+  pOperator->blocking = false;
-  pOperator->status        = OP_NOT_OPENED;
+  pOperator->status = OP_NOT_OPENED;
-  pOperator->info          = pInfo;
+  pOperator->info = pInfo;
-  pOperator->pTaskInfo     = pTaskInfo;
+  pOperator->pTaskInfo = pTaskInfo;
-  
+
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doBlockInfoScan, NULL, NULL, destroyBlockDistScanOperatorInfo, NULL, NULL, NULL);
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doBlockInfoScan, NULL, NULL,
+                                         destroyBlockDistScanOperatorInfo, NULL, NULL, NULL);
   return pOperator;
 
 _error:
@@ -890,7 +891,7 @@ static SSDataBlock* doStreamBlockScan(SOperatorInfo* pOperator) {
       return NULL;
     }
 
-    int32_t current = pInfo->validBlockIndex++;
+    int32_t      current = pInfo->validBlockIndex++;
     SSDataBlock* pBlock = taosArrayGetP(pInfo->pBlockLists, current);
     blockDataUpdateTsWindow(pBlock, 0);
     return pBlock;
@@ -1058,7 +1059,7 @@ SOperatorInfo* createStreamScanOperatorInfo(void* pDataReader, SReadHandle* pHan
   SScanPhysiNode* pScanPhyNode = &pTableScanNode->scan;
 
   SDataBlockDescNode* pDescNode = pScanPhyNode->node.pOutputDataBlockDesc;
-  SOperatorInfo* pTableScanDummy = createTableScanOperatorInfo(pTableScanNode, pDataReader, pHandle, pTaskInfo);
+  SOperatorInfo*      pTableScanDummy = createTableScanOperatorInfo(pTableScanNode, pDataReader, pHandle, pTaskInfo);
 
   STableScanInfo* pSTInfo = (STableScanInfo*)pTableScanDummy->info;
 
@@ -1512,7 +1513,6 @@ static SSDataBlock* doSysTableScan(SOperatorInfo* pOperator) {
                pRsp->numOfRows, pInfo->loadInfo.totalRows);
 
         if (pRsp->numOfRows == 0) {
-
           taosMemoryFree(pRsp);
           return NULL;
         }
@@ -1849,3 +1849,498 @@ _error:
   terrno = TSDB_CODE_OUT_OF_MEMORY;
   return NULL;
 }
+
+typedef struct STableMergeScanInfo {
+  SArray*     dataReaders;  // array of tsdbReaderT*
+  SReadHandle readHandle;
+
+  int32_t  bufPageSize;
+  uint32_t sortBufSize;  // max buffer size for in-memory sort
+
+  SArray*      pSortInfo;
+  SSortHandle* pSortHandle;
+
+  SSDataBlock* pSortInputBlock;
+  int64_t      startTs;  // sort start time
+
+  bool          hasGroupId;
+  uint64_t      groupId;
+  STupleHandle* prefetchedTuple;
+
+  SArray* sortSourceParams;
+
+  SFileBlockLoadRecorder readRecorder;
+  int64_t                numOfRows;
+  //  int32_t         prevGroupId;  // previous table group id
+  SScanInfo       scanInfo;
+  int32_t         scanTimes;
+  SNode*          pFilterNode;  // filter info, which is push down by optimizer
+  SqlFunctionCtx* pCtx;         // which belongs to the direct upstream operator operator query context
+  SResultRowInfo* pResultRowInfo;
+  int32_t*        rowCellInfoOffset;
+  SExprInfo*      pExpr;
+  SSDataBlock*    pResBlock;
+  SArray*         pColMatchInfo;
+  int32_t         numOfOutput;
+
+  SExprInfo*      pPseudoExpr;
+  int32_t         numOfPseudoExpr;
+  SqlFunctionCtx* pPseudoCtx;
+  //  int32_t*        rowCellInfoOffset;
+
+  SQueryTableDataCond cond;
+  int32_t             scanFlag;  // table scan flag to denote if it is a repeat/reverse/main scan
+  int32_t             dataBlockLoadFlag;
+  SInterval interval;  // if the upstream is an interval operator, the interval info is also kept here to get the time
+                       // window to check if current data block needs to be loaded.
+
+  SSampleExecInfo sample;  // sample execution info
+  int32_t         curTWinIdx;
+
+} STableMergeScanInfo;
+
+int32_t createMultipleDataReaders(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
+                                  STableListInfo* pTableListInfo, SArray* arrayReader, uint64_t queryId,
+                                  uint64_t taskId, SNode* pTagCond) {
+  int32_t code =
+      getTableList(pHandle->meta, pTableScanNode->scan.tableType, pTableScanNode->scan.uid, pTableListInfo, pTagCond);
+  if (code != TSDB_CODE_SUCCESS) {
+    goto _error;
+  }
+
+  if (taosArrayGetSize(pTableListInfo->pTableList) == 0) {
+    qDebug("no table qualified for query, TID:0x%" PRIx64 ", QID:0x%" PRIx64, taskId, queryId);
+    goto _error;
+  }
+
+  SQueryTableDataCond cond = {0};
+  code = initQueryTableDataCond(&cond, pTableScanNode);
+  if (code != TSDB_CODE_SUCCESS) {
+    goto _error;
+  }
+  // TODO: free the sublist info and the table list in it
+  for (int32_t i = 0; i < taosArrayGetSize(pTableListInfo->pTableList); ++i) {
+    STableListInfo* subListInfo = taosMemoryCalloc(1, sizeof(subListInfo));
+    subListInfo->pTableList = taosArrayInit(1, sizeof(STableKeyInfo));
+    taosArrayPush(subListInfo->pTableList, taosArrayGet(pTableListInfo->pTableList, i));
+
+    tsdbReaderT* pReader = tsdbReaderOpen(pHandle->vnode, &cond, subListInfo, queryId, taskId);
+    taosArrayPush(arrayReader, &pReader);
+
+    taosArrayDestroy(subListInfo->pTableList);
+    taosMemoryFree(subListInfo);
+  }
+  clearupQueryTableDataCond(&cond);
+
+  return 0;
+
+_error:
+  return code;
+}
+
+static int32_t loadDataBlockFromOneTable(SOperatorInfo* pOperator, STableMergeScanInfo* pTableScanInfo,
+                                         int32_t readerIdx, SSDataBlock* pBlock, uint32_t* status) {
+  SExecTaskInfo*  pTaskInfo = pOperator->pTaskInfo;
+  STableMergeScanInfo* pInfo = pOperator->info;
+
+  SFileBlockLoadRecorder* pCost = &pTableScanInfo->readRecorder;
+
+  pCost->totalBlocks += 1;
+  pCost->totalRows += pBlock->info.rows;
+
+  *status = pInfo->dataBlockLoadFlag;
+  if (pTableScanInfo->pFilterNode != NULL ||
+      overlapWithTimeWindow(&pTableScanInfo->interval, &pBlock->info, pTableScanInfo->cond.order)) {
+    (*status) = FUNC_DATA_REQUIRED_DATA_LOAD;
+  }
+
+  SDataBlockInfo* pBlockInfo = &pBlock->info;
+  taosMemoryFreeClear(pBlock->pBlockAgg);
+
+  if (*status == FUNC_DATA_REQUIRED_FILTEROUT) {
+    qDebug("%s data block filter out, brange:%" PRId64 "-%" PRId64 ", rows:%d", GET_TASKID(pTaskInfo),
+           pBlockInfo->window.skey, pBlockInfo->window.ekey, pBlockInfo->rows);
+    pCost->filterOutBlocks += 1;
+    return TSDB_CODE_SUCCESS;
+  } else if (*status == FUNC_DATA_REQUIRED_NOT_LOAD) {
+    qDebug("%s data block skipped, brange:%" PRId64 "-%" PRId64 ", rows:%d", GET_TASKID(pTaskInfo),
+           pBlockInfo->window.skey, pBlockInfo->window.ekey, pBlockInfo->rows);
+    pCost->skipBlocks += 1;
+
+    // clear all data in pBlock that are set when handing the previous block
+    for (int32_t i = 0; i < pBlockInfo->numOfCols; ++i) {
+      SColumnInfoData* pcol = taosArrayGet(pBlock->pDataBlock, i);
+      pcol->pData = NULL;
+    }
+
+    return TSDB_CODE_SUCCESS;
+  } else if (*status == FUNC_DATA_REQUIRED_STATIS_LOAD) {
+    pCost->loadBlockStatis += 1;
+
+    bool             allColumnsHaveAgg = true;
+    SColumnDataAgg** pColAgg = NULL;
+    tsdbReaderT*     reader = taosArrayGetP(pTableScanInfo->dataReaders, readerIdx);
+    tsdbRetrieveDataBlockStatisInfo(reader, &pColAgg, &allColumnsHaveAgg);
+
+    if (allColumnsHaveAgg == true) {
+      int32_t numOfCols = pBlock->info.numOfCols;
+
+      // todo create this buffer during creating operator
+      if (pBlock->pBlockAgg == NULL) {
+        pBlock->pBlockAgg = taosMemoryCalloc(numOfCols, POINTER_BYTES);
+      }
+
+      for (int32_t i = 0; i < numOfCols; ++i) {
+        SColMatchInfo* pColMatchInfo = taosArrayGet(pTableScanInfo->pColMatchInfo, i);
+        if (!pColMatchInfo->output) {
+          continue;
+        }
+        pBlock->pBlockAgg[pColMatchInfo->targetSlotId] = pColAgg[i];
+      }
+
+      return TSDB_CODE_SUCCESS;
+    } else {  // failed to load the block sma data, data block statistics does not exist, load data block instead
+      *status = FUNC_DATA_REQUIRED_DATA_LOAD;
+    }
+  }
+
+  ASSERT(*status == FUNC_DATA_REQUIRED_DATA_LOAD);
+
+  // todo filter data block according to the block sma data firstly
+#if 0
+  if (!doFilterByBlockStatistics(pBlock->pBlockStatis, pTableScanInfo->pCtx, pBlockInfo->rows)) {
+    pCost->filterOutBlocks += 1;
+    qDebug("%s data block filter out, brange:%" PRId64 "-%" PRId64 ", rows:%d", GET_TASKID(pTaskInfo), pBlockInfo->window.skey,
+           pBlockInfo->window.ekey, pBlockInfo->rows);
+    (*status) = FUNC_DATA_REQUIRED_FILTEROUT;
+    return TSDB_CODE_SUCCESS;
+  }
+#endif
+
+  pCost->totalCheckedRows += pBlock->info.rows;
+  pCost->loadBlocks += 1;
+
+  tsdbReaderT* reader = taosArrayGetP(pTableScanInfo->dataReaders, readerIdx);
+  SArray*      pCols = tsdbRetrieveDataBlock(reader, NULL);
+  if (pCols == NULL) {
+    return terrno;
+  }
+
+  relocateColumnData(pBlock, pTableScanInfo->pColMatchInfo, pCols);
+
+  // currently only the tbname pseudo column
+  if (pTableScanInfo->numOfPseudoExpr > 0) {
+    addTagPseudoColumnData(&pTableScanInfo->readHandle, pTableScanInfo->pPseudoExpr, pTableScanInfo->numOfPseudoExpr,
+                           pBlock);
+  }
+
+  int64_t st = taosGetTimestampMs();
+  doFilter(pTableScanInfo->pFilterNode, pBlock);
+
+  int64_t et = taosGetTimestampMs();
+  pTableScanInfo->readRecorder.filterTime += (et - st);
+
+  if (pBlock->info.rows == 0) {
+    pCost->filterOutBlocks += 1;
+    qDebug("%s data block filter out, brange:%" PRId64 "-%" PRId64 ", rows:%d", GET_TASKID(pTaskInfo),
+           pBlockInfo->window.skey, pBlockInfo->window.ekey, pBlockInfo->rows);
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
+typedef struct STableMergeScanSortSourceParam {
+  SOperatorInfo* pOperator;
+  int32_t        readerIdx;
+  SSDataBlock*   inputBlock;
+} STableMergeScanSortSourceParam;
+
+static SSDataBlock* getTableDataBlock(void* param) {
+  STableMergeScanSortSourceParam* source = param;
+  SOperatorInfo*                  pOperator = source->pOperator;
+  int32_t                         readerIdx = source->readerIdx;
+  SSDataBlock*                    pBlock = source->inputBlock;
+  STableMergeScanInfo*            pTableScanInfo = pOperator->info;
+
+  int64_t st = taosGetTimestampUs();
+
+  blockDataCleanup(pBlock);
+
+  tsdbReaderT* reader = taosArrayGetP(pTableScanInfo->dataReaders, readerIdx);
+  while (tsdbNextDataBlock(reader)) {
+    if (isTaskKilled(pOperator->pTaskInfo)) {
+      longjmp(pOperator->pTaskInfo->env, TSDB_CODE_TSC_QUERY_CANCELLED);
+    }
+
+    // process this data block based on the probabilities
+    bool processThisBlock = processBlockWithProbability(&pTableScanInfo->sample);
+    if (!processThisBlock) {
+      continue;
+    }
+
+    tsdbRetrieveDataBlockInfo(reader, &pBlock->info);
+
+    uint32_t status = 0;
+    int32_t  code = loadDataBlockFromOneTable(pOperator, pTableScanInfo, readerIdx, pBlock, &status);
+    //    int32_t  code = loadDataBlockOnDemand(pOperator->pRuntimeEnv, pTableScanInfo, pBlock, &status);
+    if (code != TSDB_CODE_SUCCESS) {
+      longjmp(pOperator->pTaskInfo->env, code);
+    }
+
+    // current block is filter out according to filter condition, continue load the next block
+    if (status == FUNC_DATA_REQUIRED_FILTEROUT || pBlock->info.rows == 0) {
+      continue;
+    }
+
+    uint64_t* groupId = taosHashGet(pOperator->pTaskInfo->tableqinfoList.map, &pBlock->info.uid, sizeof(int64_t));
+    if (groupId) {
+      pBlock->info.groupId = *groupId;
+    }
+
+    pOperator->resultInfo.totalRows = pTableScanInfo->readRecorder.totalRows;
+    pTableScanInfo->readRecorder.elapsedTime += (taosGetTimestampUs() - st) / 1000.0;
+
+    return pBlock;
+  }
+  return NULL;
+}
+
+SArray* generateSortByTsInfo(int32_t order) {
+  SArray*         pList = taosArrayInit(1, sizeof(SBlockOrderInfo));
+  SBlockOrderInfo bi = {0};
+  bi.order = order;
+  bi.slotId = 0;
+  bi.nullFirst = NULL_ORDER_FIRST;
+
+  taosArrayPush(pList, &bi);
+
+  return pList;
+}
+
+int32_t doOpenTableMergeScanOperator(SOperatorInfo* pOperator) {
+  STableMergeScanInfo* pInfo = pOperator->info;
+  SExecTaskInfo*       pTaskInfo = pOperator->pTaskInfo;
+
+  if (OPTR_IS_OPENED(pOperator)) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+  int32_t numOfBufPage = pInfo->sortBufSize / pInfo->bufPageSize;
+
+  pInfo->pSortHandle =
+      tsortCreateSortHandle(pInfo->pSortInfo, pInfo->pColMatchInfo, SORT_MULTISOURCE_MERGE, pInfo->bufPageSize,
+                            numOfBufPage, pInfo->pSortInputBlock, pTaskInfo->id.str);
+
+  tsortSetFetchRawDataFp(pInfo->pSortHandle, getTableDataBlock, NULL, NULL);
+
+  size_t numReaders = taosArrayGetSize(pInfo->dataReaders);
+  for (int32_t i = 0; i < numReaders; ++i) {
+    SSortSource*                    ps = taosMemoryCalloc(1, sizeof(SSortSource));
+    STableMergeScanSortSourceParam* param = taosArrayGet(pInfo->sortSourceParams, i);
+    ps->param = param;
+    tsortAddSource(pInfo->pSortHandle, ps);
+  }
+
+  int32_t code = tsortOpen(pInfo->pSortHandle);
+
+  if (code != TSDB_CODE_SUCCESS) {
+    longjmp(pTaskInfo->env, terrno);
+  }
+
+  pOperator->status = OP_RES_TO_RETURN;
+
+  OPTR_SET_OPENED(pOperator);
+  return TSDB_CODE_SUCCESS;
+}
+
+SSDataBlock* getSortedTableMergeScanBlockData(SSortHandle* pHandle, int32_t capacity, SOperatorInfo* pOperator) {
+  STableMergeScanInfo* pInfo = pOperator->info;
+  SExecTaskInfo*       pTaskInfo = pOperator->pTaskInfo;
+
+  SSDataBlock* p = tsortGetSortedDataBlock(pHandle);
+  if (p == NULL) {
+    return NULL;
+  }
+
+  blockDataEnsureCapacity(p, capacity);
+
+  while (1) {
+    STupleHandle* pTupleHandle = NULL;
+    if (pInfo->prefetchedTuple == NULL) {
+      pTupleHandle = tsortNextTuple(pHandle);
+    } else {
+      pTupleHandle = pInfo->prefetchedTuple;
+      pInfo->groupId = tsortGetGroupId(pTupleHandle);
+      pInfo->prefetchedTuple = NULL;
+    }
+
+    if (pTupleHandle == NULL) {
+      break;
+    }
+
+    uint64_t tupleGroupId = tsortGetGroupId(pTupleHandle);
+    if (!pInfo->hasGroupId) {
+      pInfo->groupId = tupleGroupId;
+      pInfo->hasGroupId = true;
+      appendOneRowToDataBlock(p, pTupleHandle);
+    } else if (pInfo->groupId == tupleGroupId) {
+      appendOneRowToDataBlock(p, pTupleHandle);
+    } else {
+      pInfo->prefetchedTuple = pTupleHandle;
+      break;
+    }
+
+    if (p->info.rows >= capacity) {
+      break;
+    }
+  }
+
+
+  qDebug("%s get sorted row blocks, rows:%d", GET_TASKID(pTaskInfo), p->info.rows);
+  return (p->info.rows > 0) ? p : NULL;
+}
+
+SSDataBlock* doTableMergeScan(SOperatorInfo* pOperator) {
+  if (pOperator->status == OP_EXEC_DONE) {
+    return NULL;
+  }
+
+  SExecTaskInfo*       pTaskInfo = pOperator->pTaskInfo;
+  STableMergeScanInfo* pInfo = pOperator->info;
+
+  int32_t code = pOperator->fpSet._openFn(pOperator);
+  if (code != TSDB_CODE_SUCCESS) {
+    longjmp(pTaskInfo->env, code);
+  }
+
+  SSDataBlock* pBlock =
+      getSortedTableMergeScanBlockData(pInfo->pSortHandle, pOperator->resultInfo.capacity, pOperator);
+
+  if (pBlock != NULL) {
+    pOperator->resultInfo.totalRows += pBlock->info.rows;
+  } else {
+    doSetOperatorCompleted(pOperator);
+  }
+  return pBlock;
+}
+
+void destroyTableMergeScanOperatorInfo(void* param, int32_t numOfOutput) {
+  STableMergeScanInfo* pTableScanInfo = (STableMergeScanInfo*)param;
+  clearupQueryTableDataCond(&pTableScanInfo->cond);
+
+  for (int32_t i = 0; i < taosArrayGetSize(pTableScanInfo->dataReaders); ++i) {
+    tsdbReaderT* reader = taosArrayGetP(pTableScanInfo->dataReaders, i);
+    tsdbCleanupReadHandle(reader);
+  }
+  taosArrayDestroy(pTableScanInfo->dataReaders);
+
+  if (pTableScanInfo->pColMatchInfo != NULL) {
+    taosArrayDestroy(pTableScanInfo->pColMatchInfo);
+  }
+
+  taosArrayDestroy(pTableScanInfo->sortSourceParams);
+  pTableScanInfo->pResBlock = blockDataDestroy(pTableScanInfo->pResBlock);
+  pTableScanInfo->pSortInputBlock = blockDataDestroy(pTableScanInfo->pSortInputBlock);
+
+  taosArrayDestroy(pTableScanInfo->pSortInfo);
+}
+
+typedef struct STableMergeScanExecInfo {
+  SFileBlockLoadRecorder blockRecorder;
+  SSortExecInfo sortExecInfo;
+} STableMergeScanExecInfo;
+
+int32_t getTableMergeScanExplainExecInfo(SOperatorInfo* pOptr, void** pOptrExplain, uint32_t* len) {
+  ASSERT(pOptr != NULL);
+  // TODO: merge these two info into one struct
+  STableMergeScanExecInfo* execInfo = taosMemoryCalloc(1, sizeof(STableMergeScanExecInfo));
+  STableMergeScanInfo*         pInfo = pOptr->info;
+  execInfo->blockRecorder = pInfo->readRecorder;
+  execInfo->sortExecInfo = tsortGetSortExecInfo(pInfo->pSortHandle);
+
+  *pOptrExplain = execInfo;
+  *len = sizeof(STableMergeScanExecInfo);
+  
+  return TSDB_CODE_SUCCESS;
+}
+
+SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanNode, SArray* dataReaders,
+                                                SReadHandle* readHandle, SExecTaskInfo* pTaskInfo) {
+  STableMergeScanInfo* pInfo = taosMemoryCalloc(1, sizeof(STableMergeScanInfo));
+  SOperatorInfo*       pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
+
+  if (pInfo == NULL || pOperator == NULL) {
+    goto _error;
+  }
+
+  SDataBlockDescNode* pDescNode = pTableScanNode->scan.node.pOutputDataBlockDesc;
+
+  int32_t numOfCols = 0;
+  SArray* pColList =
+      extractColMatchInfo(pTableScanNode->scan.pScanCols, pDescNode, &numOfCols, pTaskInfo, COL_MATCH_FROM_COL_ID);
+
+  int32_t code = initQueryTableDataCond(&pInfo->cond, pTableScanNode);
+  if (code != TSDB_CODE_SUCCESS) {
+    goto _error;
+  }
+
+  if (pTableScanNode->scan.pScanPseudoCols != NULL) {
+    pInfo->pPseudoExpr = createExprInfo(pTableScanNode->scan.pScanPseudoCols, NULL, &pInfo->numOfPseudoExpr);
+    pInfo->pPseudoCtx = createSqlFunctionCtx(pInfo->pPseudoExpr, pInfo->numOfPseudoExpr, &pInfo->rowCellInfoOffset);
+  }
+
+  pInfo->scanInfo = (SScanInfo){.numOfAsc = pTableScanNode->scanSeq[0], .numOfDesc = pTableScanNode->scanSeq[1]};
+
+  pInfo->readHandle = *readHandle;
+  pInfo->interval = extractIntervalInfo(pTableScanNode);
+  pInfo->sample.sampleRatio = pTableScanNode->ratio;
+  pInfo->sample.seed = taosGetTimestampSec();
+
+  pInfo->dataBlockLoadFlag = pTableScanNode->dataRequired;
+  pInfo->pFilterNode = pTableScanNode->scan.node.pConditions;
+  pInfo->dataReaders = dataReaders;
+  pInfo->scanFlag = MAIN_SCAN;
+  pInfo->pColMatchInfo = pColList;
+  pInfo->curTWinIdx = 0;
+
+  pInfo->pResBlock = createResDataBlock(pDescNode);
+
+  pInfo->sortSourceParams = taosArrayInit(taosArrayGetSize(dataReaders), sizeof(STableMergeScanSortSourceParam));
+  for (int32_t i = 0; i < taosArrayGetSize(dataReaders); ++i) {
+    STableMergeScanSortSourceParam* param = taosMemoryCalloc(1, sizeof(STableMergeScanSortSourceParam));
+    param->readerIdx = i;
+    param->pOperator = pOperator;
+    param->inputBlock = createOneDataBlock(pInfo->pResBlock, false);
+    taosArrayPush(pInfo->sortSourceParams, param);
+    taosMemoryFree(param);
+  }
+  pInfo->pSortInfo = generateSortByTsInfo(pInfo->cond.order);
+  pInfo->pSortInputBlock = createOneDataBlock(pInfo->pResBlock, false);
+  int32_t rowSize = pInfo->pResBlock->info.rowSize;
+  pInfo->bufPageSize = rowSize < 1024 ? 1024 : rowSize * 2;
+  pInfo->sortBufSize = pInfo->bufPageSize * 16;
+  pInfo->hasGroupId = false;
+  pInfo->prefetchedTuple = NULL;
+
+  pOperator->name = "TableMergeScanOperator";
+  // TODO : change it
+  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN;
+  pOperator->blocking = false;
+  pOperator->status = OP_NOT_OPENED;
+  pOperator->info = pInfo;
+  pOperator->numOfExprs = numOfCols;
+  pOperator->pTaskInfo = pTaskInfo;
+  initResultSizeInfo(pOperator, 1024);
+
+  pOperator->fpSet =
+      createOperatorFpSet(doOpenTableMergeScanOperator, doTableMergeScan, NULL, NULL, destroyTableMergeScanOperatorInfo,
+                          NULL, NULL, getTableMergeScanExplainExecInfo);
+  pOperator->cost.openCost = 0;
+  return pOperator;
+
+_error:
+  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
+  taosMemoryFree(pInfo);
+  taosMemoryFree(pOperator);
+  return NULL;
+}

