merge 3.0

examples/c/tmq.c

@@ -26,6 +26,14 @@ static void msg_process(TAOS_RES* msg) {
   printf("topic: %s\n", tmq_get_topic_name(msg));
   printf("db: %s\n", tmq_get_db_name(msg));
   printf("vg: %d\n", tmq_get_vgroup_id(msg));
+  if (tmq_get_res_type(msg) == TMQ_RES_TABLE_META) {
+    void*   meta;
+    int32_t metaLen;
+    tmq_get_raw_meta(msg, &meta, &metaLen);
+
+    printf("meta, len is %d\n", metaLen);
+    return;
+  }
   while (1) {
     TAOS_ROW row = taos_fetch_row(msg);
     if (row == NULL) break;
@@ -76,19 +84,41 @@ int32_t init_env() {
   }
   taos_free_result(pRes);
 
-  pRes = taos_query(pConn, "create table if not exists ct1 using st1 tags(2000)");
+  pRes = taos_query(pConn, "insert into ct0 values(now, 1, 2, 'a')");
   if (taos_errno(pRes) != 0) {
-    printf("failed to create child table tu2, reason:%s\n", taos_errstr(pRes));
+    printf("failed to insert into ct0, reason:%s\n", taos_errstr(pRes));
     return -1;
   }
+  taos_free_result(pRes);
+
+  pRes = taos_query(pConn, "create table if not exists ct1 using st1 tags(2000)");
+  if (taos_errno(pRes) != 0) {
+    printf("failed to create child table ct1, reason:%s\n", taos_errstr(pRes));
+    return -1;
+  }
+  taos_free_result(pRes);
+
+  pRes = taos_query(pConn, "insert into ct1 values(now, 3, 4, 'b')");
+  if (taos_errno(pRes) != 0) {
+    printf("failed to insert into ct1, reason:%s\n", taos_errstr(pRes));
+    return -1;
+  }
+  taos_free_result(pRes);
 
   pRes = taos_query(pConn, "create table if not exists ct3 using st1 tags(3000)");
   if (taos_errno(pRes) != 0) {
-    printf("failed to create child table tu3, reason:%s\n", taos_errstr(pRes));
+    printf("failed to create child table ct3, reason:%s\n", taos_errstr(pRes));
     return -1;
   }
-
   taos_free_result(pRes);
+
+  pRes = taos_query(pConn, "insert into ct3 values(now, 5, 6, 'c')");
+  if (taos_errno(pRes) != 0) {
+    printf("failed to insert into ct3, reason:%s\n", taos_errstr(pRes));
+    return -1;
+  }
+  taos_free_result(pRes);
+
   return 0;
 }
 
@@ -107,8 +137,8 @@ int32_t create_topic() {
   }
   taos_free_result(pRes);
 
-  /*pRes = taos_query(pConn, "create topic topic_ctb_column as database abc1");*/
+  pRes = taos_query(pConn, "create topic topic_ctb_column with meta as database abc1");
-  pRes = taos_query(pConn, "create topic topic_ctb_column as select ts, c1, c2, c3 from st1");
+  /*pRes = taos_query(pConn, "create topic topic_ctb_column as select ts, c1, c2, c3 from st1");*/
   if (taos_errno(pRes) != 0) {
     printf("failed to create topic topic_ctb_column, reason:%s\n", taos_errstr(pRes));
     return -1;
@@ -168,6 +198,9 @@ tmq_t* build_consumer() {
   tmq_conf_set(conf, "td.connect.pass", "taosdata");
   tmq_conf_set(conf, "msg.with.table.name", "true");
   tmq_conf_set(conf, "enable.auto.commit", "true");
+
+  tmq_conf_set(conf, "experiment.use.snapshot", "false");
+
   tmq_conf_set_auto_commit_cb(conf, tmq_commit_cb_print, NULL);
   tmq_t* tmq = tmq_consumer_new(conf, NULL, 0);
   assert(tmq);

include/client/taos.h

@@ -261,7 +261,7 @@ enum tmq_res_t {
 typedef enum tmq_res_t tmq_res_t;
 
 DLL_EXPORT tmq_res_t   tmq_get_res_type(TAOS_RES *res);
-DLL_EXPORT int32_t     tmq_get_raw_meta(TAOS_RES *res, const void **raw_meta, int32_t *raw_meta_len);
+DLL_EXPORT int32_t     tmq_get_raw_meta(TAOS_RES *res, void **raw_meta, int32_t *raw_meta_len);
 DLL_EXPORT const char *tmq_get_topic_name(TAOS_RES *res);
 DLL_EXPORT const char *tmq_get_db_name(TAOS_RES *res);
 DLL_EXPORT int32_t     tmq_get_vgroup_id(TAOS_RES *res);

include/common/tcommon.h

@@ -46,11 +46,15 @@ typedef enum EStreamType {
   STREAM_INVALID,
   STREAM_GET_ALL,
   STREAM_DELETE,
+  STREAM_RETRIEVE,
 } EStreamType;
 
 typedef struct {
   SArray*   pTableList;
   SHashObj* map;  // speedup acquire the tableQueryInfo by table uid
+  void*     pTagCond;
+  void*     pTagIndexCond;
+  uint64_t  suid;
 } STableListInfo;
 
 typedef struct SColumnDataAgg {
@@ -112,8 +116,8 @@ typedef struct SQueryTableDataCond {
   int32_t      type;              // data block load type:
   int32_t      numOfTWindows;
   STimeWindow* twindows;
-  int32_t      startVersion;
+  int64_t      startVersion;
-  int32_t      endVersion;
+  int64_t      endVersion;
 } SQueryTableDataCond;
 
 void*   blockDataDestroy(SSDataBlock* pBlock);

include/common/tdatablock.h

@@ -71,7 +71,8 @@ SEpSet getEpSet_s(SCorEpSet* pEpSet);
 #define colDataGetData(p1_, r_) \
   ((IS_VAR_DATA_TYPE((p1_)->info.type)) ? colDataGetVarData(p1_, r_) : colDataGetNumData(p1_, r_))
 
-#define IS_JSON_NULL(type, data) ((type) == TSDB_DATA_TYPE_JSON && *(data) == TSDB_DATA_TYPE_NULL)
+#define IS_JSON_NULL(type, data) ((type) == TSDB_DATA_TYPE_JSON && \
+                                  (*(data) == TSDB_DATA_TYPE_NULL || tTagIsJsonNull(data)))
 
 static FORCE_INLINE bool colDataIsNull_s(const SColumnInfoData* pColumnInfoData, uint32_t row) {
   if (!pColumnInfoData->hasNull) {

include/common/tdataformat.h

@@ -70,6 +70,8 @@ int32_t tGetTSRow(uint8_t *p, STSRow2 *pRow);
 // STag
 int32_t tTagNew(SArray *pArray, int32_t version, int8_t isJson, STag **ppTag);
 void    tTagFree(STag *pTag);
+bool    tTagIsJson(const void *pTag);
+bool    tTagIsJsonNull(void *tagVal);
 bool    tTagGet(const STag *pTag, STagVal *pTagVal);
 char   *tTagValToData(const STagVal *pTagVal, bool isJson);
 int32_t tEncodeTag(SEncoder *pEncoder, const STag *pTag);

include/common/tglobal.h

@@ -137,6 +137,8 @@ extern bool tsSmlDataFormat;
 // internal
 extern int32_t tsTransPullupInterval;
 extern int32_t tsMqRebalanceInterval;
+extern int32_t tsTtlUnit;
+extern int32_t tsTtlPushInterval;
 
 #define NEEDTO_COMPRESSS_MSG(size) (tsCompressMsgSize != -1 && (size) > tsCompressMsgSize)
 

include/common/tmsg.h

@@ -834,6 +834,14 @@ typedef struct {
 int32_t tSerializeSQnodeListReq(void* buf, int32_t bufLen, SQnodeListReq* pReq);
 int32_t tDeserializeSQnodeListReq(void* buf, int32_t bufLen, SQnodeListReq* pReq);
 
+typedef struct {
+  int32_t rowNum;
+} SDnodeListReq;
+
+int32_t tSerializeSDnodeListReq(void* buf, int32_t bufLen, SDnodeListReq* pReq);
+int32_t tDeserializeSDnodeListReq(void* buf, int32_t bufLen, SDnodeListReq* pReq);
+
+
 typedef struct SQueryNodeAddr {
   int32_t nodeId;  // vgId or qnodeId
   SEpSet  epSet;
@@ -852,6 +860,15 @@ int32_t tSerializeSQnodeListRsp(void* buf, int32_t bufLen, SQnodeListRsp* pRsp);
 int32_t tDeserializeSQnodeListRsp(void* buf, int32_t bufLen, SQnodeListRsp* pRsp);
 void    tFreeSQnodeListRsp(SQnodeListRsp* pRsp);
 
+typedef struct {
+  SArray* dnodeList;  // SArray<SEpSet>
+} SDnodeListRsp;
+
+int32_t tSerializeSDnodeListRsp(void* buf, int32_t bufLen, SDnodeListRsp* pRsp);
+int32_t tDeserializeSDnodeListRsp(void* buf, int32_t bufLen, SDnodeListRsp* pRsp);
+void    tFreeSDnodeListRsp(SDnodeListRsp* pRsp);
+
+
 typedef struct {
   SArray* pArray;  // Array of SUseDbRsp
 } SUseDbBatchRsp;
@@ -2654,6 +2671,7 @@ typedef struct {
   SMsgHead head;
   char     subKey[TSDB_SUBSCRIBE_KEY_LEN];
   int8_t   withTbName;
+  int8_t   useSnapshot;
   int32_t  epoch;
   uint64_t reqId;
   int64_t  consumerId;

include/common/tmsgdef.h

@@ -81,6 +81,7 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_DND_SERVER_STATUS, "server-status", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_DND_NET_TEST, "net-test", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_DND_CONFIG_DNODE, "config-dnode", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_DND_SYSTABLE_RETRIEVE, "dnode-retrieve", NULL, NULL)
 
   TD_NEW_MSG_SEG(TDMT_MND_MSG)
   TD_DEF_MSG_TYPE(TDMT_MND_CONNECT, "connect", NULL, NULL)
@@ -101,6 +102,7 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_MND_ALTER_QNODE, "alter-qnode", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_DROP_QNODE, "drop-qnode", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_QNODE_LIST, "qnode-list", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_MND_DNODE_LIST, "dnode-list", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_CREATE_SNODE, "create-snode", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_ALTER_SNODE, "alter-snode", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_DROP_SNODE, "drop-snode", NULL, NULL)
@@ -145,13 +147,14 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_MND_MQ_TIMER, "mq-tmr", SMTimerReq, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_TELEM_TIMER, "telem-tmr", SMTimerReq, SMTimerReq)
   TD_DEF_MSG_TYPE(TDMT_MND_TRANS_TIMER, "trans-tmr", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_MND_TTL_TIMER, "ttl-tmr", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_KILL_TRANS, "kill-trans", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_KILL_QUERY, "kill-query", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_KILL_CONN, "kill-conn", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_HEARTBEAT, "heartbeat", SClientHbBatchReq, SClientHbBatchRsp)
   TD_DEF_MSG_TYPE(TDMT_MND_STATUS, "status", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_SHOW, "show", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_MND_SYSTABLE_RETRIEVE, "retrieve", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_MND_SYSTABLE_RETRIEVE, "mnd-retrieve", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_GRANT, "grant", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_AUTH, "auth", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_APPLY_MSG, "mnode-apply", NULL, NULL)
@@ -201,15 +204,14 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_VND_ALTER_HASHRANGE, "alter-hashrange", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_COMPACT, "compact", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_DROP_TTL_TABLE, "drop-ttl-stb", NULL, NULL)
-  TD_NEW_MSG_SEG(TDMT_QND_MSG)
 
-  //shared by snode and vnode
   TD_NEW_MSG_SEG(TDMT_STREAM_MSG)
   TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_DEPLOY, "stream-task-deploy", SStreamTaskDeployReq, SStreamTaskDeployRsp)
   TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_DROP, "stream-task-drop", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_RUN, "stream-task-run", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_DISPATCH, "stream-task-dispatch", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_RECOVER, "stream-task-recover", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_STREAM_RETRIEVE, "stream-retrieve", NULL, NULL)
 
   TD_NEW_MSG_SEG(TDMT_SCH_MSG)
   TD_DEF_MSG_TYPE(TDMT_SCH_LINK_BROKEN, "link-broken", NULL, NULL)

include/libs/executor/dataSinkMgt.h

@@ -33,7 +33,7 @@ struct SDataSink;
 struct SSDataBlock;
 
 typedef struct SDeleterRes {
-  uint64_t uid;
+  uint64_t suid;
   SArray*  uidList;
   int64_t  skey;
   int64_t  ekey;
@@ -41,6 +41,7 @@ typedef struct SDeleterRes {
 } SDeleterRes;
 
 typedef struct SDeleterParam {
+  uint64_t suid;
   SArray*  pUidList;
 } SDeleterParam;
 

include/libs/executor/executor.h

@@ -36,11 +36,13 @@ typedef struct SReadHandle {
   void*   vnode;
   void*   mnd;
   SMsgCb* pMsgCb;
+  int8_t  initTsdbReader;
 } SReadHandle;
 
 enum {
   STREAM_DATA_TYPE_SUBMIT_BLOCK = 1,
   STREAM_DATA_TYPE_SSDATA_BLOCK = 2,
+  STREAM_DATA_TYPE_FROM_SNAPSHOT = 3,
 };
 
 typedef enum {
@@ -56,6 +58,13 @@ typedef enum {
  */
 qTaskInfo_t qCreateStreamExecTaskInfo(void* msg, void* streamReadHandle);
 
+/**
+ * Switch the stream scan to snapshot mode
+ * @param tinfo
+ * @return
+ */
+int32_t qStreamScanSnapshot(qTaskInfo_t tinfo);
+
 /**
  * Set the input data block for the stream scan.
  * @param tinfo

include/libs/nodes/cmdnodes.h

@@ -36,6 +36,10 @@ extern "C" {
 #define SHOW_CREATE_TB_RESULT_FIELD1_LEN (TSDB_TABLE_NAME_LEN + VARSTR_HEADER_SIZE)
 #define SHOW_CREATE_TB_RESULT_FIELD2_LEN (TSDB_MAX_BINARY_LEN + VARSTR_HEADER_SIZE)
 
+#define SHOW_LOCAL_VARIABLES_RESULT_COLS       2
+#define SHOW_LOCAL_VARIABLES_RESULT_FIELD1_LEN (TSDB_CONFIG_OPTION_LEN + VARSTR_HEADER_SIZE)
+#define SHOW_LOCAL_VARIABLES_RESULT_FIELD2_LEN (TSDB_CONFIG_VALUE_LEN + VARSTR_HEADER_SIZE)
+
 #define PRIVILEGE_TYPE_MASK(n) (1 << n)
 
 #define PRIVILEGE_TYPE_ALL   PRIVILEGE_TYPE_MASK(0)

include/libs/nodes/nodes.h

@@ -227,6 +227,7 @@ typedef enum ENodeType {
   QUERY_NODE_PHYSICAL_PLAN_EXCHANGE,
   QUERY_NODE_PHYSICAL_PLAN_MERGE,
   QUERY_NODE_PHYSICAL_PLAN_SORT,
+  QUERY_NODE_PHYSICAL_PLAN_GROUP_SORT,
   QUERY_NODE_PHYSICAL_PLAN_HASH_INTERVAL,
   QUERY_NODE_PHYSICAL_PLAN_MERGE_INTERVAL,
   QUERY_NODE_PHYSICAL_PLAN_MERGE_ALIGNED_INTERVAL,

include/libs/nodes/plannodes.h

@@ -68,6 +68,7 @@ typedef struct SScanLogicNode {
   int8_t        intervalUnit;
   int8_t        slidingUnit;
   SNode*        pTagCond;
+  SNode*        pTagIndexCond;
   int8_t        triggerType;
   int64_t       watermark;
   int16_t       tsColId;
@@ -416,6 +417,8 @@ typedef struct SSortPhysiNode {
   SNodeList* pTargets;
 } SSortPhysiNode;
 
+typedef SSortPhysiNode SGroupSortPhysiNode;
+
 typedef struct SPartitionPhysiNode {
   SPhysiNode node;
   SNodeList* pExprs;  // these are expression list of partition_by_clause
@@ -462,6 +465,7 @@ typedef struct SSubplan {
   SPhysiNode*    pNode;         // physical plan of current subplan
   SDataSinkNode* pDataSink;     // data of the subplan flow into the datasink
   SNode*         pTagCond;
+  SNode*         pTagIndexCond;
 } SSubplan;
 
 typedef enum EExplainMode { EXPLAIN_MODE_DISABLE = 1, EXPLAIN_MODE_STATIC, EXPLAIN_MODE_ANALYZE } EExplainMode;

include/libs/nodes/querynodes.h

@@ -292,7 +292,7 @@ typedef struct SDeleteStmt {
   SNode*      pFromTable;  // FROM clause
   SNode*      pWhere;      // WHERE clause
   SNode*      pCountFunc;  // count the number of rows affected
-  SNode*      pTagIndexCond;  // pWhere divided into pTagIndexCond and timeRange
+  SNode*      pTagCond;    // pWhere divided into pTagCond and timeRange
   STimeWindow timeRange;
   uint8_t     precision;
   bool        deleteZeroRows;
@@ -397,7 +397,8 @@ void    nodesValueNodeToVariant(const SValueNode* pNode, SVariant* pVal);
 
 char*   nodesGetFillModeString(EFillMode mode);
 int32_t nodesMergeConds(SNode** pDst, SNodeList** pSrc);
-int32_t nodesPartitionCond(SNode** pCondition, SNode** pPrimaryKeyCond, SNode** pTagCond, SNode** pOtherCond);
+int32_t nodesPartitionCond(SNode** pCondition, SNode** pPrimaryKeyCond, SNode** pTagIndexCond, SNode** pTagCond,
+                           SNode** pOtherCond);
 
 #ifdef __cplusplus
 }

include/libs/qworker/qworker.h

@@ -32,7 +32,7 @@ enum {
 };
 
 typedef struct SDeleteRes {
-  uint64_t uid;
+  uint64_t suid;
   SArray*  uidList;
   int64_t  skey;
   int64_t  ekey;

include/libs/stream/tstream.h

@@ -77,7 +77,7 @@ typedef struct {
 typedef struct {
   int8_t type;
 
-  int32_t sourceVg;
+  int32_t srcVgId;
   int64_t sourceVer;
 
   SArray* blocks;  // SArray<SSDataBlock*>
@@ -145,11 +145,6 @@ void streamDataSubmitRefDec(SStreamDataSubmit* pDataSubmit);
 
 SStreamDataSubmit* streamSubmitRefClone(SStreamDataSubmit* pSubmit);
 
-#if 0
-int32_t streamDataBlockEncode(void** buf, const SStreamDataBlock* pOutput);
-void*   streamDataBlockDecode(const void* buf, SStreamDataBlock* pInput);
-#endif
-
 typedef struct {
   char* qmsg;
   // followings are not applicable to encoder and decoder
@@ -234,26 +229,33 @@ enum {
   TASK_TRIGGER_STATUS__ACTIVE,
 };
 
+typedef struct {
+  int32_t nodeId;
+  int32_t childId;
+  int32_t taskId;
+  SEpSet  epSet;
+} SStreamChildEpInfo;
+
 struct SStreamTask {
   int64_t streamId;
   int32_t taskId;
-  int8_t  inputType;
+  int8_t  isDataScan;
-  int8_t  taskStatus;
-
-  int8_t execStatus;
-
   int8_t  execType;
   int8_t  sinkType;
   int8_t  dispatchType;
   int16_t dispatchMsgType;
 
-  int8_t dataScan;
+  int8_t taskStatus;
+  int8_t execStatus;
 
   // node info
-  int32_t childId;
+  int32_t selfChildId;
   int32_t nodeId;
   SEpSet  epSet;
 
+  // children info
+  SArray* childEpInfo;  // SArray<SStreamChildEpInfo*>
+
   // exec
   STaskExec exec;
 
@@ -291,6 +293,9 @@ struct SStreamTask {
   SMsgCb* pMsgCb;
 };
 
+int32_t tEncodeStreamEpInfo(SEncoder* pEncoder, const SStreamChildEpInfo* pInfo);
+int32_t tDecodeStreamEpInfo(SDecoder* pDecoder, SStreamChildEpInfo* pInfo);
+
 SStreamTask* tNewSStreamTask(int64_t streamId);
 int32_t      tEncodeSStreamTask(SEncoder* pEncoder, const SStreamTask* pTask);
 int32_t      tDecodeSStreamTask(SDecoder* pDecoder, SStreamTask* pTask);
@@ -369,9 +374,9 @@ typedef struct {
 typedef struct {
   int64_t streamId;
   int32_t taskId;
-  int32_t sourceTaskId;
+  int32_t dataSrcVgId;
-  int32_t sourceVg;
+  int32_t upstreamTaskId;
-  int32_t sourceChildId;
+  int32_t upstreamChildId;
   int32_t upstreamNodeId;
 #if 0
   int64_t sourceVer;
@@ -387,6 +392,23 @@ typedef struct {
   int8_t  inputStatus;
 } SStreamDispatchRsp;
 
+typedef struct {
+  int64_t            streamId;
+  int32_t            srcTaskId;
+  int32_t            srcNodeId;
+  int32_t            dstTaskId;
+  int32_t            dstNodeId;
+  int32_t            retrieveLen;
+  SRetrieveTableRsp* pRetrieve;
+} SStreamRetrieveReq;
+
+typedef struct {
+  int64_t streamId;
+  int32_t childId;
+  int32_t rspFromTaskId;
+  int32_t rspToTaskId;
+} SStreamRetrieveRsp;
+
 typedef struct {
   int64_t streamId;
   int32_t taskId;
@@ -401,6 +423,7 @@ typedef struct {
 } SStreamTaskRecoverRsp;
 
 int32_t tDecodeStreamDispatchReq(SDecoder* pDecoder, SStreamDispatchReq* pReq);
+int32_t tDecodeStreamRetrieveReq(SDecoder* pDecoder, SStreamRetrieveReq* pReq);
 
 int32_t streamLaunchByWrite(SStreamTask* pTask, int32_t vgId);
 int32_t streamSetupTrigger(SStreamTask* pTask);
@@ -411,6 +434,9 @@ int32_t streamProcessDispatchRsp(SStreamTask* pTask, SStreamDispatchRsp* pRsp);
 int32_t streamProcessRecoverReq(SStreamTask* pTask, SStreamTaskRecoverReq* pReq, SRpcMsg* pMsg);
 int32_t streamProcessRecoverRsp(SStreamTask* pTask, SStreamTaskRecoverRsp* pRsp);
 
+int32_t streamProcessRetrieveReq(SStreamTask* pTask, SStreamRetrieveReq* pReq, SRpcMsg* pMsg);
+int32_t streamProcessRetrieveRsp(SStreamTask* pTask, SStreamRetrieveRsp* pRsp);
+
 #ifdef __cplusplus
 }
 #endif

include/libs/sync/sync.h

@@ -28,6 +28,7 @@ extern bool gRaftDetailLog;
 
 #define SYNC_INDEX_BEGIN   0
 #define SYNC_INDEX_INVALID -1
+#define SYNC_TERM_INVALID  0xFFFFFFFFFFFFFFFF
 
 typedef uint64_t SyncNodeId;
 typedef int32_t  SyncGroupId;
@@ -199,7 +200,7 @@ const char* syncGetMyRoleStr(int64_t rid);
 SyncTerm    syncGetMyTerm(int64_t rid);
 SyncGroupId syncGetVgId(int64_t rid);
 void        syncGetEpSet(int64_t rid, SEpSet* pEpSet);
-int32_t     syncPropose(int64_t rid, const SRpcMsg* pMsg, bool isWeak);
+int32_t     syncPropose(int64_t rid, SRpcMsg* pMsg, bool isWeak);
 bool        syncEnvIsStart();
 const char* syncStr(ESyncState state);
 bool        syncIsRestoreFinish(int64_t rid);

include/libs/sync/syncTools.h

@@ -43,7 +43,7 @@ void setElectTimerMS(int64_t rid, int32_t electTimerMS);
 void setHeartbeatTimerMS(int64_t rid, int32_t hbTimerMS);
 
 // for compatibility, the same as syncPropose
-int32_t syncForwardToPeer(int64_t rid, const SRpcMsg* pMsg, bool isWeak);
+int32_t syncForwardToPeer(int64_t rid, SRpcMsg* pMsg, bool isWeak);
 
 // utils
 const char* syncUtilState2String(ESyncState state);
@@ -489,7 +489,6 @@ void syncLeaderTransferPrint2(char* s, const SyncLeaderTransfer* pMsg);
 void syncLeaderTransferLog(const SyncLeaderTransfer* pMsg);
 void syncLeaderTransferLog2(char* s, const SyncLeaderTransfer* pMsg);
 
-
 // ---------------------------------------------
 typedef struct SyncReconfigFinish {
   uint32_t  bytes;
@@ -521,8 +520,6 @@ void syncReconfigFinishPrint2(char* s, const SyncReconfigFinish* pMsg);
 void syncReconfigFinishLog(const SyncReconfigFinish* pMsg);
 void syncReconfigFinishLog2(char* s, const SyncReconfigFinish* pMsg);
 
-
-
 // on message ----------------------
 int32_t syncNodeOnPingCb(SSyncNode* ths, SyncPing* pMsg);
 int32_t syncNodeOnPingReplyCb(SSyncNode* ths, SyncPingReply* pMsg);

include/libs/wal/wal.h

@@ -195,7 +195,6 @@ void            walCloseReadHandle(SWalReadHandle *);
 int32_t         walReadWithHandle(SWalReadHandle *pRead, int64_t ver);
 
 // only for tq usage
-// int32_t walReadWithHandle_s(SWalReadHandle *pRead, int64_t ver, SWalReadHead **ppHead);
 void    walSetReaderCapacity(SWalReadHandle *pRead, int32_t capacity);
 int32_t walFetchHead(SWalReadHandle *pRead, int64_t ver, SWalHead *pHead);
 int32_t walFetchBody(SWalReadHandle *pRead, SWalHead **ppHead);
@@ -211,13 +210,8 @@ void     walCloseRef(SWalRef *);
 int32_t  walRefVer(SWalRef *, int64_t ver);
 int32_t  walUnrefVer(SWal *);
 
-// deprecated
-#if 0
-int32_t walRead(SWal *, SWalHead **, int64_t ver);
-int32_t walReadWithFp(SWal *, FWalWrite writeFp, int64_t verStart, int32_t readNum);
-#endif
-
 // lifecycle check
+bool    walIsEmpty(SWal *);
 int64_t walGetFirstVer(SWal *);
 int64_t walGetSnapshotVer(SWal *);
 int64_t walGetLastVer(SWal *);

include/os/os.h

@@ -104,8 +104,6 @@ extern "C" {
 #include "osTimezone.h"
 #include "osEnv.h"
 
-void osDefaultInit();
-
 #ifdef __cplusplus
 }
 #endif

include/util/taoserror.h

@@ -71,6 +71,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_OUT_OF_RPC_MEMORY_QUEUE       TAOS_DEF_ERROR_CODE(0, 0x0029)
 #define TSDB_CODE_INVALID_TIMESTAMP             TAOS_DEF_ERROR_CODE(0, 0x0030)
 #define TSDB_CODE_MSG_DECODE_ERROR              TAOS_DEF_ERROR_CODE(0, 0x0031)
+#define TSDB_CODE_NO_AVAIL_DISK                 TAOS_DEF_ERROR_CODE(0, 0x0032)
 
 #define TSDB_CODE_REF_NO_MEMORY                 TAOS_DEF_ERROR_CODE(0, 0x0040)
 #define TSDB_CODE_REF_FULL                      TAOS_DEF_ERROR_CODE(0, 0x0041)

include/util/tconfig.h

@@ -104,6 +104,8 @@ int32_t cfgAddTimezone(SConfig *pCfg, const char *name, const char *defaultVal);
 const char *cfgStypeStr(ECfgSrcType type);
 const char *cfgDtypeStr(ECfgDataType type);
 
+void cfgDumpItemValue(SConfigItem *pItem, char* buf, int32_t bufSize, int32_t* pLen);
+
 void cfgDumpCfg(SConfig *pCfg, bool tsc, bool dump);
 
 int32_t cfgGetApollUrl(const char **envCmd, const char *envFile, char* apolloUrl);

include/util/tdef.h

@@ -443,8 +443,8 @@ enum {
 #define VNODE_HANDLE   -3
 #define BNODE_HANDLE   -4
 
-#define TSDB_CONFIG_OPTION_LEN 16
+#define TSDB_CONFIG_OPTION_LEN 32
-#define TSDB_CONIIG_VALUE_LEN  48
+#define TSDB_CONFIG_VALUE_LEN  64
 #define TSDB_CONFIG_NUMBER     8
 
 #define QUERY_ID_SIZE      20

source/client/src/clientImpl.c

@@ -1507,7 +1507,7 @@ static int32_t estimateJsonLen(SReqResultInfo* pResultInfo, int32_t numOfCols, i
         char*   jsonInnerData = data + CHAR_BYTES;
         if (jsonInnerType == TSDB_DATA_TYPE_NULL) {
           len += (VARSTR_HEADER_SIZE + strlen(TSDB_DATA_NULL_STR_L));
-        } else if (jsonInnerType & TD_TAG_JSON) {
+        } else if (tTagIsJson(data)) {
           len += (VARSTR_HEADER_SIZE + ((const STag*)(data))->len);
         } else if (jsonInnerType == TSDB_DATA_TYPE_NCHAR) {  // value -> "value"
           len += varDataTLen(jsonInnerData) + CHAR_BYTES * 2;
@@ -1592,7 +1592,7 @@ static int32_t doConvertJson(SReqResultInfo* pResultInfo, int32_t numOfCols, int
         if (jsonInnerType == TSDB_DATA_TYPE_NULL) {
           sprintf(varDataVal(dst), "%s", TSDB_DATA_NULL_STR_L);
           varDataSetLen(dst, strlen(varDataVal(dst)));
-        } else if (jsonInnerType & TD_TAG_JSON) {
+        } else if (tTagIsJson(data)) {
           char* jsonString = parseTagDatatoJson(data);
           STR_TO_VARSTR(dst, jsonString);
           taosMemoryFree(jsonString);

source/client/src/clientMain.c

@@ -290,6 +290,8 @@ TAOS_ROW taos_fetch_row(TAOS_RES *res) {
       pResultInfo->current += 1;
       return pResultInfo->row;
     }
+  } else if (TD_RES_TMQ_META(res)) {
+    return NULL;
   } else {
     // assert to avoid un-initialization error
     ASSERT(0);

source/client/src/tmq.c

@@ -54,6 +54,7 @@ struct tmq_conf_t {
   int8_t   autoCommit;
   int8_t   resetOffset;
   int8_t   withTbName;
+  int8_t   useSnapshot;
   uint16_t port;
   int32_t  autoCommitInterval;
   char*    ip;
@@ -69,6 +70,7 @@ struct tmq_t {
   char           groupId[TSDB_CGROUP_LEN];
   char           clientId[256];
   int8_t         withTbName;
+  int8_t         useSnapshot;
   int8_t         autoCommit;
   int32_t        autoCommitInterval;
   int32_t        resetOffsetCfg;
@@ -282,6 +284,18 @@ tmq_conf_res_t tmq_conf_set(tmq_conf_t* conf, const char* key, const char* value
     }
   }
 
+  if (strcmp(key, "experiment.use.snapshot") == 0) {
+    if (strcmp(value, "true") == 0) {
+      conf->useSnapshot = true;
+      return TMQ_CONF_OK;
+    } else if (strcmp(value, "false") == 0) {
+      conf->useSnapshot = false;
+      return TMQ_CONF_OK;
+    } else {
+      return TMQ_CONF_INVALID;
+    }
+  }
+
   if (strcmp(key, "td.connect.ip") == 0) {
     conf->ip = strdup(value);
     return TMQ_CONF_OK;
@@ -953,6 +967,7 @@ tmq_t* tmq_consumer_new(tmq_conf_t* conf, char* errstr, int32_t errstrLen) {
   strcpy(pTmq->clientId, conf->clientId);
   strcpy(pTmq->groupId, conf->groupId);
   pTmq->withTbName = conf->withTbName;
+  pTmq->useSnapshot = conf->useSnapshot;
   pTmq->autoCommit = conf->autoCommit;
   pTmq->autoCommitInterval = conf->autoCommitInterval;
   pTmq->commitCb = conf->commitCb;
@@ -1145,8 +1160,6 @@ int32_t tmqPollCb(void* param, const SDataBuf* pMsg, int32_t code) {
 
   // handle meta rsp
   int8_t rspType = ((SMqRspHead*)pMsg->pData)->mqMsgType;
-  if (rspType == TMQ_MSG_TYPE__POLL_META_RSP) {
-  }
 
   SMqPollRspWrapper* pRspWrapper = taosAllocateQitem(sizeof(SMqPollRspWrapper), DEF_QITEM);
   if (pRspWrapper == NULL) {
@@ -1159,19 +1172,19 @@ int32_t tmqPollCb(void* param, const SDataBuf* pMsg, int32_t code) {
   pRspWrapper->vgHandle = pVg;
   pRspWrapper->topicHandle = pTopic;
 
-  memcpy(&pRspWrapper->dataRsp, pMsg->pData, sizeof(SMqRspHead));
-
   if (rspType == TMQ_MSG_TYPE__POLL_RSP) {
+    memcpy(&pRspWrapper->dataRsp, pMsg->pData, sizeof(SMqRspHead));
     tDecodeSMqDataBlkRsp(POINTER_SHIFT(pMsg->pData, sizeof(SMqRspHead)), &pRspWrapper->dataRsp);
   } else {
     ASSERT(rspType == TMQ_MSG_TYPE__POLL_META_RSP);
+    memcpy(&pRspWrapper->metaRsp, pMsg->pData, sizeof(SMqRspHead));
     tDecodeSMqMetaRsp(POINTER_SHIFT(pMsg->pData, sizeof(SMqRspHead)), &pRspWrapper->metaRsp);
   }
 
   taosMemoryFree(pMsg->pData);
 
-  tscDebug("consumer %ld recv poll: vg %d, req offset %ld, rsp offset %ld", tmq->consumerId, pVg->vgId,
+  tscDebug("consumer %ld recv poll: vg %d, req offset %ld, rsp offset %ld, type %d", tmq->consumerId, pVg->vgId,
-           pRspWrapper->dataRsp.reqOffset, pRspWrapper->dataRsp.rspOffset);
+           pRspWrapper->dataRsp.reqOffset, pRspWrapper->dataRsp.rspOffset, rspType);
 
   taosWriteQitem(tmq->mqueue, pRspWrapper);
   tsem_post(&tmq->rspSem);
@@ -1534,6 +1547,8 @@ SMqPollReq* tmqBuildConsumeReqImpl(tmq_t* tmq, int64_t timeout, SMqClientTopic*
   pReq->currentOffset = reqOffset;
   pReq->reqId = generateRequestId();
 
+  pReq->useSnapshot = tmq->useSnapshot;
+
   pReq->head.vgId = htonl(pVg->vgId);
   pReq->head.contLen = htonl(sizeof(SMqPollReq));
   return pReq;
@@ -1541,7 +1556,7 @@ SMqPollReq* tmqBuildConsumeReqImpl(tmq_t* tmq, int64_t timeout, SMqClientTopic*
 
 SMqMetaRspObj* tmqBuildMetaRspFromWrapper(SMqPollRspWrapper* pWrapper) {
   SMqMetaRspObj* pRspObj = taosMemoryCalloc(1, sizeof(SMqMetaRspObj));
-  pRspObj->resType = RES_TYPE__TMQ;
+  pRspObj->resType = RES_TYPE__TMQ_META;
   tstrncpy(pRspObj->topic, pWrapper->topicHandle->topicName, TSDB_TOPIC_FNAME_LEN);
   tstrncpy(pRspObj->db, pWrapper->topicHandle->db, TSDB_DB_FNAME_LEN);
   pRspObj->vgId = pWrapper->vgHandle->vgId;
@@ -1659,7 +1674,7 @@ int32_t tmqHandleNoPollRsp(tmq_t* tmq, SMqRspWrapper* rspWrapper, bool* pReset)
   return 0;
 }
 
