Merge pull request #16959 from taosdata/FIX/TD-19011-2.4

fix: synchronize initialization, access, and destruction of subqueries and their states

src/client/inc/tscSubquery.h

@@ -50,7 +50,7 @@ void tscUnlockByThread(int64_t *lockedBy);
 
 int tsInsertInitialCheck(SSqlObj *pSql);
 
-void doCleanupSubqueries(SSqlObj *pSql, int32_t numOfSubs);
+void doCleanupSubqueries(SSqlObj *pSql);
 
 void tscFreeRetrieveSup(void **param);
 

src/client/inc/tscUtil.h

@@ -342,7 +342,7 @@ void doAddGroupColumnForSubquery(SQueryInfo* pQueryInfo, int32_t tagIndex, SSqlC
 
 int16_t tscGetJoinTagColIdByUid(STagCond* pTagCond, uint64_t uid);
 int16_t tscGetTagColIndexById(STableMeta* pTableMeta, int16_t colId);
-int32_t doInitSubState(SSqlObj* pSql, int32_t numOfSubqueries);
+int32_t doReInitSubState(SSqlObj* pSql, int32_t numOfSubqueries);
 
 void tscPrintSelNodeList(SSqlObj* pSql, int32_t subClauseIndex);
 

src/client/inc/tsclient.h

@@ -359,6 +359,7 @@ typedef struct SSubqueryState {
   int8_t  *states;
   int32_t  numOfSub;            // the number of total sub-queries
   uint64_t numOfRetrievedRows;  // total number of points in this query
+  uint32_t version;
 } SSubqueryState;
 
 typedef struct SSqlObj {
@@ -386,7 +387,6 @@ typedef struct SSqlObj {
   SSqlRes          res;
 
   SSubqueryState   subState;
-  pthread_mutex_t  mtxSubs;      // avoid double access pSubs after failure
   struct SSqlObj **pSubs;
   struct SSqlObj  *rootObj;
 
@@ -434,6 +434,12 @@ typedef struct SSqlStream {
   struct SSqlStream *prev, *next;
 } SSqlStream;
 
+SSqlObj* tscAllocSqlObj();
+uint32_t tscGetVersionOfSubStateWithoutLock(SSqlObj *pSql);
+SSqlObj* tscAcquireRefOfSubobj(SSqlObj *pSql, int32_t idx, uint32_t stateVersion);
+void tscReleaseRefOfSubobj(SSqlObj *pSql);
+void tscResetAllSubStates(SSqlObj* pSql);
+
 void tscSetStreamDestTable(SSqlStream* pStream, const char* dstTable);
 
 int  tscAcquireRpc(const char *key, const char *user, const char *secret,void **pRpcObj);

src/client/src/tscAsync.c

@@ -46,8 +46,6 @@ void doAsyncQuery(STscObj* pObj, SSqlObj* pSql, __async_cb_func_t fp, void* para
   pSql->fetchFp   = fp;
   pSql->rootObj   = pSql;
 
-  pthread_mutex_init(&pSql->mtxSubs, NULL);
-
   registerSqlObj(pSql);
 
   pSql->sqlstr = calloc(1, sqlLen + 1);
@@ -108,7 +106,7 @@ TAOS_RES * taos_query_ra(TAOS *taos, const char *sqlstr, __async_cb_func_t fp, v
   
   nPrintTsc("%s", sqlstr);
   
-  SSqlObj *pSql = (SSqlObj *)calloc(1, sizeof(SSqlObj));
+  SSqlObj *pSql = tscAllocSqlObj();
   if (pSql == NULL) {
     tscError("failed to malloc sqlObj");
     tscQueueAsyncError(fp, param, TSDB_CODE_TSC_OUT_OF_MEMORY);

src/client/src/tscLocal.c

@@ -611,7 +611,7 @@ static int32_t tscRebuildDDLForSubTable(SSqlObj *pSql, const char *tableName, ch
   STableMetaInfo *pTableMetaInfo = tscGetMetaInfo(pQueryInfo, 0);
   STableMeta *    pMeta = pTableMetaInfo->pTableMeta;
 
-  SSqlObj *pInterSql = (SSqlObj *)calloc(1, sizeof(SSqlObj));
+  SSqlObj *pInterSql = tscAllocSqlObj();
   if (pInterSql == NULL) {
     return TSDB_CODE_TSC_OUT_OF_MEMORY;
   }
@@ -763,7 +763,7 @@ static int32_t tscProcessShowCreateDatabase(SSqlObj *pSql) {
 
   STableMetaInfo *pTableMetaInfo = tscGetMetaInfo(pQueryInfo, 0);
 