source/libs/executor/src/sortoperator.c

@@ -76,7 +76,9 @@ void appendOneRowToDataBlock(SSDataBlock* pBlock, STupleHandle* pTupleHandle) {
       colDataAppendNULL(pColInfo, pBlock->info.rows);
     } else {
       char* pData = tsortGetValue(pTupleHandle, i);
-      colDataAppend(pColInfo, pBlock->info.rows, pData, false);
+      if (pData != NULL) {
+        colDataAppend(pColInfo, pBlock->info.rows, pData, false);
+      }
     }
   }
 
@@ -427,7 +429,7 @@ SOperatorInfo* createMultiwaySortMergeOperatorInfo(SOperatorInfo** downStreams,
   pInfo->pInputBlock = pInputBlock;
   pOperator->name = "MultiwaySortMerge";
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_MERGE;
-  pOperator->blocking = true;
+  pOperator->blocking = false;
   pOperator->status = OP_NOT_OPENED;
   pOperator->info = pInfo;
 

source/libs/executor/src/timewindowoperator.c

@@ -17,6 +17,7 @@
 #include "functionMgt.h"
 #include "tdatablock.h"
 #include "ttime.h"
+#include "tfill.h"
 
 typedef enum SResultTsInterpType {
   RESULT_ROW_START_INTERP = 1,
@@ -30,18 +31,18 @@ static int64_t* extractTsCol(SSDataBlock* pBlock, const SIntervalAggOperatorInfo
 static SResultRowPosition addToOpenWindowList(SResultRowInfo* pResultRowInfo, const SResultRow* pResult);
 static void doCloseWindow(SResultRowInfo* pResultRowInfo, const SIntervalAggOperatorInfo* pInfo, SResultRow* pResult);
 
-/*
+///*
- * There are two cases to handle:
+// * There are two cases to handle:
- *
+// *
- * 1. Query range is not set yet (queryRangeSet = 0). we need to set the query range info, including
+// * 1. Query range is not set yet (queryRangeSet = 0). we need to set the query range info, including
- * pQueryAttr->lastKey, pQueryAttr->window.skey, and pQueryAttr->eKey.
+// * pQueryAttr->lastKey, pQueryAttr->window.skey, and pQueryAttr->eKey.
- * 2. Query range is set and query is in progress. There may be another result with the same query ranges to be
+// * 2. Query range is set and query is in progress. There may be another result with the same query ranges to be
- *    merged during merge stage. In this case, we need the pTableQueryInfo->lastResRows to decide if there
+// *    merged during merge stage. In this case, we need the pTableQueryInfo->lastResRows to decide if there
- *    is a previous result generated or not.
+// *    is a previous result generated or not.
- */
+// */
-static void setIntervalQueryRange(STableQueryInfo* pTableQueryInfo, TSKEY key, STimeWindow* pQRange) {
+//static void setIntervalQueryRange(STableQueryInfo* pTableQueryInfo, TSKEY key, STimeWindow* pQRange) {
-  // do nothing
+//  // do nothing
-}
+//}
 
 static TSKEY getStartTsKey(STimeWindow* win, const TSKEY* tsCols) { return tsCols == NULL ? win->skey : tsCols[0]; }
 
@@ -327,8 +328,7 @@ void doTimeWindowInterpolation(SIntervalAggOperatorInfo* pInfo, int32_t numOfExp
 
   int32_t index = 1;
   for (int32_t k = 0; k < numOfExprs; ++k) {
-    // todo use flag instead of function name
+    if (!fmIsIntervalInterpoFunc(pCtx[k].functionId)) {
-    if (strcmp(pCtx[k].pExpr->pExpr->_function.functionName, "twa") != 0) {
       pCtx[k].start.key = INT64_MIN;
       continue;
     }
@@ -341,7 +341,7 @@ void doTimeWindowInterpolation(SIntervalAggOperatorInfo* pInfo, int32_t numOfExp
     SFunctParam*     pParam = &pCtx[k].param[0];
     SColumnInfoData* pColInfo = taosArrayGet(pDataBlock, pParam->pCol->slotId);
 
-    ASSERT(pColInfo->info.colId == pParam->pCol->colId && curTs != windowKey);
+    ASSERT(pColInfo->info.type == pParam->pCol->type && curTs != windowKey);
 
     double v1 = 0, v2 = 0, v = 0;
     if (prevRowIndex == -1) {
@@ -958,9 +958,6 @@ static int32_t doOpenIntervalAgg(SOperatorInfo* pOperator) {
 