-SMqRspObj* tmqHandleAllRsp(tmq_t* tmq, int64_t timeout, bool pollIfReset) {
+void* tmqHandleAllRsp(tmq_t* tmq, int64_t timeout, bool pollIfReset) {
   while (1) {
     SMqRspWrapper* rspWrapper = NULL;
     taosGetQitem(tmq->qall, (void**)&rspWrapper);
@@ -1699,18 +1714,18 @@ SMqRspObj* tmqHandleAllRsp(tmq_t* tmq, int64_t timeout, bool pollIfReset) {
     } else if (rspWrapper->tmqRspType == TMQ_MSG_TYPE__POLL_META_RSP) {
       SMqPollRspWrapper* pollRspWrapper = (SMqPollRspWrapper*)rspWrapper;
       int32_t            consumerEpoch = atomic_load_32(&tmq->epoch);
-      if (pollRspWrapper->dataRsp.head.epoch == consumerEpoch) {
+      if (pollRspWrapper->metaRsp.head.epoch == consumerEpoch) {
         SMqClientVg* pVg = pollRspWrapper->vgHandle;
         /*printf("vg %d offset %ld up to %ld\n", pVg->vgId, pVg->currentOffset, rspMsg->msg.rspOffset);*/
-        pVg->currentOffset = pollRspWrapper->dataRsp.rspOffset;
+        pVg->currentOffset = pollRspWrapper->metaRsp.rspOffset;
         atomic_store_32(&pVg->vgStatus, TMQ_VG_STATUS__IDLE);
         // build rsp
-        SMqRspObj* pRsp = tmqBuildRspFromWrapper(pollRspWrapper);
+        SMqMetaRspObj* pRsp = tmqBuildMetaRspFromWrapper(pollRspWrapper);
         taosFreeQitem(pollRspWrapper);
         return pRsp;
       } else {
         tscDebug("msg discard since epoch mismatch: msg epoch %d, consumer epoch %d\n",
-                 pollRspWrapper->dataRsp.head.epoch, consumerEpoch);
+                 pollRspWrapper->metaRsp.head.epoch, consumerEpoch);
         taosFreeQitem(pollRspWrapper);
       }
     } else {
@@ -1727,7 +1742,7 @@ SMqRspObj* tmqHandleAllRsp(tmq_t* tmq, int64_t timeout, bool pollIfReset) {
 }
 
 TAOS_RES* tmq_consumer_poll(tmq_t* tmq, int64_t timeout) {
-  SMqRspObj* rspObj;
+  void*   rspObj;
   int64_t startTime = taosGetTimestampMs();
 
 #if 0
@@ -1856,7 +1871,7 @@ const char* tmq_get_table_name(TAOS_RES* res) {
   return NULL;
 }
 
-int32_t tmq_get_raw_meta(TAOS_RES* res, const void** raw_meta, int32_t* raw_meta_len) {
+int32_t tmq_get_raw_meta(TAOS_RES* res, void** raw_meta, int32_t* raw_meta_len) {
   if (TD_RES_TMQ_META(res)) {
     SMqMetaRspObj* pMetaRspObj = (SMqMetaRspObj*)res;
     *raw_meta = pMetaRspObj->metaRsp.metaRsp;

source/common/src/systable.c

@@ -231,7 +231,13 @@ static const SSysDbTableSchema transSchema[] = {
 
 static const SSysDbTableSchema configSchema[] = {
     {.name = "name", .bytes = TSDB_CONFIG_OPTION_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
-    {.name = "value", .bytes = TSDB_CONIIG_VALUE_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
+    {.name = "value", .bytes = TSDB_CONFIG_VALUE_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
+};
+
+static const SSysDbTableSchema variablesSchema[] = {
+    {.name = "dnode_id", .bytes = 4, .type = TSDB_DATA_TYPE_INT},
+    {.name = "name", .bytes = TSDB_CONFIG_OPTION_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
+    {.name = "value", .bytes = TSDB_CONFIG_VALUE_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
 };
 
 static const SSysTableMeta infosMeta[] = {
@@ -253,6 +259,7 @@ static const SSysTableMeta infosMeta[] = {
     {TSDB_INS_TABLE_LICENCES, grantsSchema, tListLen(grantsSchema)},
     {TSDB_INS_TABLE_VGROUPS, vgroupsSchema, tListLen(vgroupsSchema)},
     {TSDB_INS_TABLE_CONFIGS, configSchema, tListLen(configSchema)},
+    {TSDB_INS_TABLE_DNODE_VARIABLES, variablesSchema, tListLen(variablesSchema)},
 };
 
 static const SSysDbTableSchema connectionsSchema[] = {

source/common/src/tdatablock.c

@@ -110,7 +110,7 @@ int32_t getJsonValueLen(const char* data) {
     dataLen = DOUBLE_BYTES + CHAR_BYTES;
   } else if (*data == TSDB_DATA_TYPE_BOOL) {
     dataLen = CHAR_BYTES + CHAR_BYTES;
-  } else if (*data & TD_TAG_JSON) {  // json string
+  } else if (tTagIsJson(data)) {  // json string
     dataLen = ((STag*)(data))->len;
   } else {
     ASSERT(0);

source/common/src/tdataformat.c

@@ -924,6 +924,18 @@ static int32_t tGetTagVal(uint8_t *p, STagVal *pTagVal, int8_t isJson) {
 
   return n;
 }
+
+bool tTagIsJson(const void *pTag){
+  return (((const STag *)pTag)->flags & TD_TAG_JSON);
+}
+
+bool tTagIsJsonNull(void *data){
+  STag *pTag = (STag*)data;
+  int8_t   isJson = tTagIsJson(pTag);
+  if(!isJson) return false;
+  return ((STag*)data)->nTag == 0;
+}
+
 int32_t tTagNew(SArray *pArray, int32_t version, int8_t isJson, STag **ppTag) {
   int32_t  code = 0;
   uint8_t *p = NULL;

source/common/src/tglobal.c

@@ -187,6 +187,9 @@ bool tsStartUdfd = true;
 // internal
 int32_t tsTransPullupInterval = 2;
 int32_t tsMqRebalanceInterval = 2;
+int32_t tsTtlUnit = 86400;
+int32_t tsTtlPushInterval = 60;
+
 
 void taosAddDataDir(int32_t index, char *v1, int32_t level, int32_t primary) {
   tstrncpy(tsDiskCfg[index].dir, v1, TSDB_FILENAME_LEN);
@@ -467,6 +470,8 @@ static int32_t taosAddServerCfg(SConfig *pCfg) {
 
   if (cfgAddInt32(pCfg, "transPullupInterval", tsTransPullupInterval, 1, 10000, 1) != 0) return -1;
   if (cfgAddInt32(pCfg, "mqRebalanceInterval", tsMqRebalanceInterval, 1, 10000, 1) != 0) return -1;
+  if (cfgAddInt32(pCfg, "ttlUnit", tsTtlUnit, 1, 86400*365, 1) != 0) return -1;
+  if (cfgAddInt32(pCfg, "ttlPushInterval", tsTtlPushInterval, 1, 10000, 1) != 0) return -1;
 
   if (cfgAddBool(pCfg, "udf", tsStartUdfd, 0) != 0) return -1;
   return 0;
@@ -619,6 +624,8 @@ static int32_t taosSetServerCfg(SConfig *pCfg) {
 
   tsTransPullupInterval = cfgGetItem(pCfg, "transPullupInterval")->i32;
   tsMqRebalanceInterval = cfgGetItem(pCfg, "mqRebalanceInterval")->i32;
+  tsTtlUnit = cfgGetItem(pCfg, "ttlUnit")->i32;
+  tsTtlPushInterval = cfgGetItem(pCfg, "ttlPushInterval")->i32;
 
   tsStartUdfd = cfgGetItem(pCfg, "udf")->bval;
 
@@ -631,7 +638,7 @@ static int32_t taosSetServerCfg(SConfig *pCfg) {
 
 int32_t taosCreateLog(const char *logname, int32_t logFileNum, const char *cfgDir, const char **envCmd,
                       const char *envFile, char *apolloUrl, SArray *pArgs, bool tsc) {
-  osDefaultInit();
+  if (tsCfg == NULL) osDefaultInit();
 
   SConfig *pCfg = cfgInit();
   if (pCfg == NULL) return -1;

source/common/src/tmsg.c

@@ -2198,6 +2198,32 @@ int32_t tDeserializeSQnodeListReq(void *buf, int32_t bufLen, SQnodeListReq *pReq
   return 0;
 }
 
+int32_t tSerializeSDnodeListReq(void *buf, int32_t bufLen, SDnodeListReq *pReq) {
+  SEncoder encoder = {0};
+  tEncoderInit(&encoder, buf, bufLen);
+
+  if (tStartEncode(&encoder) < 0) return -1;
+  if (tEncodeI32(&encoder, pReq->rowNum) < 0) return -1;
+  tEndEncode(&encoder);
+
+  int32_t tlen = encoder.pos;
+  tEncoderClear(&encoder);
+  return tlen;
+}
+
+int32_t tDeserializeSDnodeListReq(void *buf, int32_t bufLen, SDnodeListReq *pReq) {
+  SDecoder decoder = {0};
+  tDecoderInit(&decoder, buf, bufLen);
+
+  if (tStartDecode(&decoder) < 0) return -1;
+  if (tDecodeI32(&decoder, &pReq->rowNum) < 0) return -1;
+  tEndDecode(&decoder);
+
+  tDecoderClear(&decoder);
+  return 0;
+}
+
+
 int32_t tSerializeSQnodeListRsp(void *buf, int32_t bufLen, SQnodeListRsp *pRsp) {
   SEncoder encoder = {0};
   tEncoderInit(&encoder, buf, bufLen);
@@ -2241,6 +2267,50 @@ int32_t tDeserializeSQnodeListRsp(void *buf, int32_t bufLen, SQnodeListRsp *pRsp
 
 void tFreeSQnodeListRsp(SQnodeListRsp *pRsp) { taosArrayDestroy(pRsp->qnodeList); }
 
+int32_t tSerializeSDnodeListRsp(void *buf, int32_t bufLen, SDnodeListRsp *pRsp) {
+  SEncoder encoder = {0};
+  tEncoderInit(&encoder, buf, bufLen);
+
+  if (tStartEncode(&encoder) < 0) return -1;
+  int32_t num = taosArrayGetSize(pRsp->dnodeList);
+  if (tEncodeI32(&encoder, num) < 0) return -1;
+  for (int32_t i = 0; i < num; ++i) {
+    SEpSet *pEpSet = taosArrayGet(pRsp->dnodeList, i);
+    if (tEncodeSEpSet(&encoder, pEpSet) < 0) return -1;
+  }
+  tEndEncode(&encoder);
+
+  int32_t tlen = encoder.pos;
+  tEncoderClear(&encoder);
+  return tlen;
+}
+
+int32_t tDeserializeSDnodeListRsp(void *buf, int32_t bufLen, SDnodeListRsp *pRsp) {
+  SDecoder decoder = {0};
+  tDecoderInit(&decoder, buf, bufLen);
+
+  if (tStartDecode(&decoder) < 0) return -1;
+  int32_t num = 0;
+  if (tDecodeI32(&decoder, &num) < 0) return -1;
+  if (NULL == pRsp->dnodeList) {
+    pRsp->dnodeList = taosArrayInit(num, sizeof(SEpSet));
+    if (NULL == pRsp->dnodeList) return -1;
+  }
+
+  for (int32_t i = 0; i < num; ++i) {
+    SEpSet epSet = {0};
+    if (tDecodeSEpSet(&decoder, &epSet) < 0) return -1;
+    taosArrayPush(pRsp->dnodeList, &epSet);
+  }
+  tEndDecode(&decoder);
+
+  tDecoderClear(&decoder);
+  return 0;
+}
+
+void tFreeSDnodeListRsp(SDnodeListRsp *pRsp) { taosArrayDestroy(pRsp->dnodeList); }
+
+
 int32_t tSerializeSCompactDbReq(void *buf, int32_t bufLen, SCompactDbReq *pReq) {
   SEncoder encoder = {0};
   tEncoderInit(&encoder, buf, bufLen);

source/common/src/ttime.c

@@ -76,22 +76,28 @@ void           deltaToUtcInitOnce() {
 
 static int64_t parseFraction(char* str, char** end, int32_t timePrec);
 static int32_t parseTimeWithTz(const char* timestr, int64_t* time, int32_t timePrec, char delim);
-static int32_t parseLocaltime(char* timestr, int32_t len, int64_t* utime, int32_t timePrec);
+static int32_t parseLocaltime(char* timestr, int32_t len, int64_t* utime, int32_t timePrec, char delim);
-static int32_t parseLocaltimeDst(char* timestr, int32_t len, int64_t* utime, int32_t timePrec);
+static int32_t parseLocaltimeDst(char* timestr, int32_t len, int64_t* utime, int32_t timePrec, char delim);
 static char*   forwardToTimeStringEnd(char* str);
 static bool    checkTzPresent(const char* str, int32_t len);
 
-static int32_t (*parseLocaltimeFp[])(char* timestr, int32_t len, int64_t* utime, int32_t timePrec) = {parseLocaltime,
+static int32_t (*parseLocaltimeFp[])(char* timestr, int32_t len, int64_t* utime, int32_t timePrec, char delim) = {parseLocaltime,
                                                                                                                   parseLocaltimeDst};
 
 int32_t taosParseTime(const char* timestr, int64_t* utime, int32_t len, int32_t timePrec, int8_t day_light) {
   /* parse datatime string in with tz */
   if (strnchr(timestr, 'T', len, false) != NULL) {
+    if (checkTzPresent(timestr, len)) {
       return parseTimeWithTz(timestr, utime, timePrec, 'T');
-  } else if (checkTzPresent(timestr, len)) {
+    } else {
+      return (*parseLocaltimeFp[day_light])((char*)timestr, len, utime, timePrec, 'T');
+    }
+  } else {
+    if (checkTzPresent(timestr, len)) {
       return parseTimeWithTz(timestr, utime, timePrec, 0);
     } else {
-    return (*parseLocaltimeFp[day_light])((char*)timestr, len, utime, timePrec);
+      return (*parseLocaltimeFp[day_light])((char*)timestr, len, utime, timePrec, 0);
+    }
   }
 }
 
@@ -333,14 +339,26 @@ static FORCE_INLINE bool validateTm(struct tm* pTm) {
   return  true;
 }
 
-int32_t parseLocaltime(char* timestr, int32_t len, int64_t* time, int32_t timePrec) {
+int32_t parseLocaltime(char* timestr, int32_t len, int64_t* time, int32_t timePrec, char delim) {
   *time = 0;
   struct tm tm = {0};
 
-  char* str = taosStrpTime(timestr, "%Y-%m-%d %H:%M:%S", &tm);
+  char *str;
+  if (delim == 'T') {
+    str = taosStrpTime(timestr, "%Y-%m-%dT%H:%M:%S", &tm);
+  } else if (delim == 0) {
+    str = taosStrpTime(timestr, "%Y-%m-%d %H:%M:%S", &tm);
+  } else {
+    str = NULL;
+  }
+
+  if (str == NULL || (((str - timestr) < len) && (*str != '.')) || !validateTm(&tm)) {
+    //if parse failed, try "%Y-%m-%d" format
+    str = taosStrpTime(timestr, "%Y-%m-%d", &tm);
     if (str == NULL || (((str - timestr) < len) && (*str != '.')) || !validateTm(&tm)) {
       return -1;
     }
+  }
 
 #ifdef _MSC_VER
 #if _MSC_VER >= 1900
@@ -367,15 +385,27 @@ int32_t parseLocaltime(char* timestr, int32_t len, int64_t* time, int32_t timePr
   return 0;
 }
 
-int32_t parseLocaltimeDst(char* timestr, int32_t len, int64_t* time, int32_t timePrec) {
+int32_t parseLocaltimeDst(char* timestr, int32_t len, int64_t* time, int32_t timePrec, char delim) {
   *time = 0;
   struct tm tm = {0};
   tm.tm_isdst = -1;
 
-  char* str = taosStrpTime(timestr, "%Y-%m-%d %H:%M:%S", &tm);
+  char *str;
+  if (delim == 'T') {
+    str = taosStrpTime(timestr, "%Y-%m-%dT%H:%M:%S", &tm);
+  } else if (delim == 0) {
+    str = taosStrpTime(timestr, "%Y-%m-%d %H:%M:%S", &tm);
+  } else {
+    str = NULL;
+  }
+
+  if (str == NULL || (((str - timestr) < len) && (*str != '.')) || !validateTm(&tm)) {
+    //if parse failed, try "%Y-%m-%d" format
+    str = taosStrpTime(timestr, "%Y-%m-%d", &tm);
     if (str == NULL || (((str - timestr) < len) && (*str != '.')) || !validateTm(&tm)) {
       return -1;
     }
+  }
 
   /* mktime will be affected by TZ, set by using taos_options */
   int64_t seconds = taosMktime(&tm);

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

@@ -45,6 +45,7 @@ int32_t dmProcessConfigReq(SDnodeMgmt *pMgmt, SRpcMsg *pMsg);
 int32_t dmProcessAuthRsp(SDnodeMgmt *pMgmt, SRpcMsg *pMsg);
 int32_t dmProcessGrantRsp(SDnodeMgmt *pMgmt, SRpcMsg *pMsg);
 int32_t dmProcessServerRunStatus(SDnodeMgmt *pMgmt, SRpcMsg *pMsg);
+int32_t dmProcessRetrieve(SDnodeMgmt *pMgmt, SRpcMsg *pMsg);
 
 // dmWorker.c
 int32_t dmPutNodeMsgToMgmtQueue(SDnodeMgmt *pMgmt, SRpcMsg *pMsg);

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

@@ -15,6 +15,10 @@
 
 #define _DEFAULT_SOURCE
 #include "dmInt.h"
+#include "systable.h"
+
+
+extern SConfig *tsCfg;
 
 static void dmUpdateDnodeCfg(SDnodeMgmt *pMgmt, SDnodeCfg *pCfg) {
   if (pMgmt->pData->dnodeId == 0 || pMgmt->pData->clusterId == 0) {
@@ -175,6 +179,130 @@ int32_t dmProcessServerRunStatus(SDnodeMgmt *pMgmt, SRpcMsg *pMsg) {
   return 0;
 }
 
+SSDataBlock* dmBuildVariablesBlock(void) {
+  SSDataBlock* pBlock = taosMemoryCalloc(1, sizeof(SSDataBlock));
+  size_t               size = 0;
+  const SSysTableMeta* pMeta = NULL;
+  getInfosDbMeta(&pMeta, &size);
+
+  int32_t index = 0;
+  for (int32_t i = 0; i < size; ++i) {
+    if (strcmp(pMeta[i].name, TSDB_INS_TABLE_DNODE_VARIABLES) == 0) {
+      index = i;
+      break;
+    }
+  }
+
+  pBlock->pDataBlock = taosArrayInit(pMeta[index].colNum, sizeof(SColumnInfoData));
+
+  for (int32_t i = 0; i < pMeta[index].colNum; ++i) {
+    SColumnInfoData colInfoData = {0};
+    colInfoData.info.colId = i + 1;
+    colInfoData.info.type = pMeta[index].schema[i].type;
+    colInfoData.info.bytes = pMeta[index].schema[i].bytes;
+    taosArrayPush(pBlock->pDataBlock, &colInfoData);
+  }
+
+  pBlock->info.numOfCols = pMeta[index].colNum;
+  pBlock->info.hasVarCol = true;
+
+  return pBlock;
+}
+
+int32_t dmAppendVariablesToBlock(SSDataBlock* pBlock, int32_t dnodeId) {
+  int32_t numOfCfg = taosArrayGetSize(tsCfg->array);
+  int32_t numOfRows = 0;
+  blockDataEnsureCapacity(pBlock, numOfCfg);
+
+  for (int32_t i = 0, c = 0; i < numOfCfg; ++i, c = 0) {
+    SConfigItem *pItem = taosArrayGet(tsCfg->array, i);
+
+    SColumnInfoData *pColInfo = taosArrayGet(pBlock->pDataBlock, c++);
+    colDataAppend(pColInfo, i, (const char *)&dnodeId, false);
+
+    char name[TSDB_CONFIG_OPTION_LEN + VARSTR_HEADER_SIZE] = {0};
+    STR_WITH_MAXSIZE_TO_VARSTR(name, pItem->name, TSDB_CONFIG_OPTION_LEN + VARSTR_HEADER_SIZE);
+    pColInfo = taosArrayGet(pBlock->pDataBlock, c++);
+    colDataAppend(pColInfo, i, name, false);
+    
+    char value[TSDB_CONFIG_VALUE_LEN + VARSTR_HEADER_SIZE] = {0};
+    int32_t valueLen = 0;
+    cfgDumpItemValue(pItem, &value[VARSTR_HEADER_SIZE], TSDB_CONFIG_VALUE_LEN, &valueLen);
+    varDataSetLen(value, valueLen);
+    pColInfo = taosArrayGet(pBlock->pDataBlock, c++);
+    colDataAppend(pColInfo, i, value, false);
+
+    numOfRows++;
+  }
+
+
+  pBlock->info.rows = numOfRows;
+  
+  return TSDB_CODE_SUCCESS;
+}
+
+
+
+int32_t dmProcessRetrieve(SDnodeMgmt *pMgmt, SRpcMsg *pMsg) {
+  int32_t    size = 0;
+  int32_t    rowsRead = 0;
+
+  SRetrieveTableReq retrieveReq = {0};
+  if (tDeserializeSRetrieveTableReq(pMsg->pCont, pMsg->contLen, &retrieveReq) != 0) {
+    terrno = TSDB_CODE_INVALID_MSG;
+    return -1;
+  }
+
+  if (strcasecmp(retrieveReq.tb, TSDB_INS_TABLE_DNODE_VARIABLES)) {
+    terrno = TSDB_CODE_INVALID_MSG;
+    return -1;
+  }
+
+  SSDataBlock* pBlock = dmBuildVariablesBlock();
+
+  dmAppendVariablesToBlock(pBlock, pMgmt->pData->dnodeId);
+
+  size = sizeof(SRetrieveMetaTableRsp) + sizeof(int32_t) + sizeof(SSysTableSchema) * pBlock->info.numOfCols +
+         blockDataGetSize(pBlock) + blockDataGetSerialMetaSize(pBlock->info.numOfCols);
+
+  SRetrieveMetaTableRsp *pRsp = rpcMallocCont(size);
+  if (pRsp == NULL) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    dError("failed to retrieve data since %s", terrstr());
+    blockDataDestroy(pBlock);
+    return -1;
+  }
+
+  char    *pStart = pRsp->data;
+  *(int32_t *)pStart = htonl(pBlock->info.numOfCols);
+  pStart += sizeof(int32_t);  // number of columns
+
+  for (int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
+    SSysTableSchema *pSchema = (SSysTableSchema *)pStart;
+    SColumnInfoData *pColInfo = taosArrayGet(pBlock->pDataBlock, i);
+    
+    pSchema->bytes = htonl(pColInfo->info.bytes);
+    pSchema->colId = htons(pColInfo->info.colId);
+    pSchema->type = pColInfo->info.type;
+
+    pStart += sizeof(SSysTableSchema);
+  }
+
+  int32_t len = 0;
+  blockCompressEncode(pBlock, pStart, &len, pBlock->info.numOfCols, false);
+
+  pRsp->numOfRows = htonl(pBlock->info.rows);
+  pRsp->precision = TSDB_TIME_PRECISION_MILLI;  // millisecond time precision
+  pRsp->completed = 1;
+  pMsg->info.rsp = pRsp;
+  pMsg->info.rspLen = size;
+  dDebug("dnode variables retrieve completed");
+
+  blockDataDestroy(pBlock);
+  return TSDB_CODE_SUCCESS;
+}
+
+
 SArray *dmGetMsgHandles() {
   int32_t code = -1;
   SArray *pArray = taosArrayInit(16, sizeof(SMgmtHandle));
@@ -191,6 +319,7 @@ SArray *dmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_DND_DROP_BNODE, dmPutNodeMsgToMgmtQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_DND_CONFIG_DNODE, dmPutNodeMsgToMgmtQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_DND_SERVER_STATUS, dmPutNodeMsgToMgmtQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_DND_SYSTABLE_RETRIEVE, dmPutNodeMsgToMgmtQueue, 0) == NULL) goto _OVER;
 
   // Requests handled by MNODE
   if (dmSetMgmtHandle(pArray, TDMT_MND_GRANT_RSP, dmPutNodeMsgToMgmtQueue, 0) == NULL) goto _OVER;

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

@@ -141,6 +141,9 @@ static void dmProcessMgmtQueue(SQueueInfo *pInfo, SRpcMsg *pMsg) {
     case TDMT_DND_SERVER_STATUS:
       code = dmProcessServerRunStatus(pMgmt, pMsg);
       break;
+    case TDMT_DND_SYSTABLE_RETRIEVE:
+      code = dmProcessRetrieve(pMgmt, pMsg);
+      break;
     default:
       terrno = TSDB_CODE_MSG_NOT_PROCESSED;
       break;

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

@@ -161,6 +161,7 @@ SArray *mmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_MND_CREATE_QNODE, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_MND_DROP_QNODE, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_MND_QNODE_LIST, mmPutMsgToReadQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_MND_DNODE_LIST, mmPutMsgToReadQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_MND_CREATE_SNODE, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_MND_DROP_SNODE, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_MND_CREATE_BNODE, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;

source/dnode/mgmt/mgmt_snode/src/smHandle.c

@@ -101,6 +101,8 @@ SArray *smGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DISPATCH_RSP, smPutNodeMsgToSharedQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RECOVER, smPutNodeMsgToSharedQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RECOVER_RSP, smPutNodeMsgToSharedQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_STREAM_RETRIEVE, smPutNodeMsgToSharedQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_STREAM_RETRIEVE_RSP, smPutNodeMsgToSharedQueue, 1) == NULL) goto _OVER;
 
   code = 0;
 _OVER:

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

@@ -359,6 +359,8 @@ SArray *vmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DISPATCH_RSP, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RECOVER, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RECOVER_RSP, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_STREAM_RETRIEVE, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_STREAM_RETRIEVE_RSP, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
 
   if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_REPLICA, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_CONFIG, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;

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

@@ -85,6 +85,7 @@ static void dmProcessRpcMsg(SDnode *pDnode, SRpcMsg *pRpc, SEpSet *pEpSet) {
       dmProcessNetTestReq(pDnode, pRpc);
       return;
     case TDMT_MND_SYSTABLE_RETRIEVE_RSP:
+    case TDMT_DND_SYSTABLE_RETRIEVE_RSP:
     case TDMT_VND_FETCH_RSP:
       qWorkerProcessFetchRsp(NULL, NULL, pRpc, 0);
       return;

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

@@ -47,6 +47,7 @@ static SSdbRow *mndDnodeActionDecode(SSdbRaw *pRaw);
 static int32_t  mndDnodeActionInsert(SSdb *pSdb, SDnodeObj *pDnode);
 static int32_t  mndDnodeActionDelete(SSdb *pSdb, SDnodeObj *pDnode);
 static int32_t  mndDnodeActionUpdate(SSdb *pSdb, SDnodeObj *pOld, SDnodeObj *pNew);
+static int32_t  mndProcessDnodeListReq(SRpcMsg *pReq);
 
 static int32_t mndProcessCreateDnodeReq(SRpcMsg *pReq);
 static int32_t mndProcessDropDnodeReq(SRpcMsg *pReq);
@@ -76,6 +77,7 @@ int32_t mndInitDnode(SMnode *pMnode) {
   mndSetMsgHandle(pMnode, TDMT_MND_CONFIG_DNODE, mndProcessConfigDnodeReq);
   mndSetMsgHandle(pMnode, TDMT_DND_CONFIG_DNODE_RSP, mndProcessConfigDnodeRsp);
   mndSetMsgHandle(pMnode, TDMT_MND_STATUS, mndProcessStatusReq);
+  mndSetMsgHandle(pMnode, TDMT_MND_DNODE_LIST, mndProcessDnodeListReq);
 
   mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_CONFIGS, mndRetrieveConfigs);
   mndAddShowFreeIterHandle(pMnode, TSDB_MGMT_TABLE_CONFIGS, mndCancelGetNextConfig);
@@ -499,6 +501,60 @@ _OVER:
   return code;
 }
 
+static int32_t mndProcessDnodeListReq(SRpcMsg *pReq) {
+  SMnode *pMnode = pReq->info.node;
+  SSdb   *pSdb = pMnode->pSdb;
+  SDnodeObj *pObj = NULL;
+  void *pIter = NULL;
+  SDnodeListRsp rsp = {0};
+  int32_t code = -1;
+  
+  rsp.dnodeList = taosArrayInit(5, sizeof(SEpSet));
+  if (NULL == rsp.dnodeList) {
+    mError("failed to alloc epSet while process dnode list req");
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    goto _OVER;
+  }
+  
+  while (1) {
+    pIter = sdbFetch(pSdb, SDB_DNODE, pIter, (void **)&pObj);
+    if (pIter == NULL) break;
+
+    SEpSet epSet = {0};
+    epSet.numOfEps = 1;
+    tstrncpy(epSet.eps[0].fqdn, pObj->fqdn, TSDB_FQDN_LEN);
+    epSet.eps[0].port = pObj->port;
+
+    (void)taosArrayPush(rsp.dnodeList, &epSet);
+
+    sdbRelease(pSdb, pObj);
+  }
+
+  int32_t rspLen = tSerializeSDnodeListRsp(NULL, 0, &rsp);
+  void   *pRsp = rpcMallocCont(rspLen);
+  if (pRsp == NULL) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    goto _OVER;
+  }
+
+  tSerializeSDnodeListRsp(pRsp, rspLen, &rsp);
+
+  pReq->info.rspLen = rspLen;
+  pReq->info.rsp = pRsp;
+  code = 0;
+
+_OVER:
+
+  if (code != 0) {
+    mError("failed to get dnode list since %s", terrstr());
+  }
+
+  tFreeSDnodeListRsp(&rsp);
+
+  return code;
+}
+
+
 static int32_t mndProcessCreateDnodeReq(SRpcMsg *pReq) {
   SMnode         *pMnode = pReq->info.node;
   int32_t         code = -1;
@@ -580,6 +636,7 @@ static int32_t mndDropDnode(SMnode *pMnode, SRpcMsg *pReq, SDnodeObj *pDnode, SM
     mInfo("trans:%d, %d vnodes on dnode:%d will be dropped", pTrans->id, numOfVnodes, pDnode->id);
     if (mndSetMoveVgroupsInfoToTrans(pMnode, pTrans, pDnode->id) != 0) goto _OVER;
   }
+
   if (mndTransPrepare(pMnode, pTrans) != 0) goto _OVER;
 
   code = 0;
@@ -597,25 +654,25 @@ static int32_t mndProcessDropDnodeReq(SRpcMsg *pReq) {
   SMnodeObj    *pMObj = NULL;
   SQnodeObj    *pQObj = NULL;
   SSnodeObj    *pSObj = NULL;
-  SMDropMnodeReq dropReq = {0};
+  SDropDnodeReq dropReq = {0};
 
-  if (tDeserializeSCreateDropMQSBNodeReq(pReq->pCont, pReq->contLen, &dropReq) != 0) {
+  if (tDeserializeSDropDnodeReq(pReq->pCont, pReq->contLen, &dropReq) != 0) {
     terrno = TSDB_CODE_INVALID_MSG;
     goto _OVER;
   }
 
-  mInfo("dnode:%d, start to drop", dropReq.dnodeId);
+  mInfo("dnode:%d, start to drop, ep:%s:%d", dropReq.dnodeId, dropReq.fqdn, dropReq.port);
-
-  if (dropReq.dnodeId <= 0) {
-    terrno = TSDB_CODE_MND_INVALID_DNODE_ID;
-    goto _OVER;
-  }
 
   pDnode = mndAcquireDnode(pMnode, dropReq.dnodeId);
+  if (pDnode == NULL) {
+    char ep[TSDB_EP_LEN + 1] = {0};
+    snprintf(ep, sizeof(ep), dropReq.fqdn, dropReq.port);
+    pDnode = mndAcquireDnodeByEp(pMnode, ep);
     if (pDnode == NULL) {
       terrno = TSDB_CODE_MND_DNODE_NOT_EXIST;
       goto _OVER;
     }
+  }
 
   pQObj = mndAcquireQnode(pMnode, dropReq.dnodeId);
   pSObj = mndAcquireSnode(pMnode, dropReq.dnodeId);
@@ -669,6 +726,7 @@ static int32_t mndProcessConfigDnodeReq(SRpcMsg *pReq) {
     return -1;
   }
 
+  mInfo("dnode:%d, start to config, option:%s, value:%s", cfgReq.dnodeId, cfgReq.config, cfgReq.value);
   SDnodeObj *pDnode = mndAcquireDnode(pMnode, cfgReq.dnodeId);
   if (pDnode == NULL) {
     mError("dnode:%d, failed to config since %s ", cfgReq.dnodeId, terrstr());
@@ -685,7 +743,7 @@ static int32_t mndProcessConfigDnodeReq(SRpcMsg *pReq) {
   tSerializeSMCfgDnodeReq(pBuf, bufLen, &cfgReq);
 
   mDebug("dnode:%d, send config req to dnode, app:%p", cfgReq.dnodeId, pReq->info.ahandle);
-  SRpcMsg rpcMsg = {.msgType = TDMT_DND_CONFIG_DNODE, .pCont = pBuf, .contLen = bufLen, .info = pReq->info};
+  SRpcMsg rpcMsg = {.msgType = TDMT_DND_CONFIG_DNODE, .pCont = pBuf, .contLen = bufLen};
   return tmsgSendReq(&epSet, &rpcMsg);
 }
 
@@ -699,28 +757,28 @@ static int32_t mndRetrieveConfigs(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *p
   int32_t totalRows = 0;
   int32_t numOfRows = 0;
   char   *cfgOpts[TSDB_CONFIG_NUMBER] = {0};
-  char    cfgVals[TSDB_CONFIG_NUMBER][TSDB_CONIIG_VALUE_LEN + 1] = {0};
+  char    cfgVals[TSDB_CONFIG_NUMBER][TSDB_CONFIG_VALUE_LEN + 1] = {0};
   char   *pWrite = NULL;
   int32_t cols = 0;
 
   cfgOpts[totalRows] = "statusInterval";
-  snprintf(cfgVals[totalRows], TSDB_CONIIG_VALUE_LEN, "%d", tsStatusInterval);
+  snprintf(cfgVals[totalRows], TSDB_CONFIG_VALUE_LEN, "%d", tsStatusInterval);
   totalRows++;
 
   cfgOpts[totalRows] = "timezone";
-  snprintf(cfgVals[totalRows], TSDB_CONIIG_VALUE_LEN, "%s", tsTimezoneStr);
+  snprintf(cfgVals[totalRows], TSDB_CONFIG_VALUE_LEN, "%s", tsTimezoneStr);
   totalRows++;
 
   cfgOpts[totalRows] = "locale";
-  snprintf(cfgVals[totalRows], TSDB_CONIIG_VALUE_LEN, "%s", tsLocale);
+  snprintf(cfgVals[totalRows], TSDB_CONFIG_VALUE_LEN, "%s", tsLocale);
   totalRows++;
 
   cfgOpts[totalRows] = "charset";
-  snprintf(cfgVals[totalRows], TSDB_CONIIG_VALUE_LEN, "%s", tsCharset);
+  snprintf(cfgVals[totalRows], TSDB_CONFIG_VALUE_LEN, "%s", tsCharset);
   totalRows++;
 
   char buf[TSDB_CONFIG_OPTION_LEN + VARSTR_HEADER_SIZE] = {0};
-  char bufVal[TSDB_CONIIG_VALUE_LEN + VARSTR_HEADER_SIZE] = {0};
+  char bufVal[TSDB_CONFIG_VALUE_LEN + VARSTR_HEADER_SIZE] = {0};
 
   for (int32_t i = 0; i < totalRows; i++) {
     cols = 0;
@@ -729,7 +787,7 @@ static int32_t mndRetrieveConfigs(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *p
     SColumnInfoData *pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
     colDataAppend(pColInfo, numOfRows, (const char *)buf, false);
 
-    STR_WITH_MAXSIZE_TO_VARSTR(bufVal, cfgVals[i], TSDB_CONIIG_VALUE_LEN);
+    STR_WITH_MAXSIZE_TO_VARSTR(bufVal, cfgVals[i], TSDB_CONFIG_VALUE_LEN);
     pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
     colDataAppend(pColInfo, numOfRows, (const char *)bufVal, false);
 

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

@@ -65,6 +65,13 @@ static void mndPullupTrans(SMnode *pMnode) {
   }
 }
 
+static void mndTtlTimer(SMnode *pMnode) {
+  int32_t contLen = 0;
+  void   *pReq = mndBuildTimerMsg(&contLen);
+  SRpcMsg rpcMsg = {.msgType = TDMT_MND_TTL_TIMER, .pCont = pReq, .contLen = contLen};
+  tmsgPutToQueue(&pMnode->msgCb, WRITE_QUEUE, &rpcMsg);
+}
+
 static void mndCalMqRebalance(SMnode *pMnode) {
   int32_t contLen = 0;
   void   *pReq = mndBuildTimerMsg(&contLen);
@@ -83,41 +90,6 @@ static void mndPullupTelem(SMnode *pMnode) {
   }
 }
 