-  SSqlObj *pInterSql = (SSqlObj *)calloc(1, sizeof(SSqlObj)); 
+  SSqlObj *pInterSql = tscAllocSqlObj();
   if (pInterSql == NULL) {
     return TSDB_CODE_TSC_OUT_OF_MEMORY;
   }  

src/client/src/tscParseLineProtocol.c

@@ -558,7 +558,7 @@ static int32_t getSuperTableMetaFromLocalCache(TAOS* taos, char* tableName, STab
   int32_t     code = 0;
   STableMeta* tableMeta = NULL;
 
-  SSqlObj* pSql = calloc(1, sizeof(SSqlObj));
+  SSqlObj* pSql = tscAllocSqlObj();
   if (pSql == NULL) {
     tscError("SML:0x%" PRIx64 " failed to allocate memory, reason:%s", info->id, strerror(errno));
     code = TSDB_CODE_TSC_OUT_OF_MEMORY;
@@ -2763,7 +2763,7 @@ static int32_t convertPrecisionType(int precision, SMLTimeStampType *tsType) {
 
 //make a dummy SSqlObj
 static SSqlObj* createSmlQueryObj(TAOS* taos, int32_t affected_rows, int32_t code) {
-  SSqlObj *pNew = (SSqlObj*)calloc(1, sizeof(SSqlObj));
+  SSqlObj *pNew = tscAllocSqlObj();
   if (pNew == NULL) {
     return NULL;
   }
@@ -2771,7 +2771,6 @@ static SSqlObj* createSmlQueryObj(TAOS* taos, int32_t affected_rows, int32_t cod
   pNew->pTscObj = taos;
   pNew->fp = NULL;
 
-  tsem_init(&pNew->rspSem, 0, 0);
   registerSqlObj(pNew);
 
   pNew->res.numOfRows = affected_rows;

src/client/src/tscPrepare.c

@@ -1261,8 +1261,6 @@ static void insertBatchClean(STscStmt* pStmt) {
   taosHashClear(pCmd->insertParam.pTableBlockHashList);
   tscFreeSqlResult(pSql);
   tscFreeSubobj(pSql);
-  tfree(pSql->pSubs);
-  pSql->subState.numOfSub = 0;
 }
 
 static int insertBatchStmtExecute(STscStmt* pStmt) {
@@ -1589,7 +1587,7 @@ TAOS_STMT* taos_stmt_init(TAOS* taos) {
   }
   pStmt->taos = pObj;
 
-  SSqlObj* pSql = calloc(1, sizeof(SSqlObj));
+  SSqlObj* pSql = tscAllocSqlObj();
 
   if (pSql == NULL) {
     free(pStmt);
@@ -1606,7 +1604,6 @@ TAOS_STMT* taos_stmt_init(TAOS* taos) {
     return NULL;
   }
 
-  tsem_init(&pSql->rspSem, 0, 0);
   pSql->signature   = pSql;
   pSql->pTscObj     = pObj;
   pSql->maxRetry    = TSDB_MAX_REPLICA;

src/client/src/tscProfile.c

@@ -280,7 +280,8 @@ int tscBuildQueryStreamDesc(void *pMsg, STscObj *pObj) {
 //      } else {
 //        pQdesc->stableQuery = 0;
 //      }
-      pthread_mutex_lock(&pSql->subState.mutex);
+      { pthread_mutex_lock(&pSql->subState.mutex);
+
       if (pSql->pSubs != NULL && pSql->subState.states != NULL) {
         for (int32_t i = 0; i < pSql->subState.numOfSub; ++i) {
           // because subState maybe free on anytime by any thread, check validate from here
@@ -297,7 +298,8 @@ int tscBuildQueryStreamDesc(void *pMsg, STscObj *pObj) {
         }
       }
       pQdesc->numOfSub = pSql->subState.numOfSub;
-      pthread_mutex_unlock(&pSql->subState.mutex);
+
+      pthread_mutex_unlock(&pSql->subState.mutex); }
     }
 
     pQdesc->numOfSub = htonl(pQdesc->numOfSub);

src/client/src/tscServer.c

@@ -353,8 +353,8 @@ void checkBrokenQueries(STscObj *pTscObj) {
         pSql->lastAlive = taosGetTimestampMs();
       }
     } else {
-      // lock subs
-      pthread_mutex_lock(&pSql->subState.mutex);
+      { pthread_mutex_lock(&pSql->subState.mutex);
+
       if (pSql->pSubs) {
         // have sub sql
         for (int i = 0; i < pSql->subState.numOfSub; i++) {
@@ -375,8 +375,8 @@ void checkBrokenQueries(STscObj *pTscObj) {
           }
         }
       }
-      // unlock
-      pthread_mutex_unlock(&pSql->subState.mutex);
+
+      pthread_mutex_unlock(&pSql->subState.mutex); }
     }
 