     // the pDataBlock are always the same one, no need to call this again
     setInputDataBlock(pOperator, pInfo->binfo.pCtx, pBlock, pInfo->order, scanFlag, true);
-    STableQueryInfo* pTableQueryInfo = pInfo->pCurrent;
-
-    setIntervalQueryRange(pTableQueryInfo, pBlock->info.window.skey, &pTaskInfo->window);
     hashIntervalAgg(pOperator, &pInfo->binfo.resultRowInfo, pBlock, scanFlag, NULL);
 
 #if 0  // test for encode/decode result info
@@ -1414,7 +1411,7 @@ static bool timeWindowinterpNeeded(SqlFunctionCtx* pCtx, int32_t numOfCols, SInt
   for (int32_t i = 0; i < numOfCols; ++i) {
     SExprInfo* pExpr = pCtx[i].pExpr;
 
-    if (strcmp(pExpr->pExpr->_function.functionName, "twa") == 0) {
+    if (fmIsIntervalInterpoFunc(pCtx[i].functionId)) {
       SFunctParam* pParam = &pExpr->base.pParam[0];
 
       SColumn c = *pParam->pCol;
@@ -1475,11 +1472,9 @@ SOperatorInfo* createIntervalOperatorInfo(SOperatorInfo* downstream, SExprInfo*
   pInfo->timeWindowInterpo = timeWindowinterpNeeded(pInfo->binfo.pCtx, numOfCols, pInfo);
   if (pInfo->timeWindowInterpo) {
     pInfo->binfo.resultRowInfo.openWindow = tdListNew(sizeof(SResultRowPosition));
-  }
+    if (pInfo->binfo.resultRowInfo.openWindow == NULL) {
-
+      goto _error;
-  //  pInfo->pTableQueryInfo = initTableQueryInfo(pTableGroupInfo);
+    }
-  if (code != TSDB_CODE_SUCCESS /* || pInfo->pTableQueryInfo == NULL*/) {
-    goto _error;
   }
 
   initResultRowInfo(&pInfo->binfo.resultRowInfo, (int32_t)1);
@@ -1694,24 +1689,44 @@ static SSDataBlock* doSessionWindowAgg(SOperatorInfo* pOperator) {
   return (rows == 0) ? NULL : pBInfo->pRes;
 }
 
-static SSDataBlock* doAllIntervalAgg(SOperatorInfo* pOperator) {
+static void doKeepPrevRows(STimeSliceOperatorInfo* pSliceInfo, const SSDataBlock* pBlock) {
+  int32_t numOfCols = taosArrayGetSize(pBlock->pDataBlock);
+  for(int32_t i = 0; i < numOfCols; ++i) {
+    SColumnInfoData*  pColInfoData = taosArrayGet(pBlock->pDataBlock, i);
+
+    // null data should not be kept since it can not be used to perform interpolation
+    if (!colDataIsNull_s(pColInfoData, i)) {
+      SGroupKeys* pkey = taosArrayGet(pSliceInfo->pPrevRow, i);
+
+      pkey->isNull = false;
+      char* val = colDataGetData(pColInfoData, i);
+      memcpy(pkey->pData, val, pkey->bytes);
+    }
+  }
+}
+
+static SSDataBlock* doTimeslice(SOperatorInfo* pOperator) {
   if (pOperator->status == OP_EXEC_DONE) {
     return NULL;
   }
 
   STimeSliceOperatorInfo* pSliceInfo = pOperator->info;
-  if (pOperator->status == OP_RES_TO_RETURN) {
+  SSDataBlock* pResBlock = pSliceInfo->binfo.pRes;
-    //    doBuildResultDatablock(&pRuntimeEnv->groupResInfo, pRuntimeEnv, pIntervalInfo->pRes);
-    if (pSliceInfo->binfo.pRes->info.rows == 0 || !hashRemainDataInGroupInfo(&pSliceInfo->groupResInfo)) {
-      doSetOperatorCompleted(pOperator);
-    }
 
-    return pSliceInfo->binfo.pRes;
+//  if (pOperator->status == OP_RES_TO_RETURN) {
-  }
+//    //    doBuildResultDatablock(&pRuntimeEnv->groupResInfo, pRuntimeEnv, pIntervalInfo->pRes);
+//    if (pResBlock->info.rows == 0 || !hashRemainDataInGroupInfo(&pSliceInfo->groupResInfo)) {
+//      doSetOperatorCompleted(pOperator);
+//    }
+//
+//    return pResBlock;
+//  }
 
-  int32_t        order = TSDB_ORDER_ASC;
+  int32_t order = TSDB_ORDER_ASC;
+  SInterval* pInterval = &pSliceInfo->interval;
   SOperatorInfo* downstream = pOperator->pDownstream[0];
 
+  int32_t numOfRows = 0;
   while (1) {
     SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
     if (pBlock == NULL) {
@@ -1720,48 +1735,198 @@ static SSDataBlock* doAllIntervalAgg(SOperatorInfo* pOperator) {
 
     // the pDataBlock are always the same one, no need to call this again
     setInputDataBlock(pOperator, pSliceInfo->binfo.pCtx, pBlock, order, MAIN_SCAN, true);
-    //    hashAllIntervalAgg(pOperator, &pSliceInfo->binfo.resultRowInfo, pBlock, 0);
+
+    SColumnInfoData* pTsCol = taosArrayGet(pBlock->pDataBlock, 0);
+    for(int32_t i = 0; i < pBlock->info.rows; ++i) {
+      int64_t ts = *(int64_t*) colDataGetData(pTsCol, i);
+
+      if (ts == pSliceInfo->current) {
+        for(int32_t j = 0; j < pOperator->numOfExprs; ++j) {
+          SExprInfo* pExprInfo = &pOperator->pExpr[j];
+          int32_t dstSlot = pExprInfo->base.resSchema.slotId;
+          int32_t srcSlot = pExprInfo->base.pParam[0].pCol->slotId;
+
+          SColumnInfoData* pSrc = taosArrayGet(pBlock->pDataBlock, srcSlot);
+          SColumnInfoData* pDst = taosArrayGet(pBlock->pDataBlock, dstSlot);
+
+          char* v = colDataGetData(pSrc, i);
+          colDataAppend(pDst, numOfRows, v, false);
+        }
+
+        numOfRows += 1;
+
+        pSliceInfo->current += taosTimeAdd(pSliceInfo->current, pInterval->interval, pInterval->intervalUnit, pInterval->precision);
+        if (pSliceInfo->current > pSliceInfo->win.ekey) {
+          doSetOperatorCompleted(pOperator);
+          break;
+        }
+      } else if (ts < pSliceInfo->current) {
+        if (i != pBlock->info.window.ekey) {
+          int64_t nextTs = *(int64_t*) colDataGetData(pTsCol, i + 1);
+          if (nextTs > pSliceInfo->current) {
+            // output the result
+            for (int32_t j = 0; j < pOperator->numOfExprs; ++j) {
+              SExprInfo* pExprInfo = &pOperator->pExpr[j];
+              int32_t    dstSlot = pExprInfo->base.resSchema.slotId;
+              int32_t    srcSlot = pExprInfo->base.pParam[0].pCol->slotId;
+
+              SColumnInfoData* pSrc = taosArrayGet(pBlock->pDataBlock, srcSlot);
+              SColumnInfoData* pDst = taosArrayGet(pBlock->pDataBlock, dstSlot);
+
+              switch (pSliceInfo->fillType) {
+                case TSDB_FILL_NULL:
+                  colDataAppendNULL(pDst, numOfRows);
+                  break;
+
+                case TSDB_FILL_SET_VALUE: {
+                  SVariant* pVar = &pSliceInfo->pFillColInfo[i].fillVal;
+
+                  if (pDst->info.type == TSDB_DATA_TYPE_FLOAT) {
+                    float v = 0;
+                    GET_TYPED_DATA(v, float, pVar->nType, &pVar->i);
+                    colDataAppend(pDst, numOfRows, (char*)&v, false);
+                  } else if (pDst->info.type == TSDB_DATA_TYPE_DOUBLE) {
+                    double v = 0;
+                    GET_TYPED_DATA(v, double, pVar->nType, &pVar->i);
+                    colDataAppend(pDst, numOfRows, (char*)&v, false);
+                  } else if (IS_SIGNED_NUMERIC_TYPE(pDst->info.type)) {
+                    int64_t v = 0;
+                    GET_TYPED_DATA(v, int64_t, pVar->nType, &pVar->i);
+                    colDataAppend(pDst, numOfRows, (char*)&v, false);
+                  }
+                }
+                break;
+
+                case TSDB_FILL_LINEAR:
+#if 0
+                if (pCtx->start.key == INT64_MIN || pCtx->start.key > pCtx->startTs
+                    || pCtx->end.key == INT64_MIN || pCtx->end.key < pCtx->startTs) {
+//                  goto interp_exit;
+                }
+
+              double v1 = -1, v2 = -1;
+              GET_TYPED_DATA(v1, double, pCtx->inputType, &pCtx->start.val);
+              GET_TYPED_DATA(v2, double, pCtx->inputType, &pCtx->end.val);
+
+              SPoint point1 = {.key = ts, .val = &v1};
+              SPoint point2 = {.key = nextTs, .val = &v2};
+              SPoint point  = {.key = pCtx->startTs, .val = pCtx->pOutput};
+
+              int32_t srcType = pCtx->inputType;
+              if (isNull((char *)&pCtx->start.val, srcType) || isNull((char *)&pCtx->end.val, srcType)) {
+                setNull(pCtx->pOutput, srcType, pCtx->inputBytes);
+              } else {
+                bool exceedMax = false, exceedMin = false;
+                taosGetLinearInterpolationVal(&point, pCtx->outputType, &point1, &point2, TSDB_DATA_TYPE_DOUBLE, &exceedMax, &exceedMin);
+                if (exceedMax || exceedMin) {
+                  __compar_fn_t func = getComparFunc((int32_t)pCtx->inputType, 0);
+                  if (func(&pCtx->start.val, &pCtx->end.val) <= 0) {
+                    COPY_TYPED_DATA(pCtx->pOutput, pCtx->inputType, exceedMax ? &pCtx->start.val : &pCtx->end.val);
+                  } else {
+                    COPY_TYPED_DATA(pCtx->pOutput, pCtx->inputType, exceedMax ? &pCtx->end.val : &pCtx->start.val);
+                  }
+                }
+              }
+#endif
+                  break;
+
+                case TSDB_FILL_PREV: {
+                  SGroupKeys* pkey = taosArrayGet(pSliceInfo->pPrevRow, srcSlot);
+                  colDataAppend(pDst, numOfRows, pkey->pData, false);
+                } break;
+
+                case TSDB_FILL_NEXT: {
+                } break;
+
+                case TSDB_FILL_NONE:
+                default:
+                  break;
+              }
+
+              pSliceInfo->current +=
+                  taosTimeAdd(pSliceInfo->current, pInterval->interval, pInterval->intervalUnit, pInterval->precision);
+              if (pSliceInfo->current > pSliceInfo->win.ekey) {
+                doSetOperatorCompleted(pOperator);
+                break;
+              }
+            }
+          } else {
+            // ignore current row, and do nothing
+          }
+        } else {  // it is the last row of current block
+          doKeepPrevRows(pSliceInfo, pBlock);
+        }
+      }
+    }
   }
 
   // restore the value
-  pOperator->status = OP_RES_TO_RETURN;
-  closeAllResultRows(&pSliceInfo->binfo.resultRowInfo);
   setTaskStatus(pOperator->pTaskInfo, TASK_COMPLETED);
-  //  finalizeQueryResult(pSliceInfo->binfo.pCtx, pOperator->numOfExprs);
+  if (pResBlock->info.rows == 0) {
-
-  //  initGroupedResultInfo(&pSliceInfo->groupResInfo, &pSliceInfo->binfo.resultRowInfo);
-  //  doBuildResultDatablock(&pRuntimeEnv->groupResInfo, pRuntimeEnv, pSliceInfo->pRes);
-
-  if (pSliceInfo->binfo.pRes->info.rows == 0 || !hashRemainDataInGroupInfo(&pSliceInfo->groupResInfo)) {
     pOperator->status = OP_EXEC_DONE;
   }
 
-  return pSliceInfo->binfo.pRes->info.rows == 0 ? NULL : pSliceInfo->binfo.pRes;
+  return pResBlock->info.rows == 0 ? NULL : pResBlock;
+}
+
+static int32_t initTimesliceInfo(STimeSliceOperatorInfo* pInfo, SqlFunctionCtx* pCtx, int32_t numOfCols) {
+  pInfo->pPrevRow = taosArrayInit(4, sizeof(SGroupKeys));
+  pInfo->pCols = taosArrayInit(4, sizeof(SColumn));
+
+  if (pInfo->pPrevRow == NULL || pInfo->pCols == NULL) {
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+
+  for (int32_t i = 0; i < numOfCols; ++i) {
+    SExprInfo* pExpr = pCtx[i].pExpr;
+
+    SFunctParam* pParam = &pExpr->base.pParam[0];
+
+    SColumn c = *pParam->pCol;
+    taosArrayPush(pInfo->pCols, &c);
+
+    SGroupKeys key = {0};
+    key.bytes = c.bytes;
+    key.type = c.type;
+    key.isNull = false;
+    key.pData = taosMemoryCalloc(1, c.bytes);
+    taosArrayPush(pInfo->pPrevRow, &key);
+  }
+
+  return TSDB_CODE_SUCCESS;
 }
 
 SOperatorInfo* createTimeSliceOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t numOfCols,
-                                           SSDataBlock* pResultBlock, SExecTaskInfo* pTaskInfo) {
+                                           SSDataBlock* pResultBlock, const SNodeListNode* pValNode, SExecTaskInfo* pTaskInfo) {
   STimeSliceOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(STimeSliceOperatorInfo));
   SOperatorInfo*          pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
   if (pOperator == NULL || pInfo == NULL) {
     goto _error;
   }
 
+  int32_t code = initTimesliceInfo(pInfo, pInfo->binfo.pCtx, numOfCols);
+  if (code != TSDB_CODE_SUCCESS) {
+    goto _error;
+  }
+
   initResultRowInfo(&pInfo->binfo.resultRowInfo, 8);
+  pInfo->pFillColInfo = createFillColInfo(pExprInfo, numOfCols, pValNode);
 
-  pOperator->name = "TimeSliceOperator";
+  pInfo->binfo.pRes     = pResultBlock;
+
+  pOperator->name       = "TimeSliceOperator";
   //  pOperator->operatorType = OP_AllTimeWindow;
-  pOperator->blocking = true;
+  pOperator->blocking   = true;
-  pOperator->status = OP_NOT_OPENED;
+  pOperator->status     = OP_NOT_OPENED;
-  pOperator->pExpr = pExprInfo;
+  pOperator->pExpr      = pExprInfo;
   pOperator->numOfExprs = numOfCols;
-  pOperator->info = pInfo;
+  pOperator->info       = pInfo;
-  pOperator->pTaskInfo = pTaskInfo;
+  pOperator->pTaskInfo  = pTaskInfo;
 