-static void mndPushTtlTime(SMnode *pMnode) {
-  SSdb   *pSdb = pMnode->pSdb;
-  SVgObj *pVgroup = NULL;
-  void   *pIter = NULL;
-
-  while (1) {
-    pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void **)&pVgroup);
-    if (pIter == NULL) break;
-
-    int32_t   contLen = sizeof(SMsgHead) + sizeof(int32_t);
-    SMsgHead *pHead = rpcMallocCont(contLen);
-    if (pHead == NULL) {
-      sdbCancelFetch(pSdb, pIter);
-      sdbRelease(pSdb, pVgroup);
-      continue;
-    }
-
-    pHead->contLen = htonl(contLen);
-    pHead->vgId = htonl(pVgroup->vgId);
-
-    int32_t t = taosGetTimestampSec();
-    *(int32_t *)(POINTER_SHIFT(pHead, sizeof(SMsgHead))) = htonl(t);
-
-    SRpcMsg rpcMsg = {.msgType = TDMT_VND_DROP_TTL_TABLE, .pCont = pHead, .contLen = contLen};
-    SEpSet  epSet = mndGetVgroupEpset(pMnode, pVgroup);
-    int32_t code = tmsgSendReq(&epSet, &rpcMsg);
-    if (code != 0) {
-      mError("failed to send ttl time seed msg, code:0x%x", code);
-    } else {
-      mInfo("send ttl time seed msg, time:%d", t);
-    }
-    sdbRelease(pSdb, pVgroup);
-  }
-}
-
 static void *mndThreadFp(void *param) {
   SMnode *pMnode = param;
   int64_t lastTime = 0;
@@ -125,14 +97,13 @@ static void *mndThreadFp(void *param) {
 
   while (1) {
     lastTime++;
-    
-    if (lastTime % (864000) == 0) {   // sleep 1 day for ttl
-      mndPushTtlTime(pMnode);
-    }
-
     taosMsleep(100);
     if (mndGetStop(pMnode)) break;
 
+    if (lastTime % (tsTransPullupInterval * 10) == 1) {
+      mndTtlTimer(pMnode);
+    }
+
     if (lastTime % (tsTransPullupInterval * 10) == 0) {
       mndPullupTrans(pMnode);
     }
@@ -558,12 +529,12 @@ static int32_t mndCheckMnodeState(SRpcMsg *pMsg) {
   if (!IsReq(pMsg)) return 0;
   if (mndAcquireRpcRef(pMsg->info.node) == 0) return 0;
   if (pMsg->msgType == TDMT_MND_MQ_TIMER || pMsg->msgType == TDMT_MND_TELEM_TIMER ||
-      pMsg->msgType == TDMT_MND_TRANS_TIMER) {
+      pMsg->msgType == TDMT_MND_TRANS_TIMER || TDMT_MND_TTL_TIMER) {
     return -1;
   }
 
   const STraceId *trace = &pMsg->info.traceId;
-  mGError("msg:%p, failed to check mnode state since %s, type:%s", pMsg, terrstr(), TMSG_INFO(pMsg->msgType));
+  mError("msg:%p, failed to check mnode state since %s, type:%s", pMsg, terrstr(), TMSG_INFO(pMsg->msgType));
 
   SEpSet epSet = {0};
   mndGetMnodeEpSet(pMsg->info.node, &epSet);

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

@@ -37,7 +37,7 @@ extern bool tsSchedStreamToSnode;
 
 static int32_t mndAddTaskToTaskSet(SArray* pArray, SStreamTask* pTask) {
   int32_t childId = taosArrayGetSize(pArray);
-  pTask->childId = childId;
+  pTask->selfChildId = childId;
   taosArrayPush(pArray, &pTask);
   return 0;
 }
@@ -271,7 +271,7 @@ int32_t mndAddShuffleSinkTasksToStream(SMnode* pMnode, STrans* pTrans, SStreamOb
     pTask->epSet = mndGetVgroupEpset(pMnode, pVgroup);
 
     // source
-    pTask->inputType = TASK_INPUT_TYPE__DATA_BLOCK;
+    pTask->isDataScan = 0;
 
     // exec
     pTask->execType = TASK_EXEC__NONE;
@@ -306,6 +306,8 @@ int32_t mndAddFixedSinkTaskToStream(SMnode* pMnode, STrans* pTrans, SStreamObj*
   }
   mndAddTaskToTaskSet(tasks, pTask);
 
+  ASSERT(pStream->fixedSinkVg.vgId == pStream->fixedSinkVgId);
+
   pTask->nodeId = pStream->fixedSinkVgId;
 #if 0
   SVgObj* pVgroup = mndAcquireVgroup(pMnode, pStream->fixedSinkVgId);
@@ -315,8 +317,9 @@ int32_t mndAddFixedSinkTaskToStream(SMnode* pMnode, STrans* pTrans, SStreamObj*
   pTask->epSet = mndGetVgroupEpset(pMnode, pVgroup);
 #endif
   pTask->epSet = mndGetVgroupEpset(pMnode, &pStream->fixedSinkVg);
+
   // source
-  pTask->inputType = TASK_INPUT_TYPE__DATA_BLOCK;
+  pTask->isDataScan = 0;
 
   // exec
   pTask->execType = TASK_EXEC__NONE;
@@ -384,8 +387,11 @@ int32_t mndScheduleStream(SMnode* pMnode, STrans* pTrans, SStreamObj* pStream) {
 
       pInnerTask = tNewSStreamTask(pStream->uid);
       mndAddTaskToTaskSet(taskInnerLevel, pInnerTask);
-      // input
+
-      pInnerTask->inputType = TASK_INPUT_TYPE__DATA_BLOCK;
+      pInnerTask->childEpInfo = taosArrayInit(0, sizeof(void*));
+
+      // source
+      pInnerTask->isDataScan = 0;
 
       // trigger
       pInnerTask->triggerParam = pStream->triggerParam;
@@ -446,10 +452,8 @@ int32_t mndScheduleStream(SMnode* pMnode, STrans* pTrans, SStreamObj* pStream) {
       SStreamTask* pTask = tNewSStreamTask(pStream->uid);
       mndAddTaskToTaskSet(taskSourceLevel, pTask);
 
-      pTask->dataScan = 1;
+      // source
-
+      pTask->isDataScan = 1;
-      // input
-      pTask->inputType = TASK_INPUT_TYPE__SUMBIT_BLOCK;
 
       // add fixed vg dispatch
       pTask->sinkType = TASK_SINK__NONE;
@@ -467,6 +471,20 @@ int32_t mndScheduleStream(SMnode* pMnode, STrans* pTrans, SStreamObj* pStream) {
         qDestroyQueryPlan(pPlan);
         return -1;
       }
+
+      SStreamChildEpInfo* pEpInfo = taosMemoryMalloc(sizeof(SStreamChildEpInfo));
+      if (pEpInfo == NULL) {
+        ASSERT(0);
+        terrno = TSDB_CODE_OUT_OF_MEMORY;
+        sdbRelease(pSdb, pVgroup);
+        qDestroyQueryPlan(pPlan);
+        return -1;
+      }
+      pEpInfo->childId = pTask->selfChildId;
+      pEpInfo->epSet = pTask->epSet;
+      pEpInfo->nodeId = pTask->nodeId;
+      pEpInfo->taskId = pTask->taskId;
+      taosArrayPush(pInnerTask->childEpInfo, &pEpInfo);
     }
   }
 
@@ -491,10 +509,8 @@ int32_t mndScheduleStream(SMnode* pMnode, STrans* pTrans, SStreamObj* pStream) {
       SStreamTask* pTask = tNewSStreamTask(pStream->uid);
       mndAddTaskToTaskSet(taskOneLevel, pTask);
 
-      pTask->dataScan = 1;
+      // source
-
+      pTask->isDataScan = 1;
-      // input
-      pTask->inputType = TASK_INPUT_TYPE__SUMBIT_BLOCK;
 
       // trigger
       pTask->triggerParam = pStream->triggerParam;

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

@@ -37,6 +37,7 @@ static SSdbRow *mndStbActionDecode(SSdbRaw *pRaw);
 static int32_t  mndStbActionInsert(SSdb *pSdb, SStbObj *pStb);
 static int32_t  mndStbActionDelete(SSdb *pSdb, SStbObj *pStb);
 static int32_t  mndStbActionUpdate(SSdb *pSdb, SStbObj *pOld, SStbObj *pNew);
+static int32_t  mndProcessTtlTimer(SRpcMsg *pReq);
 static int32_t  mndProcessCreateStbReq(SRpcMsg *pReq);
 static int32_t  mndProcessAlterStbReq(SRpcMsg *pReq);
 static int32_t  mndProcessDropStbReq(SRpcMsg *pReq);
@@ -63,6 +64,7 @@ int32_t mndInitStb(SMnode *pMnode) {
   mndSetMsgHandle(pMnode, TDMT_VND_ALTER_STB_RSP, mndTransProcessRsp);
   mndSetMsgHandle(pMnode, TDMT_VND_DROP_STB_RSP, mndTransProcessRsp);
   mndSetMsgHandle(pMnode, TDMT_MND_TABLE_META, mndProcessTableMetaReq);
+  mndSetMsgHandle(pMnode, TDMT_MND_TTL_TIMER, mndProcessTtlTimer);
   mndSetMsgHandle(pMnode, TDMT_MND_TABLE_CFG, mndProcessTableCfgReq);
 
   mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_STB, mndRetrieveStb);
@@ -799,6 +801,43 @@ int32_t mndAddStbToTrans(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SStbObj *p
   return 0;
 }
 
+static int32_t mndProcessTtlTimer(SRpcMsg *pReq) {
+  SMnode *pMnode = pReq->info.node;
+  SSdb   *pSdb = pMnode->pSdb;
+  SVgObj *pVgroup = NULL;
+  void   *pIter = NULL;
+
+  while (1) {
+    pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void **)&pVgroup);
+    if (pIter == NULL) break;
+
+    int32_t   contLen = sizeof(SMsgHead) + sizeof(int32_t);
+    SMsgHead *pHead = rpcMallocCont(contLen);
+    if (pHead == NULL) {
+      sdbCancelFetch(pSdb, pVgroup);
+      sdbRelease(pSdb, pVgroup);
+      continue;
+    }
+    pHead->contLen = htonl(contLen);
+    pHead->vgId = htonl(pVgroup->vgId);
+
+    int32_t t = taosGetTimestampSec();
+    *(int32_t *)((char *)pHead + sizeof(SMsgHead)) = htonl(t);
+
+    SRpcMsg rpcMsg = {.msgType = TDMT_VND_DROP_TTL_TABLE, .pCont = pHead, .contLen = contLen};
+    SEpSet  epSet = mndGetVgroupEpset(pMnode, pVgroup);
+    int32_t code = tmsgSendReq(&epSet, &rpcMsg);
+    if (code != 0) {
+      mError("failed to send ttl time seed, code:0x%x", code);
+    } else {
+      mDebug("send ttl time seed success, time:%d", t);
+    }
+    sdbRelease(pSdb, pVgroup);
+  }
+
+  return 0;
+}
+
 static int32_t mndProcessCreateStbReq(SRpcMsg *pReq) {
   SMnode        *pMnode = pReq->info.node;
   int32_t        code = -1;

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

@@ -56,6 +56,7 @@ static bool    mndCannotExecuteTransAction(SMnode *pMnode) { return !pMnode->dep
 
 static void    mndTransSendRpcRsp(SMnode *pMnode, STrans *pTrans);
 static int32_t mndProcessTransReq(SRpcMsg *pReq);
+static int32_t mndProcessTtl(SRpcMsg *pReq);
 static int32_t mndProcessKillTransReq(SRpcMsg *pReq);
 
 static int32_t mndRetrieveTrans(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBlock, int32_t rows);

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

@@ -1553,10 +1553,11 @@ static int32_t mndSetBalanceVgroupInfoToTrans(SMnode *pMnode, STrans *pTrans, SD
 static int32_t mndBalanceVgroupBetweenDnode(SMnode *pMnode, STrans *pTrans, SDnodeObj *pSrc, SDnodeObj *pDst) {
   void   *pIter = NULL;
   int32_t code = -1;
+  SSdb   *pSdb = pMnode->pSdb;
 
   while (1) {
     SVgObj *pVgroup = NULL;
-    pIter = sdbFetch(pMnode->pSdb, SDB_VGROUP, pIter, (void **)&pVgroup);
+    pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void **)&pVgroup);
     if (pIter == NULL) break;
 
     bool existInSrc = false;
@@ -1568,13 +1569,15 @@ static int32_t mndBalanceVgroupBetweenDnode(SMnode *pMnode, STrans *pTrans, SDno
     }
 
     if (!existInSrc || existInDst) {
-      sdbRelease(pMnode->pSdb, pVgroup);
+      sdbRelease(pSdb, pVgroup);
+      continue;
     }
 
     SDbObj *pDb = mndAcquireDb(pMnode, pVgroup->dbName);
     code = mndSetBalanceVgroupInfoToTrans(pMnode, pTrans, pDb, pVgroup, pSrc, pDst);
     mndReleaseDb(pMnode, pDb);
-    sdbRelease(pMnode->pSdb, pVgroup);
+    sdbRelease(pSdb, pVgroup);
+    sdbCancelFetch(pSdb, pIter);
     break;
   }
 
@@ -1593,15 +1596,25 @@ static int32_t mndBalanceVgroup(SMnode *pMnode, SRpcMsg *pReq, SArray *pArray) {
 
   while (1) {
     taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);
-    SDnodeObj *pSrc = taosArrayGet(pArray, 0);
+    for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {
-    SDnodeObj *pDst = taosArrayGet(pArray, taosArrayGetSize(pArray) - 1);
+      SDnodeObj *pDnode = taosArrayGet(pArray, i);
+      mDebug("dnode:%d, equivalent vnodes:%d support:%d, score:%f", pDnode->id, pDnode->numOfVnodes,
+             pDnode->numOfSupportVnodes, (float)pDnode->numOfVnodes / pDnode->numOfSupportVnodes);
+    }
+
+    SDnodeObj *pSrc = taosArrayGet(pArray, taosArrayGetSize(pArray) - 1);
+    SDnodeObj *pDst = taosArrayGet(pArray, 0);
 
     float srcScore = (float)(pSrc->numOfVnodes - 1) / pSrc->numOfSupportVnodes;
     float dstScore = (float)(pDst->numOfVnodes + 1) / pDst->numOfSupportVnodes;
-    if (srcScore + 0.0001 < dstScore) {
+    mDebug("trans:%d, after balance, src dnode:%d score:%f, dst dnode:%d score:%f", pTrans->id, pSrc->id, srcScore,
-      mDebug("trans:%d, balance vgroup from dnode:%d to dnode:%d", pTrans->id, pSrc->id, pDst->id);
+           pDst->id, dstScore);
+
+    if (srcScore > dstScore - 0.000001) {
       code = mndBalanceVgroupBetweenDnode(pMnode, pTrans, pSrc, pDst);
       if (code == 0) {
+        pSrc->numOfVnodes--;
+        pDst->numOfVnodes++;
         numOfVgroups++;
         continue;
       } else {
@@ -1635,7 +1648,13 @@ static int32_t mndProcessBalanceVgroupMsg(SRpcMsg *pReq) {
   void   *pIter = NULL;
   int64_t curMs = taosGetTimestampMs();
 
-  mDebug("start to balance vgroup");
+  SBalanceVgroupReq req = {0};
+  if (tDeserializeSBalanceVgroupReq(pReq->pCont, pReq->contLen, &req) != 0) {
+    terrno = TSDB_CODE_INVALID_MSG;
+    goto _OVER;
+  }
+
+  mInfo("start to balance vgroup");
 
   if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_BALANCE_VGROUP) != 0) goto _OVER;
 

source/dnode/snode/src/snode.c

@@ -105,13 +105,14 @@ static int32_t sndProcessTaskDeployReq(SSnode *pNode, SRpcMsg *pMsg) {
 
   ASSERT(pTask->execType != TASK_EXEC__NONE);
 
-  ASSERT(pTask->dataScan == 0);
+  ASSERT(pTask->isDataScan == 0);
   pTask->exec.executor = qCreateStreamExecTaskInfo(pTask->exec.qmsg, NULL);
   ASSERT(pTask->exec.executor);
 
   streamSetupTrigger(pTask);
 
-  qInfo("deploy stream: stream id %ld task id %d child id %d on snode", pTask->streamId, pTask->taskId, pTask->childId);
+  qInfo("deploy stream: stream id %ld task id %d child id %d on snode", pTask->streamId, pTask->taskId,
+        pTask->selfChildId);
 
   taosHashPut(pMeta->pHash, &pTask->taskId, sizeof(int32_t), &pTask, sizeof(void *));
 
@@ -198,6 +199,34 @@ static int32_t sndProcessTaskDropReq(SSnode *pNode, SRpcMsg *pMsg) {
   return code;
 }
 
+static int32_t sndProcessTaskRetrieveReq(SSnode *pNode, SRpcMsg *pMsg) {
+  SStreamMeta *pMeta = pNode->pMeta;
+
+  char              *msgStr = pMsg->pCont;
+  char              *msgBody = POINTER_SHIFT(msgStr, sizeof(SMsgHead));
+  int32_t            msgLen = pMsg->contLen - sizeof(SMsgHead);
+  SStreamRetrieveReq req;
+  SDecoder           decoder;
+  tDecoderInit(&decoder, msgBody, msgLen);
+  tDecodeStreamRetrieveReq(&decoder, &req);
+  int32_t      taskId = req.dstTaskId;
+  SStreamTask *pTask = *(SStreamTask **)taosHashGet(pMeta->pHash, &taskId, sizeof(int32_t));
+  if (atomic_load_8(&pTask->taskStatus) != TASK_STATUS__NORMAL) {
+    return 0;
+  }
+  SRpcMsg rsp = {
+      .info = pMsg->info,
+      .code = 0,
+  };
+  streamProcessRetrieveReq(pTask, &req, &rsp);
+  return 0;
+}
+
+static int32_t sndProcessTaskRetrieveRsp(SSnode *pNode, SRpcMsg *pMsg) {
+  //
+  return 0;
+}
+
 int32_t sndProcessUMsg(SSnode *pSnode, SRpcMsg *pMsg) {
   // stream deploy
   // stream stop/resume
@@ -221,10 +250,14 @@ int32_t sndProcessSMsg(SSnode *pSnode, SRpcMsg *pMsg) {
       return sndProcessTaskDispatchReq(pSnode, pMsg);
     case TDMT_STREAM_TASK_RECOVER:
       return sndProcessTaskRecoverReq(pSnode, pMsg);
+    case TDMT_STREAM_RETRIEVE:
+      return sndProcessTaskRecoverReq(pSnode, pMsg);
     case TDMT_STREAM_TASK_DISPATCH_RSP:
       return sndProcessTaskDispatchRsp(pSnode, pMsg);
     case TDMT_STREAM_TASK_RECOVER_RSP:
       return sndProcessTaskRecoverRsp(pSnode, pMsg);
+    case TDMT_STREAM_RETRIEVE_RSP:
+      return sndProcessTaskRecoverRsp(pSnode, pMsg);
     default:
       ASSERT(0);
   }

source/dnode/vnode/CMakeLists.txt

@@ -29,6 +29,7 @@ target_sources(
     # sma
     "src/sma/sma.c"
     "src/sma/smaEnv.c"
+    "src/sma/smaUtil.c"
     "src/sma/smaOpen.c"
     "src/sma/smaRollup.c"
     "src/sma/smaTimeRange.c"

source/dnode/vnode/inc/vnode.h

@@ -195,6 +195,7 @@ struct SVnodeCfg {
 typedef struct {
   TSKEY    lastKey;
   uint64_t uid;
+  uint64_t groupId;
 } STableKeyInfo;
 
 struct SMetaEntry {

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

@@ -34,7 +34,8 @@ extern "C" {
 
 typedef struct SSmaEnv       SSmaEnv;
 typedef struct SSmaStat      SSmaStat;
-typedef struct SSmaStatItem  SSmaStatItem;
+typedef struct STSmaStat     STSmaStat;
+typedef struct SRSmaStat     SRSmaStat;
 typedef struct SSmaKey       SSmaKey;
 typedef struct SRSmaInfo     SRSmaInfo;
 typedef struct SRSmaInfoItem SRSmaInfoItem;
@@ -48,23 +49,35 @@ 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_ITEM(env) ((env)->pStat->tsmaStatItem)
 
-struct SSmaStatItem {
+struct STSmaStat {
   int8_t    state;  // ETsdbSmaStat
   STSma    *pTSma;  // cache schema
   STSchema *pTSchema;
 };
 
+struct SRSmaStat {
+  SSma     *pSma;
+  void     *tmrHandle;
+  tmr_h     tmrId;
+  int8_t    tmrStat;
+  int32_t   tmrSeconds;
+  SHashObj *rsmaInfoHash;  // key: stbUid, value: SRSmaInfo;
+};
+
 struct SSmaStat {
   union {
-    SSmaStatItem tsmaStatItem;
+    STSmaStat tsmaStat;  // time-range-wise sma
-    SHashObj    *rsmaInfoHash;  // key: stbUid, value: SRSmaInfo;
+    SRSmaStat rsmaStat;  // rollup sma
   };
   T_REF_DECLARE()
 };
-#define SMA_STAT_ITEM(s)      ((s)->tsmaStatItem)
+#define SMA_TSMA_STAT(s)       (&(s)->tsmaStat)
-#define SMA_STAT_INFO_HASH(s) ((s)->rsmaInfoHash)
+#define SMA_RSMA_STAT(s)       (&(s)->rsmaStat)
+#define SMA_RSMA_INFO_HASH(s)  ((s)->rsmaStat.rsmaInfoHash)
+#define SMA_RSMA_TMR_HANDLE(s) ((s)->rsmaStat.tmrHandle)
+#define SMA_RSMA_TMR_STAT(s)   ((s)->rsmaStat.tmrStat)
+#define RSMA_INFO_HASH(r)      ((r)->rsmaInfoHash)
 
 void  tdDestroySmaEnv(SSmaEnv *pSmaEnv);
 void *tdFreeSmaEnv(SSmaEnv *pSmaEnv);
@@ -107,53 +120,51 @@ static FORCE_INLINE int32_t tdUnLockSmaEnv(SSmaEnv *pEnv) {
   return 0;
 }
 
-static FORCE_INLINE int8_t tdSmaStat(SSmaStatItem *pStatItem) {
+static FORCE_INLINE int8_t tdSmaStat(STSmaStat *pTStat) {
-  if (pStatItem) {
+  if (pTStat) {
-    return atomic_load_8(&pStatItem->state);
+    return atomic_load_8(&pTStat->state);
   }
   return TSDB_SMA_STAT_UNKNOWN;
 }
 
-static FORCE_INLINE bool tdSmaStatIsOK(SSmaStatItem *pStatItem, int8_t *state) {
+static FORCE_INLINE bool tdSmaStatIsOK(STSmaStat *pTStat, int8_t *state) {
-  if (!pStatItem) {
+  if (!pTStat) {
     return false;
   }
 
   if (state) {
-    *state = atomic_load_8(&pStatItem->state);
+    *state = atomic_load_8(&pTStat->state);
     return *state == TSDB_SMA_STAT_OK;
   }
-  return atomic_load_8(&pStatItem->state) == TSDB_SMA_STAT_OK;
+  return atomic_load_8(&pTStat->state) == TSDB_SMA_STAT_OK;
 }
 
-static FORCE_INLINE bool tdSmaStatIsExpired(SSmaStatItem *pStatItem) {
+static FORCE_INLINE bool tdSmaStatIsExpired(STSmaStat *pTStat) {
-  return pStatItem ? (atomic_load_8(&pStatItem->state) & TSDB_SMA_STAT_EXPIRED) : true;
+  return pTStat ? (atomic_load_8(&pTStat->state) & TSDB_SMA_STAT_EXPIRED) : true;
 }
 
-static FORCE_INLINE bool tdSmaStatIsDropped(SSmaStatItem *pStatItem) {
+static FORCE_INLINE bool tdSmaStatIsDropped(STSmaStat *pTStat) {
-  return pStatItem ? (atomic_load_8(&pStatItem->state) & TSDB_SMA_STAT_DROPPED) : true;
+  return pTStat ? (atomic_load_8(&pTStat->state) & TSDB_SMA_STAT_DROPPED) : true;
 }
 
-static FORCE_INLINE void tdSmaStatSetOK(SSmaStatItem *pStatItem) {
+static FORCE_INLINE void tdSmaStatSetOK(STSmaStat *pTStat) {
-  if (pStatItem) {
+  if (pTStat) {
-    atomic_store_8(&pStatItem->state, TSDB_SMA_STAT_OK);
+    atomic_store_8(&pTStat->state, TSDB_SMA_STAT_OK);
   }
 }
 
-static FORCE_INLINE void tdSmaStatSetExpired(SSmaStatItem *pStatItem) {
+static FORCE_INLINE void tdSmaStatSetExpired(STSmaStat *pTStat) {
-  if (pStatItem) {
+  if (pTStat) {
-    atomic_or_fetch_8(&pStatItem->state, TSDB_SMA_STAT_EXPIRED);
+    atomic_or_fetch_8(&pTStat->state, TSDB_SMA_STAT_EXPIRED);
   }
 }
 
-static FORCE_INLINE void tdSmaStatSetDropped(SSmaStatItem *pStatItem) {
+static FORCE_INLINE void tdSmaStatSetDropped(STSmaStat *pTStat) {
-  if (pStatItem) {
+  if (pTStat) {
-    atomic_or_fetch_8(&pStatItem->state, TSDB_SMA_STAT_DROPPED);
+    atomic_or_fetch_8(&pTStat->state, TSDB_SMA_STAT_DROPPED);
   }
 }
 
-static int32_t tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType);
-void          *tdFreeSmaStatItem(SSmaStatItem *pSmaStatItem);
 static int32_t tdDestroySmaState(SSmaStat *pSmaStat, int8_t smaType);
 void          *tdFreeSmaState(SSmaStat *pSmaStat, int8_t smaType);
 
@@ -163,6 +174,51 @@ int32_t tdProcessTSmaCreateImpl(SSma *pSma, int64_t version, const char *pMsg);
 int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char *msg);
 int32_t tdProcessTSmaGetDaysImpl(SVnodeCfg *pCfg, void *pCont, uint32_t contLen, int32_t *days);
 
+typedef struct STFInfo STFInfo;
+typedef struct STFile  STFile;
+
+struct STFInfo {
+  uint32_t magic;
+  uint32_t ftype;
+  uint32_t fver;
+  uint64_t fsize;
+};
+
+struct STFile {
+  STFInfo   info;
+  STfsFile  f;
+  TdFilePtr pFile;
+  uint8_t   state;
+};
+
+#define TD_FILE_F(tf)            (&((tf)->f))
+#define TD_FILE_PFILE(tf)        ((tf)->pFile)
+#define TD_FILE_OPENED(tf)       (TD_FILE_PFILE(tf) != NULL)
+#define TD_FILE_FULL_NAME(tf)    (TD_FILE_F(tf)->aname)
+#define TD_FILE_REL_NAME(tf)     (TD_FILE_F(tf)->rname)
+#define TD_FILE_OPENED(tf)       (TD_FILE_PFILE(tf) != NULL)
+#define TD_FILE_CLOSED(tf)       (!TD_FILE_OPENED(tf))
+#define TD_FILE_SET_CLOSED(f)    (TD_FILE_PFILE(f) = NULL)
+#define TD_FILE_STATE(tf)        ((tf)->state)
+#define TD_FILE_SET_STATE(tf, s) ((tf)->state = (s))
+#define TD_FILE_DID(tf)          (TD_FILE_F(tf)->did)
+#define TD_FILE_IS_OK(tf)        (TD_FILE_STATE(tf) == TD_FILE_STATE_OK)
+#define TD_FILE_IS_BAD(tf)       (TD_FILE_STATE(tf) == TD_FILE_STATE_BAD)
+
+int32_t tdInitTFile(STFile *pTFile, STfs *pTfs, const char *fname);
+int32_t tdCreateTFile(STFile *pTFile, STfs *pTfs, bool updateHeader, int8_t fType);
+int32_t tdOpenTFile(STFile *pTFile, int flags);
+int64_t tdReadTFile(STFile *pTFile, void *buf, int64_t nbyte);
+int64_t tdSeekTFile(STFile *pTFile, int64_t offset, int whence);
+int64_t tdWriteTFile(STFile *pTFile, void *buf, int64_t nbyte);
+int64_t tdAppendTFile(STFile *pTFile, void *buf, int64_t nbyte, int64_t *offset);
+int32_t tdRemoveTFile(STFile *pTFile);
+int32_t tdLoadTFileHeader(STFile *pTFile, STFInfo *pInfo);
+int32_t tdUpdateTFileHeader(STFile *pTFile);
+void    tdUpdateTFileMagic(STFile *pTFile, void *pCksm);
+void    tdCloseTFile(STFile *pTFile);
+void    tdGetVndFileName(int32_t vid, const char *dname, const char *fname, char *outputName);
+
 #ifdef __cplusplus
 }
 #endif

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

@@ -150,6 +150,7 @@ int64_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, SWalHead*
 
 // tqExec
 int32_t tqDataExec(STQ* pTq, STqExecHandle* pExec, SSubmitReq* pReq, SMqDataBlkRsp* pRsp, int32_t workerId);
+int32_t tqScanSnapshot(STQ* pTq, const STqExecHandle* pExec, SMqDataBlkRsp* pRsp, int32_t workerId);
 int32_t tqSendPollRsp(STQ* pTq, const SRpcMsg* pMsg, const SMqPollReq* pReq, const SMqDataBlkRsp* pRsp);
 
 // tqMeta

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

@@ -149,6 +149,8 @@ int32_t tqProcessTaskDispatchReq(STQ* pTq, SRpcMsg* pMsg);
 int32_t tqProcessTaskRecoverReq(STQ* pTq, SRpcMsg* pMsg);
 int32_t tqProcessTaskDispatchRsp(STQ* pTq, SRpcMsg* pMsg);
 int32_t tqProcessTaskRecoverRsp(STQ* pTq, SRpcMsg* pMsg);
+int32_t tqProcessTaskRetrieveReq(STQ* pTq, SRpcMsg* pMsg);
+int32_t tqProcessTaskRetrieveRsp(STQ* pTq, SRpcMsg* pMsg);
 
 SSubmitReq* tdBlockToSubmit(const SArray* pBlocks, const STSchema* pSchema, bool createTb, int64_t suid,
                             const char* stbFullName, int32_t vgId);
@@ -245,7 +247,6 @@ struct SVnode {
 
 struct STbUidStore {
   tb_uid_t  suid;
-  tb_uid_t  uid;  // TODO: just for debugging, remove when uid provided in SSDataBlock
   SArray*   tbUids;
   SHashObj* uidHash;
 };

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

@@ -219,11 +219,9 @@ _err:
 }
 
 int metaTtlSmaller(SMeta *pMeta, uint64_t ttl, SArray *uidList){
-  metaRLock(pMeta);
   TBC *    pCur;
   int ret = tdbTbcOpen(pMeta->pTtlIdx, &pCur, NULL);
   if (ret < 0) {
-    metaULock(pMeta);
     return ret;
   }
 
@@ -249,6 +247,7 @@ int metaTtlSmaller(SMeta *pMeta, uint64_t ttl, SArray *uidList){
   tdbTbcClose(pCur);
 
   tdbFree(pKey);
+
   return 0;
 }
 
@@ -265,8 +264,8 @@ struct SMCtbCursor {
 SMCtbCursor *metaOpenCtbCursor(SMeta *pMeta, tb_uid_t uid) {
   SMCtbCursor *pCtbCur = NULL;
   SCtbIdxKey   ctbIdxKey;
-  int          ret;
+  int          ret = 0;
-  int          c;
+  int          c = 0;
 
   pCtbCur = (SMCtbCursor *)taosMemoryCalloc(1, sizeof(*pCtbCur));
   if (pCtbCur == NULL) {
@@ -613,9 +612,6 @@ const void *metaGetTableTagVal(SMetaEntry *pEntry, int16_t type, STagVal *val) {
   ASSERT(pEntry->type == TSDB_CHILD_TABLE);
   STag *tag = (STag *)pEntry->ctbEntry.pTags;
   if (type == TSDB_DATA_TYPE_JSON) {
-    if (tag->nTag == 0) {
-      return NULL;
-    }
     return tag;
   }
   bool find = tTagGet(tag, val);

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

@@ -375,6 +375,7 @@ int metaTtlDropTable(SMeta *pMeta, int64_t ttl, SArray *tbUids) {
   metaWLock(pMeta);
   int ret = metaTtlSmaller(pMeta, ttl, tbUids);
   if(ret != 0){
+    metaULock(pMeta);
     return ret;
   }
   for (int i = 0; i < taosArrayGetSize(tbUids); ++i) {
@@ -400,8 +401,7 @@ static void metaBuildTtlIdxKey(STtlIdxKey *ttlKey, const SMetaEntry *pME){
 
   if (ttlDays <= 0) return;
 
-  ttlKey->dtime = ctime / 1000 + ttlDays * 24 * 60 * 60;
+  ttlKey->dtime = ctime / 1000 + ttlDays * tsTtlUnit;
-//  ttlKey->dtime = ctime / 1000 + ttlDays;
   ttlKey->uid = pME->uid;
 }
 

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

@@ -21,9 +21,10 @@ typedef struct SSmaStat SSmaStat;
 
 // declaration of static functions
 
-static int32_t  tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType);
+static int32_t  tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType, const SSma *pSma);
 static SSmaEnv *tdNewSmaEnv(const SSma *pSma, int8_t smaType, const char *path);
 static int32_t  tdInitSmaEnv(SSma *pSma, int8_t smaType, const char *path, SSmaEnv **pEnv);
+static void    *tdFreeTSmaStat(STSmaStat *pStat);
 
 // implementation
 
@@ -45,7 +46,7 @@ static SSmaEnv *tdNewSmaEnv(const SSma *pSma, int8_t smaType, const char *path)
     return NULL;
   }
 
-  if (tdInitSmaStat(&SMA_ENV_STAT(pEnv), smaType) != TSDB_CODE_SUCCESS) {
+  if (tdInitSmaStat(&SMA_ENV_STAT(pEnv), smaType, pSma) != TSDB_CODE_SUCCESS) {
     tdFreeSmaEnv(pEnv);
     return NULL;
   }
@@ -105,7 +106,7 @@ int32_t tdUnRefSmaStat(SSma *pSma, SSmaStat *pStat) {
   return 0;
 }
 
-static int32_t tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType) {
+static int32_t tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType, const SSma *pSma) {
   ASSERT(pSmaStat != NULL);
 
   if (*pSmaStat) {  // no lock
@@ -125,10 +126,23 @@ static int32_t tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType) {
     }
 
     if (smaType == TSDB_SMA_TYPE_ROLLUP) {
-      SMA_STAT_INFO_HASH(*pSmaStat) = taosHashInit(
+      SMA_RSMA_STAT(*pSmaStat)->pSma = (SSma*)pSma;
-          RSMA_TASK_INFO_HASH_SLOT, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), true, HASH_ENTRY_LOCK);
+      // init timer
+      SMA_RSMA_TMR_HANDLE(*pSmaStat) = taosTmrInit(10000, 100, 10000, "RSMA_G");
+      if (!SMA_RSMA_TMR_HANDLE(*pSmaStat)) {
+        taosMemoryFreeClear(*pSmaStat);
+        return TSDB_CODE_FAILED;
+      }
       
-      if (!SMA_STAT_INFO_HASH(*pSmaStat)) {
+      atomic_store_8(&SMA_RSMA_TMR_STAT(*pSmaStat), TASK_TRIGGER_STATUS__ACTIVE);
+
+      // init hash
+      SMA_RSMA_INFO_HASH(*pSmaStat) = taosHashInit(
+          RSMA_TASK_INFO_HASH_SLOT, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), true, HASH_ENTRY_LOCK);
+      if (!SMA_RSMA_INFO_HASH(*pSmaStat)) {
+        if (SMA_RSMA_TMR_HANDLE(*pSmaStat)) {
+          taosTmrCleanUp(SMA_RSMA_TMR_HANDLE(*pSmaStat));
+        }
         taosMemoryFreeClear(*pSmaStat);
         return TSDB_CODE_FAILED;
       }
@@ -141,16 +155,16 @@ static int32_t tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType) {
   return TSDB_CODE_SUCCESS;
 }
 