     // kill query
@@ -2708,7 +2708,7 @@ static void createHbObj(STscObj* pObj) {
     return;
   }
 
-  SSqlObj *pSql = (SSqlObj *)calloc(1, sizeof(SSqlObj));
+  SSqlObj *pSql = tscAllocSqlObj();
   if (NULL == pSql) return;
 
   pSql->fp = tscProcessHeartBeatRsp;
@@ -2977,7 +2977,7 @@ int tscProcessRetrieveRspFromNode(SSqlObj *pSql) {
 void tscTableMetaCallBack(void *param, TAOS_RES *res, int code);
 
 static int32_t getTableMetaFromMnode(SSqlObj *pSql, STableMetaInfo *pTableMetaInfo, bool autocreate) {
-  SSqlObj *pNew = calloc(1, sizeof(SSqlObj));
+  SSqlObj *pNew = tscAllocSqlObj();
   if (NULL == pNew) {
     tscError("0x%"PRIx64" malloc failed for new sqlobj to get table meta", pSql->self);
     return TSDB_CODE_TSC_OUT_OF_MEMORY;
@@ -3043,7 +3043,7 @@ static int32_t getTableMetaFromMnode(SSqlObj *pSql, STableMetaInfo *pTableMetaIn
 }
 
 int32_t getMultiTableMetaFromMnode(SSqlObj *pSql, SArray* pNameList, SArray* pVgroupNameList, SArray* pUdfList, __async_cb_func_t fp, bool metaClone) {
-  SSqlObj *pNew = calloc(1, sizeof(SSqlObj));
+  SSqlObj *pNew = tscAllocSqlObj();
   if (NULL == pNew) {
     tscError("0x%"PRIx64" failed to allocate sqlobj to get multiple table meta", pSql->self);
     return TSDB_CODE_TSC_OUT_OF_MEMORY;
@@ -3180,7 +3180,7 @@ int tscGetTableMetaEx(SSqlObj *pSql, STableMetaInfo *pTableMetaInfo, bool create
 }
 
 int32_t tscGetUdfFromNode(SSqlObj *pSql, SQueryInfo* pQueryInfo) {
-  SSqlObj *pNew = calloc(1, sizeof(SSqlObj));
+  SSqlObj *pNew = tscAllocSqlObj();
   if (NULL == pNew) {
     tscError("%p malloc failed for new sqlobj to get user-defined functions", pSql);
     return TSDB_CODE_TSC_OUT_OF_MEMORY;
@@ -3291,10 +3291,7 @@ int tscRenewTableMeta(SSqlObj *pSql) {
   pSql->rootObj->retryReason = pSql->retryReason;
 
   SSqlObj *rootSql = pSql->rootObj;
-  pthread_mutex_lock(&rootSql->mtxSubs);
   tscFreeSubobj(rootSql);
-  pthread_mutex_unlock(&rootSql->mtxSubs);
-  tfree(rootSql->pSubs);
   tscResetSqlCmd(&rootSql->cmd, true, rootSql->self);
 
   code = getMultiTableMetaFromMnode(rootSql, pNameList, vgroupList, NULL, tscTableMetaCallBack, true);
@@ -3321,7 +3318,7 @@ int tscGetSTableVgroupInfo(SSqlObj *pSql, SQueryInfo* pQueryInfo) {
   if (allVgroupInfoRetrieved(pQueryInfo)) {
     return TSDB_CODE_SUCCESS;
   }
-  SSqlObj *pNew = calloc(1, sizeof(SSqlObj));
+  SSqlObj *pNew = tscAllocSqlObj();
   pNew->pTscObj = pSql->pTscObj;
   pNew->signature = pNew;
 

src/client/src/tscSql.c

@@ -143,7 +143,7 @@ static SSqlObj *taosConnectImpl(const char *ip, const char *user, const char *pa
 
   pthread_mutex_init(&pObj->mutex, NULL);
 
-  SSqlObj *pSql = (SSqlObj *)calloc(1, sizeof(SSqlObj));
+  SSqlObj *pSql = tscAllocSqlObj();
   if (NULL == pSql) {
     terrno = TSDB_CODE_TSC_OUT_OF_MEMORY;
     tscReleaseRpc(pRpcObj);
@@ -158,8 +158,6 @@ static SSqlObj *taosConnectImpl(const char *ip, const char *user, const char *pa
   pSql->param     = param;
   pSql->cmd.command = TSDB_SQL_CONNECT;
 
