Merge remote-tracking branch 'origin/3.0' into fix/dnode

.gitmodules

@@ -16,3 +16,6 @@
 [submodule "tools/taos-tools"]
 	path = tools/taos-tools
 	url = https://github.com/taosdata/taos-tools
+[submodule "tools/taosadapter"]
+	path = tools/taosadapter
+	url = https://github.com/taosdata/taosadapter.git

Jenkinsfile2

@@ -38,6 +38,7 @@ def pre_test(){
     sh '''
         hostname
         date
+        env
     '''
     sh '''
         cd ${WK}
@@ -82,23 +83,33 @@ def pre_test(){
         sh '''
             cd ${WKC}
             git pull >/dev/null
+            git log -5
+            echo "`date "+%Y%m%d-%H%M%S"` ${JOB_NAME}:${BRANCH_NAME}:${BUILD_ID}:${CHANGE_TARGET}" >>${WKDIR}/jenkins.log
+            echo "community log: `git log -5`" >>${WKDIR}/jenkins.log
             git fetch origin +refs/pull/${CHANGE_ID}/merge
             git checkout -qf FETCH_HEAD
             git log -5
+            echo "community log merged: `git log -5`" >>${WKDIR}/jenkins.log
             cd ${WK}
             git pull >/dev/null
             git log -5
+            echo "tdinternal log: `git log -5`" >>${WKDIR}/jenkins.log
         '''
     } else if (env.CHANGE_URL =~ /\/TDinternal\//) {
         sh '''
             cd ${WK}
             git pull >/dev/null
+            git log -5
+            echo "`date "+%Y%m%d-%H%M%S"` ${JOB_NAME}:${BRANCH_NAME}:${BUILD_ID}:${CHANGE_TARGET}" >>${WKDIR}/jenkins.log
+            echo "tdinternal log: `git log -5`" >>${WKDIR}/jenkins.log
             git fetch origin +refs/pull/${CHANGE_ID}/merge
             git checkout -qf FETCH_HEAD
             git log -5
+            echo "tdinternal log merged: `git log -5`" >>${WKDIR}/jenkins.log
             cd ${WKC}
             git pull >/dev/null
             git log -5
+            echo "community log: `git log -5`" >>${WKDIR}/jenkins.log
         '''
     } else {
         sh '''
@@ -113,6 +124,9 @@ def pre_test(){
         cd ${WKPY}
         git reset --hard
         git pull
+        git log -5
+        echo "python connector log: `git log -5`" >>${WKDIR}/jenkins.log
+        echo >>${WKDIR}/jenkins.log
     '''
     return 1
 }

cmake/cmake.define

@@ -18,6 +18,33 @@ if (NOT DEFINED TD_GRANT)
   SET(TD_GRANT FALSE)
 endif()
 
+IF ("${BUILD_HTTP}" STREQUAL "")
+  IF (TD_LINUX)
+    IF (TD_ARM_32)
+      SET(TD_BUILD_HTTP TRUE)
+    ELSE ()
+      SET(TD_BUILD_HTTP TRUE)
+    ENDIF ()
+  ELSEIF (TD_DARWIN)
+    SET(TD_BUILD_HTTP TRUE)
+  ELSE ()
+    SET(TD_BUILD_HTTP TRUE)
+  ENDIF ()
+ELSEIF (${BUILD_HTTP} MATCHES "false")
+  SET(TD_BUILD_HTTP FALSE)
+ELSEIF (${BUILD_HTTP} MATCHES "true")
+  SET(TD_BUILD_HTTP TRUE)
+ELSEIF (${BUILD_HTTP} MATCHES "internal")
+  SET(TD_BUILD_HTTP FALSE)
+  SET(TD_BUILD_TAOSA_INTERNAL TRUE)
+ELSE ()
+  SET(TD_BUILD_HTTP TRUE)
+ENDIF ()
+
+IF (TD_BUILD_HTTP)
+  ADD_DEFINITIONS(-DHTTP_EMBEDDED)
+ENDIF ()
+
 IF ("${BUILD_TOOLS}" STREQUAL "")
   IF (TD_LINUX)
     IF (TD_ARM_32)

examples/c/stream_demo.c

@@ -82,7 +82,8 @@ int32_t create_stream() {
   /*const char* sql = "select sum(k) from tu1 interval(10m)";*/
   /*pRes = tmq_create_stream(pConn, "stream1", "out1", sql);*/
   pRes = taos_query(
-      pConn, "create stream stream1 trigger at_once into outstb as select _wstartts, sum(k) from st1 interval(10m)");
+      pConn,
+      "create stream stream1 trigger max_delay 10s into outstb as select _wstartts, sum(k) from st1 interval(10m)");
   if (taos_errno(pRes) != 0) {
     printf("failed to create stream stream1, reason:%s\n", taos_errstr(pRes));
     return -1;

include/common/systable.h

@@ -52,6 +52,7 @@ extern "C" {
 #define TSDB_PERFS_TABLE_OFFSETS       "offsets"
 #define TSDB_PERFS_TABLE_TRANS         "trans"
 #define TSDB_PERFS_TABLE_STREAMS       "streams"
+#define TSDB_PERFS_TABLE_APPS          "apps"
 
 typedef struct SSysDbTableSchema {
   const char*   name;

include/common/tmsg.h

@@ -1143,11 +1143,11 @@ void    tFreeSMAlterStbRsp(SMAlterStbRsp* pRsp);
 int32_t tSerializeSTableMetaRsp(void* buf, int32_t bufLen, STableMetaRsp* pRsp);
 int32_t tDeserializeSTableMetaRsp(void* buf, int32_t bufLen, STableMetaRsp* pRsp);
 void    tFreeSTableMetaRsp(STableMetaRsp* pRsp);
-void tFreeSTableIndexRsp(void *info);
+void    tFreeSTableIndexRsp(void* info);
 
 typedef struct {
-  SArray*         pMetaRsp;  // Array of STableMetaRsp
+  SArray* pMetaRsp;   // Array of STableMetaRsp
-  SArray*         pIndexRsp;  // Array of STableIndexRsp;
+  SArray* pIndexRsp;  // Array of STableIndexRsp;
 } SSTbHbRsp;
 
 int32_t tSerializeSSTbHbRsp(void* buf, int32_t bufLen, SSTbHbRsp* pRsp);
@@ -1314,8 +1314,9 @@ int32_t tSerializeSSetStandbyReq(void* buf, int32_t bufLen, SSetStandbyReq* pReq
 int32_t tDeserializeSSetStandbyReq(void* buf, int32_t bufLen, SSetStandbyReq* pReq);
 
 typedef struct {
-  int32_t connId;
+  int32_t connId;   // todo remove
-  int32_t queryId;
+  int32_t queryId;  // todo remove
+  char    queryStrId[TSDB_QUERY_ID_LEN];
 } SKillQueryReq;
 
 int32_t tSerializeSKillQueryReq(void* buf, int32_t bufLen, SKillQueryReq* pReq);
@@ -2436,6 +2437,7 @@ typedef struct {
 
 static FORCE_INLINE void tDestroyTSma(STSma* pSma) {
   if (pSma) {
+    taosMemoryFreeClear(pSma->dstTbName);
     taosMemoryFreeClear(pSma->expr);
     taosMemoryFreeClear(pSma->tagsFilter);
   }
@@ -2464,7 +2466,7 @@ int32_t tEncodeSVCreateTSmaReq(SEncoder* pCoder, const SVCreateTSmaReq* pReq);
 int32_t tDecodeSVCreateTSmaReq(SDecoder* pCoder, SVCreateTSmaReq* pReq);
 
 int32_t tEncodeTSma(SEncoder* pCoder, const STSma* pSma);
-int32_t tDecodeTSma(SDecoder* pCoder, STSma* pSma);
+int32_t tDecodeTSma(SDecoder* pCoder, STSma* pSma, bool deepCopy);
 
 static int32_t tEncodeTSmaWrapper(SEncoder* pEncoder, const STSmaWrapper* pReq) {
   if (tEncodeI32(pEncoder, pReq->number) < 0) return -1;
@@ -2474,10 +2476,10 @@ static int32_t tEncodeTSmaWrapper(SEncoder* pEncoder, const STSmaWrapper* pReq)
   return 0;
 }
 
-static int32_t tDecodeTSmaWrapper(SDecoder* pDecoder, STSmaWrapper* pReq) {
+static int32_t tDecodeTSmaWrapper(SDecoder* pDecoder, STSmaWrapper* pReq, bool deepCopy) {
   if (tDecodeI32(pDecoder, &pReq->number) < 0) return -1;
   for (int32_t i = 0; i < pReq->number; ++i) {
-    tDecodeTSma(pDecoder, pReq->tSma + i);
+    tDecodeTSma(pDecoder, pReq->tSma + i, deepCopy);
   }
   return 0;
 }

include/common/ttokendef.h

@@ -78,7 +78,7 @@
 #define TK_BUFFER        60
 #define TK_CACHELAST     61
 #define TK_COMP          62
-#define TK_DAYS          63
+#define TK_DURATION      63
 #define TK_NK_VARIABLE   64
 #define TK_FSYNC         65
 #define TK_MAXROWS       66

include/libs/executor/executor.h

@@ -38,8 +38,10 @@ typedef struct SReadHandle {
   SMsgCb* pMsgCb;
 } SReadHandle;
 
-#define STREAM_DATA_TYPE_SUBMIT_BLOCK 0x1
+enum {
-#define STREAM_DATA_TYPE_SSDATA_BLOCK 0x2
+  STREAM_DATA_TYPE_SUBMIT_BLOCK = 1,
+  STREAM_DATA_TYPE_SSDATA_BLOCK = 2,
+};
 
 typedef enum {
   OPTR_EXEC_MODEL_BATCH = 0x1,
@@ -102,7 +104,8 @@ int32_t qCreateExecTask(SReadHandle* readHandle, int32_t vgId, uint64_t taskId,
  * @param tversion
  * @return
  */
-int32_t qGetQueriedTableSchemaVersion(qTaskInfo_t tinfo, char* dbName, char* tableName, int32_t* sversion, int32_t* tversion);
+int32_t qGetQueriedTableSchemaVersion(qTaskInfo_t tinfo, char* dbName, char* tableName, int32_t* sversion,
+                                      int32_t* tversion);
 
 /**
  * The main task execution function, including query on both table and multiple tables,

include/libs/nodes/cmdnodes.h

@@ -89,6 +89,7 @@ typedef struct STableOptions {
   ENodeType  type;
   char       comment[TSDB_TB_COMMENT_LEN];
   double     filesFactor;
+  int32_t    delay;
   SNodeList* pRollupFuncs;
   int32_t    ttl;
   SNodeList* pSma;
@@ -286,6 +287,11 @@ typedef struct SKillStmt {
   int32_t   targetId;
 } SKillStmt;
 
+typedef struct SKillQueryStmt {
+  ENodeType type;
+  char      queryId[TSDB_QUERY_ID_LEN];
+} SKillQueryStmt;
+
 typedef struct SStreamOptions {
   ENodeType type;
   int8_t    triggerType;

include/libs/nodes/nodes.h

@@ -204,6 +204,7 @@ typedef enum ENodeType {
   QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN,
   QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN,
   QUERY_NODE_PHYSICAL_PLAN_TABLE_SEQ_SCAN,
+  QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN,
   QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN,
   QUERY_NODE_PHYSICAL_PLAN_SYSTABLE_SCAN,
   QUERY_NODE_PHYSICAL_PLAN_PROJECT,

include/libs/nodes/plannodes.h

@@ -34,7 +34,13 @@ typedef struct SLogicNode {
   uint8_t            precision;
 } SLogicNode;
 
-typedef enum EScanType { SCAN_TYPE_TAG = 1, SCAN_TYPE_TABLE, SCAN_TYPE_SYSTEM_TABLE, SCAN_TYPE_STREAM } EScanType;
+typedef enum EScanType {
+  SCAN_TYPE_TAG = 1,
+  SCAN_TYPE_TABLE,
+  SCAN_TYPE_SYSTEM_TABLE,
+  SCAN_TYPE_STREAM,
+  SCAN_TYPE_TABLE_MERGE
+} EScanType;
 
 typedef struct SScanLogicNode {
   SLogicNode    node;
@@ -262,6 +268,7 @@ typedef struct STableScanPhysiNode {
 } STableScanPhysiNode;
 
 typedef STableScanPhysiNode STableSeqScanPhysiNode;
+typedef STableScanPhysiNode STableMergeScanPhysiNode;
 typedef STableScanPhysiNode SStreamScanPhysiNode;
 
 typedef struct SProjectPhysiNode {

include/libs/stream/tstream.h

@@ -58,6 +58,7 @@ enum {
 enum {
   STREAM_INPUT__DATA_SUBMIT = 1,
   STREAM_INPUT__DATA_BLOCK,
+  STREAM_INPUT__TRIGGER,
   STREAM_INPUT__CHECKPOINT,
 };
 
@@ -85,6 +86,11 @@ typedef struct {
   int8_t type;
 } SStreamCheckpoint;
 
+typedef struct {
+  int8_t       type;
+  SSDataBlock* pBlock;
+} SStreamTrigger;
+
 enum {
   STREAM_QUEUE__SUCESS = 1,
   STREAM_QUEUE__FAILED,
@@ -98,6 +104,9 @@ typedef struct {
   int8_t      status;
 } SStreamQueue;
 
+int32_t streamInit();
+void    streamCleanUp();
+
 SStreamQueue* streamQueueOpen();
 void          streamQueueClose(SStreamQueue* queue);
 
@@ -220,6 +229,11 @@ enum {
   TASK_INPUT_TYPE__DATA_BLOCK,
 };
 
+enum {
+  TASK_TRIGGER_STATUS__IN_ACTIVE = 1,
+  TASK_TRIGGER_STATUS__ACTIVE,
+};
+
 struct SStreamTask {
   int64_t streamId;
   int32_t taskId;
@@ -262,8 +276,16 @@ struct SStreamTask {
   SStreamQueue* inputQueue;
   SStreamQueue* outputQueue;
 
+  // trigger
+  int8_t  triggerStatus;
+  int64_t triggerParam;
+  void*   timer;
+
   // application storage
   // void* ahandle;
+
+  // msg handle
+  SMsgCb* pMsgCb;
 };
 
 SStreamTask* tNewSStreamTask(int64_t streamId);
@@ -292,6 +314,13 @@ static FORCE_INLINE int32_t streamTaskInput(SStreamTask* pTask, SStreamQueueItem
     taosWriteQitem(pTask->inputQueue->queue, pItem);
   } else if (pItem->type == STREAM_INPUT__CHECKPOINT) {
     taosWriteQitem(pTask->inputQueue->queue, pItem);
+  } else if (pItem->type == STREAM_INPUT__TRIGGER) {
+    taosWriteQitem(pTask->inputQueue->queue, pItem);
+  }
+
+  if (pItem->type != STREAM_INPUT__TRIGGER && pItem->type != STREAM_INPUT__CHECKPOINT && pTask->triggerParam != 0 &&
+      pTask->triggerStatus == TASK_TRIGGER_STATUS__IN_ACTIVE) {
+    atomic_store_8(&pTask->triggerStatus, TASK_TRIGGER_STATUS__ACTIVE);
   }
 
   // TODO: back pressure
@@ -370,7 +399,8 @@ typedef struct {
 
 int32_t tDecodeStreamDispatchReq(SDecoder* pDecoder, SStreamDispatchReq* pReq);
 
-int32_t streamTriggerByWrite(SStreamTask* pTask, int32_t vgId, SMsgCb* pMsgCb);
+int32_t streamLaunchByWrite(SStreamTask* pTask, int32_t vgId, SMsgCb* pMsgCb);
+int32_t streamSetupTrigger(SStreamTask* pTask);
 
 int32_t streamTaskRun(SStreamTask* pTask);
 

include/libs/sync/sync.h

@@ -24,7 +24,7 @@ extern "C" {
 #include "tdef.h"
 #include "tmsgcb.h"
 
-#define SYNC_INDEX_BEGIN 0
+#define SYNC_INDEX_BEGIN   0
 #define SYNC_INDEX_INVALID -1
 
 typedef uint64_t SyncNodeId;
@@ -44,14 +44,6 @@ typedef enum {
   TAOS_SYNC_STATE_ERROR = 103,
 } ESyncState;
 
-typedef enum {
-  TAOS_SYNC_PROPOSE_SUCCESS = 0,
-  TAOS_SYNC_PROPOSE_NOT_LEADER = 1,
-  TAOS_SYNC_ONLY_ONE_REPLICA = 2,
-  TAOS_SYNC_NOT_IN_NEW_CONFIG = 3,
-  TAOS_SYNC_OTHER_ERROR = 100,
-} ESyncProposeCode;
-
 typedef enum {
   TAOS_SYNC_FSM_CB_SUCCESS = 0,
   TAOS_SYNC_FSM_CB_OTHER_ERROR = 1,

include/util/taoserror.h

@@ -412,6 +412,8 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_SYN_INVALID_MSGTYPE           TAOS_DEF_ERROR_CODE(0, 0x090A)
 
 #define TSDB_CODE_SYN_NOT_LEADER                TAOS_DEF_ERROR_CODE(0, 0x0910)
+#define TSDB_CODE_SYN_ONE_REPLICA               TAOS_DEF_ERROR_CODE(0, 0x0911)
+#define TSDB_CODE_SYN_NOT_IN_NEW_CONFIG         TAOS_DEF_ERROR_CODE(0, 0x0912)
 #define TSDB_CODE_SYN_INTERNAL_ERROR            TAOS_DEF_ERROR_CODE(0, 0x09FF)
 
 // tq
@@ -686,12 +688,15 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_SML_INVALID_DB_CONF           TAOS_DEF_ERROR_CODE(0, 0x3003)
 
 //tsma
-#define TSDB_CODE_TSMA_ALREADY_EXIST             TAOS_DEF_ERROR_CODE(0, 0x3100)
+#define TSDB_CODE_TSMA_INIT_FAILED               TAOS_DEF_ERROR_CODE(0, 0x3100)
-#define TSDB_CODE_TSMA_NO_INDEX_IN_META          TAOS_DEF_ERROR_CODE(0, 0x3101)
+#define TSDB_CODE_TSMA_ALREADY_EXIST             TAOS_DEF_ERROR_CODE(0, 0x3101)
-#define TSDB_CODE_TSMA_INVALID_ENV               TAOS_DEF_ERROR_CODE(0, 0x3102)
+#define TSDB_CODE_TSMA_NO_INDEX_IN_META          TAOS_DEF_ERROR_CODE(0, 0x3102)
-#define TSDB_CODE_TSMA_INVALID_STAT              TAOS_DEF_ERROR_CODE(0, 0x3103)
+#define TSDB_CODE_TSMA_INVALID_ENV               TAOS_DEF_ERROR_CODE(0, 0x3103)
-#define TSDB_CODE_TSMA_NO_INDEX_IN_CACHE         TAOS_DEF_ERROR_CODE(0, 0x3104)
+#define TSDB_CODE_TSMA_INVALID_STAT              TAOS_DEF_ERROR_CODE(0, 0x3104)
-#define TSDB_CODE_TSMA_RM_SKEY_IN_HASH           TAOS_DEF_ERROR_CODE(0, 0x3105)
+#define TSDB_CODE_TSMA_INVALID_PTR               TAOS_DEF_ERROR_CODE(0, 0x3105)
+#define TSDB_CODE_TSMA_INVALID_PARA              TAOS_DEF_ERROR_CODE(0, 0x3106)
+#define TSDB_CODE_TSMA_NO_INDEX_IN_CACHE         TAOS_DEF_ERROR_CODE(0, 0x3107)
+
 
 //rsma
 #define TSDB_CODE_RSMA_INVALID_ENV               TAOS_DEF_ERROR_CODE(0, 0x3150)
@@ -700,6 +705,7 @@ int32_t* taosGetErrno();
 //index
 #define TSDB_CODE_INDEX_REBUILDING               TAOS_DEF_ERROR_CODE(0, 0x3200)
 
+
 #ifdef __cplusplus
 }
 #endif

include/util/tdef.h

@@ -222,6 +222,8 @@ typedef enum ELogicConditionType {
 #define TSDB_APP_NAME_LEN   TSDB_UNI_LEN
 #define TSDB_TB_COMMENT_LEN 1025
 
+#define TSDB_QUERY_ID_LEN 26
+
 /**
  *  In some scenarios uint16_t (0~65535) is used to store the row len.
  *  - Firstly, we use 65531(65535 - 4), as the SDataRow/SKVRow contains 4 bits header.
@@ -341,6 +343,9 @@ typedef enum ELogicConditionType {
 #define TSDB_DB_SCHEMALESS_OFF          0
 #define TSDB_DEFAULT_DB_SCHEMALESS      TSDB_DB_SCHEMALESS_OFF
 
+// #define TSDB_MIN_ROLLUP_DELAY     1
+// #define TSDB_MAX_ROLLUP_DELAY     10
+// #define TSDB_DEFAULT_ROLLUP_DELAY 1
 #define TSDB_MIN_ROLLUP_FILE_FACTOR     0
 #define TSDB_MAX_ROLLUP_FILE_FACTOR     10
 #define TSDB_DEFAULT_ROLLUP_FILE_FACTOR 0.1

source/client/src/clientSml.c

@@ -2355,34 +2355,34 @@ static int smlProcess(SSmlHandle *info, char *lines[], int numLines) {
 }
 
 static int32_t isSchemalessDb(STscObj *taos, SRequestObj *request) {
-  SCatalog *catalog = NULL;
+//  SCatalog *catalog = NULL;
-  int32_t   code = catalogGetHandle(((STscObj *)taos)->pAppInfo->clusterId, &catalog);
+//  int32_t   code = catalogGetHandle(((STscObj *)taos)->pAppInfo->clusterId, &catalog);
-  if (code != TSDB_CODE_SUCCESS) {
+//  if (code != TSDB_CODE_SUCCESS) {
-    uError("SML get catalog error %d", code);
+//    uError("SML get catalog error %d", code);
-    return code;
+//    return code;
-  }
+//  }
-
+//
-  SName name;
+//  SName name;
-  tNameSetDbName(&name, taos->acctId, taos->db, strlen(taos->db));
+//  tNameSetDbName(&name, taos->acctId, taos->db, strlen(taos->db));
-  char dbFname[TSDB_DB_FNAME_LEN] = {0};
+//  char dbFname[TSDB_DB_FNAME_LEN] = {0};
-  tNameGetFullDbName(&name, dbFname);
+//  tNameGetFullDbName(&name, dbFname);
-  SDbCfgInfo pInfo = {0};
+//  SDbCfgInfo pInfo = {0};
-
+//
-  SRequestConnInfo conn = {0};
+//  SRequestConnInfo conn = {0};
-  conn.pTrans = taos->pAppInfo->pTransporter;
+//  conn.pTrans = taos->pAppInfo->pTransporter;
-  conn.requestId = request->requestId;
+//  conn.requestId = request->requestId;
-  conn.requestObjRefId = request->self;
+//  conn.requestObjRefId = request->self;
-  conn.mgmtEps = getEpSet_s(&taos->pAppInfo->mgmtEp);
+//  conn.mgmtEps = getEpSet_s(&taos->pAppInfo->mgmtEp);
-
+//
-  code = catalogGetDBCfg(catalog, &conn, dbFname, &pInfo);
+//  code = catalogGetDBCfg(catalog, &conn, dbFname, &pInfo);
-  if (code != TSDB_CODE_SUCCESS) {
+//  if (code != TSDB_CODE_SUCCESS) {
-    return code;
+//    return code;
-  }
+//  }
-  taosArrayDestroy(pInfo.pRetensions);
+//  taosArrayDestroy(pInfo.pRetensions);
-
+//
-  if (!pInfo.schemaless) {
+//  if (!pInfo.schemaless) {
-    return TSDB_CODE_SML_INVALID_DB_CONF;
+//    return TSDB_CODE_SML_INVALID_DB_CONF;
-  }
+//  }
   return TSDB_CODE_SUCCESS;
 }
 

source/client/src/tmq.c

@@ -992,6 +992,90 @@ CREATE_MSG_FAIL:
   return -1;
 }
 
+bool tmqUpdateEp2(tmq_t* tmq, int32_t epoch, SMqAskEpRsp* pRsp) {
+  bool set = false;
+
+  int32_t topicNumGet = taosArrayGetSize(pRsp->topics);
+  char    vgKey[TSDB_TOPIC_FNAME_LEN + 22];
+  tscDebug("consumer %ld update ep epoch %d to epoch %d, topic num: %d", tmq->consumerId, tmq->epoch, epoch,
+           topicNumGet);
+
+  SArray* newTopics = taosArrayInit(topicNumGet, sizeof(SMqClientTopic));
+  if (newTopics == NULL) {
+    return false;
+  }
+
+  SHashObj* pHash = taosHashInit(64, MurmurHash3_32, false, HASH_NO_LOCK);
+  if (pHash == NULL) {
+    taosArrayDestroy(newTopics);
+    return false;
+  }
+  int32_t topicNumCur = taosArrayGetSize(tmq->clientTopics);
+  for (int32_t i = 0; i < topicNumCur; i++) {
+    // find old topic
+    SMqClientTopic* pTopicCur = taosArrayGet(tmq->clientTopics, i);
+    if (pTopicCur->vgs) {
+      int32_t vgNumCur = taosArrayGetSize(pTopicCur->vgs);
+      tscDebug("consumer %ld new vg num: %d", tmq->consumerId, vgNumCur);
+      if (vgNumCur == 0) break;
+      for (int32_t j = 0; j < vgNumCur; j++) {
+        SMqClientVg* pVgCur = taosArrayGet(pTopicCur->vgs, j);
+        sprintf(vgKey, "%s:%d", pTopicCur->topicName, pVgCur->vgId);
+        tscDebug("consumer %ld epoch %d vg %d build %s", tmq->consumerId, epoch, pVgCur->vgId, vgKey);
+        taosHashPut(pHash, vgKey, strlen(vgKey), &pVgCur->currentOffset, sizeof(int64_t));
+      }
+      break;
+    }
+  }
+
+  for (int32_t i = 0; i < topicNumGet; i++) {
+    SMqClientTopic topic = {0};
+    SMqSubTopicEp* pTopicEp = taosArrayGet(pRsp->topics, i);
+    topic.schema = pTopicEp->schema;
+    taosHashClear(pHash);
+    topic.topicName = strdup(pTopicEp->topic);
+    tstrncpy(topic.db, pTopicEp->db, TSDB_DB_FNAME_LEN);
+
+    tscDebug("consumer %ld update topic: %s", tmq->consumerId, topic.topicName);
+
+    int32_t vgNumGet = taosArrayGetSize(pTopicEp->vgs);
+    topic.vgs = taosArrayInit(vgNumGet, sizeof(SMqClientVg));
+    for (int32_t j = 0; j < vgNumGet; j++) {
+      SMqSubVgEp* pVgEp = taosArrayGet(pTopicEp->vgs, j);
+      sprintf(vgKey, "%s:%d", topic.topicName, pVgEp->vgId);
+      int64_t* pOffset = taosHashGet(pHash, vgKey, strlen(vgKey));
+      int64_t  offset = tmq->resetOffsetCfg;
+      if (pOffset != NULL) {
+        offset = *pOffset;
+      }
+
+      tscDebug("consumer %ld(epoch %d) offset of vg %d updated to %ld", tmq->consumerId, epoch, pVgEp->vgId, offset);
+      SMqClientVg clientVg = {
+          .pollCnt = 0,
+          .currentOffset = offset,
+          .vgId = pVgEp->vgId,
+          .epSet = pVgEp->epSet,
+          .vgStatus = TMQ_VG_STATUS__IDLE,
+          .vgSkipCnt = 0,
+      };
+      taosArrayPush(topic.vgs, &clientVg);
+      set = true;
+    }
+    taosArrayPush(newTopics, &topic);
+  }
+  if (tmq->clientTopics) taosArrayDestroy(tmq->clientTopics);
+  taosHashCleanup(pHash);
+  tmq->clientTopics = newTopics;
+
+  if (taosArrayGetSize(tmq->clientTopics) == 0)
+    atomic_store_8(&tmq->status, TMQ_CONSUMER_STATUS__NO_TOPIC);
+  else
+    atomic_store_8(&tmq->status, TMQ_CONSUMER_STATUS__READY);
+
+  atomic_store_32(&tmq->epoch, epoch);
+  return set;
+}
+
 bool tmqUpdateEp(tmq_t* tmq, int32_t epoch, SMqAskEpRsp* pRsp) {
   /*printf("call update ep %d\n", epoch);*/
   bool    set = false;

source/common/src/systable.c

@@ -91,8 +91,8 @@ static const SSysDbTableSchema userDBSchema[] = {
     {.name = "precision", .bytes = 2 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
     {.name = "single_stable_model", .bytes = 1, .type = TSDB_DATA_TYPE_BOOL},
     {.name = "status", .bytes = 10 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
-//    {.name = "schemaless", .bytes = 1, .type = TSDB_DATA_TYPE_BOOL},
+    //    {.name = "schemaless", .bytes = 1, .type = TSDB_DATA_TYPE_BOOL},
-    {.name = "retension", .bytes = 60 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
+    {.name = "retention", .bytes = 60 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
 
     // {.name = "update", .bytes = 1, .type = TSDB_DATA_TYPE_TINYINT},  // disable update
 };
@@ -137,7 +137,7 @@ static const SSysDbTableSchema streamSchema[] = {
     {.name = "target_table", .bytes = SYSTABLE_SCH_TABLE_NAME_LEN, .type = TSDB_DATA_TYPE_VARCHAR},
     {.name = "watermark", .bytes = 8, .type = TSDB_DATA_TYPE_BIGINT},
     {.name = "trigger", .bytes = 4, .type = TSDB_DATA_TYPE_INT},
-    };
+};
 
 static const SSysDbTableSchema userTblsSchema[] = {
     {.name = "table_name", .bytes = SYSTABLE_SCH_TABLE_NAME_LEN, .type = TSDB_DATA_TYPE_VARCHAR},
@@ -223,7 +223,9 @@ static const SSysDbTableSchema transSchema[] = {
     {.name = "db", .bytes = SYSTABLE_SCH_DB_NAME_LEN, .type = TSDB_DATA_TYPE_VARCHAR},
     {.name = "failed_times", .bytes = 4, .type = TSDB_DATA_TYPE_INT},
     {.name = "last_exec_time", .bytes = 8, .type = TSDB_DATA_TYPE_TIMESTAMP},
-    {.name = "last_action_info", .bytes = (TSDB_TRANS_ERROR_LEN - 1) + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
+    {.name = "last_action_info",
+     .bytes = (TSDB_TRANS_ERROR_LEN - 1) + VARSTR_HEADER_SIZE,
+     .type = TSDB_DATA_TYPE_VARCHAR},
 };
 
 static const SSysDbTableSchema configSchema[] = {
@@ -316,8 +318,6 @@ static const SSysDbTableSchema querySchema[] = {
     {.name = "sql", .bytes = TSDB_SHOW_SQL_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
 };
 
-
-
 static const SSysTableMeta perfsMeta[] = {
     {TSDB_PERFS_TABLE_CONNECTIONS, connectionsSchema, tListLen(connectionsSchema)},
     {TSDB_PERFS_TABLE_QUERIES, querySchema, tListLen(querySchema)},

source/common/src/tmsg.c

@@ -2667,7 +2667,7 @@ int32_t tSerializeSSTbHbRsp(void *buf, int32_t bufLen, SSTbHbRsp *pRsp) {
   }
 
   int32_t numOfIndex = taosArrayGetSize(pRsp->pIndexRsp);
-  if (tEncodeI32(&encoder, numOfIndex) < 0) return -1;  
+  if (tEncodeI32(&encoder, numOfIndex) < 0) return -1;
   for (int32_t i = 0; i < numOfIndex; ++i) {
     STableIndexRsp *pIndexRsp = taosArrayGet(pRsp->pIndexRsp, i);
     if (tEncodeCStr(&encoder, pIndexRsp->tbName) < 0) return -1;
@@ -2752,7 +2752,7 @@ int32_t tDeserializeSSTbHbRsp(void *buf, int32_t bufLen, SSTbHbRsp *pRsp) {
     }
     taosArrayPush(pRsp->pIndexRsp, &tableIndexRsp);
   }
-  
+
   tEndDecode(&decoder);
 
   tDecoderClear(&decoder);
@@ -3383,8 +3383,7 @@ int32_t tSerializeSKillQueryReq(void *buf, int32_t bufLen, SKillQueryReq *pReq)
   tEncoderInit(&encoder, buf, bufLen);
 
   if (tStartEncode(&encoder) < 0) return -1;
-  if (tEncodeI32(&encoder, pReq->connId) < 0) return -1;
+  if (tEncodeCStr(&encoder, pReq->queryStrId) < 0) return -1;
-  if (tEncodeI32(&encoder, pReq->queryId) < 0) return -1;
   tEndEncode(&encoder);
 
   int32_t tlen = encoder.pos;
@@ -3397,8 +3396,7 @@ int32_t tDeserializeSKillQueryReq(void *buf, int32_t bufLen, SKillQueryReq *pReq
   tDecoderInit(&decoder, buf, bufLen);
 
   if (tStartDecode(&decoder) < 0) return -1;
-  if (tDecodeI32(&decoder, &pReq->connId) < 0) return -1;
+  if (tDecodeCStrTo(&decoder, pReq->queryStrId) < 0) return -1;
-  if (tDecodeI32(&decoder, &pReq->queryId) < 0) return -1;
   tEndDecode(&decoder);
 
   tDecoderClear(&decoder);
@@ -4014,7 +4012,7 @@ int32_t tEncodeTSma(SEncoder *pCoder, const STSma *pSma) {
   return 0;
 }
 
-int32_t tDecodeTSma(SDecoder *pCoder, STSma *pSma) {
+int32_t tDecodeTSma(SDecoder *pCoder, STSma *pSma, bool deepCopy) {
   if (tDecodeI8(pCoder, &pSma->version) < 0) return -1;
   if (tDecodeI8(pCoder, &pSma->intervalUnit) < 0) return -1;
   if (tDecodeI8(pCoder, &pSma->slidingUnit) < 0) return -1;
@@ -4026,17 +4024,30 @@ int32_t tDecodeTSma(SDecoder *pCoder, STSma *pSma) {
   if (tDecodeI64(pCoder, &pSma->indexUid) < 0) return -1;
   if (tDecodeI64(pCoder, &pSma->tableUid) < 0) return -1;
   if (tDecodeI64(pCoder, &pSma->dstTbUid) < 0) return -1;
-  if (tDecodeCStr(pCoder, &pSma->dstTbName) < 0) return -1;
+  if (deepCopy) {
+    if (tDecodeCStrAlloc(pCoder, &pSma->dstTbName) < 0) return -1;
+  } else {
+    if (tDecodeCStr(pCoder, &pSma->dstTbName) < 0) return -1;
+  }
+
   if (tDecodeI64(pCoder, &pSma->interval) < 0) return -1;
   if (tDecodeI64(pCoder, &pSma->offset) < 0) return -1;
   if (tDecodeI64(pCoder, &pSma->sliding) < 0) return -1;
   if (pSma->exprLen > 0) {
-    if (tDecodeCStr(pCoder, &pSma->expr) < 0) return -1;
+    if (deepCopy) {
+      if (tDecodeCStrAlloc(pCoder, &pSma->expr) < 0) return -1;
+    } else {
+      if (tDecodeCStr(pCoder, &pSma->expr) < 0) return -1;
+    }
   } else {
     pSma->expr = NULL;
   }
   if (pSma->tagsFilterLen > 0) {
-    if (tDecodeCStr(pCoder, &pSma->tagsFilter) < 0) return -1;
+    if (deepCopy) {
+      if (tDecodeCStrAlloc(pCoder, &pSma->tagsFilter) < 0) return -1;
+    } else {
+      if (tDecodeCStr(pCoder, &pSma->tagsFilter) < 0) return -1;
+    }
   } else {
     pSma->tagsFilter = NULL;
   }
@@ -4059,7 +4070,7 @@ int32_t tEncodeSVCreateTSmaReq(SEncoder *pCoder, const SVCreateTSmaReq *pReq) {
 int32_t tDecodeSVCreateTSmaReq(SDecoder *pCoder, SVCreateTSmaReq *pReq) {
   if (tStartDecode(pCoder) < 0) return -1;
 
-  tDecodeTSma(pCoder, pReq);
+  tDecodeTSma(pCoder, pReq, false);
 
   tEndDecode(pCoder);
   return 0;
@@ -4893,4 +4904,3 @@ int32_t tDecodeSTqOffset(SDecoder *pDecoder, STqOffset *pOffset) {
   if (tDecodeCStrTo(pDecoder, pOffset->subKey) < 0) return -1;
   return 0;
 }
-

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

@@ -557,8 +557,8 @@ typedef struct {
   SVgObj         fixedSinkVg;
   int64_t        smaId;  // 0 for unused
   int8_t         trigger;
-  int32_t        triggerParam;
+  int64_t        triggerParam;
-  int64_t        waterMark;
+  int64_t        watermark;
   char*          sql;
   char*          physicalPlan;
   SArray*        tasks;  // SArray<SArray<SStreamTask>>

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

@@ -31,7 +31,7 @@ void        mndReleaseStream(SMnode *pMnode, SStreamObj *pStream);
 SSdbRaw *mndStreamActionEncode(SStreamObj *pStream);
 SSdbRow *mndStreamActionDecode(SSdbRaw *pRaw);
 