-void *tdFreeSmaStatItem(SSmaStatItem *pSmaStatItem) {
+static void *tdFreeTSmaStat(STSmaStat *pStat) {
-  if (pSmaStatItem) {
+  if (pStat) {
-    tDestroyTSma(pSmaStatItem->pTSma);
+    tDestroyTSma(pStat->pTSma);
-    taosMemoryFreeClear(pSmaStatItem->pTSma);
+    taosMemoryFreeClear(pStat->pTSma);
-    taosMemoryFreeClear(pSmaStatItem);
+    taosMemoryFreeClear(pStat);
   }
   return NULL;
 }
 
-void* tdFreeSmaState(SSmaStat *pSmaStat, int8_t smaType) {
+void *tdFreeSmaState(SSmaStat *pSmaStat, int8_t smaType) {
   tdDestroySmaState(pSmaStat, smaType);
   taosMemoryFreeClear(pSmaStat);
   return NULL;
@@ -165,16 +179,19 @@ void* tdFreeSmaState(SSmaStat *pSmaStat, int8_t smaType) {
 int32_t tdDestroySmaState(SSmaStat *pSmaStat, int8_t smaType) {
   if (pSmaStat) {
     if (smaType == TSDB_SMA_TYPE_TIME_RANGE) {
-      tdFreeSmaStatItem(&pSmaStat->tsmaStatItem);
+      tdFreeTSmaStat(&pSmaStat->tsmaStat);
     } else if (smaType == TSDB_SMA_TYPE_ROLLUP) {
+      if (SMA_RSMA_TMR_HANDLE(pSmaStat)) {
+        taosTmrCleanUp(SMA_RSMA_TMR_HANDLE(pSmaStat));
+      }
       // TODO: use taosHashSetFreeFp when taosHashSetFreeFp is ready.
-      void *infoHash = taosHashIterate(SMA_STAT_INFO_HASH(pSmaStat), NULL);
+      void *infoHash = taosHashIterate(SMA_RSMA_INFO_HASH(pSmaStat), NULL);
       while (infoHash) {
         SRSmaInfo *pInfoHash = *(SRSmaInfo **)infoHash;
         tdFreeRSmaInfo(pInfoHash);
-        infoHash = taosHashIterate(SMA_STAT_INFO_HASH(pSmaStat), infoHash);
+        infoHash = taosHashIterate(SMA_RSMA_INFO_HASH(pSmaStat), infoHash);
       }
-      taosHashCleanup(SMA_STAT_INFO_HASH(pSmaStat));
+      taosHashCleanup(SMA_RSMA_INFO_HASH(pSmaStat));
     } else {
       ASSERT(0);
     }

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

@@ -138,3 +138,16 @@ int32_t smaClose(SSma *pSma) {
   }
   return 0;
 }
+
+/**
+ * @brief rsma env restore
+ * 
+ * @param pSma 
+ * @return int32_t 
+ */
+int32_t smaRestore(SSma *pSma) {
+  if (!pSma) return 0;
+  // iterate all stables to restore the rsma env
+  
+  return TSDB_CODE_SUCCESS;
+}

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

@@ -16,35 +16,17 @@
 #include "sma.h"
 #include "tstream.h"
 
-static FORCE_INLINE int32_t tdUidStorePut(STbUidStore *pStore, tb_uid_t suid, tb_uid_t *uid);
+typedef enum { TD_QTASK_TMP_FILE = 0, TD_QTASK_CUR_FILE } TD_QTASK_FILE_T;
-static FORCE_INLINE int32_t tdUpdateTbUidListImpl(SSma *pSma, tb_uid_t *suid, SArray *tbUids);
+static const char *tdQTaskInfoFname[] = {"qtaskinfo.t", "qtaskinfo"};
-static FORCE_INLINE int32_t tdExecuteRSmaImpl(SSma *pSma, const void *pMsg, int32_t inputType, SRSmaInfoItem *rsmaItem,
-                                              tb_uid_t suid, int8_t level);
 
-#define SET_RSMA_INFO_ITEM_PARAMS(__idx, __level)                                                               \
+static int32_t tdUidStorePut(STbUidStore *pStore, tb_uid_t suid, tb_uid_t *uid);
-  if (param->qmsg[__idx]) {                                                                                     \
+static int32_t tdUpdateTbUidListImpl(SSma *pSma, tb_uid_t *suid, SArray *tbUids);
-    pRSmaInfo->items[__idx].pRsmaInfo = pRSmaInfo;                                                              \
+static int32_t tdSetRSmaInfoItemParams(SSma *pSma, SRSmaParam *param, SRSmaInfo *pRSmaInfo, SReadHandle *handle,
-    pRSmaInfo->items[__idx].taskInfo = qCreateStreamExecTaskInfo(param->qmsg[0], &handle);                      \
+                                       int8_t idx);
-    if (!pRSmaInfo->items[__idx].taskInfo) {                                                                    \
+static int32_t tdExecuteRSmaImpl(SSma *pSma, const void *pMsg, int32_t inputType, SRSmaInfoItem *rsmaItem,
-      goto _err;                                                                                                \
+                                 tb_uid_t suid, int8_t level);
-    }                                                                                                           \
+static void    tdRSmaFetchTrigger(void *param, void *tmrId);
-    pRSmaInfo->items[__idx].triggerStatus = TASK_TRIGGER_STATUS__IN_ACTIVE;                                     \
+static void    tdRSmaPersistTrigger(void *param, void *tmrId);
-    if (param->maxdelay[__idx] < 1) {                                                                           \
-      int64_t msInterval =                                                                                      \
-          convertTimeFromPrecisionToUnit(pRetention[__level].freq, pTsdbCfg->precision, TIME_UNIT_MILLISECOND); \
-      pRSmaInfo->items[__idx].maxDelay = msInterval;                                                            \
-    } else {                                                                                                    \
-      pRSmaInfo->items[__idx].maxDelay = param->maxdelay[__idx];                                                \
-    }                                                                                                           \
-    if (pRSmaInfo->items[__idx].maxDelay > TSDB_MAX_ROLLUP_MAX_DELAY) {                                         \
-      pRSmaInfo->items[__idx].maxDelay = TSDB_MAX_ROLLUP_MAX_DELAY;                                             \
-    }                                                                                                           \
-    pRSmaInfo->items[__idx].level = TSDB_RETENTION_L##__level;                                                  \
-    pRSmaInfo->items[__idx].tmrHandle = taosTmrInit(10000, 100, 10000, "RSMA");                                 \
-    if (!pRSmaInfo->items[__idx].tmrHandle) {                                                                   \
-      goto _err;                                                                                                \
-    }                                                                                                           \
-  }
 
 struct SRSmaInfoItem {
   SRSmaInfo *pRsmaInfo;
@@ -56,14 +38,6 @@ struct SRSmaInfoItem {
   int8_t     triggerStatus;  // TASK_TRIGGER_STATUS__IN_ACTIVE/TASK_TRIGGER_STATUS__ACTIVE
   int32_t    maxDelay;
 };
-
-typedef struct {
-  int64_t        suid;
-  SRSmaInfoItem *pItem;
-  SSma          *pSma;
-  STSchema      *pTSchema;
-} SRSmaTriggerParam;
-
 struct SRSmaInfo {
   STSchema     *pTSchema;
   SSma         *pSma;
@@ -81,7 +55,7 @@ static FORCE_INLINE void tdFreeTaskHandle(qTaskInfo_t *taskHandle) {
 
 void *tdFreeRSmaInfo(SRSmaInfo *pInfo) {
   if (pInfo) {
-    for (int32_t i = 0; i < TSDB_RETENTION_MAX; ++i) {
+    for (int32_t i = 0; i < TSDB_RETENTION_L2; ++i) {
       SRSmaInfoItem *pItem = &pInfo->items[i];
       if (pItem->taskInfo) {
         tdFreeTaskHandle(pItem->taskInfo);
@@ -118,7 +92,7 @@ static FORCE_INLINE int32_t tdUpdateTbUidListImpl(SSma *pSma, tb_uid_t *suid, SA
     return TSDB_CODE_FAILED;
   }
 
-  pRSmaInfo = taosHashGet(SMA_STAT_INFO_HASH(pStat), suid, sizeof(tb_uid_t));
+  pRSmaInfo = taosHashGet(SMA_RSMA_INFO_HASH(pStat), suid, sizeof(tb_uid_t));
   if (!pRSmaInfo || !(pRSmaInfo = *(SRSmaInfo **)pRSmaInfo)) {
     smaError("vgId:%d, failed to get rsma info for uid:%" PRIi64, SMA_VID(pSma), *suid);
     terrno = TSDB_CODE_RSMA_INVALID_STAT;
@@ -187,7 +161,7 @@ int32_t tdFetchTbUidList(SSma *pSma, STbUidStore **ppStore, tb_uid_t suid, tb_ui
 
   SSmaStat *pStat = SMA_ENV_STAT(pEnv);
   SHashObj *infoHash = NULL;
-  if (!pStat || !(infoHash = SMA_STAT_INFO_HASH(pStat))) {
+  if (!pStat || !(infoHash = SMA_RSMA_INFO_HASH(pStat))) {
     terrno = TSDB_CODE_RSMA_INVALID_STAT;
     return TSDB_CODE_FAILED;
   }
@@ -213,6 +187,40 @@ int32_t tdFetchTbUidList(SSma *pSma, STbUidStore **ppStore, tb_uid_t suid, tb_ui
   return TSDB_CODE_SUCCESS;
 }
 
+static int32_t tdSetRSmaInfoItemParams(SSma *pSma, SRSmaParam *param, SRSmaInfo *pRSmaInfo, SReadHandle *pReadHandle,
+                                       int8_t idx) {
+  SRetention *pRetention = SMA_RETENTION(pSma);
+  STsdbCfg   *pTsdbCfg = SMA_TSDB_CFG(pSma);
+
+  if (param->qmsg[idx]) {
+    SRSmaInfoItem *pItem = &(pRSmaInfo->items[idx]);
+    pItem->pRsmaInfo = pRSmaInfo;
+    pItem->taskInfo = qCreateStreamExecTaskInfo(param->qmsg[0], pReadHandle);
+    if (!pItem->taskInfo) {
+      goto _err;
+    }
+    pItem->triggerStatus = TASK_TRIGGER_STATUS__IN_ACTIVE;
+    if (param->maxdelay[idx] < TSDB_MIN_ROLLUP_MAX_DELAY) {
+      int64_t msInterval =
+          convertTimeFromPrecisionToUnit(pRetention[idx + 1].freq, pTsdbCfg->precision, TIME_UNIT_MILLISECOND);
+      pItem->maxDelay = (int32_t)msInterval;
+    } else {
+      pItem->maxDelay = (int32_t)param->maxdelay[idx];
+    }
+    if (pItem->maxDelay > TSDB_MAX_ROLLUP_MAX_DELAY) {
+      pItem->maxDelay = TSDB_MAX_ROLLUP_MAX_DELAY;
+    }
+    pItem->level = (idx == 0 ? TSDB_RETENTION_L1 : TSDB_RETENTION_L2);
+    pItem->tmrHandle = taosTmrInit(10000, 100, 10000, "RSMA");
+    if (!pItem->tmrHandle) {
+      goto _err;
+    }
+  }
+  return TSDB_CODE_SUCCESS;
+_err:
+  return TSDB_CODE_FAILED;
+}
+
 /**
  * @brief Check and init qTaskInfo_t, only applicable to stable with SRSmaParam.
  *
@@ -246,7 +254,7 @@ int32_t tdProcessRSmaCreate(SVnode *pVnode, SVCreateStbReq *pReq) {
   SSmaStat  *pStat = SMA_ENV_STAT(pEnv);
   SRSmaInfo *pRSmaInfo = NULL;
 
-  pRSmaInfo = taosHashGet(SMA_STAT_INFO_HASH(pStat), &pReq->suid, sizeof(tb_uid_t));
+  pRSmaInfo = taosHashGet(SMA_RSMA_INFO_HASH(pStat), &pReq->suid, sizeof(tb_uid_t));
   if (pRSmaInfo) {
     ASSERT(0);  // TODO: free original pRSmaInfo is exists abnormally
     smaWarn("vgId:%d, rsma info already exists for stb: %s, %" PRIi64, SMA_VID(pSma), pReq->name, pReq->suid);
@@ -282,14 +290,14 @@ int32_t tdProcessRSmaCreate(SVnode *pVnode, SVCreateStbReq *pReq) {
   pRSmaInfo->pSma = pSma;
   pRSmaInfo->suid = pReq->suid;
 
-  SRetention *pRetention = SMA_RETENTION(pSma);
+  if (tdSetRSmaInfoItemParams(pSma, param, pRSmaInfo, &handle, 0) < 0) {
-  STsdbCfg   *pTsdbCfg = SMA_TSDB_CFG(pSma);
+    goto _err;
+  }
+  if (tdSetRSmaInfoItemParams(pSma, param, pRSmaInfo, &handle, 1) < 0) {
+    goto _err;
+  }
 
-  SET_RSMA_INFO_ITEM_PARAMS(0, 1);
+  if (taosHashPut(SMA_RSMA_INFO_HASH(pStat), &pReq->suid, sizeof(tb_uid_t), &pRSmaInfo, sizeof(pRSmaInfo)) < 0) {
-  SET_RSMA_INFO_ITEM_PARAMS(1, 2);
-
-  if (taosHashPut(SMA_STAT_INFO_HASH(pStat), &pReq->suid, sizeof(tb_uid_t), &pRSmaInfo, sizeof(pRSmaInfo)) !=
-      TSDB_CODE_SUCCESS) {
     goto _err;
   } else {
     smaDebug("vgId:%d, register rsma info succeed for suid:%" PRIi64, SMA_VID(pSma), pReq->suid);
@@ -418,7 +426,6 @@ static int32_t tdFetchSubmitReqSuids(SSubmitReq *pMsg, STbUidStore *pStore) {
 
     if (!pBlock) break;
     tdUidStorePut(pStore, msgIter.suid, NULL);
-    pStore->uid = msgIter.uid;  // TODO: remove, just for debugging
   }
 
   if (terrno != TSDB_CODE_SUCCESS) return -1;
@@ -439,8 +446,9 @@ static int32_t tdFetchAndSubmitRSmaResult(SRSmaInfoItem *pItem, int8_t blkType)
     if (!output) {
       break;
     }
+
     if (!pResult) {
-      pResult = taosArrayInit(0, sizeof(SSDataBlock));
+      pResult = taosArrayInit(1, sizeof(SSDataBlock));
       if (!pResult) {
         terrno = TSDB_CODE_OUT_OF_MEMORY;
         return TSDB_CODE_FAILED;
@@ -451,7 +459,7 @@ static int32_t tdFetchAndSubmitRSmaResult(SRSmaInfoItem *pItem, int8_t blkType)
   }
 
   if (taosArrayGetSize(pResult) > 0) {
-#if 0
+#if 1
     char flag[10] = {0};
     snprintf(flag, 10, "level %" PRIi8, pItem->level);
     blockDebugShowData(pResult, flag);
@@ -459,14 +467,12 @@ static int32_t tdFetchAndSubmitRSmaResult(SRSmaInfoItem *pItem, int8_t blkType)
     STsdb      *sinkTsdb = (pItem->level == TSDB_RETENTION_L1 ? pSma->pRSmaTsdb1 : pSma->pRSmaTsdb2);
     SSubmitReq *pReq = NULL;
     if (buildSubmitReqFromDataBlock(&pReq, pResult, pRSmaInfo->pTSchema, SMA_VID(pSma), pRSmaInfo->suid) < 0) {
-      taosArrayDestroy(pResult);
+      goto _err;
-      return TSDB_CODE_FAILED;
     }
 
     if (pReq && tdProcessSubmitReq(sinkTsdb, INT64_MAX, pReq) < 0) {
-      taosArrayDestroy(pResult);
       taosMemoryFreeClear(pReq);
-      return TSDB_CODE_FAILED;
+      goto _err;
     }
 
     taosMemoryFreeClear(pReq);
@@ -479,7 +485,10 @@ static int32_t tdFetchAndSubmitRSmaResult(SRSmaInfoItem *pItem, int8_t blkType)
   }
 
   taosArrayDestroy(pResult);
-  return 0;
+  return TSDB_CODE_SUCCESS;
+_err:
+  taosArrayDestroy(pResult);
+  return TSDB_CODE_FAILED;
 }
 
 /**
@@ -488,13 +497,12 @@ static int32_t tdFetchAndSubmitRSmaResult(SRSmaInfoItem *pItem, int8_t blkType)
  * @param param
  * @param tmrId
  */
-static void rsmaTriggerByTimer(void *param, void *tmrId) {
+static void tdRSmaFetchTrigger(void *param, void *tmrId) {
-  // SRSmaTriggerParam *pParam = (SRSmaTriggerParam *)param;
-  // SRSmaInfoItem     *pItem = pParam->pItem;
   SRSmaInfoItem *pItem = param;
 
   if (atomic_load_8(&pItem->triggerStatus) == TASK_TRIGGER_STATUS__ACTIVE) {
-    smaTrace("level %" PRIi8 " status is active for tb suid:%" PRIi64, pItem->level, pItem->pRsmaInfo->suid);
+    smaWarn("%s:%d THREAD:%" PRIi64 " level %" PRIi8 " status is active for tb suid:%" PRIi64, __func__, __LINE__,
+            taosGetSelfPthreadId(), pItem->level, pItem->pRsmaInfo->suid);
     SSDataBlock dataBlock = {.info.type = STREAM_GET_ALL};
 
     atomic_store_8(&pItem->triggerStatus, TASK_TRIGGER_STATUS__IN_ACTIVE);
@@ -502,10 +510,11 @@ static void rsmaTriggerByTimer(void *param, void *tmrId) {
 
     tdFetchAndSubmitRSmaResult(pItem, STREAM_DATA_TYPE_SSDATA_BLOCK);
   } else {
-    smaTrace("level %" PRIi8 " status is inactive for tb suid:%" PRIi64, pItem->level, pItem->pRsmaInfo->suid);
+    smaWarn("%s:%d THREAD:%" PRIi64 " level %" PRIi8 " status is inactive for tb suid:%" PRIi64, __func__, __LINE__,
+            taosGetSelfPthreadId(), pItem->level, pItem->pRsmaInfo->suid);
   }
 
-  // taosTmrReset(rsmaTriggerByTimer, pItem->maxDelay, pItem, pItem->tmrHandle, &pItem->tmrId);
+  // taosTmrReset(tdRSmaFetchTrigger, pItem->maxDelay, pItem, pItem->tmrHandle, &pItem->tmrId);
 }
 
 static FORCE_INLINE int32_t tdExecuteRSmaImpl(SSma *pSma, const void *pMsg, int32_t inputType, SRSmaInfoItem *pItem,
@@ -518,16 +527,20 @@ static FORCE_INLINE int32_t tdExecuteRSmaImpl(SSma *pSma, const void *pMsg, int3
   smaDebug("vgId:%d, execute rsma %" PRIi8 " task for qTaskInfo:%p suid:%" PRIu64, SMA_VID(pSma), level,
            pItem->taskInfo, suid);
 
-  // inputType = STREAM_DATA_TYPE_SUBMIT_BLOCK(1)
+  if (qSetStreamInput(pItem->taskInfo, pMsg, inputType, true) < 0) {  // STREAM_DATA_TYPE_SUBMIT_BLOCK
-  if (qSetStreamInput(pItem->taskInfo, pMsg, inputType, true) < 0) {
     smaError("vgId:%d, rsma % " PRIi8 " qSetStreamInput failed since %s", SMA_VID(pSma), level, tstrerror(terrno));
     return TSDB_CODE_FAILED;
   }
 
-  // SRSmaTriggerParam triggerParam = {.suid = suid, .pItem = pItem, .pSma = pSma, .pTSchema = pTSchema};
   tdFetchAndSubmitRSmaResult(pItem, STREAM_DATA_TYPE_SUBMIT_BLOCK);
   atomic_store_8(&pItem->triggerStatus, TASK_TRIGGER_STATUS__ACTIVE);
-  taosTmrReset(rsmaTriggerByTimer, pItem->maxDelay, pItem, pItem->tmrHandle, &pItem->tmrId);
+  smaWarn("%s:%d THREAD:%" PRIi64 " process rsma insert", __func__, __LINE__, taosGetSelfPthreadId());
+
+  SSmaEnv   *pEnv = SMA_RSMA_ENV(pSma);
+  SRSmaStat *pStat = SMA_RSMA_STAT(pEnv->pStat);
+
+  taosTmrStart(tdRSmaPersistTrigger, 5000, pStat, pStat->tmrHandle);
+  taosTmrReset(tdRSmaFetchTrigger, pItem->maxDelay, pItem, pItem->tmrHandle, &pItem->tmrId);
 
   return TSDB_CODE_SUCCESS;
 }
@@ -542,7 +555,7 @@ static int32_t tdExecuteRSma(SSma *pSma, const void *pMsg, int32_t inputType, tb
   SSmaStat  *pStat = SMA_ENV_STAT(pEnv);
   SRSmaInfo *pRSmaInfo = NULL;
 
-  pRSmaInfo = taosHashGet(SMA_STAT_INFO_HASH(pStat), &suid, sizeof(tb_uid_t));
+  pRSmaInfo = taosHashGet(SMA_RSMA_INFO_HASH(pStat), &suid, sizeof(tb_uid_t));
 
   if (!pRSmaInfo || !(pRSmaInfo = *(SRSmaInfo **)pRSmaInfo)) {
     smaDebug("vgId:%d, return as no rsma info for suid:%" PRIu64, SMA_VID(pSma), suid);
@@ -594,3 +607,106 @@ int32_t tdProcessRSmaSubmit(SSma *pSma, void *pMsg, int32_t inputType) {
   }
   return TSDB_CODE_SUCCESS;
 }
+
+void tdRSmaQTaskGetFName(int32_t vid, int8_t ftype, char* outputName) {
+  tdGetVndFileName(vid, "rsma", tdQTaskInfoFname[ftype], outputName);
+}
+
+static void *tdRSmaPersistExec(void *param) {
+  setThreadName("rsma-task-persist");
+  SRSmaStat *pRSmaStat = param;
+  SSma      *pSma = pRSmaStat->pSma;
+  STfs      *pTfs = pSma->pVnode->pTfs;
+  int64_t    toffset = 0;
+
+  void *infoHash = taosHashIterate(RSMA_INFO_HASH(pRSmaStat), NULL);
+  if (!infoHash) {
+    goto _end;
+  }
+
+  STFile  tFile = {0};
+  int32_t vid = 2;
+  char    qTaskInfoFName[TSDB_FILENAME_LEN];
+  tdRSmaQTaskGetFName(vid, TD_QTASK_TMP_FILE, qTaskInfoFName);
+  tdInitTFile(&tFile, pTfs, qTaskInfoFName);
+  tdCreateTFile(&tFile, pTfs, true, -1);
+
+  while (infoHash) {
+    SRSmaInfo *pRSmaInfo = *(SRSmaInfo **)infoHash;
+    char      *pOutput = NULL;
+    int32_t    len = 0;
+    if (qSerializeTaskStatus(pRSmaInfo->items[0].taskInfo, &pOutput, &len) < 0) {
+      smaError("serialize rsma task for table %" PRIi64 " failed since %s", pRSmaInfo->items[0].pRsmaInfo->suid,
+               terrstr(terrno));
+    } else {
+      smaWarn("serialize rsma task for table %" PRIi64 " success and len is %d", pRSmaInfo->items[0].pRsmaInfo->suid,
+              len);
+    }
+    tdAppendTFile(&tFile, &len, sizeof(len), &toffset);
+    tdAppendTFile(&tFile, pOutput, len, &toffset);
+
+    taosMemoryFree(pOutput);
+    infoHash = taosHashIterate(RSMA_INFO_HASH(pRSmaStat), infoHash);
+  }
+_end:
+
+  if (tdUpdateTFileHeader(&tFile) < 0) {
+    smaError("vgId:%d, failed to update tfile %s header since %s", vid, TD_FILE_FULL_NAME(&tFile), tstrerror(terrno));
+    tdCloseTFile(&tFile);
+    tdRemoveTFile(&tFile);
+    return NULL;
+  }
+  
+  tdCloseTFile(&tFile);
+
+  char newFName[TSDB_FILENAME_LEN];
+  strncpy(newFName, TD_FILE_FULL_NAME(&tFile), TSDB_FILENAME_LEN);
+  char *pos = strstr(newFName, tdQTaskInfoFname[TD_QTASK_TMP_FILE]);
+  strncpy(pos, tdQTaskInfoFname[TD_QTASK_CUR_FILE], TSDB_FILENAME_LEN - POINTER_DISTANCE(pos, newFName));
+  taosRenameFile(TD_FILE_FULL_NAME(&tFile), newFName);
+
+  atomic_store_8(&pRSmaStat->tmrStat, TASK_TRIGGER_STATUS__ACTIVE);
+  return NULL;
+_err:
+  atomic_store_8(&pRSmaStat->tmrStat, TASK_TRIGGER_STATUS__ACTIVE);
+  // remove the .tmp file
+  return NULL;
+}
+
+static void tdRSmaPersistTask(SRSmaStat *pRSmaStat) {
+  smaWarn("%s:%d entry ", __func__, __LINE__);
+  TdThread     threadId;
+  TdThreadAttr thAttr;
+  taosThreadAttrInit(&thAttr);
+  taosThreadAttrSetDetachState(&thAttr, PTHREAD_CREATE_DETACHED);
+
+  if (taosThreadCreate(&threadId, &thAttr, tdRSmaPersistExec, pRSmaStat) != 0) {
+    smaError("failed to create thread to persist rsma qTaskInfo since %s", strerror(errno));
+  }
+
+  taosThreadAttrDestroy(&thAttr);
+  smaWarn("%s:%d end ", __func__, __LINE__);
+}
+
+/**
+ * @brief trigger to persist rsma qTaskInfo
+ *
+ * @param param
+ * @param tmrId
+ */
+static void tdRSmaPersistTrigger(void *param, void *tmrId) {
+  SRSmaStat *pRSmaStat = param;
+
+  if (atomic_load_8(&pRSmaStat->tmrStat) == TASK_TRIGGER_STATUS__ACTIVE) {
+    smaWarn("%s:%d THREAD:%" PRIi64 " rsma persistence start since active", __func__, __LINE__, taosGetSelfPthreadId());
+    atomic_store_8(&pRSmaStat->tmrStat, TASK_TRIGGER_STATUS__IN_ACTIVE);
+
+    // execution
+    tdRSmaPersistTask(pRSmaStat);
+  } else {
+    smaWarn("%s:%d THREAD:%" PRIi64 " rsma persistence not start since inactive", __func__, __LINE__,
+            taosGetSelfPthreadId());
+  }
+
+  taosTmrReset(tdRSmaPersistTrigger, 3600000, pRSmaStat, pRSmaStat->tmrHandle, &pRSmaStat->tmrId);
+}

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

@@ -129,7 +129,7 @@ int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char *msg) {
 
   SSmaEnv      *pEnv = SMA_TSMA_ENV(pSma);
   SSmaStat     *pStat = NULL;
-  SSmaStatItem *pItem = NULL;
+  STSmaStat    *pItem = NULL;
 
   if (!pEnv || !(pStat = SMA_ENV_STAT(pEnv))) {
     terrno = TSDB_CODE_TSMA_INVALID_STAT;
@@ -137,7 +137,7 @@ int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char *msg) {
   }
 
   tdRefSmaStat(pSma, pStat);
-  pItem = &pStat->tsmaStatItem;
+  pItem = &pStat->tsmaStat;
 
   ASSERT(pItem);
 

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

@@ -0,0 +1,238 @@
+/*
+ * 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"
+
+#define TD_FILE_HEAD_SIZE 512
+
+#define TD_FILE_STATE_OK  0
+#define TD_FILE_STATE_BAD 1
+
+#define TD_FILE_INIT_MAGIC 0xFFFFFFFF
+
+
+static int32_t tdEncodeTFInfo(void **buf, STFInfo *pInfo);
+static void   *tdDecodeTFInfo(void *buf, STFInfo *pInfo);
+
+static int32_t tdEncodeTFInfo(void **buf, STFInfo *pInfo) {
+  int32_t tlen = 0;
+
+  tlen += taosEncodeFixedU32(buf, pInfo->magic);
+  tlen += taosEncodeFixedU32(buf, pInfo->ftype);
+  tlen += taosEncodeFixedU32(buf, pInfo->fver);
+  tlen += taosEncodeFixedU64(buf, pInfo->fsize);
+
+  return tlen;
+}
+
+static void *tdDecodeTFInfo(void *buf, STFInfo *pInfo) {
+  buf = taosDecodeFixedU32(buf, &(pInfo->magic));
+  buf = taosDecodeFixedU32(buf, &(pInfo->ftype));
+  buf = taosDecodeFixedU32(buf, &(pInfo->fver));
+  buf = taosDecodeFixedU64(buf, &(pInfo->fsize));
+  return buf;
+}
+
+int64_t tdWriteTFile(STFile *pTFile, void *buf, int64_t nbyte) {
+  ASSERT(TD_FILE_OPENED(pTFile));
+
+  int64_t nwrite = taosWriteFile(pTFile->pFile, buf, nbyte);
+  if (nwrite < nbyte) {
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    return -1;
+  }
+
+  return nwrite;
+}
+
+int64_t tdSeekTFile(STFile *pTFile, int64_t offset, int whence) {
+  ASSERT(TD_FILE_OPENED(pTFile));
+
+  int64_t loffset = taosLSeekFile(TD_FILE_PFILE(pTFile), offset, whence);
+  if (loffset < 0) {
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    return -1;
+  }
+
+  return loffset;
+}
+
+int64_t tdReadTFile(STFile *pTFile, void *buf, int64_t nbyte) {
+  ASSERT(TD_FILE_OPENED(pTFile));
+
+  int64_t nread = taosReadFile(pTFile->pFile, buf, nbyte);
+  if (nread < 0) {
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    return -1;
+  }
+
+  return nread;
+}
+
+int32_t tdUpdateTFileHeader(STFile *pTFile) {
+  char buf[TD_FILE_HEAD_SIZE] = "\0";
+
+  if (tdSeekTFile(pTFile, 0, SEEK_SET) < 0) {
+    return -1;
+  }
+
+  void *ptr = buf;
+  tdEncodeTFInfo(&ptr, &(pTFile->info));
+
+  taosCalcChecksumAppend(0, (uint8_t *)buf, TD_FILE_HEAD_SIZE);
+  if (tdWriteTFile(pTFile, buf, TD_FILE_HEAD_SIZE) < 0) {
+    return -1;
+  }
+
+  return 0;
+}
+
+int32_t tdLoadTFileHeader(STFile *pTFile, STFInfo *pInfo) {
+  char     buf[TD_FILE_HEAD_SIZE] = "\0";
+  uint32_t _version;
+
+  ASSERT(TD_FILE_OPENED(pTFile));
+
+  if (tdSeekTFile(pTFile, 0, SEEK_SET) < 0) {
+    return -1;
+  }
+
+  if (tdReadTFile(pTFile, buf, TD_FILE_HEAD_SIZE) < 0) {
+    return -1;
+  }
+
+  if (!taosCheckChecksumWhole((uint8_t *)buf, TD_FILE_HEAD_SIZE)) {
+    terrno = TSDB_CODE_FILE_CORRUPTED;
+    return -1;
+  }
+
+  void *pBuf = buf;
+  pBuf = tdDecodeTFInfo(pBuf, pInfo);
+  return 0;
+}
+
+void tdUpdateTFileMagic(STFile *pTFile, void *pCksm) {
+  pTFile->info.magic = taosCalcChecksum(pTFile->info.magic, (uint8_t *)(pCksm), sizeof(TSCKSUM));
+}
+
+int64_t tdAppendTFile(STFile *pTFile, void *buf, int64_t nbyte, int64_t *offset) {
+  ASSERT(TD_FILE_OPENED(pTFile));
+
+  int64_t toffset;
+
+  if ((toffset = tdSeekTFile(pTFile, 0, SEEK_END)) < 0) {
+    return -1;
+  }
+
+  ASSERT(pTFile->info.fsize == toffset);
+
+  if (offset) {
+    *offset = toffset;
+  }
+
+  if (tdWriteTFile(pTFile, buf, nbyte) < 0) {
+    return -1;
+  }
+
+  pTFile->info.fsize += nbyte;
+
+  return nbyte;
+}
+
+int32_t tdOpenTFile(STFile *pTFile, int flags) {
+  ASSERT(!TD_FILE_OPENED(pTFile));
+
+  pTFile->pFile = taosOpenFile(TD_FILE_FULL_NAME(pTFile), flags);
+  if (pTFile->pFile == NULL) {
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    return -1;
+  }
+
+  return 0;
+}
+
+void tdCloseTFile(STFile *pTFile) {
+  if (TD_FILE_OPENED(pTFile)) {
+    taosCloseFile(&pTFile->pFile);
+    TD_FILE_SET_CLOSED(pTFile);
+  }
+}
+
+void tdGetVndFileName(int32_t vid, const char *dname, const char *fname, char *outputName) {
+  snprintf(outputName, TSDB_FILENAME_LEN, "vnode/vnode%d/%s/%s", vid, dname, fname);
+}
+
+int32_t tdInitTFile(STFile *pTFile, STfs *pTfs, const char *fname) {
+  char    fullname[TSDB_FILENAME_LEN];
+  SDiskID did = {0};
+
+  TD_FILE_SET_STATE(pTFile, TD_FILE_STATE_OK);
+  TD_FILE_SET_CLOSED(pTFile);
+
+  memset(&(pTFile->info), 0, sizeof(pTFile->info));
+  pTFile->info.magic = TD_FILE_INIT_MAGIC;
+
+  if (tfsAllocDisk(pTfs, 0, &did) < 0) {
+    terrno = TSDB_CODE_NO_AVAIL_DISK;
+    return -1;
+  }
+
+  tfsInitFile(pTfs, &(pTFile->f), did, fname);
+
+  return 0;
+}
+
+int32_t tdCreateTFile(STFile *pTFile, STfs *pTfs, bool updateHeader, int8_t fType) {
+  ASSERT(pTFile->info.fsize == 0 && pTFile->info.magic == TD_FILE_INIT_MAGIC);
+
+  pTFile->pFile = taosOpenFile(TD_FILE_FULL_NAME(pTFile), TD_FILE_CREATE | TD_FILE_WRITE | TD_FILE_TRUNC);
+  if (pTFile->pFile == NULL) {
+    if (errno == ENOENT) {
+      // Try to create directory recursively
+      char *s = strdup(TD_FILE_REL_NAME(pTFile));
+      if (tfsMkdirRecurAt(pTfs, taosDirName(s), TD_FILE_DID(pTFile)) < 0) {
+        taosMemoryFreeClear(s);
+        return -1;
+      }
+      taosMemoryFreeClear(s);
+
+      pTFile->pFile = taosOpenFile(TD_FILE_FULL_NAME(pTFile), TD_FILE_CREATE | TD_FILE_WRITE | TD_FILE_TRUNC);
+      if (pTFile->pFile == NULL) {
+        terrno = TAOS_SYSTEM_ERROR(errno);
+        return -1;
+      }
+    } else {
+      terrno = TAOS_SYSTEM_ERROR(errno);
+      return -1;
+    }
+  }
+
+  if (!updateHeader) {
+    return 0;
+  }
+
+  pTFile->info.fsize += TD_FILE_HEAD_SIZE;
+  pTFile->info.fver = 0;
+
+  if (tdUpdateTFileHeader(pTFile) < 0) {
+    tdCloseTFile(pTFile);
+    tdRemoveTFile(pTFile);
+    return -1;
+  }
+
+  return 0;
+}
+
+int32_t tdRemoveTFile(STFile *pTFile) { return tfsRemoveFile(TD_FILE_F(pTFile)); }

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

@@ -227,19 +227,16 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
     consumerEpoch = atomic_val_compare_exchange_32(&pHandle->epoch, consumerEpoch, reqEpoch);
   }
 
-  SWalHead* pHeadWithCkSum = taosMemoryMalloc(sizeof(SWalHead) + 2048);
-  if (pHeadWithCkSum == NULL) {
-    return -1;
-  }
-
-  walSetReaderCapacity(pHandle->pWalReader, 2048);
-
   SMqDataBlkRsp rsp = {0};
   rsp.reqOffset = pReq->currentOffset;
 
   rsp.blockData = taosArrayInit(0, sizeof(void*));
   rsp.blockDataLen = taosArrayInit(0, sizeof(int32_t));
 
+  if (rsp.blockData == NULL || rsp.blockDataLen == NULL) {
+    return -1;
+  }
+
   rsp.withTbName = pReq->withTbName;
   if (rsp.withTbName) {
     rsp.blockTbName = taosArrayInit(0, sizeof(void*));
@@ -253,6 +250,32 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
     rsp.blockSchema = taosArrayInit(0, sizeof(void*));
   }
 