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doAllIntervalAgg, NULL, NULL, destroyBasicOperatorInfo,
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doTimeslice, NULL, NULL, destroyBasicOperatorInfo,
                                          NULL, NULL, NULL);
 
-  int32_t code = appendDownstream(pOperator, &downstream, 1);
+  code = appendDownstream(pOperator, &downstream, 1);
   return pOperator;
 
 _error:
@@ -3344,9 +3509,6 @@ static SSDataBlock* doMergeIntervalAgg(SOperatorInfo* pOperator) {
 
       getTableScanInfo(pOperator, &iaInfo->order, &scanFlag);
       setInputDataBlock(pOperator, iaInfo->binfo.pCtx, pBlock, iaInfo->order, scanFlag, true);
-      STableQueryInfo* pTableQueryInfo = iaInfo->pCurrent;
-
-      setIntervalQueryRange(pTableQueryInfo, pBlock->info.window.skey, &pTaskInfo->window);
       doMergeIntervalAggImpl(pOperator, &iaInfo->binfo.resultRowInfo, pBlock, scanFlag, pRes);
 
       if (pRes->info.rows >= pOperator->resultInfo.threshold) {

source/libs/executor/src/tsort.c

@@ -738,7 +738,11 @@ bool tsortIsNullVal(STupleHandle* pVHandle, int32_t colIndex) {
 
 void* tsortGetValue(STupleHandle* pVHandle, int32_t colIndex) {
   SColumnInfoData* pColInfo = TARRAY_GET_ELEM(pVHandle->pBlock->pDataBlock, colIndex);
-  return colDataGetData(pColInfo, pVHandle->rowIndex);
+  if (pColInfo->pData == NULL) {
+    return NULL;
+  } else {
+    return colDataGetData(pColInfo, pVHandle->rowIndex);
+  }
 }
 
 uint64_t tsortGetGroupId(STupleHandle* pVHandle) {

source/libs/function/inc/builtinsimpl.h

@@ -97,6 +97,10 @@ bool getDiffFuncEnv(struct SFunctionNode* pFunc, SFuncExecEnv* pEnv);
 bool diffFunctionSetup(SqlFunctionCtx *pCtx, SResultRowEntryInfo* pResInfo);
 int32_t diffFunction(SqlFunctionCtx *pCtx);
 
+bool getDerivativeFuncEnv(struct SFunctionNode* pFunc, SFuncExecEnv* pEnv);
+bool derivativeFuncSetup(SqlFunctionCtx *pCtx, SResultRowEntryInfo* pResInfo);
+int32_t derivativeFunction(SqlFunctionCtx *pCtx);
+
 bool getFirstLastFuncEnv(struct SFunctionNode* pFunc, SFuncExecEnv* pEnv);
 int32_t firstFunction(SqlFunctionCtx *pCtx);
 int32_t firstFunctionMerge(SqlFunctionCtx *pCtx);

source/libs/function/inc/functionMgtInt.h

@@ -42,7 +42,8 @@ extern "C" {
 #define FUNC_MGT_SELECT_FUNC            FUNC_MGT_FUNC_CLASSIFICATION_MASK(13)
 #define FUNC_MGT_REPEAT_SCAN_FUNC       FUNC_MGT_FUNC_CLASSIFICATION_MASK(14)
 #define FUNC_MGT_FORBID_FILL_FUNC       FUNC_MGT_FUNC_CLASSIFICATION_MASK(15)
-#define FUNC_MGT_FORBID_STREAM_FUNC     FUNC_MGT_FUNC_CLASSIFICATION_MASK(16)
+#define FUNC_MGT_INTERVAL_INTERPO_FUNC  FUNC_MGT_FUNC_CLASSIFICATION_MASK(16)
+#define FUNC_MGT_FORBID_STREAM_FUNC     FUNC_MGT_FUNC_CLASSIFICATION_MASK(17)
 
 #define FUNC_MGT_TEST_MASK(val, mask) (((val) & (mask)) != 0)
 

source/libs/function/inc/texpr.h

@@ -60,12 +60,6 @@ typedef struct SExprTraverseSupp {
   void                  *pExtInfo;
 } SExprTraverseSupp;
 
-tExprNode* exprTreeFromTableName(const char* tbnameCond);
-
-bool exprTreeApplyFilter(tExprNode *pExpr, const void *pItem, SExprTraverseSupp *param);
-
-void buildFilterSetFromBinary(void **q, const char *buf, int32_t len);
-
 #ifdef __cplusplus
 }
 #endif

source/libs/function/inc/tunaryoperator.h

@@ -1,32 +0,0 @@
-/*
- * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
- *
- * This program is free software: you can use, redistribute, and/or modify
- * it under the terms of the GNU Affero General Public License, version 3
- * or later ("AGPL"), as published by the Free Software Foundation.
- *
- * This program is distributed in the hope that it will be useful, but WITHOUT
- * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
- * FITNESS FOR A PARTICULAR PURPOSE.
- *
- * You should have received a copy of the GNU Affero General Public License
- * along with this program. If not, see <http://www.gnu.org/licenses/>.
- */
-
-#ifndef _TD_COMMON_UNARY_SCALAR_OPERATOR_H_
-#define _TD_COMMON_UNARY_SCALAR_OPERATOR_H_
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-//#include "tscalarfunction.h"
-
-//typedef void (*_unary_scalar_fn_t)(SScalarParam *pLeft, SScalarParam* pOutput);
-//_unary_scalar_fn_t getUnaryScalarOperatorFn(int32_t binOperator);
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif  /*_TD_COMMON_BIN_SCALAR_OPERATOR_H_*/

source/libs/function/src/builtins.c

@@ -997,6 +997,38 @@ static int32_t translateLastRow(SFunctionNode* pFunc, char* pErrBuf, int32_t len
   return TSDB_CODE_SUCCESS;
 }
 
+static int32_t translateDerivative(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
+  if (3 != LIST_LENGTH(pFunc->pParameterList)) {
+    return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
+  }
+
+  uint8_t colType = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType.type;
+
+  // param1
+  SNode* pParamNode1 = nodesListGetNode(pFunc->pParameterList, 1);
+  if (QUERY_NODE_VALUE != nodeType(pParamNode1)) {
+    return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
+  }
+
+  SValueNode* pValue = (SValueNode*)pParamNode1;
+  pValue->notReserved = true;
+
+  if (!IS_NUMERIC_TYPE(colType)) {
+    return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
+  }
+
+  SNode* pParamNode2 = nodesListGetNode(pFunc->pParameterList, 2);
+  SValueNode* pValue2 = (SValueNode*)pParamNode2;
+  pValue2->notReserved = true;
+
+  if (pValue2->datum.i != 0 && pValue2->datum.i != 1) {
+    return invaildFuncParaValueErrMsg(pErrBuf, len, pFunc->functionName);
+  }
+
+  pFunc->node.resType = (SDataType){.bytes = tDataTypes[TSDB_DATA_TYPE_DOUBLE].bytes, .type = TSDB_DATA_TYPE_DOUBLE};
+  return TSDB_CODE_SUCCESS;
+}
+
 static int32_t translateFirstLast(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
   // first(col_list) will be rewritten as first(col)
   if (1 != LIST_LENGTH(pFunc->pParameterList)) {
@@ -1596,6 +1628,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
     .type = FUNCTION_TYPE_AVG,
     .classification = FUNC_MGT_AGG_FUNC,
     .translateFunc = translateInNumOutDou,
+    .dataRequiredFunc = statisDataRequired,
     .getEnvFunc   = getAvgFuncEnv,
     .initFunc     = avgFunctionSetup,
     .processFunc  = avgFunction,
@@ -1793,7 +1826,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "elapsed",
     .type = FUNCTION_TYPE_ELAPSED,
-    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_FORBID_STREAM_FUNC,
+    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_INTERVAL_INTERPO_FUNC | FUNC_MGT_FORBID_STREAM_FUNC,
     .dataRequiredFunc = statisDataRequired,
     .translateFunc = translateElapsed,
     .getEnvFunc   = getElapsedFuncEnv,
@@ -1831,6 +1864,26 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
     .invertFunc   = NULL,
     .combineFunc  = elapsedCombine,
   },
+  {
+    .name = "interp",
+    .type = FUNCTION_TYPE_INTERP,
+    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_INTERVAL_INTERPO_FUNC,
+    .translateFunc = translateFirstLast,
+    .getEnvFunc    = getSelectivityFuncEnv,
+    .initFunc      = functionSetup,
+    .processFunc   = NULL,
+    .finalizeFunc  = NULL
+  },
+  {
+    .name = "derivative",
+    .type = FUNCTION_TYPE_DERIVATIVE,
+    .classification = FUNC_MGT_INDEFINITE_ROWS_FUNC | FUNC_MGT_TIMELINE_FUNC,
+    .translateFunc = translateDerivative,
+    .getEnvFunc   = getDerivativeFuncEnv,
+    .initFunc     = derivativeFuncSetup,
+    .processFunc  = derivativeFunction,
+    .finalizeFunc = functionFinalize
+  },
   {
     .name = "last_row",
     .type = FUNCTION_TYPE_LAST_ROW,
@@ -1914,8 +1967,9 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "twa",
     .type = FUNCTION_TYPE_TWA,
-    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_FORBID_STREAM_FUNC,
+    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_INTERVAL_INTERPO_FUNC | FUNC_MGT_FORBID_STREAM_FUNC,
     .translateFunc = translateInNumOutDou,
+    .dataRequiredFunc = statisDataRequired,
     .getEnvFunc    = getTwaFuncEnv,
     .initFunc      = twaFunctionSetup,
     .processFunc   = twaFunction,

source/libs/function/src/builtinsimpl.c

@@ -990,9 +990,6 @@ int32_t doMinMaxHelper(SqlFunctionCtx* pCtx, int32_t isMinFunc) {
       index = pInput->pColumnDataAgg[0]->maxIndex;
     }
 
-    // the index is the original position, not the relative position
-    TSKEY key = (pCtx->ptsList != NULL) ? pCtx->ptsList[index] : TSKEY_INITIAL_VAL;
-
     if (!pBuf->assign) {
       pBuf->v = *(int64_t*)tval;
       if (pCtx->subsidiaries.num > 0) {
@@ -3424,7 +3421,7 @@ bool elapsedFunctionSetup(SqlFunctionCtx* pCtx, SResultRowEntryInfo* pResultInfo
   pInfo->min = MAX_TS_KEY;
   pInfo->max = 0;
 
-  if (pCtx->numOfParams == 2) {
+  if (pCtx->numOfParams > 2) {
     pInfo->timeUnit = pCtx->param[1].param.i;
   } else {
     pInfo->timeUnit = 1;
@@ -3474,8 +3471,8 @@ int32_t elapsedFunction(SqlFunctionCtx* pCtx) {
 
     SColumnInfoData* pCol = pInput->pData[0];
 
-    int32_t start = pInput->startRowIndex;
+    int32_t start     = pInput->startRowIndex;
-    TSKEY*  ptsList = (int64_t*)colDataGetData(pCol, start);
+    TSKEY* ptsList = (int64_t*)colDataGetData(pCol, 0);
     if (pCtx->order == TSDB_ORDER_DESC) {
       if (pCtx->start.key == INT64_MIN) {
         pInfo->max =
@@ -5166,3 +5163,198 @@ int32_t blockDistFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
 