-int32_t mndAddStreamToTrans(SMnode *pMnode, SStreamObj *pStream, const char *ast, int8_t triggerType, int64_t watermark, STrans *pTrans);
+int32_t mndAddStreamToTrans(SMnode *pMnode, SStreamObj *pStream, const char *ast, STrans *pTrans);
 
 #ifdef __cplusplus
 }

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

@@ -183,12 +183,12 @@ static SSdbRow *mndDbActionDecode(SSdbRaw *pRaw) {
     pDb->cfg.pRetensions = taosArrayInit(pDb->cfg.numOfRetensions, sizeof(SRetention));
     if (pDb->cfg.pRetensions == NULL) goto _OVER;
     for (int32_t i = 0; i < pDb->cfg.numOfRetensions; ++i) {
-      SRetention retension = {0};
+      SRetention retention = {0};
-      SDB_GET_INT64(pRaw, dataPos, &retension.freq, _OVER)
+      SDB_GET_INT64(pRaw, dataPos, &retention.freq, _OVER)
-      SDB_GET_INT64(pRaw, dataPos, &retension.keep, _OVER)
+      SDB_GET_INT64(pRaw, dataPos, &retention.keep, _OVER)
-      SDB_GET_INT8(pRaw, dataPos, &retension.freqUnit, _OVER)
+      SDB_GET_INT8(pRaw, dataPos, &retention.freqUnit, _OVER)
-      SDB_GET_INT8(pRaw, dataPos, &retension.keepUnit, _OVER)
+      SDB_GET_INT8(pRaw, dataPos, &retention.keepUnit, _OVER)
-      if (taosArrayPush(pDb->cfg.pRetensions, &retension) == NULL) {
+      if (taosArrayPush(pDb->cfg.pRetensions, &retention) == NULL) {
         goto _OVER;
       }
     }
@@ -472,7 +472,7 @@ static int32_t mndCreateDb(SMnode *pMnode, SRpcMsg *pReq, SCreateDbReq *pCreate,
   }
 
   int32_t code = -1;
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_DB, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_DB, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to create db:%s", pTrans->id, pCreate->db);
@@ -1352,7 +1352,7 @@ static void dumpDbInfoData(SSDataBlock *pBlock, SDbObj *pDb, SShowObj *pShow, in
   char *status = "ready";
   if (objStatus == SDB_STATUS_CREATING) status = "creating";
   if (objStatus == SDB_STATUS_DROPPING) status = "dropping";
-  char  statusB[24] = {0};
+  char statusB[24] = {0};
   STR_WITH_SIZE_TO_VARSTR(statusB, status, strlen(status));
 
   if (sysDb) {

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

@@ -33,8 +33,8 @@ int32_t tEncodeSStreamObj(SEncoder *pEncoder, const SStreamObj *pObj) {
   if (tEncodeI8(pEncoder, pObj->status) < 0) return -1;
   if (tEncodeI8(pEncoder, pObj->createdBy) < 0) return -1;
   if (tEncodeI8(pEncoder, pObj->trigger) < 0) return -1;
-  if (tEncodeI32(pEncoder, pObj->triggerParam) < 0) return -1;
+  if (tEncodeI64(pEncoder, pObj->triggerParam) < 0) return -1;
-  if (tEncodeI64(pEncoder, pObj->waterMark) < 0) return -1;
+  if (tEncodeI64(pEncoder, pObj->watermark) < 0) return -1;
   if (tEncodeI32(pEncoder, pObj->fixedSinkVgId) < 0) return -1;
   if (tEncodeI64(pEncoder, pObj->smaId) < 0) return -1;
   if (tEncodeCStr(pEncoder, pObj->sql) < 0) return -1;
@@ -85,8 +85,8 @@ int32_t tDecodeSStreamObj(SDecoder *pDecoder, SStreamObj *pObj) {
   if (tDecodeI8(pDecoder, &pObj->status) < 0) return -1;
   if (tDecodeI8(pDecoder, &pObj->createdBy) < 0) return -1;
   if (tDecodeI8(pDecoder, &pObj->trigger) < 0) return -1;
-  if (tDecodeI32(pDecoder, &pObj->triggerParam) < 0) return -1;
+  if (tDecodeI64(pDecoder, &pObj->triggerParam) < 0) return -1;
-  if (tDecodeI64(pDecoder, &pObj->waterMark) < 0) return -1;
+  if (tDecodeI64(pDecoder, &pObj->watermark) < 0) return -1;
   if (tDecodeI32(pDecoder, &pObj->fixedSinkVgId) < 0) return -1;
   if (tDecodeI64(pDecoder, &pObj->smaId) < 0) return -1;
   if (tDecodeCStrAlloc(pDecoder, &pObj->sql) < 0) return -1;

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

@@ -380,17 +380,19 @@ void mndStop(SMnode *pMnode) {
 int32_t mndProcessSyncMsg(SRpcMsg *pMsg) {
   SMnode    *pMnode = pMsg->info.node;
   SSyncMgmt *pMgmt = &pMnode->syncMgmt;
-  int32_t    code = TAOS_SYNC_OTHER_ERROR;
+  int32_t    code = 0;
 
   if (!syncEnvIsStart()) {
     mError("failed to process sync msg:%p type:%s since syncEnv stop", pMsg, TMSG_INFO(pMsg->msgType));
-    return TAOS_SYNC_OTHER_ERROR;
+    terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
+    return -1;
   }
 
   SSyncNode *pSyncNode = syncNodeAcquire(pMgmt->sync);
   if (pSyncNode == NULL) {
     mError("failed to process sync msg:%p type:%s since syncNode is null", pMsg, TMSG_INFO(pMsg->msgType));
-    return TAOS_SYNC_OTHER_ERROR;
+    terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
+    return -1;
   }
 
   char  logBuf[512] = {0};
@@ -451,7 +453,7 @@ int32_t mndProcessSyncMsg(SRpcMsg *pMsg) {
       tmsgSendRsp(&rsp);
     } else {
       mError("failed to process msg:%p since invalid type:%s", pMsg, TMSG_INFO(pMsg->msgType));
-      code = TAOS_SYNC_OTHER_ERROR;
+      code = -1;
     }
   } else {
     if (pMsg->msgType == TDMT_SYNC_TIMEOUT) {
@@ -492,10 +494,13 @@ int32_t mndProcessSyncMsg(SRpcMsg *pMsg) {
       tmsgSendRsp(&rsp);
     } else {
       mError("failed to process msg:%p since invalid type:%s", pMsg, TMSG_INFO(pMsg->msgType));
-      code = TAOS_SYNC_OTHER_ERROR;
+      code = -1;
     }
   }
 
+  if (code != 0) {
+    terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
+  }
   return code;
 }
 

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

@@ -387,6 +387,9 @@ int32_t mndScheduleStream(SMnode* pMnode, STrans* pTrans, SStreamObj* pStream) {
       // input
       pFinalTask->inputType = TASK_INPUT_TYPE__DATA_BLOCK;
 
+      // trigger
+      pFinalTask->triggerParam = pStream->triggerParam;
+
       // dispatch
       if (mndAddDispatcherToInnerTask(pMnode, pTrans, pStream, pFinalTask) < 0) {
         qDestroyQueryPlan(pPlan);

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

@@ -44,6 +44,7 @@ static int32_t  mndProcessGetSmaReq(SRpcMsg *pReq);
 static int32_t  mndProcessGetTbSmaReq(SRpcMsg *pReq);
 static int32_t  mndRetrieveSma(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBlock, int32_t rows);
 static void     mndCancelGetNextSma(SMnode *pMnode, void *pIter);
+static void     mndDestroySmaObj(SSmaObj *pSmaObj);
 
 int32_t mndInitSma(SMnode *pMnode) {
   SSdbTable table = {
@@ -390,7 +391,9 @@ static int32_t mndSetUpdateSmaStbCommitLogs(SMnode *pMnode, STrans *pTrans, SStb
   taosRLockLatch(&pStb->lock);
   memcpy(&stbObj, pStb, sizeof(SStbObj));
   taosRUnLockLatch(&pStb->lock);
+  stbObj.numOfColumns = 0;
   stbObj.pColumns = NULL;
+  stbObj.numOfTags = 0;
   stbObj.pTags = NULL;
   stbObj.updateTime = taosGetTimestampMs();
   stbObj.lock = 0;
@@ -404,6 +407,7 @@ static int32_t mndSetUpdateSmaStbCommitLogs(SMnode *pMnode, STrans *pTrans, SStb
   return 0;
 }
 
+#if 0
 static int32_t mndSetCreateSmaRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SSmaObj *pSma) {
   SSdb   *pSdb = pMnode->pSdb;
   SVgObj *pVgroup = NULL;
@@ -442,6 +446,7 @@ static int32_t mndSetCreateSmaRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj
 
   return 0;
 }
+#endif
 
 static int32_t mndSetCreateSmaVgroupRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup,
                                                 SSmaObj *pSma) {
@@ -501,6 +506,13 @@ static int32_t mndSetCreateSmaVgroupRedoActions(SMnode *pMnode, STrans *pTrans,
   return 0;
 }
 
+static void mndDestroySmaObj(SSmaObj *pSmaObj) {
+  if (pSmaObj) {
+    taosMemoryFreeClear(pSmaObj->schemaRow.pSchema);
+    taosMemoryFreeClear(pSmaObj->schemaTag.pSchema);
+  }
+}
+
 static int32_t mndCreateSma(SMnode *pMnode, SRpcMsg *pReq, SMCreateSmaReq *pCreate, SDbObj *pDb, SStbObj *pStb) {
   SSmaObj smaObj = {0};
   memcpy(smaObj.name, pCreate->name, TSDB_TABLE_FNAME_LEN);
@@ -524,29 +536,17 @@ static int32_t mndCreateSma(SMnode *pMnode, SRpcMsg *pReq, SMCreateSmaReq *pCrea
   smaObj.tagsFilterLen = pCreate->tagsFilterLen;
   smaObj.sqlLen = pCreate->sqlLen;
   smaObj.astLen = pCreate->astLen;
-
   if (smaObj.exprLen > 0) {
-    smaObj.expr = taosMemoryMalloc(smaObj.exprLen);
+    smaObj.expr = pCreate->expr;
-    if (smaObj.expr == NULL) goto _OVER;
-    memcpy(smaObj.expr, pCreate->expr, smaObj.exprLen);
   }
-
   if (smaObj.tagsFilterLen > 0) {
-    smaObj.tagsFilter = taosMemoryMalloc(smaObj.tagsFilterLen);
+    smaObj.tagsFilter = pCreate->tagsFilter;
-    if (smaObj.tagsFilter == NULL) goto _OVER;
-    memcpy(smaObj.tagsFilter, pCreate->tagsFilter, smaObj.tagsFilterLen);
   }
-
   if (smaObj.sqlLen > 0) {
-    smaObj.sql = taosMemoryMalloc(smaObj.sqlLen);
+    smaObj.sql = pCreate->sql;
-    if (smaObj.sql == NULL) goto _OVER;
-    memcpy(smaObj.sql, pCreate->sql, smaObj.sqlLen);
   }
-
   if (smaObj.astLen > 0) {
-    smaObj.ast = taosMemoryMalloc(smaObj.astLen);
+    smaObj.ast = pCreate->ast;
-    if (smaObj.ast == NULL) goto _OVER;
-    memcpy(smaObj.ast, pCreate->ast, smaObj.astLen);
   }
 
   SStreamObj streamObj = {0};
@@ -561,6 +561,8 @@ static int32_t mndCreateSma(SMnode *pMnode, SRpcMsg *pReq, SMCreateSmaReq *pCrea
   streamObj.sql = pCreate->sql;
   streamObj.createdBy = STREAM_CREATED_BY__SMA;
   streamObj.smaId = smaObj.uid;
+  streamObj.watermark = 0;
+  streamObj.trigger = STREAM_TRIGGER_AT_ONCE;
 
   if (mndAllocSmaVgroup(pMnode, pDb, &streamObj.fixedSinkVg) != 0) {
     mError("sma:%s, failed to create since %s", smaObj.name, terrstr());
@@ -581,14 +583,15 @@ static int32_t mndCreateSma(SMnode *pMnode, SRpcMsg *pReq, SMCreateSmaReq *pCrea
   if (mndSetCreateSmaCommitLogs(pMnode, pTrans, &smaObj) != 0) goto _OVER;
   if (mndSetCreateSmaVgroupCommitLogs(pMnode, pTrans, &streamObj.fixedSinkVg) != 0) goto _OVER;
   if (mndSetUpdateSmaStbCommitLogs(pMnode, pTrans, pStb) != 0) goto _OVER;
-  if (mndSetCreateSmaRedoActions(pMnode, pTrans, pDb, &smaObj) != 0) goto _OVER;
+  // if (mndSetCreateSmaRedoActions(pMnode, pTrans, pDb, &smaObj) != 0) goto _OVER;
   if (mndSetCreateSmaVgroupRedoActions(pMnode, pTrans, pDb, &streamObj.fixedSinkVg, &smaObj) != 0) goto _OVER;
-  if (mndAddStreamToTrans(pMnode, &streamObj, pCreate->ast, STREAM_TRIGGER_AT_ONCE, 0, pTrans) != 0) goto _OVER;
+  if (mndAddStreamToTrans(pMnode, &streamObj, pCreate->ast, pTrans) != 0) goto _OVER;
   if (mndTransPrepare(pMnode, pTrans) != 0) goto _OVER;
 
   code = 0;
 
 _OVER:
+  mndDestroySmaObj(&smaObj);
   mndTransDrop(pTrans);
   return code;
 }
@@ -728,6 +731,7 @@ static int32_t mndSetDropSmaVgroupCommitLogs(SMnode *pMnode, STrans *pTrans, SVg
   return 0;
 }
 
+#if 0
 static int32_t mndSetDropSmaRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SSmaObj *pSma) {
   SSdb   *pSdb = pMnode->pSdb;
   SVgObj *pVgroup = NULL;
@@ -768,6 +772,7 @@ static int32_t mndSetDropSmaRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *
 
   return 0;
 }
+#endif
 
 static int32_t mndSetDropSmaVgroupRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup) {
   SVnodeGid *pVgid = pVgroup->vnodeGid + 0;
@@ -818,7 +823,7 @@ static int32_t mndDropSma(SMnode *pMnode, SRpcMsg *pReq, SDbObj *pDb, SSmaObj *p
   if (mndSetDropSmaCommitLogs(pMnode, pTrans, pSma) != 0) goto _OVER;
   if (mndSetDropSmaVgroupCommitLogs(pMnode, pTrans, pVgroup) != 0) goto _OVER;
   if (mndSetUpdateSmaStbCommitLogs(pMnode, pTrans, pStb) != 0) goto _OVER;
-  if (mndSetDropSmaRedoActions(pMnode, pTrans, pDb, pSma) != 0) goto _OVER;
+  // if (mndSetDropSmaRedoActions(pMnode, pTrans, pDb, pSma) != 0) goto _OVER;
   if (mndSetDropSmaVgroupRedoActions(pMnode, pTrans, pDb, pVgroup) != 0) goto _OVER;
   if (mndTransPrepare(pMnode, pTrans) != 0) goto _OVER;
 
@@ -848,7 +853,7 @@ int32_t mndDropSmasByStb(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SStbObj *p
       if (mndSetDropSmaVgroupCommitLogs(pMnode, pTrans, pVgroup) != 0) goto _OVER;
       if (mndSetDropSmaVgroupRedoActions(pMnode, pTrans, pDb, pVgroup) != 0) goto _OVER;
       if (mndSetDropSmaCommitLogs(pMnode, pTrans, pSma) != 0) goto _OVER;
-      if (mndSetDropSmaRedoActions(pMnode, pTrans, pDb, pSma) != 0) goto _OVER;
+      // if (mndSetDropSmaRedoActions(pMnode, pTrans, pDb, pSma) != 0) goto _OVER;
       mndReleaseVgroup(pMnode, pVgroup);
       pVgroup = NULL;
     }
@@ -996,7 +1001,6 @@ int32_t mndGetTableSma(SMnode *pMnode, char *tbFName, STableIndexRsp *rsp, bool
   rsp->suid = pStb->uid;
   rsp->version = pStb->smaVer;
   mndReleaseStb(pMnode, pStb);
-  
 
   while (1) {
     pIter = sdbFetch(pSdb, SDB_SMA, pIter, (void **)&pSma);

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

@@ -323,10 +323,14 @@ static int32_t mndStbActionUpdate(SSdb *pSdb, SStbObj *pOld, SStbObj *pNew) {
   pOld->smaVer = pNew->smaVer;
   pOld->nextColId = pNew->nextColId;
   pOld->ttl = pNew->ttl;
-  pOld->numOfColumns = pNew->numOfColumns;
+  if (pNew->numOfColumns > 0) {
-  pOld->numOfTags = pNew->numOfTags;
+    pOld->numOfColumns = pNew->numOfColumns;
-  memcpy(pOld->pColumns, pNew->pColumns, pOld->numOfColumns * sizeof(SSchema));
+    memcpy(pOld->pColumns, pNew->pColumns, pOld->numOfColumns * sizeof(SSchema));
-  memcpy(pOld->pTags, pNew->pTags, pOld->numOfTags * sizeof(SSchema));
+  }
+  if (pNew->numOfTags > 0) {
+    pOld->numOfTags = pNew->numOfTags;
+    memcpy(pOld->pTags, pNew->pTags, pOld->numOfTags * sizeof(SSchema));
+  }
   if (pNew->commentLen != 0) {
     memcpy(pOld->comment, pNew->comment, pNew->commentLen);
   }

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

@@ -228,23 +228,22 @@ static int32_t mndStreamGetPlanString(const char *ast, int8_t triggerType, int64
         .pAstRoot = pAst,
         .topicQuery = false,
         .streamQuery = true,
-        .triggerType = triggerType,
+        .triggerType = triggerType == STREAM_TRIGGER_MAX_DELAY ? STREAM_TRIGGER_WINDOW_CLOSE : triggerType,
         .watermark = watermark,
     };
     code = qCreateQueryPlan(&cxt, &pPlan, NULL);
   }
 
   if (TSDB_CODE_SUCCESS == code) {
-    code = nodesNodeToString((SNode*)pPlan, false, pStr, NULL);
+    code = nodesNodeToString((SNode *)pPlan, false, pStr, NULL);
   }
   nodesDestroyNode(pAst);
-  nodesDestroyNode((SNode*)pPlan);
+  nodesDestroyNode((SNode *)pPlan);
   terrno = code;
   return code;
 }
 
-int32_t mndAddStreamToTrans(SMnode *pMnode, SStreamObj *pStream, const char *ast, int8_t triggerType, int64_t watermark,
+int32_t mndAddStreamToTrans(SMnode *pMnode, SStreamObj *pStream, const char *ast, STrans *pTrans) {
-                            STrans *pTrans) {
   SNode *pAst = NULL;
 
   if (nodesStringToNode(ast, &pAst) < 0) {
@@ -258,7 +257,6 @@ int32_t mndAddStreamToTrans(SMnode *pMnode, SStreamObj *pStream, const char *ast
   // free
   nodesDestroyNode(pAst);
 
-
 #if 0
   printf("|");
   for (int i = 0; i < pStream->outputSchema.nCols; i++) {
@@ -268,7 +266,7 @@ int32_t mndAddStreamToTrans(SMnode *pMnode, SStreamObj *pStream, const char *ast
 
 #endif
 
-  if (TSDB_CODE_SUCCESS != mndStreamGetPlanString(ast, triggerType, watermark, &pStream->physicalPlan)) {
+  if (TSDB_CODE_SUCCESS != mndStreamGetPlanString(ast, pStream->trigger, pStream->watermark, &pStream->physicalPlan)) {
     mError("topic:%s, failed to get plan since %s", pStream->name, terrstr());
     return -1;
   }
@@ -385,7 +383,8 @@ static int32_t mndCreateStream(SMnode *pMnode, SRpcMsg *pReq, SCMCreateStreamReq
   streamObj.smaId = 0;
   /*streamObj.physicalPlan = "";*/
   streamObj.trigger = pCreate->triggerType;
-  streamObj.waterMark = pCreate->watermark;
+  streamObj.watermark = pCreate->watermark;
+  streamObj.triggerParam = pCreate->maxDelay;
 
   if (streamObj.targetSTbName[0]) {
     pDb = mndAcquireDbByStb(pMnode, streamObj.targetSTbName);
@@ -403,7 +402,7 @@ static int32_t mndCreateStream(SMnode *pMnode, SRpcMsg *pReq, SCMCreateStreamReq
   }
   mDebug("trans:%d, used to create stream:%s", pTrans->id, pCreate->name);
 
-  if (mndAddStreamToTrans(pMnode, &streamObj, pCreate->ast, pCreate->triggerType, pCreate->watermark, pTrans) != 0) {
+  if (mndAddStreamToTrans(pMnode, &streamObj, pCreate->ast, pTrans) != 0) {
     mError("trans:%d, failed to add stream since %s", pTrans->id, terrstr());
     mndTransDrop(pTrans);
     return -1;
@@ -553,7 +552,7 @@ static int32_t mndRetrieveStream(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pB
     colDataAppend(pColInfo, numOfRows, (const char *)&pStream->targetSTbName, true);
 
     pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
-    colDataAppend(pColInfo, numOfRows, (const char *)&pStream->waterMark, false);
+    colDataAppend(pColInfo, numOfRows, (const char *)&pStream->watermark, false);
 
     pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
     colDataAppend(pColInfo, numOfRows, (const char *)&pStream->trigger, false);

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

@@ -234,9 +234,9 @@ int32_t mndSyncPropose(SMnode *pMnode, SSdbRaw *pRaw, int32_t transId) {
   int32_t    code = syncPropose(pMgmt->sync, &rsp, isWeak);
   if (code == 0) {
     tsem_wait(&pMgmt->syncSem);
-  } else if (code == TAOS_SYNC_PROPOSE_NOT_LEADER) {
+  } else if (code == -1 && terrno == TSDB_CODE_SYN_NOT_LEADER) {
     terrno = TSDB_CODE_APP_NOT_READY;
-  } else if (code == TAOS_SYNC_OTHER_ERROR) {
+  } else if (code == -1 && terrno == TSDB_CODE_SYN_INTERNAL_ERROR) {
     terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
   } else {
     terrno = TSDB_CODE_APP_ERROR;
@@ -257,13 +257,13 @@ void mndSyncStart(SMnode *pMnode) {
   syncStart(pMgmt->sync);
   mDebug("mnode sync started, id:%" PRId64 " standby:%d", pMgmt->sync, pMgmt->standby);
 
-/*
+  /*
-  if (pMgmt->standby) {
+    if (pMgmt->standby) {
-    syncStartStandBy(pMgmt->sync);
+      syncStartStandBy(pMgmt->sync);
-  } else {
+    } else {
-    syncStart(pMgmt->sync);
+      syncStart(pMgmt->sync);
-  }
+    }
-*/  
+  */
 }
 
 void mndSyncStop(SMnode *pMnode) {}

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

@@ -1347,13 +1347,11 @@ int32_t mndKillTrans(SMnode *pMnode, STrans *pTrans) {
 
   for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {
     STransAction *pAction = taosArrayGet(pArray, i);
-    if (pAction->errCode != 0) {
+    mInfo("trans:%d, %s:%d set processed for kill msg received, errCode from %s to success", pTrans->id,
-      mInfo("trans:%d, %s:%d set processed for kill msg received, errCode from %s to success", pTrans->id,
+          mndTransStr(pAction->stage), i, tstrerror(pAction->errCode));
-            mndTransStr(pAction->stage), i, tstrerror(pAction->errCode));
+    pAction->msgSent = 1;
-      pAction->msgSent = 1;
+    pAction->msgReceived = 1;
-      pAction->msgReceived = 1;
+    pAction->errCode = 0;
-      pAction->errCode = 0;
-    }
   }
 
   mndTransExecute(pMnode, pTrans);

source/dnode/vnode/CMakeLists.txt

@@ -31,7 +31,7 @@ target_sources(
     "src/sma/smaEnv.c"
     "src/sma/smaOpen.c"
     "src/sma/smaRollup.c"
-    "src/sma/smaTimeRange2.c"
+    "src/sma/smaTimeRange.c"
 
     # tsdb
     "src/tsdb/tsdbCommit.c"

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

@@ -38,8 +38,6 @@ typedef struct SSmaStatItem SSmaStatItem;
 typedef struct SSmaKey      SSmaKey;
 typedef struct SRSmaInfo    SRSmaInfo;
 
-#define SMA_IVLD_FID INT_MIN
-
 struct SSmaEnv {
   TdThreadRwlock lock;
   int8_t         type;
@@ -49,45 +47,38 @@ struct SSmaEnv {
 #define SMA_ENV_LOCK(env)       ((env)->lock)
 #define SMA_ENV_TYPE(env)       ((env)->type)
 #define SMA_ENV_STAT(env)       ((env)->pStat)
-#define SMA_ENV_STAT_ITEMS(env) ((env)->pStat->smaStatItems)
+#define SMA_ENV_STAT_ITEM(env) ((env)->pStat->tsmaStatItem)
 
 struct SSmaStatItem {
-  int8_t state;  // ETsdbSmaStat
+  int8_t    state;  // ETsdbSmaStat
-  STSma *pTSma;  // cache schema
+  STSma    *pTSma;  // cache schema
+  STSchema *pTSchema;
 };
 
 struct SSmaStat {
   union {
-    SHashObj *smaStatItems;  // key: indexUid, value: SSmaStatItem for tsma
+    SSmaStatItem tsmaStatItem;
-    SHashObj *rsmaInfoHash;  // key: stbUid, value: SRSmaInfo;
+    SHashObj    *rsmaInfoHash;  // key: stbUid, value: SRSmaInfo;
   };
   T_REF_DECLARE()
 };
-#define SMA_STAT_ITEMS(s)     ((s)->smaStatItems)
+#define SMA_STAT_ITEM(s)      ((s)->tsmaStatItem)
 #define SMA_STAT_INFO_HASH(s) ((s)->rsmaInfoHash)
 
 void  tdDestroySmaEnv(SSmaEnv *pSmaEnv);
 void *tdFreeSmaEnv(SSmaEnv *pSmaEnv);
-#if 0
-int32_t tbGetTSmaStatus(SSma *pSma, STSma *param, void *result);
-int32_t tbRemoveTSmaData(SSma *pSma, STSma *param, STimeWindow *pWin);
-#endif
 
-int32_t tdInitSma(SSma *pSma);
 int32_t tdDropTSma(SSma *pSma, char *pMsg);
 int32_t tdDropTSmaData(SSma *pSma, int64_t indexUid);
 int32_t tdInsertRSmaData(SSma *pSma, char *msg);
 
 int32_t tdRefSmaStat(SSma *pSma, SSmaStat *pStat);
 int32_t tdUnRefSmaStat(SSma *pSma, SSmaStat *pStat);
-int32_t tdCheckAndInitSmaEnv(SSma *pSma, int8_t smaType, bool onlyCheck);
+int32_t tdCheckAndInitSmaEnv(SSma *pSma, int8_t smaType);
 
 int32_t tdLockSma(SSma *pSma);
 int32_t tdUnLockSma(SSma *pSma);
 
-static FORCE_INLINE int16_t tdTSmaAdd(SSma *pSma, int16_t n) { return atomic_add_fetch_16(&SMA_TSMA_NUM(pSma), n); }
-static FORCE_INLINE int16_t tdTSmaSub(SSma *pSma, int16_t n) { return atomic_sub_fetch_16(&SMA_TSMA_NUM(pSma), n); }
-
 static FORCE_INLINE int32_t tdRLockSmaEnv(SSmaEnv *pEnv) {
   int code = taosThreadRwlockRdlock(&(pEnv->lock));
   if (code != 0) {
@@ -160,11 +151,10 @@ static FORCE_INLINE void tdSmaStatSetDropped(SSmaStatItem *pStatItem) {
   }
 }
 
-static int32_t  tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType);
+static int32_t tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType);
-void           *tdFreeSmaStatItem(SSmaStatItem *pSmaStatItem);
+void          *tdFreeSmaStatItem(SSmaStatItem *pSmaStatItem);
-static int32_t  tdDestroySmaState(SSmaStat *pSmaStat, int8_t smaType);
+static int32_t tdDestroySmaState(SSmaStat *pSmaStat, int8_t smaType);
-static SSmaEnv *tdNewSmaEnv(const SSma *pSma, int8_t smaType, const char *path, SDiskID did);
+void          *tdFreeSmaState(SSmaStat *pSmaStat, int8_t smaType);
-static int32_t  tdInitSmaEnv(SSma *pSma, int8_t smaType, const char *path, SDiskID did, SSmaEnv **pEnv);
 
 void *tdFreeRSmaInfo(SRSmaInfo *pInfo);
 

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

@@ -147,6 +147,9 @@ int32_t tqProcessTaskRecoverReq(STQ* pTq, SRpcMsg* pMsg);
 int32_t tqProcessTaskDispatchRsp(STQ* pTq, SRpcMsg* pMsg);
 int32_t tqProcessTaskRecoverRsp(STQ* pTq, SRpcMsg* pMsg);
 
+SSubmitReq* tdBlockToSubmit(const SArray* pBlocks, const STSchema* pSchema, bool createTb, int64_t suid,
+                            const char* stbFullName, int32_t vgId);
+
 // sma
 int32_t smaOpen(SVnode* pVnode);
 int32_t smaClose(SSma* pSma);
@@ -245,7 +248,6 @@ struct STbUidStore {
 };
 
 struct SSma {
-  int16_t       nTSma;
   bool          locked;
   TdThreadMutex mutex;
   SVnode*       pVnode;
@@ -261,7 +263,6 @@ struct SSma {
 #define SMA_META(s)       ((s)->pVnode->pMeta)
 #define SMA_VID(s)        TD_VID((s)->pVnode)
 #define SMA_TFS(s)        ((s)->pVnode->pTfs)
-#define SMA_TSMA_NUM(s)   ((s)->nTSma)
 #define SMA_TSMA_ENV(s)   ((s)->pTSmaEnv)
 #define SMA_RSMA_ENV(s)   ((s)->pRSmaEnv)
 #define SMA_RSMA_TSDB0(s) ((s)->pVnode->pTsdb)

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

@@ -75,7 +75,7 @@ int metaDecodeEntry(SDecoder *pCoder, SMetaEntry *pME) {
       terrno = TSDB_CODE_OUT_OF_MEMORY;
       return -1;
     }
-    if (tDecodeTSma(pCoder, pME->smaEntry.tsma) < 0) return -1;
+    if (tDecodeTSma(pCoder, pME->smaEntry.tsma, true) < 0) return -1;
   } else {
     ASSERT(0);
   }

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

@@ -44,3 +44,209 @@ int32_t smaGetTSmaDays(SVnodeCfg* pCfg, void* pCont, uint32_t contLen, int32_t*
   smaDebug("vgId:%d, get tsma days %d", pCfg->vgId, *days);
   return code;
 }
+
+#if 0
+
+/**
+ * @brief TODO: Assume that the final generated result it less than 3M
+ *
+ * @param pReq
+ * @param pDataBlocks
+ * @param vgId
+ * @param suid  // TODO: check with Liao whether suid response is reasonable
+ *
+ * TODO: colId should be set
+ */
+int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SArray* pDataBlocks, STSchema* pTSchema, int32_t vgId,
+                                    tb_uid_t suid, const char* stbName, bool isCreateCtb) {
+  int32_t sz = taosArrayGetSize(pDataBlocks);
+  int32_t bufSize = sizeof(SSubmitReq);
+  for (int32_t i = 0; i < sz; ++i) {
+    SDataBlockInfo* pBlkInfo = &((SSDataBlock*)taosArrayGet(pDataBlocks, i))->info;
+    bufSize += pBlkInfo->rows * (TD_ROW_HEAD_LEN + pBlkInfo->rowSize + BitmapLen(pBlkInfo->numOfCols));
+    bufSize += sizeof(SSubmitBlk);
+  }
+
+  *pReq = taosMemoryCalloc(1, bufSize);
+  if (!(*pReq)) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    return TSDB_CODE_FAILED;
+  }
+  void* pDataBuf = *pReq;
+
+  SArray*     pTagArray = NULL;
+  int32_t     msgLen = sizeof(SSubmitReq);
+  int32_t     numOfBlks = 0;
+  int32_t     schemaLen = 0;
+  SRowBuilder rb = {0};
+  tdSRowInit(&rb, pTSchema->version);
+
+  for (int32_t i = 0; i < sz; ++i) {
+    SSDataBlock*    pDataBlock = taosArrayGet(pDataBlocks, i);
+    SDataBlockInfo* pDataBlkInfo = &pDataBlock->info;
+    int32_t         colNum = pDataBlkInfo->numOfCols;
+    int32_t         rows = pDataBlkInfo->rows;
+    int32_t         rowSize = pDataBlkInfo->rowSize;
+    int64_t         groupId = pDataBlkInfo->groupId;
+
+    if (rb.nCols != colNum) {
+      tdSRowSetTpInfo(&rb, colNum, pTSchema->flen);
+    }
+
+   if(isCreateCtb) {
+     SMetaReader  mr = {0};
+     const char*  ctbName = buildCtbNameByGroupId(stbName, pDataBlock->info.groupId);
+     if (metaGetTableEntryByName(&mr, ctbName) != 0) {
+       smaDebug("vgId:%d, no tsma ctb %s exists", vgId, ctbName);
+     }
+     SVCreateTbReq ctbReq = {0};
+     ctbReq.name = ctbName;
+     ctbReq.type = TSDB_CHILD_TABLE;
+     ctbReq.ctb.suid = suid;
+
+     STagVal tagVal = {.cid = colNum + PRIMARYKEY_TIMESTAMP_COL_ID,
+                       .type = TSDB_DATA_TYPE_BIGINT,
+                       .i64 = groupId};
+     STag*   pTag = NULL;
+     if(!pTagArray) {
+       pTagArray = taosArrayInit(1, sizeof(STagVal));
+       if (!pTagArray) goto _err;
+     }
+     taosArrayClear(pTagArray);
+     taosArrayPush(pTagArray, &tagVal);
+     tTagNew(pTagArray, 1, false, &pTag);
+     if (pTag == NULL) {
+       tdDestroySVCreateTbReq(&ctbReq);
+       goto _err;
+      }
+      ctbReq.ctb.pTag = (uint8_t*)pTag;
+
+      int32_t code;
+      tEncodeSize(tEncodeSVCreateTbReq, &ctbReq, schemaLen, code);
+
+      tdDestroySVCreateTbReq(&ctbReq);
+      if (code < 0) {
+        goto _err;
+      }
+   }
+
+
+
+    SSubmitBlk* pSubmitBlk = POINTER_SHIFT(pDataBuf, msgLen);
+    pSubmitBlk->suid = suid;
+    pSubmitBlk->uid = groupId;
+    pSubmitBlk->numOfRows = rows;
+
+    msgLen += sizeof(SSubmitBlk);
+    int32_t dataLen = 0;
+    for (int32_t j = 0; j < rows; ++j) {                     // iterate by row
+      tdSRowResetBuf(&rb, POINTER_SHIFT(pDataBuf, msgLen));  // set row buf
+      bool    isStartKey = false;
+      int32_t offset = 0;
+      for (int32_t k = 0; k < colNum; ++k) {  // iterate by column
+        SColumnInfoData* pColInfoData = taosArrayGet(pDataBlock->pDataBlock, k);
+        STColumn*        pCol = &pTSchema->columns[k];
+        void*            var = POINTER_SHIFT(pColInfoData->pData, j * pColInfoData->info.bytes);
+        switch (pColInfoData->info.type) {
+          case TSDB_DATA_TYPE_TIMESTAMP:
+            if (!isStartKey) {
+              isStartKey = true;
+              tdAppendColValToRow(&rb, PRIMARYKEY_TIMESTAMP_COL_ID, TSDB_DATA_TYPE_TIMESTAMP, TD_VTYPE_NORM, var, true,
+                                  offset, k);
+
+            } else {
+              tdAppendColValToRow(&rb, PRIMARYKEY_TIMESTAMP_COL_ID + k, TSDB_DATA_TYPE_TIMESTAMP, TD_VTYPE_NORM, var,
+                                  true, offset, k);
+            }
+            break;
+          case TSDB_DATA_TYPE_NCHAR: {
+            tdAppendColValToRow(&rb, PRIMARYKEY_TIMESTAMP_COL_ID + k, TSDB_DATA_TYPE_NCHAR, TD_VTYPE_NORM, var, true,
+                                offset, k);
+            break;
+          }
+          case TSDB_DATA_TYPE_VARCHAR: {  // TSDB_DATA_TYPE_BINARY
+            tdAppendColValToRow(&rb, PRIMARYKEY_TIMESTAMP_COL_ID + k, TSDB_DATA_TYPE_VARCHAR, TD_VTYPE_NORM, var, true,
+                                offset, k);
+            break;
+          }
+          case TSDB_DATA_TYPE_VARBINARY:
+          case TSDB_DATA_TYPE_DECIMAL:
+          case TSDB_DATA_TYPE_BLOB:
+          case TSDB_DATA_TYPE_JSON:
+          case TSDB_DATA_TYPE_MEDIUMBLOB:
+            uError("the column type %" PRIi16 " is defined but not implemented yet", pColInfoData->info.type);
+            TASSERT(0);
+            break;
+          default:
+            if (pColInfoData->info.type < TSDB_DATA_TYPE_MAX && pColInfoData->info.type > TSDB_DATA_TYPE_NULL) {
+              if (pCol->type == pColInfoData->info.type) {
+                tdAppendColValToRow(&rb, PRIMARYKEY_TIMESTAMP_COL_ID + k, pCol->type, TD_VTYPE_NORM, var, true, offset,
+                                    k);
+              } else {
+                char tv[8] = {0};
+                if (pColInfoData->info.type == TSDB_DATA_TYPE_FLOAT) {
+                  float v = 0;
+                  GET_TYPED_DATA(v, float, pColInfoData->info.type, var);
+                  SET_TYPED_DATA(&tv, pCol->type, v);
+                } else if (pColInfoData->info.type == TSDB_DATA_TYPE_DOUBLE) {
+                  double v = 0;
+                  GET_TYPED_DATA(v, double, pColInfoData->info.type, var);
+                  SET_TYPED_DATA(&tv, pCol->type, v);
+                } else if (IS_SIGNED_NUMERIC_TYPE(pColInfoData->info.type)) {
+                  int64_t v = 0;
+                  GET_TYPED_DATA(v, int64_t, pColInfoData->info.type, var);
+                  SET_TYPED_DATA(&tv, pCol->type, v);
+                } else {
+                  uint64_t v = 0;
+                  GET_TYPED_DATA(v, uint64_t, pColInfoData->info.type, var);
+                  SET_TYPED_DATA(&tv, pCol->type, v);
+                }
+                tdAppendColValToRow(&rb, PRIMARYKEY_TIMESTAMP_COL_ID + k, pCol->type, TD_VTYPE_NORM, tv, true, offset,
+                                    k);
+              }
+            } else {
+              uError("the column type %" PRIi16 " is undefined\n", pColInfoData->info.type);
+              TASSERT(0);
+            }
+            break;
+        }
+        offset += TYPE_BYTES[pCol->type];  // sum/avg would convert to int64_t/uint64_t/double during aggregation
+      }
+      dataLen += TD_ROW_LEN(rb.pBuf);
+#ifdef TD_DEBUG_PRINT_ROW
+      tdSRowPrint(rb.pBuf, pTSchema, __func__);
+#endif
+    }
+
+    ++numOfBlks;
+
+    pSubmitBlk->dataLen = dataLen;
+    msgLen += pSubmitBlk->dataLen;
+  }
+
+  (*pReq)->length = msgLen;
+
+  (*pReq)->header.vgId = htonl(vgId);
+  (*pReq)->header.contLen = htonl(msgLen);
+  (*pReq)->length = (*pReq)->header.contLen;
+  (*pReq)->numOfBlocks = htonl(numOfBlks);
+  SSubmitBlk* blk = (SSubmitBlk*)((*pReq) + 1);
+  while (numOfBlks--) {
+    int32_t dataLen = blk->dataLen;
+    blk->uid = htobe64(blk->uid);
+    blk->suid = htobe64(blk->suid);
+    blk->padding = htonl(blk->padding);
+    blk->sversion = htonl(blk->sversion);
+    blk->dataLen = htonl(blk->dataLen);
+    blk->schemaLen = htonl(blk->schemaLen);
+    blk->numOfRows = htons(blk->numOfRows);
+    blk = (SSubmitBlk*)(blk->data + dataLen);
+  }
+  return TSDB_CODE_SUCCESS;
+_err:
+  taosMemoryFreeClear(*pReq);
+  taosArrayDestroy(pTagArray);
+
+  return TSDB_CODE_FAILED;
+}
+#endif