+#if 1
+  if (pReq->useSnapshot) {
+    // TODO set ver into snapshot
+    int64_t lastVer = walGetCommittedVer(pTq->pWal);
+    if (rsp.reqOffset < lastVer) {
+      tqInfo("retrieve using snapshot req offset %ld last ver %ld", rsp.reqOffset, lastVer);
+      tqScanSnapshot(pTq, &pHandle->execHandle, &rsp, workerId);
+
+      if (rsp.blockNum != 0) {
+        rsp.withTbName = false;
+        rsp.rspOffset = lastVer;
+        tqInfo("direct send by snapshot req offset %ld rsp offset %ld", rsp.reqOffset, rsp.rspOffset);
+        fetchOffset = lastVer;
+        goto SEND_RSP;
+      }
+    }
+  }
+#endif
+
+  SWalHead* pHeadWithCkSum = taosMemoryMalloc(sizeof(SWalHead) + 2048);
+  if (pHeadWithCkSum == NULL) {
+    return -1;
+  }
+
+  walSetReaderCapacity(pHandle->pWalReader, 2048);
+
   while (1) {
     consumerEpoch = atomic_load_32(&pHandle->epoch);
     if (consumerEpoch > reqEpoch) {
@@ -283,7 +306,7 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
              pHead->msgType == TDMT_VND_DROP_STB || pHead->msgType == TDMT_VND_CREATE_TABLE ||
              pHead->msgType == TDMT_VND_ALTER_TABLE || pHead->msgType == TDMT_VND_DROP_TABLE ||
              pHead->msgType == TDMT_VND_DROP_TTL_TABLE);
-      // return
+      tqInfo("fetch meta msg, ver: %ld, type: %d", pHead->version, pHead->msgType);
       SMqMetaRsp metaRsp = {0};
       metaRsp.reqOffset = pReq->currentOffset;
       metaRsp.rspOffset = fetchOffset;
@@ -292,6 +315,7 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
       metaRsp.metaRsp = pHead->body;
       if (tqSendMetaPollRsp(pTq, pMsg, pReq, &metaRsp) < 0) {
         code = -1;
+        goto OVER;
       }
       code = 0;
       goto OVER;
@@ -308,6 +332,7 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
 
   taosMemoryFree(pHeadWithCkSum);
 
+SEND_RSP:
   ASSERT(taosArrayGetSize(rsp.blockData) == rsp.blockNum);
   ASSERT(taosArrayGetSize(rsp.blockDataLen) == rsp.blockNum);
   if (rsp.withSchema) {
@@ -364,6 +389,7 @@ int32_t tqProcessVgChangeReq(STQ* pTq, char* msg, int32_t msgLen) {
     pHandle->epoch = -1;
 
     pHandle->execHandle.subType = req.subType;
+    pHandle->fetchMeta = req.withMeta;
 
     pHandle->pWalReader = walOpenReadHandle(pTq->pVnode->pWal);
     for (int32_t i = 0; i < 5; i++) {
@@ -376,6 +402,8 @@ int32_t tqProcessVgChangeReq(STQ* pTq, char* msg, int32_t msgLen) {
         SReadHandle handle = {
             .reader = pHandle->execHandle.pExecReader[i],
             .meta = pTq->pVnode->pMeta,
+            .vnode = pTq->pVnode,
+            .initTsdbReader = 1,
         };
         pHandle->execHandle.execCol.task[i] = qCreateStreamExecTaskInfo(pHandle->execHandle.execCol.qmsg, &handle);
         ASSERT(pHandle->execHandle.execCol.task[i]);
@@ -426,6 +454,7 @@ int32_t tqProcessTaskDeployReq(STQ* pTq, char* msg, int32_t msgLen) {
     ASSERT(0);
   }
   tDecoderClear(&decoder);
+  ASSERT(pTask->isDataScan == 0 || pTask->isDataScan == 1);
 
   pTask->execStatus = TASK_EXEC_STATUS__IDLE;
 
@@ -441,12 +470,13 @@ int32_t tqProcessTaskDeployReq(STQ* pTq, char* msg, int32_t msgLen) {
   // exec
   if (pTask->execType != TASK_EXEC__NONE) {
     // expand runners
-    if (pTask->dataScan) {
+    if (pTask->isDataScan) {
       STqReadHandle* pStreamReader = tqInitSubmitMsgScanner(pTq->pVnode->pMeta);
       SReadHandle    handle = {
              .reader = pStreamReader,
              .meta = pTq->pVnode->pMeta,
              .vnode = pTq->pVnode,
+             .initTsdbReader = 1,
       };
       /*pTask->exec.inputHandle = pStreamReader;*/
       pTask->exec.executor = qCreateStreamExecTaskInfo(pTask->exec.qmsg, &handle);
@@ -476,7 +506,7 @@ int32_t tqProcessTaskDeployReq(STQ* pTq, char* msg, int32_t msgLen) {
 
   streamSetupTrigger(pTask);
 
-  tqInfo("deploy stream task id %d child id %d on vg %d", pTask->taskId, pTask->childId, TD_VID(pTq->pVnode));
+  tqInfo("deploy stream task id %d child id %d on vg %d", pTask->taskId, pTask->selfChildId, TD_VID(pTq->pVnode));
 
   taosHashPut(pTq->pStreamTasks, &pTask->taskId, sizeof(int32_t), &pTask, sizeof(void*));
 
@@ -505,7 +535,7 @@ int32_t tqProcessStreamTrigger(STQ* pTq, SSubmitReq* pReq) {
     if (atomic_load_8(&pTask->taskStatus) == TASK_STATUS__DROPPING) {
       continue;
     }
-    if (pTask->inputType != STREAM_INPUT__DATA_SUBMIT) continue;
+    if (!pTask->isDataScan) continue;
 
     if (!failed) {
       if (streamTaskInput(pTask, (SStreamQueueItem*)pSubmit) < 0) {
@@ -616,3 +646,29 @@ int32_t tqProcessTaskDropReq(STQ* pTq, char* msg, int32_t msgLen) {
   return code;
 #endif
 }
+
+int32_t tqProcessTaskRetrieveReq(STQ* pTq, SRpcMsg* pMsg) {
+  char*              msgStr = pMsg->pCont;
+  char*              msgBody = POINTER_SHIFT(msgStr, sizeof(SMsgHead));
+  int32_t            msgLen = pMsg->contLen - sizeof(SMsgHead);
+  SStreamRetrieveReq req;
+  SDecoder           decoder;
+  tDecoderInit(&decoder, msgBody, msgLen);
+  tDecodeStreamRetrieveReq(&decoder, &req);
+  int32_t      taskId = req.dstTaskId;
+  SStreamTask* pTask = *(SStreamTask**)taosHashGet(pTq->pStreamTasks, &taskId, sizeof(int32_t));
+  if (atomic_load_8(&pTask->taskStatus) != TASK_STATUS__NORMAL) {
+    return 0;
+  }
+  SRpcMsg rsp = {
+      .info = pMsg->info,
+      .code = 0,
+  };
+  streamProcessRetrieveReq(pTask, &req, &rsp);
+  return 0;
+}
+
+int32_t tqProcessTaskRetrieveRsp(STQ* pTq, SRpcMsg* pMsg) {
+  //
+  return 0;
+}

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

@@ -60,6 +60,30 @@ static int32_t tqAddTbNameToRsp(const STQ* pTq, const STqExecHandle* pExec, SMqD
   return 0;
 }
 
+int32_t tqScanSnapshot(STQ* pTq, const STqExecHandle* pExec, SMqDataBlkRsp* pRsp, int32_t workerId) {
+  ASSERT(pExec->subType == TOPIC_SUB_TYPE__COLUMN);
+  qTaskInfo_t task = pExec->execCol.task[workerId];
+  if (qStreamScanSnapshot(task) < 0) {
+    ASSERT(0);
+  }
+  while (1) {
+    SSDataBlock* pDataBlock = NULL;
+    uint64_t     ts = 0;
+    if (qExecTask(task, &pDataBlock, &ts) < 0) {
+      ASSERT(0);
+    }
+    if (pDataBlock == NULL) break;
+
+    ASSERT(pDataBlock->info.rows != 0);
+    ASSERT(pDataBlock->info.numOfCols != 0);
+
+    tqAddBlockDataToRsp(pDataBlock, pRsp);
+    pRsp->blockNum++;
+  }
+
+  return 0;
+}
+
 int32_t tqDataExec(STQ* pTq, STqExecHandle* pExec, SSubmitReq* pReq, SMqDataBlkRsp* pRsp, int32_t workerId) {
   if (pExec->subType == TOPIC_SUB_TYPE__COLUMN) {
     qTaskInfo_t task = pExec->execCol.task[workerId];

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

@@ -345,8 +345,8 @@ int32_t tqUpdateTbUidList(STQ* pTq, const SArray* tbUidList, bool isAdd) {
   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) {
+    if (pTask->isDataScan) {
       int32_t code = qUpdateQualifiedTableId(pTask->exec.executor, tbUidList, isAdd);
       ASSERT(code == 0);
     }

source/dnode/vnode/src/tsdb/tsdbRead.c

@@ -2845,7 +2845,7 @@ int32_t tsdbGetAllTableList(SMeta* pMeta, uint64_t uid, SArray* list) {
       break;
     }
 
-    STableKeyInfo info = {.lastKey = TSKEY_INITIAL_VAL, uid = id};
+    STableKeyInfo info = {.lastKey = TSKEY_INITIAL_VAL, uid = id, .groupId = 0};
     taosArrayPush(list, &info);
   }
 

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

@@ -265,10 +265,14 @@ int32_t vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg, SQueueInfo *pInfo) {
       return tqProcessTaskDispatchReq(pVnode->pTq, pMsg);
     case TDMT_STREAM_TASK_RECOVER:
       return tqProcessTaskRecoverReq(pVnode->pTq, pMsg);
+    case TDMT_STREAM_RETRIEVE:
+      return tqProcessTaskRetrieveReq(pVnode->pTq, pMsg);
     case TDMT_STREAM_TASK_DISPATCH_RSP:
       return tqProcessTaskDispatchRsp(pVnode->pTq, pMsg);
     case TDMT_STREAM_TASK_RECOVER_RSP:
       return tqProcessTaskRecoverRsp(pVnode->pTq, pMsg);
+    case TDMT_STREAM_RETRIEVE_RSP:
+      return tqProcessTaskRetrieveRsp(pVnode->pTq, pMsg);
     default:
       vError("unknown msg type:%d in fetch queue", pMsg->msgType);
       return TSDB_CODE_VND_APP_ERROR;
@@ -311,7 +315,7 @@ static int32_t vnodeProcessDropTtlTbReq(SVnode *pVnode, int64_t version, void *p
   if (tbUids == NULL) return TSDB_CODE_OUT_OF_MEMORY;
 
   int32_t t = ntohl(*(int32_t *)pReq);
-  vError("rec ttl time:%d", t);
+  vDebug("vgId:%d, recv ttl msg, time:%d", pVnode->config.vgId, t);
   int32_t ret = metaTtlDropTable(pVnode->pMeta, t, tbUids);
   if (ret != 0) {
     goto end;

source/libs/catalog/inc/catalogInt.h

@@ -65,6 +65,7 @@ enum {
 
 typedef enum {
   CTG_TASK_GET_QNODE = 0,
+  CTG_TASK_GET_DNODE,
   CTG_TASK_GET_DB_VGROUP,
   CTG_TASK_GET_DB_CFG,
   CTG_TASK_GET_DB_INFO,
@@ -216,6 +217,7 @@ typedef struct SCtgJob {
   int32_t           dbVgNum;
   int32_t           udfNum;
   int32_t           qnodeNum;
+  int32_t           dnodeNum;
   int32_t           dbCfgNum;
   int32_t           indexNum;
   int32_t           userNum;
@@ -565,6 +567,7 @@ int32_t ctgGetTbHashVgroupFromCache(SCatalog *pCtg, const SName *pTableName, SVg
 int32_t ctgProcessRspMsg(void* out, int32_t reqType, char* msg, int32_t msgSize, int32_t rspCode, char* target);
 int32_t ctgGetDBVgInfoFromMnode(SCatalog* pCtg, SRequestConnInfo *pConn, SBuildUseDBInput *input, SUseDbOutput *out, SCtgTask* pTask);
 int32_t ctgGetQnodeListFromMnode(SCatalog* pCtg, SRequestConnInfo *pConn, SArray *out, SCtgTask* pTask);
+int32_t ctgGetDnodeListFromMnode(SCatalog* pCtg, SRequestConnInfo *pConn, SArray **out, SCtgTask* pTask);
 int32_t ctgGetDBCfgFromMnode(SCatalog* pCtg, SRequestConnInfo *pConn, const char *dbFName, SDbCfgInfo *out, SCtgTask* pTask);
 int32_t ctgGetIndexInfoFromMnode(SCatalog* pCtg, SRequestConnInfo *pConn, const char *indexName, SIndexInfo *out, SCtgTask* pTask);
 int32_t ctgGetTbIndexFromMnode(SCatalog* pCtg, SRequestConnInfo *pConn, SName *name, STableIndex* out, SCtgTask* pTask);

source/libs/catalog/src/catalog.c

@@ -1099,8 +1099,19 @@ _return:
   CTG_API_LEAVE(TSDB_CODE_SUCCESS);
 }
 
-int32_t catalogGetDnodeList(SCatalog* pCatalog, SRequestConnInfo* pConn, SArray** pDnodeList) {
+int32_t catalogGetDnodeList(SCatalog* pCtg, SRequestConnInfo* pConn, SArray** pDnodeList) {
-  return TSDB_CODE_CTG_INVALID_INPUT;
+  CTG_API_ENTER();
+
+  int32_t code = 0;
+  if (NULL == pCtg || NULL == pConn || NULL == pDnodeList) {
+    CTG_API_LEAVE(TSDB_CODE_CTG_INVALID_INPUT);
+  }
+
+  CTG_ERR_JRET(ctgGetDnodeListFromMnode(pCtg, pConn, pDnodeList, NULL));
+
+_return:
+
+  CTG_API_LEAVE(TSDB_CODE_SUCCESS);
 }
 
 int32_t catalogGetExpiredSTables(SCatalog* pCtg, SSTableVersion **stables, uint32_t *num) {

source/libs/catalog/src/ctgAsync.c

@@ -168,6 +168,21 @@ int32_t ctgInitGetQnodeTask(SCtgJob *pJob, int32_t taskIdx, void* param) {
   return TSDB_CODE_SUCCESS;
 }
 
+int32_t ctgInitGetDnodeTask(SCtgJob *pJob, int32_t taskIdx, void* param) {
+  SCtgTask task = {0};
+
+  task.type = CTG_TASK_GET_DNODE;
+  task.taskId = taskIdx;
+  task.pJob = pJob;
+  task.taskCtx = NULL;
+
+  taosArrayPush(pJob->pTasks, &task);
+
+  qDebug("QID:0x%" PRIx64 " the %d task type %s initialized", pJob->queryId, taskIdx, ctgTaskTypeStr(task.type));
+
+  return TSDB_CODE_SUCCESS;
+}
+
 int32_t ctgInitGetIndexTask(SCtgJob *pJob, int32_t taskIdx, void* param) {
   char *name = (char*)param;
   SCtgTask task = {0};
@@ -405,6 +420,7 @@ int32_t ctgInitJob(SCatalog* pCtg, SRequestConnInfo *pConn, SCtgJob** job, uint6
   int32_t tbHashNum = (int32_t)taosArrayGetSize(pReq->pTableHash);
   int32_t udfNum = (int32_t)taosArrayGetSize(pReq->pUdf);
   int32_t qnodeNum = pReq->qNodeRequired ? 1 : 0;
+  int32_t dnodeNum = pReq->dNodeRequired ? 1 : 0;
   int32_t dbCfgNum = (int32_t)taosArrayGetSize(pReq->pDbCfg);
   int32_t indexNum = (int32_t)taosArrayGetSize(pReq->pIndex);
   int32_t userNum = (int32_t)taosArrayGetSize(pReq->pUser);
@@ -412,7 +428,7 @@ int32_t ctgInitJob(SCatalog* pCtg, SRequestConnInfo *pConn, SCtgJob** job, uint6
   int32_t tbIndexNum = (int32_t)taosArrayGetSize(pReq->pTableIndex);
   int32_t tbCfgNum = (int32_t)taosArrayGetSize(pReq->pTableCfg);
 
-  *taskNum = tbMetaNum + dbVgNum + udfNum + tbHashNum + qnodeNum + dbCfgNum + indexNum + userNum + dbInfoNum + tbIndexNum + tbCfgNum;
+  *taskNum = tbMetaNum + dbVgNum + udfNum + tbHashNum + qnodeNum + dnodeNum + dbCfgNum + indexNum + userNum + dbInfoNum + tbIndexNum + tbCfgNum;
   if (*taskNum <= 0) {
     ctgDebug("Empty input for job, no need to retrieve meta, reqId:0x%" PRIx64, reqId);
     return TSDB_CODE_SUCCESS;
@@ -435,6 +451,7 @@ int32_t ctgInitJob(SCatalog* pCtg, SRequestConnInfo *pConn, SCtgJob** job, uint6
   pJob->tbMetaNum = tbMetaNum;
   pJob->tbHashNum = tbHashNum;
   pJob->qnodeNum = qnodeNum;
+  pJob->dnodeNum = dnodeNum;
   pJob->dbVgNum = dbVgNum;
   pJob->udfNum = udfNum;
   pJob->dbCfgNum = dbCfgNum;
@@ -509,6 +526,10 @@ int32_t ctgInitJob(SCatalog* pCtg, SRequestConnInfo *pConn, SCtgJob** job, uint6
     CTG_ERR_JRET(ctgInitTask(pJob, CTG_TASK_GET_QNODE, NULL, NULL));
   }
 
+  if (dnodeNum) {
+    CTG_ERR_JRET(ctgInitTask(pJob, CTG_TASK_GET_DNODE, NULL, NULL));
+  }
+
   pJob->refId = taosAddRef(gCtgMgmt.jobPool, pJob);
   if (pJob->refId < 0) {
     ctgError("add job to ref failed, error: %s", tstrerror(terrno));
@@ -631,6 +652,22 @@ int32_t ctgDumpQnodeRes(SCtgTask* pTask) {
   return TSDB_CODE_SUCCESS;
 }
 
+int32_t ctgDumpDnodeRes(SCtgTask* pTask) {
+  SCtgJob* pJob = pTask->pJob;
+  if (NULL == pJob->jobRes.pDnodeList) {
+    pJob->jobRes.pDnodeList = taosArrayInit(1, sizeof(SMetaRes));
+    if (NULL == pJob->jobRes.pDnodeList) {
+      CTG_ERR_RET(TSDB_CODE_OUT_OF_MEMORY);
+    }
+  }
+
+  SMetaRes res = {.code = pTask->code, .pRes = pTask->res};
+  taosArrayPush(pJob->jobRes.pDnodeList, &res);
+
+  return TSDB_CODE_SUCCESS;
+}
+
+
 int32_t ctgDumpDbCfgRes(SCtgTask* pTask) {
   SCtgJob* pJob = pTask->pJob;
   if (NULL == pJob->jobRes.pDbCfg) {
@@ -1036,6 +1073,19 @@ _return:
   CTG_RET(code);
 }
 
+int32_t ctgHandleGetDnodeRsp(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));
+
+  TSWAP(pTask->res, pTask->msgCtx.out);
+  
+_return:
+
+  ctgHandleTaskEnd(pTask, code);
+
+  CTG_RET(code);
+}
+
 int32_t ctgHandleGetIndexRsp(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));
@@ -1311,6 +1361,15 @@ int32_t ctgLaunchGetQnodeTask(SCtgTask *pTask) {
   return TSDB_CODE_SUCCESS;
 }
 
+int32_t ctgLaunchGetDnodeTask(SCtgTask *pTask) {
+  SCatalog* pCtg = pTask->pJob->pCtg; 
+  SRequestConnInfo* pConn = &pTask->pJob->conn;
+
+  CTG_ERR_RET(ctgGetDnodeListFromMnode(pCtg, pConn, NULL, pTask));
+  return TSDB_CODE_SUCCESS;
+}
+
+
 int32_t ctgLaunchGetDbCfgTask(SCtgTask *pTask) {
   SCatalog* pCtg = pTask->pJob->pCtg; 
   SRequestConnInfo* pConn = &pTask->pJob->conn;
@@ -1462,6 +1521,7 @@ int32_t ctgCloneDbVg(SCtgTask* pTask, void** pRes) {
 
 SCtgAsyncFps gCtgAsyncFps[] = {
   {ctgInitGetQnodeTask,   ctgLaunchGetQnodeTask,   ctgHandleGetQnodeRsp,   ctgDumpQnodeRes,   NULL,               NULL},
+  {ctgInitGetDnodeTask,   ctgLaunchGetDnodeTask,   ctgHandleGetDnodeRsp,   ctgDumpDnodeRes,   NULL,               NULL},
   {ctgInitGetDbVgTask,    ctgLaunchGetDbVgTask,    ctgHandleGetDbVgRsp,    ctgDumpDbVgRes,    ctgCompDbVgTasks,   ctgCloneDbVg},
   {ctgInitGetDbCfgTask,   ctgLaunchGetDbCfgTask,   ctgHandleGetDbCfgRsp,   ctgDumpDbCfgRes,   NULL,               NULL},
   {ctgInitGetDbInfoTask,  ctgLaunchGetDbInfoTask,  ctgHandleGetDbInfoRsp,  ctgDumpDbInfoRes,  NULL,               NULL},

source/libs/catalog/src/ctgRemote.c

@@ -40,6 +40,21 @@ int32_t ctgProcessRspMsg(void* out, int32_t reqType, char* msg, int32_t msgSize,
       qDebug("Got qnode list from mnode, listNum:%d", (int32_t)taosArrayGetSize(out));
       break;
     }
+    case TDMT_MND_DNODE_LIST: {
+      if (TSDB_CODE_SUCCESS != rspCode) {
+        qError("error rsp for dnode list, error:%s", tstrerror(rspCode));
+        CTG_ERR_RET(rspCode);
+      }
+      
+      code = queryProcessMsgRsp[TMSG_INDEX(reqType)](out, msg, msgSize);
+      if (code) {
+        qError("Process dnode list rsp failed, error:%s", tstrerror(rspCode));
+        CTG_ERR_RET(code);
+      }
+      
+      qDebug("Got dnode list from mnode, listNum:%d", (int32_t)taosArrayGetSize(*(SArray**)out));
+      break;
+    }
     case TDMT_MND_USE_DB: {
       if (TSDB_CODE_SUCCESS != rspCode) {
         qError("error rsp for use db, error:%s, dbFName:%s", tstrerror(rspCode), target);
@@ -309,9 +324,6 @@ _return:
   CTG_RET(code);
 }
 
-
-
-
 int32_t ctgGetQnodeListFromMnode(SCatalog* pCtg, SRequestConnInfo *pConn, SArray *out, SCtgTask* pTask) {
   char *msg = NULL;
   int32_t msgLen = 0;
@@ -349,6 +361,39 @@ int32_t ctgGetQnodeListFromMnode(SCatalog* pCtg, SRequestConnInfo *pConn, SArray
   return TSDB_CODE_SUCCESS;
 }
 
+int32_t ctgGetDnodeListFromMnode(SCatalog* pCtg, SRequestConnInfo *pConn, SArray **out, SCtgTask* pTask) {
+  char *msg = NULL;
+  int32_t msgLen = 0;
+  int32_t reqType = TDMT_MND_DNODE_LIST;
+  void*(*mallocFp)(int32_t) = pTask ? taosMemoryMalloc : rpcMallocCont;
+
+  ctgDebug("try to get dnode list from mnode, mgmtEpInUse:%d", pConn->mgmtEps.inUse);
+
+  int32_t code = queryBuildMsg[TMSG_INDEX(reqType)](NULL, &msg, 0, &msgLen, mallocFp);
+  if (code) {
+    ctgError("Build dnode list msg failed, error:%s", tstrerror(code));
+    CTG_ERR_RET(code);
+  }
+
+  if (pTask) {
+    CTG_ERR_RET(ctgUpdateMsgCtx(&pTask->msgCtx, reqType, NULL, NULL));
+    CTG_RET(ctgAsyncSendMsg(pCtg, pConn, pTask, reqType, msg, msgLen));
+  }
+  
+  SRpcMsg rpcMsg = {
+      .msgType = reqType,
+      .pCont   = msg,
+      .contLen = msgLen,
+  };
+
+  SRpcMsg rpcRsp = {0};
+  rpcSendRecv(pConn->pTrans, &pConn->mgmtEps, &rpcMsg, &rpcRsp);
+
+  CTG_ERR_RET(ctgProcessRspMsg(out, reqType, rpcRsp.pCont, rpcRsp.contLen, rpcRsp.code, NULL));
+
+  return TSDB_CODE_SUCCESS;
+}
+
 
 int32_t ctgGetDBVgInfoFromMnode(SCatalog* pCtg, SRequestConnInfo *pConn, SBuildUseDBInput *input, SUseDbOutput *out, SCtgTask* pTask) {
   char *msg = NULL;

source/libs/catalog/src/ctgUtil.c

@@ -23,6 +23,8 @@ char *ctgTaskTypeStr(CTG_TASK_TYPE type) {
   switch (type) {
     case CTG_TASK_GET_QNODE:
       return "[get qnode list]";
+    case CTG_TASK_GET_DNODE:
+      return "[get dnode list]";
     case CTG_TASK_GET_DB_VGROUP:
       return "[get db vgroup]";
     case CTG_TASK_GET_DB_CFG:
@@ -349,6 +351,11 @@ void ctgFreeTaskRes(CTG_TASK_TYPE type, void **pRes) {
       *pRes = NULL;
       break;
     }
+    case CTG_TASK_GET_DNODE: {
+      taosArrayDestroy((SArray*)*pRes);
+      *pRes = NULL;
+      break;
+    }
     case CTG_TASK_GET_TB_META: {
       taosMemoryFreeClear(*pRes);
       break;
@@ -413,6 +420,11 @@ void ctgFreeSubTaskRes(CTG_TASK_TYPE type, void **pRes) {
       *pRes = NULL;
       break;
     }
+    case CTG_TASK_GET_DNODE: {
+      taosArrayDestroy((SArray*)*pRes);
+      *pRes = NULL;
+      break;
+    }
     case CTG_TASK_GET_TB_META: {
       taosMemoryFreeClear(*pRes);
       break;

source/libs/command/src/command.c

@@ -18,6 +18,7 @@
 #include "tdatablock.h"
 #include "tglobal.h"
 
+extern SConfig *tsCfg;
 static int32_t getSchemaBytes(const SSchema* pSchema) {
   switch (pSchema->type) {
     case TSDB_DATA_TYPE_BINARY:
@@ -551,7 +552,85 @@ static int32_t execShowCreateSTable(SShowCreateTableStmt* pStmt, SRetrieveTableR
 
 static int32_t execAlterLocal(SAlterLocalStmt* pStmt) { return TSDB_CODE_FAILED; }
 
-static int32_t execShowLocalVariables() { return TSDB_CODE_FAILED; }
+static SSDataBlock* buildLocalVariablesResultDataBlock() {
+  SSDataBlock* pBlock = taosMemoryCalloc(1, sizeof(SSDataBlock));
+  pBlock->info.numOfCols = SHOW_LOCAL_VARIABLES_RESULT_COLS;
+  pBlock->info.hasVarCol = true;
+
+  pBlock->pDataBlock = taosArrayInit(pBlock->info.numOfCols, sizeof(SColumnInfoData));
+
+  SColumnInfoData infoData = {0};
+  infoData.info.type = TSDB_DATA_TYPE_VARCHAR;
+  infoData.info.bytes = SHOW_LOCAL_VARIABLES_RESULT_FIELD1_LEN;
+
+  taosArrayPush(pBlock->pDataBlock, &infoData);
+
+  infoData.info.type = TSDB_DATA_TYPE_VARCHAR;
+  infoData.info.bytes = SHOW_LOCAL_VARIABLES_RESULT_FIELD2_LEN;
+  taosArrayPush(pBlock->pDataBlock, &infoData);
+
+  return pBlock;
+}
+
+
+int32_t setLocalVariablesResultIntoDataBlock(SSDataBlock* pBlock) {
+  int32_t numOfCfg = taosArrayGetSize(tsCfg->array);
+  int32_t numOfRows = 0;
+  blockDataEnsureCapacity(pBlock, numOfCfg);
+
+  for (int32_t i = 0, c = 0; i < numOfCfg; ++i, c = 0) {
+    SConfigItem *pItem = taosArrayGet(tsCfg->array, i);
+
+    char name[TSDB_CONFIG_OPTION_LEN + VARSTR_HEADER_SIZE] = {0};
+    STR_WITH_MAXSIZE_TO_VARSTR(name, pItem->name, TSDB_CONFIG_OPTION_LEN + VARSTR_HEADER_SIZE);
+    SColumnInfoData *pColInfo = taosArrayGet(pBlock->pDataBlock, c++);
+    colDataAppend(pColInfo, i, name, false);
+    
+    char value[TSDB_CONFIG_VALUE_LEN + VARSTR_HEADER_SIZE] = {0};
+    int32_t valueLen = 0;
+    cfgDumpItemValue(pItem, &value[VARSTR_HEADER_SIZE], TSDB_CONFIG_VALUE_LEN, &valueLen);
+    varDataSetLen(value, valueLen);
+    pColInfo = taosArrayGet(pBlock->pDataBlock, c++);
+    colDataAppend(pColInfo, i, value, false);
+
+    numOfRows++;
+  }
+
+
+  pBlock->info.rows = numOfRows;
+  
+  return TSDB_CODE_SUCCESS;
+}
+
+
+static int32_t execShowLocalVariables(SRetrieveTableRsp** pRsp) {
+  SSDataBlock* pBlock = buildLocalVariablesResultDataBlock();
+  int32_t code = setLocalVariablesResultIntoDataBlock(pBlock);
+  if (code) {
+    return code;
+  }
+
+  size_t rspSize = sizeof(SRetrieveTableRsp) + blockGetEncodeSize(pBlock);
+  *pRsp = taosMemoryCalloc(1, rspSize);
+  if (NULL == *pRsp) {
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+
+  (*pRsp)->useconds = 0;
+  (*pRsp)->completed = 1;
+  (*pRsp)->precision = 0;
+  (*pRsp)->compressed = 0;
+  (*pRsp)->compLen = 0;
+  (*pRsp)->numOfRows = htonl(pBlock->info.rows);
+  (*pRsp)->numOfCols = htonl(SHOW_LOCAL_VARIABLES_RESULT_COLS);
+
+  int32_t len = 0;
+  blockCompressEncode(pBlock, (*pRsp)->data, &len, SHOW_LOCAL_VARIABLES_RESULT_COLS, false);
+  ASSERT(len == rspSize - sizeof(SRetrieveTableRsp));
+
+  blockDataDestroy(pBlock);
+  return TSDB_CODE_SUCCESS;
+}
 
 int32_t qExecCommand(SNode* pStmt, SRetrieveTableRsp** pRsp) {
   switch (nodeType(pStmt)) {
@@ -568,7 +647,7 @@ int32_t qExecCommand(SNode* pStmt, SRetrieveTableRsp** pRsp) {
     case QUERY_NODE_ALTER_LOCAL_STMT:
       return execAlterLocal((SAlterLocalStmt*)pStmt);
     case QUERY_NODE_SHOW_LOCAL_VARIABLES_STMT:
-      return execShowLocalVariables();
+      return execShowLocalVariables(pRsp);
     default:
       break;
   }

source/libs/executor/inc/executil.h

@@ -106,7 +106,8 @@ int32_t getNumOfTotalRes(SGroupResInfo* pGroupResInfo);
 
 SSDataBlock* createResDataBlock(SDataBlockDescNode* pNode);
 
-int32_t getTableList(void* metaHandle, SScanPhysiNode* pScanNode, STableListInfo* pListInfo, SNode* pTagCond);
+EDealRes doTranslateTagExpr(SNode** pNode, void* pContext);
+int32_t getTableList(void* metaHandle, SScanPhysiNode* pScanNode, STableListInfo* pListInfo);
 SArray* createSortInfo(SNodeList* pNodeList);
 SArray* extractPartitionColInfo(SNodeList* pNodeList);
 SArray* extractColMatchInfo(SNodeList* pNodeList, SDataBlockDescNode* pOutputNodeList, int32_t* numOfOutputCols, int32_t type);

source/libs/executor/inc/executorimpl.h

@@ -282,7 +282,6 @@ typedef struct STagScanInfo {
   int32_t          curPos;
   SReadHandle      readHandle;
   STableListInfo  *pTableList;
-  SNode*           pFilterNode;  // filter info,
 } STagScanInfo;
 
 typedef enum EStreamScanMode {
@@ -341,7 +340,7 @@ typedef struct SStreamBlockScanInfo {
   SReadHandle     readHandle;
   uint64_t        tableUid;         // queried super table uid
   EStreamScanMode scanMode;
-  SOperatorInfo* pOperatorDumy;
+  SOperatorInfo* pSnapshotReadOp;
   SInterval      interval;     // if the upstream is an interval operator, the interval info is also kept here.
   SArray*        childIds;
   SessionWindowSupporter sessionSup;
@@ -838,14 +837,17 @@ int32_t finalizeResultRowIntoResultDataBlock(SDiskbasedBuf* pBuf, SResultRowPosi
                                        SqlFunctionCtx* pCtx, SExprInfo* pExprInfo, int32_t numOfExprs, const int32_t* rowCellOffset,
                                        SSDataBlock* pBlock, SExecTaskInfo* pTaskInfo);
 
-int32_t createMultipleDataReaders(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
+int32_t createScanTableListInfo(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
-                                  STableListInfo* pTableListInfo, SArray* arrayReader, uint64_t queryId,
+                                STableListInfo* pTableListInfo, uint64_t queryId, uint64_t taskId);
-                                  uint64_t taskId, SNode* pTagCond);
+SOperatorInfo* createGroupSortOperatorInfo(SOperatorInfo* downstream, SGroupSortPhysiNode* pSortPhyNode,
-SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanNode, SArray* dataReaders,
+                                           SExecTaskInfo* pTaskInfo);
-                                                SReadHandle* readHandle, SExecTaskInfo* pTaskInfo);
+SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanNode, STableListInfo *pTableListInfo,
+                                                SReadHandle* readHandle, SExecTaskInfo* pTaskInfo, uint64_t queryId, uint64_t taskId);
 
 void copyUpdateDataBlock(SSDataBlock* pDest, SSDataBlock* pSource, int32_t tsColIndex);
 
+int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle, SArray* groupKey);
+
 #ifdef __cplusplus
 }
 #endif

source/libs/executor/src/dataDeleter.c

@@ -86,7 +86,7 @@ static void toDataCacheEntry(SDataDeleterHandle* pHandle, const SInputData* pInp
   SColumnInfoData* pColRes = (SColumnInfoData*)taosArrayGet(pInput->pData->pDataBlock, 0);
 
   SDeleterRes* pRes = (SDeleterRes*)pEntry->data;
-  pRes->uid = pHandle->pDeleter->tableId;
+  pRes->suid = pHandle->pParam->suid;
   pRes->uidList = pHandle->pParam->pUidList;
   pRes->skey = pHandle->pDeleter->deleteTimeRange.skey;
   pRes->ekey = pHandle->pDeleter->deleteTimeRange.ekey;

source/libs/executor/src/executil.c

@@ -214,40 +214,138 @@ SSDataBlock* createResDataBlock(SDataBlockDescNode* pNode) {
   return pBlock;
 }
 