   return row;
 }
+
+typedef struct SDerivInfo {
+  double   prevValue;     // previous value
+  TSKEY    prevTs;        // previous timestamp
+  bool     ignoreNegative;// ignore the negative value
+  int64_t  tsWindow;      // time window for derivative
+  bool     valueSet;      // the value has been set already
+} SDerivInfo;
+
+bool getDerivativeFuncEnv(struct SFunctionNode* pFunc, SFuncExecEnv* pEnv) {
+  pEnv->calcMemSize = sizeof(SDerivInfo);
+  return true;
+}
+
+bool derivativeFuncSetup(SqlFunctionCtx *pCtx, SResultRowEntryInfo* pResInfo) {
+  if (!functionSetup(pCtx, pResInfo)) {
+    return false;  // not initialized since it has been initialized
+  }
+
+  SDerivInfo* pDerivInfo = GET_ROWCELL_INTERBUF(pResInfo);
+
+  pDerivInfo->ignoreNegative = pCtx->param[2].param.i;
+  pDerivInfo->prevTs   = -1;
+  pDerivInfo->tsWindow = pCtx->param[1].param.i;
+  pDerivInfo->valueSet = false;
+  return true;
+}
+
+int32_t derivativeFunction(SqlFunctionCtx *pCtx) {
+  SResultRowEntryInfo *pResInfo = GET_RES_INFO(pCtx);
+  SDerivInfo* pDerivInfo = GET_ROWCELL_INTERBUF(pResInfo);
+
+  SInputColumnInfoData* pInput = &pCtx->input;
+  SColumnInfoData* pInputCol = pInput->pData[0];
+
+  int32_t numOfElems = 0;
+  SColumnInfoData* pOutput = (SColumnInfoData*)pCtx->pOutput;
+  SColumnInfoData* pTsOutput = pCtx->pTsOutput;
+
+  int32_t i = pInput->startRowIndex;
+  TSKEY* tsList = (int64_t*)pInput->pPTS->pData;
+
+  double v = 0;
+
+  if (pCtx->order == TSDB_ORDER_ASC) {
+    for (; i < pInput->numOfRows + pInput->startRowIndex; i += 1) {
+      if (colDataIsNull_f(pInputCol->nullbitmap, i)) {
+        continue;
+      }
+
+      char*  d = (char*)pInputCol->pData + pInputCol->info.bytes * i;
+      GET_TYPED_DATA(v, double, pInputCol->info.type, d);
+
+      int32_t pos = pCtx->offset + numOfElems;
+      if (!pDerivInfo->valueSet) {  // initial value is not set yet
+        pDerivInfo->valueSet = true;
+      } else {
+        double r = ((v - pDerivInfo->prevValue) * pDerivInfo->tsWindow) / (tsList[i] - pDerivInfo->prevTs);
+        if (pDerivInfo->ignoreNegative && r < 0) {
+        } else {
+          colDataAppend(pOutput, pos, (const char*)&r, false);
+          if (pTsOutput != NULL) {
+            colDataAppendInt64(pTsOutput, pos, &tsList[i]);
+          }
+          numOfElems++;
+        }
+      }
+
+      pDerivInfo->prevValue = v;
+      pDerivInfo->prevTs = tsList[i];
+    }
+  } else {
+    for (; i < pInput->numOfRows + pInput->startRowIndex; i += 1) {
+      if (colDataIsNull_f(pInputCol->nullbitmap, i)) {
+        continue;
+      }
+
+      char*  d = (char*)pInputCol->pData + pInputCol->info.bytes * i;
+      GET_TYPED_DATA(v, double, pInputCol->info.type, d);
+
+      int32_t pos = pCtx->offset + numOfElems;
+      if (!pDerivInfo->valueSet) {  // initial value is not set yet
+        pDerivInfo->valueSet = true;
+      } else {
+        double r = ((pDerivInfo->prevValue - v) * pDerivInfo->tsWindow) / (pDerivInfo->prevTs - tsList[i]);
+        if (pDerivInfo->ignoreNegative && r < 0) {
+        } else {
+          colDataAppend(pOutput, pos, (const char*)&r, false);
+          if (pTsOutput != NULL) {
+            colDataAppendInt64(pTsOutput, pos, &pDerivInfo->prevTs);
+          }
+          numOfElems++;
+        }
+      }
+
+      pDerivInfo->prevValue = v;
+      pDerivInfo->prevTs = tsList[i];
+    }
+  }
+
+  return numOfElems;
+}
+
+int32_t interpFunction(SqlFunctionCtx *pCtx) {
+#if 0
+  int32_t fillType = (int32_t) pCtx->param[2].i64;
+  //bool ascQuery = (pCtx->order == TSDB_ORDER_ASC);
+
+  if (pCtx->start.key == pCtx->startTs) {
+    assert(pCtx->start.key != INT64_MIN);
+
+    COPY_TYPED_DATA(pCtx->pOutput, pCtx->inputType, &pCtx->start.val);
+
+    goto interp_success_exit;
+  } else if (pCtx->end.key == pCtx->startTs && pCtx->end.key != INT64_MIN && fillType == TSDB_FILL_NEXT) {
+    COPY_TYPED_DATA(pCtx->pOutput, pCtx->inputType, &pCtx->end.val);
+
+    goto interp_success_exit;
+  }
+
+  switch (fillType) {
+    case TSDB_FILL_NULL:
+      setNull(pCtx->pOutput, pCtx->outputType, pCtx->outputBytes);
+      break;
+
+    case TSDB_FILL_SET_VALUE:
+      tVariantDump(&pCtx->param[1], pCtx->pOutput, pCtx->inputType, true);
+      break;
+
+    case TSDB_FILL_LINEAR:
+      if (pCtx->start.key == INT64_MIN || pCtx->start.key > pCtx->startTs
+          || pCtx->end.key == INT64_MIN || pCtx->end.key < pCtx->startTs) {
+        goto interp_exit;
+      }
+
+      double v1 = -1, v2 = -1;
+      GET_TYPED_DATA(v1, double, pCtx->inputType, &pCtx->start.val);
+      GET_TYPED_DATA(v2, double, pCtx->inputType, &pCtx->end.val);
+
+      SPoint point1 = {.key = pCtx->start.key, .val = &v1};
+      SPoint point2 = {.key = pCtx->end.key, .val = &v2};
+      SPoint point  = {.key = pCtx->startTs, .val = pCtx->pOutput};
+
+      int32_t srcType = pCtx->inputType;
+      if (isNull((char *)&pCtx->start.val, srcType) || isNull((char *)&pCtx->end.val, srcType)) {
+        setNull(pCtx->pOutput, srcType, pCtx->inputBytes);
+      } else {
+        bool exceedMax = false, exceedMin = false;
+        taosGetLinearInterpolationVal(&point, pCtx->outputType, &point1, &point2, TSDB_DATA_TYPE_DOUBLE, &exceedMax, &exceedMin);
+        if (exceedMax || exceedMin) {
+          __compar_fn_t func = getComparFunc((int32_t)pCtx->inputType, 0);
+          if (func(&pCtx->start.val, &pCtx->end.val) <= 0) {
+            COPY_TYPED_DATA(pCtx->pOutput, pCtx->inputType, exceedMax ? &pCtx->start.val : &pCtx->end.val);
+          } else {
+            COPY_TYPED_DATA(pCtx->pOutput, pCtx->inputType, exceedMax ? &pCtx->end.val : &pCtx->start.val);
+          }
+        }
+      }
+      break;
+
+    case TSDB_FILL_PREV:
+      if (pCtx->start.key == INT64_MIN || pCtx->start.key > pCtx->startTs) {
+        goto interp_exit;
+      }
+
+      COPY_TYPED_DATA(pCtx->pOutput, pCtx->inputType, &pCtx->start.val);
+      break;
+
+    case TSDB_FILL_NEXT:
+      if (pCtx->end.key == INT64_MIN || pCtx->end.key < pCtx->startTs) {
+        goto interp_exit;
+      }
+
+      COPY_TYPED_DATA(pCtx->pOutput, pCtx->inputType, &pCtx->end.val);
+      break;
+
+    case TSDB_FILL_NONE:
+      // do nothing
+    default:
+      goto interp_exit;
+  }
+
+
+  interp_success_exit:
+  *(TSKEY*)pCtx->ptsOutputBuf = pCtx->startTs;
+  INC_INIT_VAL(pCtx, 1);
+
+  interp_exit:
+  pCtx->start.key = INT64_MIN;
+  pCtx->end.key = INT64_MIN;
+  pCtx->endTs = pCtx->startTs;
+#endif
+
+  return TSDB_CODE_SUCCESS;
+}

source/libs/function/src/functionMgt.c

@@ -161,6 +161,8 @@ bool fmIsUserDefinedFunc(int32_t funcId) { return funcId > FUNC_UDF_ID_START; }
 
 bool fmIsForbidFillFunc(int32_t funcId) { return isSpecificClassifyFunc(funcId, FUNC_MGT_FORBID_FILL_FUNC); }
 
+bool fmIsIntervalInterpoFunc(int32_t funcId) { return isSpecificClassifyFunc(funcId, FUNC_MGT_INTERVAL_INTERPO_FUNC); }
+
 bool fmIsForbidStreamFunc(int32_t funcId) { return isSpecificClassifyFunc(funcId, FUNC_MGT_FORBID_STREAM_FUNC); }
 
 void fmFuncMgtDestroy() {

source/libs/function/src/taggfunction.c

@@ -161,13 +161,13 @@ typedef struct SRateInfo {
   bool    isIRate;    // true for IRate functions, false for Rate functions
 } SRateInfo;
 
-typedef struct SDerivInfo {
+//typedef struct SDerivInfo {
-  double   prevValue;     // previous value
+//  double   prevValue;     // previous value
-  TSKEY    prevTs;        // previous timestamp
+//  TSKEY    prevTs;        // previous timestamp
-  bool     ignoreNegative;// ignore the negative value
+//  bool     ignoreNegative;// ignore the negative value
-  int64_t  tsWindow;      // time window for derivative
+//  int64_t  tsWindow;      // time window for derivative
-  bool     valueSet;      // the value has been set already
+//  bool     valueSet;      // the value has been set already
-} SDerivInfo;
+//} SDerivInfo;
 
 typedef struct SResPair {
   TSKEY  key;

source/libs/function/src/texpr.c

@@ -48,34 +48,6 @@ static void doExprTreeDestroy(tExprNode **pExpr, void (*fp)(void *)) {
   *pExpr = NULL;
 }
 
-bool exprTreeApplyFilter(tExprNode *pExpr, const void *pItem, SExprTraverseSupp *param) {
-#if 0
-  //non-leaf nodes, recursively traverse the expression tree in the post-root order
-  if (pLeft->nodeType == TEXPR_BINARYEXPR_NODE && pRight->nodeType == TEXPR_BINARYEXPR_NODE) {
-    if (pExpr->_node.optr == LOGIC_COND_TYPE_OR) {  // or
-      if (exprTreeApplyFilter(pLeft, pItem, param)) {
-        return true;
-      }
-
-      // left child does not satisfy the query condition, try right child
-      return exprTreeApplyFilter(pRight, pItem, param);
-    } else {  // and
-      if (!exprTreeApplyFilter(pLeft, pItem, param)) {
-        return false;
-      }
-
-      return exprTreeApplyFilter(pRight, pItem, param);
-    }
-  }
-
-  // handle the leaf node
-  param->setupInfoFn(pExpr, param->pExtInfo);
-  return param->nodeFilterFn(pItem, pExpr->_node.info);
-#endif
-
-  return 0;
-}
-
 // TODO: these three functions should be made global
 static void* exception_calloc(size_t nmemb, size_t size) {
   void* p = taosMemoryCalloc(nmemb, size);
@@ -101,214 +73,3 @@ static UNUSED_FUNC char* exception_strdup(const char* str) {
   return p;
 }
 
-void buildFilterSetFromBinary(void **q, const char *buf, int32_t len) {
-  SBufferReader br = tbufInitReader(buf, len, false); 
-  uint32_t type  = tbufReadUint32(&br);     
-  SHashObj *pObj = taosHashInit(256, taosGetDefaultHashFunction(type), true, false);
-  
-//  taosHashSetEqualFp(pObj, taosGetDefaultEqualFunction(type));
-
-  int dummy = -1;
-  int32_t sz = tbufReadInt32(&br);
-  for (int32_t i = 0; i < sz; i++) {
-    if (type == TSDB_DATA_TYPE_BOOL || IS_SIGNED_NUMERIC_TYPE(type)) {
-      int64_t val = tbufReadInt64(&br); 
-      taosHashPut(pObj, (char *)&val, sizeof(val),  &dummy, sizeof(dummy));
-    } else if (IS_UNSIGNED_NUMERIC_TYPE(type)) {
-      uint64_t val = tbufReadUint64(&br); 
-      taosHashPut(pObj, (char *)&val, sizeof(val),  &dummy, sizeof(dummy));
-    }
-    else if (type == TSDB_DATA_TYPE_TIMESTAMP) {
-      int64_t val = tbufReadInt64(&br); 
-      taosHashPut(pObj, (char *)&val, sizeof(val),  &dummy, sizeof(dummy));
-    } else if (type == TSDB_DATA_TYPE_DOUBLE || type == TSDB_DATA_TYPE_FLOAT) {
-      double  val = tbufReadDouble(&br);
-      taosHashPut(pObj, (char *)&val, sizeof(val), &dummy, sizeof(dummy));
-    } else if (type == TSDB_DATA_TYPE_BINARY) {
-      size_t  t = 0;
-      const char *val = tbufReadBinary(&br, &t);
-      taosHashPut(pObj, (char *)val, t, &dummy, sizeof(dummy));
-    } else if (type == TSDB_DATA_TYPE_NCHAR) {
-      size_t  t = 0;
-      const char *val = tbufReadBinary(&br, &t);      
-      taosHashPut(pObj, (char *)val, t, &dummy, sizeof(dummy));
-    }
-  } 
-  *q = (void *)pObj;
-}
-
-void convertFilterSetFromBinary(void **q, const char *buf, int32_t len, uint32_t tType) {
-  SBufferReader br = tbufInitReader(buf, len, false); 
-  uint32_t sType  = tbufReadUint32(&br);     
-  SHashObj *pObj = taosHashInit(256, taosGetDefaultHashFunction(tType), true, false);
-  
-//  taosHashSetEqualFp(pObj, taosGetDefaultEqualFunction(tType));
-  
-  int dummy = -1;
-  SVariant tmpVar = {0};  
-  size_t  t = 0;
-  int32_t sz = tbufReadInt32(&br);
-  void *pvar = NULL;  
-  int64_t val = 0;
-  int32_t bufLen = 0;
-  if (IS_NUMERIC_TYPE(sType)) {
-    bufLen = 60;  // The maximum length of string that a number is converted to.
-  } else {
-    bufLen = 128;
-  }
-
-  char *tmp = taosMemoryCalloc(1, bufLen * TSDB_NCHAR_SIZE);
-    
-  for (int32_t i = 0; i < sz; i++) {
-    switch (sType) {
-    case TSDB_DATA_TYPE_BOOL:
-    case TSDB_DATA_TYPE_UTINYINT:
-    case TSDB_DATA_TYPE_TINYINT: {
-      *(uint8_t *)&val = (uint8_t)tbufReadInt64(&br); 
-      t = sizeof(val);
-      pvar = &val;
-      break;
-    }
-    case TSDB_DATA_TYPE_USMALLINT:
-    case TSDB_DATA_TYPE_SMALLINT: {
-      *(uint16_t *)&val = (uint16_t)tbufReadInt64(&br); 
-      t = sizeof(val);
-      pvar = &val;
-      break;
-    }
-    case TSDB_DATA_TYPE_UINT:
-    case TSDB_DATA_TYPE_INT: {
-      *(uint32_t *)&val = (uint32_t)tbufReadInt64(&br); 
-      t = sizeof(val);
-      pvar = &val;
-      break;
-    }
-    case TSDB_DATA_TYPE_TIMESTAMP:
-    case TSDB_DATA_TYPE_UBIGINT:
-    case TSDB_DATA_TYPE_BIGINT: {
-      *(uint64_t *)&val = (uint64_t)tbufReadInt64(&br); 
-      t = sizeof(val);
-      pvar = &val;
-      break;
-    }
-    case TSDB_DATA_TYPE_DOUBLE: {
-      *(double *)&val = tbufReadDouble(&br);
-      t = sizeof(val);
-      pvar = &val;
-      break;
-    }
-    case TSDB_DATA_TYPE_FLOAT: {
-      *(float *)&val = (float)tbufReadDouble(&br);
-      t = sizeof(val);
-      pvar = &val;
-      break;
-    }
-    case TSDB_DATA_TYPE_BINARY: {
-      pvar = (char *)tbufReadBinary(&br, &t);
-      break;
-    }
-    case TSDB_DATA_TYPE_NCHAR: {
-      pvar = (char *)tbufReadBinary(&br, &t);      
-      break;
-    }
-    default:
-      taosHashCleanup(pObj);
-      *q = NULL;
-      return;
-    }
-    
-    taosVariantCreateFromBinary(&tmpVar, (char *)pvar, t, sType);
-
-    if (bufLen < t) {
-      tmp = taosMemoryRealloc(tmp, t * TSDB_NCHAR_SIZE);
-      bufLen = (int32_t)t;
-    }
-
-    switch (tType) {
-      case TSDB_DATA_TYPE_BOOL:
-      case TSDB_DATA_TYPE_UTINYINT:
-      case TSDB_DATA_TYPE_TINYINT: {
-        if (taosVariantDump(&tmpVar, (char *)&val, tType, false)) {
-          goto err_ret;
-        }
-        pvar = &val;
-        t = sizeof(val);
-        break;
-      }
-      case TSDB_DATA_TYPE_USMALLINT:
-      case TSDB_DATA_TYPE_SMALLINT: {
-        if (taosVariantDump(&tmpVar, (char *)&val, tType, false)) {
-          goto err_ret;
-        }
-        pvar = &val;
-        t = sizeof(val);
-        break;
-      }
-      case TSDB_DATA_TYPE_UINT:
-      case TSDB_DATA_TYPE_INT: {
-        if (taosVariantDump(&tmpVar, (char *)&val, tType, false)) {
-          goto err_ret;
-        }
-        pvar = &val;
-        t = sizeof(val);
-        break;
-      }
-      case TSDB_DATA_TYPE_TIMESTAMP:
-      case TSDB_DATA_TYPE_UBIGINT:
-      case TSDB_DATA_TYPE_BIGINT: {
-        if (taosVariantDump(&tmpVar, (char *)&val, tType, false)) {
-          goto err_ret;
-        }
-        pvar = &val;
-        t = sizeof(val);
-        break;
-      }
-      case TSDB_DATA_TYPE_DOUBLE: {
-        if (taosVariantDump(&tmpVar, (char *)&val, tType, false)) {
-          goto err_ret;
-        }
-        pvar = &val;
-        t = sizeof(val);
-        break;
-      }
-      case TSDB_DATA_TYPE_FLOAT: {
-        if (taosVariantDump(&tmpVar, (char *)&val, tType, false)) {
-          goto err_ret;
-        }
-        pvar = &val;
-        t = sizeof(val);
-        break;
-      }
-      case TSDB_DATA_TYPE_BINARY: {
-        if (taosVariantDump(&tmpVar, tmp, tType, true)) {
-          goto err_ret;
-        }
-        t = varDataLen(tmp);
-        pvar = varDataVal(tmp);
-        break;
-      }
-      case TSDB_DATA_TYPE_NCHAR: {
-        if (taosVariantDump(&tmpVar, tmp, tType, true)) {
-          goto err_ret;
-        }
-        t = varDataLen(tmp);
-        pvar = varDataVal(tmp);        
-        break;
-      }
-      default:
-        goto err_ret;
-    }
-    
-    taosHashPut(pObj, (char *)pvar, t,  &dummy, sizeof(dummy));
-    taosVariantDestroy(&tmpVar);
-    memset(&tmpVar, 0, sizeof(tmpVar));
-  } 
-
-  *q = (void *)pObj;
-  pObj = NULL;
-  
-err_ret:  
-  taosVariantDestroy(&tmpVar);
-  taosHashCleanup(pObj);
-  taosMemoryFreeClear(tmp);
-}