-  tsem_init(&pSql->rspSem, 0, 0);
-
   if (TSDB_CODE_SUCCESS != tscAllocPayload(&pSql->cmd, TSDB_DEFAULT_PAYLOAD_SIZE)) {
     terrno = TSDB_CODE_TSC_OUT_OF_MEMORY;
     tscReleaseRpc(pRpcObj);
@@ -307,10 +305,6 @@ void taos_close(TAOS *taos) {
 
       tscDebug("0x%"PRIx64" HB is freed", pHb->self);
       taosReleaseRef(tscObjRef, pHb->self);
-#ifdef __APPLE__
-      // to satisfy later tsem_destroy in taos_free_result
-      tsem_init(&pHb->rspSem, 0, 0);
-#endif // __APPLE__
       taos_free_result(pHb);
     }
   }
@@ -346,14 +340,13 @@ TAOS_RES* taos_query_c(TAOS *taos, const char *sqlstr, uint32_t sqlLen, int64_t*
 
   nPrintTsc("%s", sqlstr);
 
-  SSqlObj* pSql = calloc(1, sizeof(SSqlObj));
+  SSqlObj* pSql = tscAllocSqlObj();
   if (pSql == NULL) {
     tscError("failed to malloc sqlObj");
     terrno = TSDB_CODE_TSC_OUT_OF_MEMORY;
     return NULL;
   }
   
-  tsem_init(&pSql->rspSem, 0, 0);
   doAsyncQuery(pObj, pSql, waitForQueryRsp, taos, sqlstr, sqlLen);
 
   if (res != NULL) {
@@ -730,7 +723,9 @@ static void tscKillSTableQuery(SSqlObj *pSql) {
   pSql->res.code = TSDB_CODE_TSC_QUERY_CANCELLED;
 
   tscLockByThread(&pSql->squeryLock);
-  
+
+  { pthread_mutex_lock(&pSql->subState.mutex);
+
   for (int i = 0; i < pSql->subState.numOfSub; ++i) {
     // NOTE: pSub may have been released already here
     SSqlObj *pSub = pSql->pSubs[i];
@@ -750,6 +745,8 @@ static void tscKillSTableQuery(SSqlObj *pSql) {
     // taosRelekaseRef(tscObjRef, pSubObj->self);
   }
 
+  pthread_mutex_unlock(&pSql->subState.mutex); }
+
   if (pSql->subState.numOfSub <= 0) {
     tscAsyncResultOnError(pSql);
   }
@@ -923,7 +920,7 @@ int taos_validate_sql(TAOS *taos, const char *sql) {
     return TSDB_CODE_TSC_DISCONNECTED;
   }
 
-  SSqlObj* pSql = calloc(1, sizeof(SSqlObj));
+  SSqlObj* pSql = tscAllocSqlObj();
 
   pSql->pTscObj  = taos;
   pSql->signature = pSql;
@@ -1041,7 +1038,7 @@ int taos_load_table_info(TAOS *taos, const char *tableNameList) {
     return TSDB_CODE_TSC_OUT_OF_MEMORY;
   }
 
-  SSqlObj* pSql = calloc(1, sizeof(SSqlObj));
+  SSqlObj* pSql = tscAllocSqlObj();
   tscAllocPayload(&pSql->cmd, 1024);
 
   pSql->pTscObj   = taos;

src/client/src/tscStream.c

@@ -208,8 +208,6 @@ static void tscProcessStreamQueryCallback(void *param, TAOS_RES *tres, int numOf
 
     tscFreeSqlResult(pSql);
     tscFreeSubobj(pSql);
-    tfree(pSql->pSubs);
-    pSql->subState.numOfSub = 0;
 
     pSql->parseRetry = 0;
     int32_t code = tsParseSql(pSql, true);
@@ -332,8 +330,6 @@ static void tscProcessStreamRetrieveResult(void *param, TAOS_RES *res, int numOf
 
     tscFreeSqlResult(pSql);
     tscFreeSubobj(pSql);    
-    tfree(pSql->pSubs);
-    pSql->subState.numOfSub = 0;
     pTableMetaInfo->vgroupList = tscVgroupInfoClear(pTableMetaInfo->vgroupList);
     tscSetNextLaunchTimer(pStream, pSql);
   }
@@ -677,7 +673,7 @@ TAOS_STREAM *taos_open_stream_withname(TAOS *taos, const char* dstTable, const c
     return NULL;
   }
 
-  SSqlObj *pSql = (SSqlObj *)calloc(1, sizeof(SSqlObj));
+  SSqlObj *pSql = tscAllocSqlObj();
   if (pSql == NULL) {
     return NULL;
   }
@@ -722,7 +718,6 @@ TAOS_STREAM *taos_open_stream_withname(TAOS *taos, const char* dstTable, const c
   pSql->fetchFp = tscCreateStream;
   pSql->cmd.resColumnId = TSDB_RES_COL_ID;
 