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

@@ -17,123 +17,17 @@
 
 typedef struct SSmaStat SSmaStat;
 
-static const char *TSDB_SMA_DNAME[] = {
-    "",      // TSDB_SMA_TYPE_BLOCK
-    "tsma",  // TSDB_SMA_TYPE_TIME_RANGE
-    "rsma",  // TSDB_SMA_TYPE_ROLLUP
-};
-
-#define SMA_TEST_INDEX_NAME "smaTestIndexName"  // TODO: just for test
-#define SMA_TEST_INDEX_UID  2000000001          // TODO: just for test
-#define SMA_STATE_HASH_SLOT 4
-
 #define RSMA_TASK_INFO_HASH_SLOT 8
 
-typedef struct SPoolMem {
-  int64_t          size;
-  struct SPoolMem *prev;
-  struct SPoolMem *next;
-} SPoolMem;
-
 // declaration of static functions
 
-// insert data
+static int32_t  tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType);
-
+static SSmaEnv *tdNewSmaEnv(const SSma *pSma, int8_t smaType, const char *path);
-static void tdGetSmaDir(int32_t vgId, ETsdbSmaType smaType, char dirName[]);
+static int32_t  tdInitSmaEnv(SSma *pSma, int8_t smaType, const char *path, SSmaEnv **pEnv);
-
-// Pool Memory
-static SPoolMem *openPool();
-static void      clearPool(SPoolMem *pPool);
-static void      closePool(SPoolMem *pPool);
-static void     *poolMalloc(void *arg, size_t size);
-static void      poolFree(void *arg, void *ptr);
 
 // implementation
 
-static SPoolMem *openPool() {
+static SSmaEnv *tdNewSmaEnv(const SSma *pSma, int8_t smaType, const char *path) {
-  SPoolMem *pPool = (SPoolMem *)taosMemoryMalloc(sizeof(*pPool));
-
-  pPool->prev = pPool->next = pPool;
-  pPool->size = 0;
-
-  return pPool;
-}
-
-static void clearPool(SPoolMem *pPool) {
-  if (!pPool) return;
-
-  SPoolMem *pMem;
-
-  do {
-    pMem = pPool->next;
-
-    if (pMem == pPool) break;
-
-    pMem->next->prev = pMem->prev;
-    pMem->prev->next = pMem->next;
-    pPool->size -= pMem->size;
-
-    taosMemoryFree(pMem);
-  } while (1);
-
-  assert(pPool->size == 0);
-}
-
-static void closePool(SPoolMem *pPool) {
-  if (pPool) {
-    clearPool(pPool);
-    taosMemoryFree(pPool);
-  }
-}
-
-static void *poolMalloc(void *arg, size_t size) {
-  void     *ptr = NULL;
-  SPoolMem *pPool = (SPoolMem *)arg;
-  SPoolMem *pMem;
-
-  pMem = (SPoolMem *)taosMemoryMalloc(sizeof(*pMem) + size);
-  if (!pMem) {
-    assert(0);
-  }
-
-  pMem->size = sizeof(*pMem) + size;
-  pMem->next = pPool->next;
-  pMem->prev = pPool;
-
-  pPool->next->prev = pMem;
-  pPool->next = pMem;
-  pPool->size += pMem->size;
-
-  ptr = (void *)(&pMem[1]);
-  return ptr;
-}
-
-static void poolFree(void *arg, void *ptr) {
-  SPoolMem *pPool = (SPoolMem *)arg;
-  SPoolMem *pMem;
-
-  pMem = &(((SPoolMem *)ptr)[-1]);
-
-  pMem->next->prev = pMem->prev;
-  pMem->prev->next = pMem->next;
-  pPool->size -= pMem->size;
-
-  taosMemoryFree(pMem);
-}
-
-int32_t tdInitSma(SSma *pSma) {
-  int32_t numOfTSma = taosArrayGetSize(metaGetSmaTbUids(SMA_META(pSma)));
-  if (numOfTSma > 0) {
-    atomic_store_16(&SMA_TSMA_NUM(pSma), (int16_t)numOfTSma);
-  }
-  return TSDB_CODE_SUCCESS;
-}
-
-static void tdGetSmaDir(int32_t vgId, ETsdbSmaType smaType, char dirName[]) {
-  snprintf(dirName, TSDB_FILENAME_LEN, "vnode%svnode%d%s%s", TD_DIRSEP, vgId, TD_DIRSEP, TSDB_SMA_DNAME[smaType]);
-}
-
-static SSmaEnv *tdNewSmaEnv(const SSma *pSma, int8_t smaType, const char *path, SDiskID did) {
   SSmaEnv *pEnv = NULL;
 
   pEnv = (SSmaEnv *)taosMemoryCalloc(1, sizeof(SSmaEnv));
@@ -156,18 +50,17 @@ static SSmaEnv *tdNewSmaEnv(const SSma *pSma, int8_t smaType, const char *path,
     return NULL;
   }
 
-
   return pEnv;
 }
 
-static int32_t tdInitSmaEnv(SSma *pSma, int8_t smaType, const char *path, SDiskID did, SSmaEnv **pEnv) {
+static int32_t tdInitSmaEnv(SSma *pSma, int8_t smaType, const char *path, SSmaEnv **pEnv) {
   if (!pEnv) {
     terrno = TSDB_CODE_INVALID_PTR;
     return TSDB_CODE_FAILED;
   }
 
   if (!(*pEnv)) {
-    if (!(*pEnv = tdNewSmaEnv(pSma, smaType, path, did))) {
+    if (!(*pEnv = tdNewSmaEnv(pSma, smaType, path))) {
       return TSDB_CODE_FAILED;
     }
   }
@@ -183,15 +76,16 @@ static int32_t tdInitSmaEnv(SSma *pSma, int8_t smaType, const char *path, SDiskI
  */
 void tdDestroySmaEnv(SSmaEnv *pSmaEnv) {
   if (pSmaEnv) {
-    tdDestroySmaState(pSmaEnv->pStat, SMA_ENV_TYPE(pSmaEnv));
+    pSmaEnv->pStat = tdFreeSmaState(pSmaEnv->pStat, SMA_ENV_TYPE(pSmaEnv));
-    taosMemoryFreeClear(pSmaEnv->pStat);
     taosThreadRwlockDestroy(&(pSmaEnv->lock));
   }
 }
 
 void *tdFreeSmaEnv(SSmaEnv *pSmaEnv) {
-  tdDestroySmaEnv(pSmaEnv);
+  if (pSmaEnv) {
-  taosMemoryFreeClear(pSmaEnv);
+    tdDestroySmaEnv(pSmaEnv);
+    taosMemoryFreeClear(pSmaEnv);
+  }
   return NULL;
 }
 
@@ -239,13 +133,7 @@ static int32_t tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType) {
         return TSDB_CODE_FAILED;
       }
     } else if (smaType == TSDB_SMA_TYPE_TIME_RANGE) {
-      SMA_STAT_ITEMS(*pSmaStat) =
+      // TODO
-          taosHashInit(SMA_STATE_HASH_SLOT, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), true, HASH_ENTRY_LOCK);
-
-      if (!SMA_STAT_ITEMS(*pSmaStat)) {
-        taosMemoryFreeClear(*pSmaStat);
-        return TSDB_CODE_FAILED;
-      }
     } else {
       ASSERT(0);
     }
@@ -262,6 +150,12 @@ void *tdFreeSmaStatItem(SSmaStatItem *pSmaStatItem) {
   return NULL;
 }
 
+void* tdFreeSmaState(SSmaStat *pSmaStat, int8_t smaType) {
+  tdDestroySmaState(pSmaStat, smaType);
+  taosMemoryFreeClear(pSmaStat);
+  return NULL;
+}
+
 /**
  * @brief Release resources allocated for its member fields, not including itself.
  *
@@ -270,16 +164,10 @@ void *tdFreeSmaStatItem(SSmaStatItem *pSmaStatItem) {
  */
 int32_t tdDestroySmaState(SSmaStat *pSmaStat, int8_t smaType) {
   if (pSmaStat) {
-    // TODO: use taosHashSetFreeFp when taosHashSetFreeFp is ready.
     if (smaType == TSDB_SMA_TYPE_TIME_RANGE) {
-      void *item = taosHashIterate(SMA_STAT_ITEMS(pSmaStat), NULL);
+      tdFreeSmaStatItem(&pSmaStat->tsmaStatItem);
-      while (item) {
-        SSmaStatItem *pItem = *(SSmaStatItem **)item;
-        tdFreeSmaStatItem(pItem);
-        item = taosHashIterate(SMA_STAT_ITEMS(pSmaStat), item);
-      }
-      taosHashCleanup(SMA_STAT_ITEMS(pSmaStat));
     } else if (smaType == TSDB_SMA_TYPE_ROLLUP) {
+      // TODO: use taosHashSetFreeFp when taosHashSetFreeFp is ready.
       void *infoHash = taosHashIterate(SMA_STAT_INFO_HASH(pSmaStat), NULL);
       while (infoHash) {
         SRSmaInfo *pInfoHash = *(SRSmaInfo **)infoHash;
@@ -317,7 +205,7 @@ int32_t tdUnLockSma(SSma *pSma) {
   return 0;
 }
 
-int32_t tdCheckAndInitSmaEnv(SSma *pSma, int8_t smaType, bool onlyCheck) {
+int32_t tdCheckAndInitSmaEnv(SSma *pSma, int8_t smaType) {
   SSmaEnv *pEnv = NULL;
 
   // return if already init
@@ -344,26 +232,7 @@ int32_t tdCheckAndInitSmaEnv(SSma *pSma, int8_t smaType, bool onlyCheck) {
   if (!pEnv) {
     char rname[TSDB_FILENAME_LEN] = {0};
 
-    SDiskID did = {0};
+    if (tdInitSmaEnv(pSma, smaType, rname, &pEnv) < 0) {
-    if (tfsAllocDisk(SMA_TFS(pSma), TFS_PRIMARY_LEVEL, &did) < 0) {
-      tdUnLockSma(pSma);
-      return TSDB_CODE_FAILED;
-    }
-
-    if (did.level < 0 || did.id < 0) {
-      tdUnLockSma(pSma);
-      smaError("vgId:%d, init sma env failed since invalid did(%d,%d)", SMA_VID(pSma), did.level, did.id);
-      return TSDB_CODE_FAILED;
-    }
-
-    tdGetSmaDir(SMA_VID(pSma), smaType, rname);
-
-    if (tfsMkdirRecurAt(SMA_TFS(pSma), rname, did) < 0) {
-      tdUnLockSma(pSma);
-      return TSDB_CODE_FAILED;
-    }
-
-    if (tdInitSmaEnv(pSma, smaType, rname, did, &pEnv) < 0) {
       tdUnLockSma(pSma);
       return TSDB_CODE_FAILED;
     }

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

@@ -132,7 +132,9 @@ int32_t smaClose(SSma *pSma) {
     if SMA_RSMA_TSDB0 (pSma) tsdbClose(&SMA_RSMA_TSDB0(pSma));
     if SMA_RSMA_TSDB1 (pSma) tsdbClose(&SMA_RSMA_TSDB1(pSma));
     if SMA_RSMA_TSDB2 (pSma) tsdbClose(&SMA_RSMA_TSDB2(pSma));
-    taosMemoryFree(pSma);
+    // SMA_TSMA_ENV(pSma) = tdFreeSmaEnv(SMA_TSMA_ENV(pSma));
+    // SMA_RSMA_ENV(pSma) = tdFreeSmaEnv(SMA_RSMA_ENV(pSma));
+    taosMemoryFreeClear(pSma);
   }
   return 0;
 }

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

@@ -181,7 +181,7 @@ int32_t tdProcessRSmaCreate(SVnode *pVnode, SVCreateStbReq *pReq) {
     return TSDB_CODE_SUCCESS;
   }
 
-  if (tdCheckAndInitSmaEnv(pSma, TSDB_SMA_TYPE_ROLLUP, false) != TSDB_CODE_SUCCESS) {
+  if (tdCheckAndInitSmaEnv(pSma, TSDB_SMA_TYPE_ROLLUP) != TSDB_CODE_SUCCESS) {
     terrno = TSDB_CODE_TDB_INIT_FAILED;
     return TSDB_CODE_FAILED;
   }

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

@@ -142,7 +142,6 @@ int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char *msg) {
   ASSERT(pItem);
 
   if (!pItem->pTSma) {
-    // cache smaMeta
     STSma *pTSma = metaGetSmaInfoByIndex(SMA_META(pSma), indexUid);
     if (!pTSma) {
       terrno = TSDB_CODE_TSMA_NO_INDEX_IN_META;
@@ -150,27 +149,28 @@ int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char *msg) {
       return TSDB_CODE_FAILED;
     }
     pItem->pTSma = pTSma;
+    pItem->pTSchema = metaGetTbTSchema(SMA_META(pSma), pTSma->dstTbUid, -1);
+    ASSERT(pItem->pTSchema);  // TODO
   }
 
-  STSma *pTSma = pItem->pTSma;
+  ASSERT(pItem->pTSma->indexUid == indexUid);
-
-  ASSERT(pTSma->indexUid == indexUid);
-
-  SMetaReader mr = {0};
-
-  const char *dbName = "testDb";
-  if (metaGetTableEntryByName(&mr, dbName) != 0) {
-    smaDebug("vgId:%d, tsma no table testTb exists for smaIndex %" PRIi64 " since %s", SMA_VID(pSma), indexUid, tstrerror(terrno));
-    SVCreateStbReq pReq = {0};
-    pReq.name = dbName;
-    pReq.suid = pTSma->dstTbUid;
-    pReq.schemaRow = pCfg->schemaRow;
-    pReq.schemaTag = pCfg->schemaTag;
-  }
 
   SSubmitReq *pSubmitReq = NULL;
-  buildSubmitReqFromDataBlock(&pSubmitReq, (const SArray *)msg, NULL, pItem->pTSma->dstVgId,
+
-                              pItem->pTSma->dstTbUid);
+  pSubmitReq = tdBlockToSubmit((const SArray *)msg, pItem->pTSchema, true, pItem->pTSma->dstTbUid,
+                               pItem->pTSma->dstTbName, pItem->pTSma->dstVgId);
+
+  ASSERT(pSubmitReq);  // TODO
+
+  ASSERT(!strncasecmp("td.tsma.rst.tb", pItem->pTSma->dstTbName, 14));
+
+  SRpcMsg submitReqMsg = {
+      .msgType = TDMT_VND_SUBMIT,
+      .pCont = pSubmitReq,
+      .contLen = ntohl(pSubmitReq->length),
+  };
+
+  ASSERT(tmsgPutToQueue(&pSma->pVnode->msgCb, WRITE_QUEUE, &submitReqMsg) == 0);
 
   tdUnRefSmaStat(pSma, pStat);
 

source/dnode/vnode/src/sma/smaTimeRange2.c

@@ -1,170 +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/>.
- */
-
-#include "sma.h"
-#include "tsdb.h"
-
-typedef STsdbCfg STSmaKeepCfg;
-
-#undef _TEST_SMA_PRINT_DEBUG_LOG_
-#define SMA_STORAGE_MINUTES_MAX  86400
-#define SMA_STORAGE_MINUTES_DAY  1440
-#define SMA_STORAGE_MINUTES_MIN  1440
-#define SMA_STORAGE_TSDB_MINUTES 86400
-#define SMA_STORAGE_TSDB_TIMES   10
-#define SMA_STORAGE_SPLIT_FACTOR 14400  // least records in tsma file TODO: the feasible value?
-#define SMA_KEY_LEN              16     // TSKEY+groupId 8+8
-#define SMA_DROP_EXPIRED_TIME    10     // default is 10 seconds
-
-#define SMA_STATE_ITEM_HASH_SLOT 32
-
-// static func
-
-/**
- * @brief Judge the tsma file split days
- *
- * @param pCfg
- * @param pCont
- * @param contLen
- * @param days unit is minute
- * @return int32_t
- */
-int32_t tdProcessTSmaGetDaysImpl(SVnodeCfg *pCfg, void *pCont, uint32_t contLen, int32_t *days) {
-  SDecoder coder = {0};
-  tDecoderInit(&coder, pCont, contLen);
-
-  STSma tsma = {0};
-  if (tDecodeSVCreateTSmaReq(&coder, &tsma) < 0) {
-    terrno = TSDB_CODE_MSG_DECODE_ERROR;
-    goto _err;
-  }
-  STsdbCfg *pTsdbCfg = &pCfg->tsdbCfg;
-  int64_t   sInterval = convertTimeFromPrecisionToUnit(tsma.interval, pTsdbCfg->precision, TIME_UNIT_SECOND);
-  if (sInterval <= 0) {
-    *days = pTsdbCfg->days;
-    return 0;
-  }
-  int64_t records = pTsdbCfg->days * 60 / sInterval;
-  if (records >= SMA_STORAGE_SPLIT_FACTOR) {
-    *days = pTsdbCfg->days;
-  } else {
-    int64_t mInterval = convertTimeFromPrecisionToUnit(tsma.interval, pTsdbCfg->precision, TIME_UNIT_MINUTE);
-    int64_t daysPerFile = mInterval * SMA_STORAGE_MINUTES_DAY * 2;
-
-    if (daysPerFile > SMA_STORAGE_MINUTES_MAX) {
-      *days = SMA_STORAGE_MINUTES_MAX;
-    } else {
-      *days = (int32_t)daysPerFile;
-    }
-
-    if (*days < pTsdbCfg->days) {
-      *days = pTsdbCfg->days;
-    }
-  }
-  tDecoderClear(&coder);
-  return 0;
-_err:
-  tDecoderClear(&coder);
-  return -1;
-}
-
-// read data
-
-// implementation
-
-/**
- * @brief Insert/Update Time-range-wise SMA data.
- *  - If interval < SMA_STORAGE_SPLIT_HOURS(e.g. 24), save the SMA data as a part of DFileSet to e.g.
- * v3f1900.tsma.${sma_index_name}. The days is the same with that for TS data files.
- *  - If interval >= SMA_STORAGE_SPLIT_HOURS, save the SMA data to e.g. vnode3/tsma/v3f632.tsma.${sma_index_name}. The
- * days is 30 times of the interval, and the minimum days is SMA_STORAGE_TSDB_DAYS(30d).
- *  - The destination file of one data block for some interval is determined by its start TS key.
- *
- * @param pSma
- * @param msg
- * @return int32_t
- */
-int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char *msg) {
-  STsdbCfg *pCfg = SMA_TSDB_CFG(pSma);
-
-  const SArray *pDataBlocks = (const SArray *)msg;
-
-  // TODO: destroy SSDataBlocks(msg)
-
-  // For super table aggregation, the sma data is stored in vgroup calculated from the hash value of stable name. Thus
-  // the sma data would arrive ahead of the update-expired-window msg.
-  if (tdCheckAndInitSmaEnv(pSma, TSDB_SMA_TYPE_TIME_RANGE, false) != TSDB_CODE_SUCCESS) {
-    terrno = TSDB_CODE_TDB_INIT_FAILED;
-    return TSDB_CODE_FAILED;
-  }
-
-  if (!pDataBlocks) {
-    terrno = TSDB_CODE_INVALID_PTR;
-    smaWarn("vgId:%d, insert tsma data failed since pDataBlocks is NULL", SMA_VID(pSma));
-    return terrno;
-  }
-
-  if (taosArrayGetSize(pDataBlocks) <= 0) {
-    terrno = TSDB_CODE_INVALID_PARA;
-    smaWarn("vgId:%d, insert tsma data failed since pDataBlocks is empty", SMA_VID(pSma));
-    return TSDB_CODE_FAILED;
-  }
-
-  SSmaEnv      *pEnv = SMA_TSMA_ENV(pSma);
-  SSmaStat     *pStat = SMA_ENV_STAT(pEnv);
-  SSmaStatItem *pItem = NULL;
-
-  tdRefSmaStat(pSma, pStat);
-
-  if (pStat && SMA_STAT_ITEMS(pStat)) {
-    pItem = taosHashGet(SMA_STAT_ITEMS(pStat), &indexUid, sizeof(indexUid));
-  }
-
-  if (!pItem || !(pItem = *(SSmaStatItem **)pItem) || tdSmaStatIsDropped(pItem)) {
-    terrno = TSDB_CODE_TSMA_INVALID_STAT;
-    tdUnRefSmaStat(pSma, pStat);
-    return TSDB_CODE_FAILED;
-  }
-
-  STSma *pTSma = pItem->pTSma;
-
-  tdUnRefSmaStat(pSma, pStat);
-
-  return TSDB_CODE_SUCCESS;
-}
-
-int32_t tdProcessTSmaCreateImpl(SSma *pSma, int64_t version, const char *pMsg) {
-  SSmaCfg *pCfg = (SSmaCfg *)pMsg;
-
-  if (metaCreateTSma(SMA_META(pSma), version, pCfg) < 0) {
-    return -1;
-  }
-
-  if (TD_VID(pSma->pVnode) == pCfg->dstVgId) {
-    // create stable to save tsma result in dstVgId
-    SVCreateStbReq pReq = {0};
-    pReq.name = pCfg->dstTbName;
-    pReq.suid = pCfg->dstTbUid;
-    pReq.schemaRow = pCfg->schemaRow;
-    pReq.schemaTag = pCfg->schemaTag;
-
-    if (metaCreateSTable(SMA_META(pSma), version, &pReq) < 0) {
-      return -1;
-    }
-  }
-
-  tdTSmaAdd(pSma, 1);
-  return 0;
-}

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

@@ -375,6 +375,8 @@ int32_t tqProcessTaskDeploy(STQ* pTq, char* msg, int32_t msgLen) {
 
   if (pTask->inputQueue == NULL || pTask->outputQueue == NULL) goto FAIL;
 
+  pTask->pMsgCb = &pTq->pVnode->msgCb;
+
   // exec
   if (pTask->execType != TASK_EXEC__NONE) {
     // expand runners
@@ -406,9 +408,12 @@ int32_t tqProcessTaskDeploy(STQ* pTq, char* msg, int32_t msgLen) {
         tdGetSTSChemaFromSSChema(&pTask->tbSink.pSchemaWrapper->pSchema, pTask->tbSink.pSchemaWrapper->nCols);
     ASSERT(pTask->tbSink.pTSchema);
   }
+
+  streamSetupTrigger(pTask);
+
   tqInfo("deploy stream task id %d child id %d on vg %d", pTask->taskId, pTask->childId, pTq->pVnode->config.vgId);
 
-  taosHashPut(pTq->pStreamTasks, &pTask->taskId, sizeof(int32_t), pTask, sizeof(SStreamTask));
+  taosHashPut(pTq->pStreamTasks, &pTask->taskId, sizeof(int32_t), &pTask, sizeof(void*));
 
   return 0;
 FAIL:
@@ -431,7 +436,7 @@ int32_t tqProcessStreamTrigger(STQ* pTq, SSubmitReq* pReq) {
   while (1) {
     pIter = taosHashIterate(pTq->pStreamTasks, pIter);
     if (pIter == NULL) break;
-    SStreamTask* pTask = (SStreamTask*)pIter;
+    SStreamTask* pTask = *(SStreamTask**)pIter;
     if (pTask->inputType != STREAM_INPUT__DATA_SUBMIT) continue;
 
     if (!failed) {
@@ -439,7 +444,7 @@ int32_t tqProcessStreamTrigger(STQ* pTq, SSubmitReq* pReq) {
         continue;
       }
 
-      if (streamTriggerByWrite(pTask, pTq->pVnode->config.vgId, &pTq->pVnode->msgCb) < 0) {
+      if (streamLaunchByWrite(pTask, pTq->pVnode->config.vgId, &pTq->pVnode->msgCb) < 0) {
         continue;
       }
     } else {
@@ -459,7 +464,7 @@ int32_t tqProcessTaskRunReq(STQ* pTq, SRpcMsg* pMsg) {
   //
   SStreamTaskRunReq* pReq = pMsg->pCont;
   int32_t            taskId = pReq->taskId;
-  SStreamTask*       pTask = taosHashGet(pTq->pStreamTasks, &taskId, sizeof(int32_t));
+  SStreamTask*       pTask = *(SStreamTask**)taosHashGet(pTq->pStreamTasks, &taskId, sizeof(int32_t));
   streamTaskProcessRunReq(pTask, &pTq->pVnode->msgCb);
   return 0;
 }
@@ -473,7 +478,7 @@ int32_t tqProcessTaskDispatchReq(STQ* pTq, SRpcMsg* pMsg) {
   tDecoderInit(&decoder, msgBody, msgLen);
   tDecodeStreamDispatchReq(&decoder, &req);
   int32_t      taskId = req.taskId;
-  SStreamTask* pTask = taosHashGet(pTq->pStreamTasks, &taskId, sizeof(int32_t));
+  SStreamTask* pTask = *(SStreamTask**)taosHashGet(pTq->pStreamTasks, &taskId, sizeof(int32_t));
   SRpcMsg      rsp = {
            .info = pMsg->info,
            .code = 0,
@@ -485,7 +490,7 @@ int32_t tqProcessTaskDispatchReq(STQ* pTq, SRpcMsg* pMsg) {
 int32_t tqProcessTaskRecoverReq(STQ* pTq, SRpcMsg* pMsg) {
   SStreamTaskRecoverReq* pReq = pMsg->pCont;
   int32_t                taskId = pReq->taskId;
-  SStreamTask*           pTask = taosHashGet(pTq->pStreamTasks, &taskId, sizeof(int32_t));
+  SStreamTask*           pTask = *(SStreamTask**)taosHashGet(pTq->pStreamTasks, &taskId, sizeof(int32_t));
   streamProcessRecoverReq(pTask, &pTq->pVnode->msgCb, pReq, pMsg);
   return 0;
 }
@@ -493,7 +498,7 @@ int32_t tqProcessTaskRecoverReq(STQ* pTq, SRpcMsg* pMsg) {
 int32_t tqProcessTaskDispatchRsp(STQ* pTq, SRpcMsg* pMsg) {
   SStreamDispatchRsp* pRsp = POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead));
   int32_t             taskId = pRsp->taskId;
-  SStreamTask*        pTask = taosHashGet(pTq->pStreamTasks, &taskId, sizeof(int32_t));
+  SStreamTask*        pTask = *(SStreamTask**)taosHashGet(pTq->pStreamTasks, &taskId, sizeof(int32_t));
   streamProcessDispatchRsp(pTask, &pTq->pVnode->msgCb, pRsp);
   return 0;
 }
@@ -501,7 +506,7 @@ int32_t tqProcessTaskDispatchRsp(STQ* pTq, SRpcMsg* pMsg) {
 int32_t tqProcessTaskRecoverRsp(STQ* pTq, SRpcMsg* pMsg) {
   SStreamTaskRecoverRsp* pRsp = pMsg->pCont;
   int32_t                taskId = pRsp->taskId;
-  SStreamTask*           pTask = taosHashGet(pTq->pStreamTasks, &taskId, sizeof(int32_t));
+  SStreamTask*           pTask = *(SStreamTask**)taosHashGet(pTq->pStreamTasks, &taskId, sizeof(int32_t));
   streamProcessRecoverRsp(pTask, pRsp);
   return 0;
 }

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

@@ -15,10 +15,7 @@
 
 #include "tq.h"
 
-static SSubmitReq* tdBlockToSubmit(const SArray* pBlocks, const STSchema* pSchema, bool createTb, int64_t suid,
+SSubmitReq* tdBlockToSubmit(const SArray* pBlocks, const STSchema* pTSchema, bool createTb, int64_t suid,
-                                   const char* stbFullName, int32_t vgId);
-
-static SSubmitReq* tdBlockToSubmit(const SArray* pBlocks, const STSchema* pTSchema, bool createTb, int64_t suid,
                                    const char* stbFullName, int32_t vgId) {
   SSubmitReq* ret = NULL;
   SArray*     tagArray = taosArrayInit(1, sizeof(STagVal));

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

@@ -284,7 +284,7 @@ int32_t vnodeProcessWriteMsg(SVnode *pVnode, int64_t version, SRpcMsg *pMsg, SRp
 void smaHandleRes(void *pVnode, int64_t smaId, const SArray *data) {
   // TODO
 
-  // blockDebugShowData(data, __func__);
+  blockDebugShowData(data, __func__);
   tdProcessTSmaInsert(((SVnode *)pVnode)->pSma, smaId, (const char *)data);
 }
 
@@ -296,7 +296,7 @@ void vnodeUpdateMetaRsp(SVnode *pVnode, STableMetaRsp *pMetaRsp) {
 }
 
 int32_t vnodeProcessSyncReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
-  int32_t ret = TAOS_SYNC_OTHER_ERROR;
+  int32_t ret = 0;
 
   if (syncEnvIsStart()) {
     SSyncNode *pSyncNode = syncNodeAcquire(pVnode->sync);
@@ -381,15 +381,18 @@ int32_t vnodeProcessSyncReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
       tmsgSendRsp(&rsp);
     } else {
       vError("==vnodeProcessSyncReq== error msg type:%d", pRpcMsg->msgType);
-      ret = TAOS_SYNC_OTHER_ERROR;
+      ret = -1;
     }
 
     syncNodeRelease(pSyncNode);
   } else {
     vError("==vnodeProcessSyncReq== error syncEnv stop");
-    ret = TAOS_SYNC_OTHER_ERROR;
+    ret = -1;
   }
 
+  if (ret != 0) {
+    terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
+  }
   return ret;
 }
 

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

@@ -98,7 +98,8 @@ void vnodeProposeMsg(SQueueInfo *pInfo, STaosQall *qall, int32_t numOfMsgs) {
 
     if (code == 0) {
       vnodeAccumBlockMsg(pVnode, pMsg->msgType);
-    } else if (code == TAOS_SYNC_PROPOSE_NOT_LEADER) {
+
+    } else if (code == -1 && terrno == TSDB_CODE_SYN_NOT_LEADER) {
       SEpSet newEpSet = {0};
       syncGetEpSet(pVnode->sync, &newEpSet);
       SEp *pEp = &newEpSet.eps[newEpSet.inUse];
@@ -247,29 +248,17 @@ static void vnodeSyncRollBackMsg(SSyncFSM *pFsm, const SRpcMsg *pMsg, SFsmCbMeta
   syncRpcMsgLog2(logBuf, (SRpcMsg *)pMsg);
 }
 
-int32_t vnodeSnapshotStartRead(struct SSyncFSM *pFsm, void **ppReader) {
+int32_t vnodeSnapshotStartRead(struct SSyncFSM *pFsm, void **ppReader) { return 0; }
-  return 0;
-}
 
-int32_t vnodeSnapshotStopRead(struct SSyncFSM *pFsm, void *pReader) {
+int32_t vnodeSnapshotStopRead(struct SSyncFSM *pFsm, void *pReader) { return 0; }
-  return 0;
-}
 
-int32_t vnodeSnapshotDoRead(struct SSyncFSM *pFsm, void *pReader, void **ppBuf, int32_t *len) {
+int32_t vnodeSnapshotDoRead(struct SSyncFSM *pFsm, void *pReader, void **ppBuf, int32_t *len) { return 0; }
-  return 0;
-}
 
-int32_t vnodeSnapshotStartWrite(struct SSyncFSM *pFsm, void **ppWriter) {
+int32_t vnodeSnapshotStartWrite(struct SSyncFSM *pFsm, void **ppWriter) { return 0; }
-  return 0;
-}
 
-int32_t vnodeSnapshotStopWrite(struct SSyncFSM *pFsm, void *pWriter, bool isApply) {
+int32_t vnodeSnapshotStopWrite(struct SSyncFSM *pFsm, void *pWriter, bool isApply) { return 0; }
-  return 0;
-}
 
-int32_t vnodeSnapshotDoWrite(struct SSyncFSM *pFsm, void *pWriter, void *pBuf, int32_t len) {
+int32_t vnodeSnapshotDoWrite(struct SSyncFSM *pFsm, void *pWriter, void *pBuf, int32_t len) { return 0; }
-  return 0;
-}
 
 static SSyncFSM *vnodeSyncMakeFsm(SVnode *pVnode) {
   SSyncFSM *pFsm = taosMemoryCalloc(1, sizeof(SSyncFSM));

source/dnode/vnode/test/tsdbSmaTest.cpp

@@ -121,7 +121,7 @@ TEST(testCase, tSma_Meta_Encode_Decode_Test) {
 
   // decode
   STSmaWrapper dstTSmaWrapper = {0};
-  void        *result = tDecodeTSmaWrapper(pSW, &dstTSmaWrapper);
+  void        *result = tDecodeTSmaWrapper(pSW, &dstTSmaWrapper, false);
   EXPECT_NE(result, nullptr);
 
   EXPECT_EQ(tSmaWrapper.number, dstTSmaWrapper.number);

source/libs/catalog/src/ctgAsync.c

@@ -261,54 +261,48 @@ int32_t ctgInitGetTbIndexTask(SCtgJob *pJob, int32_t taskIdx, SName *name) {
 }
 