source/libs/function/src/udfd.c

@@ -468,8 +468,8 @@ int32_t udfdConnectToMnode() {
   char pass[TSDB_PASSWORD_LEN + 1] = {0};
   taosEncryptPass_c((uint8_t *)(TSDB_DEFAULT_PASS), strlen(TSDB_DEFAULT_PASS), pass);
   tstrncpy(connReq.passwd, pass, sizeof(connReq.passwd));
-  connReq.pid = htonl(taosGetPId());
+  connReq.pid = taosGetPId();
-  connReq.startTime = htobe64(taosGetTimestampMs());
+  connReq.startTime = taosGetTimestampMs();
 
   int32_t contLen = tSerializeSConnectReq(NULL, 0, &connReq);
   void *  pReq = rpcMallocCont(contLen);

source/libs/nodes/src/nodesUtilFuncs.c

@@ -199,6 +199,7 @@ SNode* nodesMakeNode(ENodeType type) {
     case QUERY_NODE_SHOW_SUBSCRIBES_STMT:
     case QUERY_NODE_SHOW_SMAS_STMT:
     case QUERY_NODE_SHOW_CONFIGS_STMT:
+    case QUERY_NODE_SHOW_CONNECTIONS_STMT:
     case QUERY_NODE_SHOW_QUERIES_STMT:
     case QUERY_NODE_SHOW_VNODES_STMT:
     case QUERY_NODE_SHOW_APPS_STMT:

source/libs/parser/src/parAstParser.c

@@ -377,6 +377,16 @@ static int32_t collectMetaKeyFromShowTopics(SCollectMetaKeyCxt* pCxt, SShowStmt*
                                  pCxt->pMetaCache);
 }
 
+static int32_t collectMetaKeyFromShowConnections(SCollectMetaKeyCxt* pCxt, SShowStmt* pStmt) {
+  return reserveTableMetaInCache(pCxt->pParseCxt->acctId, TSDB_PERFORMANCE_SCHEMA_DB, TSDB_PERFS_TABLE_CONNECTIONS,
+                                 pCxt->pMetaCache);
+}
+
+static int32_t collectMetaKeyFromShowQueries(SCollectMetaKeyCxt* pCxt, SShowStmt* pStmt) {
+  return reserveTableMetaInCache(pCxt->pParseCxt->acctId, TSDB_PERFORMANCE_SCHEMA_DB, TSDB_PERFS_TABLE_QUERIES,
+                                 pCxt->pMetaCache);
+}
+
 static int32_t collectMetaKeyFromShowTransactions(SCollectMetaKeyCxt* pCxt, SShowStmt* pStmt) {
   return reserveTableMetaInCache(pCxt->pParseCxt->acctId, TSDB_PERFORMANCE_SCHEMA_DB, TSDB_PERFS_TABLE_TRANS,
                                  pCxt->pMetaCache);
@@ -447,6 +457,10 @@ static int32_t collectMetaKeyFromQuery(SCollectMetaKeyCxt* pCxt, SNode* pStmt) {
       return collectMetaKeyFromShowVgroups(pCxt, (SShowStmt*)pStmt);
     case QUERY_NODE_SHOW_TOPICS_STMT:
       return collectMetaKeyFromShowTopics(pCxt, (SShowStmt*)pStmt);
+    case QUERY_NODE_SHOW_CONNECTIONS_STMT:
+      return collectMetaKeyFromShowConnections(pCxt, (SShowStmt*)pStmt);
+    case QUERY_NODE_SHOW_QUERIES_STMT:
+      return collectMetaKeyFromShowQueries(pCxt, (SShowStmt*)pStmt);
     case QUERY_NODE_SHOW_TRANSACTIONS_STMT:
       return collectMetaKeyFromShowTransactions(pCxt, (SShowStmt*)pStmt);
     case QUERY_NODE_DELETE_STMT:

source/libs/parser/src/parTranslater.c

@@ -42,6 +42,7 @@ typedef struct STranslateContext {
   SExplainOptions* pExplainOpt;
   SParseMetaCache* pMetaCache;
   bool             createStream;
+  bool             stableQuery;
 } STranslateContext;
 
 typedef struct SFullDatabaseName {
@@ -1508,6 +1509,9 @@ static int32_t translateTable(STranslateContext* pCxt, SNode* pTable) {
       if (TSDB_CODE_SUCCESS == code) {
         code = addNamespace(pCxt, pRealTable);
       }
+      if (TSDB_SUPER_TABLE == pRealTable->pMeta->tableType) {
+        pCxt->stableQuery = true;
+      }
       break;
     }
     case QUERY_NODE_TEMP_TABLE: {
@@ -4805,6 +4809,7 @@ static int32_t buildDropTableVgroupHashmap(STranslateContext* pCxt, SDropTableCl
 
   if (TSDB_CODE_PAR_TABLE_NOT_EXIST == code && pClause->ignoreNotExists) {
     code = TSDB_CODE_SUCCESS;
+    goto over;
   }
 
   *pIsSuperTable = false;
@@ -4898,7 +4903,7 @@ static int32_t rewriteDropTable(STranslateContext* pCxt, SQuery* pQuery) {
     }
   }
 
-  if (isSuperTable) {
+  if (isSuperTable || 0 == taosHashGetSize(pVgroupHashmap)) {
     taosHashCleanup(pVgroupHashmap);
     return TSDB_CODE_SUCCESS;
   }
@@ -5343,6 +5348,8 @@ static int32_t setQuery(STranslateContext* pCxt, SQuery* pQuery) {
       break;
   }
 
+  pQuery->stableQuery = pCxt->stableQuery;
+
   if (pQuery->haveResultSet) {
     if (TSDB_CODE_SUCCESS != extractResultSchema(pQuery->pRoot, &pQuery->numOfResCols, &pQuery->pResSchema)) {
       return TSDB_CODE_OUT_OF_MEMORY;

source/libs/planner/src/planLogicCreater.c

@@ -498,7 +498,7 @@ static int32_t createWindowLogicNodeFinalize(SLogicPlanContext* pCxt, SSelectStm
   }
 
   if (pCxt->pPlanCxt->rSmaQuery) {
-    pWindow->filesFactor = pCxt->pPlanCxt->filesFactor;
+    /*pWindow->filesFactor = pCxt->pPlanCxt->filesFactor;*/
   }
 
   if (TSDB_CODE_SUCCESS == code) {

source/libs/planner/src/planPhysiCreater.c

@@ -470,8 +470,8 @@ static ENodeType getScanOperatorType(EScanType scanType) {
     case SCAN_TYPE_STREAM:
       return QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN;
     case SCAN_TYPE_TABLE_MERGE:
-      return QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN;
+      // return QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN;
-      // return QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN;
+      return QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN;
     default:
       break;
   }

source/libs/scheduler/src/scheduler.c

@@ -151,7 +151,9 @@ int32_t schedulerGetTasksStatus(int64_t job, SArray *pSub) {
 
     for (int32_t m = 0; m < pLevel->taskNum; ++m) {
       SSchTask     *pTask = taosArrayGet(pLevel->subTasks, m);
-      SQuerySubDesc subDesc = {.tid = pTask->taskId, .status = pTask->status};
+      SQuerySubDesc subDesc = {0};
+      subDesc.tid = pTask->taskId;
+      strcpy(subDesc.status, jobTaskStatusStr(pTask->status));
 
       taosArrayPush(pSub, &subDesc);
     }

source/libs/sync/src/syncMain.c

@@ -150,12 +150,14 @@ int32_t syncSetStandby(int64_t rid) {
   SSyncNode* pSyncNode = (SSyncNode*)taosAcquireRef(tsNodeRefId, rid);
   if (pSyncNode == NULL) {
     terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
+    sError("failed to set standby since accquire ref error, rid:%" PRId64, rid);
     return -1;
   }
 
   if (pSyncNode->state == TAOS_SYNC_STATE_LEADER) {
     taosReleaseRef(tsNodeRefId, pSyncNode->rid);
     terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
+    sError("failed to set standby since it is leader, rid:%" PRId64, rid);
     return -1;
   }
 

source/libs/wal/src/walRead.c

@@ -125,7 +125,7 @@ static int32_t walReadSeekVer(SWalReadHandle *pRead, int64_t ver) {
   }
   if (ver > pWal->vers.lastVer || ver < pWal->vers.firstVer) {
     wError("invalid version: % " PRId64 ", first ver %ld, last ver %ld", ver, pWal->vers.firstVer, pWal->vers.lastVer);
-    terrno = TSDB_CODE_WAL_INVALID_VER;
+    terrno = TSDB_CODE_WAL_LOG_NOT_EXIST;
     return -1;
   }
   if (ver < pWal->vers.snapshotVer) {

source/libs/wal/src/walWrite.c

@@ -64,7 +64,10 @@ int32_t walRestoreFromSnapshot(SWal *pWal, int64_t ver) {
 int32_t walCommit(SWal *pWal, int64_t ver) {
   ASSERT(pWal->vers.commitVer >= pWal->vers.snapshotVer);
   ASSERT(pWal->vers.commitVer <= pWal->vers.lastVer);
-  if (ver < pWal->vers.commitVer || ver > pWal->vers.lastVer) {
+  if (ver < pWal->vers.commitVer) {
+    return 0;
+  }
+  if (ver > pWal->vers.lastVer) {
     terrno = TSDB_CODE_WAL_INVALID_VER;
     return -1;
   }

source/util/src/tcache.c

@@ -53,19 +53,19 @@ typedef struct SCacheEntry {
   SCacheNode *next;
 } SCacheEntry;
 
-typedef struct STrashElem {
+struct STrashElem {
   struct STrashElem *prev;
   struct STrashElem *next;
   SCacheNode        *pData;
-} STrashElem;
+};
 
-typedef struct SCacheIter {
+struct SCacheIter {
   SCacheObj   *pCacheObj;
   SCacheNode **pCurrent;
   int32_t      entryIndex;
   int32_t      index;
   int32_t      numOfObj;
-} SCacheIter;
+};
 
 /*
  * to accommodate the old data which has the same key value of new one in hashList
@@ -931,15 +931,15 @@ bool taosCacheIterNext(SCacheIter *pIter) {
   SCacheObj *pCacheObj = pIter->pCacheObj;
 
   if (pIter->index + 1 >= pIter->numOfObj) {
-    if (pIter->entryIndex + 1 >= pCacheObj->capacity) {
-      return false;
-    }
-
     // release the reference for all objects in the snapshot
     for (int32_t i = 0; i < pIter->numOfObj; ++i) {
       char *p = pIter->pCurrent[i]->data;
       taosCacheRelease(pCacheObj, (void **)&p, false);
       pIter->pCurrent[i] = NULL;
+    }   
+
+    if (pIter->entryIndex + 1 >= pCacheObj->capacity) {
+      return false;
     }
 
     while (1) {

source/util/src/terror.c

@@ -424,6 +424,7 @@ TAOS_DEFINE_ERROR(TSDB_CODE_WAL_APP_ERROR,                "Unexpected generic er
 TAOS_DEFINE_ERROR(TSDB_CODE_WAL_FILE_CORRUPTED,           "WAL file is corrupted")
 TAOS_DEFINE_ERROR(TSDB_CODE_WAL_SIZE_LIMIT,               "WAL size exceeds limit")
 TAOS_DEFINE_ERROR(TSDB_CODE_WAL_INVALID_VER,              "WAL use invalid version")
+TAOS_DEFINE_ERROR(TSDB_CODE_WAL_LOG_NOT_EXIST,            "WAL log not exist")
 