-  tsem_init(&pSql->rspSem, 0, 0);
   registerSqlObj(pSql);
 
   tscDebugL("0x%"PRIx64" SQL: %s", pSql->self, pSql->sqlstr);

src/client/src/tscSub.c

@@ -83,7 +83,6 @@ void tscUpdateSubscriptionProgress(void* sub, int64_t uid, TSKEY ts) {
   }
 }
 
-
 static void asyncCallback(void *param, TAOS_RES *tres, int code) {
   assert(param != NULL);
   SSub *pSub = ((SSub *)param);
@@ -117,7 +116,7 @@ static SSub* tscCreateSubscription(STscObj* pObj, const char* topic, const char*
     goto fail;
   }
 
-  pSql = calloc(1, sizeof(SSqlObj));
+  pSql = tscAllocSqlObj();
   if (pSql == NULL) {
     line = __LINE__;
     code = TSDB_CODE_TSC_OUT_OF_MEMORY;
@@ -132,11 +131,6 @@ static SSub* tscCreateSubscription(STscObj* pObj, const char* topic, const char*
 
   SSqlCmd* pCmd = &pSql->cmd;
   SSqlRes* pRes = &pSql->res;
-  if (tsem_init(&pSql->rspSem, 0, 0) == -1) {
-    line = __LINE__;
-    code = TAOS_SYSTEM_ERROR(errno);
-    goto fail;
-  }
 
   pSql->param = pSub;
   pSql->maxRetry = TSDB_MAX_REPLICA;
@@ -432,7 +426,7 @@ TAOS_SUB *taos_subscribe(TAOS *taos, int restart, const char* topic, const char
 }
 
 SSqlObj* recreateSqlObj(SSub* pSub) {
-  SSqlObj* pSql = calloc(1, sizeof(SSqlObj));
+  SSqlObj* pSql = tscAllocSqlObj();
   if (pSql == NULL) {
     return NULL;
   }
@@ -442,10 +436,6 @@ SSqlObj* recreateSqlObj(SSub* pSub) {
 
   SSqlCmd* pCmd = &pSql->cmd;
   SSqlRes* pRes = &pSql->res;
-  if (tsem_init(&pSql->rspSem, 0, 0) == -1) {
-    tscFreeSqlObj(pSql);
-    return NULL;
-  }
 
   pSql->param = pSub;
   pSql->maxRetry = TSDB_MAX_REPLICA;

src/client/src/tscUtil.c

@@ -1656,29 +1656,30 @@ void tscFreeSqlResult(SSqlObj* pSql) {
 }
 
 void tscFreeSubobj(SSqlObj* pSql) {
+  pthread_mutex_lock(&pSql->subState.mutex);
   if (pSql->subState.numOfSub == 0) {
-    return;
+    goto _out;
   }
+  pSql->subState.version ++;
 
   tscDebug("0x%"PRIx64" start to free sub SqlObj, numOfSub:%d", pSql->self, pSql->subState.numOfSub);
 
   for(int32_t i = 0; i < pSql->subState.numOfSub; ++i) {
-    if (pSql->pSubs[i] != NULL) {
-      tscDebug("0x%"PRIx64" free sub SqlObj:0x%"PRIx64", index:%d", pSql->self, pSql->pSubs[i]->self, i);
-    } else {
-      /* just for python error test case */
+    if (!pSql->pSubs[i]) {
       tscDebug("0x%"PRIx64" free sub SqlObj:0x0, index:%d", pSql->self, i);
+      continue;
     }
+    tscDebug("0x%"PRIx64" free sub SqlObj:0x%"PRIx64", index:%d", pSql->self, pSql->pSubs[i]->self, i);
     taos_free_result(pSql->pSubs[i]);
     pSql->pSubs[i] = NULL;
   }
 
-  if (pSql->subState.states) {
-    pthread_mutex_destroy(&pSql->subState.mutex);
-  }
-
+  tfree(pSql->pSubs);
   tfree(pSql->subState.states);
   pSql->subState.numOfSub = 0;
+
+_out:
+  pthread_mutex_unlock(&pSql->subState.mutex);
 }
 