-int32_t getTableList(void* metaHandle, SScanPhysiNode* pScanNode, STableListInfo* pListInfo, SNode* pTagCond) {
+EDealRes doTranslateTagExpr(SNode** pNode, void* pContext) {
+  SMetaReader* mr = (SMetaReader*)pContext;
+  if(nodeType(*pNode) == QUERY_NODE_COLUMN){
+    SColumnNode* pSColumnNode = *(SColumnNode**)pNode;
+
+    SValueNode *res = (SValueNode *)nodesMakeNode(QUERY_NODE_VALUE);
+    if (NULL == res) {
+      return DEAL_RES_ERROR;
+    }
+
+    res->translate = true;
+    res->node.resType = pSColumnNode->node.resType;
+
+    STagVal tagVal = {0};
+    tagVal.cid = pSColumnNode->colId;
+    const char* p = metaGetTableTagVal(&mr->me, pSColumnNode->node.resType.type, &tagVal);
+    if (p == NULL) {
+      res->node.resType.type = TSDB_DATA_TYPE_NULL;
+    }else if (pSColumnNode->node.resType.type == TSDB_DATA_TYPE_JSON) {
+      int32_t len = ((const STag*)p) -> len;
+      res->datum.p = taosMemoryCalloc(len + 1, 1);
+      memcpy(res->datum.p, p, len);
+    } else if (IS_VAR_DATA_TYPE(pSColumnNode->node.resType.type)) {
+      res->datum.p = taosMemoryCalloc(tagVal.nData + VARSTR_HEADER_SIZE + 1, 1);
+      memcpy(varDataVal(res->datum.p), tagVal.pData, tagVal.nData);
+      varDataSetLen(res->datum.p, tagVal.nData);
+    } else {
+      nodesSetValueNodeValue(res, &(tagVal.i64));
+    }
+    nodesDestroyNode(*pNode);
+    *pNode = (SNode*)res;
+  }else if (nodeType(*pNode) == QUERY_NODE_FUNCTION){
+    SFunctionNode * pFuncNode = *(SFunctionNode**)pNode;
+    if(pFuncNode->funcType == FUNCTION_TYPE_TBNAME){
+      SValueNode *res = (SValueNode *)nodesMakeNode(QUERY_NODE_VALUE);
+      if (NULL == res) {
+        return DEAL_RES_ERROR;
+      }
+
+      res->translate = true;
+      res->node.resType = pFuncNode->node.resType;
+
+      int32_t len = strlen(mr->me.name);
+      res->datum.p = taosMemoryCalloc(len + VARSTR_HEADER_SIZE + 1, 1);
+      memcpy(varDataVal(res->datum.p), mr->me.name, len);
+      varDataSetLen(res->datum.p, len);
+      nodesDestroyNode(*pNode);
+      *pNode = (SNode*)res;
+    }
+  }
+
+  return DEAL_RES_CONTINUE;
+}
+
+static bool isTableOk(STableKeyInfo* info, SNode *pTagCond, SMeta *metaHandle){
+  SMetaReader    mr = {0};
+  metaReaderInit(&mr, metaHandle, 0);
+  metaGetTableEntryByUid(&mr, info->uid);
+
+  SNode *pTagCondTmp = nodesCloneNode(pTagCond);
+
+  nodesRewriteExprPostOrder(&pTagCondTmp, doTranslateTagExpr, &mr);
+  metaReaderClear(&mr);
+
+  SNode*  pNew = NULL;
+  int32_t code = scalarCalculateConstants(pTagCondTmp, &pNew);
+  if (TSDB_CODE_SUCCESS != code) {
+    nodesDestroyNode(pTagCondTmp);
+    return false;
+  }
+
+  ASSERT(nodeType(pNew) == QUERY_NODE_VALUE);
+  SValueNode *pValue = (SValueNode *)pNew;
+
+  ASSERT(pValue->node.resType.type == TSDB_DATA_TYPE_BOOL);
+  bool result = pValue->datum.b;
+  nodesDestroyNode(pNew);
+  return result;
+}
+
+int32_t getTableList(void* metaHandle, SScanPhysiNode* pScanNode, STableListInfo* pListInfo) {
   int32_t code = TSDB_CODE_SUCCESS;
   pListInfo->pTableList = taosArrayInit(8, sizeof(STableKeyInfo));
 
   uint64_t tableUid = pScanNode->uid;
 
+  pListInfo->suid = pScanNode->suid;
+  
+  SNode* pTagCond = (SNode*)pListInfo->pTagCond;
+  SNode* pTagIndexCond = (SNode*)pListInfo->pTagIndexCond;
   if (pScanNode->tableType == TSDB_SUPER_TABLE) {
-    if (pTagCond) {
+    if (pTagIndexCond) {
       SIndexMetaArg metaArg = {
           .metaEx = metaHandle, .idx = tsdbGetIdx(metaHandle), .ivtIdx = tsdbGetIvtIdx(metaHandle), .suid = tableUid};
 
       SArray* res = taosArrayInit(8, sizeof(uint64_t));
-      code = doFilterTag(pTagCond, &metaArg, res);
+      //code = doFilterTag(pTagIndexCond, &metaArg, res);
-      if (code == TSDB_CODE_INDEX_REBUILDING) {  // todo
+      code = TSDB_CODE_INDEX_REBUILDING;
-        // doFilter();
+      if (code == TSDB_CODE_INDEX_REBUILDING) {
+        code = tsdbGetAllTableList(metaHandle, tableUid, pListInfo->pTableList);
       } else if (code != TSDB_CODE_SUCCESS) {
         qError("failed  to  get tableIds, reason: %s, suid: %" PRIu64 "", tstrerror(code), tableUid);
         taosArrayDestroy(res);
         terrno = code;
         return code;
       } else {
-        qDebug("success to  get tableIds, size: %d, suid: %" PRIu64 "", (int)taosArrayGetSize(res), tableUid);
+        qDebug("sucess to  get tableIds, size: %d, suid: %" PRIu64 "", (int)taosArrayGetSize(res), tableUid);
       }
 
       for (int i = 0; i < taosArrayGetSize(res); i++) {
-        STableKeyInfo info = {.lastKey = TSKEY_INITIAL_VAL, .uid = *(uint64_t*)taosArrayGet(res, i)};
+        STableKeyInfo info = {.lastKey = TSKEY_INITIAL_VAL, .uid = *(uint64_t*)taosArrayGet(res, i), .groupId = 0};
         taosArrayPush(pListInfo->pTableList, &info);
       }
       taosArrayDestroy(res);
     } else {
       code = tsdbGetAllTableList(metaHandle, tableUid, pListInfo->pTableList);
     }
-  } else {  // Create one table group.
+
-    STableKeyInfo info = {.lastKey = 0, .uid = tableUid};
+    if(pTagCond){
+      int32_t i = 0;
+      while(i < taosArrayGetSize(pListInfo->pTableList)) {
+        STableKeyInfo* info = taosArrayGet(pListInfo->pTableList, i);
+        bool isOk = isTableOk(info, pTagCond, metaHandle);
+        if(!isOk){
+          taosArrayRemove(pListInfo->pTableList, i);
+          continue;
+        }
+        i++;
+      }
+    }
+  }else {  // Create one table group.
+    STableKeyInfo info = {.lastKey = 0, .uid = tableUid, .groupId = 0};
     taosArrayPush(pListInfo->pTableList, &info);
   }
 

source/libs/executor/src/executor.c

@@ -40,7 +40,7 @@ static int32_t doSetStreamBlock(SOperatorInfo* pOperator, void* input, size_t nu
     SStreamBlockScanInfo* pInfo = pOperator->info;
     pInfo->assignBlockUid = assignUid;
 
-    // the block type can not be changed in the streamscan operators
+    // no need to check
 #if 0
     if (pInfo->blockType == 0) {
       pInfo->blockType = type;
@@ -49,10 +49,7 @@ static int32_t doSetStreamBlock(SOperatorInfo* pOperator, void* input, size_t nu
       return TSDB_CODE_QRY_APP_ERROR;
     }
 #endif
-    // rollup sma, the same qTaskInfo is used to insert data by SubmitReq and fetch result by SSDataBlock
-    if (pInfo->blockType != type) { 
     pInfo->blockType = type;
-    }
 
     if (type == STREAM_DATA_TYPE_SUBMIT_BLOCK) {
       if (tqReadHandleSetMsg(pInfo->streamBlockReader, input, 0) < 0) {
@@ -70,6 +67,9 @@ static int32_t doSetStreamBlock(SOperatorInfo* pOperator, void* input, size_t nu
         taosArrayAddAll(p->pDataBlock, pDataBlock->pDataBlock);
         taosArrayPush(pInfo->pBlockLists, &p);
       }
+    } else if (type == STREAM_DATA_TYPE_FROM_SNAPSHOT) {
+      // do nothing
+      ASSERT(pInfo->blockType == STREAM_DATA_TYPE_FROM_SNAPSHOT);
     } else {
       ASSERT(0);
     }
@@ -78,6 +78,14 @@ static int32_t doSetStreamBlock(SOperatorInfo* pOperator, void* input, size_t nu
   }
 }
 
+int32_t qStreamScanSnapshot(qTaskInfo_t tinfo) {
+  if (tinfo == NULL) {
+    return TSDB_CODE_QRY_APP_ERROR;
+  }
+  SExecTaskInfo* pTaskInfo = (SExecTaskInfo*)tinfo;
+  return doSetStreamBlock(pTaskInfo->pRoot, NULL, 0, STREAM_DATA_TYPE_FROM_SNAPSHOT, 0, NULL);
+}
+
 int32_t qSetStreamInput(qTaskInfo_t tinfo, const void* input, int32_t type, bool assignUid) {
   return qSetMultiStreamInput(tinfo, input, 1, type, assignUid);
 }
@@ -109,14 +117,6 @@ qTaskInfo_t qCreateStreamExecTaskInfo(void* msg, void* streamReadHandle) {
     return NULL;
   }
 
-  // print those info into log
-#if 0
-  pMsg->sId = pMsg->sId;
-  pMsg->queryId = pMsg->queryId;
-  pMsg->taskId = pMsg->taskId;
-  pMsg->contentLen = pMsg->contentLen;
-#endif
-
   /*qDebugL("stream task string %s", (const char*)msg);*/
 
   struct SSubplan* plan = NULL;

source/libs/executor/src/executorimpl.c

@@ -3876,8 +3876,7 @@ static SExecTaskInfo* createExecTaskInfo(uint64_t queryId, uint64_t taskId, EOPT
 }
 
 static tsdbReaderT doCreateDataReader(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
-                                      STableListInfo* pTableListInfo, uint64_t queryId, uint64_t taskId,
+                                      STableListInfo* pTableListInfo, uint64_t queryId, uint64_t taskId);
-                                      SNode* pTagCond);
 
 static SArray* extractColumnInfo(SNodeList* pNodeList);
 
@@ -3972,14 +3971,16 @@ int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle,
         }
       }
     }
-
     int32_t   len = (int32_t)(pStart - (char*)keyBuf);
-    uint64_t* groupId = taosHashGet(pTableListInfo->map, keyBuf, len);
+
-    if (groupId) {
+    uint64_t* pGroupId = taosHashGet(pTableListInfo->map, keyBuf, len);
-      taosHashPut(pTableListInfo->map, &(info->uid), sizeof(uint64_t), groupId, sizeof(uint64_t));
+
-    } else {
+    if (!pGroupId) {
       uint64_t tmpId = calcGroupId(keyBuf, len);
+      info->groupId = tmpId;
       taosHashPut(pTableListInfo->map, &(info->uid), sizeof(uint64_t), &tmpId, sizeof(uint64_t));
+    } else {
+      info->groupId = *pGroupId;
     }
 
     metaReaderClear(&mr);
@@ -3989,7 +3990,7 @@ int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle,
 }
 
 SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo, SReadHandle* pHandle,
-                                  uint64_t queryId, uint64_t taskId, STableListInfo* pTableListInfo, SNode* pTagCond) {
+                                  uint64_t queryId, uint64_t taskId, STableListInfo* pTableListInfo) {
   int32_t type = nodeType(pPhyNode);
 
   if (pPhyNode->pChildren == NULL || LIST_LENGTH(pPhyNode->pChildren) == 0) {
@@ -3997,7 +3998,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
       STableScanPhysiNode* pTableScanNode = (STableScanPhysiNode*)pPhyNode;
 
       tsdbReaderT pDataReader =
-          doCreateDataReader(pTableScanNode, pHandle, pTableListInfo, (uint64_t)queryId, taskId, pTagCond);
+          doCreateDataReader(pTableScanNode, pHandle, pTableListInfo, (uint64_t)queryId, taskId);
       if (pDataReader == NULL && terrno != 0) {
         pTaskInfo->code = terrno;
         return NULL;
@@ -4025,14 +4026,9 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
       return pOperator;
     } else if (QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN == type) {
       STableMergeScanPhysiNode* pTableScanNode = (STableMergeScanPhysiNode*)pPhyNode;
-
+      createScanTableListInfo(pTableScanNode, pHandle, pTableListInfo, queryId, taskId);
-      SArray* dataReaders = taosArrayInit(8, POINTER_BYTES);
-      createMultipleDataReaders(pTableScanNode, pHandle, pTableListInfo, dataReaders, queryId, taskId, pTagCond);
       extractTableSchemaVersion(pHandle, pTableScanNode->scan.uid, pTaskInfo);
-      SArray* groupKeys = extractPartitionColInfo(pTableScanNode->pPartitionTags);
+      SOperatorInfo*  pOperator = createTableMergeScanOperatorInfo(pTableScanNode, pTableListInfo, pHandle, pTaskInfo, queryId, taskId);
-      generateGroupIdMap(pTableListInfo, pHandle, groupKeys);  // todo for json
-      taosArrayDestroy(groupKeys);
-      SOperatorInfo*  pOperator = createTableMergeScanOperatorInfo(pTableScanNode, dataReaders, pHandle, pTaskInfo);
       STableScanInfo* pScanInfo = pOperator->info;
       pTaskInfo->cost.pRecoder = &pScanInfo->readRecorder;
       return pOperator;
@@ -4049,16 +4045,17 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
       tsdbReaderT pDataReader = NULL;
 
       if (pHandle) {
-        if (pHandle->vnode) {
+        if (pHandle->initTsdbReader) {
-          // for stram
+          // for stream
+          ASSERT(pHandle->vnode);
           pDataReader =
-              doCreateDataReader(pTableScanNode, pHandle, pTableListInfo, (uint64_t)queryId, taskId, pTagCond);
+              doCreateDataReader(pTableScanNode, pHandle, pTableListInfo, (uint64_t)queryId, taskId);
         } else {
           // for tq
-          getTableList(pHandle->meta, pScanPhyNode, pTableListInfo, pTagCond);
+          ASSERT(pHandle->meta);
+          getTableList(pHandle->meta, pScanPhyNode, pTableListInfo);
         }
       }
-
       if (pDataReader == NULL && terrno != 0) {
         qDebug("%s pDataReader is NULL", GET_TASKID(pTaskInfo));
         // return NULL;
@@ -4083,7 +4080,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
     } else if (QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN == type) {
       STagScanPhysiNode* pScanPhyNode = (STagScanPhysiNode*)pPhyNode;
 
-      int32_t code = getTableList(pHandle->meta, pScanPhyNode, pTableListInfo, pScanPhyNode->node.pConditions);
+      int32_t code = getTableList(pHandle->meta, pScanPhyNode, pTableListInfo);
       if (code != TSDB_CODE_SUCCESS) {
         pTaskInfo->code = terrno;
         return NULL;
@@ -4101,7 +4098,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
           return NULL;
         }
       } else {  // Create one table group.
-        STableKeyInfo info = {.lastKey = 0, .uid = pBlockNode->uid};
+        STableKeyInfo info = {.lastKey = 0, .uid = pBlockNode->uid, .groupId = 0};
         taosArrayPush(pTableListInfo->pTableList, &info);
       }
 
@@ -4141,7 +4138,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
   SOperatorInfo** ops = taosMemoryCalloc(size, POINTER_BYTES);
   for (int32_t i = 0; i < size; ++i) {
     SPhysiNode* pChildNode = (SPhysiNode*)nodesListGetNode(pPhyNode->pChildren, i);
-    ops[i] = createOperatorTree(pChildNode, pTaskInfo, pHandle, queryId, taskId, pTableListInfo, pTagCond);
+    ops[i] = createOperatorTree(pChildNode, pTaskInfo, pHandle, queryId, taskId, pTableListInfo);
     if (ops[i] == NULL) {
       return NULL;
     }
@@ -4232,6 +4229,8 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
     pOptr = createStreamFinalIntervalOperatorInfo(ops[0], pPhyNode, pTaskInfo, children);
   } else if (QUERY_NODE_PHYSICAL_PLAN_SORT == type) {
     pOptr = createSortOperatorInfo(ops[0], (SSortPhysiNode*)pPhyNode, pTaskInfo);
+  } else if (QUERY_NODE_PHYSICAL_PLAN_GROUP_SORT == type) {
+    pOptr = createGroupSortOperatorInfo(ops[0], (SGroupSortPhysiNode*)pPhyNode, pTaskInfo);
   } else if (QUERY_NODE_PHYSICAL_PLAN_MERGE == type) {
     SMergePhysiNode* pMergePhyNode = (SMergePhysiNode*)pPhyNode;
 
@@ -4343,8 +4342,8 @@ SArray* extractColumnInfo(SNodeList* pNodeList) {
 }
 
 tsdbReaderT doCreateDataReader(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
-                               STableListInfo* pTableListInfo, uint64_t queryId, uint64_t taskId, SNode* pTagCond) {
+                               STableListInfo* pTableListInfo, uint64_t queryId, uint64_t taskId) {
-  int32_t code = getTableList(pHandle->meta, &pTableScanNode->scan, pTableListInfo, pTagCond);
+  int32_t code = getTableList(pHandle->meta, &pTableScanNode->scan, pTableListInfo);
   if (code != TSDB_CODE_SUCCESS) {
     goto _error;
   }
@@ -4470,6 +4469,7 @@ int32_t createDataSinkParam(SDataSinkNode* pNode, void** pParam, qTaskInfo_t* pT
         return TSDB_CODE_OUT_OF_MEMORY;
       }
       int32_t tbNum = taosArrayGetSize(pTask->tableqinfoList.pTableList);
+      pDeleterParam->suid = pTask->tableqinfoList.suid;
       pDeleterParam->pUidList = taosArrayInit(tbNum, sizeof(uint64_t));
       if (NULL == pDeleterParam->pUidList) {
         taosMemoryFree(pDeleterParam);
@@ -4502,8 +4502,10 @@ int32_t createExecTaskInfoImpl(SSubplan* pPlan, SExecTaskInfo** pTaskInfo, SRead
   }
 
   (*pTaskInfo)->sql = sql;
+  (*pTaskInfo)->tableqinfoList.pTagCond = pPlan->pTagCond;
+  (*pTaskInfo)->tableqinfoList.pTagIndexCond = pPlan->pTagIndexCond;
   (*pTaskInfo)->pRoot = createOperatorTree(pPlan->pNode, *pTaskInfo, pHandle, queryId, taskId,
-                                           &(*pTaskInfo)->tableqinfoList, pPlan->pTagCond);
+                                           &(*pTaskInfo)->tableqinfoList);
   if (NULL == (*pTaskInfo)->pRoot) {
     code = (*pTaskInfo)->code;
     goto _complete;

source/libs/executor/src/scanoperator.c

@@ -337,7 +337,8 @@ void addTagPseudoColumnData(SReadHandle* pHandle, SExprInfo* pPseudoExpr, int32_
       }
 
       for (int32_t i = 0; i < pBlock->info.rows; ++i) {
-        colDataAppend(pColInfoData, i, data, (data == NULL));
+        colDataAppend(pColInfoData, i, data,
+                      (data == NULL) || (pColInfoData->info.type == TSDB_DATA_TYPE_JSON && tTagIsJsonNull(data)));
       }
 
       if (data && (pColInfoData->info.type != TSDB_DATA_TYPE_JSON) && p != NULL &&
@@ -616,7 +617,7 @@ static int32_t doGetTableRowSize(void* pMeta, uint64_t uid) {
   metaGetTableEntryByUid(&mr, uid);
   if (mr.me.type == TSDB_SUPER_TABLE) {
     int32_t numOfCols = mr.me.stbEntry.schemaRow.nCols;
-    for(int32_t i = 0; i < numOfCols; ++i) {
+    for (int32_t i = 0; i < numOfCols; ++i) {
       rowLen += mr.me.stbEntry.schemaRow.pSchema[i].bytes;
     }
   } else if (mr.me.type == TSDB_CHILD_TABLE) {
@@ -624,12 +625,12 @@ static int32_t doGetTableRowSize(void* pMeta, uint64_t uid) {
     metaGetTableEntryByUid(&mr, suid);
     int32_t numOfCols = mr.me.stbEntry.schemaRow.nCols;
 
-    for(int32_t i = 0; i < numOfCols; ++i) {
+    for (int32_t i = 0; i < numOfCols; ++i) {
       rowLen += mr.me.stbEntry.schemaRow.pSchema[i].bytes;
     }
   } else if (mr.me.type == TSDB_NORMAL_TABLE) {
     int32_t numOfCols = mr.me.ntbEntry.schemaRow.nCols;
-    for(int32_t i = 0; i < numOfCols; ++i) {
+    for (int32_t i = 0; i < numOfCols; ++i) {
       rowLen += mr.me.ntbEntry.schemaRow.pSchema[i].bytes;
     }
   }
@@ -776,7 +777,7 @@ static bool prepareDataScan(SStreamBlockScanInfo* pInfo) {
   if (!needRead) {
     return false;
   }
-  STableScanInfo* pTableScanInfo = pInfo->pOperatorDumy->info;
+  STableScanInfo* pTableScanInfo = pInfo->pSnapshotReadOp->info;
   pTableScanInfo->cond.twindows[0] = win;
   pTableScanInfo->curTWinIdx = 0;
   tsdbResetReadHandle(pTableScanInfo->dataReader, &pTableScanInfo->cond, 0);
@@ -821,11 +822,11 @@ static uint64_t getGroupId(SOperatorInfo* pOperator, SSDataBlock* pBlock, int32_
 static SSDataBlock* doDataScan(SStreamBlockScanInfo* pInfo) {
   while (1) {
     SSDataBlock* pResult = NULL;
-    pResult = doTableScan(pInfo->pOperatorDumy);
+    pResult = doTableScan(pInfo->pSnapshotReadOp);
     if (pResult == NULL) {
       if (prepareDataScan(pInfo)) {
         // scan next window data
-        pResult = doTableScan(pInfo->pOperatorDumy);
+        pResult = doTableScan(pInfo->pSnapshotReadOp);
       }
     }
     if (!pResult) {
@@ -860,11 +861,11 @@ static void setUpdateData(SStreamBlockScanInfo* pInfo, SSDataBlock* pBlock, SSDa
     ASSERT(pBlock->info.numOfCols == pUpdateBlock->info.numOfCols);
 
     int32_t rowId = *(int32_t*)taosArrayGet(pInfo->tsArray, pInfo->tsArrayIndex);
-    pInfo->groupId = getGroupId(pInfo->pOperatorDumy, pBlock, rowId);
+    pInfo->groupId = getGroupId(pInfo->pSnapshotReadOp, 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);
+      uint64_t id = getGroupId(pInfo->pSnapshotReadOp, pBlock, rowId);
       if (pInfo->groupId != id) {
         break;
       }
@@ -916,6 +917,7 @@ static SSDataBlock* doStreamBlockScan(SOperatorInfo* pOperator) {
   }
 
   size_t total = taosArrayGetSize(pInfo->pBlockLists);
+  // TODO: refactor
   if (pInfo->blockType == STREAM_DATA_TYPE_SSDATA_BLOCK) {
     if (pInfo->validBlockIndex >= total) {
       /*doClearBufferedBlocks(pInfo);*/
@@ -927,7 +929,7 @@ static SSDataBlock* doStreamBlockScan(SOperatorInfo* pOperator) {
     SSDataBlock* pBlock = taosArrayGetP(pInfo->pBlockLists, current);
     blockDataUpdateTsWindow(pBlock, 0);
     return pBlock;
-  } else {
+  } else if (pInfo->blockType == STREAM_DATA_TYPE_SUBMIT_BLOCK) {
     if (pInfo->scanMode == STREAM_SCAN_FROM_RES) {
       blockDataDestroy(pInfo->pUpdateRes);
       pInfo->scanMode = STREAM_SCAN_FROM_READERHANDLE;
@@ -1061,6 +1063,15 @@ static SSDataBlock* doStreamBlockScan(SOperatorInfo* pOperator) {
     }
 
     return (pBlockInfo->rows == 0) ? NULL : pInfo->pRes;
+  } else if (pInfo->blockType == STREAM_DATA_TYPE_FROM_SNAPSHOT) {
+    SSDataBlock* pResult = doTableScan(pInfo->pSnapshotReadOp);
+    if (pResult) {
+      return pResult->info.rows > 0 ? pResult : NULL;
+    }
+    return NULL;
+  } else {
+    ASSERT(0);
+    return NULL;
   }
 }
 
@@ -1125,7 +1136,7 @@ SOperatorInfo* createStreamScanOperatorInfo(void* pDataReader, SReadHandle* pHan
     } else {
       pInfo->pUpdateInfo = NULL;
     }
-    pInfo->pOperatorDumy = pTableScanDummy;
+    pInfo->pSnapshotReadOp = pTableScanDummy;
     pInfo->interval = pSTInfo->interval;
 
     pInfo->readHandle = *pHandle;
@@ -1547,11 +1558,15 @@ static SSDataBlock* doSysTableScan(SOperatorInfo* pOperator) {
         return NULL;
       }
 
+      int32_t msgType = (strcasecmp(name, TSDB_INS_TABLE_DNODE_VARIABLES) == 0) ? TDMT_DND_SYSTABLE_RETRIEVE
+                                                                                : TDMT_MND_SYSTABLE_RETRIEVE;
+
       pMsgSendInfo->param = pOperator;
       pMsgSendInfo->msgInfo.pData = buf1;
       pMsgSendInfo->msgInfo.len = contLen;
-      pMsgSendInfo->msgType = TDMT_MND_SYSTABLE_RETRIEVE;
+      pMsgSendInfo->msgType = msgType;
       pMsgSendInfo->fp = loadSysTableCallback;
+      pMsgSendInfo->requestId = pTaskInfo->id.queryId;
 
       int64_t transporterId = 0;
       int32_t code =
@@ -1586,6 +1601,8 @@ static SSDataBlock* doSysTableScan(SOperatorInfo* pOperator) {
       taosMemoryFree(pRsp);
       if (pInfo->pRes->info.rows > 0) {
         return pInfo->pRes;
+      } else if (pOperator->status == OP_EXEC_DONE) {
+        return NULL;
       }
     }
   }
@@ -1828,7 +1845,8 @@ static SSDataBlock* doTagScan(SOperatorInfo* pOperator) {
         } else {
           data = (char*)p;
         }
-        colDataAppend(pDst, count, data, (data == NULL));
+        colDataAppend(pDst, count, data,
+                      (data == NULL) || (pDst->info.type == TSDB_DATA_TYPE_JSON && tTagIsJsonNull(data)));
 
         if (pDst->info.type != TSDB_DATA_TYPE_JSON && p != NULL && IS_VAR_DATA_TYPE(((const STagVal*)p)->type) &&
             data != NULL) {
@@ -1851,9 +1869,7 @@ static SSDataBlock* doTagScan(SOperatorInfo* pOperator) {
   }
 
   pRes->info.rows = count;
-  doFilter(pInfo->pFilterNode, pRes);
+  pOperator->resultInfo.totalRows += count;
-
-  pOperator->resultInfo.totalRows += pRes->info.rows;
 
   return (pRes->info.rows == 0) ? NULL : pInfo->pRes;
 }
@@ -1878,7 +1894,6 @@ SOperatorInfo* createTagScanOperatorInfo(SReadHandle* pReadHandle, STagScanPhysi
   SExprInfo* pExprInfo = createExprInfo(pPhyNode->pScanPseudoCols, NULL, &numOfExprs);
   SArray*    colList = extractColMatchInfo(pPhyNode->pScanPseudoCols, pDescNode, &num, COL_MATCH_FROM_COL_ID);
 
-
   int32_t code = initExprSupp(&pOperator->exprSupp, pExprInfo, numOfExprs);
   if (code != TSDB_CODE_SUCCESS) {
     goto _error;
@@ -1889,8 +1904,6 @@ SOperatorInfo* createTagScanOperatorInfo(SReadHandle* pReadHandle, STagScanPhysi
   pInfo->pRes = createResDataBlock(pDescNode);
   pInfo->readHandle = *pReadHandle;
   pInfo->curPos = 0;
-  pInfo->pFilterNode      = pPhyNode->node.pConditions;
-
   pOperator->name = "TagScanOperator";
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN;
 
@@ -1915,6 +1928,12 @@ _error:
 }
 
 typedef struct STableMergeScanInfo {
+  STableListInfo* tableListInfo;
+  int32_t         tableStartIndex;
+  int32_t         tableEndIndex;
+  bool            hasGroupId;
+  uint64_t        groupId;
+
   SArray*     dataReaders;  // array of tsdbReaderT*
   SReadHandle readHandle;
 
@@ -1927,11 +1946,9 @@ typedef struct STableMergeScanInfo {
   SSDataBlock* pSortInputBlock;
   int64_t      startTs;  // sort start time
 
-  bool          hasGroupId;
-  uint64_t      groupId;
-  STupleHandle* prefetchedTuple;
-
   SArray*  sortSourceParams;
+  uint64_t queryId;
+  uint64_t taskId;
 
   SFileBlockLoadRecorder readRecorder;
   int64_t                numOfRows;
@@ -1959,29 +1976,42 @@ typedef struct STableMergeScanInfo {
                        // window to check if current data block needs to be loaded.
 
   SSampleExecInfo sample;  // sample execution info
-  int32_t         curTWinIdx;
-
 } STableMergeScanInfo;
 
-int32_t createMultipleDataReaders(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
+int32_t compareTableKeyInfoByGid(const void* p1, const void* p2) {
-                                  STableListInfo* pTableListInfo, SArray* arrayReader, uint64_t queryId,
+  const STableKeyInfo* info1 = p1;
-                                  uint64_t taskId, SNode* pTagCond) {
+  const STableKeyInfo* info2 = p2;
-  int32_t code = getTableList(pHandle->meta, &pTableScanNode->scan, pTableListInfo, pTagCond);
+  return info1->groupId - info2->groupId;
+}
+
+int32_t createScanTableListInfo(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
+                                STableListInfo* pTableListInfo, uint64_t queryId, uint64_t taskId) {
+  int32_t code = getTableList(pHandle->meta, &pTableScanNode->scan, pTableListInfo);
   if (code != TSDB_CODE_SUCCESS) {
-    goto _error;
+    return code;
   }
 
   if (taosArrayGetSize(pTableListInfo->pTableList) == 0) {
     qDebug("no table qualified for query, TID:0x%" PRIx64 ", QID:0x%" PRIx64, taskId, queryId);
-    goto _error;
+    return TSDB_CODE_SUCCESS;
   }
+  SArray* groupKeys = extractPartitionColInfo(pTableScanNode->pPartitionTags);
+  generateGroupIdMap(pTableListInfo, pHandle, groupKeys);  // todo for json
+  if (groupKeys) {
+    taosArraySort(pTableListInfo->pTableList, compareTableKeyInfoByGid);
+  }
+  taosArrayDestroy(groupKeys);
+  return TSDB_CODE_SUCCESS;
+}
 
+int32_t doCreateMultipleDataReaders(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
+                                    STableListInfo* pTableListInfo, SArray* arrayReader, uint64_t queryId,
+                                    uint64_t taskId) {
   SQueryTableDataCond cond = {0};
-  code = initQueryTableDataCond(&cond, pTableScanNode);
+  int32_t             code = initQueryTableDataCond(&cond, pTableScanNode);
   if (code != TSDB_CODE_SUCCESS) {
     goto _error;
   }
-  // TODO: free the sublist info and the table list in it
   for (int32_t i = 0; i < taosArrayGetSize(pTableListInfo->pTableList); ++i) {
     STableListInfo* subListInfo = taosMemoryCalloc(1, sizeof(subListInfo));
     subListInfo->pTableList = taosArrayInit(1, sizeof(STableKeyInfo));
@@ -1995,12 +2025,30 @@ int32_t createMultipleDataReaders(STableScanPhysiNode* pTableScanNode, SReadHand
   }
   cleanupQueryTableDataCond(&cond);
 
-  return 0;
+  return TSDB_CODE_SUCCESS;
 
 _error:
   return code;
 }
 
+int32_t createMultipleDataReaders(SQueryTableDataCond* pQueryCond, SReadHandle* pHandle, STableListInfo* pTableListInfo,
+                                  int32_t tableStartIdx, int32_t tableEndIdx, SArray* arrayReader, uint64_t queryId,
+                                  uint64_t taskId) {
+  for (int32_t i = tableStartIdx; i <= tableEndIdx; ++i) {
+    STableListInfo* subListInfo = taosMemoryCalloc(1, sizeof(subListInfo));
+    subListInfo->pTableList = taosArrayInit(1, sizeof(STableKeyInfo));
+    taosArrayPush(subListInfo->pTableList, taosArrayGet(pTableListInfo->pTableList, i));
+
+    tsdbReaderT* pReader = tsdbReaderOpen(pHandle->vnode, pQueryCond, subListInfo, queryId, taskId);
+    taosArrayPush(arrayReader, &pReader);
+
+    taosArrayDestroy(subListInfo->pTableList);
+    taosMemoryFree(subListInfo);
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
 static int32_t loadDataBlockFromOneTable(SOperatorInfo* pOperator, STableMergeScanInfo* pTableScanInfo,
                                          int32_t readerIdx, SSDataBlock* pBlock, uint32_t* status) {
   SExecTaskInfo*       pTaskInfo = pOperator->pTaskInfo;
@@ -2180,22 +2228,47 @@ SArray* generateSortByTsInfo(int32_t order) {
   return pList;
 }
 
-int32_t doOpenTableMergeScanOperator(SOperatorInfo* pOperator) {
+int32_t startGroupTableMergeScan(SOperatorInfo* pOperator) {
   STableMergeScanInfo* pInfo = pOperator->info;
   SExecTaskInfo*       pTaskInfo = pOperator->pTaskInfo;
 
-  if (OPTR_IS_OPENED(pOperator)) {
+  {
-    return TSDB_CODE_SUCCESS;
+    size_t  tableListSize = taosArrayGetSize(pInfo->tableListInfo->pTableList);
+    int32_t i = pInfo->tableStartIndex + 1;
+    for (; i < tableListSize; ++i) {
+      STableKeyInfo* tableKeyInfo = taosArrayGet(pInfo->tableListInfo->pTableList, i);
+      if (tableKeyInfo->groupId != pInfo->groupId) {
+        break;
+      }
+    }
+    pInfo->tableEndIndex = i - 1;
   }
 
-  int32_t numOfBufPage = pInfo->sortBufSize / pInfo->bufPageSize;
+  int32_t tableStartIdx = pInfo->tableStartIndex;
+  int32_t tableEndIdx = pInfo->tableEndIndex;
 
+  STableListInfo* tableListInfo = pInfo->tableListInfo;
+  createMultipleDataReaders(&pInfo->cond, &pInfo->readHandle, tableListInfo, tableStartIdx, tableEndIdx,
+                            pInfo->dataReaders, pInfo->queryId, pInfo->taskId);
+
+  // todo the total available buffer should be determined by total capacity of buffer of this task.
+  // the additional one is reserved for merge result
+  pInfo->sortBufSize = pInfo->bufPageSize * (tableEndIdx - tableStartIdx + 1 + 1);
+  int32_t numOfBufPage = pInfo->sortBufSize / pInfo->bufPageSize;
   pInfo->pSortHandle = tsortCreateSortHandle(pInfo->pSortInfo, SORT_MULTISOURCE_MERGE, pInfo->bufPageSize, numOfBufPage,
                                              pInfo->pSortInputBlock, pTaskInfo->id.str);
 
   tsortSetFetchRawDataFp(pInfo->pSortHandle, getTableDataBlock, NULL, NULL);
 
   size_t numReaders = taosArrayGetSize(pInfo->dataReaders);
+  for (int32_t i = 0; i < numReaders; ++i) {
+    STableMergeScanSortSourceParam param = {0};
+    param.readerIdx = i;
+    param.pOperator = pOperator;
+    param.inputBlock = createOneDataBlock(pInfo->pResBlock, false);
+    taosArrayPush(pInfo->sortSourceParams, &param);
+  }
+
   for (int32_t i = 0; i < numReaders; ++i) {
     SSortSource*                    ps = taosMemoryCalloc(1, sizeof(SSortSource));
     STableMergeScanSortSourceParam* param = taosArrayGet(pInfo->sortSourceParams, i);
@@ -2209,9 +2282,22 @@ int32_t doOpenTableMergeScanOperator(SOperatorInfo* pOperator) {
     longjmp(pTaskInfo->env, terrno);
   }
 
-  pOperator->status = OP_RES_TO_RETURN;
+  return TSDB_CODE_SUCCESS;
+}
+
+int32_t stopGroupTableMergeScan(SOperatorInfo* pOperator) {
+  STableMergeScanInfo* pInfo = pOperator->info;
+  SExecTaskInfo*       pTaskInfo = pOperator->pTaskInfo;
+
+  tsortDestroySortHandle(pInfo->pSortHandle);
+  taosArrayClear(pInfo->sortSourceParams);
+
+  for (int32_t i = 0; i < taosArrayGetSize(pInfo->dataReaders); ++i) {
+    tsdbReaderT* reader = taosArrayGetP(pInfo->dataReaders, i);
+    tsdbCleanupReadHandle(reader);
+  }
+  taosArrayDestroy(pInfo->dataReaders);
 
-  OPTR_SET_OPENED(pOperator);
   return TSDB_CODE_SUCCESS;
 }
 
@@ -2254,14 +2340,38 @@ SSDataBlock* doTableMergeScan(SOperatorInfo* pOperator) {
   if (code != TSDB_CODE_SUCCESS) {
     longjmp(pTaskInfo->env, code);
   }
+  size_t tableListSize = taosArrayGetSize(pInfo->tableListInfo->pTableList);
+  if (!pInfo->hasGroupId) {
+    pInfo->hasGroupId = true;
 