 // tfs
 TAOS_DEFINE_ERROR(TSDB_CODE_FS_APP_ERROR,                 "tfs out of memory")

tests/pytest/util/dnodes.py

@@ -334,22 +334,22 @@ class TDDnode:
                 bkey2 = bytes(key2, encoding="utf8")
                 logFile = self.logDir + "/taosdlog.0"
                 i = 0
-                while not os.path.exists(logFile):
+                # while not os.path.exists(logFile):
-                    sleep(0.1)
+                #     sleep(0.1)
-                    i += 1
+                #     i += 1
-                    if i > 10:
+                #     if i > 10:
-                        break
+                #         break
-                tailCmdStr = 'tail -f '
+                # tailCmdStr = 'tail -f '
-                if platform.system().lower() == 'windows':
+                # if platform.system().lower() == 'windows':
-                    tailCmdStr = 'tail -n +0 -f '
+                #     tailCmdStr = 'tail -n +0 -f '
-                popen = subprocess.Popen(
+                # popen = subprocess.Popen(
-                    tailCmdStr + logFile,
+                #     tailCmdStr + logFile,
-                    stdout=subprocess.PIPE,
+                #     stdout=subprocess.PIPE,
-                    stderr=subprocess.PIPE,
+                #     stderr=subprocess.PIPE,
-                    shell=True)
+                #     shell=True)
-                pid = popen.pid
+                # pid = popen.pid
-                # print('Popen.pid:' + str(pid))
+                # # print('Popen.pid:' + str(pid))
-                timeout = time.time() + 60 * 2
+                # timeout = time.time() + 60 * 2
                 # while True:
                 #     line = popen.stdout.readline().strip()
                 #     print(line)

tests/pytest/util/sql.py

@@ -156,6 +156,7 @@ class TDSql:
     def checkRows(self, expectRows):
         if self.queryRows == expectRows:
             tdLog.info("sql:%s, queryRows:%d == expect:%d" % (self.sql, self.queryRows, expectRows))
+            return True
         else:
             caller = inspect.getframeinfo(inspect.stack()[1][0])
             args = (caller.filename, caller.lineno, self.sql, self.queryRows, expectRows)

tests/system-test/2-query/json_tag.py

@@ -543,7 +543,7 @@ class TDTestCase:
         tdSql.checkData(0, 0, 1.5)
         #tdSql.query("select last_row(dataint) from jsons1 where jtag->'tag1'>1")
         #tdSql.checkData(0, 0, 11)
-        tdSql.error("select interp(dataint) from jsons1 where ts = '2020-06-02 09:17:08.000' and jtag->'tag1'>1")
+        #tdSql.error("select interp(dataint) from jsons1 where ts = '2020-06-02 09:17:08.000' and jtag->'tag1'>1")
         #
         # #test calculation function:diff/derivative/spread/ceil/floor/round/
         #tdSql.error("select diff(dataint) from jsons1 where jtag->'tag1'>1")

tests/system-test/6-cluster/5dnode2mnode.py

@@ -54,7 +54,9 @@ class TDTestCase:
         valgrind = 0  
         hostname = socket.gethostname()
         dnodes = []
-        start_port = 6030
+        start_port = 6030        
+        start_port_sec = 6130
+
         for num in range(1, dnodes_nums+1):
             dnode = TDDnode(num)
             dnode.addExtraCfg("firstEp", f"{hostname}:{start_port}")
@@ -62,6 +64,8 @@ class TDTestCase:
             dnode.addExtraCfg("serverPort", f"{start_port + (num-1)*100}")
             dnode.addExtraCfg("monitorFqdn", hostname)
             dnode.addExtraCfg("monitorPort", 7043)
+            dnode.addExtraCfg("secondEp", f"{hostname}:{start_port_sec}")
+ 
             dnodes.append(dnode)
         
         self.TDDnodes = MyDnodes(dnodes)

tests/system-test/6-cluster/5dnode3mnodeDrop.py

@@ -78,6 +78,7 @@ class TDTestCase:
         print(hostname)
         dnodes = []
         start_port = 6030
+        start_port_sec = 6130
         for num in range(1, dnodes_nums+1):
             dnode = TDDnode(num)
             dnode.addExtraCfg("firstEp", f"{hostname}:{start_port}")
@@ -85,6 +86,7 @@ class TDTestCase:
             dnode.addExtraCfg("serverPort", f"{start_port + (num-1)*100}")
             dnode.addExtraCfg("monitorFqdn", hostname)
             dnode.addExtraCfg("monitorPort", 7043)
+            dnode.addExtraCfg("secondEp", f"{hostname}:{start_port_sec}")
             dnodes.append(dnode)
         
         self.TDDnodes = MyDnodes(dnodes)
@@ -116,25 +118,28 @@ class TDTestCase:
         while count < 10:
             time.sleep(1)
             tdSql.query("show mnodes;")
-            if tdSql.checkRows(3) :                
+            if tdSql.checkRows(3) :     
-                if  tdSql.queryResult[0][2]=='leader' :
+                print("mnode is  three nodes")           
-                    if  tdSql.queryResult[1][2]=='follower':
+            if  tdSql.queryResult[0][2]=='leader' :
-                        if  tdSql.queryResult[2][2]=='follower':
+                if  tdSql.queryResult[1][2]=='follower':
-                            print("three mnodes is ready in 10s")
+                    if  tdSql.queryResult[2][2]=='follower':
-                            break
+                        print("three mnodes is ready in 10s")
-                elif tdSql.queryResult[0][2]=='follower' :
+                        break
-                    if  tdSql.queryResult[1][2]=='leader':
+            elif tdSql.queryResult[0][2]=='follower' :
-                        if  tdSql.queryResult[2][2]=='follower':
+                if  tdSql.queryResult[1][2]=='leader':
-                            print("three mnodes is ready in 10s")
+                    if  tdSql.queryResult[2][2]=='follower':
-                            break      
+                        print("three mnodes is ready in 10s")
-                elif tdSql.queryResult[0][2]=='follower' :
+                        break      
-                    if  tdSql.queryResult[1][2]=='follower':
+            elif tdSql.queryResult[0][2]=='follower' :
-                        if  tdSql.queryResult[2][2]=='leader':
+                if  tdSql.queryResult[1][2]=='follower':
-                            print("three mnodes is ready in 10s")
+                    if  tdSql.queryResult[2][2]=='leader':
-                            break                   
+                        print("three mnodes is ready in 10s")
+                        break                   
             count+=1
         else:
+            print(tdSql.queryResult)
             print("three mnodes is not ready in 10s ")
+            return -1
 
         tdSql.query("show mnodes;")       
         tdSql.checkRows(3) 
@@ -151,18 +156,21 @@ class TDTestCase:
             time.sleep(1)
             tdSql.query("show mnodes;")
             if tdSql.checkRows(3) :
-                if  tdSql.queryResult[0][2]=='offline' :
+                print("mnode is  three nodes")
-                    if  tdSql.queryResult[1][2]=='leader':
+            if  tdSql.queryResult[0][2]=='offline' :
-                        if  tdSql.queryResult[2][2]=='follower':
+                if  tdSql.queryResult[1][2]=='leader':
-                            print("stop mnodes  on dnode 2 successfully in 10s")
+                    if  tdSql.queryResult[2][2]=='follower':
-                            break
+                        print("stop mnodes  on dnode 2 successfully in 10s")
-                    elif tdSql.queryResult[1][2]=='follower':
+                        break
-                        if  tdSql.queryResult[2][2]=='leader':
+                elif tdSql.queryResult[1][2]=='follower':
-                            print("stop mnodes  on dnode 2 successfully in 10s")
+                    if  tdSql.queryResult[2][2]=='leader':
-                            break
+                        print("stop mnodes  on dnode 2 successfully in 10s")
+                        break
             count+=1
         else:
             print("stop mnodes  on dnode 2 failed in 10s ")
+            return -1 
+        tdSql.error("drop mnode on dnode 1;")
 
         tdSql.query("show mnodes;")       
         tdSql.checkRows(3) 
@@ -174,20 +182,23 @@ class TDTestCase:
         tdSql.checkData(2,1,'%s:6230'%self.host)
         tdSql.checkData(2,3,'ready')
 
-    def check3mnode2drop(self):
+    def check3mnode2off(self):
         count=0
-        while count < 10:
+        while count < 40:
             time.sleep(1)
             tdSql.query("show mnodes;")
             if tdSql.checkRows(3) :
-                if  tdSql.queryResult[0][2]=='leader' :
+                print("mnode is  three nodes")
-                    if  tdSql.queryResult[1][2]=='offline':
+            if  tdSql.queryResult[0][2]=='leader' :
-                        if  tdSql.queryResult[2][2]=='follower':
+                if  tdSql.queryResult[1][2]=='offline':
-                            print("stop mnodes  on dnode 2 successfully in 10s")
+                    if  tdSql.queryResult[2][2]=='follower':
-                            break
+                        print("stop mnodes  on dnode 2 successfully in 10s")
+                        break
             count+=1
         else:
             print("stop mnodes  on dnode 2 failed in 10s ")
+            return -1
+        tdSql.error("drop mnode on dnode 2;")
 
         tdSql.query("show mnodes;")       
         tdSql.checkRows(3) 
@@ -197,7 +208,7 @@ class TDTestCase:
         tdSql.checkData(1,1,'%s:6130'%self.host)
         tdSql.checkData(1,2,'offline')
         tdSql.checkData(1,3,'ready')
-        tdSql.checkData(2,1,'%s:6230')
+        tdSql.checkData(2,1,'%s:6230'%self.host)
         tdSql.checkData(2,2,'follower')
         tdSql.checkData(2,3,'ready')
 
@@ -207,15 +218,17 @@ class TDTestCase:
             time.sleep(1)
             tdSql.query("show mnodes;")
             if tdSql.checkRows(3) :
-                if  tdSql.queryResult[0][2]=='leader' :
+                print("mnode is  three nodes")
-                    if  tdSql.queryResult[2][2]=='offline':
+            if  tdSql.queryResult[0][2]=='leader' :
-                        if  tdSql.queryResult[1][2]=='follower':
+                if  tdSql.queryResult[2][2]=='offline':
-                            print("stop mnodes  on dnode 3 successfully in 10s")
+                    if  tdSql.queryResult[1][2]=='follower':
-                            break
+                        print("stop mnodes  on dnode 3 successfully in 10s")
+                        break
             count+=1
         else:
             print("stop mnodes  on dnode 3 failed in 10s")
-            
+            return -1
+        tdSql.error("drop mnode on dnode 3;")
         tdSql.query("show mnodes;")       
         tdSql.checkRows(3) 
         tdSql.checkData(0,1,'%s:6030'%self.host)
@@ -262,8 +275,25 @@ class TDTestCase:
             for i in range(1,3):
                 tdLog.debug("drop mnode on dnode %d"%(i+1))
                 tdSql.execute("drop mnode on dnode %d"%(i+1))
+                tdSql.query("show mnodes;")
+                count=0
+                while count<10:
+                    time.sleep(1)
+                    tdSql.query("show mnodes;")
+                    if tdSql.checkRows(2):
+                        print("drop mnode %d successfully"%(i+1))
+                        break
+                    count+=1
                 tdLog.debug("create mnode on dnode %d"%(i+1))
                 tdSql.execute("create mnode on dnode %d"%(i+1))
+                count=0
+                while count<10:
+                    time.sleep(1)
+                    tdSql.query("show mnodes;")
+                    if tdSql.checkRows(3):
+                        print("drop mnode %d successfully"%(i+1))
+                        break
+                    count+=1
             dropcount+=1
         self.check3mnode()
 

tests/system-test/6-cluster/5dnode3mnodeStop.py

@@ -115,25 +115,27 @@ class TDTestCase:
         while count < 10:
             time.sleep(1)
             tdSql.query("show mnodes;")
-            if tdSql.checkRows(3) :                
+            if tdSql.checkRows(3) :     
-                if  tdSql.queryResult[0][2]=='leader' :
+                print("mnode is  three nodes")           
-                    if  tdSql.queryResult[1][2]=='follower':
+            if  tdSql.queryResult[0][2]=='leader' :
-                        if  tdSql.queryResult[2][2]=='follower':
+                if  tdSql.queryResult[1][2]=='follower':
-                            print("three mnodes is ready in 10s")
+                    if  tdSql.queryResult[2][2]=='follower':
-                            break
+                        print("three mnodes is ready in 10s")
-                elif tdSql.queryResult[0][2]=='follower' :
+                        break
-                    if  tdSql.queryResult[1][2]=='leader':
+            elif tdSql.queryResult[0][2]=='follower' :
-                        if  tdSql.queryResult[2][2]=='follower':
+                if  tdSql.queryResult[1][2]=='leader':
-                            print("three mnodes is ready in 10s")
+                    if  tdSql.queryResult[2][2]=='follower':
-                            break      
+                        print("three mnodes is ready in 10s")
-                elif tdSql.queryResult[0][2]=='follower' :
+                        break      
-                    if  tdSql.queryResult[1][2]=='follower':
+            elif tdSql.queryResult[0][2]=='follower' :
-                        if  tdSql.queryResult[2][2]=='leader':
+                if  tdSql.queryResult[1][2]=='follower':
-                            print("three mnodes is ready in 10s")
+                    if  tdSql.queryResult[2][2]=='leader':
-                            break                   
+                        print("three mnodes is ready in 10s")
+                        break                   
             count+=1
         else:
             print("three mnodes is not ready in 10s ")
+            return -1
 
         tdSql.query("show mnodes;")       
         tdSql.checkRows(3) 
@@ -145,24 +147,26 @@ class TDTestCase:
         tdSql.checkData(2,3,'ready')
 
     def check3mnode1off(self):
-        tdSql.error("drop mnode on dnode 1;")
         count=0
         while count < 10:
             time.sleep(1)
             tdSql.query("show mnodes;")
             if tdSql.checkRows(3) :
-                if  tdSql.queryResult[0][2]=='offline' :
+                print("mnode is  three nodes")
-                    if  tdSql.queryResult[1][2]=='leader':
+            if  tdSql.queryResult[0][2]=='offline' :
-                        if  tdSql.queryResult[2][2]=='follower':
+                if  tdSql.queryResult[1][2]=='leader':
-                            print("stop mnodes  on dnode 2 successfully in 10s")
+                    if  tdSql.queryResult[2][2]=='follower':
-                            break
+                        print("stop mnodes  on dnode 2 successfully in 10s")
-                    elif tdSql.queryResult[1][2]=='follower':
+                        break
-                        if  tdSql.queryResult[2][2]=='leader':
+                elif tdSql.queryResult[1][2]=='follower':
-                            print("stop mnodes  on dnode 2 successfully in 10s")
+                    if  tdSql.queryResult[2][2]=='leader':
-                            break
+                        print("stop mnodes  on dnode 2 successfully in 10s")
+                        break
             count+=1
         else:
             print("stop mnodes  on dnode 2 failed in 10s ")
+            return -1
+        tdSql.error("drop mnode on dnode 1;")
 
         tdSql.query("show mnodes;")       
         tdSql.checkRows(3) 
@@ -175,20 +179,22 @@ class TDTestCase:
         tdSql.checkData(2,3,'ready')
 
     def check3mnode2off(self):
-        tdSql.error("drop mnode on dnode 2;")
         count=0
-        while count < 10:
+        while count < 40:
             time.sleep(1)
             tdSql.query("show mnodes;")
             if tdSql.checkRows(3) :
-                if  tdSql.queryResult[0][2]=='leader' :
+                print("mnode is  three nodes")
-                    if  tdSql.queryResult[1][2]=='offline':
+            if  tdSql.queryResult[0][2]=='leader' :
-                        if  tdSql.queryResult[2][2]=='follower':
+                if  tdSql.queryResult[1][2]=='offline':
-                            print("stop mnodes  on dnode 2 successfully in 10s")
+                    if  tdSql.queryResult[2][2]=='follower':
-                            break
+                        print("stop mnodes  on dnode 2 successfully in 10s")
+                        break
             count+=1
         else:
             print("stop mnodes  on dnode 2 failed in 10s ")
+            return -1
+        tdSql.error("drop mnode on dnode 2;")
 
         tdSql.query("show mnodes;")       
         tdSql.checkRows(3) 
@@ -203,21 +209,22 @@ class TDTestCase:
         tdSql.checkData(2,3,'ready')
 
     def check3mnode3off(self):
-        tdSql.error("drop mnode on dnode 3;")
         count=0
         while count < 10:
             time.sleep(1)
             tdSql.query("show mnodes;")
             if tdSql.checkRows(3) :
-                if  tdSql.queryResult[0][2]=='leader' :
+                print("mnode is  three nodes")
-                    if  tdSql.queryResult[2][2]=='offline':
+            if  tdSql.queryResult[0][2]=='leader' :
-                        if  tdSql.queryResult[1][2]=='follower':
+                if  tdSql.queryResult[2][2]=='offline':
-                            print("stop mnodes  on dnode 3 successfully in 10s")
+                    if  tdSql.queryResult[1][2]=='follower':
-                            break
+                        print("stop mnodes  on dnode 3 successfully in 10s")
+                        break
             count+=1
         else:
             print("stop mnodes  on dnode 3 failed in 10s")
-            
+            return -1
+        tdSql.error("drop mnode on dnode 3;")
         tdSql.query("show mnodes;")       
         tdSql.checkRows(3) 
         tdSql.checkData(0,1,'%s:6030'%self.host)
@@ -264,13 +271,13 @@ class TDTestCase:
 
         self.TDDnodes.stoptaosd(3)
         self.check3mnode3off()
-        self.TDDnodes.starttaosd(2)
+        self.TDDnodes.starttaosd(3)
 
         self.TDDnodes.stoptaosd(1)
         self.check3mnode1off()
         self.TDDnodes.starttaosd(1)
 