 /**
@@ -1715,6 +1716,52 @@ void tscFreeMetaSqlObj(int64_t *rid){
   }
 }
 
+SSqlObj* tscAllocSqlObj() {
+    SSqlObj *pNew = calloc(1, sizeof(SSqlObj));
+    if (!pNew) {
+        return NULL;
+    }
+    int rc = tsem_init(&pNew->rspSem, 0, 0);
+    assert(rc == 0 && "tsem_init failure");
+    rc = pthread_mutex_init(&pNew->subState.mutex, NULL);
+    assert(rc == 0 && "pthread_mutex_init failure");
+    return pNew;
+}
+
+SSqlObj* tscAcquireRefOfSubobj(SSqlObj* pSql, int32_t idx, uint32_t stateVersion) {
+    assert (pSql != NULL);
+    SSqlObj *pSub = NULL;
+
+    { pthread_mutex_lock(&pSql->subState.mutex);
+    if (stateVersion != tscGetVersionOfSubStateWithoutLock(pSql) ||
+        idx < 0 ||
+        idx >= pSql->subState.numOfSub ||
+        !pSql->pSubs[idx]) {
+        goto _out;
+    }
+    pSub = taosAcquireRef(tscObjRef, pSql->pSubs[idx]->self);
+    assert (pSql->pSubs[idx] == pSub && "Refcounted subquery obj mismatch");
+
+_out:
+    pthread_mutex_unlock(&pSql->subState.mutex); }
+    return pSub;
+}
+
+void tscReleaseRefOfSubobj(SSqlObj* pSub) {
+    assert (pSub != NULL && pSub->self != 0 && "Subquery obj not refcounted");
+    taosReleaseRef(tscObjRef, pSub->self);
+}
+
+uint32_t tscGetVersionOfSubStateWithoutLock(SSqlObj *pSql) {
+    return pSql->subState.version;
+}
+
+void tscResetAllSubStates(SSqlObj* pSql) {
+    pthread_mutex_lock(&pSql->subState.mutex);
+    memset(pSql->subState.states, 0, sizeof(pSql->subState.states[0]) * pSql->subState.numOfSub);
+    pthread_mutex_unlock(&pSql->subState.mutex);
+}
+
 void tscFreeSqlObj(SSqlObj* pSql) {
   if (pSql == NULL || pSql->signature != pSql) {
     return;
@@ -1738,17 +1785,11 @@ void tscFreeSqlObj(SSqlObj* pSql) {
 
   tscFreeSubobj(pSql);
 
-  if (pSql && (pSql == pSql->rootObj)) {
-    pthread_mutex_destroy(&pSql->mtxSubs);
-  }
-
   pSql->signature = NULL;
   pSql->fp = NULL;
   tfree(pSql->sqlstr);
   tfree(pSql->pBuf);
 
-  tfree(pSql->pSubs);
-  pSql->subState.numOfSub = 0;
   pSql->self = 0;
 
   tscFreeSqlResult(pSql);
@@ -1759,6 +1800,7 @@ void tscFreeSqlObj(SSqlObj* pSql) {
 
   tscDebug("0x%"PRIx64" addr:%p free completed", sid, pSql);
 
+  pthread_mutex_destroy(&pSql->subState.mutex);
   tsem_destroy(&pSql->rspSem);
   memset(pSql, 0, sizeof(*pSql));
   free(pSql);
@@ -3744,7 +3786,7 @@ void registerSqlObj(SSqlObj* pSql) {
 }
 
 SSqlObj* createSimpleSubObj(SSqlObj* pSql, __async_cb_func_t fp, void* param, int32_t cmd) {
-  SSqlObj* pNew = (SSqlObj*)calloc(1, sizeof(SSqlObj));
+  SSqlObj* pNew = tscAllocSqlObj();
   if (pNew == NULL) {
     tscError("0x%"PRIx64" new subquery failed, tableIndex:%d", pSql->self, 0);
     return NULL;
@@ -3756,7 +3798,6 @@ SSqlObj* createSimpleSubObj(SSqlObj* pSql, __async_cb_func_t fp, void* param, in
 
   SSqlCmd* pCmd = &pNew->cmd;
   pCmd->command = cmd;
-  tsem_init(&pNew->rspSem, 0 ,0);
 
   if (tscAddQueryInfo(pCmd) != TSDB_CODE_SUCCESS) {
     tscFreeSqlObj(pNew);
@@ -3821,7 +3862,7 @@ static void doSetSqlExprAndResultFieldInfo(SQueryInfo* pNewQueryInfo, int64_t ui
 SSqlObj* createSubqueryObj(SSqlObj* pSql, int16_t tableIndex, __async_cb_func_t fp, void* param, int32_t cmd, SSqlObj* pPrevSql) {
   SSqlCmd* pCmd = &pSql->cmd;
 
-  SSqlObj* pNew = (SSqlObj*)calloc(1, sizeof(SSqlObj));
+  SSqlObj* pNew = tscAllocSqlObj();
   if (pNew == NULL) {
     tscError("0x%"PRIx64" new subquery failed, tableIndex:%d", pSql->self, tableIndex);
     terrno = TSDB_CODE_TSC_OUT_OF_MEMORY;
@@ -3835,7 +3876,6 @@ SSqlObj* createSubqueryObj(SSqlObj* pSql, int16_t tableIndex, __async_cb_func_t
   pNew->signature = pNew;
   pNew->sqlstr    = strdup(pSql->sqlstr);
   pNew->rootObj   = pSql->rootObj;
-  tsem_init(&pNew->rspSem, 0, 0);
 