 
-int32_t ctgHandleForceUpdate(SCatalog* pCtg, SCtgJob *pJob, const SCatalogReq* pReq) {
+int32_t ctgHandleForceUpdate(SCatalog* pCtg, int32_t taskNum, SCtgJob *pJob, const SCatalogReq* pReq) {
-  int32_t dbNum = pJob->dbCfgNum + pJob->dbVgNum + pJob->dbInfoNum;
+  SHashObj* pDb = taosHashInit(taskNum, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), false, HASH_NO_LOCK);
-  if (dbNum > 0) {
+  if (NULL == pDb) {
-    if (dbNum > pJob->dbCfgNum && dbNum > pJob->dbVgNum && dbNum > pJob->dbInfoNum) {
+    CTG_ERR_RET(TSDB_CODE_OUT_OF_MEMORY);
-      SHashObj* pDb = taosHashInit(dbNum, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), false, HASH_NO_LOCK);
-      if (NULL == pDb) {
-        CTG_ERR_RET(TSDB_CODE_OUT_OF_MEMORY);
-      }
-      
-      for (int32_t i = 0; i < pJob->dbVgNum; ++i) {
-        char* dbFName = taosArrayGet(pReq->pDbVgroup, i);
-        taosHashPut(pDb, dbFName, strlen(dbFName), dbFName, TSDB_DB_FNAME_LEN);
-      }
-
-      for (int32_t i = 0; i < pJob->dbCfgNum; ++i) {
-        char* dbFName = taosArrayGet(pReq->pDbCfg, i);
-        taosHashPut(pDb, dbFName, strlen(dbFName), dbFName, TSDB_DB_FNAME_LEN);
-      }
-
-      for (int32_t i = 0; i < pJob->dbInfoNum; ++i) {
-        char* dbFName = taosArrayGet(pReq->pDbInfo, i);
-        taosHashPut(pDb, dbFName, strlen(dbFName), dbFName, TSDB_DB_FNAME_LEN);
-      }
-
-      char* dbFName = taosHashIterate(pDb, NULL);
-      while (dbFName) {
-        ctgDropDbVgroupEnqueue(pCtg, dbFName, true);
-        dbFName = taosHashIterate(pDb, dbFName);
-      }
-
-      taosHashCleanup(pDb);      
-    } else {
-      for (int32_t i = 0; i < pJob->dbVgNum; ++i) {
-        char* dbFName = taosArrayGet(pReq->pDbVgroup, i);
-        CTG_ERR_RET(ctgDropDbVgroupEnqueue(pCtg, dbFName, true));
-      }
-      
-      for (int32_t i = 0; i < pJob->dbCfgNum; ++i) {
-        char* dbFName = taosArrayGet(pReq->pDbCfg, i);
-        CTG_ERR_RET(ctgDropDbVgroupEnqueue(pCtg, dbFName, true));
-      }
-      
-      for (int32_t i = 0; i < pJob->dbInfoNum; ++i) {
-        char* dbFName = taosArrayGet(pReq->pDbInfo, i);
-        CTG_ERR_RET(ctgDropDbVgroupEnqueue(pCtg, dbFName, true));
-      }
-    }
   }
+  
+  for (int32_t i = 0; i < pJob->dbVgNum; ++i) {
+    char* dbFName = taosArrayGet(pReq->pDbVgroup, i);
+    taosHashPut(pDb, dbFName, strlen(dbFName), dbFName, TSDB_DB_FNAME_LEN);
+  }
+
+  for (int32_t i = 0; i < pJob->dbCfgNum; ++i) {
+    char* dbFName = taosArrayGet(pReq->pDbCfg, i);
+    taosHashPut(pDb, dbFName, strlen(dbFName), dbFName, TSDB_DB_FNAME_LEN);
+  }
+
+  for (int32_t i = 0; i < pJob->dbInfoNum; ++i) {
+    char* dbFName = taosArrayGet(pReq->pDbInfo, i);
+    taosHashPut(pDb, dbFName, strlen(dbFName), dbFName, TSDB_DB_FNAME_LEN);
+  }
+
+  for (int32_t i = 0; i < pJob->tbMetaNum; ++i) {
+    SName* name = taosArrayGet(pReq->pTableMeta, i);
+    char dbFName[TSDB_DB_FNAME_LEN];
+    tNameGetFullDbName(name, dbFName);
+    taosHashPut(pDb, dbFName, strlen(dbFName), dbFName, TSDB_DB_FNAME_LEN);
+  }
+  
+  for (int32_t i = 0; i < pJob->tbHashNum; ++i) {
+    SName* name = taosArrayGet(pReq->pTableHash, i);
+    char dbFName[TSDB_DB_FNAME_LEN];
+    tNameGetFullDbName(name, dbFName);
+    taosHashPut(pDb, dbFName, strlen(dbFName), dbFName, TSDB_DB_FNAME_LEN);
+  }
+
+  char* dbFName = taosHashIterate(pDb, NULL);
+  while (dbFName) {
+    ctgDropDbVgroupEnqueue(pCtg, dbFName, true);
+    dbFName = taosHashIterate(pDb, dbFName);
+  }
+
+  taosHashCleanup(pDb);
 
   int32_t tbNum = pJob->tbMetaNum + pJob->tbHashNum;
   if (tbNum > 0) {
@@ -404,7 +398,7 @@ int32_t ctgInitJob(SCatalog* pCtg, SRequestConnInfo *pConn, SCtgJob** job, uint6
   }
 
   if (pReq->forceUpdate) {
-    CTG_ERR_JRET(ctgHandleForceUpdate(pCtg, pJob, pReq));
+    CTG_ERR_JRET(ctgHandleForceUpdate(pCtg, *taskNum, pJob, pReq));
   }
 
   int32_t taskIdx = 0;
@@ -790,8 +784,7 @@ int32_t ctgHandleGetTbMetaRsp(SCtgTask* pTask, int32_t reqType, const SDataBuf *
 _return:
 
   if (dbCache) {
-    ctgRUnlockVgInfo(dbCache);
+    ctgReleaseVgInfoToCache(pCtg, dbCache);
-    ctgReleaseDBCache(pCtg, dbCache);
   }
 
   ctgHandleTaskEnd(pTask, code);
@@ -870,7 +863,7 @@ _return:
 
 int32_t ctgHandleGetTbIndexRsp(SCtgTask* pTask, int32_t reqType, const SDataBuf *pMsg, int32_t rspCode) {
   int32_t code = 0;
-  CTG_ERR_JRET(ctgProcessRspMsg(&pTask->msgCtx.out, reqType, pMsg->pData, pMsg->len, rspCode, pTask->msgCtx.target));
+  CTG_ERR_JRET(ctgProcessRspMsg(pTask->msgCtx.out, reqType, pMsg->pData, pMsg->len, rspCode, pTask->msgCtx.target));
 
   STableIndex* pOut = (STableIndex*)pTask->msgCtx.out;
   SArray* pInfo = NULL;
@@ -949,7 +942,6 @@ _return:
 
 int32_t ctgHandleGetUserRsp(SCtgTask* pTask, int32_t reqType, const SDataBuf *pMsg, int32_t rspCode) {
   int32_t code = 0;
-  SCtgDBCache *dbCache = NULL;
   SCtgUserCtx* ctx = (SCtgUserCtx*)pTask->taskCtx;
   SCatalog* pCtg = pTask->pJob->pCtg; 
   bool pass = false;
@@ -1018,7 +1010,7 @@ int32_t ctgAsyncRefreshTbMeta(SCtgTask *pTask) {
   CTG_ERR_RET(ctgAcquireVgInfoFromCache(pCtg, dbFName, &dbCache));
   if (dbCache) {
     SVgroupInfo vgInfo = {0};
-    CTG_ERR_RET(ctgGetVgInfoFromHashValue(pCtg, dbCache->vgCache.vgInfo, ctx->pName, &vgInfo));
+    CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, dbCache->vgCache.vgInfo, ctx->pName, &vgInfo));
 
     ctgDebug("will refresh tbmeta, not supposed to be stb, tbName:%s, flag:%d", tNameGetTableName(ctx->pName), ctx->flag);
 
@@ -1067,6 +1059,9 @@ int32_t ctgLaunchGetDbVgTask(SCtgTask *pTask) {
   CTG_ERR_RET(ctgAcquireVgInfoFromCache(pCtg, pCtx->dbFName, &dbCache));
   if (NULL != dbCache) {
     CTG_ERR_JRET(ctgGenerateVgList(pCtg, dbCache->vgCache.vgInfo->vgHash, (SArray**)&pTask->res));
+
+    ctgReleaseVgInfoToCache(pCtg, dbCache);
+    dbCache = NULL;
     
     CTG_ERR_JRET(ctgHandleTaskEnd(pTask, 0));
   } else {
@@ -1101,6 +1096,9 @@ int32_t ctgLaunchGetTbHashTask(SCtgTask *pTask) {
       CTG_ERR_JRET(TSDB_CODE_OUT_OF_MEMORY);
     }
     CTG_ERR_JRET(ctgGetVgInfoFromHashValue(pCtg, dbCache->vgCache.vgInfo, pCtx->pName, (SVgroupInfo*)pTask->res));
+
+    ctgReleaseVgInfoToCache(pCtg, dbCache);
+    dbCache = NULL;
     
     CTG_ERR_JRET(ctgHandleTaskEnd(pTask, 0));
   } else {
@@ -1176,6 +1174,9 @@ int32_t ctgLaunchGetDbInfoTask(SCtgTask *pTask) {
     pInfo->vgVer = dbCache->vgCache.vgInfo->vgVersion;
     pInfo->dbId = dbCache->dbId;
     pInfo->tbNum = dbCache->vgCache.vgInfo->numOfTable;
+
+    ctgReleaseVgInfoToCache(pCtg, dbCache);
+    dbCache = NULL;
   } else {
     pInfo->vgVer = CTG_DEFAULT_INVALID_VERSION;
   }

source/libs/catalog/src/ctgCache.c

@@ -1899,7 +1899,7 @@ _return:
 
 
 void ctgUpdateThreadUnexpectedStopped(void) {
-  if (CTG_IS_LOCKED(&gCtgMgmt.lock) > 0) CTG_UNLOCK(CTG_READ, &gCtgMgmt.lock);
+  if (!atomic_load_8((int8_t*)&gCtgMgmt.exit) && CTG_IS_LOCKED(&gCtgMgmt.lock) > 0) CTG_UNLOCK(CTG_READ, &gCtgMgmt.lock);
 }
 
 void ctgCleanupCacheQueue(void) {

source/libs/catalog/src/ctgDbg.c

@@ -19,7 +19,7 @@
 #include "catalogInt.h"
 
 extern SCatalogMgmt gCtgMgmt;
-SCtgDebug gCTGDebug = {0};
+SCtgDebug gCTGDebug = {.lockEnable = true, .apiEnable = true};
 
 void ctgdUserCallback(SMetaData* pResult, void* param, int32_t code) {
   ASSERT(*(int32_t*)param == 1);

source/libs/catalog/src/ctgUtil.c

@@ -675,7 +675,8 @@ int32_t ctgCloneTableIndex(SArray* pIndex, SArray** pRes) {
 
   for (int32_t i = 0; i < num; ++i) {
     STableIndexInfo *pInfo = taosArrayGet(pIndex, i);
-    taosArrayPush(*pRes, pInfo);
+    pInfo = taosArrayPush(*pRes, pInfo);
+    pInfo->expr = strdup(pInfo->expr);
   }
 
   return TSDB_CODE_SUCCESS;

source/libs/executor/inc/executorimpl.h

@@ -365,7 +365,9 @@ typedef struct SCatchSupporter {
 } SCatchSupporter;
 
 typedef struct SStreamAggSupporter {
-  SArray*        pResultRows;
+  SHashObj*      pResultRows;
+  SArray*        pCurWins;
+  int32_t        valueSize;
   int32_t        keySize;
   char*          pKeyBuf;              // window key buffer
   SDiskbasedBuf* pResultBuf;           // query result buffer based on blocked-wised disk file
@@ -391,7 +393,9 @@ typedef struct SStreamBlockScanInfo {
   void*           streamBlockReader;// stream block reader handle
   SArray*         pColMatchInfo;    //
   SNode*          pCondition;
+  int32_t         tsArrayIndex;
   SArray*         tsArray;
+  uint64_t        groupId;
   SUpdateInfo*    pUpdateInfo;
 
   SExprInfo*      pPseudoExpr;
@@ -582,6 +586,7 @@ typedef struct SPartitionOperatorInfo {
   int32_t*       columnOffset;  // start position for each column data
   void*          pGroupIter;  // group iterator
   int32_t        pageIndex;   // page index of current group
+  SSDataBlock*   pUpdateRes;
 } SPartitionOperatorInfo;
 
 typedef struct SWindowRowsSup {
@@ -896,9 +901,9 @@ int32_t getNumOfRowsInTimeWindow(SDataBlockInfo* pDataBlockInfo, TSKEY* pPrimary
     __block_search_fn_t searchFn, STableQueryInfo* item, int32_t order);
 int32_t binarySearchForKey(char* pValue, int num, TSKEY key, int order);
 int32_t initStreamAggSupporter(SStreamAggSupporter* pSup, const char* pKey,
-    SqlFunctionCtx* pCtx, int32_t numOfOutput, size_t size);
+    SqlFunctionCtx* pCtx, int32_t numOfOutput, int32_t size);
 SResultRow* getNewResultRow(SDiskbasedBuf* pResultBuf, int64_t tableGroupId, int32_t interBufSize);
-SResultWindowInfo* getSessionTimeWindow(SArray* pWinInfos, TSKEY ts, int64_t gap, int32_t* pIndex);
+SResultWindowInfo* getSessionTimeWindow(SStreamAggSupporter* pAggSup, TSKEY ts, uint64_t groupId, int64_t gap, int32_t* pIndex);
 int32_t updateSessionWindowInfo(SResultWindowInfo* pWinInfo, TSKEY* pTs, int32_t rows,
     int32_t start, int64_t gap, SHashObj* pStDeleted);
 bool functionNeedToExecute(SqlFunctionCtx* pCtx);
@@ -907,6 +912,7 @@ 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);
+void copyUpdateDataBlock(SSDataBlock* pDest, SSDataBlock* pSource, int32_t tsColIndex);
 
 #ifdef __cplusplus
 }

source/libs/executor/src/executor.c

@@ -19,7 +19,8 @@
 #include "tdatablock.h"
 #include "vnode.h"
 
-static int32_t doSetStreamBlock(SOperatorInfo* pOperator, void* input, size_t numOfBlocks, int32_t type, bool assignUid, char* id) {
+static int32_t doSetStreamBlock(SOperatorInfo* pOperator, void* input, size_t numOfBlocks, int32_t type, bool assignUid,
+                                char* id) {
   ASSERT(pOperator != NULL);
   if (pOperator->operatorType != QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN) {
     if (pOperator->numOfDownstream == 0) {
@@ -43,6 +44,7 @@ static int32_t doSetStreamBlock(SOperatorInfo* pOperator, void* input, size_t nu
     if (pInfo->blockType == 0) {
       pInfo->blockType = type;
     } else if (pInfo->blockType != type) {
+      ASSERT(0);
       return TSDB_CODE_QRY_APP_ERROR;
     }
 
@@ -51,7 +53,7 @@ static int32_t doSetStreamBlock(SOperatorInfo* pOperator, void* input, size_t nu
         qError("submit msg messed up when initing stream block, %s" PRIx64, id);
         return TSDB_CODE_QRY_APP_ERROR;
       }
-    } else {
+    } else if (type == STREAM_DATA_TYPE_SSDATA_BLOCK) {
       for (int32_t i = 0; i < numOfBlocks; ++i) {
         SSDataBlock* pDataBlock = &((SSDataBlock*)input)[i];
 
@@ -62,6 +64,8 @@ static int32_t doSetStreamBlock(SOperatorInfo* pOperator, void* input, size_t nu
         taosArrayAddAll(p->pDataBlock, pDataBlock->pDataBlock);
         taosArrayPush(pInfo->pBlockLists, &p);
       }
+    } else {
+      ASSERT(0);
     }
 
     return TSDB_CODE_SUCCESS;
@@ -83,7 +87,8 @@ int32_t qSetMultiStreamInput(qTaskInfo_t tinfo, const void* pBlocks, size_t numO
 
   SExecTaskInfo* pTaskInfo = (SExecTaskInfo*)tinfo;
 
-  int32_t code = doSetStreamBlock(pTaskInfo->pRoot, (void**)pBlocks, numOfBlocks, type, assignUid, GET_TASKID(pTaskInfo));
+  int32_t code =
+      doSetStreamBlock(pTaskInfo->pRoot, (void**)pBlocks, numOfBlocks, type, assignUid, GET_TASKID(pTaskInfo));
   if (code != TSDB_CODE_SUCCESS) {
     qError("%s failed to set the stream block data", GET_TASKID(pTaskInfo));
   } else {
@@ -162,7 +167,7 @@ int32_t qUpdateQualifiedTableId(qTaskInfo_t tinfo, const SArray* tableIdList, bo
     pInfo = pInfo->pDownstream[0];
   }
 
-  int32_t code = 0;
+  int32_t               code = 0;
   SStreamBlockScanInfo* pScanInfo = pInfo->info;
   if (isAdd) {  // add new table id
     SArray* qa = filterQualifiedChildTables(pScanInfo, tableIdList);
@@ -178,9 +183,10 @@ int32_t qUpdateQualifiedTableId(qTaskInfo_t tinfo, const SArray* tableIdList, bo
   return code;
 }
 
-int32_t qGetQueriedTableSchemaVersion(qTaskInfo_t tinfo, char* dbName, char* tableName, int32_t* sversion, int32_t* tversion) {
+int32_t qGetQueriedTableSchemaVersion(qTaskInfo_t tinfo, char* dbName, char* tableName, int32_t* sversion,
+                                      int32_t* tversion) {
   ASSERT(tinfo != NULL && dbName != NULL && tableName != NULL);
-  SExecTaskInfo* pTaskInfo = (SExecTaskInfo*) tinfo;
+  SExecTaskInfo* pTaskInfo = (SExecTaskInfo*)tinfo;
 
   *sversion = pTaskInfo->schemaVer.sversion;
   *tversion = pTaskInfo->schemaVer.tversion;
@@ -196,4 +202,4 @@ int32_t qGetQueriedTableSchemaVersion(qTaskInfo_t tinfo, char* dbName, char* tab
   }
 
   return 0;
-}
+}

source/libs/executor/src/executorimpl.c

@@ -4814,6 +4814,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
         .calTrigger = pIntervalPhyNode->window.triggerType,
         .maxTs = INT64_MIN,
     };
+    ASSERT(as.calTrigger != STREAM_TRIGGER_MAX_DELAY);
 
     int32_t tsSlotId = ((SColumnNode*)pIntervalPhyNode->window.pTspk)->slotId;
     bool    isStream = (QUERY_NODE_PHYSICAL_PLAN_STREAM_INTERVAL == type);
@@ -5498,14 +5499,16 @@ int32_t getOperatorExplainExecInfo(SOperatorInfo* operatorInfo, SExplainExecInfo
 }
 
 int32_t initStreamAggSupporter(SStreamAggSupporter* pSup, const char* pKey, SqlFunctionCtx* pCtx, int32_t numOfOutput,
-                               size_t size) {
+                               int32_t size) {
   pSup->resultRowSize = getResultRowSize(pCtx, numOfOutput);
   pSup->keySize = sizeof(int64_t) + sizeof(TSKEY);
   pSup->pKeyBuf = taosMemoryCalloc(1, pSup->keySize);
-  pSup->pResultRows = taosArrayInit(1024, size);
+  _hash_fn_t hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);
+  pSup->pResultRows = taosHashInit(1024, hashFn, false, HASH_NO_LOCK);
   if (pSup->pKeyBuf == NULL || pSup->pResultRows == NULL) {
     return TSDB_CODE_OUT_OF_MEMORY;
   }
+  pSup->valueSize = size;
 
   pSup->pScanWindow = taosArrayInit(4, sizeof(STimeWindow));
 

source/libs/executor/src/scanoperator.c

@@ -716,7 +716,7 @@ static bool prepareDataScan(SStreamBlockScanInfo* pInfo) {
       int64_t              gap = pInfo->sessionSup.gap;
       int32_t              winIndex = 0;
       SResultWindowInfo*   pCurWin =
-          getSessionTimeWindow(pAggSup->pResultRows, tsCols[pInfo->updateResIndex], gap, &winIndex);
+          getSessionTimeWindow(pAggSup, tsCols[pInfo->updateResIndex], pSDB->info.groupId, gap, &winIndex);
       win = pCurWin->win;
       pInfo->updateResIndex +=
           updateSessionWindowInfo(pCurWin, tsCols, pSDB->info.rows, pInfo->updateResIndex, gap, NULL);
@@ -750,16 +750,103 @@ static bool prepareDataScan(SStreamBlockScanInfo* pInfo) {
   return true;
 }
 
+static void copyOneRow(SSDataBlock* dest, SSDataBlock* source, int32_t sourceRowId) {
+  for (int32_t j = 0; j < source->info.numOfCols; j++) {
+    SColumnInfoData* pDestCol = (SColumnInfoData*)taosArrayGet(dest->pDataBlock, j);
+    SColumnInfoData* pSourceCol = (SColumnInfoData*)taosArrayGet(source->pDataBlock, j);
+    if (colDataIsNull_s(pSourceCol, sourceRowId)) {
+      colDataAppendNULL(pDestCol, dest->info.rows);
+    } else {
+      colDataAppend(pDestCol, dest->info.rows, colDataGetData(pSourceCol, sourceRowId), false);
+    }
+  }
+  dest->info.rows++;
+}
+
+static uint64_t getGroupId(SOperatorInfo* pOperator, SSDataBlock* pBlock, int32_t rowId) {
+  uint64_t* groupId = taosHashGet(pOperator->pTaskInfo->tableqinfoList.map, &pBlock->info.uid, sizeof(int64_t));
+  if (groupId) {
+    return *groupId;
+  }
+  return 0;
+  /* Todo(liuyao) for partition by column
+  recordNewGroupKeys(pTableScanInfo->pGroupCols, pTableScanInfo->pGroupColVals, pBlock, rowId);
+  int32_t len = buildGroupKeys(pTableScanInfo->keyBuf, pTableScanInfo->pGroupColVals);
+  uint64_t resId = 0;
+  uint64_t* groupId = taosHashGet(pTableScanInfo->pGroupSet, pTableScanInfo->keyBuf, len);
+  if (groupId) {
+    return *groupId;
+  } else if (len != 0) {
+    resId = calcGroupId(pTableScanInfo->keyBuf, len);
+    taosHashPut(pTableScanInfo->pGroupSet, pTableScanInfo->keyBuf, len, &resId, sizeof(uint64_t));
+  }
+  return resId;
+  */
+}
+
 static SSDataBlock* doDataScan(SStreamBlockScanInfo* pInfo) {
-  SSDataBlock* pResult = NULL;
+  while (1) {
-  pResult = doTableScan(pInfo->pOperatorDumy);
+    SSDataBlock* pResult = NULL;
-  if (pResult == NULL) {
+    pResult = doTableScan(pInfo->pOperatorDumy);
-    if (prepareDataScan(pInfo)) {
+    if (pResult == NULL) {
-      // scan next window data
+      if (prepareDataScan(pInfo)) {
-      pResult = doTableScan(pInfo->pOperatorDumy);
+        // scan next window data
+        pResult = doTableScan(pInfo->pOperatorDumy);
+      }
+    }
+    if (!pResult) {
+      return NULL;
+    }
+    
+    if (pResult->info.groupId == pInfo->groupId) {
+      return pResult;
+    }
+  }
+
+/* Todo(liuyao) for partition by column
+  SSDataBlock* pBlock = createOneDataBlock(pResult, true);
+  blockDataCleanup(pResult);
+  for (int32_t i = 0; i < pBlock->info.rows; i++) {
+    uint64_t id = getGroupId(pInfo->pOperatorDumy, pBlock, i);
+    if (id == pInfo->groupId) {
+      copyOneRow(pResult, pBlock, i);
     }
   }
   return pResult;
+*/
+}
+
+static void setUpdateData(SStreamBlockScanInfo* pInfo, SSDataBlock* pBlock, SSDataBlock* pUpdateBlock) {
+  blockDataCleanup(pUpdateBlock);
+  int32_t size = taosArrayGetSize(pInfo->tsArray);
+  if (pInfo->tsArrayIndex < size) {
+    SColumnInfoData* pCol = (SColumnInfoData*)taosArrayGet(pUpdateBlock->pDataBlock, pInfo->primaryTsIndex);
+    ASSERT(pCol->info.type == TSDB_DATA_TYPE_TIMESTAMP);
+    blockDataEnsureCapacity(pUpdateBlock, size);
+    ASSERT(pBlock->info.numOfCols == pUpdateBlock->info.numOfCols);
+
+    int32_t rowId = *(int32_t*)taosArrayGet(pInfo->tsArray, pInfo->tsArrayIndex);
+    pInfo->groupId = getGroupId(pInfo->pOperatorDumy, pBlock, rowId);
+    int32_t i = 0;
+    for ( ; i < size; i++) {
+      rowId = *(int32_t*)taosArrayGet(pInfo->tsArray, i + pInfo->tsArrayIndex);
+      uint64_t id = getGroupId(pInfo->pOperatorDumy, pBlock, rowId);
+      if (pInfo->groupId != id) {
+        break;
+      }
+      copyOneRow(pUpdateBlock, pBlock, rowId);
+    }
+    pUpdateBlock->info.rows = i;
+    pInfo->tsArrayIndex += i;
+    pUpdateBlock->info.groupId = pInfo->groupId;
+    pUpdateBlock->info.type = STREAM_REPROCESS;
+    blockDataUpdateTsWindow(pUpdateBlock, 0);
+  }
+  // all rows have same group id
+  ASSERT(pInfo->tsArrayIndex >= size);
+  if (size > 0 && pInfo->tsArrayIndex == size) {
+    taosArrayClear(pInfo->tsArray);
+  }
 }
 
 static void getUpdateDataBlock(SStreamBlockScanInfo* pInfo, bool invertible, SSDataBlock* pBlock,
@@ -767,41 +854,21 @@ static void getUpdateDataBlock(SStreamBlockScanInfo* pInfo, bool invertible, SSD
   SColumnInfoData* pColDataInfo = taosArrayGet(pBlock->pDataBlock, pInfo->primaryTsIndex);
   ASSERT(pColDataInfo->info.type == TSDB_DATA_TYPE_TIMESTAMP);
   TSKEY* ts = (TSKEY*)pColDataInfo->pData;
-  for (int32_t i = 0; i < pBlock->info.rows; i++) {
+  for (int32_t rowId = 0; rowId < pBlock->info.rows; rowId++) {
-    if (updateInfoIsUpdated(pInfo->pUpdateInfo, pBlock->info.uid, ts[i])) {
+    if (updateInfoIsUpdated(pInfo->pUpdateInfo, pBlock->info.uid, ts[rowId])) {
-      taosArrayPush(pInfo->tsArray, ts + i);
+      taosArrayPush(pInfo->tsArray, &rowId);
     }
   }
   if (!pUpdateBlock) {
     taosArrayClear(pInfo->tsArray);
     return;
   }
-  int32_t size = taosArrayGetSize(pInfo->tsArray);
+  setUpdateData(pInfo, pBlock, pUpdateBlock);
-  if (size > 0 && invertible) {
+  // Todo(liuyao) get from tsdb
-    // Todo(liuyao) get from tsdb
+  //  SSDataBlock* p = createOneDataBlock(pBlock, true);
-    //  SSDataBlock* p = createOneDataBlock(pBlock, true);
+  //  p->info.type = STREAM_INVERT;
-    //  p->info.type = STREAM_INVERT;
+  //  taosArrayClear(pInfo->tsArray);
-    //  taosArrayClear(pInfo->tsArray);
+  //  return p;
-    //  return p;
-    SColumnInfoData* pCol = (SColumnInfoData*)taosArrayGet(pUpdateBlock->pDataBlock, pInfo->primaryTsIndex);
-    ASSERT(pCol->info.type == TSDB_DATA_TYPE_TIMESTAMP);
-    blockDataEnsureCapacity(pUpdateBlock, size);
-    for (int32_t i = 0; i < size; i++) {
-      TSKEY* pTs = (TSKEY*)taosArrayGet(pInfo->tsArray, i);
-      colDataAppend(pCol, i, (char*)pTs, false);
-    }
-    for (int32_t i = 0; i < pUpdateBlock->info.numOfCols; i++) {
-      if (i == pInfo->primaryTsIndex) {
-        continue;
-      }
-      SColumnInfoData* pCol = (SColumnInfoData*)taosArrayGet(pUpdateBlock->pDataBlock, i);
-      colDataAppendNNULL(pCol, 0, size);
-    }
-    pUpdateBlock->info.rows = size;
-    pUpdateBlock->info.type = STREAM_REPROCESS;
-    blockDataUpdateTsWindow(pUpdateBlock, 0);
-    taosArrayClear(pInfo->tsArray);
-  }
 }
 
 static SSDataBlock* doStreamBlockScan(SOperatorInfo* pOperator) {
@@ -833,7 +900,6 @@ static SSDataBlock* doStreamBlockScan(SOperatorInfo* pOperator) {
       pInfo->scanMode = STREAM_SCAN_FROM_READERHANDLE;
       return pInfo->pRes;
     } else if (pInfo->scanMode == STREAM_SCAN_FROM_UPDATERES) {
-      blockDataCleanup(pInfo->pRes);
       pInfo->scanMode = STREAM_SCAN_FROM_DATAREADER;
       if (!isStateWindow(pInfo)) {
         prepareDataScan(pInfo);
@@ -848,7 +914,15 @@ static SSDataBlock* doStreamBlockScan(SOperatorInfo* pOperator) {
       if (pInfo->scanMode == STREAM_SCAN_FROM_DATAREADER) {
         SSDataBlock* pSDB = doDataScan(pInfo);
         if (pSDB == NULL) {
-          pInfo->scanMode = STREAM_SCAN_FROM_READERHANDLE;
+          setUpdateData(pInfo, pInfo->pRes, pInfo->pUpdateRes);
+          if (pInfo->pUpdateRes->info.rows > 0) {
+            if (!isStateWindow(pInfo)) {
+              prepareDataScan(pInfo);
+            }
+            return pInfo->pUpdateRes;
+          } else {
+            pInfo->scanMode = STREAM_SCAN_FROM_READERHANDLE;
+          }
         } else {
           getUpdateDataBlock(pInfo, true, pSDB, NULL);
           return pSDB;
@@ -941,7 +1015,7 @@ static SSDataBlock* doStreamBlockScan(SOperatorInfo* pOperator) {
     if (rows == 0) {
       pOperator->status = OP_EXEC_DONE;
     } else if (pInfo->pUpdateInfo) {
-      blockDataCleanup(pInfo->pUpdateRes);
+      pInfo->tsArrayIndex = 0;
       getUpdateDataBlock(pInfo, true, pInfo->pRes, pInfo->pUpdateRes);
       if (pInfo->pUpdateRes->info.rows > 0) {
         if (pInfo->pUpdateRes->info.type == STREAM_REPROCESS) {
@@ -1020,7 +1094,7 @@ SOperatorInfo* createStreamScanOperatorInfo(void* pDataReader, SReadHandle* pHan
     goto _error;
   }
 
-  pInfo->tsArray = taosArrayInit(4, sizeof(TSKEY));
+  pInfo->tsArray = taosArrayInit(4, sizeof(int32_t));
   if (pInfo->tsArray == NULL) {
     goto _error;
   }
@@ -1047,6 +1121,8 @@ SOperatorInfo* createStreamScanOperatorInfo(void* pDataReader, SReadHandle* pHan
   pInfo->pOperatorDumy = pTableScanDummy;
   pInfo->interval = pSTInfo->interval;
   pInfo->sessionSup = (SessionWindowSupporter){.pStreamAggSup = NULL, .gap = -1};
+  pInfo->groupId = 0;
+  
   pOperator->name = "StreamBlockScanOperator";
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN;
   pOperator->blocking = false;

source/libs/executor/src/timewindowoperator.c

@@ -1320,6 +1320,15 @@ static SSDataBlock* doStreamIntervalAgg(SOperatorInfo* pOperator) {
       break;
     }
 
+    if (pBlock->info.type == STREAM_REPROCESS) {
+      doClearWindows(&pInfo->aggSup, &pInfo->binfo, &pInfo->interval, 0, pOperator->numOfExprs, pBlock, NULL);
+      qDebug("%s clear existed time window results for updates checked", GET_TASKID(pTaskInfo));
+      continue;
+    } else if (pBlock->info.type == STREAM_GET_ALL) {
+      getAllIntervalWindow(pInfo->aggSup.pResultRowHashTable, pUpdated);
+      continue;
+    }
+
     // The timewindow that overlaps the timestamps of the input pBlock need to be recalculated and return to the
     // caller. Note that all the time window are not close till now.
     // the pDataBlock are always the same one, no need to call this again
@@ -1328,16 +1337,6 @@ static SSDataBlock* doStreamIntervalAgg(SOperatorInfo* pOperator) {
       setInverFunction(pInfo->binfo.pCtx, pOperator->numOfExprs, pBlock->info.type);
     }
 
-    if (pBlock->info.type == STREAM_REPROCESS) {
-      doClearWindows(&pInfo->aggSup, &pInfo->binfo, &pInfo->interval, 0, pOperator->numOfExprs, pBlock, NULL);
-      qDebug("%s clear existed time window results for updates checked", GET_TASKID(pTaskInfo));
-      continue;
-    } else if (pBlock->info.type == STREAM_GET_ALL &&
-        pInfo->twAggSup.calTrigger == STREAM_TRIGGER_MAX_DELAY) {
-      getAllIntervalWindow(pInfo->aggSup.pResultRowHashTable, pUpdated);
-      continue;
-    }
-
     pInfo->twAggSup.maxTs = TMAX(pInfo->twAggSup.maxTs, pBlock->info.window.ekey);
     hashIntervalAgg(pOperator, &pInfo->binfo.resultRowInfo, pBlock, MAIN_SCAN, pUpdated);
   }
@@ -1985,7 +1984,7 @@ static void clearUpdateDataBlock(SSDataBlock* pBlock) {
   blockDataCleanup(pBlock);
 }
 
-static void copyUpdateDataBlock(SSDataBlock* pDest, SSDataBlock* pSource, int32_t tsColIndex) {
+void copyUpdateDataBlock(SSDataBlock* pDest, SSDataBlock* pSource, int32_t tsColIndex) {
   ASSERT(pDest->info.capacity >= pSource->info.rows);
   clearUpdateDataBlock(pDest);
   SColumnInfoData* pDestCol = taosArrayGet(pDest->pDataBlock, 0);
@@ -1997,6 +1996,8 @@ static void copyUpdateDataBlock(SSDataBlock* pDest, SSDataBlock* pSource, int32_
     colDataAppendNNULL(pCol, 0, pSource->info.rows);
   }
   pDest->info.rows = pSource->info.rows;
+  pDest->info.groupId = pSource->info.groupId;
+  pDest->info.type = pSource->info.type;
   blockDataUpdateTsWindow(pDest, 0);
 }
 
@@ -2036,7 +2037,6 @@ static SSDataBlock* doStreamFinalIntervalAgg(SOperatorInfo* pOperator) {
       break;
     }
 
-    setInputDataBlock(pOperator, pInfo->binfo.pCtx, pBlock, pInfo->order, MAIN_SCAN, true);
     if (pBlock->info.type == STREAM_REPROCESS) {
       SArray* pUpWins = taosArrayInit(8, sizeof(STimeWindow));
       doClearWindows(&pInfo->aggSup, &pInfo->binfo, &pInfo->interval, pInfo->primaryTsIndex, pOperator->numOfExprs,
@@ -2056,12 +2056,12 @@ static SSDataBlock* doStreamFinalIntervalAgg(SOperatorInfo* pOperator) {
       copyUpdateDataBlock(pInfo->pUpdateRes, pBlock, pInfo->primaryTsIndex);
       taosArrayDestroy(pUpWins);
       break;
-    } else if (pBlock->info.type == STREAM_GET_ALL && isFinalInterval(pInfo) &&
+    } else if (pBlock->info.type == STREAM_GET_ALL && isFinalInterval(pInfo)) {
-        pInfo->twAggSup.calTrigger == STREAM_TRIGGER_MAX_DELAY) {
       getAllIntervalWindow(pInfo->aggSup.pResultRowHashTable, pUpdated);
       continue;
     }
 
+    setInputDataBlock(pOperator, pInfo->binfo.pCtx, pBlock, pInfo->order, MAIN_SCAN, true);
     if (isFinalInterval(pInfo)) {
       int32_t chIndex = getChildIndex(pBlock);
       int32_t size = taosArrayGetSize(pInfo->pChildren);
@@ -2123,6 +2123,7 @@ SOperatorInfo* createStreamFinalIntervalOperatorInfo(SOperatorInfo* downstream,
       .calTrigger = pIntervalPhyNode->window.triggerType,
       .maxTs = INT64_MIN,
   };
+  ASSERT(pInfo->twAggSup.calTrigger != STREAM_TRIGGER_MAX_DELAY);
   pInfo->primaryTsIndex = ((SColumnNode*)pIntervalPhyNode->window.pTspk)->slotId;
   size_t keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
   initResultSizeInfo(pOperator, 4096);
@@ -2188,8 +2189,13 @@ _error:
 }
 
 void destroyStreamAggSupporter(SStreamAggSupporter* pSup) {
-  taosArrayDestroy(pSup->pResultRows);
   taosMemoryFreeClear(pSup->pKeyBuf);
+  void **pIte = NULL;
+  while ((pIte = taosHashIterate(pSup->pResultRows, pIte)) != NULL) {
+    SArray *pWins = (SArray *) (*pIte);
+    taosArrayDestroy(pWins);
+  }
+  taosHashCleanup(pSup->pResultRows);
   destroyDiskbasedBuf(pSup->pResultBuf);
 }
 
@@ -2331,7 +2337,22 @@ static SResultWindowInfo* addNewSessionWindow(SArray* pWinInfos, TSKEY ts) {
   return taosArrayPush(pWinInfos, &win);
 }
 