-        # self.check3mnode()
+        self.check3mnode()
         stopcount =0 
         while stopcount <= 2:
             for i in range(dnodenumber):

tests/system-test/6-cluster/5dnode3mnodeStopInsert.py

@@ -13,13 +13,17 @@ import time
 import socket
 import subprocess
 from multiprocessing import Process
-import threading as thd
+import threading 
+import time
+import inspect
+import ctypes
 class MyDnodes(TDDnodes):
     def __init__(self ,dnodes_lists):
         super(MyDnodes,self).__init__()
         self.dnodes = dnodes_lists  # dnode must be TDDnode instance
         self.simDeployed = False
-        
+
+
 class TDTestCase:
 
     def init(self,conn ,logSql):
@@ -49,11 +53,31 @@ class TDTestCase:
                     buildPath = root[:len(root) - len("/build/bin")]
                     break
         return buildPath
-    
+
+    def _async_raise(self, tid, exctype):
+        """raises the exception, performs cleanup if needed"""
+        if not inspect.isclass(exctype):
+            exctype = type(exctype)
+        res = ctypes.pythonapi.PyThreadState_SetAsyncExc(tid, ctypes.py_object(exctype))
+        if res == 0:
+            raise ValueError("invalid thread id")
+        elif res != 1:
+            # """if it returns a number greater than one, you're in trouble, 
+            # and you should call it again with exc=NULL to revert the effect"""
+            ctypes.pythonapi.PyThreadState_SetAsyncExc(tid, None)
+            raise SystemError("PyThreadState_SetAsyncExc failed")
+
+    def stop_thread(self,thread):
+        self._async_raise(thread.ident, SystemExit)
+
+
     def insert_data(self,countstart,countstop):
         # fisrt add data : db\stable\childtable\general table
+        
         for couti in range(countstart,countstop):
+            tdLog.debug("drop database if exists db%d" %couti)
             tdSql.execute("drop database if exists db%d" %couti)
+            print("create database if not exists db%d replica 1 duration 300" %couti)
             tdSql.execute("create database if not exists db%d replica 1 duration 300" %couti)
             tdSql.execute("use db%d" %couti)
             tdSql.execute(
@@ -78,6 +102,7 @@ class TDTestCase:
         hostname = socket.gethostname()
         dnodes = []
         start_port = 6030
+        start_port_sec = 6130
         for num in range(1, dnodes_nums+1):
             dnode = TDDnode(num)
             dnode.addExtraCfg("firstEp", f"{hostname}:{start_port}")
@@ -85,6 +110,7 @@ class TDTestCase:
             dnode.addExtraCfg("serverPort", f"{start_port + (num-1)*100}")
             dnode.addExtraCfg("monitorFqdn", hostname)
             dnode.addExtraCfg("monitorPort", 7043)
+            dnode.addExtraCfg("secondEp", f"{hostname}:{start_port_sec}")
             dnodes.append(dnode)
         
         self.TDDnodes = MyDnodes(dnodes)
@@ -111,30 +137,58 @@ class TDTestCase:
         time.sleep(2)
         tdLog.info(" create cluster with %d dnode  done! " %dnodes_nums)
 
+    def checkdnodes(self,dnodenumber):
+        count=0
+        while count < 10:
+            time.sleep(1)
+            statusReadyBumber=0
+            tdSql.query("show dnodes;")
+            if tdSql.checkRows(dnodenumber) :     
+                print("dnode is %d nodes"%dnodenumber)   
+            for i in range(dnodenumber):
+                if tdSql.queryResult[i][4] !='ready'  :
+                    status=tdSql.queryResult[i][4]
+                    print("dnode:%d status is %s "%(i,status))
+                    break
+                else:
+                    statusReadyBumber+=1
+            print(statusReadyBumber)
+            if statusReadyBumber == dnodenumber :
+                print("all of %d mnodes is ready in 10s "%dnodenumber)
+                return True
+                break
+            count+=1
+        else:
+            print("%d mnodes is not ready in 10s "%dnodenumber)
+            return False
+           
+
     def check3mnode(self):
         count=0
         while count < 10:
             time.sleep(1)
             tdSql.query("show mnodes;")
-            if tdSql.checkRows(3) :                
+            if tdSql.checkRows(3) :     
-                if  tdSql.queryResult[0][2]=='leader' :
+                print("mnode is  three nodes")           
-                    if  tdSql.queryResult[1][2]=='follower':
+            if  tdSql.queryResult[0][2]=='leader' :
-                        if  tdSql.queryResult[2][2]=='follower':
+                if  tdSql.queryResult[1][2]=='follower':
-                            print("three mnodes is ready in 10s")
+                    if  tdSql.queryResult[2][2]=='follower':
-                            break
+                        print("three mnodes is ready in 10s")
-                elif tdSql.queryResult[0][2]=='follower' :
+                        break
-                    if  tdSql.queryResult[1][2]=='leader':
+            elif tdSql.queryResult[0][2]=='follower' :
-                        if  tdSql.queryResult[2][2]=='follower':
+                if  tdSql.queryResult[1][2]=='leader':
-                            print("three mnodes is ready in 10s")
+                    if  tdSql.queryResult[2][2]=='follower':
-                            break      
+                        print("three mnodes is ready in 10s")
-                elif tdSql.queryResult[0][2]=='follower' :
+                        break      
-                    if  tdSql.queryResult[1][2]=='follower':
+            elif tdSql.queryResult[0][2]=='follower' :
-                        if  tdSql.queryResult[2][2]=='leader':
+                if  tdSql.queryResult[1][2]=='follower':
-                            print("three mnodes is ready in 10s")
+                    if  tdSql.queryResult[2][2]=='leader':
-                            break                   
+                        print("three mnodes is ready in 10s")
+                        break                   
             count+=1
         else:
             print("three mnodes is not ready in 10s ")
+            return -1
 
         tdSql.query("show mnodes;")       
         tdSql.checkRows(3) 
@@ -146,24 +200,26 @@ class TDTestCase:
         tdSql.checkData(2,3,'ready')
 
     def check3mnode1off(self):
-        tdSql.error("drop mnode on dnode 1;")
         count=0
         while count < 10:
             time.sleep(1)
             tdSql.query("show mnodes;")
             if tdSql.checkRows(3) :
-                if  tdSql.queryResult[0][2]=='offline' :
+                print("mnode is  three nodes")
-                    if  tdSql.queryResult[1][2]=='leader':
+            if  tdSql.queryResult[0][2]=='offline' :
-                        if  tdSql.queryResult[2][2]=='follower':
+                if  tdSql.queryResult[1][2]=='leader':
-                            print("stop mnodes  on dnode 2 successfully in 10s")
+                    if  tdSql.queryResult[2][2]=='follower':
-                            break
+                        print("stop mnodes  on dnode 2 successfully in 10s")
-                    elif tdSql.queryResult[1][2]=='follower':
+                        break
-                        if  tdSql.queryResult[2][2]=='leader':
+                elif tdSql.queryResult[1][2]=='follower':
-                            print("stop mnodes  on dnode 2 successfully in 10s")
+                    if  tdSql.queryResult[2][2]=='leader':
-                            break
+                        print("stop mnodes  on dnode 2 successfully in 10s")
+                        break
             count+=1
         else:
             print("stop mnodes  on dnode 2 failed in 10s ")
+            return -1
+        tdSql.error("drop mnode on dnode 1;")
 
         tdSql.query("show mnodes;")       
         tdSql.checkRows(3) 
@@ -176,20 +232,22 @@ class TDTestCase:
         tdSql.checkData(2,3,'ready')
 
     def check3mnode2off(self):
-        tdSql.error("drop mnode on dnode 2;")
         count=0
         while count < 40:
             time.sleep(1)
             tdSql.query("show mnodes;")
             if tdSql.checkRows(3) :
-                if  tdSql.queryResult[0][2]=='leader' :
+                print("mnode is  three nodes")
-                    if  tdSql.queryResult[1][2]=='offline':
+            if  tdSql.queryResult[0][2]=='leader' :
-                        if  tdSql.queryResult[2][2]=='follower':
+                if  tdSql.queryResult[1][2]=='offline':
-                            print("stop mnodes  on dnode 2 successfully in 10s")
+                    if  tdSql.queryResult[2][2]=='follower':
-                            break
+                        print("stop mnodes  on dnode 2 successfully in 10s")
+                        break
             count+=1
         else:
             print("stop mnodes  on dnode 2 failed in 10s ")
+            return -1
+        tdSql.error("drop mnode on dnode 2;")
 
         tdSql.query("show mnodes;")       
         tdSql.checkRows(3) 
@@ -204,21 +262,22 @@ class TDTestCase:
         tdSql.checkData(2,3,'ready')
 
     def check3mnode3off(self):
-        tdSql.error("drop mnode on dnode 3;")
         count=0
         while count < 10:
             time.sleep(1)
             tdSql.query("show mnodes;")
             if tdSql.checkRows(3) :
-                if  tdSql.queryResult[0][2]=='leader' :
+                print("mnode is  three nodes")
-                    if  tdSql.queryResult[2][2]=='offline':
+            if  tdSql.queryResult[0][2]=='leader' :
-                        if  tdSql.queryResult[1][2]=='follower':
+                if  tdSql.queryResult[2][2]=='offline':
-                            print("stop mnodes  on dnode 3 successfully in 10s")
+                    if  tdSql.queryResult[1][2]=='follower':
-                            break
+                        print("stop mnodes  on dnode 3 successfully in 10s")
+                        break
             count+=1
         else:
             print("stop mnodes  on dnode 3 failed in 10s")
-            
+            return -1
+        tdSql.error("drop mnode on dnode 3;")
         tdSql.query("show mnodes;")       
         tdSql.checkRows(3) 
         tdSql.checkData(0,1,'%s:6030'%self.host)
@@ -231,8 +290,6 @@ class TDTestCase:
         tdSql.checkData(2,2,'offline')
         tdSql.checkData(2,3,'ready')
 
-
-
     def five_dnode_three_mnode(self,dnodenumber):
         tdSql.query("show dnodes;")
         tdSql.checkData(0,1,'%s:6030'%self.host)
@@ -252,43 +309,29 @@ class TDTestCase:
         # fisrt check statut ready
         self.check3mnode()
 
-
         tdSql.error("create mnode on dnode 2")
-
         tdSql.query("show dnodes;")
         print(tdSql.queryResult)
+        tdLog.debug("stop all of mnode ") 
 
-        tdLog.debug("stop and follower of mnode") 
-        self.TDDnodes.stoptaosd(2)
-        self.check3mnode2off()
-        self.TDDnodes.starttaosd(2)
-
-        self.TDDnodes.stoptaosd(3)
-        self.check3mnode3off()
-        self.TDDnodes.starttaosd(3)
-
-        self.TDDnodes.stoptaosd(1)
-        self.check3mnode1off()
-        self.TDDnodes.starttaosd(1)
-
-        # self.check3mnode()
         stopcount =0 
         while stopcount <= 2:
             for i in range(dnodenumber):
-                threads = []
+                # threads=[]
-                threads.append(thd.Thread(target=self.insert_data, args=(i*2,i*2+2))) 
+                # threads = MyThreadFunc(self.insert_data(i*2,i*2+2)) 
-                threads[0].start()
+                threads=threading.Thread(target=self.insert_data, args=((stopcount+i)*2,(i+stopcount)*2+2))
+                threads.start()
                 self.TDDnodes.stoptaosd(i+1)
-                # if i == 1 :
-                #     self.check3mnode2off()
-                # elif i == 2 :
-                #     self.check3mnode3off()
-                # elif i == 0:
-                #     self.check3mnode1off()
-
                 self.TDDnodes.starttaosd(i+1)
-                threads[0].join()
 
+                if self.checkdnodes(5):
+                    print("123")
+                    threads.join()
+                else:
+                    print("456")
+                    self.stop_thread(threads)
+                    assert 1 == 2 ,"some dnode started failed"
+                    return False
                 # self.check3mnode()
             stopcount+=1
         self.check3mnode()

tests/system-test/7-tmq/tmqError.py

@@ -59,8 +59,8 @@ class TDTestCase:
     def initConsumerTable(self,cdbName='cdb'):        
         tdLog.info("create consume database, and consume info table, and consume result table")
         tdSql.query("create database if not exists %s vgroups 1"%(cdbName))
-        # tdSql.query("drop table if exists %s.consumeinfo "%(cdbName))
+        tdSql.query("drop table if exists %s.consumeinfo "%(cdbName))
-        # tdSql.query("drop table if exists %s.consumeresult "%(cdbName))
+        tdSql.query("drop table if exists %s.consumeresult "%(cdbName))
 
         tdSql.query("create table %s.consumeinfo (ts timestamp, consumerid int, topiclist binary(1024), keylist binary(1024), expectmsgcnt bigint, ifcheckdata int, ifmanualcommit int)"%cdbName)
         tdSql.query("create table %s.consumeresult (ts timestamp, consumerid int, consummsgcnt bigint, consumrowcnt bigint, checkresult int)"%cdbName)

tests/system-test/fulltest.sh

@@ -101,7 +101,7 @@ python3 ./test.py -f 2-query/tail.py
 
 python3 ./test.py -f 6-cluster/5dnode1mnode.py 
 python3 ./test.py -f 6-cluster/5dnode2mnode.py 
-# BUG python3 ./test.py -f 6-cluster/5dnode3mnodeStop.py
+python3 ./test.py -f 6-cluster/5dnode3mnodeStop.py
 python3 ./test.py -f 6-cluster/5dnode3mnodeDrop.py
 # BUG python3 ./test.py -f 6-cluster/5dnode3mnodeStopInsert.py 
 

tools/taosadapter

@@ -1 +0,0 @@
-Subproject commit 9ce3f5c98ef95d9c7c596c4ed7302b0ed69a92b2