   SSqlCmd* pnCmd  = &pNew->cmd;
   memcpy(pnCmd, pCmd, sizeof(SSqlCmd));
@@ -4124,10 +4164,7 @@ static void tscSubqueryCompleteCallback(void* param, TAOS_RES* tres, int code) {
     }
 
     tscFreeSubobj(pParentSql);
-    tfree(pParentSql->pSubs);
-
     tscFreeSubobj(rootObj);
-    tfree(rootObj->pSubs);
 
     rootObj->res.code = TSDB_CODE_SUCCESS;
     rootObj->retry++;
@@ -4163,34 +4200,25 @@ static void tscSubqueryCompleteCallback(void* param, TAOS_RES* tres, int code) {
     return;
   }
 
-
   taos_fetch_rows_a(tres, tscSubqueryRetrieveCallback, param);
 }
 
-int32_t doInitSubState(SSqlObj* pSql, int32_t numOfSubqueries) {
-  //bug fix. Above doInitSubState level, the loop invocation with the same SSqlObj will be fail.
-  //assert(pSql->subState.numOfSub == 0 && pSql->pSubs == NULL && pSql->subState.states == NULL);  
-  if(pSql->pSubs) {
-    free(pSql->pSubs);
-    pSql->pSubs = NULL;
-  }
-  
-  if(pSql->subState.states) {
-    free(pSql->subState.states);
-    pSql->subState.states = NULL;
-  }
-  
-  pSql->subState.numOfSub = numOfSubqueries;
-  
-  pSql->pSubs = calloc(pSql->subState.numOfSub, POINTER_BYTES);
-  pSql->subState.states = calloc(pSql->subState.numOfSub, sizeof(int8_t));
+int32_t doReInitSubState(SSqlObj* pSql, int32_t numOfSubqueries) {
+  tscFreeSubobj(pSql);
+  int32_t code = TSDB_CODE_SUCCESS;
 
-  int32_t code = pthread_mutex_init(&pSql->subState.mutex, NULL);
-  if (pSql->pSubs == NULL || pSql->subState.states == NULL || code != 0) {
-    return TSDB_CODE_TSC_OUT_OF_MEMORY;
+  { pthread_mutex_lock(&pSql->subState.mutex);
+
+  pSql->pSubs = calloc(numOfSubqueries, POINTER_BYTES);
+  pSql->subState.states = calloc(numOfSubqueries, sizeof(int8_t));
+  if (pSql->pSubs == NULL || pSql->subState.states == NULL) {
+    code = TSDB_CODE_TSC_OUT_OF_MEMORY;
   }
+  pSql->subState.numOfSub = numOfSubqueries;
+  pSql->subState.version ++;
 
-  return TSDB_CODE_SUCCESS;
+  pthread_mutex_unlock(&pSql->subState.mutex); }
+  return code;
 }
 
 // do execute the query according to the query execution plan
@@ -4209,21 +4237,26 @@ void executeQuery(SSqlObj* pSql, SQueryInfo* pQueryInfo) {
 
   // upstream may be freed before retry
   if (pQueryInfo->pUpstream && taosArrayGetSize(pQueryInfo->pUpstream) > 0) {  // nest query. do execute it firstly
-    code = doInitSubState(pSql, (int32_t) taosArrayGetSize(pQueryInfo->pUpstream));
+    code = doReInitSubState(pSql, (int32_t) taosArrayGetSize(pQueryInfo->pUpstream));
     if (code != TSDB_CODE_SUCCESS) {
       goto _error;
     }
 
+    uint32_t stateVersion = 0;
+    int errflag = 0;
+    { pthread_mutex_lock(&pSql->subState.mutex);
+
     for(int32_t i = 0; i < pSql->subState.numOfSub; ++i) {
       SQueryInfo* pSub = taosArrayGetP(pQueryInfo->pUpstream, i);
 
       pSql->cmd.active = pSub;
       pSql->cmd.command = TSDB_SQL_SELECT;
 
-      SSqlObj* pNew = (SSqlObj*)calloc(1, sizeof(SSqlObj));
+      SSqlObj* pNew = tscAllocSqlObj();
       if (pNew == NULL) {
         code = TSDB_CODE_TSC_OUT_OF_MEMORY;
-        goto _error;
+        errflag = 1;
+        break;
       }
 
       pNew->pTscObj   = pSql->pTscObj;
@@ -4236,24 +4269,26 @@ void executeQuery(SSqlObj* pSql, SQueryInfo* pQueryInfo) {
 
       pNew->cmd.resColumnId = TSDB_RES_COL_ID;
 