-SResultWindowInfo* getSessionTimeWindow(SArray* pWinInfos, TSKEY ts, int64_t gap, int32_t* pIndex) {
+SArray* getWinInfos(SStreamAggSupporter* pAggSup, uint64_t groupId) {
+  void** ite = taosHashGet(pAggSup->pResultRows, &groupId, sizeof(uint64_t));
+  SArray* pWinInfos = NULL;
+  if (ite == NULL) {
+    pWinInfos = taosArrayInit(1024, pAggSup->valueSize);
+    taosHashPut(pAggSup->pResultRows, &groupId, sizeof(uint64_t), &pWinInfos, sizeof(void *));
+  } else {
+    pWinInfos = *ite;
+  }
+  return pWinInfos;
+}
+
+SResultWindowInfo* getSessionTimeWindow(SStreamAggSupporter* pAggSup, TSKEY ts, uint64_t groupId, int64_t gap, int32_t* pIndex) {
+  SArray* pWinInfos = getWinInfos(pAggSup, groupId);
+  pAggSup->pCurWins = pWinInfos;
+
   int32_t size = taosArrayGetSize(pWinInfos);
   if (size == 0) {
     *pIndex = 0;
@@ -2387,7 +2408,7 @@ static int32_t setWindowOutputBuf(SResultWindowInfo* pWinInfo, SResultRow** pRes
                                   SStreamAggSupporter* pAggSup, SExecTaskInfo* pTaskInfo) {
   assert(pWinInfo->win.skey <= pWinInfo->win.ekey);
   // too many time window in query
-  int32_t size = taosArrayGetSize(pAggSup->pResultRows);
+  int32_t size = taosArrayGetSize(pAggSup->pCurWins);
   if (size > MAX_INTERVAL_TIME_WINDOW) {
     longjmp(pTaskInfo->env, TSDB_CODE_QRY_TOO_MANY_TIMEWINDOW);
   }
@@ -2447,25 +2468,6 @@ static int32_t doOneStateWindowAgg(SStreamStateAggOperatorInfo* pInfo, SSDataBlo
                             pSDataBlock, pCurWin, pResult, startIndex, winRows, numOutput, pTaskInfo);
 }
 
-int32_t copyWinInfoToDataBlock(SSDataBlock* pBlock, SStreamAggSupporter* pAggSup, int32_t start, int32_t num,
-                               int32_t numOfExprs, SOptrBasicInfo* pBinfo) {
-  for (int32_t i = start; i < num; i += 1) {
-    SResultWindowInfo* pWinInfo = taosArrayGet(pAggSup->pResultRows, start);
-    SFilePage*         bufPage = getBufPage(pAggSup->pResultBuf, pWinInfo->pos.pageId);
-    SResultRow*        pRow = (SResultRow*)((char*)bufPage + pWinInfo->pos.offset);
-    for (int32_t j = 0; j < numOfExprs; ++j) {
-      SResultRowEntryInfo* pResultInfo = getResultCell(pRow, j, pBinfo->rowCellInfoOffset);
-      SColumnInfoData*     pColInfoData = taosArrayGet(pBlock->pDataBlock, j);
-      char*                in = GET_ROWCELL_INTERBUF(pBinfo->pCtx[j].resultInfo);
-      colDataAppend(pColInfoData, pBlock->info.rows, in, pResultInfo->isNullRes);
-    }
-    pBlock->info.rows += pRow->numOfRows;
-    releaseBufPage(pAggSup->pResultBuf, bufPage);
-  }
-  blockDataUpdateTsWindow(pBlock, -1);
-  return TSDB_CODE_SUCCESS;
-}
-
 int32_t getNumCompactWindow(SArray* pWinInfos, int32_t startIndex, int64_t gap) {
   SResultWindowInfo* pCurWin = taosArrayGet(pWinInfos, startIndex);
   int32_t            size = taosArrayGetSize(pWinInfos);
@@ -2482,15 +2484,15 @@ int32_t getNumCompactWindow(SArray* pWinInfos, int32_t startIndex, int64_t gap)
 
 void compactTimeWindow(SStreamSessionAggOperatorInfo* pInfo, int32_t startIndex, int32_t num, int32_t groupId,
                        int32_t numOfOutput, SExecTaskInfo* pTaskInfo, SHashObj* pStUpdated, SHashObj* pStDeleted) {
-  SResultWindowInfo* pCurWin = taosArrayGet(pInfo->streamAggSup.pResultRows, startIndex);
+  SResultWindowInfo* pCurWin = taosArrayGet(pInfo->streamAggSup.pCurWins, startIndex);
   SResultRow*        pCurResult = NULL;
   setWindowOutputBuf(pCurWin, &pCurResult, pInfo->binfo.pCtx, groupId, numOfOutput, pInfo->binfo.rowCellInfoOffset,
                      &pInfo->streamAggSup, pTaskInfo);
   num += startIndex + 1;
-  ASSERT(num <= taosArrayGetSize(pInfo->streamAggSup.pResultRows));
+  ASSERT(num <= taosArrayGetSize(pInfo->streamAggSup.pCurWins));
   // Just look for the window behind StartIndex
   for (int32_t i = startIndex + 1; i < num; i++) {
-    SResultWindowInfo* pWinInfo = taosArrayGet(pInfo->streamAggSup.pResultRows, i);
+    SResultWindowInfo* pWinInfo = taosArrayGet(pInfo->streamAggSup.pCurWins, i);
     SResultRow*        pWinResult = NULL;
     setWindowOutputBuf(pWinInfo, &pWinResult, pInfo->pDummyCtx, groupId, numOfOutput, pInfo->binfo.rowCellInfoOffset,
                        &pInfo->streamAggSup, pTaskInfo);
@@ -2501,7 +2503,7 @@ void compactTimeWindow(SStreamSessionAggOperatorInfo* pInfo, int32_t startIndex,
       taosHashPut(pStDeleted, &pWinInfo->pos, sizeof(SResultRowPosition), &pWinInfo->win.skey, sizeof(TSKEY));
       pWinInfo->isOutput = false;
     }
-    taosArrayRemove(pInfo->streamAggSup.pResultRows, i);
+    taosArrayRemove(pInfo->streamAggSup.pCurWins, i);
   }
 }
 
@@ -2531,7 +2533,7 @@ static void doStreamSessionAggImpl(SOperatorInfo* pOperator, SSDataBlock* pSData
   SStreamAggSupporter* pAggSup = &pInfo->streamAggSup;
   for (int32_t i = 0; i < pSDataBlock->info.rows;) {
     int32_t            winIndex = 0;
-    SResultWindowInfo* pCurWin = getSessionTimeWindow(pAggSup->pResultRows, tsCols[i], gap, &winIndex);
+    SResultWindowInfo* pCurWin = getSessionTimeWindow(pAggSup, tsCols[i], pSDataBlock->info.groupId, gap, &winIndex);
     winRows = updateSessionWindowInfo(pCurWin, tsCols, pSDataBlock->info.rows, i, pInfo->gap, pStDeleted);
     code = doOneWindowAgg(pInfo, pSDataBlock, pCurWin, &pResult, i, winRows, numOfOutput, pTaskInfo);
     if (code != TSDB_CODE_SUCCESS || pResult == NULL) {
@@ -2541,7 +2543,7 @@ static void doStreamSessionAggImpl(SOperatorInfo* pOperator, SSDataBlock* pSData
     // doWindowBorderInterpolation(pOperatorInfo, pSDataBlock, pInfo->binfo.pCtx, pResult, &nextWin, startPos,
     // forwardRows,
     //                             pInfo->order, false);
-    int32_t winNum = getNumCompactWindow(pAggSup->pResultRows, winIndex, gap);
+    int32_t winNum = getNumCompactWindow(pAggSup->pCurWins, winIndex, gap);
     if (winNum > 0) {
       compactTimeWindow(pInfo, winIndex, winNum, groupId, numOfOutput, pTaskInfo, pStUpdated, pStDeleted);
     }
@@ -2564,7 +2566,7 @@ static void doClearSessionWindows(SStreamAggSupporter* pAggSup, SOptrBasicInfo*
   int32_t          step = 0;
   for (int32_t i = 0; i < pBlock->info.rows; i += step) {
     int32_t            winIndex = 0;
-    SResultWindowInfo* pCurWin = getSessionTimeWindow(pAggSup->pResultRows, tsCols[i], gap, &winIndex);
+    SResultWindowInfo* pCurWin = getSessionTimeWindow(pAggSup, tsCols[i], pBlock->info.groupId, gap, &winIndex);
     step = updateSessionWindowInfo(pCurWin, tsCols, pBlock->info.rows, i, gap, NULL);
     ASSERT(isInWindow(pCurWin, tsCols[i], gap));
     doClearWindowImpl(&pCurWin->pos, pAggSup->pResultBuf, pBinfo, numOfOutput);
@@ -2625,7 +2627,7 @@ static void rebuildTimeWindow(SStreamSessionAggOperatorInfo* pInfo, SArray* pWin
     for (int32_t j = 0; j < numOfChildren; j++) {
       SOperatorInfo*                 pChild = taosArrayGetP(pInfo->pChildren, j);
       SStreamSessionAggOperatorInfo* pChInfo = pChild->info;
-      SArray*                        pChWins = pChInfo->streamAggSup.pResultRows;
+      SArray*                        pChWins = getWinInfos(&pChInfo->streamAggSup, groupId);
       int32_t                        chWinSize = taosArrayGetSize(pChWins);
       int32_t index = binarySearch(pChWins, chWinSize, pParentWin->win.skey, TSDB_ORDER_DESC, getSessionWindowEndkey);
       for (int32_t k = index; k > 0 && k < chWinSize; k++) {
@@ -2649,36 +2651,44 @@ typedef SResultWindowInfo* (*__get_win_info_)(void*);
 SResultWindowInfo* getSessionWinInfo(void* pData) { return (SResultWindowInfo*)pData; }
 SResultWindowInfo* getStateWinInfo(void* pData) { return &((SStateWindowInfo*)pData)->winInfo; }
 
-int32_t closeSessionWindow(SArray* pWins, STimeWindowAggSupp* pTwSup, SArray* pClosed,
+int32_t closeSessionWindow(SHashObj* pHashMap, STimeWindowAggSupp* pTwSup, SArray* pClosed,
     __get_win_info_ fn) {
   // Todo(liuyao) save window to tdb
-  int32_t size = taosArrayGetSize(pWins);
+  void **pIte = NULL;
-  for (int32_t i = 0; i < size; i++) {
+  while ((pIte = taosHashIterate(pHashMap, pIte)) != NULL) {
-    void*              pWin = taosArrayGet(pWins, i);
+    SArray *pWins = (SArray *) (*pIte);
-    SResultWindowInfo* pSeWin = fn(pWin);
+    int32_t size = taosArrayGetSize(pWins);
-    if (pSeWin->win.ekey < pTwSup->maxTs - pTwSup->waterMark) {
+    for (int32_t i = 0; i < size; i++) {
-      if (!pSeWin->isClosed) {
+      void*              pWin = taosArrayGet(pWins, i);
-        pSeWin->isClosed = true;
+      SResultWindowInfo* pSeWin = fn(pWin);
-        if (pTwSup->calTrigger == STREAM_TRIGGER_WINDOW_CLOSE) {
+      if (pSeWin->win.ekey < pTwSup->maxTs - pTwSup->waterMark) {
-          int32_t code = saveResult(pSeWin->win.skey, pSeWin->pos.pageId, pSeWin->pos.offset, 0, pClosed);
+        if (!pSeWin->isClosed) {
-          pSeWin->isOutput = true;
+          pSeWin->isClosed = true;
+          if (pTwSup->calTrigger == STREAM_TRIGGER_WINDOW_CLOSE) {
+            int32_t code = saveResult(pSeWin->win.skey, pSeWin->pos.pageId, pSeWin->pos.offset, 0, pClosed);
+            pSeWin->isOutput = true;
+          }
         }
+        continue;
       }
-      continue;
+      break;
     }
-    break;
   }
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t getAllSessionWindow(SArray* pWins, SArray* pClosed, __get_win_info_ fn) {
+int32_t getAllSessionWindow(SHashObj* pHashMap, SArray* pClosed, __get_win_info_ fn) {
-  int32_t size = taosArrayGetSize(pWins);
+  void **pIte = NULL;
-  for (int32_t i = 0; i < size; i++) {
+  while ((pIte = taosHashIterate(pHashMap, pIte)) != NULL) {
-    void*              pWin = taosArrayGet(pWins, i);
+    SArray *pWins = (SArray *) (*pIte);
-    SResultWindowInfo* pSeWin = fn(pWin);
+    int32_t size = taosArrayGetSize(pWins);
-    if (!pSeWin->isClosed) {
+    for (int32_t i = 0; i < size; i++) {
-      int32_t code = saveResult(pSeWin->win.skey, pSeWin->pos.pageId, pSeWin->pos.offset, 0, pClosed);
+      void*              pWin = taosArrayGet(pWins, i);
-      pSeWin->isOutput = true;
+      SResultWindowInfo* pSeWin = fn(pWin);
+      if (!pSeWin->isClosed) {
+        int32_t code = saveResult(pSeWin->win.skey, pSeWin->pos.pageId, pSeWin->pos.offset, 0, pClosed);
+        pSeWin->isOutput = true;
+      }
     }
   }
   return TSDB_CODE_SUCCESS;
@@ -2712,8 +2722,7 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
     if (pBlock == NULL) {
       break;
     }
-    // the pDataBlock are always the same one, no need to call this again
+
-    setInputDataBlock(pOperator, pBInfo->pCtx, pBlock, TSDB_ORDER_ASC, MAIN_SCAN, true);
     if (pBlock->info.type == STREAM_REPROCESS) {
       SArray* pWins = taosArrayInit(16, sizeof(SResultWindowInfo));
       doClearSessionWindows(&pInfo->streamAggSup, &pInfo->binfo, pBlock, 0, pOperator->numOfExprs, pInfo->gap, pWins);
@@ -2727,12 +2736,13 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
       }
       taosArrayDestroy(pWins);
       continue;
-    } else if (pBlock->info.type == STREAM_GET_ALL &&
+    } else if (pBlock->info.type == STREAM_GET_ALL) {
-        pInfo->twAggSup.calTrigger == STREAM_TRIGGER_MAX_DELAY) {
       getAllSessionWindow(pInfo->streamAggSup.pResultRows, pUpdated, getSessionWinInfo);
       continue;
     }
 
+    // the pDataBlock are always the same one, no need to call this again
+    setInputDataBlock(pOperator, pBInfo->pCtx, pBlock, TSDB_ORDER_ASC, MAIN_SCAN, true);
     if (isFinalSession(pInfo)) {
       int32_t         childIndex = 0;  // Todo(liuyao) get child id from SSDataBlock
       SOptrBasicInfo* pChildOp = taosArrayGetP(pInfo->pChildren, childIndex);
@@ -2871,7 +2881,9 @@ bool isEqualStateKey(SStateWindowInfo* pWin, char* pKeyData) {
   return pKeyData && compareVal(pKeyData, &pWin->stateKey);
 }
 
-SStateWindowInfo* getStateWindowByTs(SArray* pWinInfos, TSKEY ts, int32_t* pIndex) {
+SStateWindowInfo* getStateWindowByTs(SStreamAggSupporter* pAggSup, TSKEY ts, uint64_t groupId, int32_t* pIndex) {
+  SArray* pWinInfos = getWinInfos(pAggSup, groupId);
+  pAggSup->pCurWins = pWinInfos;
   int32_t           size = taosArrayGetSize(pWinInfos);
   int32_t           index = binarySearch(pWinInfos, size, ts, TSDB_ORDER_DESC, getStateWinTsKey);
   SStateWindowInfo* pWin = NULL;
@@ -2894,7 +2906,10 @@ SStateWindowInfo* getStateWindowByTs(SArray* pWinInfos, TSKEY ts, int32_t* pInde
   return NULL;
 }
 
-SStateWindowInfo* getStateWindow(SArray* pWinInfos, TSKEY ts, char* pKeyData, SColumn* pCol, int32_t* pIndex) {
+SStateWindowInfo* getStateWindow(SStreamAggSupporter* pAggSup, TSKEY ts,
+    uint64_t groupId, char* pKeyData, SColumn* pCol, int32_t* pIndex) {
+  SArray* pWinInfos = getWinInfos(pAggSup, groupId);
+  pAggSup->pCurWins = pWinInfos;
   int32_t size = taosArrayGetSize(pWinInfos);
   if (size == 0) {
     *pIndex = 0;
@@ -2985,16 +3000,16 @@ static void doClearStateWindows(SStreamAggSupporter* pAggSup, SSDataBlock* pBloc
   for (int32_t i = 0; i < pBlock->info.rows; i += step) {
     char*             pKeyData = colDataGetData(pKeyColInfo, i);
     int32_t           winIndex = 0;
-    SStateWindowInfo* pCurWin = getStateWindowByTs(pAggSup->pResultRows, tsCol[i], &winIndex);
+    SStateWindowInfo* pCurWin = getStateWindowByTs(pAggSup, tsCol[i], pBlock->info.groupId, &winIndex);
     if (!pCurWin) {
       continue;
     }
-    step = updateStateWindowInfo(pAggSup->pResultRows, winIndex, tsCol, pKeyColInfo, pBlock->info.rows, i, &allEqual,
+    step = updateStateWindowInfo(pAggSup->pCurWins, winIndex, tsCol, pKeyColInfo, pBlock->info.rows, i, &allEqual,
                                  pSeDeleted);
     ASSERT(isTsInWindow(pCurWin, tsCol[i]) || isEqualStateKey(pCurWin, pKeyData));
     taosArrayPush(pAggSup->pScanWindow, &pCurWin->winInfo.win);
     taosHashRemove(pSeUpdated, &pCurWin->winInfo.pos, sizeof(SResultRowPosition));
-    deleteWindow(pAggSup->pResultRows, winIndex);
+    deleteWindow(pAggSup->pCurWins, winIndex);
   }
 }
 
@@ -3024,13 +3039,15 @@ static void doStreamStateAggImpl(SOperatorInfo* pOperator, SSDataBlock* pSDataBl
     char*             pKeyData = colDataGetData(pKeyColInfo, i);
     int32_t           winIndex = 0;
     bool              allEqual = true;
-    SStateWindowInfo* pCurWin = getStateWindow(pAggSup->pResultRows, tsCols[i], pKeyData, &pInfo->stateCol, &winIndex);
+    SStateWindowInfo* pCurWin = 
-    winRows = updateStateWindowInfo(pAggSup->pResultRows, winIndex, tsCols, pKeyColInfo, pSDataBlock->info.rows, i,
+        getStateWindow(pAggSup, tsCols[i], pSDataBlock->info.groupId, pKeyData,
-                                    &allEqual, pInfo->pSeDeleted);
+            &pInfo->stateCol, &winIndex);
+    winRows = updateStateWindowInfo(pAggSup->pCurWins, winIndex, tsCols, pKeyColInfo,
+        pSDataBlock->info.rows, i, &allEqual, pInfo->pSeDeleted);
     if (!allEqual) {
       taosArrayPush(pAggSup->pScanWindow, &pCurWin->winInfo.win);
       taosHashRemove(pSeUpdated, &pCurWin->winInfo.pos, sizeof(SResultRowPosition));
-      deleteWindow(pAggSup->pResultRows, winIndex);
+      deleteWindow(pAggSup->pCurWins, winIndex);
       continue;
     }
     code = doOneStateWindowAgg(pInfo, pSDataBlock, &pCurWin->winInfo, &pResult, i, winRows, numOfOutput, pTaskInfo);
@@ -3077,17 +3094,18 @@ static SSDataBlock* doStreamStateAgg(SOperatorInfo* pOperator) {
     if (pBlock == NULL) {
       break;
     }
-    // the pDataBlock are always the same one, no need to call this again
+
-    setInputDataBlock(pOperator, pBInfo->pCtx, pBlock, TSDB_ORDER_ASC, MAIN_SCAN, true);
     if (pBlock->info.type == STREAM_REPROCESS) {
       doClearStateWindows(&pInfo->streamAggSup, pBlock, pInfo->primaryTsIndex, &pInfo->stateCol, pInfo->stateCol.slotId,
                           pSeUpdated, pInfo->pSeDeleted);
       continue;
-    } else if (pBlock->info.type == STREAM_GET_ALL &&
+    } else if (pBlock->info.type == STREAM_GET_ALL) {
-        pInfo->twAggSup.calTrigger == STREAM_TRIGGER_MAX_DELAY) {
       getAllSessionWindow(pInfo->streamAggSup.pResultRows, pUpdated, getStateWinInfo);
       continue;
     }
+
+    // the pDataBlock are always the same one, no need to call this again
+    setInputDataBlock(pOperator, pBInfo->pCtx, pBlock, TSDB_ORDER_ASC, MAIN_SCAN, true);
     doStreamStateAggImpl(pOperator, pBlock, pSeUpdated, pInfo->pSeDeleted);
     pInfo->twAggSup.maxTs = TMAX(pInfo->twAggSup.maxTs, pBlock->info.window.ekey);
   }

source/libs/function/src/builtins.c

@@ -1682,7 +1682,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "top",
     .type = FUNCTION_TYPE_TOP,
-    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC | FUNC_MGT_INDEFINITE_ROWS_FUNC,
+    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC | FUNC_MGT_INDEFINITE_ROWS_FUNC | FUNC_MGT_FORBID_STREAM_FUNC,
     .translateFunc = translateTopBot,
     .getEnvFunc   = getTopBotFuncEnv,
     .initFunc     = topBotFunctionSetup,
@@ -1717,7 +1717,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "bottom",
     .type = FUNCTION_TYPE_BOTTOM,
-    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC | FUNC_MGT_INDEFINITE_ROWS_FUNC,
+    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC | FUNC_MGT_INDEFINITE_ROWS_FUNC | FUNC_MGT_FORBID_STREAM_FUNC,
     .translateFunc = translateTopBot,
     .getEnvFunc   = getTopBotFuncEnv,
     .initFunc     = topBotFunctionSetup,

source/libs/function/src/builtinsimpl.c

@@ -2345,14 +2345,7 @@ int32_t apercentileCombine(SqlFunctionCtx* pDestCtx, SqlFunctionCtx* pSourceCtx)
   SResultRowEntryInfo* pSResInfo = GET_RES_INFO(pSourceCtx);
   SAPercentileInfo*    pSBuf = GET_ROWCELL_INTERBUF(pSResInfo);
   ASSERT(pDBuf->algo == pSBuf->algo);
-  if (pDBuf->algo == APERCT_ALGO_TDIGEST) {
+  apercentileTransferInfo(pSBuf, pDBuf);
-    tdigestMerge(pDBuf->pTDigest, pSBuf->pTDigest);
-  } else {
-    SHistogramInfo* pTmp = tHistogramMerge(pDBuf->pHisto, pSBuf->pHisto, MAX_HISTOGRAM_BIN);
-    memcpy(pDBuf->pHisto, pTmp, sizeof(SHistogramInfo) + sizeof(SHistBin) * (MAX_HISTOGRAM_BIN + 1));
-    pDBuf->pHisto->elems = (SHistBin*)((char*)pDBuf->pHisto + sizeof(SHistogramInfo));
-    tHistogramDestroy(&pTmp);
-  }
   pDResInfo->numOfRes = TMAX(pDResInfo->numOfRes, pSResInfo->numOfRes);
   return TSDB_CODE_SUCCESS;
 }
@@ -2392,6 +2385,7 @@ int32_t firstFunction(SqlFunctionCtx* pCtx) {
   SInputColumnInfoData* pInput = &pCtx->input;
   SColumnInfoData*      pInputCol = pInput->pData[0];
 
+  int32_t type  = pInputCol->info.type;
   int32_t bytes = pInputCol->info.bytes;
   pInfo->bytes  = bytes;
 
@@ -2428,6 +2422,10 @@ int32_t firstFunction(SqlFunctionCtx* pCtx) {
       TSKEY cts = getRowPTs(pInput->pPTS, i);
 
       if (pResInfo->numOfRes == 0 || *(TSKEY*)(pInfo->buf + bytes) > cts) {
+        if (IS_VAR_DATA_TYPE(type)) {
+          bytes = varDataTLen(data);
+          pInfo->bytes = bytes;
+        }
         memcpy(pInfo->buf, data, bytes);
         *(TSKEY*)(pInfo->buf + bytes) = cts;
         pInfo->hasResult = true;
@@ -2458,6 +2456,10 @@ int32_t firstFunction(SqlFunctionCtx* pCtx) {
       TSKEY cts = getRowPTs(pInput->pPTS, i);
 
       if (pResInfo->numOfRes == 0 || *(TSKEY*)(pInfo->buf + bytes) > cts) {
+        if (IS_VAR_DATA_TYPE(type)) {
+          bytes = varDataTLen(data);
+          pInfo->bytes = bytes;
+        }
         memcpy(pInfo->buf, data, bytes);
         *(TSKEY*)(pInfo->buf + bytes) = cts;
         pInfo->hasResult = true;
@@ -2481,6 +2483,7 @@ int32_t lastFunction(SqlFunctionCtx* pCtx) {
   SInputColumnInfoData* pInput = &pCtx->input;
   SColumnInfoData*      pInputCol = pInput->pData[0];
 
+  int32_t type  = pInputCol->info.type;
   int32_t bytes = pInputCol->info.bytes;
   pInfo->bytes  = bytes;
 
@@ -2508,6 +2511,10 @@ int32_t lastFunction(SqlFunctionCtx* pCtx) {
       char* data = colDataGetData(pInputCol, i);
       TSKEY cts = getRowPTs(pInput->pPTS, i);
       if (pResInfo->numOfRes == 0 || *(TSKEY*)(pInfo->buf + bytes) < cts) {
+        if (IS_VAR_DATA_TYPE(type)) {
+          bytes = varDataTLen(data);
+          pInfo->bytes = bytes;
+        }
         memcpy(pInfo->buf, data, bytes);
         *(TSKEY*)(pInfo->buf + bytes) = cts;
         //        DO_UPDATE_TAG_COLUMNS(pCtx, ts);
@@ -2527,6 +2534,10 @@ int32_t lastFunction(SqlFunctionCtx* pCtx) {
       char* data = colDataGetData(pInputCol, i);
       TSKEY cts = getRowPTs(pInput->pPTS, i);
       if (pResInfo->numOfRes == 0 || *(TSKEY*)(pInfo->buf + bytes) < cts) {
+        if (IS_VAR_DATA_TYPE(type)) {
+          bytes = varDataTLen(data);
+          pInfo->bytes = bytes;
+        }
         memcpy(pInfo->buf, data, bytes);
         *(TSKEY*)(pInfo->buf + bytes) = cts;
         pInfo->hasResult = true;

source/libs/nodes/src/nodesCodeFuncs.c

@@ -1480,14 +1480,6 @@ static int32_t jsonToPhysiTableScanNode(const SJson* pJson, void* pObj) {
   return code;
 }
 
-static int32_t physiStreamScanNodeToJson(const void* pObj, SJson* pJson) {
-  return physiTableScanNodeToJson(pObj, pJson);
-}
-
-static int32_t jsonToPhysiStreamScanNode(const SJson* pJson, void* pObj) {
-  return jsonToPhysiTableScanNode(pJson, pObj);
-}
-
 static const char* jkSysTableScanPhysiPlanMnodeEpSet = "MnodeEpSet";
 static const char* jkSysTableScanPhysiPlanShowRewrite = "ShowRewrite";
 static const char* jkSysTableScanPhysiPlanAccountId = "AccountId";
@@ -3964,9 +3956,9 @@ static int32_t specificNodeToJson(const void* pObj, SJson* pJson) {
     case QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN:
       return physiTagScanNodeToJson(pObj, pJson);
     case QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN:
-      return physiTableScanNodeToJson(pObj, pJson);
+    case QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN:
     case QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN:
-      return physiStreamScanNodeToJson(pObj, pJson);
+      return physiTableScanNodeToJson(pObj, pJson);
     case QUERY_NODE_PHYSICAL_PLAN_SYSTABLE_SCAN:
       return physiSysTableScanNodeToJson(pObj, pJson);
     case QUERY_NODE_PHYSICAL_PLAN_PROJECT:
@@ -4097,9 +4089,9 @@ static int32_t jsonToSpecificNode(const SJson* pJson, void* pObj) {
     case QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN:
       return jsonToPhysiTagScanNode(pJson, pObj);
     case QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN:
-      return jsonToPhysiTableScanNode(pJson, pObj);
+    case QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN:
     case QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN:
-      return jsonToPhysiStreamScanNode(pJson, pObj);
+      return jsonToPhysiTableScanNode(pJson, pObj);
     case QUERY_NODE_PHYSICAL_PLAN_SYSTABLE_SCAN:
       return jsonToPhysiSysTableScanNode(pJson, pObj);
     case QUERY_NODE_PHYSICAL_PLAN_PROJECT:

source/libs/nodes/src/nodesUtilFuncs.c

@@ -209,9 +209,10 @@ SNode* nodesMakeNode(ENodeType type) {
     case QUERY_NODE_SHOW_CREATE_STABLE_STMT:
     case QUERY_NODE_SHOW_TRANSACTIONS_STMT:
       return makeNode(type, sizeof(SShowStmt));
-    case QUERY_NODE_KILL_CONNECTION_STMT:
     case QUERY_NODE_KILL_QUERY_STMT:
+      return makeNode(type, sizeof(SKillQueryStmt));
     case QUERY_NODE_KILL_TRANSACTION_STMT:
+    case QUERY_NODE_KILL_CONNECTION_STMT:
       return makeNode(type, sizeof(SKillStmt));
     case QUERY_NODE_DELETE_STMT:
       return makeNode(type, sizeof(SDeleteStmt));
@@ -251,6 +252,8 @@ SNode* nodesMakeNode(ENodeType type) {
       return makeNode(type, sizeof(STableScanPhysiNode));
     case QUERY_NODE_PHYSICAL_PLAN_TABLE_SEQ_SCAN:
       return makeNode(type, sizeof(STableSeqScanPhysiNode));
+    case QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN:
+      return makeNode(type, sizeof(STableMergeScanPhysiNode));
     case QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN:
       return makeNode(type, sizeof(SStreamScanPhysiNode));
     case QUERY_NODE_PHYSICAL_PLAN_SYSTABLE_SCAN:

source/libs/parser/inc/parAst.h

@@ -60,6 +60,7 @@ typedef enum EDatabaseOptionType {
 typedef enum ETableOptionType {
   TABLE_OPTION_COMMENT = 1,
   TABLE_OPTION_FILE_FACTOR,
+  TABLE_OPTION_DELAY,
   TABLE_OPTION_ROLLUP,
   TABLE_OPTION_TTL,
   TABLE_OPTION_SMA
@@ -187,6 +188,7 @@ SNode* createCreateStreamStmt(SAstCreateContext* pCxt, bool ignoreExists, const
                               SNode* pOptions, SNode* pQuery);
 SNode* createDropStreamStmt(SAstCreateContext* pCxt, bool ignoreNotExists, const SToken* pStreamName);
 SNode* createKillStmt(SAstCreateContext* pCxt, ENodeType type, const SToken* pId);
+SNode* createKillQueryStmt(SAstCreateContext* pCxt, const SToken* pQueryId);
 SNode* createBalanceVgroupStmt(SAstCreateContext* pCxt);
 SNode* createMergeVgroupStmt(SAstCreateContext* pCxt, const SToken* pVgId1, const SToken* pVgId2);
 SNode* createRedistributeVgroupStmt(SAstCreateContext* pCxt, const SToken* pVgId, SNodeList* pDnodes);

source/libs/parser/inc/sql.y

@@ -168,8 +168,8 @@ db_options(A) ::= .
 db_options(A) ::= db_options(B) BUFFER NK_INTEGER(C).                             { A = setDatabaseOption(pCxt, B, DB_OPTION_BUFFER, &C); }
 db_options(A) ::= db_options(B) CACHELAST NK_INTEGER(C).                          { A = setDatabaseOption(pCxt, B, DB_OPTION_CACHELAST, &C); }
 db_options(A) ::= db_options(B) COMP NK_INTEGER(C).                               { A = setDatabaseOption(pCxt, B, DB_OPTION_COMP, &C); }
-db_options(A) ::= db_options(B) DAYS NK_INTEGER(C).                               { A = setDatabaseOption(pCxt, B, DB_OPTION_DAYS, &C); }
+db_options(A) ::= db_options(B) DURATION NK_INTEGER(C).                           { A = setDatabaseOption(pCxt, B, DB_OPTION_DAYS, &C); }
-db_options(A) ::= db_options(B) DAYS NK_VARIABLE(C).                              { A = setDatabaseOption(pCxt, B, DB_OPTION_DAYS, &C); }
+db_options(A) ::= db_options(B) DURATION NK_VARIABLE(C).                          { A = setDatabaseOption(pCxt, B, DB_OPTION_DAYS, &C); }
 db_options(A) ::= db_options(B) FSYNC NK_INTEGER(C).                              { A = setDatabaseOption(pCxt, B, DB_OPTION_FSYNC, &C); }
 db_options(A) ::= db_options(B) MAXROWS NK_INTEGER(C).                            { A = setDatabaseOption(pCxt, B, DB_OPTION_MAXROWS, &C); }
 db_options(A) ::= db_options(B) MINROWS NK_INTEGER(C).                            { A = setDatabaseOption(pCxt, B, DB_OPTION_MINROWS, &C); }
@@ -317,8 +317,9 @@ tags_def(A) ::= TAGS NK_LP column_def_list(B) NK_RP.
 
 table_options(A) ::= .                                                            { A = createDefaultTableOptions(pCxt); }
 table_options(A) ::= table_options(B) COMMENT NK_STRING(C).                       { A = setTableOption(pCxt, B, TABLE_OPTION_COMMENT, &C); }
+//table_options(A) ::= table_options(B) DELAY NK_INTEGER(C).                        { A = setTableOption(pCxt, B, TABLE_OPTION_DELAY, &C); }
 table_options(A) ::= table_options(B) FILE_FACTOR NK_FLOAT(C).                    { A = setTableOption(pCxt, B, TABLE_OPTION_FILE_FACTOR, &C); }
-table_options(A) ::= table_options(B) ROLLUP NK_LP func_name_list(C) NK_RP.       { A = setTableOption(pCxt, B, TABLE_OPTION_ROLLUP, C); }
+table_options(A) ::= table_options(B) ROLLUP NK_LP rollup_func_list(C) NK_RP.     { A = setTableOption(pCxt, B, TABLE_OPTION_ROLLUP, C); }
 table_options(A) ::= table_options(B) TTL NK_INTEGER(C).                          { A = setTableOption(pCxt, B, TABLE_OPTION_TTL, &C); }
 table_options(A) ::= table_options(B) SMA NK_LP col_name_list(C) NK_RP.           { A = setTableOption(pCxt, B, TABLE_OPTION_SMA, C); }
 
@@ -330,6 +331,15 @@ alter_table_options(A) ::= alter_table_options(B) alter_table_option(C).
 alter_table_option(A) ::= COMMENT NK_STRING(B).                                   { A.type = TABLE_OPTION_COMMENT; A.val = B; }
 alter_table_option(A) ::= TTL NK_INTEGER(B).                                      { A.type = TABLE_OPTION_TTL; A.val = B; }
 
+%type rollup_func_list                                                            { SNodeList* }
+%destructor rollup_func_list                                                      { nodesDestroyList($$); }
+rollup_func_list(A) ::= rollup_func_name(B).                                      { A = createNodeList(pCxt, B); }
+rollup_func_list(A) ::= rollup_func_list(B) NK_COMMA rollup_func_name(C).         { A = addNodeToList(pCxt, B, C); }
+
+rollup_func_name(A) ::= function_name(B).                                         { A = createFunctionNode(pCxt, &B, NULL); }
+rollup_func_name(A) ::= FIRST(B).                                                 { A = createFunctionNode(pCxt, &B, NULL); }
+rollup_func_name(A) ::= LAST(B).                                                  { A = createFunctionNode(pCxt, &B, NULL); }
+
 %type col_name_list                                                               { SNodeList* }
 %destructor col_name_list                                                         { nodesDestroyList($$); }
 col_name_list(A) ::= col_name(B).                                                 { A = createNodeList(pCxt, B); }
@@ -378,13 +388,6 @@ table_name_cond(A) ::= table_name(B).
 from_db_opt(A) ::= .                                                              { A = createDefaultDatabaseCondValue(pCxt); }
 from_db_opt(A) ::= FROM db_name(B).                                               { A = createValueNode(pCxt, TSDB_DATA_TYPE_BINARY, &B); }
 
-%type func_name_list                                                              { SNodeList* }
-%destructor func_name_list                                                        { nodesDestroyList($$); }
-func_name_list(A) ::= func_name(B).                                               { A = createNodeList(pCxt, B); }
-func_name_list(A) ::= func_name_list(B) NK_COMMA func_name(C).                    { A = addNodeToList(pCxt, B, C); }
-
-func_name(A) ::= function_name(B).                                                { A = createFunctionNode(pCxt, &B, NULL); }
-
 /************************************************ create index ********************************************************/
 cmd ::= CREATE SMA INDEX not_exists_opt(D) 
   index_name(A) ON table_name(B) index_options(C).                                { pCxt->pRootNode = createCreateIndexStmt(pCxt, INDEX_TYPE_SMA, D, &A, &B, NULL, C); }
@@ -466,7 +469,7 @@ stream_options(A) ::= stream_options(B) WATERMARK duration_literal(C).
 