-  SSDataBlock* pBlock = getSortedTableMergeScanBlockData(pInfo->pSortHandle, pOperator->resultInfo.capacity, pOperator);
+    if (tableListSize == 0) {
-
-  if (pBlock != NULL) {
-    pOperator->resultInfo.totalRows += pBlock->info.rows;
-  } else {
       doSetOperatorCompleted(pOperator);
+      return NULL;
     }
+    pInfo->tableStartIndex = 0;
+    pInfo->groupId = ((STableKeyInfo*)taosArrayGet(pInfo->tableListInfo->pTableList, pInfo->tableStartIndex))->groupId;
+    startGroupTableMergeScan(pOperator);
+  }
+  SSDataBlock* pBlock = NULL;
+  while (pInfo->tableStartIndex < tableListSize) {
+    pBlock = getSortedTableMergeScanBlockData(pInfo->pSortHandle, pOperator->resultInfo.capacity, pOperator);
+    if (pBlock != NULL) {
+      pBlock->info.groupId = pInfo->groupId;
+      pOperator->resultInfo.totalRows += pBlock->info.rows;
+      return pBlock;
+    } else {
+      stopGroupTableMergeScan(pOperator);
+      if (pInfo->tableEndIndex >= tableListSize - 1) {
+        doSetOperatorCompleted(pOperator);
+        break;
+      }
+      pInfo->tableStartIndex = pInfo->tableEndIndex + 1;
+      pInfo->groupId =
+          ((STableKeyInfo*)taosArrayGet(pInfo->tableListInfo->pTableList, pInfo->tableStartIndex))->groupId;
+      startGroupTableMergeScan(pOperator);
+    }
+  }
+
   return pBlock;
 }
 
@@ -2269,17 +2379,10 @@ void destroyTableMergeScanOperatorInfo(void* param, int32_t numOfOutput) {
   STableMergeScanInfo* pTableScanInfo = (STableMergeScanInfo*)param;
   cleanupQueryTableDataCond(&pTableScanInfo->cond);
 
-  for (int32_t i = 0; i < taosArrayGetSize(pTableScanInfo->dataReaders); ++i) {
-    tsdbReaderT* reader = taosArrayGetP(pTableScanInfo->dataReaders, i);
-    tsdbCleanupReadHandle(reader);
-  }
-  taosArrayDestroy(pTableScanInfo->dataReaders);
-
   if (pTableScanInfo->pColMatchInfo != NULL) {
     taosArrayDestroy(pTableScanInfo->pColMatchInfo);
   }
 
-  taosArrayDestroy(pTableScanInfo->sortSourceParams);
   pTableScanInfo->pResBlock = blockDataDestroy(pTableScanInfo->pResBlock);
   pTableScanInfo->pSortInputBlock = blockDataDestroy(pTableScanInfo->pSortInputBlock);
 
@@ -2305,8 +2408,9 @@ int32_t getTableMergeScanExplainExecInfo(SOperatorInfo* pOptr, void** pOptrExpla
   return TSDB_CODE_SUCCESS;
 }
 
-SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanNode, SArray* dataReaders,
+SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanNode, STableListInfo* pTableListInfo,
-                                                SReadHandle* readHandle, SExecTaskInfo* pTaskInfo) {
+                                                SReadHandle* readHandle, SExecTaskInfo* pTaskInfo, uint64_t queryId,
+                                                uint64_t taskId) {
   STableMergeScanInfo* pInfo = taosMemoryCalloc(1, sizeof(STableMergeScanInfo));
   SOperatorInfo*       pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
   if (pInfo == NULL || pOperator == NULL) {
@@ -2336,22 +2440,16 @@ SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanN
   pInfo->sample.seed = taosGetTimestampSec();
   pInfo->dataBlockLoadFlag = pTableScanNode->dataRequired;
   pInfo->pFilterNode = pTableScanNode->scan.node.pConditions;
-  pInfo->dataReaders = dataReaders;
+  pInfo->tableListInfo = pTableListInfo;
   pInfo->scanFlag = MAIN_SCAN;
   pInfo->pColMatchInfo = pColList;
-  pInfo->curTWinIdx = 0;
 
   pInfo->pResBlock = createResDataBlock(pDescNode);
+  pInfo->dataReaders = taosArrayInit(64, POINTER_BYTES);
+  pInfo->queryId = queryId;
+  pInfo->taskId = taskId;
 
-  pInfo->sortSourceParams = taosArrayInit(taosArrayGetSize(dataReaders), sizeof(STableMergeScanSortSourceParam));
+  pInfo->sortSourceParams = taosArrayInit(64, sizeof(STableMergeScanSortSourceParam));
-  for (int32_t i = 0; i < taosArrayGetSize(dataReaders); ++i) {
-    STableMergeScanSortSourceParam* param = taosMemoryCalloc(1, sizeof(STableMergeScanSortSourceParam));
-    param->readerIdx = i;
-    param->pOperator = pOperator;
-    param->inputBlock = createOneDataBlock(pInfo->pResBlock, false);
-    taosArrayPush(pInfo->sortSourceParams, param);
-    taosMemoryFree(param);
-  }
 
   pInfo->pSortInfo = generateSortByTsInfo(pInfo->cond.order);
   pInfo->pSortInputBlock = createOneDataBlock(pInfo->pResBlock, false);
@@ -2359,14 +2457,7 @@ SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanN
   int32_t rowSize = pInfo->pResBlock->info.rowSize;
   pInfo->bufPageSize = getProperSortPageSize(rowSize);
 
-  // todo the total available buffer should be determined by total capacity of buffer of this task.
-  // the additional one is reserved for merge result
-  pInfo->sortBufSize = pInfo->bufPageSize * (taosArrayGetSize(dataReaders) + 1);
-  pInfo->hasGroupId = false;
-  pInfo->prefetchedTuple = NULL;
-
   pOperator->name = "TableMergeScanOperator";
-  // TODO : change it
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN;
   pOperator->blocking = false;
   pOperator->status = OP_NOT_OPENED;
@@ -2376,8 +2467,8 @@ SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanN
   initResultSizeInfo(pOperator, 1024);
 
   pOperator->fpSet =
-      createOperatorFpSet(doOpenTableMergeScanOperator, doTableMergeScan, NULL, NULL, destroyTableMergeScanOperatorInfo,
+      createOperatorFpSet(operatorDummyOpenFn, doTableMergeScan, NULL, NULL, destroyTableMergeScanOperatorInfo, NULL,
-                          NULL, NULL, getTableMergeScanExplainExecInfo);
+                          NULL, getTableMergeScanExplainExecInfo);
   pOperator->cost.openCost = 0;
   return pOperator;
 

source/libs/executor/src/sortoperator.c

@@ -424,10 +424,17 @@ int32_t getGroupSortExplainExecInfo(SOperatorInfo* pOptr, void** pOptrExplain, u
   return TSDB_CODE_SUCCESS;
 }
 
-// TODO:
+void destroyGroupSortOperatorInfo(void* param, int32_t numOfOutput) {
-SOperatorInfo* createGroupSortOperatorInfo(SOperatorInfo* downstream, SSortPhysiNode* pSortPhyNode,
+  SGroupSortOperatorInfo* pInfo = (SGroupSortOperatorInfo*)param;
+  pInfo->binfo.pRes = blockDataDestroy(pInfo->binfo.pRes);
+
+  taosArrayDestroy(pInfo->pSortInfo);
+  taosArrayDestroy(pInfo->pColMatchInfo);
+}
+
+SOperatorInfo* createGroupSortOperatorInfo(SOperatorInfo* downstream, SGroupSortPhysiNode* pSortPhyNode,
                                            SExecTaskInfo* pTaskInfo) {
-  SSortOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SSortOperatorInfo));
+  SGroupSortOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SGroupSortOperatorInfo));
   SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
   if (pInfo == NULL || pOperator == NULL /* || rowSize > 100 * 1024 * 1024*/) {
     goto _error;
@@ -452,8 +459,7 @@ SOperatorInfo* createGroupSortOperatorInfo(SOperatorInfo* downstream, SSortPhysi
   ;
   pInfo->pColMatchInfo = pColMatchColInfo;
   pOperator->name = "GroupSortOperator";
-  // TODO
+  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_GROUP_SORT;
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_SORT;
   pOperator->blocking = true;
   pOperator->status = OP_NOT_OPENED;
   pOperator->info = pInfo;
@@ -461,7 +467,7 @@ SOperatorInfo* createGroupSortOperatorInfo(SOperatorInfo* downstream, SSortPhysi
   pOperator->exprSupp.numOfExprs = numOfCols;
   pOperator->pTaskInfo = pTaskInfo;
 
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doGroupSort, NULL, NULL, destroyOrderOperatorInfo, NULL,
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doGroupSort, NULL, NULL, destroyGroupSortOperatorInfo, NULL,
                                          NULL, getGroupSortExplainExecInfo);
 
   int32_t code = appendDownstream(pOperator, &downstream, 1);
@@ -478,18 +484,6 @@ _error:
   return NULL;
 }
 
-void destroyGroupSortOperatorInfo(void* param, int32_t numOfOutput) {
-  SGroupSortOperatorInfo* pInfo = (SGroupSortOperatorInfo*)param;
-  pInfo->binfo.pRes = blockDataDestroy(pInfo->binfo.pRes);
-
-  taosArrayDestroy(pInfo->pSortInfo);
-  taosArrayDestroy(pInfo->pColMatchInfo);
-}
-
-// TODO: sort group
-// TODO: msortCompare compare group id in multiway merge sort.
-// TODO: table merge scan, group first, then for each group, multiple readers
-
 //=====================================================================================
 // Multiway Sort Merge operator
 typedef struct SMultiwaySortMergeOperatorInfo {

source/libs/executor/src/timewindowoperator.c

@@ -1656,10 +1656,9 @@ static SSDataBlock* doSessionWindowAgg(SOperatorInfo* pOperator) {
     doBuildResultDatablock(pOperator, pBInfo, &pInfo->groupResInfo, pInfo->aggSup.pResultBuf);
     if (pBInfo->pRes->info.rows == 0 || !hasDataInGroupInfo(&pInfo->groupResInfo)) {
       doSetOperatorCompleted(pOperator);
-      return NULL;
     }
 
-    return pBInfo->pRes;
+    return pBInfo->pRes->info.rows > 0 ? pBInfo->pRes : NULL;
   }
 
   int64_t st = taosGetTimestampUs();
@@ -2239,7 +2238,8 @@ static void clearUpdateDataBlock(SSDataBlock* pBlock) {
 }
 
 void copyUpdateDataBlock(SSDataBlock* pDest, SSDataBlock* pSource, int32_t tsColIndex) {
-  ASSERT(pDest->info.capacity >= pSource->info.rows);
+  // ASSERT(pDest->info.capacity >= pSource->info.rows);
+  blockDataEnsureCapacity(pDest, pSource->info.rows);
   clearUpdateDataBlock(pDest);
   SColumnInfoData* pDestCol = taosArrayGet(pDest->pDataBlock, 0);
   SColumnInfoData* pSourceCol = taosArrayGet(pSource->pDataBlock, tsColIndex);

source/libs/nodes/src/nodesCloneFuncs.c

@@ -346,6 +346,7 @@ static int32_t logicScanCopy(const SScanLogicNode* pSrc, SScanLogicNode* pDst) {
   COPY_SCALAR_FIELD(intervalUnit);
   COPY_SCALAR_FIELD(slidingUnit);
   CLONE_NODE_FIELD(pTagCond);
+  CLONE_NODE_FIELD(pTagIndexCond);
   COPY_SCALAR_FIELD(triggerType);
   COPY_SCALAR_FIELD(watermark);
   COPY_SCALAR_FIELD(tsColId);

source/libs/nodes/src/nodesCodeFuncs.c

@@ -2298,6 +2298,7 @@ static const char* jkSubplanNodeAddr = "NodeAddr";
 static const char* jkSubplanRootNode = "RootNode";
 static const char* jkSubplanDataSink = "DataSink";
 static const char* jkSubplanTagCond = "TagCond";
+static const char* jkSubplanTagIndexCond = "TagIndexCond";
 
 static int32_t subplanToJson(const void* pObj, SJson* pJson) {
   const SSubplan* pNode = (const SSubplan*)pObj;
@@ -2327,6 +2328,9 @@ static int32_t subplanToJson(const void* pObj, SJson* pJson) {
   if (TSDB_CODE_SUCCESS == code) {
     code = tjsonAddObject(pJson, jkSubplanTagCond, nodeToJson, pNode->pTagCond);
   }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = tjsonAddObject(pJson, jkSubplanTagIndexCond, nodeToJson, pNode->pTagIndexCond);
+  }
 
   return code;
 }
@@ -2360,6 +2364,9 @@ static int32_t jsonToSubplan(const SJson* pJson, void* pObj) {
   if (TSDB_CODE_SUCCESS == code) {
     code = jsonToNodeObject(pJson, jkSubplanTagCond, (SNode**)&pNode->pTagCond);
   }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = jsonToNodeObject(pJson, jkSubplanTagIndexCond, (SNode**)&pNode->pTagIndexCond);
+  }
 
   return code;
 }
@@ -3951,7 +3958,7 @@ static int32_t deleteStmtToJson(const void* pObj, SJson* pJson) {
     code = tjsonAddObject(pJson, jkDeleteStmtCountFunc, nodeToJson, pNode->pCountFunc);
   }
   if (TSDB_CODE_SUCCESS == code) {
-    code = tjsonAddObject(pJson, jkDeleteStmtTagIndexCond, nodeToJson, pNode->pTagIndexCond);
+    code = tjsonAddObject(pJson, jkDeleteStmtTagIndexCond, nodeToJson, pNode->pTagCond);
   }
   if (TSDB_CODE_SUCCESS == code) {
     code = tjsonAddIntegerToObject(pJson, jkDeleteStmtTimeRangeStartKey, pNode->timeRange.skey);
@@ -3980,7 +3987,7 @@ static int32_t jsonToDeleteStmt(const SJson* pJson, void* pObj) {
     code = jsonToNodeObject(pJson, jkDeleteStmtCountFunc, &pNode->pCountFunc);
   }
   if (TSDB_CODE_SUCCESS == code) {
-    code = jsonToNodeObject(pJson, jkDeleteStmtTagIndexCond, &pNode->pTagIndexCond);
+    code = jsonToNodeObject(pJson, jkDeleteStmtTagIndexCond, &pNode->pTagCond);
   }
   if (TSDB_CODE_SUCCESS == code) {
     code = tjsonGetBigIntValue(pJson, jkDeleteStmtTimeRangeStartKey, &pNode->timeRange.skey);

source/libs/nodes/src/nodesUtilFuncs.c

@@ -671,7 +671,7 @@ void nodesDestroyNode(SNode* pNode) {
       nodesDestroyNode(pStmt->pFromTable);
       nodesDestroyNode(pStmt->pWhere);
       nodesDestroyNode(pStmt->pCountFunc);
-      nodesDestroyNode(pStmt->pTagIndexCond);
+      nodesDestroyNode(pStmt->pTagCond);
       break;
     }
     case QUERY_NODE_QUERY: {
@@ -690,7 +690,13 @@ void nodesDestroyNode(SNode* pNode) {
       SScanLogicNode* pLogicNode = (SScanLogicNode*)pNode;
       destroyLogicNode((SLogicNode*)pLogicNode);
       nodesDestroyList(pLogicNode->pScanCols);
+      nodesDestroyList(pLogicNode->pScanPseudoCols);
       taosMemoryFreeClear(pLogicNode->pVgroupList);
+      nodesDestroyList(pLogicNode->pDynamicScanFuncs);
+      nodesDestroyNode(pLogicNode->pTagCond);
+      nodesDestroyNode(pLogicNode->pTagIndexCond);
+      taosArrayDestroy(pLogicNode->pSmaIndexes);
+      nodesDestroyList(pLogicNode->pPartTags);
       break;
     }
     case QUERY_NODE_LOGIC_PLAN_JOIN: {
@@ -899,6 +905,8 @@ void nodesDestroyNode(SNode* pNode) {
       nodesDestroyList(pSubplan->pChildren);
       nodesDestroyNode((SNode*)pSubplan->pNode);
       nodesDestroyNode((SNode*)pSubplan->pDataSink);
+      nodesDestroyNode((SNode*)pSubplan->pTagCond);
+      nodesDestroyNode((SNode*)pSubplan->pTagIndexCond);
       nodesClearList(pSubplan->pParents);
       break;
     }
@@ -1132,6 +1140,7 @@ void* nodesGetValueFromNode(SValueNode* pNode) {
     case TSDB_DATA_TYPE_NCHAR:
     case TSDB_DATA_TYPE_VARCHAR:
     case TSDB_DATA_TYPE_VARBINARY:
+    case TSDB_DATA_TYPE_JSON:
       return (void*)pNode->datum.p;
     default:
       break;
@@ -1661,6 +1670,7 @@ int32_t nodesMergeConds(SNode** pDst, SNodeList** pSrc) {
 typedef struct SClassifyConditionCxt {
   bool hasPrimaryKey;
   bool hasTagIndexCol;
+  bool hasTagCol;
   bool hasOtherCol;
 } SClassifyConditionCxt;
 
@@ -1672,6 +1682,9 @@ static EDealRes classifyConditionImpl(SNode* pNode, void* pContext) {
       pCxt->hasPrimaryKey = true;
     } else if (pCol->hasIndex) {
       pCxt->hasTagIndexCol = true;
+      pCxt->hasTagCol = true;
+    } else if (COLUMN_TYPE_TAG == pCol->colType) {
+      pCxt->hasTagCol = true;
     } else {
       pCxt->hasOtherCol = true;
     }
@@ -1680,23 +1693,31 @@ static EDealRes classifyConditionImpl(SNode* pNode, void* pContext) {
   return DEAL_RES_CONTINUE;
 }
 
-typedef enum EConditionType { COND_TYPE_PRIMARY_KEY = 1, COND_TYPE_TAG_INDEX, COND_TYPE_NORMAL } EConditionType;
+typedef enum EConditionType {
+  COND_TYPE_PRIMARY_KEY = 1,
+  COND_TYPE_TAG_INDEX,
+  COND_TYPE_TAG,
+  COND_TYPE_NORMAL
+} EConditionType;
 
 static EConditionType classifyCondition(SNode* pNode) {
   SClassifyConditionCxt cxt = {.hasPrimaryKey = false, .hasTagIndexCol = false, .hasOtherCol = false};
   nodesWalkExpr(pNode, classifyConditionImpl, &cxt);
   return cxt.hasOtherCol ? COND_TYPE_NORMAL
-                         : (cxt.hasPrimaryKey && cxt.hasTagIndexCol
+                         : (cxt.hasPrimaryKey && cxt.hasTagCol
                                 ? COND_TYPE_NORMAL
-                                : (cxt.hasPrimaryKey ? COND_TYPE_PRIMARY_KEY : COND_TYPE_TAG_INDEX));
+                                : (cxt.hasPrimaryKey ? COND_TYPE_PRIMARY_KEY
+                                                     : (cxt.hasTagIndexCol ? COND_TYPE_TAG_INDEX : COND_TYPE_TAG)));
 }
 
-static int32_t partitionLogicCond(SNode** pCondition, SNode** pPrimaryKeyCond, SNode** pTagCond, SNode** pOtherCond) {
+static int32_t partitionLogicCond(SNode** pCondition, SNode** pPrimaryKeyCond, SNode** pTagIndexCond, SNode** pTagCond,
+                                  SNode** pOtherCond) {
   SLogicConditionNode* pLogicCond = (SLogicConditionNode*)(*pCondition);
 
   int32_t code = TSDB_CODE_SUCCESS;
 
   SNodeList* pPrimaryKeyConds = NULL;
+  SNodeList* pTagIndexConds = NULL;
   SNodeList* pTagConds = NULL;
   SNodeList* pOtherConds = NULL;
   SNode*     pCond = NULL;
@@ -1708,6 +1729,14 @@ static int32_t partitionLogicCond(SNode** pCondition, SNode** pPrimaryKeyCond, S
         }
         break;
       case COND_TYPE_TAG_INDEX:
+        if (NULL != pTagIndexCond) {
+          code = nodesListMakeAppend(&pTagIndexConds, nodesCloneNode(pCond));
+        }
+        if (NULL != pTagCond) {
+          code = nodesListMakeAppend(&pTagConds, nodesCloneNode(pCond));
+        }
+        break;
+      case COND_TYPE_TAG:
         if (NULL != pTagCond) {
           code = nodesListMakeAppend(&pTagConds, nodesCloneNode(pCond));
         }
@@ -1725,11 +1754,15 @@ static int32_t partitionLogicCond(SNode** pCondition, SNode** pPrimaryKeyCond, S
   }
 
   SNode* pTempPrimaryKeyCond = NULL;
+  SNode* pTempTagIndexCond = NULL;
   SNode* pTempTagCond = NULL;
   SNode* pTempOtherCond = NULL;
   if (TSDB_CODE_SUCCESS == code) {
     code = nodesMergeConds(&pTempPrimaryKeyCond, &pPrimaryKeyConds);
   }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = nodesMergeConds(&pTempTagIndexCond, &pTagIndexConds);
+  }
   if (TSDB_CODE_SUCCESS == code) {
     code = nodesMergeConds(&pTempTagCond, &pTagConds);
   }
@@ -1741,6 +1774,9 @@ static int32_t partitionLogicCond(SNode** pCondition, SNode** pPrimaryKeyCond, S
     if (NULL != pPrimaryKeyCond) {
       *pPrimaryKeyCond = pTempPrimaryKeyCond;
     }
+    if (NULL != pTagIndexCond) {
+      *pTagIndexCond = pTempTagIndexCond;
+    }
     if (NULL != pTagCond) {
       *pTagCond = pTempTagCond;
     }
@@ -1751,9 +1787,11 @@ static int32_t partitionLogicCond(SNode** pCondition, SNode** pPrimaryKeyCond, S
     *pCondition = NULL;
   } else {
     nodesDestroyList(pPrimaryKeyConds);
+    nodesDestroyList(pTagIndexConds);
     nodesDestroyList(pTagConds);
     nodesDestroyList(pOtherConds);
     nodesDestroyNode(pTempPrimaryKeyCond);
+    nodesDestroyNode(pTempTagIndexCond);
     nodesDestroyNode(pTempTagCond);
     nodesDestroyNode(pTempOtherCond);
   }
@@ -1761,10 +1799,11 @@ static int32_t partitionLogicCond(SNode** pCondition, SNode** pPrimaryKeyCond, S
   return code;
 }
 
-int32_t nodesPartitionCond(SNode** pCondition, SNode** pPrimaryKeyCond, SNode** pTagCond, SNode** pOtherCond) {
+int32_t nodesPartitionCond(SNode** pCondition, SNode** pPrimaryKeyCond, SNode** pTagIndexCond, SNode** pTagCond,
+                           SNode** pOtherCond) {
   if (QUERY_NODE_LOGIC_CONDITION == nodeType(*pCondition) &&
       LOGIC_COND_TYPE_AND == ((SLogicConditionNode*)*pCondition)->condType) {
-    return partitionLogicCond(pCondition, pPrimaryKeyCond, pTagCond, pOtherCond);
+    return partitionLogicCond(pCondition, pPrimaryKeyCond, pTagIndexCond, pTagCond, pOtherCond);
   }
 
   switch (classifyCondition(*pCondition)) {
@@ -1774,6 +1813,21 @@ int32_t nodesPartitionCond(SNode** pCondition, SNode** pPrimaryKeyCond, SNode**
       }
       break;
     case COND_TYPE_TAG_INDEX:
+      if (NULL != pTagIndexCond) {
+        *pTagIndexCond = *pCondition;
+      }
+      if (NULL != pTagCond) {
+        SNode* pTempCond = *pCondition;
+        if (NULL != pTagIndexCond) {
+          pTempCond = nodesCloneNode(*pCondition);
+          if (NULL == pTempCond) {
+            return TSDB_CODE_OUT_OF_MEMORY;
+          }
+        }
+        *pTagCond = pTempCond;
+      }
+      break;
+    case COND_TYPE_TAG:
       if (NULL != pTagCond) {
         *pTagCond = *pCondition;
       }

source/libs/parser/src/parAstParser.c

@@ -137,6 +137,9 @@ static int32_t collectMetaKeyFromRealTableImpl(SCollectMetaKeyCxt* pCxt, SRealTa
     code = reserveTableIndexInCache(pCxt->pParseCxt->acctId, pRealTable->table.dbName, pRealTable->table.tableName,
                                     pCxt->pMetaCache);
   }
+  if (TSDB_CODE_SUCCESS == code && (0 == strcmp(pRealTable->table.tableName, TSDB_INS_TABLE_DNODE_VARIABLES))) {
+    code = reserveDnodeRequiredInCache(pCxt->pMetaCache);
+  }
   return code;
 }
 

source/libs/parser/src/parTranslater.c

@@ -2017,7 +2017,7 @@ static int32_t getFillTimeRange(STranslateContext* pCxt, SNode* pWhere, STimeWin
   }
 
   SNode* pPrimaryKeyCond = NULL;
-  nodesPartitionCond(&pCond, &pPrimaryKeyCond, NULL, NULL);
+  nodesPartitionCond(&pCond, &pPrimaryKeyCond, NULL, NULL, NULL);
 
   int32_t code = TSDB_CODE_SUCCESS;
   if (NULL != pPrimaryKeyCond) {
@@ -2165,9 +2165,6 @@ static EDealRes checkStateExpr(SNode* pNode, void* pContext) {
     if (COLUMN_TYPE_TAG == pCol->colType) {
       return generateDealNodeErrMsg(pCxt, TSDB_CODE_PAR_INVALID_STATE_WIN_COL);
     }
-    if (TSDB_SUPER_TABLE == pCol->tableType) {
-      return generateDealNodeErrMsg(pCxt, TSDB_CODE_PAR_INVALID_STATE_WIN_TABLE);
-    }
   }
   return DEAL_RES_CONTINUE;
 }
@@ -2699,7 +2696,7 @@ static int32_t partitionDeleteWhere(STranslateContext* pCxt, SDeleteStmt* pDelet
 
   SNode*  pPrimaryKeyCond = NULL;
   SNode*  pOtherCond = NULL;
-  int32_t code = nodesPartitionCond(&pDelete->pWhere, &pPrimaryKeyCond, &pDelete->pTagIndexCond, &pOtherCond);
+  int32_t code = nodesPartitionCond(&pDelete->pWhere, &pPrimaryKeyCond, NULL, &pDelete->pTagCond, &pOtherCond);
   if (TSDB_CODE_SUCCESS == code && NULL != pOtherCond) {
     code = generateSyntaxErrMsg(&pCxt->msgBuf, TSDB_CODE_PAR_INVALID_DELETE_WHERE);
   }
@@ -4643,6 +4640,25 @@ static int32_t extractShowCreateTableResultSchema(int32_t* numOfCols, SSchema**
   return TSDB_CODE_SUCCESS;
 }
 
+static int32_t extractShowLocalVariablesResultSchema(int32_t* numOfCols, SSchema** pSchema) {
+  *numOfCols = 2;
+  *pSchema = taosMemoryCalloc((*numOfCols), sizeof(SSchema));
+  if (NULL == (*pSchema)) {
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+
+  (*pSchema)[0].type = TSDB_DATA_TYPE_BINARY;
+  (*pSchema)[0].bytes = TSDB_CONFIG_OPTION_LEN;
+  strcpy((*pSchema)[0].name, "name");
+
+  (*pSchema)[1].type = TSDB_DATA_TYPE_BINARY;
+  (*pSchema)[1].bytes = TSDB_CONFIG_VALUE_LEN;
+  strcpy((*pSchema)[1].name, "value");
+
+  return TSDB_CODE_SUCCESS;
+}
+
+
 int32_t extractResultSchema(const SNode* pRoot, int32_t* numOfCols, SSchema** pSchema) {
   if (NULL == pRoot) {
     return TSDB_CODE_SUCCESS;
@@ -4661,6 +4677,8 @@ int32_t extractResultSchema(const SNode* pRoot, int32_t* numOfCols, SSchema** pS
     case QUERY_NODE_SHOW_CREATE_TABLE_STMT:
     case QUERY_NODE_SHOW_CREATE_STABLE_STMT:
       return extractShowCreateTableResultSchema(numOfCols, pSchema);
+    case QUERY_NODE_SHOW_LOCAL_VARIABLES_STMT:
+      return extractShowLocalVariablesResultSchema(numOfCols, pSchema);
     default:
       break;
   }
@@ -5985,12 +6003,12 @@ static int32_t setQuery(STranslateContext* pCxt, SQuery* pQuery) {
     case QUERY_NODE_SHOW_CREATE_DATABASE_STMT:
     case QUERY_NODE_SHOW_CREATE_TABLE_STMT:
     case QUERY_NODE_SHOW_CREATE_STABLE_STMT:
+    case QUERY_NODE_SHOW_LOCAL_VARIABLES_STMT:
       pQuery->execMode = QUERY_EXEC_MODE_LOCAL;
       pQuery->haveResultSet = true;
       break;
     case QUERY_NODE_RESET_QUERY_CACHE_STMT:
     case QUERY_NODE_ALTER_LOCAL_STMT:
-    case QUERY_NODE_SHOW_LOCAL_VARIABLES_STMT:
       pQuery->execMode = QUERY_EXEC_MODE_LOCAL;
       break;
     default:

source/libs/parser/src/parUtil.c

@@ -928,7 +928,12 @@ int32_t reserveDnodeRequiredInCache(SParseMetaCache* pMetaCache) {
 }
 
 int32_t getDnodeListFromCache(SParseMetaCache* pMetaCache, SArray** pDnodes) {
-  *pDnodes = taosArrayDup(pMetaCache->pDnodes);
+  SMetaRes* pRes = taosArrayGet(pMetaCache->pDnodes, 0);
+  if (pRes->code) {
+    return pRes->code;
+  }
+  
+  *pDnodes = taosArrayDup((SArray*)pRes->pRes);
   if (NULL == *pDnodes) {
     return TSDB_CODE_OUT_OF_MEMORY;
   }

source/libs/parser/test/mockCatalogService.cpp

@@ -166,10 +166,13 @@ class MockCatalogServiceImpl {
   }
 
   int32_t catalogGetDnodeList(SArray** pDnodes) const {
-    *pDnodes = taosArrayInit(dnode_.size(), sizeof(SEpSet));
+    SMetaRes res = {0};
+    res.pRes = taosArrayInit(dnode_.size(), sizeof(SEpSet));
     for (const auto& dnode : dnode_) {
-      taosArrayPush(*pDnodes, &dnode.second);
+      taosArrayPush((SArray*)res.pRes, &dnode.second);
     }
+    *pDnodes = taosArrayInit(1, sizeof(SMetaRes));
+    taosArrayPush(*pDnodes, &res);
     return TSDB_CODE_SUCCESS;
   }
 

source/libs/planner/src/planLogicCreater.c

@@ -1133,8 +1133,8 @@ static int32_t createDeleteScanLogicNode(SLogicPlanContext* pCxt, SDeleteStmt* p
     }
   }
 
-  if (TSDB_CODE_SUCCESS == code && NULL != pDelete->pTagIndexCond) {
+  if (TSDB_CODE_SUCCESS == code && NULL != pDelete->pTagCond) {
-    pScan->pTagCond = nodesCloneNode(pDelete->pTagIndexCond);
+    pScan->pTagCond = nodesCloneNode(pDelete->pTagCond);
     if (NULL == pScan->pTagCond) {
       code = TSDB_CODE_OUT_OF_MEMORY;
     }

source/libs/planner/src/planOptimizer.c

@@ -15,7 +15,6 @@
 
 #include "filter.h"
 #include "functionMgt.h"
-#include "index.h"
 #include "planInt.h"
 #include "ttime.h"
 
@@ -309,32 +308,6 @@ static int32_t cpdCalcTimeRange(SOptimizeContext* pCxt, SScanLogicNode* pScan, S
   return code;
 }
 
-static int32_t cpdApplyTagIndex(SScanLogicNode* pScan, SNode** pTagCond, SNode** pOtherCond) {
-  int32_t       code = TSDB_CODE_SUCCESS;
-  SIdxFltStatus idxStatus = idxGetFltStatus(*pTagCond);
-  switch (idxStatus) {
-    case SFLT_NOT_INDEX:
-      code = cpdCondAppend(pOtherCond, pTagCond);
-      break;
-    case SFLT_COARSE_INDEX:
-      pScan->pTagCond = nodesCloneNode(*pTagCond);
-      if (NULL == pScan->pTagCond) {
-        code = TSDB_CODE_OUT_OF_MEMORY;
-        break;
-      }
-      code = cpdCondAppend(pOtherCond, pTagCond);
-      break;
-    case SFLT_ACCURATE_INDEX:
-      pScan->pTagCond = *pTagCond;
-      *pTagCond = NULL;
-      break;
-    default:
-      code = TSDB_CODE_FAILED;
-      break;
-  }
-  return code;
-}
-
 static int32_t cpdOptimizeScanCondition(SOptimizeContext* pCxt, SScanLogicNode* pScan) {
   if (NULL == pScan->node.pConditions || OPTIMIZE_FLAG_TEST_MASK(pScan->node.optimizedFlag, OPTIMIZE_FLAG_CPD) ||
       TSDB_SYSTEM_TABLE == pScan->tableType) {
@@ -342,15 +315,12 @@ static int32_t cpdOptimizeScanCondition(SOptimizeContext* pCxt, SScanLogicNode*
   }
 
   SNode*  pPrimaryKeyCond = NULL;
-  SNode*  pTagCond = NULL;
   SNode*  pOtherCond = NULL;
-  int32_t code = nodesPartitionCond(&pScan->node.pConditions, &pPrimaryKeyCond, &pTagCond, &pOtherCond);
+  int32_t code = nodesPartitionCond(&pScan->node.pConditions, &pPrimaryKeyCond, &pScan->pTagIndexCond, &pScan->pTagCond,
+                                    &pOtherCond);
   if (TSDB_CODE_SUCCESS == code && NULL != pPrimaryKeyCond) {
     code = cpdCalcTimeRange(pCxt, pScan, &pPrimaryKeyCond, &pOtherCond);
   }
-  if (TSDB_CODE_SUCCESS == code && NULL != pTagCond) {
-    code = cpdApplyTagIndex(pScan, &pTagCond, &pOtherCond);
-  }
   if (TSDB_CODE_SUCCESS == code) {
     pScan->node.pConditions = pOtherCond;
   }

source/libs/planner/src/planPhysiCreater.c

@@ -439,6 +439,15 @@ static int32_t createScanPhysiNodeFinalize(SPhysiPlanContext* pCxt, SSubplan* pS
     }
   }
 
+  if (TSDB_CODE_SUCCESS == code) {
+    if (NULL != pScanLogicNode->pTagIndexCond) {
+      pSubplan->pTagIndexCond = nodesCloneNode(pScanLogicNode->pTagIndexCond);
+      if (NULL == pSubplan->pTagIndexCond) {
+        code = TSDB_CODE_OUT_OF_MEMORY;
+      }
+    }
+  }
+
   if (TSDB_CODE_SUCCESS == code) {
     *pPhyNode = (SPhysiNode*)pScanPhysiNode;
   } else {

source/libs/planner/src/planSpliter.c

@@ -166,6 +166,31 @@ static bool stbSplHasMultiTbScan(bool streamQuery, SLogicNode* pNode) {
   return (QUERY_NODE_LOGIC_PLAN_SCAN == nodeType(pChild) && stbSplIsMultiTbScan(streamQuery, (SScanLogicNode*)pChild));
 }
 