-      tsem_init(&pNew->rspSem, 0, 0);
-
       SRetrieveSupport* ps = calloc(1, sizeof(SRetrieveSupport));  // todo use object id
       if (ps == NULL) {
         tscFreeSqlObj(pNew);
-        goto _error;
+        errflag = 1;
+        break;
       }
 
       ps->pParentSql = pSql;
       ps->subqueryIndex = i;
-
       pNew->param = ps;
+
+      registerSqlObj(pNew);
+
       pSql->pSubs[i] = pNew;
 
       SSqlCmd* pCmd = &pNew->cmd;
       pCmd->command = TSDB_SQL_SELECT;
       if ((code = tscAddQueryInfo(pCmd)) != TSDB_CODE_SUCCESS) {
-        goto _error;
+        errflag = 1;
+        break;
       }
 
       SQueryInfo* pNewQueryInfo = tscGetQueryInfo(pCmd);
@@ -4263,9 +4298,23 @@ void executeQuery(SSqlObj* pSql, SQueryInfo* pQueryInfo) {
       numOfInit++;
     }
 
+    stateVersion = tscGetVersionOfSubStateWithoutLock(pSql);
+
+    pthread_mutex_unlock(&pSql->subState.mutex); }
+    if (errflag) {
+        goto _error;
+    }
+
     for(int32_t i = 0; i < pSql->subState.numOfSub; ++i) {
-      SSqlObj* psub = pSql->pSubs[i];
-      registerSqlObj(psub);
+      SSqlObj* psub = tscAcquireRefOfSubobj(pSql, i, stateVersion);  // ACQ ref
+      if (!psub) {
+        if (stateVersion == tscGetVersionOfSubStateWithoutLock(pSql)) {
+            continue;
+        }
+        tscError("0x%"PRIx64"subqueries objs reset unexpectedly. numOfSub:%d", pSql->self, pSql->subState.numOfSub);
+        code = TSDB_CODE_FAILED;
+        goto _error;
+      }
 
       // create sub query to handle the sub query.
       SQueryInfo* pq = tscGetQueryInfo(&psub->cmd);
@@ -4275,29 +4324,21 @@ void executeQuery(SSqlObj* pSql, SQueryInfo* pQueryInfo) {
         psub->cmd.command = TSDB_SQL_RETRIEVE_EMPTY_RESULT;
       }
       executeQuery(psub, pq);
+
+      tscReleaseRefOfSubobj(psub); // REL ref
     }
 
     return;
-  } else if (hasMoreClauseToTry(pSql)) {
-    if (pthread_mutex_init(&pSql->subState.mutex, NULL) != 0) {
-      goto _error;
-    }
   }
 
   pSql->cmd.active = pQueryInfo;
   doExecuteQuery(pSql, pQueryInfo);
   return;
 
-  _error:
-  for(int32_t i = 0; i < numOfInit; ++i) {
-    SSqlObj* p = pSql->pSubs[i];
-    tscFreeSqlObj(p);
-  }
-
+_error:
   pSql->res.code = code;
-  pSql->subState.numOfSub = 0;   // not initialized sub query object will not be freed
-  tfree(pSql->subState.states);
-  tfree(pSql->pSubs);
+  tscFreeSubobj(pSql);
+
   tscAsyncResultOnError(pSql);
 }
 
@@ -4532,7 +4573,7 @@ void tscTryQueryNextVnode(SSqlObj* pSql, __async_cb_func_t fp) {
      *
      * For super table join with projection query, if anyone of the subquery is exhausted, the query completed.
      */
-    pSql->subState.numOfSub = 0;
+    tscFreeSubobj(pSql);
     pCmd->command = TSDB_SQL_SELECT;
 
     tscResetForNextRetrieve(pRes);
@@ -4564,16 +4605,7 @@ void tscTryQueryNextClause(SSqlObj* pSql, __async_cb_func_t fp) {
   pRes->final = finalBk;
   pRes->numOfTotal = num;
 
-  pthread_mutex_lock(&pSql->subState.mutex);
-  for(int32_t i = 0; i < pSql->subState.numOfSub; ++i) {
-    taos_free_result(pSql->pSubs[i]);
-  }
-
-  tfree(pSql->pSubs);
-  tfree(pSql->subState.states);
-  pSql->subState.numOfSub = 0;
-  pthread_mutex_unlock(&pSql->subState.mutex);
-  pthread_mutex_destroy(&pSql->subState.mutex);
+  tscFreeSubobj(pSql);
 
   pSql->fp = fp;
 

src/util/src/tref.c

@@ -195,7 +195,6 @@ int taosRemoveRef(int rsetId, int64_t rid)
   return taosDecRefCount(rsetId, rid, 1);
 }
 
-// if rid is 0, return the first p in hash list, otherwise, return the next after current rid
 void *taosAcquireRef(int rsetId, int64_t rid)
 {
   int       hash;