 /************************************************ kill connection/query ***********************************************/
 cmd ::= KILL CONNECTION NK_INTEGER(A).                                            { pCxt->pRootNode = createKillStmt(pCxt, QUERY_NODE_KILL_CONNECTION_STMT, &A); }
-cmd ::= KILL QUERY NK_INTEGER(A).                                                 { pCxt->pRootNode = createKillStmt(pCxt, QUERY_NODE_KILL_QUERY_STMT, &A); }
+cmd ::= KILL QUERY NK_STRING(A).                                                  { pCxt->pRootNode = createKillQueryStmt(pCxt, &A); }
 cmd ::= KILL TRANSACTION NK_INTEGER(A).                                           { pCxt->pRootNode = createKillStmt(pCxt, QUERY_NODE_KILL_TRANSACTION_STMT, &A); }
 
 /************************************************ merge/redistribute/ vgroup ******************************************/

source/libs/parser/src/parAstCreater.c

@@ -804,10 +804,10 @@ SNode* setDatabaseOption(SAstCreateContext* pCxt, SNode* pOptions, EDatabaseOpti
     case DB_OPTION_RETENTIONS:
       ((SDatabaseOptions*)pOptions)->pRetentions = pVal;
       break;
-    case DB_OPTION_SCHEMALESS:
+      //    case DB_OPTION_SCHEMALESS:
       //      ((SDatabaseOptions*)pOptions)->schemaless = taosStr2Int8(((SToken*)pVal)->z, NULL, 10);
-      ((SDatabaseOptions*)pOptions)->schemaless = 1;
+      //      ((SDatabaseOptions*)pOptions)->schemaless = 0;
-      break;
+      //      break;
     default:
       break;
   }
@@ -867,6 +867,7 @@ SNode* createDefaultTableOptions(SAstCreateContext* pCxt) {
   CHECK_PARSER_STATUS(pCxt);
   STableOptions* pOptions = (STableOptions*)nodesMakeNode(QUERY_NODE_TABLE_OPTIONS);
   CHECK_OUT_OF_MEM(pOptions);
+  // pOptions->delay = TSDB_DEFAULT_ROLLUP_DELAY;
   pOptions->filesFactor = TSDB_DEFAULT_ROLLUP_FILE_FACTOR;
   pOptions->ttl = TSDB_DEFAULT_TABLE_TTL;
   return (SNode*)pOptions;
@@ -876,7 +877,7 @@ SNode* createAlterTableOptions(SAstCreateContext* pCxt) {
   CHECK_PARSER_STATUS(pCxt);
   STableOptions* pOptions = (STableOptions*)nodesMakeNode(QUERY_NODE_TABLE_OPTIONS);
   CHECK_OUT_OF_MEM(pOptions);
-  pOptions->filesFactor = -1;
+  pOptions->delay = -1;
   pOptions->ttl = -1;
   return (SNode*)pOptions;
 }
@@ -890,8 +891,8 @@ SNode* setTableOption(SAstCreateContext* pCxt, SNode* pOptions, ETableOptionType
                                   sizeof(((STableOptions*)pOptions)->comment));
       }
       break;
-    case TABLE_OPTION_FILE_FACTOR:
+    case TABLE_OPTION_DELAY:
-      ((STableOptions*)pOptions)->filesFactor = taosStr2Double(((SToken*)pVal)->z, NULL);
+      ((STableOptions*)pOptions)->delay = taosStr2Int32(((SToken*)pVal)->z, NULL, 10);
       break;
     case TABLE_OPTION_ROLLUP:
       ((STableOptions*)pOptions)->pRollupFuncs = pVal;
@@ -1431,7 +1432,7 @@ SNode* createDropStreamStmt(SAstCreateContext* pCxt, bool ignoreNotExists, const
   CHECK_PARSER_STATUS(pCxt);
   SDropStreamStmt* pStmt = (SDropStreamStmt*)nodesMakeNode(QUERY_NODE_DROP_STREAM_STMT);
   CHECK_OUT_OF_MEM(pStmt);
-  strncpy(pStmt->streamName, pStreamName->z, pStreamName->n);
+  strncpy(pStmt->streamName, pStreamName->z, TMIN(pStreamName->n, sizeof(pStmt->streamName) - 1));
   pStmt->ignoreNotExists = ignoreNotExists;
   return (SNode*)pStmt;
 }
@@ -1444,6 +1445,14 @@ SNode* createKillStmt(SAstCreateContext* pCxt, ENodeType type, const SToken* pId
   return (SNode*)pStmt;
 }
 
+SNode* createKillQueryStmt(SAstCreateContext* pCxt, const SToken* pQueryId) {
+  CHECK_PARSER_STATUS(pCxt);
+  SKillQueryStmt* pStmt = (SKillQueryStmt*)nodesMakeNode(QUERY_NODE_KILL_QUERY_STMT);
+  CHECK_OUT_OF_MEM(pStmt);
+  strncpy(pStmt->queryId, pQueryId->z, TMIN(pQueryId->n, sizeof(pStmt->queryId) - 1));
+  return (SNode*)pStmt;
+}
+
 SNode* createBalanceVgroupStmt(SAstCreateContext* pCxt) {
   CHECK_PARSER_STATUS(pCxt);
   SBalanceVgroupStmt* pStmt = (SBalanceVgroupStmt*)nodesMakeNode(QUERY_NODE_BALANCE_VGROUP_STMT);

source/libs/parser/src/parTokenizer.c

@@ -68,7 +68,7 @@ static SKeyword keywordTable[] = {
     {"CONTAINS",      TK_CONTAINS}, 
     {"DATABASE",      TK_DATABASE},
     {"DATABASES",     TK_DATABASES},
-    {"DAYS",          TK_DAYS},
+    // {"DAYS",          TK_DAYS},
     {"DBS",           TK_DBS},
     {"DELETE",        TK_DELETE},
     {"DESC",          TK_DESC},
@@ -78,6 +78,7 @@ static SKeyword keywordTable[] = {
     {"DNODES",        TK_DNODES},
     {"DOUBLE",        TK_DOUBLE},
     {"DROP",          TK_DROP},
+    {"DURATION",      TK_DURATION},
     {"EXISTS",        TK_EXISTS},
     {"EXPLAIN",       TK_EXPLAIN},
     {"FILE_FACTOR",   TK_FILE_FACTOR},

source/libs/parser/src/parTranslater.c

@@ -2839,6 +2839,9 @@ static int32_t checkCreateTable(STranslateContext* pCxt, SCreateTableStmt* pStmt
   if (TSDB_CODE_SUCCESS == code) {
     code = checTableFactorOption(pCxt, pStmt->pOptions->filesFactor);
   }
+  // if (TSDB_CODE_SUCCESS == code) {
+  //   code = checkRangeOption(pCxt, "delay", pStmt->pOptions->delay, TSDB_MIN_ROLLUP_DELAY, TSDB_MAX_ROLLUP_DELAY);
+  // }
   if (TSDB_CODE_SUCCESS == code) {
     code = checkTableRollupOption(pCxt, pStmt->pOptions->pRollupFuncs);
   }
@@ -3081,6 +3084,7 @@ static int32_t buildRollupAst(STranslateContext* pCxt, SCreateTableStmt* pStmt,
 
 static int32_t buildCreateStbReq(STranslateContext* pCxt, SCreateTableStmt* pStmt, SMCreateStbReq* pReq) {
   pReq->igExists = pStmt->ignoreExists;
+  // pReq->delay = pStmt->pOptions->delay;
   pReq->xFilesFactor = pStmt->pOptions->filesFactor;
   pReq->ttl = pStmt->pOptions->ttl;
   columnDefNodeToField(pStmt->pCols, &pReq->pColumns);
@@ -3628,9 +3632,9 @@ static int32_t translateKillConnection(STranslateContext* pCxt, SKillStmt* pStmt
   return buildCmdMsg(pCxt, TDMT_MND_KILL_CONN, (FSerializeFunc)tSerializeSKillQueryReq, &killReq);
 }
 
-static int32_t translateKillQuery(STranslateContext* pCxt, SKillStmt* pStmt) {
+static int32_t translateKillQuery(STranslateContext* pCxt, SKillQueryStmt* pStmt) {
   SKillQueryReq killReq = {0};
-  killReq.queryId = pStmt->targetId;
+  strcpy(killReq.queryStrId, pStmt->queryId);
   return buildCmdMsg(pCxt, TDMT_MND_KILL_QUERY, (FSerializeFunc)tSerializeSKillQueryReq, &killReq);
 }
 
@@ -3972,7 +3976,7 @@ static int32_t translateQuery(STranslateContext* pCxt, SNode* pNode) {
       code = translateKillConnection(pCxt, (SKillStmt*)pNode);
       break;
     case QUERY_NODE_KILL_QUERY_STMT:
-      code = translateKillQuery(pCxt, (SKillStmt*)pNode);
+      code = translateKillQuery(pCxt, (SKillQueryStmt*)pNode);
       break;
     case QUERY_NODE_KILL_TRANSACTION_STMT:
       code = translateKillTransaction(pCxt, (SKillStmt*)pNode);
@@ -4126,12 +4130,14 @@ static const char* getSysDbName(ENodeType type) {
     case QUERY_NODE_SHOW_SNODES_STMT:
     case QUERY_NODE_SHOW_LICENCE_STMT:
     case QUERY_NODE_SHOW_CLUSTER_STMT:
+    case QUERY_NODE_SHOW_VARIABLE_STMT:
       return TSDB_INFORMATION_SCHEMA_DB;
     case QUERY_NODE_SHOW_CONNECTIONS_STMT:
     case QUERY_NODE_SHOW_QUERIES_STMT:
     case QUERY_NODE_SHOW_TOPICS_STMT:
     case QUERY_NODE_SHOW_STREAMS_STMT:
     case QUERY_NODE_SHOW_TRANSACTIONS_STMT:
+    case QUERY_NODE_SHOW_APPS_STMT:
       return TSDB_PERFORMANCE_SCHEMA_DB;
     default:
       break;
@@ -4181,6 +4187,10 @@ static const char* getSysTableName(ENodeType type) {
       return TSDB_PERFS_TABLE_TOPICS;
     case QUERY_NODE_SHOW_TRANSACTIONS_STMT:
       return TSDB_PERFS_TABLE_TRANS;
+    case QUERY_NODE_SHOW_VARIABLE_STMT:
+      return TSDB_INS_TABLE_CONFIGS;
+    case QUERY_NODE_SHOW_APPS_STMT:
+      return TSDB_PERFS_TABLE_APPS;
     default:
       break;
   }
@@ -5235,6 +5245,8 @@ static int32_t rewriteQuery(STranslateContext* pCxt, SQuery* pQuery) {
     case QUERY_NODE_SHOW_CLUSTER_STMT:
     case QUERY_NODE_SHOW_TOPICS_STMT:
     case QUERY_NODE_SHOW_TRANSACTIONS_STMT:
+    case QUERY_NODE_SHOW_VARIABLE_STMT:
+    case QUERY_NODE_SHOW_APPS_STMT:
       code = rewriteShow(pCxt, pQuery);
       break;
     case QUERY_NODE_CREATE_TABLE_STMT:

source/libs/parser/src/sql.c

@@ -564,7 +564,7 @@ static const YYCODETYPE yy_lookahead[] = {
  /*  1260 */     0,    0,   40,    0,  271,   72,    0,   47,  175,  175,
  /*  1270 */    47,   47,  279,  310,   47,    0,  313,  314,  315,  316,
  /*  1280 */   317,  318,  289,  320,   47,   47,  293,  243,  175,    0,
- /*  1290 */   175,    0,   47,    0,   47,    0,  243,   47,    0,   81,
+ /*  1290 */   175,    0,   47,    0,   22,    0,  243,   47,    0,   81,
  /*  1300 */   113,  160,  156,  310,  159,    0,  313,  314,  315,  316,
  /*  1310 */   317,  318,    0,  320,  152,  271,  323,  151,    0,  356,
  /*  1320 */   357,  328,    0,  279,  271,   44,    0,    0,    0,    0,
@@ -693,7 +693,7 @@ static const unsigned short int yy_shift_ofst[] = {
  /*   310 */   929,  931,  826,  875,  934,  952,  962,  965,  974,  976,
  /*   320 */   859,  935, 1260, 1261, 1222, 1263, 1193, 1266, 1220, 1093,
  /*   330 */  1223, 1224, 1227, 1094, 1275, 1237, 1238, 1113, 1289, 1115,
- /*   340 */  1291, 1245, 1293, 1247, 1295, 1250, 1298, 1218, 1141, 1145,
+ /*   340 */  1291, 1245, 1293, 1272, 1295, 1250, 1298, 1218, 1141, 1145,
  /*   350 */  1187, 1146, 1305, 1312, 1162, 1166, 1318, 1322, 1281, 1326,
  /*   360 */  1327, 1328, 1329, 1330, 1331, 1334, 1335, 1336, 1338, 1339,
  /*   370 */  1340, 1341, 1343, 1344, 1345, 1347, 1348, 1309, 1351, 1352,
@@ -898,7 +898,7 @@ static const YYCODETYPE yyFallback[] = {
     0,  /*     BUFFER => nothing */
     0,  /*  CACHELAST => nothing */
     0,  /*       COMP => nothing */
-    0,  /*       DAYS => nothing */
+    0,  /*   DURATION => nothing */
     0,  /* NK_VARIABLE => nothing */
     0,  /*      FSYNC => nothing */
     0,  /*    MAXROWS => nothing */
@@ -1225,7 +1225,7 @@ static const char *const yyTokenName[] = {
   /*   60 */ "BUFFER",
   /*   61 */ "CACHELAST",
   /*   62 */ "COMP",
-  /*   63 */ "DAYS",
+  /*   63 */ "DURATION",
   /*   64 */ "NK_VARIABLE",
   /*   65 */ "FSYNC",
   /*   66 */ "MAXROWS",
@@ -1600,8 +1600,8 @@ static const char *const yyRuleName[] = {
  /*  68 */ "db_options ::= db_options BUFFER NK_INTEGER",
  /*  69 */ "db_options ::= db_options CACHELAST NK_INTEGER",
  /*  70 */ "db_options ::= db_options COMP NK_INTEGER",
- /*  71 */ "db_options ::= db_options DAYS NK_INTEGER",
+ /*  71 */ "db_options ::= db_options DURATION NK_INTEGER",
- /*  72 */ "db_options ::= db_options DAYS NK_VARIABLE",
+ /*  72 */ "db_options ::= db_options DURATION NK_VARIABLE",
  /*  73 */ "db_options ::= db_options FSYNC NK_INTEGER",
  /*  74 */ "db_options ::= db_options MAXROWS NK_INTEGER",
  /*  75 */ "db_options ::= db_options MINROWS NK_INTEGER",
@@ -1783,7 +1783,7 @@ static const char *const yyRuleName[] = {
  /* 251 */ "stream_options ::= stream_options TRIGGER MAX_DELAY duration_literal",
  /* 252 */ "stream_options ::= stream_options WATERMARK duration_literal",
  /* 253 */ "cmd ::= KILL CONNECTION NK_INTEGER",
- /* 254 */ "cmd ::= KILL QUERY NK_INTEGER",
+ /* 254 */ "cmd ::= KILL QUERY NK_STRING",
  /* 255 */ "cmd ::= KILL TRANSACTION NK_INTEGER",
  /* 256 */ "cmd ::= BALANCE VGROUP",
  /* 257 */ "cmd ::= MERGE VGROUP NK_INTEGER NK_INTEGER",
@@ -2646,8 +2646,8 @@ static const struct {
   {  254,   -3 }, /* (68) db_options ::= db_options BUFFER NK_INTEGER */
   {  254,   -3 }, /* (69) db_options ::= db_options CACHELAST NK_INTEGER */
   {  254,   -3 }, /* (70) db_options ::= db_options COMP NK_INTEGER */
-  {  254,   -3 }, /* (71) db_options ::= db_options DAYS NK_INTEGER */
+  {  254,   -3 }, /* (71) db_options ::= db_options DURATION NK_INTEGER */
-  {  254,   -3 }, /* (72) db_options ::= db_options DAYS NK_VARIABLE */
+  {  254,   -3 }, /* (72) db_options ::= db_options DURATION NK_VARIABLE */
   {  254,   -3 }, /* (73) db_options ::= db_options FSYNC NK_INTEGER */
   {  254,   -3 }, /* (74) db_options ::= db_options MAXROWS NK_INTEGER */
   {  254,   -3 }, /* (75) db_options ::= db_options MINROWS NK_INTEGER */
@@ -2829,7 +2829,7 @@ static const struct {
   {  305,   -4 }, /* (251) stream_options ::= stream_options TRIGGER MAX_DELAY duration_literal */
   {  305,   -3 }, /* (252) stream_options ::= stream_options WATERMARK duration_literal */
   {  240,   -3 }, /* (253) cmd ::= KILL CONNECTION NK_INTEGER */
-  {  240,   -3 }, /* (254) cmd ::= KILL QUERY NK_INTEGER */
+  {  240,   -3 }, /* (254) cmd ::= KILL QUERY NK_STRING */
   {  240,   -3 }, /* (255) cmd ::= KILL TRANSACTION NK_INTEGER */
   {  240,   -2 }, /* (256) cmd ::= BALANCE VGROUP */
   {  240,   -4 }, /* (257) cmd ::= MERGE VGROUP NK_INTEGER NK_INTEGER */
@@ -3326,8 +3326,8 @@ static YYACTIONTYPE yy_reduce(
 { yylhsminor.yy632 = setDatabaseOption(pCxt, yymsp[-2].minor.yy632, DB_OPTION_COMP, &yymsp[0].minor.yy0); }
   yymsp[-2].minor.yy632 = yylhsminor.yy632;
         break;
-      case 71: /* db_options ::= db_options DAYS NK_INTEGER */
+      case 71: /* db_options ::= db_options DURATION NK_INTEGER */
-      case 72: /* db_options ::= db_options DAYS NK_VARIABLE */ yytestcase(yyruleno==72);
+      case 72: /* db_options ::= db_options DURATION NK_VARIABLE */ yytestcase(yyruleno==72);
 { yylhsminor.yy632 = setDatabaseOption(pCxt, yymsp[-2].minor.yy632, DB_OPTION_DAYS, &yymsp[0].minor.yy0); }
   yymsp[-2].minor.yy632 = yylhsminor.yy632;
         break;
@@ -3907,8 +3907,8 @@ static YYACTIONTYPE yy_reduce(
       case 253: /* cmd ::= KILL CONNECTION NK_INTEGER */
 { pCxt->pRootNode = createKillStmt(pCxt, QUERY_NODE_KILL_CONNECTION_STMT, &yymsp[0].minor.yy0); }
         break;
-      case 254: /* cmd ::= KILL QUERY NK_INTEGER */
+      case 254: /* cmd ::= KILL QUERY NK_STRING */
-{ pCxt->pRootNode = createKillStmt(pCxt, QUERY_NODE_KILL_QUERY_STMT, &yymsp[0].minor.yy0); }
+{ pCxt->pRootNode = createKillQueryStmt(pCxt, &yymsp[0].minor.yy0); }
         break;
       case 255: /* cmd ::= KILL TRANSACTION NK_INTEGER */
 { pCxt->pRootNode = createKillStmt(pCxt, QUERY_NODE_KILL_TRANSACTION_STMT, &yymsp[0].minor.yy0); }

source/libs/parser/test/parInitialCTest.cpp

@@ -46,7 +46,7 @@ TEST_F(ParserInitialCTest, createBnode) {
  *     BUFFER value
  *   | CACHELAST value
  *   | COMP {0 | 1 | 2}
- *   | DAYS value
+ *   | DURATION value
  *   | FSYNC value
  *   | MAXROWS value
  *   | MINROWS value
@@ -155,7 +155,7 @@ TEST_F(ParserInitialCTest, createDatabase) {
     ASSERT_EQ(req.replications, expect.replications);
     ASSERT_EQ(req.strict, expect.strict);
     ASSERT_EQ(req.cacheLastRow, expect.cacheLastRow);
-    ASSERT_EQ(req.schemaless, expect.schemaless);
+    // ASSERT_EQ(req.schemaless, expect.schemaless);
     ASSERT_EQ(req.ignoreExist, expect.ignoreExist);
     ASSERT_EQ(req.numOfRetensions, expect.numOfRetensions);
     if (expect.numOfRetensions > 0) {
@@ -202,7 +202,7 @@ TEST_F(ParserInitialCTest, createDatabase) {
       "BUFFER 64 "
       "CACHELAST 2 "
       "COMP 1 "
-      "DAYS 100 "
+      "DURATION 100 "
       "FSYNC 100 "
       "MAXROWS 1000 "
       "MINROWS 100 "
@@ -223,7 +223,7 @@ TEST_F(ParserInitialCTest, createDatabase) {
   setDbDaysFunc(100);
   setDbKeepFunc(1440, 300 * 60, 400 * 1440);
   run("CREATE DATABASE IF NOT EXISTS wxy_db "
-      "DAYS 100m "
+      "DURATION 100m "
       "KEEP 1440m,300h,400d ");
   clearCreateDbReq();
 }

source/libs/planner/src/planOptimizer.c

@@ -104,10 +104,14 @@ static bool osdMayBeOptimized(SLogicNode* pNode) {
     return false;
   }
   if (NULL == pNode->pParent || (QUERY_NODE_LOGIC_PLAN_WINDOW != nodeType(pNode->pParent) &&
-                                 QUERY_NODE_LOGIC_PLAN_AGG != nodeType(pNode->pParent))) {
+                                 QUERY_NODE_LOGIC_PLAN_AGG != nodeType(pNode->pParent) &&
+                                 QUERY_NODE_LOGIC_PLAN_PARTITION != nodeType(pNode->pParent))) {
     return false;
   }
-  if (QUERY_NODE_LOGIC_PLAN_WINDOW == nodeType(pNode->pParent)) {
+  if (QUERY_NODE_LOGIC_PLAN_WINDOW == nodeType(pNode->pParent) ||
+      (QUERY_NODE_LOGIC_PLAN_PARTITION == nodeType(pNode->pParent) &&
+          pNode->pParent->pParent &&
+          QUERY_NODE_LOGIC_PLAN_WINDOW == nodeType(pNode->pParent->pParent)) ) {
     return true;
   }
   return !osdHaveNormalCol(((SAggLogicNode*)pNode->pParent)->pGroupKeys);
@@ -217,16 +221,22 @@ static int32_t osdGetDataRequired(SNodeList* pFuncs) {
 }
 
 static void setScanWindowInfo(SScanLogicNode* pScan) {
-  if (QUERY_NODE_LOGIC_PLAN_WINDOW == nodeType(pScan->node.pParent)) {
+  SLogicNode* pParent = pScan->node.pParent;
-    pScan->interval = ((SWindowLogicNode*)pScan->node.pParent)->interval;
+  if (QUERY_NODE_LOGIC_PLAN_PARTITION == nodeType(pParent) &&
-    pScan->offset = ((SWindowLogicNode*)pScan->node.pParent)->offset;
+          pParent->pParent &&
-    pScan->sliding = ((SWindowLogicNode*)pScan->node.pParent)->sliding;
+          QUERY_NODE_LOGIC_PLAN_WINDOW == nodeType(pParent->pParent)) {
-    pScan->intervalUnit = ((SWindowLogicNode*)pScan->node.pParent)->intervalUnit;
+    pParent = pParent->pParent;
-    pScan->slidingUnit = ((SWindowLogicNode*)pScan->node.pParent)->slidingUnit;
+  }
-    pScan->triggerType = ((SWindowLogicNode*)pScan->node.pParent)->triggerType;
+  if (QUERY_NODE_LOGIC_PLAN_WINDOW == nodeType(pParent)) {
-    pScan->watermark = ((SWindowLogicNode*)pScan->node.pParent)->watermark;
+    pScan->interval = ((SWindowLogicNode*)pParent)->interval;
-    pScan->tsColId = ((SColumnNode*)((SWindowLogicNode*)pScan->node.pParent)->pTspk)->colId;
+    pScan->offset = ((SWindowLogicNode*)pParent)->offset;
-    pScan->filesFactor = ((SWindowLogicNode*)pScan->node.pParent)->filesFactor;
+    pScan->sliding = ((SWindowLogicNode*)pParent)->sliding;
+    pScan->intervalUnit = ((SWindowLogicNode*)pParent)->intervalUnit;
+    pScan->slidingUnit = ((SWindowLogicNode*)pParent)->slidingUnit;
+    pScan->triggerType = ((SWindowLogicNode*)pParent)->triggerType;
+    pScan->watermark = ((SWindowLogicNode*)pParent)->watermark;
+    pScan->tsColId = ((SColumnNode*)((SWindowLogicNode*)pParent)->pTspk)->colId;
+    pScan->filesFactor = ((SWindowLogicNode*)pParent)->filesFactor;
   }
 }
 

source/libs/planner/src/planPhysiCreater.c

@@ -181,7 +181,7 @@ static int16_t getUnsetSlotId(const SArray* pSlotIdsInfo) {
 }
 
 static int32_t addDataBlockSlotsImpl(SPhysiPlanContext* pCxt, SNodeList* pList, SDataBlockDescNode* pDataBlockDesc,
-                                     const char* pStmtName, bool output,  bool reserve) {
+                                     const char* pStmtName, bool output, bool reserve) {
   if (NULL == pList) {
     return TSDB_CODE_SUCCESS;
   }
@@ -463,10 +463,25 @@ static int32_t createTagScanPhysiNode(SPhysiPlanContext* pCxt, SSubplan* pSubpla
   return createScanPhysiNodeFinalize(pCxt, pSubplan, pScanLogicNode, (SScanPhysiNode*)pTagScan, pPhyNode);
 }
 
+static ENodeType getScanOperatorType(EScanType scanType) {
+  switch (scanType) {
+    case SCAN_TYPE_TABLE:
+      return QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN;
+    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_MERGE_SCAN;
+    default:
+      break;
+  }
+  return QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN;
+}
+
 static int32_t createTableScanPhysiNode(SPhysiPlanContext* pCxt, SSubplan* pSubplan, SScanLogicNode* pScanLogicNode,
                                         SPhysiNode** pPhyNode) {
-  STableScanPhysiNode* pTableScan =
+  STableScanPhysiNode* pTableScan = (STableScanPhysiNode*)makePhysiNode(pCxt, (SLogicNode*)pScanLogicNode,
-      (STableScanPhysiNode*)makePhysiNode(pCxt, (SLogicNode*)pScanLogicNode, QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN);
+                                                                        getScanOperatorType(pScanLogicNode->scanType));
   if (NULL == pTableScan) {
     return TSDB_CODE_OUT_OF_MEMORY;
   }
@@ -528,12 +543,12 @@ static int32_t createSystemTableScanPhysiNode(SPhysiPlanContext* pCxt, SSubplan*
 
 static int32_t createStreamScanPhysiNode(SPhysiPlanContext* pCxt, SSubplan* pSubplan, SScanLogicNode* pScanLogicNode,
                                          SPhysiNode** pPhyNode) {
-  int32_t res = createTableScanPhysiNode(pCxt, pSubplan, pScanLogicNode, pPhyNode);
+  return createTableScanPhysiNode(pCxt, pSubplan, pScanLogicNode, pPhyNode);
-  if (res == TSDB_CODE_SUCCESS) {
+}
-    ENodeType type = QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN;
+
-    setNodeType(*pPhyNode, type);
+static int32_t createTableMergeScanPhysiNode(SPhysiPlanContext* pCxt, SSubplan* pSubplan,
-  }
+                                             SScanLogicNode* pScanLogicNode, SPhysiNode** pPhyNode) {
-  return res;
+  return createTableScanPhysiNode(pCxt, pSubplan, pScanLogicNode, pPhyNode);
 }
 
 static int32_t createScanPhysiNode(SPhysiPlanContext* pCxt, SSubplan* pSubplan, SScanLogicNode* pScanLogicNode,
@@ -547,6 +562,8 @@ static int32_t createScanPhysiNode(SPhysiPlanContext* pCxt, SSubplan* pSubplan,
       return createSystemTableScanPhysiNode(pCxt, pSubplan, pScanLogicNode, pPhyNode);
     case SCAN_TYPE_STREAM:
       return createStreamScanPhysiNode(pCxt, pSubplan, pScanLogicNode, pPhyNode);
+    case SCAN_TYPE_TABLE_MERGE:
+      return createTableMergeScanPhysiNode(pCxt, pSubplan, pScanLogicNode, pPhyNode);
     default:
       break;
   }

source/libs/planner/src/planSpliter.c

@@ -15,6 +15,7 @@
 
 #include "functionMgt.h"
 #include "planInt.h"
+#include "tglobal.h"
 
 #define SPLIT_FLAG_MASK(n) (1 << n)
 
@@ -37,7 +38,8 @@ typedef struct SSplitRule {
   FSplit splitFunc;
 } SSplitRule;
 
-typedef bool (*FSplFindSplitNode)(SSplitContext* pCxt, SLogicSubplan* pSubplan, void* pInfo);
+// typedef bool (*FSplFindSplitNode)(SSplitContext* pCxt, SLogicSubplan* pSubplan, void* pInfo);
+typedef bool (*FSplFindSplitNode)(SSplitContext* pCxt, SLogicSubplan* pSubplan, SLogicNode* pNode, void* pInfo);
 
 static void splSetSubplanVgroups(SLogicSubplan* pSubplan, SLogicNode* pNode) {
   if (QUERY_NODE_LOGIC_PLAN_SCAN == nodeType(pNode)) {
@@ -95,9 +97,23 @@ static int32_t splCreateExchangeNodeForSubplan(SSplitContext* pCxt, SLogicSubpla
   return code;
 }
 
+static bool splMatchByNode(SSplitContext* pCxt, SLogicSubplan* pSubplan, SLogicNode* pNode, FSplFindSplitNode func,
+                           void* pInfo) {
+  if (func(pCxt, pSubplan, pNode, pInfo)) {
+    return true;
+  }
+  SNode* pChild;
+  FOREACH(pChild, pNode->pChildren) {
+    if (splMatchByNode(pCxt, pSubplan, (SLogicNode*)pChild, func, pInfo)) {
+      return true;
+    }
+  }
+  return NULL;
+}
+
 static bool splMatch(SSplitContext* pCxt, SLogicSubplan* pSubplan, int32_t flag, FSplFindSplitNode func, void* pInfo) {
   if (!SPLIT_FLAG_TEST_MASK(pSubplan->splitFlag, flag)) {
-    if (func(pCxt, pSubplan, pInfo)) {
+    if (splMatchByNode(pCxt, pSubplan, pSubplan->pNode, func, pInfo)) {
       return true;
     }
   }
@@ -110,6 +126,11 @@ static bool splMatch(SSplitContext* pCxt, SLogicSubplan* pSubplan, int32_t flag,
   return false;
 }
 
+static void splSetParent(SLogicNode* pNode) {
+  SNode* pChild = NULL;
+  FOREACH(pChild, pNode->pChildren) { ((SLogicNode*)pChild)->pParent = pNode; }
+}
+
 typedef struct SStableSplitInfo {
   SLogicNode*    pSplitNode;
   SLogicSubplan* pSubplan;
@@ -136,11 +157,21 @@ static bool stbSplHasMultiTbScan(bool streamQuery, SLogicNode* pNode) {
     return false;
   }
   SNode* pChild = nodesListGetNode(pNode->pChildren, 0);
+  if (QUERY_NODE_LOGIC_PLAN_PARTITION == nodeType(pChild)) {
+    if (1 != LIST_LENGTH(((SLogicNode*)pChild)->pChildren)) {
+      return false;
+    }
+    pChild = nodesListGetNode(((SLogicNode*)pChild)->pChildren, 0);
+  }
   return (QUERY_NODE_LOGIC_PLAN_SCAN == nodeType(pChild) && stbSplIsMultiTbScan(streamQuery, (SScanLogicNode*)pChild));
 }
 
 static bool stbSplNeedSplit(bool streamQuery, SLogicNode* pNode) {
   switch (nodeType(pNode)) {
+    case QUERY_NODE_LOGIC_PLAN_SCAN:
+      return stbSplIsMultiTbScan(streamQuery, (SScanLogicNode*)pNode);
+    case QUERY_NODE_LOGIC_PLAN_JOIN:
+      return !(((SJoinLogicNode*)pNode)->isSingleTableJoin);
     case QUERY_NODE_LOGIC_PLAN_AGG:
       return !stbSplHasGatherExecFunc(((SAggLogicNode*)pNode)->pAggFuncs) && stbSplHasMultiTbScan(streamQuery, pNode);
     case QUERY_NODE_LOGIC_PLAN_WINDOW: {
@@ -152,35 +183,20 @@ static bool stbSplNeedSplit(bool streamQuery, SLogicNode* pNode) {
     }
     case QUERY_NODE_LOGIC_PLAN_SORT:
       return stbSplHasMultiTbScan(streamQuery, pNode);
-    case QUERY_NODE_LOGIC_PLAN_SCAN:
-      return stbSplIsMultiTbScan(streamQuery, (SScanLogicNode*)pNode);
     default:
       break;
   }
   return false;
 }
 
-static SLogicNode* stbSplMatchByNode(bool streamQuery, SLogicNode* pNode) {
+static bool stbSplFindSplitNode(SSplitContext* pCxt, SLogicSubplan* pSubplan, SLogicNode* pNode,
-  if (stbSplNeedSplit(streamQuery, pNode)) {
+                                SStableSplitInfo* pInfo) {
-    return pNode;
+  if (stbSplNeedSplit(pCxt->pPlanCxt->streamQuery, pNode)) {
-  }
+    pInfo->pSplitNode = pNode;
-  SNode* pChild;
-  FOREACH(pChild, pNode->pChildren) {
-    SLogicNode* pSplitNode = stbSplMatchByNode(streamQuery, (SLogicNode*)pChild);
-    if (NULL != pSplitNode) {
-      return pSplitNode;
-    }
-  }
-  return NULL;
-}
-
-static bool stbSplFindSplitNode(SSplitContext* pCxt, SLogicSubplan* pSubplan, SStableSplitInfo* pInfo) {
-  SLogicNode* pSplitNode = stbSplMatchByNode(pCxt->pPlanCxt->streamQuery, pSubplan->pNode);
-  if (NULL != pSplitNode) {
-    pInfo->pSplitNode = pSplitNode;
     pInfo->pSubplan = pSubplan;
+    return true;
   }
-  return NULL != pSplitNode;
+  return false;
 }
 
 static int32_t stbSplRewriteFuns(const SNodeList* pFuncs, SNodeList** pPartialFuncs, SNodeList** pMergeFuncs) {
@@ -258,6 +274,7 @@ static int32_t stbSplCreatePartWindowNode(SWindowLogicNode* pMergeWindow, SLogic
   if (TSDB_CODE_SUCCESS == code) {
     pMergeWindow->node.pTargets = pTargets;
     pPartWin->node.pChildren = pChildren;
+    splSetParent((SLogicNode*)pPartWin);
     code = stbSplRewriteFuns(pFunc, &pPartWin->pFuncs, &pMergeWindow->pFuncs);
   }
   int32_t index = 0;
@@ -285,13 +302,24 @@ static int32_t stbSplCreatePartWindowNode(SWindowLogicNode* pMergeWindow, SLogic
   return code;
 }
 
+static int32_t stbSplGetNumOfVgroups(SLogicNode* pNode) {
+  if (QUERY_NODE_LOGIC_PLAN_SCAN == nodeType(pNode)) {
+    return ((SScanLogicNode*)pNode)->pVgroupList->numOfVgroups;
+  } else {
+    if (1 == LIST_LENGTH(pNode->pChildren)) {
+      return stbSplGetNumOfVgroups((SLogicNode*)nodesListGetNode(pNode->pChildren, 0));
+    }
+  }
+  return 0;
+}
+
 static int32_t stbSplCreateMergeNode(SSplitContext* pCxt, SLogicSubplan* pSubplan, SLogicNode* pSplitNode,
                                      SNodeList* pMergeKeys, SLogicNode* pPartChild) {
   SMergeLogicNode* pMerge = (SMergeLogicNode*)nodesMakeNode(QUERY_NODE_LOGIC_PLAN_MERGE);
   if (NULL == pMerge) {
     return TSDB_CODE_OUT_OF_MEMORY;
   }
-  pMerge->numOfChannels = ((SScanLogicNode*)nodesListGetNode(pPartChild->pChildren, 0))->pVgroupList->numOfVgroups;
+  pMerge->numOfChannels = stbSplGetNumOfVgroups(pPartChild);
   pMerge->srcGroupId = pCxt->groupId;
   pMerge->node.precision = pPartChild->precision;
   pMerge->pMergeKeys = pMergeKeys;
@@ -329,12 +357,12 @@ static int32_t stbSplCreateExchangeNode(SSplitContext* pCxt, SLogicNode* pParent
   return code;
 }
 