+static bool stbSplNeedSplitWindow(bool streamQuery, SLogicNode* pNode) {
+  SWindowLogicNode* pWindow = (SWindowLogicNode*)pNode;
+  if (WINDOW_TYPE_INTERVAL == pWindow->winType) {
+    return !stbSplHasGatherExecFunc(pWindow->pFuncs) && stbSplHasMultiTbScan(streamQuery, pNode);
+  }
+
+  if (WINDOW_TYPE_SESSION == pWindow->winType) {
+    if (!streamQuery) {
+      return stbSplHasMultiTbScan(streamQuery, pNode);
+    } else {
+      return !stbSplHasGatherExecFunc(pWindow->pFuncs) && stbSplHasMultiTbScan(streamQuery, pNode);
+    }
+  }
+  
+  if (WINDOW_TYPE_STATE == pWindow->winType) {
+    if (!streamQuery) {
+      return stbSplHasMultiTbScan(streamQuery, pNode);
+    } else {
+      return false;
+    }
+  }
+
+  return false;
+}
+
 static bool stbSplNeedSplit(bool streamQuery, SLogicNode* pNode) {
   switch (nodeType(pNode)) {
     case QUERY_NODE_LOGIC_PLAN_SCAN:
@@ -174,13 +199,8 @@ static bool stbSplNeedSplit(bool streamQuery, SLogicNode* pNode) {
       return !(((SJoinLogicNode*)pNode)->isSingleTableJoin);
     case QUERY_NODE_LOGIC_PLAN_AGG:
       return !stbSplHasGatherExecFunc(((SAggLogicNode*)pNode)->pAggFuncs) && stbSplHasMultiTbScan(streamQuery, pNode);
-    case QUERY_NODE_LOGIC_PLAN_WINDOW: {
+    case QUERY_NODE_LOGIC_PLAN_WINDOW:
-      SWindowLogicNode* pWindow = (SWindowLogicNode*)pNode;
+      return stbSplNeedSplitWindow(streamQuery, pNode);
-      if (WINDOW_TYPE_STATE == pWindow->winType || (!streamQuery && WINDOW_TYPE_SESSION == pWindow->winType)) {
-        return false;
-      }
-      return !stbSplHasGatherExecFunc(pWindow->pFuncs) && stbSplHasMultiTbScan(streamQuery, pNode);
-    }
     case QUERY_NODE_LOGIC_PLAN_SORT:
       return stbSplHasMultiTbScan(streamQuery, pNode);
     default:
@@ -477,11 +497,64 @@ static int32_t stbSplSplitSessionForStream(SSplitContext* pCxt, SStableSplitInfo
   return code;
 }
 
+static void splSetTableScanType(SLogicNode* pNode, EScanType scanType) {
+  if (QUERY_NODE_LOGIC_PLAN_SCAN == nodeType(pNode)) {
+    ((SScanLogicNode*)pNode)->scanType = scanType;
+  } else {
+    if (1 == LIST_LENGTH(pNode->pChildren)) {
+      splSetTableScanType((SLogicNode*)nodesListGetNode(pNode->pChildren, 0), scanType);
+    }
+  }
+}
+
+static int32_t stbSplSplitSessionOrStateForBatch(SSplitContext* pCxt, SStableSplitInfo* pInfo) {
+  SLogicNode* pWindow = pInfo->pSplitNode;
+  SLogicNode* pChild = (SLogicNode*)nodesListGetNode(pWindow->pChildren, 0);
+
+  SNodeList* pMergeKeys = NULL;
+  int32_t code = stbSplCreateMergeKeysByPrimaryKey(((SWindowLogicNode*)pWindow)->pTspk, &pMergeKeys);
+
+  if (TSDB_CODE_SUCCESS == code) {
+    code = stbSplCreateMergeNode(pCxt, pInfo->pSubplan, pChild, pMergeKeys, (SLogicNode*)pChild);
+  }
+
+  if (TSDB_CODE_SUCCESS == code) {
+    code = nodesListMakeStrictAppend(&pInfo->pSubplan->pChildren,
+                                     (SNode*)splCreateScanSubplan(pCxt, pChild, SPLIT_FLAG_STABLE_SPLIT));
+  }
+
+  if (TSDB_CODE_SUCCESS == code) {
+    splSetTableScanType(pChild, SCAN_TYPE_TABLE_MERGE);
+    ++(pCxt->groupId);
+  }
+
+  if (TSDB_CODE_SUCCESS == code) {
+    pInfo->pSubplan->subplanType = SUBPLAN_TYPE_MERGE;
+    SPLIT_FLAG_SET_MASK(pInfo->pSubplan->splitFlag, SPLIT_FLAG_STABLE_SPLIT);
+  } else {
+    nodesDestroyList(pMergeKeys);
+  }
+
+  return code;
+}
+
 static int32_t stbSplSplitSession(SSplitContext* pCxt, SStableSplitInfo* pInfo) {
   if (pCxt->pPlanCxt->streamQuery) {
     return stbSplSplitSessionForStream(pCxt, pInfo);
   } else {
+    return stbSplSplitSessionOrStateForBatch(pCxt, pInfo);
+  }
+}
+
+static int32_t stbSplSplitStateForStream(SSplitContext* pCxt, SStableSplitInfo* pInfo) {
   return TSDB_CODE_PLAN_INTERNAL_ERROR;
+}
+
+static int32_t stbSplSplitState(SSplitContext* pCxt, SStableSplitInfo* pInfo) {
+  if (pCxt->pPlanCxt->streamQuery) {
+    return stbSplSplitStateForStream(pCxt, pInfo);
+  } else {
+    return stbSplSplitSessionOrStateForBatch(pCxt, pInfo);
   }
 }
 
@@ -511,6 +584,8 @@ static int32_t stbSplSplitWindowForMergeTable(SSplitContext* pCxt, SStableSplitI
       return stbSplSplitInterval(pCxt, pInfo);
     case WINDOW_TYPE_SESSION:
       return stbSplSplitSession(pCxt, pInfo);
+    case WINDOW_TYPE_STATE:
+      return stbSplSplitState(pCxt, pInfo);
     default:
       break;
   }

source/libs/planner/test/planOptimizeTest.cpp

@@ -40,6 +40,8 @@ TEST_F(PlanOptimizeTest, ConditionPushDown) {
   run("SELECT ts, c1 FROM st1 WHERE tag1 > 4 or tag1 < 2");
 
   run("SELECT ts, c1 FROM st1 WHERE tag1 > 4 AND tag2 = 'hello'");
+
+  run("SELECT ts, c1 FROM st1 WHERE tag1 > 4 AND tag2 = 'hello' AND c1 > 10");
 }
 
 TEST_F(PlanOptimizeTest, orderByPrimaryKey) {

source/libs/planner/test/planSessionTest.cpp

@@ -34,3 +34,13 @@ TEST_F(PlanSessionTest, selectFunc) {
   // select function along with the columns of select row, and with SESSION clause
   run("SELECT MAX(c1), c2 FROM t1 SESSION(ts, 10s)");
 }
+
+TEST_F(PlanSessionTest, stable) {
+  useDb("root", "test");
+
+  // select function for SESSION clause
+  run("SELECT MAX(c1), MIN(c1) FROM st1 SESSION(ts, 10s)");
+  // select function along with the columns of select row, and with SESSION clause
+  run("SELECT MAX(c1), c2 FROM st1 SESSION(ts, 10s)");
+  run("SELECT count(ts) FROM st1 PARTITION BY c1 SESSION(ts, 10s)");
+}

source/libs/planner/test/planStateTest.cpp

@@ -40,3 +40,12 @@ TEST_F(PlanStateTest, selectFunc) {
   // select function along with the columns of select row, and with STATE_WINDOW clause
   run("SELECT MAX(c1), c2 FROM t1 STATE_WINDOW(c3)");
 }
+
+TEST_F(PlanStateTest, stable) {
+  useDb("root", "test");
+
+  // select function for STATE_WINDOW clause
+  run("SELECT MAX(c1), MIN(c1) FROM st1 STATE_WINDOW(c2)");
+  // select function along with the columns of select row, and with STATE_WINDOW clause
+  run("SELECT MAX(c1), c2 FROM st1 STATE_WINDOW(c2)");
+}

source/libs/qcom/src/querymsg.c

@@ -126,6 +126,25 @@ int32_t queryBuildQnodeListMsg(void *input, char **msg, int32_t msgSize, int32_t
   return TSDB_CODE_SUCCESS;
 }
 
+int32_t queryBuildDnodeListMsg(void *input, char **msg, int32_t msgSize, int32_t *msgLen, void*(*mallcFp)(int32_t)) {
+  if (NULL == msg || NULL == msgLen) {
+    return TSDB_CODE_TSC_INVALID_INPUT;
+  }
+
+  SDnodeListReq dnodeListReq = {0};
+  dnodeListReq.rowNum = -1;
+
+  int32_t bufLen = tSerializeSDnodeListReq(NULL, 0, &dnodeListReq);
+  void   *pBuf = (*mallcFp)(bufLen);
+  tSerializeSDnodeListReq(pBuf, bufLen, &dnodeListReq);
+
+  *msg = pBuf;
+  *msgLen = bufLen;
+
+  return TSDB_CODE_SUCCESS;
+}
+
+
 int32_t queryBuildGetDBCfgMsg(void *input, char **msg, int32_t msgSize, int32_t *msgLen, void*(*mallcFp)(int32_t)) {
   if (NULL == msg || NULL == msgLen) {
     return TSDB_CODE_TSC_INVALID_INPUT;
@@ -428,6 +447,27 @@ int32_t queryProcessQnodeListRsp(void *output, char *msg, int32_t msgSize) {
   return code;
 }
 
+int32_t queryProcessDnodeListRsp(void *output, char *msg, int32_t msgSize) {
+  SDnodeListRsp out = {0};
+  int32_t       code = 0;
+
+  if (NULL == output || NULL == msg || msgSize <= 0) {
+    code = TSDB_CODE_TSC_INVALID_INPUT;
+    return code;
+  }
+
+  if (tDeserializeSDnodeListRsp(msg, msgSize, &out) != 0) {
+    qError("invalid dnode list rsp msg, msgSize:%d", msgSize);
+    code = TSDB_CODE_INVALID_MSG;
+    return code;
+  }
+
+  *(SArray**)output = out.dnodeList;
+
+  return code;
+}
+
+
 int32_t queryProcessGetDbCfgRsp(void *output, char *msg, int32_t msgSize) {
   SDbCfgRsp out = {0};
 
@@ -535,6 +575,7 @@ void initQueryModuleMsgHandle() {
   queryBuildMsg[TMSG_INDEX(TDMT_MND_TABLE_META)]       = queryBuildTableMetaReqMsg;
   queryBuildMsg[TMSG_INDEX(TDMT_MND_USE_DB)]           = queryBuildUseDbMsg;
   queryBuildMsg[TMSG_INDEX(TDMT_MND_QNODE_LIST)]       = queryBuildQnodeListMsg;
+  queryBuildMsg[TMSG_INDEX(TDMT_MND_DNODE_LIST)]       = queryBuildDnodeListMsg;
   queryBuildMsg[TMSG_INDEX(TDMT_MND_GET_DB_CFG)]       = queryBuildGetDBCfgMsg;
   queryBuildMsg[TMSG_INDEX(TDMT_MND_GET_INDEX)]        = queryBuildGetIndexMsg;
   queryBuildMsg[TMSG_INDEX(TDMT_MND_RETRIEVE_FUNC)]    = queryBuildRetrieveFuncMsg;
@@ -547,6 +588,7 @@ void initQueryModuleMsgHandle() {
   queryProcessMsgRsp[TMSG_INDEX(TDMT_MND_TABLE_META)]      = queryProcessTableMetaRsp;
   queryProcessMsgRsp[TMSG_INDEX(TDMT_MND_USE_DB)]          = queryProcessUseDBRsp;
   queryProcessMsgRsp[TMSG_INDEX(TDMT_MND_QNODE_LIST)]      = queryProcessQnodeListRsp;
+  queryProcessMsgRsp[TMSG_INDEX(TDMT_MND_DNODE_LIST)]      = queryProcessDnodeListRsp;
   queryProcessMsgRsp[TMSG_INDEX(TDMT_MND_GET_DB_CFG)]      = queryProcessGetDbCfgRsp;
   queryProcessMsgRsp[TMSG_INDEX(TDMT_MND_GET_INDEX)]       = queryProcessGetIndexRsp;
   queryProcessMsgRsp[TMSG_INDEX(TDMT_MND_RETRIEVE_FUNC)]   = queryProcessRetrieveFuncRsp;

source/libs/qworker/src/qworker.c

@@ -271,7 +271,7 @@ int32_t qwGetDeleteResFromSink(QW_FPARAMS_DEF, SQWTaskCtx *ctx, int32_t *dataLen
   SDeleterRes* pDelRes = (SDeleterRes*)output.pData;
   
   rsp.affectedRows = pDelRes->affectedRows;
-  pRes->uid = pDelRes->uid;
+  pRes->suid = pDelRes->suid;
   pRes->uidList = pDelRes->uidList;
   pRes->skey = pDelRes->skey;
   pRes->ekey = pDelRes->ekey;

source/libs/scalar/src/sclvector.c

@@ -23,6 +23,7 @@
 #include "sclvector.h"
 #include "tcompare.h"
 #include "tdatablock.h"
+#include "tdataformat.h"
 #include "ttypes.h"
 #include "ttime.h"
 
@@ -506,6 +507,16 @@ bool convertJsonValue(__compar_fn_t *fp, int32_t optr, int8_t typeLeft, int8_t t
     }
   }
 
+  // if types can not comparable
+  if((IS_NUMERIC_TYPE(typeLeft) && !IS_NUMERIC_TYPE(typeRight)) ||
+     (IS_NUMERIC_TYPE(typeRight) && !IS_NUMERIC_TYPE(typeLeft)) ||
+     (IS_VAR_DATA_TYPE(typeLeft) && !IS_VAR_DATA_TYPE(typeRight)) ||
+     (IS_VAR_DATA_TYPE(typeRight) && !IS_VAR_DATA_TYPE(typeLeft)) ||
+     ((typeLeft == TSDB_DATA_TYPE_BOOL) && (typeRight != TSDB_DATA_TYPE_BOOL)) ||
+     ((typeRight == TSDB_DATA_TYPE_BOOL) && (typeLeft != TSDB_DATA_TYPE_BOOL)))
+    return false;
+
+
   if(typeLeft == TSDB_DATA_TYPE_NULL || typeRight == TSDB_DATA_TYPE_NULL){
     *isNull = true;
     return true;
@@ -519,24 +530,28 @@ bool convertJsonValue(__compar_fn_t *fp, int32_t optr, int8_t typeLeft, int8_t t
 
   *fp = filterGetCompFunc(type, optr);
 
-  if(IS_NUMERIC_TYPE(type) || IS_FLOAT_TYPE(type)){
+  if(IS_NUMERIC_TYPE(type)){
     if(typeLeft == TSDB_DATA_TYPE_NCHAR) {
-      convertNcharToDouble(*pLeftData, pLeftOut);
+      ASSERT(0);
-      *pLeftData = pLeftOut;
+//      convertNcharToDouble(*pLeftData, pLeftOut);
+//      *pLeftData = pLeftOut;
     } else if(typeLeft == TSDB_DATA_TYPE_BINARY) {
-      convertBinaryToDouble(*pLeftData, pLeftOut);
+      ASSERT(0);
-      *pLeftData = pLeftOut;
+//      convertBinaryToDouble(*pLeftData, pLeftOut);
+//      *pLeftData = pLeftOut;
     } else if(typeLeft != type) {
       convertNumberToNumber(*pLeftData, pLeftOut, typeLeft, type);
       *pLeftData = pLeftOut;
     }
 
     if(typeRight == TSDB_DATA_TYPE_NCHAR) {
-      convertNcharToDouble(*pRightData, pRightOut);
+      ASSERT(0);
-      *pRightData = pRightOut;
+//      convertNcharToDouble(*pRightData, pRightOut);
+//      *pRightData = pRightOut;
     } else if(typeRight == TSDB_DATA_TYPE_BINARY) {
-      convertBinaryToDouble(*pRightData, pRightOut);
+      ASSERT(0);
-      *pRightData = pRightOut;
+//      convertBinaryToDouble(*pRightData, pRightOut);
+//      *pRightData = pRightOut;
     } else if(typeRight != type) {
       convertNumberToNumber(*pRightData, pRightOut, typeRight, type);
       *pRightData = pRightOut;
@@ -1693,6 +1708,13 @@ void vectorIsTrue(SScalarParam* pLeft, SScalarParam* pRight, SScalarParam *pOut,
 
 STagVal getJsonValue(char *json, char *key, bool *isExist) {
   STagVal val = {.pKey = key};
+  if (tTagIsJson((const STag *)json) == false){
+    if(isExist){
+      *isExist = false;
+    }
+    return val;
+  }
+
   bool find = tTagGet(((const STag *)json), &val);  // json value is null and not exist is different
   if(isExist){
     *isExist = find;

source/libs/scalar/test/scalar/scalarTests.cpp

@@ -1310,11 +1310,11 @@ TEST(columnTest, json_column_logic_op) {
  printf("--------------------json string--0 {1, 8, 2, 2, 3, 0, 0, 0, 0}-------------------\n");
 
  key = "k2";
- bool eRes1[len+len1] = {false, false, true, true, false, false, false, true, false, true, false, true, true};
+ bool eRes1[len+len1] = {false, false, false, false, false, false, false, true, false, true, false, true, true};
  for(int i = 0; i < len; i++){
    makeCalculate(row, key, TSDB_DATA_TYPE_INT, &input[i], eRes1[i], op[i], false);
  }
- bool eRes_1[len0] = {true, true, false, false, false, false};
+ bool eRes_1[len0] = {false, false, false, false, false, false};
  for(int i = 0; i < len0; i++){
    makeCalculate(row, key, TSDB_DATA_TYPE_INT, &input[i], eRes_1[i], op[i], true);
  }
@@ -1346,11 +1346,11 @@ TEST(columnTest, json_column_logic_op) {
  printf("--------------------json bool--1 {1, 8, 2, 2, 3, 0, 0, 0, 0}-------------------\n");
 
  key = "k4";
- bool eRes3[len+len1] = {false, false, true, true, false, true, false, true, true, false, false, false, false};
+ bool eRes3[len+len1] = {false, false, false, false, false, false, false, true, true, false, false, false, false};
  for(int i = 0; i < len; i++){
    makeCalculate(row, key, TSDB_DATA_TYPE_INT, &input[i], eRes3[i], op[i], false);
  }
- bool eRes_3[len0] = {false, true, false, false, false, true};
+ bool eRes_3[len0] = {false, false, false, false, false, false};
  for(int i = 0; i < len0; i++){
    makeCalculate(row, key, TSDB_DATA_TYPE_INT, &input[i], eRes_3[i], op[i], true);
  }
@@ -1419,11 +1419,11 @@ TEST(columnTest, json_column_logic_op) {
  printf("--------------------json bool--  0 {1, 8, 2, 2, 3, 0, 0, 0, 0}-------------------\n");
 
  key = "k8";
- bool eRes7[len+len1] = {false, false, true, true, false, false, false, true, false, false, false, false, false};
+ bool eRes7[len+len1] = {false, false, false, false, false, false, false, true, false, false, false, false, false};
  for(int i = 0; i < len; i++){
    makeCalculate(row, key, TSDB_DATA_TYPE_INT, &input[i], eRes7[i], op[i], false);
  }
- bool eRes_7[len0] = {true, true, false, false, false, false};
+ bool eRes_7[len0] = {false, false, false, false, false, false};
  for(int i = 0; i < len0; i++) {
    makeCalculate(row, key, TSDB_DATA_TYPE_INT, &input[i], eRes_7[i], op[i], true);
  }
@@ -1438,11 +1438,11 @@ TEST(columnTest, json_column_logic_op) {
  printf("--------------------json string--  6.6hello {1, 8, 2, 2, 3, 0, 0, 0, 0}-------------------\n");
 
  key = "k9";
- bool eRes8[len+len1] = {true, false, false, false, false, true, false, true, true, false, true, false, true};
+ bool eRes8[len+len1] = {false, false, false, false, false, false, false, true, true, false, true, false, true};
  for(int i = 0; i < len; i++){
    makeCalculate(row, key, TSDB_DATA_TYPE_INT, &input[i], eRes8[i], op[i], false);
  }
- bool eRes_8[len0] = {false, true, true, true, false, true};
+ bool eRes_8[len0] = {false, false, false, false, false, false};
  for(int i = 0; i < len0; i++) {
    makeCalculate(row, key, TSDB_DATA_TYPE_INT, &input[i], eRes_8[i], op[i], true);
  }

source/libs/stream/inc/streamInc.h

@@ -33,8 +33,13 @@ 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);
+int32_t streamRetrieveReqToData(const SStreamRetrieveReq* pReq, SStreamDataBlock* pData);
 int32_t streamBuildDispatchMsg(SStreamTask* pTask, SStreamDataBlock* data, SRpcMsg* pMsg, SEpSet** ppEpSet);
 
+int32_t streamBroadcastToChildren(SStreamTask* pTask, const SSDataBlock* pBlock);
+
+int32_t tEncodeStreamRetrieveReq(SEncoder* pEncoder, const SStreamRetrieveReq* pReq);
+
 #ifdef __cplusplus
 }
 #endif

source/libs/stream/src/stream.c

@@ -112,7 +112,7 @@ int32_t streamTaskEnqueue(SStreamTask* pTask, SStreamDispatchReq* pReq, SRpcMsg*
   // enqueue
   if (pData != NULL) {
     pData->type = STREAM_DATA_TYPE_SSDATA_BLOCK;
-    pData->sourceVg = pReq->sourceVg;
+    pData->srcVgId = pReq->dataSrcVgId;
     // decode
     /*pData->blocks = pReq->data;*/
     /*pBlock->sourceVer = pReq->sourceVer;*/
@@ -133,7 +133,42 @@ int32_t streamTaskEnqueue(SStreamTask* pTask, SStreamDispatchReq* pReq, SRpcMsg*
   SStreamDispatchRsp* pCont = POINTER_SHIFT(buf, sizeof(SMsgHead));
   pCont->inputStatus = status;
   pCont->streamId = pReq->streamId;
-  pCont->taskId = pReq->sourceTaskId;
+  pCont->taskId = pReq->upstreamTaskId;
+  pRsp->pCont = buf;
+  pRsp->contLen = sizeof(SMsgHead) + sizeof(SStreamDispatchRsp);
+  tmsgSendRsp(pRsp);
+  return status == TASK_INPUT_STATUS__NORMAL ? 0 : -1;
+}
+
+int32_t streamTaskEnqueueRetrieve(SStreamTask* pTask, SStreamRetrieveReq* pReq, SRpcMsg* pRsp) {
+  SStreamDataBlock* pData = taosAllocateQitem(sizeof(SStreamDataBlock), DEF_QITEM);
+  int8_t            status = TASK_INPUT_STATUS__NORMAL;
+
+  // enqueue
+  if (pData != NULL) {
+    pData->type = STREAM_DATA_TYPE_SSDATA_BLOCK;
+    pData->srcVgId = 0;
+    // decode
+    /*pData->blocks = pReq->data;*/
+    /*pBlock->sourceVer = pReq->sourceVer;*/
+    streamRetrieveReqToData(pReq, pData);
+    if (streamTaskInput(pTask, (SStreamQueueItem*)pData) == 0) {
+      status = TASK_INPUT_STATUS__NORMAL;
+    } else {
+      status = TASK_INPUT_STATUS__FAILED;
+    }
+  } else {
+    /*streamTaskInputFail(pTask);*/
+    /*status = TASK_INPUT_STATUS__FAILED;*/
+  }
+
+  // rsp by input status
+  void* buf = rpcMallocCont(sizeof(SMsgHead) + sizeof(SStreamRetrieveRsp));
+  ((SMsgHead*)buf)->vgId = htonl(pReq->srcNodeId);
+  SStreamRetrieveRsp* pCont = POINTER_SHIFT(buf, sizeof(SMsgHead));
+  pCont->streamId = pReq->streamId;
+  pCont->rspToTaskId = pReq->srcTaskId;
+  pCont->rspFromTaskId = pReq->dstTaskId;
   pRsp->pCont = buf;
   pRsp->contLen = sizeof(SMsgHead) + sizeof(SStreamDispatchRsp);
   tmsgSendRsp(pRsp);
@@ -141,7 +176,7 @@ int32_t streamTaskEnqueue(SStreamTask* pTask, SStreamDispatchReq* pReq, SRpcMsg*
 }
 
 int32_t streamProcessDispatchReq(SStreamTask* pTask, SStreamDispatchReq* pReq, SRpcMsg* pRsp) {
-  qInfo("task %d receive dispatch req from node %d task %d", pTask->taskId, pReq->upstreamNodeId, pReq->sourceTaskId);
+  qInfo("task %d receive dispatch req from node %d task %d", pTask->taskId, pReq->upstreamNodeId, pReq->upstreamTaskId);
 
   // 1. handle input
   streamTaskEnqueue(pTask, pReq, pRsp);
@@ -208,3 +243,22 @@ int32_t streamProcessRecoverRsp(SStreamTask* pTask, SStreamTaskRecoverRsp* pRsp)
   //
   return 0;
 }
+
+int32_t streamProcessRetrieveReq(SStreamTask* pTask, SStreamRetrieveReq* pReq, SRpcMsg* pRsp) {
+  qInfo("task %d receive retrieve req from node %d task %d", pTask->taskId, pReq->srcNodeId, pReq->srcTaskId);
+
+  streamTaskEnqueueRetrieve(pTask, pReq, pRsp);
+
+  ASSERT(pTask->execType != TASK_EXEC__NONE);
+  streamExec(pTask, pTask->pMsgCb);
+
+  ASSERT(pTask->dispatchType != TASK_DISPATCH__NONE);
+  streamDispatch(pTask, pTask->pMsgCb);
+
+  return 0;
+}
+
+int32_t streamProcessRetrieveRsp(SStreamTask* pTask, SStreamRetrieveRsp* pRsp) {
+  //
+  return 0;
+}

source/libs/stream/src/streamData.c

@@ -15,27 +15,6 @@
 
 #include "streamInc.h"
 
-#if 0
-int32_t streamDataBlockEncode(void** buf, const SStreamDataBlock* pOutput) {
-  int32_t tlen = 0;
-  tlen += taosEncodeFixedI8(buf, pOutput->type);
-  tlen += taosEncodeFixedI32(buf, pOutput->sourceVg);
-  tlen += taosEncodeFixedI64(buf, pOutput->sourceVer);
-  ASSERT(pOutput->type == STREAM_INPUT__DATA_BLOCK);
-  tlen += tEncodeDataBlocks(buf, pOutput->blocks);
-  return tlen;
-}
-
-void* streamDataBlockDecode(const void* buf, SStreamDataBlock* pInput) {
-  buf = taosDecodeFixedI8(buf, &pInput->type);
-  buf = taosDecodeFixedI32(buf, &pInput->sourceVg);
-  buf = taosDecodeFixedI64(buf, &pInput->sourceVer);
-  ASSERT(pInput->type == STREAM_INPUT__DATA_BLOCK);
-  buf = tDecodeDataBlocks(buf, &pInput->blocks);
-  return (void*)buf;
-}
-#endif
-
 int32_t streamDispatchReqToData(const SStreamDispatchReq* pReq, SStreamDataBlock* pData) {
   int32_t blockNum = pReq->blockNum;
   SArray* pArray = taosArrayInit(blockNum, sizeof(SSDataBlock));
@@ -54,12 +33,25 @@ int32_t streamDispatchReqToData(const SStreamDispatchReq* pReq, SStreamDataBlock
     blockCompressDecode(pDataBlock, htonl(pRetrieve->numOfCols), htonl(pRetrieve->numOfRows), pRetrieve->data);
     // TODO: refactor
     pDataBlock->info.type = pRetrieve->streamBlockType;
-    pDataBlock->info.childId = pReq->sourceChildId;
+    pDataBlock->info.childId = pReq->upstreamChildId;
   }
   pData->blocks = pArray;
   return 0;
 }
 
+int32_t streamRetrieveReqToData(const SStreamRetrieveReq* pReq, SStreamDataBlock* pData) {
+  SArray* pArray = taosArrayInit(1, sizeof(SSDataBlock));
+  if (pArray == NULL) {
+    return -1;
+  }
+  taosArraySetSize(pArray, 1);
+  SRetrieveTableRsp* pRetrieve = pReq->pRetrieve;
+  SSDataBlock*       pBlock = taosArrayGet(pArray, 0);
+  blockCompressDecode(pBlock, htonl(pRetrieve->numOfCols), htonl(pRetrieve->numOfRows), pRetrieve->data);
+  pData->blocks = pArray;
+  return 0;
+}
+
 SStreamDataSubmit* streamDataSubmitNew(SSubmitReq* pReq) {
   SStreamDataSubmit* pDataSubmit = (SStreamDataSubmit*)taosAllocateQitem(sizeof(SStreamDataSubmit), DEF_QITEM);
   if (pDataSubmit == NULL) return NULL;

source/libs/stream/src/streamDispatch.c

@@ -19,9 +19,9 @@ int32_t tEncodeStreamDispatchReq(SEncoder* pEncoder, const SStreamDispatchReq* p
   if (tStartEncode(pEncoder) < 0) return -1;
   if (tEncodeI64(pEncoder, pReq->streamId) < 0) return -1;
   if (tEncodeI32(pEncoder, pReq->taskId) < 0) return -1;
-  if (tEncodeI32(pEncoder, pReq->sourceTaskId) < 0) return -1;
+  if (tEncodeI32(pEncoder, pReq->upstreamTaskId) < 0) return -1;
-  if (tEncodeI32(pEncoder, pReq->sourceVg) < 0) return -1;
+  if (tEncodeI32(pEncoder, pReq->dataSrcVgId) < 0) return -1;
-  if (tEncodeI32(pEncoder, pReq->sourceChildId) < 0) return -1;
+  if (tEncodeI32(pEncoder, pReq->upstreamChildId) < 0) return -1;
   if (tEncodeI32(pEncoder, pReq->upstreamNodeId) < 0) return -1;
   if (tEncodeI32(pEncoder, pReq->blockNum) < 0) return -1;
   ASSERT(taosArrayGetSize(pReq->data) == pReq->blockNum);
@@ -40,9 +40,9 @@ int32_t tDecodeStreamDispatchReq(SDecoder* pDecoder, SStreamDispatchReq* pReq) {
   if (tStartDecode(pDecoder) < 0) return -1;
   if (tDecodeI64(pDecoder, &pReq->streamId) < 0) return -1;
   if (tDecodeI32(pDecoder, &pReq->taskId) < 0) return -1;
-  if (tDecodeI32(pDecoder, &pReq->sourceTaskId) < 0) return -1;
+  if (tDecodeI32(pDecoder, &pReq->upstreamTaskId) < 0) return -1;
-  if (tDecodeI32(pDecoder, &pReq->sourceVg) < 0) return -1;
+  if (tDecodeI32(pDecoder, &pReq->dataSrcVgId) < 0) return -1;
-  if (tDecodeI32(pDecoder, &pReq->sourceChildId) < 0) return -1;
+  if (tDecodeI32(pDecoder, &pReq->upstreamChildId) < 0) return -1;
   if (tDecodeI32(pDecoder, &pReq->upstreamNodeId) < 0) return -1;
   if (tDecodeI32(pDecoder, &pReq->blockNum) < 0) return -1;
   ASSERT(pReq->blockNum > 0);
@@ -62,6 +62,102 @@ int32_t tDecodeStreamDispatchReq(SDecoder* pDecoder, SStreamDispatchReq* pReq) {
   return 0;
 }
 
+int32_t tEncodeStreamRetrieveReq(SEncoder* pEncoder, const SStreamRetrieveReq* pReq) {
+  //
+  if (tStartEncode(pEncoder) < 0) return -1;
+  if (tEncodeI64(pEncoder, pReq->streamId) < 0) return -1;
+  if (tEncodeI32(pEncoder, pReq->dstNodeId) < 0) return -1;
+  if (tEncodeI32(pEncoder, pReq->dstTaskId) < 0) return -1;
+  if (tEncodeI32(pEncoder, pReq->srcNodeId) < 0) return -1;
+  if (tEncodeI32(pEncoder, pReq->srcTaskId) < 0) return -1;
+  if (tEncodeBinary(pEncoder, (const uint8_t*)&pReq->pRetrieve, pReq->retrieveLen) < 0) return -1;
+  tEndEncode(pEncoder);
+  return pEncoder->pos;
+}
+
+int32_t tDecodeStreamRetrieveReq(SDecoder* pDecoder, SStreamRetrieveReq* pReq) {
+  int32_t tlen = 0;
+  if (tStartDecode(pDecoder) < 0) return -1;
+  if (tDecodeI64(pDecoder, &pReq->streamId) < 0) return -1;
+  if (tDecodeI32(pDecoder, &pReq->dstNodeId) < 0) return -1;
+  if (tDecodeI32(pDecoder, &pReq->dstTaskId) < 0) return -1;
+  if (tDecodeI32(pDecoder, &pReq->srcNodeId) < 0) return -1;
+  if (tDecodeI32(pDecoder, &pReq->srcTaskId) < 0) return -1;
+  if (tDecodeBinary(pDecoder, (uint8_t**)&pReq->pRetrieve, &pReq->retrieveLen) < 0) return -1;
+  tEndDecode(pDecoder);
+  return 0;
+}
+
+int32_t streamBroadcastToChildren(SStreamTask* pTask, const SSDataBlock* pBlock) {
+  SRetrieveTableRsp* pRetrieve = NULL;
+  void*              buf = NULL;
+  int32_t            dataStrLen = sizeof(SRetrieveTableRsp) + blockGetEncodeSize(pBlock);
+
+  pRetrieve = taosMemoryCalloc(1, dataStrLen);
+  if (pRetrieve == NULL) return -1;
+
+  pRetrieve->useconds = 0;
+  pRetrieve->precision = TSDB_DEFAULT_PRECISION;
+  pRetrieve->compressed = 0;
+  pRetrieve->completed = 1;
+  pRetrieve->streamBlockType = pBlock->info.type;
+  pRetrieve->numOfRows = htonl(pBlock->info.rows);
+  pRetrieve->numOfCols = htonl(pBlock->info.numOfCols);
+
+  int32_t actualLen = 0;
+  blockCompressEncode(pBlock, pRetrieve->data, &actualLen, pBlock->info.numOfCols, false);
+
+  SStreamRetrieveReq req = {
+      .streamId = pTask->streamId,
+      .srcNodeId = pTask->nodeId,
+      .srcTaskId = pTask->taskId,
+      .pRetrieve = pRetrieve,
+  };
+
+  int32_t sz = taosArrayGetSize(pTask->childEpInfo);
+  ASSERT(sz > 0);
+  for (int32_t i = 0; i < sz; i++) {
+    SStreamChildEpInfo* pEpInfo = taosArrayGetP(pTask->childEpInfo, i);
+    req.dstNodeId = pEpInfo->nodeId;
+    req.dstTaskId = pEpInfo->taskId;
+    int32_t code;
+    int32_t len;
+    tEncodeSize(tEncodeStreamRetrieveReq, &req, len, code);
+    if (code < 0) {
+      ASSERT(0);
+      return -1;
+    }
+
+    buf = rpcMallocCont(sizeof(SMsgHead) + len);
+    if (buf == NULL) {
+      goto FAIL;
+    }
+
+    ((SMsgHead*)buf)->vgId = htonl(pEpInfo->nodeId);
+    void*    abuf = POINTER_SHIFT(buf, sizeof(SMsgHead));
+    SEncoder encoder;
+    tEncoderInit(&encoder, abuf, len);
+    tEncodeStreamRetrieveReq(&encoder, &req);
+
+    SRpcMsg rpcMsg = {
+        .code = 0,
+        .msgType = TDMT_STREAM_RETRIEVE,
+        .pCont = buf,
+        .contLen = len,
+    };
+
+    if (tmsgSendReq(&pEpInfo->epSet, &rpcMsg) < 0) {
+      ASSERT(0);
+      return -1;
+    }
+  }
+  return 0;
+FAIL:
+  if (pRetrieve) taosMemoryFree(pRetrieve);
+  if (buf) taosMemoryFree(buf);
+  return -1;
+}
+
 static int32_t streamAddBlockToDispatchMsg(const SSDataBlock* pBlock, SStreamDispatchReq* pReq) {
   int32_t dataStrLen = sizeof(SRetrieveTableRsp) + blockGetEncodeSize(pBlock);
   void*   buf = taosMemoryCalloc(1, dataStrLen);
@@ -94,9 +190,9 @@ int32_t streamBuildDispatchMsg(SStreamTask* pTask, SStreamDataBlock* data, SRpcM
 
   SStreamDispatchReq req = {
       .streamId = pTask->streamId,
-      .sourceTaskId = pTask->taskId,
+      .dataSrcVgId = data->srcVgId,
-      .sourceVg = data->sourceVg,
+      .upstreamTaskId = pTask->taskId,
-      .sourceChildId = pTask->childId,
+      .upstreamChildId = pTask->selfChildId,
       .upstreamNodeId = pTask->nodeId,
       .blockNum = blockNum,
   };
@@ -147,7 +243,7 @@ int32_t streamBuildDispatchMsg(SStreamTask* pTask, SStreamDataBlock* data, SRpcM
   ASSERT(vgId > 0 || vgId == SNODE_HANDLE);
   req.taskId = downstreamTaskId;
 
-  qInfo("dispatch from task %d (child id %d) to down stream task %d in vnode %d", pTask->taskId, pTask->childId,
+  qInfo("dispatch from task %d (child id %d) to down stream task %d in vnode %d", pTask->taskId, pTask->selfChildId,
         downstreamTaskId, vgId);
 
   // serialize