-static int32_t stbSplCreateMergeKeysForInterval(SNode* pWStartTs, SNodeList** pMergeKeys) {
+static int32_t stbSplCreateMergeKeysByPrimaryKey(SNode* pPrimaryKey, SNodeList** pMergeKeys) {
   SOrderByExprNode* pMergeKey = (SOrderByExprNode*)nodesMakeNode(QUERY_NODE_ORDER_BY_EXPR);
   if (NULL == pMergeKey) {
     return TSDB_CODE_OUT_OF_MEMORY;
   }
-  pMergeKey->pExpr = nodesCloneNode(pWStartTs);
+  pMergeKey->pExpr = nodesCloneNode(pPrimaryKey);
   if (NULL == pMergeKey->pExpr) {
     nodesDestroyNode((SNode*)pMergeKey);
     return TSDB_CODE_OUT_OF_MEMORY;
@@ -351,7 +379,7 @@ static int32_t stbSplSplitIntervalForBatch(SSplitContext* pCxt, SStableSplitInfo
     ((SWindowLogicNode*)pPartWindow)->intervalAlgo = INTERVAL_ALGO_HASH;
     ((SWindowLogicNode*)pInfo->pSplitNode)->intervalAlgo = INTERVAL_ALGO_MERGE;
     SNodeList* pMergeKeys = NULL;
-    code = stbSplCreateMergeKeysForInterval(((SWindowLogicNode*)pInfo->pSplitNode)->pTspk, &pMergeKeys);
+    code = stbSplCreateMergeKeysByPrimaryKey(((SWindowLogicNode*)pInfo->pSplitNode)->pTspk, &pMergeKeys);
     if (TSDB_CODE_SUCCESS == code) {
       code = stbSplCreateMergeNode(pCxt, NULL, pInfo->pSplitNode, pMergeKeys, pPartWindow);
     }
@@ -364,6 +392,7 @@ static int32_t stbSplSplitIntervalForBatch(SSplitContext* pCxt, SStableSplitInfo
                                      (SNode*)splCreateScanSubplan(pCxt, pPartWindow, SPLIT_FLAG_STABLE_SPLIT));
   }
   pInfo->pSubplan->subplanType = SUBPLAN_TYPE_MERGE;
+  ++(pCxt->groupId);
   return code;
 }
 
@@ -380,6 +409,7 @@ static int32_t stbSplSplitIntervalForStream(SSplitContext* pCxt, SStableSplitInf
                                      (SNode*)splCreateScanSubplan(pCxt, pPartWindow, SPLIT_FLAG_STABLE_SPLIT));
   }
   pInfo->pSubplan->subplanType = SUBPLAN_TYPE_MERGE;
+  ++(pCxt->groupId);
   return code;
 }
 
@@ -439,6 +469,7 @@ static int32_t stbSplCreatePartAggNode(SAggLogicNode* pMergeAgg, SLogicNode** pO
     pMergeAgg->node.pConditions = pConditions;
     pMergeAgg->node.pTargets = pTargets;
     pPartAgg->node.pChildren = pChildren;
+    splSetParent((SLogicNode*)pPartAgg);
 
     code = stbSplRewriteFuns(pFunc, &pPartAgg->pAggFuncs, &pMergeAgg->pAggFuncs);
   }
@@ -467,6 +498,7 @@ static int32_t stbSplSplitAggNode(SSplitContext* pCxt, SStableSplitInfo* pInfo)
                                      (SNode*)splCreateScanSubplan(pCxt, pPartAgg, SPLIT_FLAG_STABLE_SPLIT));
   }
   pInfo->pSubplan->subplanType = SUBPLAN_TYPE_MERGE;
+  ++(pCxt->groupId);
   return code;
 }
 
@@ -553,6 +585,7 @@ static int32_t stbSplCreatePartSortNode(SSortLogicNode* pSort, SLogicNode** pOut
   SNodeList* pMergeKeys = NULL;
   if (TSDB_CODE_SUCCESS == code) {
     pPartSort->node.pChildren = pChildren;
+    splSetParent((SLogicNode*)pPartSort);
     pPartSort->pSortKeys = pSortKeys;
     code = stbSplCreateMergeKeys(pPartSort->pSortKeys, pPartSort->node.pTargets, &pMergeKeys);
   }
@@ -580,6 +613,7 @@ static int32_t stbSplSplitSortNode(SSplitContext* pCxt, SStableSplitInfo* pInfo)
                                      (SNode*)splCreateScanSubplan(pCxt, pPartSort, SPLIT_FLAG_STABLE_SPLIT));
   }
   pInfo->pSubplan->subplanType = SUBPLAN_TYPE_MERGE;
+  ++(pCxt->groupId);
   return code;
 }
 
@@ -589,6 +623,59 @@ static int32_t stbSplSplitScanNode(SSplitContext* pCxt, SStableSplitInfo* pInfo)
     code = nodesListMakeStrictAppend(&pInfo->pSubplan->pChildren,
                                      (SNode*)splCreateScanSubplan(pCxt, pInfo->pSplitNode, SPLIT_FLAG_STABLE_SPLIT));
   }
+  ++(pCxt->groupId);
+  return code;
+}
+
+static SNode* stbSplFindPrimaryKeyFromScan(SScanLogicNode* pScan) {
+  SNode* pCol = NULL;
+  FOREACH(pCol, pScan->pScanCols) {
+    if (PRIMARYKEY_TIMESTAMP_COL_ID == ((SColumnNode*)pCol)->colId) {
+      return pCol;
+    }
+  }
+  return NULL;
+}
+
+static int32_t stbSplSplitScanNodeForJoin(SSplitContext* pCxt, SLogicSubplan* pSubplan, SScanLogicNode* pScan) {
+  SNodeList* pMergeKeys = NULL;
+  int32_t    code = stbSplCreateMergeKeysByPrimaryKey(stbSplFindPrimaryKeyFromScan(pScan), &pMergeKeys);
+  if (TSDB_CODE_SUCCESS == code) {
+    code = stbSplCreateMergeNode(pCxt, pSubplan, (SLogicNode*)pScan, pMergeKeys, (SLogicNode*)pScan);
+  }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = nodesListMakeStrictAppend(&pSubplan->pChildren,
+                                     (SNode*)splCreateScanSubplan(pCxt, (SLogicNode*)pScan, SPLIT_FLAG_STABLE_SPLIT));
+  }
+  pScan->scanType = SCAN_TYPE_TABLE_MERGE;
+  ++(pCxt->groupId);
+  return code;
+}
+
+static int32_t stbSplSplitJoinNodeImpl(SSplitContext* pCxt, SLogicSubplan* pSubplan, SJoinLogicNode* pJoin) {
+  int32_t code = TSDB_CODE_SUCCESS;
+  SNode*  pChild = NULL;
+  FOREACH(pChild, pJoin->node.pChildren) {
+    if (QUERY_NODE_LOGIC_PLAN_SCAN == nodeType(pChild)) {
+      code = stbSplSplitScanNodeForJoin(pCxt, pSubplan, (SScanLogicNode*)pChild);
+    } else if (QUERY_NODE_LOGIC_PLAN_JOIN == nodeType(pChild)) {
+      code = stbSplSplitJoinNodeImpl(pCxt, pSubplan, (SJoinLogicNode*)pChild);
+    } else {
+      code = TSDB_CODE_PLAN_INTERNAL_ERROR;
+    }
+    if (TSDB_CODE_SUCCESS != code) {
+      break;
+    }
+  }
+  return code;
+}
+
+static int32_t stbSplSplitJoinNode(SSplitContext* pCxt, SStableSplitInfo* pInfo) {
+  int32_t code = stbSplSplitJoinNodeImpl(pCxt, pInfo->pSubplan, (SJoinLogicNode*)pInfo->pSplitNode);
+  if (TSDB_CODE_SUCCESS == code) {
+    pInfo->pSubplan->subplanType = SUBPLAN_TYPE_MERGE;
+    SPLIT_FLAG_SET_MASK(pInfo->pSubplan->splitFlag, SPLIT_FLAG_STABLE_SPLIT);
+  }
   return code;
 }
 
@@ -604,6 +691,12 @@ static int32_t stableSplit(SSplitContext* pCxt, SLogicSubplan* pSubplan) {
 
   int32_t code = TSDB_CODE_SUCCESS;
   switch (nodeType(info.pSplitNode)) {
+    case QUERY_NODE_LOGIC_PLAN_SCAN:
+      code = stbSplSplitScanNode(pCxt, &info);
+      break;
+    case QUERY_NODE_LOGIC_PLAN_JOIN:
+      code = stbSplSplitJoinNode(pCxt, &info);
+      break;
     case QUERY_NODE_LOGIC_PLAN_AGG:
       code = stbSplSplitAggNode(pCxt, &info);
       break;
@@ -613,14 +706,10 @@ static int32_t stableSplit(SSplitContext* pCxt, SLogicSubplan* pSubplan) {
     case QUERY_NODE_LOGIC_PLAN_SORT:
       code = stbSplSplitSortNode(pCxt, &info);
       break;
-    case QUERY_NODE_LOGIC_PLAN_SCAN:
-      code = stbSplSplitScanNode(pCxt, &info);
-      break;
     default:
       break;
   }
 
-  ++(pCxt->groupId);
   pCxt->split = true;
   return code;
 }
@@ -631,7 +720,12 @@ typedef struct SSigTbJoinSplitInfo {
   SLogicSubplan*  pSubplan;
 } SSigTbJoinSplitInfo;
 
-static bool sigTbJoinSplNeedSplit(SJoinLogicNode* pJoin) {
+static bool sigTbJoinSplNeedSplit(SLogicNode* pNode) {
+  if (QUERY_NODE_LOGIC_PLAN_JOIN != nodeType(pNode)) {
+    return false;
+  }
+
+  SJoinLogicNode* pJoin = (SJoinLogicNode*)pNode;
   if (!pJoin->isSingleTableJoin) {
     return false;
   }
@@ -639,28 +733,15 @@ static bool sigTbJoinSplNeedSplit(SJoinLogicNode* pJoin) {
          QUERY_NODE_LOGIC_PLAN_EXCHANGE != nodeType(nodesListGetNode(pJoin->node.pChildren, 1));
 }
 
-static SJoinLogicNode* sigTbJoinSplMatchByNode(SLogicNode* pNode) {
+static bool sigTbJoinSplFindSplitNode(SSplitContext* pCxt, SLogicSubplan* pSubplan, SLogicNode* pNode,
-  if (QUERY_NODE_LOGIC_PLAN_JOIN == nodeType(pNode) && sigTbJoinSplNeedSplit((SJoinLogicNode*)pNode)) {
+                                      SSigTbJoinSplitInfo* pInfo) {
-    return (SJoinLogicNode*)pNode;
+  if (sigTbJoinSplNeedSplit(pNode)) {
-  }
+    pInfo->pJoin = (SJoinLogicNode*)pNode;
-  SNode* pChild;
+    pInfo->pSplitNode = (SLogicNode*)nodesListGetNode(pNode->pChildren, 1);
-  FOREACH(pChild, pNode->pChildren) {
-    SJoinLogicNode* pSplitNode = sigTbJoinSplMatchByNode((SLogicNode*)pChild);
-    if (NULL != pSplitNode) {
-      return pSplitNode;
-    }
-  }
-  return NULL;
-}
-
-static bool sigTbJoinSplFindSplitNode(SSplitContext* pCxt, SLogicSubplan* pSubplan, SSigTbJoinSplitInfo* pInfo) {
-  SJoinLogicNode* pJoin = sigTbJoinSplMatchByNode(pSubplan->pNode);
-  if (NULL != pJoin) {
-    pInfo->pJoin = pJoin;
-    pInfo->pSplitNode = (SLogicNode*)nodesListGetNode(pJoin->node.pChildren, 1);
     pInfo->pSubplan = pSubplan;
+    return true;
   }
-  return NULL != pJoin;
+  return false;
 }
 
 static int32_t singleTableJoinSplit(SSplitContext* pCxt, SLogicSubplan* pSubplan) {
@@ -753,27 +834,14 @@ typedef struct SUnionAllSplitInfo {
   SLogicSubplan*     pSubplan;
 } SUnionAllSplitInfo;
 
-static SLogicNode* unAllSplMatchByNode(SLogicNode* pNode) {
+static bool unAllSplFindSplitNode(SSplitContext* pCxt, SLogicSubplan* pSubplan, SLogicNode* pNode,
+                                  SUnionAllSplitInfo* pInfo) {
   if (QUERY_NODE_LOGIC_PLAN_PROJECT == nodeType(pNode) && LIST_LENGTH(pNode->pChildren) > 1) {
-    return pNode;
+    pInfo->pProject = (SProjectLogicNode*)pNode;
-  }
-  SNode* pChild;
-  FOREACH(pChild, pNode->pChildren) {
-    SLogicNode* pSplitNode = unAllSplMatchByNode((SLogicNode*)pChild);
-    if (NULL != pSplitNode) {
-      return pSplitNode;
-    }
-  }
-  return NULL;
-}
-
-static bool unAllSplFindSplitNode(SSplitContext* pCxt, SLogicSubplan* pSubplan, SUnionAllSplitInfo* pInfo) {
-  SLogicNode* pSplitNode = unAllSplMatchByNode(pSubplan->pNode);
-  if (NULL != pSplitNode) {
-    pInfo->pProject = (SProjectLogicNode*)pSplitNode;
     pInfo->pSubplan = pSubplan;
+    return true;
   }
-  return NULL != pSplitNode;
+  return false;
 }
 
 static int32_t unAllSplCreateExchangeNode(SSplitContext* pCxt, SLogicSubplan* pSubplan, SProjectLogicNode* pProject) {
@@ -828,20 +896,6 @@ typedef struct SUnionDistinctSplitInfo {
   SLogicSubplan* pSubplan;
 } SUnionDistinctSplitInfo;
 
-static SLogicNode* unDistSplMatchByNode(SLogicNode* pNode) {
-  if (QUERY_NODE_LOGIC_PLAN_AGG == nodeType(pNode) && LIST_LENGTH(pNode->pChildren) > 1) {
-    return pNode;
-  }
-  SNode* pChild;
-  FOREACH(pChild, pNode->pChildren) {
-    SLogicNode* pSplitNode = unDistSplMatchByNode((SLogicNode*)pChild);
-    if (NULL != pSplitNode) {
-      return pSplitNode;
-    }
-  }
-  return NULL;
-}
-
 static int32_t unDistSplCreateExchangeNode(SSplitContext* pCxt, SLogicSubplan* pSubplan, SAggLogicNode* pAgg) {
   SExchangeLogicNode* pExchange = (SExchangeLogicNode*)nodesMakeNode(QUERY_NODE_LOGIC_PLAN_EXCHANGE);
   if (NULL == pExchange) {
@@ -859,13 +913,14 @@ static int32_t unDistSplCreateExchangeNode(SSplitContext* pCxt, SLogicSubplan* p
   return nodesListMakeAppend(&pAgg->node.pChildren, (SNode*)pExchange);
 }
 
-static bool unDistSplFindSplitNode(SSplitContext* pCxt, SLogicSubplan* pSubplan, SUnionDistinctSplitInfo* pInfo) {
+static bool unDistSplFindSplitNode(SSplitContext* pCxt, SLogicSubplan* pSubplan, SLogicNode* pNode,
-  SLogicNode* pSplitNode = unDistSplMatchByNode(pSubplan->pNode);
+                                   SUnionDistinctSplitInfo* pInfo) {
-  if (NULL != pSplitNode) {
+  if (QUERY_NODE_LOGIC_PLAN_AGG == nodeType(pNode) && LIST_LENGTH(pNode->pChildren) > 1) {
-    pInfo->pAgg = (SAggLogicNode*)pSplitNode;
+    pInfo->pAgg = (SAggLogicNode*)pNode;
     pInfo->pSubplan = pSubplan;
+    return true;
   }
-  return NULL != pSplitNode;
+  return false;
 }
 
 static int32_t unionDistinctSplit(SSplitContext* pCxt, SLogicSubplan* pSubplan) {
@@ -888,27 +943,14 @@ typedef struct SSmaIndexSplitInfo {
   SLogicSubplan*   pSubplan;
 } SSmaIndexSplitInfo;
 
-static SLogicNode* smaIdxSplMatchByNode(SLogicNode* pNode) {
+static bool smaIdxSplFindSplitNode(SSplitContext* pCxt, SLogicSubplan* pSubplan, SLogicNode* pNode,
+                                   SSmaIndexSplitInfo* pInfo) {
   if (QUERY_NODE_LOGIC_PLAN_MERGE == nodeType(pNode) && LIST_LENGTH(pNode->pChildren) > 1) {
-    return pNode;
+    pInfo->pMerge = (SMergeLogicNode*)pNode;
-  }
-  SNode* pChild;
-  FOREACH(pChild, pNode->pChildren) {
-    SLogicNode* pSplitNode = smaIdxSplMatchByNode((SLogicNode*)pChild);
-    if (NULL != pSplitNode) {
-      return pSplitNode;
-    }
-  }
-  return NULL;
-}
-
-static bool smaIdxSplFindSplitNode(SSplitContext* pCxt, SLogicSubplan* pSubplan, SSmaIndexSplitInfo* pInfo) {
-  SLogicNode* pSplitNode = smaIdxSplMatchByNode(pSubplan->pNode);
-  if (NULL != pSplitNode) {
-    pInfo->pMerge = (SMergeLogicNode*)pSplitNode;
     pInfo->pSubplan = pSubplan;
+    return true;
   }
-  return NULL != pSplitNode;
+  return false;
 }
 
 static int32_t smaIndexSplit(SSplitContext* pCxt, SLogicSubplan* pSubplan) {
@@ -926,13 +968,47 @@ static int32_t smaIndexSplit(SSplitContext* pCxt, SLogicSubplan* pSubplan) {
   return code;
 }
 
+typedef struct SQnodeSplitInfo {
+  SLogicNode*    pSplitNode;
+  SLogicSubplan* pSubplan;
+} SQnodeSplitInfo;
+
+static bool qndSplFindSplitNode(SSplitContext* pCxt, SLogicSubplan* pSubplan, SLogicNode* pNode,
+                                SQnodeSplitInfo* pInfo) {
+  if (QUERY_NODE_LOGIC_PLAN_SCAN == nodeType(pNode) && NULL != pNode->pParent) {
+    pInfo->pSplitNode = pNode;
+    pInfo->pSubplan = pSubplan;
+    return true;
+  }
+  return false;
+}
+
+static int32_t qnodeSplit(SSplitContext* pCxt, SLogicSubplan* pSubplan) {
+  if (QUERY_POLICY_QNODE != tsQueryPolicy) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+  SQnodeSplitInfo info = {0};
+  if (!splMatch(pCxt, pSubplan, 0, (FSplFindSplitNode)qndSplFindSplitNode, &info)) {
+    return TSDB_CODE_SUCCESS;
+  }
+  int32_t code = splCreateExchangeNodeForSubplan(pCxt, info.pSubplan, info.pSplitNode, info.pSubplan->subplanType);
+  if (TSDB_CODE_SUCCESS == code) {
+    code = nodesListMakeStrictAppend(&info.pSubplan->pChildren, (SNode*)splCreateScanSubplan(pCxt, info.pSplitNode, 0));
+  }
+  ++(pCxt->groupId);
+  pCxt->split = true;
+  return code;
+}
+
 // clang-format off
 static const SSplitRule splitRuleSet[] = {
   {.pName = "SuperTableSplit",      .splitFunc = stableSplit},
   {.pName = "SingleTableJoinSplit", .splitFunc = singleTableJoinSplit},
   {.pName = "UnionAllSplit",        .splitFunc = unionAllSplit},
   {.pName = "UnionDistinctSplit",   .splitFunc = unionDistinctSplit},
-  {.pName = "SmaIndexSplit",        .splitFunc = smaIndexSplit}
+  {.pName = "SmaIndexSplit",        .splitFunc = smaIndexSplit},
+  {.pName = "QnodeSplit",           .splitFunc = qnodeSplit}
 };
 // clang-format on
 

source/libs/planner/test/planGroupByTest.cpp

@@ -83,5 +83,7 @@ TEST_F(PlanGroupByTest, stable) {
 
   run("SELECT COUNT(*) FROM st1 GROUP BY c1");
 
+  run("SELECT COUNT(*) FROM st1 PARTITION BY c2 GROUP BY c1");
+
   run("SELECT SUM(c1) FROM st1 GROUP BY c2 HAVING SUM(c1) IS NOT NULL");
 }

source/libs/planner/test/planIntervalTest.cpp

@@ -60,4 +60,6 @@ TEST_F(PlanIntervalTest, stable) {
   run("SELECT COUNT(*) FROM st1 INTERVAL(10s)");
 
   run("SELECT _WSTARTTS, COUNT(*) FROM st1 INTERVAL(10s)");
+
+  run("SELECT _WSTARTTS, COUNT(*) FROM st1 PARTITION BY TBNAME INTERVAL(10s)");
 }

source/libs/planner/test/planJoinTest.cpp

@@ -50,3 +50,9 @@ TEST_F(PlanJoinTest, multiJoin) {
 
   run("SELECT t1.c1, t2.c1 FROM st1s1 t1 JOIN st1s2 t2 ON t1.ts = t2.ts JOIN st1s3 t3 ON t1.ts = t3.ts");
 }
+
+TEST_F(PlanJoinTest, stable) {
+  useDb("root", "test");
+
+  run("SELECT t1.c1, t2.c1 FROM st1 t1 JOIN st2 t2 ON t1.ts = t2.ts ");
+}

source/libs/planner/test/planOrderByTest.cpp

@@ -49,4 +49,6 @@ TEST_F(PlanOrderByTest, stable) {
 
   // ORDER BY key is not in the projection list
   run("SELECT c2 FROM st1 ORDER BY c1");
+
+  run("SELECT c2 FROM st1 PARTITION BY c2 ORDER BY c1");
 }

source/libs/planner/test/planOtherTest.cpp

@@ -83,3 +83,10 @@ TEST_F(PlanOtherTest, delete) {
 
   run("DELETE FROM st1 WHERE ts > now - 2d and ts < now - 1d AND tag1 = 10");
 }
+
+TEST_F(PlanOtherTest, queryPolicy) {
+  useDb("root", "test");
+
+  tsQueryPolicy = QUERY_POLICY_QNODE;
+  run("SELECT COUNT(*) FROM st1");
+}

source/libs/planner/test/planTestUtil.h

@@ -18,6 +18,10 @@
 
 #include <gtest/gtest.h>
 
+#define ALLOW_FORBID_FUNC
+
+#include "planInt.h"
+
 class PlannerTestBaseImpl;
 struct TAOS_MULTI_BIND;
 

source/libs/stream/inc/streamInc.h

@@ -23,6 +23,13 @@
 extern "C" {
 #endif
 
+typedef struct {
+  int8_t inited;
+  void*  timer;
+} SStreamGlobalEnv;
+
+static SStreamGlobalEnv streamEnv;
+
 int32_t streamExec(SStreamTask* pTask, SMsgCb* pMsgCb);
 int32_t streamDispatch(SStreamTask* pTask, SMsgCb* pMsgCb);
 int32_t streamDispatchReqToData(const SStreamDispatchReq* pReq, SStreamDataBlock* pData);

source/libs/stream/src/stream.c

@@ -14,8 +14,74 @@
  */
 
 #include "streamInc.h"
+#include "ttimer.h"
 
-int32_t streamTriggerByWrite(SStreamTask* pTask, int32_t vgId, SMsgCb* pMsgCb) {
+int32_t streamInit() {
+  int8_t old;
+  while (1) {
+    old = atomic_val_compare_exchange_8(&streamEnv.inited, 0, 2);
+    if (old != 2) break;
+  }
+
+  if (old == 0) {
+    streamEnv.timer = taosTmrInit(10000, 100, 10000, "STREAM");
+    if (streamEnv.timer == NULL) {
+      atomic_store_8(&streamEnv.inited, 0);
+      return -1;
+    }
+    atomic_store_8(&streamEnv.inited, 1);
+  }
+  return 0;
+}
+
+void streamCleanUp() {
+  int8_t old;
+  while (1) {
+    old = atomic_val_compare_exchange_8(&streamEnv.inited, 1, 2);
+    if (old != 2) break;
+  }
+
+  if (old == 1) {
+    taosTmrCleanUp(streamEnv.timer);
+    atomic_store_8(&streamEnv.inited, 0);
+  }
+}
+
+void streamTriggerByTimer(void* param, void* tmrId) {
+  SStreamTask* pTask = (void*)param;
+
+  if (atomic_load_8(&pTask->triggerStatus) == TASK_TRIGGER_STATUS__ACTIVE) {
+    SStreamTrigger* trigger = taosAllocateQitem(sizeof(SStreamTrigger), DEF_QITEM);
+    if (trigger == NULL) return;
+    trigger->type = STREAM_INPUT__TRIGGER;
+    trigger->pBlock = taosMemoryCalloc(1, sizeof(SSDataBlock));
+    if (trigger->pBlock == NULL) {
+      taosFreeQitem(trigger);
+      return;
+    }
+    trigger->pBlock->info.type = STREAM_GET_ALL;
+
+    atomic_store_8(&pTask->triggerStatus, TASK_TRIGGER_STATUS__IN_ACTIVE);
+
+    streamTaskInput(pTask, (SStreamQueueItem*)trigger);
+    streamLaunchByWrite(pTask, pTask->nodeId, pTask->pMsgCb);
+  }
+
+  taosTmrReset(streamTriggerByTimer, (int32_t)pTask->triggerParam, pTask, streamEnv.timer, &pTask->timer);
+}
+
+int32_t streamSetupTrigger(SStreamTask* pTask) {
+  if (pTask->triggerParam != 0) {
+    if (streamInit() < 0) {
+      return -1;
+    }
+    pTask->timer = taosTmrStart(streamTriggerByTimer, (int32_t)pTask->triggerParam, pTask, streamEnv.timer);
+    pTask->triggerStatus = TASK_TRIGGER_STATUS__IN_ACTIVE;
+  }
+  return 0;
+}
+
+int32_t streamLaunchByWrite(SStreamTask* pTask, int32_t vgId, SMsgCb* pMsgCb) {
   int8_t execStatus = atomic_load_8(&pTask->status);
   if (execStatus == TASK_STATUS__IDLE || execStatus == TASK_STATUS__CLOSING) {
     SStreamTaskRunReq* pRunReq = rpcMallocCont(sizeof(SStreamTaskRunReq));

source/libs/stream/src/streamExec.c

@@ -20,15 +20,17 @@ static int32_t streamTaskExecImpl(SStreamTask* pTask, void* data, SArray* pRes)
   void* exec = pTask->exec.executor;
 
   // set input
-  if (pTask->inputType == STREAM_INPUT__DATA_SUBMIT) {
+  SStreamQueueItem* pItem = (SStreamQueueItem*)data;
+  if (pItem->type == STREAM_INPUT__TRIGGER) {
+    SStreamTrigger* pTrigger = (SStreamTrigger*)data;
+    qSetMultiStreamInput(exec, pTrigger->pBlock, 1, STREAM_DATA_TYPE_SSDATA_BLOCK, false);
+  } else if (pItem->type == STREAM_INPUT__DATA_SUBMIT) {
     SStreamDataSubmit* pSubmit = (SStreamDataSubmit*)data;
-    ASSERT(pSubmit->type == STREAM_INPUT__DATA_SUBMIT);
+    ASSERT(pTask->inputType == STREAM_INPUT__DATA_SUBMIT);
-
     qSetStreamInput(exec, pSubmit->data, STREAM_DATA_TYPE_SUBMIT_BLOCK, false);
-  } else if (pTask->inputType == STREAM_INPUT__DATA_BLOCK) {
+  } else if (pItem->type == STREAM_INPUT__DATA_BLOCK) {
     SStreamDataBlock* pBlock = (SStreamDataBlock*)data;
-    ASSERT(pBlock->type == STREAM_INPUT__DATA_BLOCK);
+    ASSERT(pTask->inputType == STREAM_INPUT__DATA_BLOCK);
-
     SArray* blocks = pBlock->blocks;
     qSetMultiStreamInput(exec, blocks->pData, blocks->size, STREAM_DATA_TYPE_SSDATA_BLOCK, false);
   }

source/libs/stream/src/streamTask.c

@@ -72,6 +72,7 @@ int32_t tEncodeSStreamTask(SEncoder* pEncoder, const SStreamTask* pTask) {
     if (tSerializeSUseDbRspImp(pEncoder, &pTask->shuffleDispatcher.dbInfo) < 0) return -1;
     /*if (tEncodeI8(pEncoder, pTask->shuffleDispatcher.hashMethod) < 0) return -1;*/
   }
+  if (tEncodeI64(pEncoder, pTask->triggerParam) < 0) return -1;
 
   /*tEndEncode(pEncoder);*/
   return pEncoder->pos;
@@ -121,6 +122,7 @@ int32_t tDecodeSStreamTask(SDecoder* pDecoder, SStreamTask* pTask) {
     /*if (tDecodeI8(pDecoder, &pTask->shuffleDispatcher.hashMethod) < 0) return -1;*/
     if (tDeserializeSUseDbRspImp(pDecoder, &pTask->shuffleDispatcher.dbInfo) < 0) return -1;
   }
+  if (tDecodeI64(pDecoder, &pTask->triggerParam) < 0) return -1;
 
   /*tEndDecode(pDecoder);*/
   return 0;

source/libs/sync/src/syncMain.c

@@ -149,12 +149,14 @@ void syncStop(int64_t rid) {
 int32_t syncSetStandby(int64_t rid) {
   SSyncNode* pSyncNode = (SSyncNode*)taosAcquireRef(tsNodeRefId, rid);
   if (pSyncNode == NULL) {
-    return TAOS_SYNC_OTHER_ERROR;
+    terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
+    return -1;
   }
 
   if (pSyncNode->state == TAOS_SYNC_STATE_LEADER) {
     taosReleaseRef(tsNodeRefId, pSyncNode->rid);
-    return TAOS_SYNC_OTHER_ERROR;
+    terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
+    return -1;
   }
 
   // state change
@@ -177,7 +179,8 @@ int32_t syncSetStandby(int64_t rid) {
 int32_t syncReconfigBuild(int64_t rid, const SSyncCfg* pNewCfg, SRpcMsg* pRpcMsg) {
   SSyncNode* pSyncNode = (SSyncNode*)taosAcquireRef(tsNodeRefId, rid);
   if (pSyncNode == NULL) {
-    return TAOS_SYNC_OTHER_ERROR;
+    terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
+    return -1;
   }
   ASSERT(rid == pSyncNode->rid);
 
@@ -201,7 +204,8 @@ int32_t syncReconfigBuild(int64_t rid, const SSyncCfg* pNewCfg, SRpcMsg* pRpcMsg
 
   if (!IamInNew) {
     taosReleaseRef(tsNodeRefId, pSyncNode->rid);
-    return TAOS_SYNC_NOT_IN_NEW_CONFIG;
+    terrno = TSDB_CODE_SYN_NOT_IN_NEW_CONFIG;
+    return -1;
   }
 
   char* newconfig = syncCfg2Str((SSyncCfg*)pNewCfg);
@@ -219,7 +223,8 @@ int32_t syncReconfigBuild(int64_t rid, const SSyncCfg* pNewCfg, SRpcMsg* pRpcMsg
 int32_t syncReconfig(int64_t rid, const SSyncCfg* pNewCfg) {
   SSyncNode* pSyncNode = (SSyncNode*)taosAcquireRef(tsNodeRefId, rid);
   if (pSyncNode == NULL) {
-    return TAOS_SYNC_OTHER_ERROR;
+    terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
+    return -1;
   }
   ASSERT(rid == pSyncNode->rid);
 
@@ -246,7 +251,8 @@ int32_t syncReconfig(int64_t rid, const SSyncCfg* pNewCfg) {
   if (!IamInNew) {
     sError("sync reconfig error, not in new config");
     taosReleaseRef(tsNodeRefId, pSyncNode->rid);
-    return TAOS_SYNC_NOT_IN_NEW_CONFIG;
+    terrno = TSDB_CODE_SYN_NOT_IN_NEW_CONFIG;
+    return -1;
   }
 
   char* newconfig = syncCfg2Str((SSyncCfg*)pNewCfg);
@@ -272,13 +278,15 @@ int32_t syncReconfig(int64_t rid, const SSyncCfg* pNewCfg) {
 int32_t syncLeaderTransfer(int64_t rid) {
   SSyncNode* pSyncNode = (SSyncNode*)taosAcquireRef(tsNodeRefId, rid);
   if (pSyncNode == NULL) {
-    return TAOS_SYNC_OTHER_ERROR;
+    terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
+    return -1;
   }
   ASSERT(rid == pSyncNode->rid);
 
   if (pSyncNode->peersNum == 0) {
     taosReleaseRef(tsNodeRefId, pSyncNode->rid);
-    return TAOS_SYNC_OTHER_ERROR;
+    terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
+    return -1;
   }
 
   SNodeInfo newLeader = (pSyncNode->peersNodeInfo)[0];
@@ -291,7 +299,8 @@ int32_t syncLeaderTransfer(int64_t rid) {
 int32_t syncLeaderTransferTo(int64_t rid, SNodeInfo newLeader) {
   SSyncNode* pSyncNode = (SSyncNode*)taosAcquireRef(tsNodeRefId, rid);
   if (pSyncNode == NULL) {
-    return TAOS_SYNC_OTHER_ERROR;
+    terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
+    return -1;
   }
   ASSERT(rid == pSyncNode->rid);
   int32_t ret = 0;
@@ -299,7 +308,8 @@ int32_t syncLeaderTransferTo(int64_t rid, SNodeInfo newLeader) {
   if (pSyncNode->replicaNum == 1) {
     sError("only one replica, cannot drop leader");
     taosReleaseRef(tsNodeRefId, pSyncNode->rid);
-    return TAOS_SYNC_ONLY_ONE_REPLICA;
+    terrno = TSDB_CODE_SYN_ONE_REPLICA;
+    return -1;
   }
 
   SyncLeaderTransfer* pMsg = syncLeaderTransferBuild(pSyncNode->vgId);
@@ -538,11 +548,12 @@ void setHeartbeatTimerMS(int64_t rid, int32_t hbTimerMS) {
 }
 
 int32_t syncPropose(int64_t rid, const SRpcMsg* pMsg, bool isWeak) {
-  int32_t ret = TAOS_SYNC_PROPOSE_SUCCESS;
+  int32_t ret = 0;
 
   SSyncNode* pSyncNode = taosAcquireRef(tsNodeRefId, rid);
   if (pSyncNode == NULL) {
-    return TAOS_SYNC_OTHER_ERROR;
+    terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
+    return -1;
   }
   assert(rid == pSyncNode->rid);
   sDebug("vgId:%d sync event propose msgType:%s", pSyncNode->vgId, TMSG_INFO(pMsg->msgType));
@@ -553,7 +564,7 @@ int32_t syncPropose(int64_t rid, const SRpcMsg* pMsg, bool isWeak) {
 }
 
 int32_t syncNodePropose(SSyncNode* pSyncNode, const SRpcMsg* pMsg, bool isWeak) {
-  int32_t ret = TAOS_SYNC_PROPOSE_SUCCESS;
+  int32_t ret = 0;
   sDebug("vgId:%d sync event propose msgType:%s", pSyncNode->vgId, TMSG_INFO(pMsg->msgType));
 
   if (pSyncNode->state == TAOS_SYNC_STATE_LEADER) {
@@ -561,22 +572,23 @@ int32_t syncNodePropose(SSyncNode* pSyncNode, const SRpcMsg* pMsg, bool isWeak)
     stub.createTime = taosGetTimestampMs();
     stub.rpcMsg = *pMsg;
     uint64_t seqNum = syncRespMgrAdd(pSyncNode->pSyncRespMgr, &stub);
-    sDebug("vgId:%d sync event propose, type:%s seq:%" PRIu64 " handle:%p", pSyncNode->vgId, TMSG_INFO(pMsg->msgType),
-           seqNum, pMsg->info.handle);
 
     SyncClientRequest* pSyncMsg = syncClientRequestBuild2(pMsg, seqNum, isWeak, pSyncNode->vgId);
     SRpcMsg            rpcMsg;
     syncClientRequest2RpcMsg(pSyncMsg, &rpcMsg);
 
     if (pSyncNode->FpEqMsg != NULL && (*pSyncNode->FpEqMsg)(pSyncNode->msgcb, &rpcMsg) == 0) {
-      ret = TAOS_SYNC_PROPOSE_SUCCESS;
+      ret = 0;
     } else {
+      ret = -1;
+      terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
       sError("syncPropose pSyncNode->FpEqMsg is NULL");
     }
     syncClientRequestDestroy(pSyncMsg);
   } else {
+    ret = -1;
+    terrno = TSDB_CODE_SYN_NOT_LEADER;
     sError("syncPropose not leader, %s", syncUtilState2String(pSyncNode->state));
-    ret = TAOS_SYNC_PROPOSE_NOT_LEADER;
   }
 
   return ret;
@@ -771,7 +783,7 @@ SSyncNode* syncNodeOpen(const SSyncInfo* pOldSyncInfo) {
   }
 
   // tools
-  pSyncNode->pSyncRespMgr = syncRespMgrCreate(NULL, 0);
+  pSyncNode->pSyncRespMgr = syncRespMgrCreate(pSyncNode, 0);
   assert(pSyncNode->pSyncRespMgr != NULL);
 
   // restore state
@@ -947,9 +959,13 @@ int32_t syncNodePingAll(SSyncNode* pSyncNode) {
 // timer control --------------
 int32_t syncNodeStartPingTimer(SSyncNode* pSyncNode) {
   int32_t ret = 0;
-  taosTmrReset(pSyncNode->FpPingTimerCB, pSyncNode->pingTimerMS, pSyncNode, gSyncEnv->pTimerManager,
+  if (syncEnvIsStart()) {
-               &pSyncNode->pPingTimer);
+    taosTmrReset(pSyncNode->FpPingTimerCB, pSyncNode->pingTimerMS, pSyncNode, gSyncEnv->pTimerManager,
-  atomic_store_64(&pSyncNode->pingTimerLogicClock, pSyncNode->pingTimerLogicClockUser);
+                 &pSyncNode->pPingTimer);
+    atomic_store_64(&pSyncNode->pingTimerLogicClock, pSyncNode->pingTimerLogicClockUser);
+  } else {
+    sError("sync env is stop, syncNodeStartPingTimer");
+  }
   return ret;
 }
 
@@ -963,10 +979,14 @@ int32_t syncNodeStopPingTimer(SSyncNode* pSyncNode) {
 
 int32_t syncNodeStartElectTimer(SSyncNode* pSyncNode, int32_t ms) {
   int32_t ret = 0;
-  pSyncNode->electTimerMS = ms;
+  if (syncEnvIsStart()) {
-  taosTmrReset(pSyncNode->FpElectTimerCB, pSyncNode->electTimerMS, pSyncNode, gSyncEnv->pTimerManager,
+    pSyncNode->electTimerMS = ms;
-               &pSyncNode->pElectTimer);
+    taosTmrReset(pSyncNode->FpElectTimerCB, pSyncNode->electTimerMS, pSyncNode, gSyncEnv->pTimerManager,
-  atomic_store_64(&pSyncNode->electTimerLogicClock, pSyncNode->electTimerLogicClockUser);
+                 &pSyncNode->pElectTimer);
+    atomic_store_64(&pSyncNode->electTimerLogicClock, pSyncNode->electTimerLogicClockUser);
+  } else {
+    sError("sync env is stop, syncNodeStartElectTimer");
+  }
   return ret;
 }
 
@@ -1000,9 +1020,13 @@ int32_t syncNodeResetElectTimer(SSyncNode* pSyncNode) {
 
 int32_t syncNodeStartHeartbeatTimer(SSyncNode* pSyncNode) {
   int32_t ret = 0;
-  taosTmrReset(pSyncNode->FpHeartbeatTimerCB, pSyncNode->heartbeatTimerMS, pSyncNode, gSyncEnv->pTimerManager,
+  if (syncEnvIsStart()) {
-               &pSyncNode->pHeartbeatTimer);
+    taosTmrReset(pSyncNode->FpHeartbeatTimerCB, pSyncNode->heartbeatTimerMS, pSyncNode, gSyncEnv->pTimerManager,
-  atomic_store_64(&pSyncNode->heartbeatTimerLogicClock, pSyncNode->heartbeatTimerLogicClockUser);
+                 &pSyncNode->pHeartbeatTimer);
+    atomic_store_64(&pSyncNode->heartbeatTimerLogicClock, pSyncNode->heartbeatTimerLogicClockUser);
+  } else {
+    sError("sync env is stop, syncNodeStartHeartbeatTimer");
+  }
   return ret;
 }
 
@@ -1722,14 +1746,25 @@ static void syncNodeEqPingTimer(void* param, void* tmrId) {
     syncTimeout2RpcMsg(pSyncMsg, &rpcMsg);
     syncRpcMsgLog2((char*)"==syncNodeEqPingTimer==", &rpcMsg);
     if (pSyncNode->FpEqMsg != NULL) {
-      pSyncNode->FpEqMsg(pSyncNode->msgcb, &rpcMsg);
+      int32_t code = pSyncNode->FpEqMsg(pSyncNode->msgcb, &rpcMsg);
+      if (code != 0) {
+        sError("vgId:%d sync enqueue ping msg error, code:%d", pSyncNode->vgId, code);
+        rpcFreeCont(rpcMsg.pCont);
+        syncTimeoutDestroy(pSyncMsg);
+        return;
+      }
     } else {
       sTrace("syncNodeEqPingTimer pSyncNode->FpEqMsg is NULL");
     }
     syncTimeoutDestroy(pSyncMsg);
 
-    taosTmrReset(syncNodeEqPingTimer, pSyncNode->pingTimerMS, pSyncNode, gSyncEnv->pTimerManager,
+    if (syncEnvIsStart()) {
-                 &pSyncNode->pPingTimer);
+      taosTmrReset(syncNodeEqPingTimer, pSyncNode->pingTimerMS, pSyncNode, gSyncEnv->pTimerManager,
+                   &pSyncNode->pPingTimer);
+    } else {
+      sError("sync env is stop, syncNodeEqPingTimer");
+    }
+
   } else {
     sTrace("==syncNodeEqPingTimer== pingTimerLogicClock:%" PRIu64 ", pingTimerLogicClockUser:%" PRIu64 "",
            pSyncNode->pingTimerLogicClock, pSyncNode->pingTimerLogicClockUser);
@@ -1745,16 +1780,26 @@ static void syncNodeEqElectTimer(void* param, void* tmrId) {
     syncTimeout2RpcMsg(pSyncMsg, &rpcMsg);
     syncRpcMsgLog2((char*)"==syncNodeEqElectTimer==", &rpcMsg);
     if (pSyncNode->FpEqMsg != NULL) {
-      pSyncNode->FpEqMsg(pSyncNode->msgcb, &rpcMsg);
+      int32_t code = pSyncNode->FpEqMsg(pSyncNode->msgcb, &rpcMsg);
+      if (code != 0) {
+        sError("vgId:%d sync enqueue elect msg error, code:%d", pSyncNode->vgId, code);
+        rpcFreeCont(rpcMsg.pCont);
+        syncTimeoutDestroy(pSyncMsg);
+        return;
+      }
     } else {
-      sTrace("syncNodeEqElectTimer pSyncNode->FpEqMsg is NULL");
+      sTrace("syncNodeEqElectTimer FpEqMsg is NULL");
     }
     syncTimeoutDestroy(pSyncMsg);
 
     // reset timer ms
-    pSyncNode->electTimerMS = syncUtilElectRandomMS(pSyncNode->electBaseLine, 2 * pSyncNode->electBaseLine);
+    if (syncEnvIsStart()) {
-    taosTmrReset(syncNodeEqElectTimer, pSyncNode->electTimerMS, pSyncNode, gSyncEnv->pTimerManager,
+      pSyncNode->electTimerMS = syncUtilElectRandomMS(pSyncNode->electBaseLine, 2 * pSyncNode->electBaseLine);
-                 &pSyncNode->pElectTimer);
+      taosTmrReset(syncNodeEqElectTimer, pSyncNode->electTimerMS, pSyncNode, gSyncEnv->pTimerManager,
+                   &pSyncNode->pElectTimer);
+    } else {
+      sError("sync env is stop, syncNodeEqElectTimer");
+    }
   } else {
     sTrace("==syncNodeEqElectTimer== electTimerLogicClock:%" PRIu64 ", electTimerLogicClockUser:%" PRIu64 "",
            pSyncNode->electTimerLogicClock, pSyncNode->electTimerLogicClockUser);
@@ -1776,19 +1821,19 @@ static void syncNodeEqHeartbeatTimer(void* param, void* tmrId) {
       if (code != 0) {
         sError("vgId:%d sync enqueue timer msg error, code:%d", pSyncNode->vgId, code);
         rpcFreeCont(rpcMsg.pCont);
+        syncTimeoutDestroy(pSyncMsg);
         return;
       }
-
     } else {
-      sTrace("syncNodeEqHeartbeatTimer pSyncNode->FpEqMsg is NULL");
+      sError("syncNodeEqHeartbeatTimer FpEqMsg is NULL");
     }
     syncTimeoutDestroy(pSyncMsg);
 
-    if (gSyncEnv != NULL) {
+    if (syncEnvIsStart()) {
-    taosTmrReset(syncNodeEqHeartbeatTimer, pSyncNode->heartbeatTimerMS, pSyncNode, gSyncEnv->pTimerManager,
+      taosTmrReset(syncNodeEqHeartbeatTimer, pSyncNode->heartbeatTimerMS, pSyncNode, gSyncEnv->pTimerManager,
-                 &pSyncNode->pHeartbeatTimer);
+                   &pSyncNode->pHeartbeatTimer);
     } else {
-      sError("sync env is already stop");
+      sError("sync env is stop, syncNodeEqHeartbeatTimer");
     }
   } else {
     sTrace("==syncNodeEqHeartbeatTimer== heartbeatTimerLogicClock:%" PRIu64 ", heartbeatTimerLogicClockUser:%" PRIu64

source/libs/sync/src/syncRespMgr.c

@@ -44,6 +44,10 @@ int64_t syncRespMgrAdd(SSyncRespMgr *pObj, SRespStub *pStub) {
   uint64_t keyCode = ++(pObj->seqNum);
   taosHashPut(pObj->pRespHash, &keyCode, sizeof(keyCode), pStub, sizeof(SRespStub));
 
+  SSyncNode *pSyncNode = pObj->data;
+  sDebug("vgId:%d sync event resp mgr add, type:%s seq:%lu handle:%p", pSyncNode->vgId,
+         TMSG_INFO(pStub->rpcMsg.msgType), keyCode, pStub->rpcMsg.info.handle);
+
   taosThreadMutexUnlock(&(pObj->mutex));
   return keyCode;
 }
@@ -63,6 +67,11 @@ int32_t syncRespMgrGet(SSyncRespMgr *pObj, uint64_t index, SRespStub *pStub) {
   void *pTmp = taosHashGet(pObj->pRespHash, &index, sizeof(index));
   if (pTmp != NULL) {
     memcpy(pStub, pTmp, sizeof(SRespStub));
+
+    SSyncNode *pSyncNode = pObj->data;
+    sDebug("vgId:%d sync event resp mgr get, type:%s seq:%lu handle:%p", pSyncNode->vgId,
+           TMSG_INFO(pStub->rpcMsg.msgType), index, pStub->rpcMsg.info.handle);
+
     taosThreadMutexUnlock(&(pObj->mutex));
     return 1;  // get one object
   }
@@ -76,6 +85,11 @@ int32_t syncRespMgrGetAndDel(SSyncRespMgr *pObj, uint64_t index, SRespStub *pStu
   void *pTmp = taosHashGet(pObj->pRespHash, &index, sizeof(index));
   if (pTmp != NULL) {
     memcpy(pStub, pTmp, sizeof(SRespStub));
+
+    SSyncNode *pSyncNode = pObj->data;
+    sDebug("vgId:%d sync event resp mgr get and del, type:%s seq:%lu handle:%p", pSyncNode->vgId,
+           TMSG_INFO(pStub->rpcMsg.msgType), index, pStub->rpcMsg.info.handle);
+
     taosHashRemove(pObj->pRespHash, &index, sizeof(index));
     taosThreadMutexUnlock(&(pObj->mutex));
     return 1;  // get one object

source/libs/sync/src/syncUtil.c

@@ -14,6 +14,8 @@
  */
 
 #include "syncUtil.h"
+#include <stdio.h>
+
 #include "syncEnv.h"
 
 void addEpIntoEpSet(SEpSet* pEpSet, const char* fqdn, uint16_t port);
@@ -21,8 +23,31 @@ void addEpIntoEpSet(SEpSet* pEpSet, const char* fqdn, uint16_t port);
 // ---- encode / decode
 uint64_t syncUtilAddr2U64(const char* host, uint16_t port) {
   uint64_t u64;
+
+  uint32_t hostU32 = taosGetIpv4FromFqdn(host);
+  if (hostU32 == (uint32_t)-1) {
+    sError("Get IP address error");
+    return -1;
+  }
+
+  /*
   uint32_t hostU32 = (uint32_t)taosInetAddr(host);
-  // assert(hostU32 != (uint32_t)-1);
+  if (hostU32 == (uint32_t)-1) {
+    struct hostent* hostEnt = gethostbyname(host);
+    if (hostEnt == NULL) {
+      sError("Get IP address error");
+      return -1;
+    }
+
+    const char* newHost = taosInetNtoa(*(struct in_addr*)(hostEnt->h_addr_list[0]));
+    hostU32 = (uint32_t)taosInetAddr(newHost);
+    if (hostU32 == (uint32_t)-1) {
+      sError("change %s to id, error", newHost);
+    }
+    // ASSERT(hostU32 != (uint32_t)-1);
+  }
+  */
+
   u64 = (((uint64_t)hostU32) << 32) | (((uint32_t)port) << 16);
   return u64;
 }

source/libs/sync/test/syncConfigChangeSnapshotTest.cpp

@@ -338,7 +338,7 @@ int main(int argc, char** argv) {
     if (alreadySend < writeRecordNum) {
       SRpcMsg* pRpcMsg = createRpcMsg(alreadySend, writeRecordNum, myIndex);
       int32_t  ret = syncPropose(rid, pRpcMsg, false);
-      if (ret == TAOS_SYNC_PROPOSE_NOT_LEADER) {
+      if (ret == -1 && terrno == TSDB_CODE_SYN_NOT_LEADER) {
         sTrace("%s value%d write not leader", s, alreadySend);
       } else {
         assert(ret == 0);

source/libs/sync/test/syncConfigChangeTest.cpp

@@ -251,7 +251,7 @@ int main(int argc, char** argv) {
     if (alreadySend < writeRecordNum) {
       SRpcMsg* pRpcMsg = createRpcMsg(alreadySend, writeRecordNum, myIndex);
       int32_t  ret = syncPropose(rid, pRpcMsg, false);
-      if (ret == TAOS_SYNC_PROPOSE_NOT_LEADER) {
+      if (ret == -1 && terrno == TSDB_CODE_SYN_NOT_LEADER) {
         sTrace("%s value%d write not leader", s, alreadySend);
       } else {
         assert(ret == 0);

source/libs/sync/test/syncReplicateTest.cpp

@@ -188,7 +188,7 @@ int main(int argc, char** argv) {
     if (alreadySend < writeRecordNum) {
       SRpcMsg* pRpcMsg = createRpcMsg(alreadySend, writeRecordNum, myIndex);
       int32_t  ret = syncPropose(rid, pRpcMsg, false);
-      if (ret == TAOS_SYNC_PROPOSE_NOT_LEADER) {
+      if (ret == -1 && terrno == TSDB_CODE_SYN_NOT_LEADER) {
         sTrace("%s value%d write not leader", s, alreadySend);
       } else {
         assert(ret == 0);

source/libs/sync/test/syncTestTool.cpp

@@ -391,7 +391,7 @@ int main(int argc, char** argv) {
     if (alreadySend < writeRecordNum) {
       SRpcMsg* pRpcMsg = createRpcMsg(alreadySend, writeRecordNum, myIndex);
       int32_t  ret = syncPropose(rid, pRpcMsg, false);
-      if (ret == TAOS_SYNC_PROPOSE_NOT_LEADER) {
+      if (ret == -1 && terrno == TSDB_CODE_SYN_NOT_LEADER) {
         sTrace("%s value%d write not leader, leaderTransferWait:%d", simpleStr, alreadySend, leaderTransferWait);
       } else {
         assert(ret == 0);

source/libs/transport/src/transCli.c

@@ -41,7 +41,7 @@ typedef struct SCliConn {
 
   // debug and log info
   struct sockaddr_in addr;
-  struct sockaddr_in locaddr;
+  struct sockaddr_in localAddr;
 } SCliConn;
 
 typedef struct SCliMsg {
@@ -54,7 +54,8 @@ typedef struct SCliMsg {
 } SCliMsg;
 
 typedef struct SCliThrdObj {
-  TdThread    thread;
+  TdThread    thread;  // tid
+  int64_t     pid;     // pid
   uv_loop_t*  loop;
   SAsyncPool* asyncPool;
   uv_timer_t  timer;
@@ -325,7 +326,7 @@ void cliHandleResp(SCliConn* conn) {
 
   tDebug("%s cli conn %p %s received from %s:%d, local info: %s:%d, msg size: %d", pTransInst->label, conn,
          TMSG_INFO(pHead->msgType), taosInetNtoa(conn->addr.sin_addr), ntohs(conn->addr.sin_port),
-         taosInetNtoa(conn->locaddr.sin_addr), ntohs(conn->locaddr.sin_port), transMsg.contLen);
+         taosInetNtoa(conn->localAddr.sin_addr), ntohs(conn->localAddr.sin_port), transMsg.contLen);
 
   if (pCtx == NULL && CONN_NO_PERSIST_BY_APP(conn)) {
     tTrace("except, server continue send while cli ignore it");
@@ -643,7 +644,7 @@ void cliSend(SCliConn* pConn) {
   uv_buf_t wb = uv_buf_init((char*)pHead, msgLen);
   tDebug("%s cli conn %p %s is send to %s:%d, local info %s:%d", CONN_GET_INST_LABEL(pConn), pConn,
          TMSG_INFO(pHead->msgType), taosInetNtoa(pConn->addr.sin_addr), ntohs(pConn->addr.sin_port),
-         taosInetNtoa(pConn->locaddr.sin_addr), ntohs(pConn->locaddr.sin_port));
+         taosInetNtoa(pConn->localAddr.sin_addr), ntohs(pConn->localAddr.sin_port));
 
   if (pHead->persist == 1) {
     CONN_SET_PERSIST_BY_APP(pConn);
@@ -668,8 +669,8 @@ void cliConnCb(uv_connect_t* req, int status) {
   int addrlen = sizeof(pConn->addr);
   uv_tcp_getpeername((uv_tcp_t*)pConn->stream, (struct sockaddr*)&pConn->addr, &addrlen);
 
-  addrlen = sizeof(pConn->locaddr);
+  addrlen = sizeof(pConn->localAddr);
-  uv_tcp_getsockname((uv_tcp_t*)pConn->stream, (struct sockaddr*)&pConn->locaddr, &addrlen);
+  uv_tcp_getsockname((uv_tcp_t*)pConn->stream, (struct sockaddr*)&pConn->localAddr, &addrlen);
 
   tTrace("%s cli conn %p connect to server successfully", CONN_GET_INST_LABEL(pConn), pConn);
   assert(pConn->stream == req->handle);
@@ -742,8 +743,7 @@ void cliMayCvtFqdnToIp(SEpSet* pEpSet, SCvtAddr* pCvtAddr) {
 void cliHandleReq(SCliMsg* pMsg, SCliThrdObj* pThrd) {
   uint64_t et = taosGetTimestampUs();
   uint64_t el = et - pMsg->st;
-  tTrace("%s cli msg tran time cost: %" PRIu64 "us, threadID: %" PRId64 "", ((STrans*)pThrd->pTransInst)->label, el,
+  // tTrace("%s cli msg tran time cost: %" PRIu64 "us", ((STrans*)pThrd->pTransInst)->label, el);
-         pThrd->thread);
 
   STransConnCtx* pCtx = pMsg->ctx;
   STrans*        pTransInst = pThrd->pTransInst;
@@ -822,6 +822,7 @@ static void cliAsyncCb(uv_async_t* handle) {
 
 static void* cliWorkThread(void* arg) {
   SCliThrdObj* pThrd = (SCliThrdObj*)arg;
+  pThrd->pid = taosGetSelfPthreadId();
   setThreadName("trans-cli-work");
   uv_run(pThrd->loop, UV_RUN_DEFAULT);
   return NULL;
@@ -966,30 +967,31 @@ int cliAppCb(SCliConn* pConn, STransMsg* pResp, SCliMsg* pMsg) {
     pMsg->st = taosGetTimestampUs();
     pCtx->retryCount += 1;
     if (pResp->code == TSDB_CODE_RPC_NETWORK_UNAVAIL) {
-      if (pCtx->retryCount < pEpSet->numOfEps) {
+      if (pCtx->retryCount < pEpSet->numOfEps * 3) {
         pEpSet->inUse = (++pEpSet->inUse) % pEpSet->numOfEps;
 
         STaskArg* arg = taosMemoryMalloc(sizeof(STaskArg));
         arg->param1 = pMsg;
         arg->param2 = pThrd;
         transDQSched(pThrd->delayQueue, doDelayTask, arg, TRANS_RETRY_INTERVAL);
+        tTrace("use local epset, current in use: %d, retry count:%d, limit: %d", pEpSet->inUse, pCtx->retryCount + 1,
+               pEpSet->numOfEps * 3);
         transUnrefCliHandle(pConn);
         return -1;
       }
     } else if (pCtx->retryCount < TRANS_RETRY_COUNT_LIMIT) {
       if (pResp->contLen == 0) {
         pEpSet->inUse = (++pEpSet->inUse) % pEpSet->numOfEps;
+        tTrace("use local epset, current in use: %d, retry count:%d, limit: %d", pEpSet->inUse, pCtx->retryCount + 1,
+               TRANS_RETRY_COUNT_LIMIT);
       } else {
         SEpSet epSet = {0};
         tDeserializeSEpSet(pResp->pCont, pResp->contLen, &epSet);
         pCtx->epSet = epSet;
-        if (!transEpSetIsEqual(&epSet, &pCtx->epSet)) {
+        tTrace("use remote epset, current in use: %d, retry count:%d, limit: %d", pEpSet->inUse, pCtx->retryCount + 1,
-          pCtx->retryCount = 0;
+               TRANS_RETRY_COUNT_LIMIT);
-        }
       }
       addConnToPool(pThrd->pool, pConn);
-      tTrace("use remote epset, current in use: %d, retry count:%d, try limit: %d", pEpSet->inUse, pCtx->retryCount + 1,
-             TRANS_RETRY_COUNT_LIMIT);
 
       STaskArg* arg = taosMemoryMalloc(sizeof(STaskArg));
       arg->param1 = pMsg;
@@ -1089,7 +1091,7 @@ void transSendRequest(void* shandle, const SEpSet* pEpSet, STransMsg* pReq, STra
 
   SCliThrdObj* thrd = ((SCliObj*)pTransInst->tcphandle)->pThreadObj[index];
 
-  tDebug("send request at thread:%d, threadID: %" PRId64 ",  msg: %p, dst: %s:%d, app:%p", index, thrd->thread, pReq,
+  tDebug("send request at thread:%d, threadID: %08" PRId64 ",  msg: %p, dst: %s:%d, app:%p", index, thrd->pid, pReq,
          EPSET_GET_INUSE_IP(&pCtx->epSet), EPSET_GET_INUSE_PORT(&pCtx->epSet), pReq->info.ahandle);
   ASSERT(transSendAsync(thrd->asyncPool, &(cliMsg->q)) == 0);
 }
@@ -1118,7 +1120,7 @@ void transSendRecv(void* shandle, const SEpSet* pEpSet, STransMsg* pReq, STransM
   cliMsg->type = Normal;
 
   SCliThrdObj* thrd = ((SCliObj*)pTransInst->tcphandle)->pThreadObj[index];
-  tDebug("send request at thread:%d, threadID:%" PRId64 ", msg: %p, dst: %s:%d, app:%p", index, thrd->thread, pReq,
+  tDebug("send request at thread:%d, threadID:%08" PRId64 ", msg: %p, dst: %s:%d, app:%p", index, thrd->pid, pReq,
          EPSET_GET_INUSE_IP(&pCtx->epSet), EPSET_GET_INUSE_PORT(&pCtx->epSet), pReq->info.ahandle);
 
   transSendAsync(thrd->asyncPool, &(cliMsg->q));
@@ -1149,7 +1151,7 @@ void transSetDefaultAddr(void* ahandle, const char* ip, const char* fqdn) {
     cliMsg->type = Update;
 
     SCliThrdObj* thrd = ((SCliObj*)pTransInst->tcphandle)->pThreadObj[i];
-    tDebug("update epset at thread:%d, threadID:%" PRId64 "", i, thrd->thread);
+    tDebug("update epset at thread:%d, threadID:%08" PRId64 "", i, thrd->pid);
 
     transSendAsync(thrd->asyncPool, &(cliMsg->q));
   }

source/libs/transport/src/transSvr.c

@@ -48,7 +48,7 @@ typedef struct SSvrConn {
 
   ConnStatus         status;
   struct sockaddr_in addr;
-  struct sockaddr_in locaddr;
+  struct sockaddr_in localAddr;
 
   int64_t refId;
   int     spi;
@@ -286,12 +286,12 @@ static void uvHandleReq(SSvrConn* pConn) {
   if (pConn->status == ConnNormal && pHead->noResp == 0) {
     transRefSrvHandle(pConn);
     tDebug("server conn %p %s received from %s:%d, local info: %s:%d, msg size: %d", pConn, TMSG_INFO(transMsg.msgType),
-           taosInetNtoa(pConn->addr.sin_addr), ntohs(pConn->addr.sin_port), taosInetNtoa(pConn->locaddr.sin_addr),
+           taosInetNtoa(pConn->addr.sin_addr), ntohs(pConn->addr.sin_port), taosInetNtoa(pConn->localAddr.sin_addr),
-           ntohs(pConn->locaddr.sin_port), transMsg.contLen);
+           ntohs(pConn->localAddr.sin_port), transMsg.contLen);
   } else {
     tDebug("server conn %p %s received from %s:%d, local info: %s:%d, msg size: %d, resp:%d ", pConn,
            TMSG_INFO(transMsg.msgType), taosInetNtoa(pConn->addr.sin_addr), ntohs(pConn->addr.sin_port),
-           taosInetNtoa(pConn->locaddr.sin_addr), ntohs(pConn->locaddr.sin_port), transMsg.contLen, pHead->noResp);
+           taosInetNtoa(pConn->localAddr.sin_addr), ntohs(pConn->localAddr.sin_port), transMsg.contLen, pHead->noResp);
     // no ref here
   }
 
@@ -454,8 +454,8 @@ static void uvPrepareSendData(SSvrMsg* smsg, uv_buf_t* wb) {
   char*   msg = (char*)pHead;
   int32_t len = transMsgLenFromCont(pMsg->contLen);
   tDebug("server conn %p %s is sent to %s:%d, local info: %s:%d, msglen:%d", pConn, TMSG_INFO(pHead->msgType),
-         taosInetNtoa(pConn->addr.sin_addr), ntohs(pConn->addr.sin_port), taosInetNtoa(pConn->locaddr.sin_addr),
+         taosInetNtoa(pConn->addr.sin_addr), ntohs(pConn->addr.sin_port), taosInetNtoa(pConn->localAddr.sin_addr),
-         ntohs(pConn->locaddr.sin_port), len);
+         ntohs(pConn->localAddr.sin_port), len);
   pHead->msgLen = htonl(len);
 
   wb->base = msg;
@@ -686,8 +686,8 @@ void uvOnConnectionCb(uv_stream_t* q, ssize_t nread, const uv_buf_t* buf) {
       return;
     }
 
-    addrlen = sizeof(pConn->locaddr);
+    addrlen = sizeof(pConn->localAddr);
-    if (0 != uv_tcp_getsockname(pConn->pTcp, (struct sockaddr*)&pConn->locaddr, &addrlen)) {
+    if (0 != uv_tcp_getsockname(pConn->pTcp, (struct sockaddr*)&pConn->localAddr, &addrlen)) {
       tError("server conn %p failed to get local info", pConn);
       transUnrefSrvHandle(pConn);
       return;

source/libs/transport/test/transUT.cpp

@@ -381,28 +381,29 @@ TEST_F(TransEnv, srvReleaseHandle) {
   }
   //////////////////
 }
-TEST_F(TransEnv, cliReleaseHandleExcept) {
+// reopen later
-  SRpcMsg resp = {0};
+// TEST_F(TransEnv, cliReleaseHandleExcept) {
-  SRpcMsg req = {0};
+//  SRpcMsg resp = {0};
-  for (int i = 0; i < 3; i++) {
+//  SRpcMsg req = {0};
-    memset(&req, 0, sizeof(req));
+//  for (int i = 0; i < 3; i++) {
-    req.info = resp.info;
+//    memset(&req, 0, sizeof(req));
-    req.info.persistHandle = 1;
+//    req.info = resp.info;
-    req.info.ahandle = (void *)1234;
+//    req.info.persistHandle = 1;
-    req.msgType = 1;
+//    req.info.ahandle = (void *)1234;
-    req.pCont = rpcMallocCont(10);
+//    req.msgType = 1;
-    req.contLen = 10;
+//    req.pCont = rpcMallocCont(10);
-    tr->cliSendAndRecv(&req, &resp);
+//    req.contLen = 10;
-    if (i == 1) {
+//    tr->cliSendAndRecv(&req, &resp);
-      std::cout << "stop server" << std::endl;
+//    if (i == 1) {
-      tr->StopSrv();
+//      std::cout << "stop server" << std::endl;
-    }
+//      tr->StopSrv();
-    if (i > 1) {
+//    }
-      EXPECT_TRUE(resp.code != 0);
+//    if (i > 1) {
-    }
+//      EXPECT_TRUE(resp.code != 0);
-  }
+//    }
-  //////////////////
+//  }
-}
+//  //////////////////
+//}
 TEST_F(TransEnv, srvContinueSend) {
   tr->SetSrvContinueSend(processContinueSend);
   SRpcMsg req = {0}, resp = {0};

source/util/src/tarray.c

@@ -174,7 +174,11 @@ void taosArrayRemoveDuplicate(SArray* pArray, __compar_fn_t comparFn, void (*fp)
 }
 
 void* taosArrayAddAll(SArray* pArray, const SArray* pInput) {
-  return taosArrayAddBatch(pArray, pInput->pData, (int32_t)taosArrayGetSize(pInput));
+  if (pInput) {
+    return taosArrayAddBatch(pArray, pInput->pData, (int32_t)taosArrayGetSize(pInput));
+  } else {
+    return NULL;
+  }
 }
 
 void* taosArrayPop(SArray* pArray) {

source/util/src/terror.c

@@ -414,6 +414,8 @@ TAOS_DEFINE_ERROR(TSDB_CODE_SYN_INVALID_MSGLEN,           "Invalid msg length")
 TAOS_DEFINE_ERROR(TSDB_CODE_SYN_INVALID_MSGTYPE,          "Invalid msg type")
 
 TAOS_DEFINE_ERROR(TSDB_CODE_SYN_NOT_LEADER,               "Sync not leader")
+TAOS_DEFINE_ERROR(TSDB_CODE_SYN_ONE_REPLICA,              "Sync one replica")
+TAOS_DEFINE_ERROR(TSDB_CODE_SYN_NOT_IN_NEW_CONFIG,        "Sync not in new config")
 TAOS_DEFINE_ERROR(TSDB_CODE_SYN_INTERNAL_ERROR,           "Sync internal error")
 
 // wal
@@ -560,13 +562,16 @@ TAOS_DEFINE_ERROR(TSDB_CODE_TSMA_ALREADY_EXIST,             "Tsma already exists
 TAOS_DEFINE_ERROR(TSDB_CODE_TSMA_NO_INDEX_IN_META,          "No tsma index in meta")
 TAOS_DEFINE_ERROR(TSDB_CODE_TSMA_INVALID_ENV,               "Invalid tsma env")
 TAOS_DEFINE_ERROR(TSDB_CODE_TSMA_INVALID_STAT,              "Invalid tsma state")
+TAOS_DEFINE_ERROR(TSDB_CODE_TSMA_INVALID_PTR,               "Invalid tsma pointer")
+TAOS_DEFINE_ERROR(TSDB_CODE_TSMA_INVALID_PARA,              "Invalid tsma parameters")
 TAOS_DEFINE_ERROR(TSDB_CODE_TSMA_NO_INDEX_IN_CACHE,         "No tsma index in cache")
-TAOS_DEFINE_ERROR(TSDB_CODE_TSMA_RM_SKEY_IN_HASH,           "Rm tsma skey in cache")
+
 
 //rsma
 TAOS_DEFINE_ERROR(TSDB_CODE_RSMA_INVALID_ENV,               "Invalid rsma env")
 TAOS_DEFINE_ERROR(TSDB_CODE_RSMA_INVALID_STAT,              "Invalid rsma state")
 
+
 TAOS_DEFINE_ERROR(TSDB_CODE_INDEX_REBUILDING,               "Index is rebuilding")
 
 #ifdef TAOS_ERROR_C

tests/parallel_test/collect_cases.sh

@@ -41,5 +41,16 @@ fi
 cat ../script/jenkins/basic.txt |grep -v "^#"|grep -v "^$"|sed "s/^/,,script,/" >>$case_file
 grep "^python" ../system-test/fulltest.sh |sed "s/^/,,system-test,/" >>$case_file
 
+# tar source code for run.sh to use
+# if [ $ent -eq 0 ]; then
+#     cd ../../../
+#     rm -rf TDengine.tar.gz
+#     tar --exclude=TDengine/debug --exclude=TDengine/sim --exclude=TDengine/release -czf TDengine.tar.gz TDengine taos-connector-python
+# else
+#     cd ../../../../
+#     rm -rf TDinternal.tar.gz
+#     tar --exclude=TDinternal/debug --exclude=TDinternal/sim --exclude=TDinternal/community/debug --exclude=TDinternal/community/release --exclude=TDinternal/community/sim -czf TDinternal.tar.gz TDinternal taos-connector-python
+# fi
+
 exit 0
 

tests/parallel_test/run.sh

@@ -255,7 +255,6 @@ function run_thread() {
             $cmd # 2>/dev/null
             local case_info=`echo "$line"|cut -d, -f 3,4`
             local corefile=`ls $log_dir/${case_file}.coredump/`
-            corefile=`find $log_dir/${case_file}.coredump/ -name "core.*"`
             echo -e "$case_info \e[31m failed\e[0m"
             echo "=========================log============================"
             cat $log_dir/$case_file.log
@@ -291,6 +290,19 @@ function run_thread() {
             fi
             cmd="$scpcmd:${remote_sim_tar} $log_dir/${case_file}.sim.tar.gz"
             $cmd
+            # backup source code (disabled)
+            source_tar_dir=$log_dir/TDengine_${hosts[index]}
+            source_tar_file=TDengine.tar.gz
+            if [ $ent -ne 0 ]; then
+                source_tar_dir=$log_dir/TDinternal_${hosts[index]}
+                source_tar_file=TDinternal.tar.gz
+            fi
+            mkdir $source_tar_dir 2>/dev/null
+            if [ $? -eq 0 ]; then
+                cmd="$scpcmd:${workdirs[index]}/$source_tar_file $source_tar_dir"
+                # echo "$cmd"
+                # $cmd
+            fi
         fi
     done
 }

tests/pytest/alter/alter_replica.py

@@ -42,7 +42,7 @@ class TDTestCase:
         tdDnodes.start(3)
 
     def run(self):
-        tdSql.execute('create database db replica 3 days 7')
+        tdSql.execute('create database db replica 3 duration 7')
         tdSql.execute('use db')
         for tid in range(1, 11):
             tdSql.execute('create table tb%d(ts timestamp, i int)' % tid)

tests/pytest/functions/function_max.py

@@ -83,7 +83,7 @@ class TDTestCase:
         tdSql.checkData(0, 0, np.max(floatData))
 
         # test case: https://jira.taosdata.com:18080/browse/TD-2583
-        tdSql.execute("create database test days 2")
+        tdSql.execute("create database test duration 2")
         tdSql.execute("create table car(ts timestamp, speed int)")
         tdSql.execute("insert into car values(now, -1)")
         tdSql.execute("insert into car values(now-10d, null)")

tests/pytest/functions/function_min.py

@@ -83,7 +83,7 @@ class TDTestCase:
         tdSql.checkData(0, 0, np.min(floatData))
 
         # test case: https://jira.taosdata.com:18080/browse/TD-2583
-        tdSql.execute("create database test days 2")
+        tdSql.execute("create database test duration 2")
         tdSql.execute("create table car(ts timestamp, speed int)")
         tdSql.execute("insert into car values(now, 1)")
         tdSql.execute("insert into car values(now-10d, null)")

tests/pytest/functions/queryTestCases.py

@@ -264,7 +264,7 @@ class TDTestCase:
 
     def td4288(self):
         tdLog.printNoPrefix("==========TD-4288==========")
-        # keep ~ [days,365000]
+        # keep ~ [duration,365000]
         tdSql.execute("drop database if exists db")
         tdSql.execute("create database  if not exists db")
         tdSql.query("show variables")

tests/pytest/insert/before_1970.py

@@ -32,7 +32,7 @@ class TDTestCase:
 
         print("==============step1")
         tdSql.execute("create database if not exists demo keep 36500;");
-        print("==============create db demo keep 365000 days")
+        print("==============create db demo keep 365000 duration")
         tdSql.execute("use demo;")
         tdSql.execute("CREATE table if not exists test (ts timestamp, f1 int);")
         print("==============create table test")

tests/pytest/insert/retentionpolicy.py

@@ -51,7 +51,7 @@ class TDTestRetetion:
     def run(self):
         
         tdLog.info("=============== step1")
-        tdSql.execute('create database test keep 3 days 1;')
+        tdSql.execute('create database test keep 3 duration 1;')
         tdSql.execute('use test;')
         tdSql.execute('create table test(ts timestamp,i int);')
 

tests/pytest/multilevel/fileDistributionSameLevel.py

@@ -66,7 +66,7 @@ class TDTestCase:
         tdDnodes.deploy(1,cfg)
         tdDnodes.startWithoutSleep(1)
 
-        tdSql.execute("create database test days 1")
+        tdSql.execute("create database test duration 1")
         tdSql.execute("use test")
 
         tdSql.execute("create table stb(ts timestamp, c int) tags(t int)")
@@ -85,7 +85,7 @@ class TDTestCase:
         tdLog.info("================= step3")
         tdSql.execute('drop database test')
         for i in range(50):
-            tdSql.execute("create database test%d days 1" %(i))
+            tdSql.execute("create database test%d duration 1" %(i))
             tdSql.execute("use test%d" %(i))
             tdSql.execute("create table tb (ts timestamp,i int)") 
             for j in range(10):           

tests/pytest/multilevel/retentionTest.py

@@ -56,7 +56,7 @@ class TDTestCase:
         tdDnodes.deploy(1,cfg)
         tdDnodes.startWithoutSleep(1)
         
-        tdSql.execute("create database test days 1 keep 15,5,10")
+        tdSql.execute("create database test duration 1 keep 15,5,10")
         tdSql.execute("use test")
 
         tdSql.execute("create table tb(ts timestamp, c int)")