merge from 3.0

include/common/tcommon.h

@@ -52,6 +52,8 @@ typedef enum EStreamType {
 typedef struct {
   SArray*   pTableList;
   SHashObj* map;  // speedup acquire the tableQueryInfo by table uid
+  void*     pTagCond;
+  void*     pTagIndexCond;
 } STableListInfo;
 
 typedef struct SColumnDataAgg {

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;

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)

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)
 

include/libs/nodes/nodes.h

@@ -226,6 +226,7 @@ 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

@@ -66,6 +66,7 @@ typedef struct SScanLogicNode {
   int8_t        intervalUnit;
   int8_t        slidingUnit;
   SNode*        pTagCond;
+  SNode*        pTagIndexCond;
   int8_t        triggerType;
   int64_t       watermark;
   int16_t       tsColId;
@@ -420,6 +421,8 @@ typedef struct SSortPhysiNode {
   SNodeList* pTargets;
 } SSortPhysiNode;
 
+typedef SSortPhysiNode SGroupSortPhysiNode;
+
 typedef struct SPartitionPhysiNode {
   SPhysiNode node;
   SNodeList* pExprs;  // these are expression list of partition_by_clause
@@ -466,6 +469,7 @@ 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

@@ -288,11 +288,11 @@ typedef enum ESqlClause {
 } ESqlClause;
 
 typedef struct SDeleteStmt {
-  ENodeType   type;           // QUERY_NODE_DELETE_STMT
+  ENodeType   type;        // QUERY_NODE_DELETE_STMT
-  SNode*      pFromTable;     // FROM clause
+  SNode*      pFromTable;  // FROM clause
-  SNode*      pWhere;         // WHERE clause
+  SNode*      pWhere;      // WHERE clause
-  SNode*      pCountFunc;     // count the number of rows affected
+  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/sync/sync.h

@@ -26,8 +26,9 @@ extern "C" {
 
 extern bool gRaftDetailLog;
 
-#define SYNC_INDEX_BEGIN 0
+#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);
@@ -468,7 +468,7 @@ typedef struct SyncLeaderTransfer {
    SRaftId  destId;
    */
   SNodeInfo newNodeInfo;
-  SRaftId newLeaderId;
+  SRaftId   newLeaderId;
 } SyncLeaderTransfer;
 
 SyncLeaderTransfer* syncLeaderTransferBuild(int32_t vgId);
@@ -489,17 +489,16 @@ 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;
+  uint32_t  bytes;
-  int32_t  vgId;
+  int32_t   vgId;
-  uint32_t msgType;
+  uint32_t  msgType;
-  SSyncCfg oldCfg;
+  SSyncCfg  oldCfg;
-  SSyncCfg newCfg;
+  SSyncCfg  newCfg;
   SyncIndex newCfgIndex;
-  SyncTerm newCfgTerm;
+  SyncTerm  newCfgTerm;
-  uint64_t newCfgSeqNum;
+  uint64_t  newCfgSeqNum;
 
 } SyncReconfigFinish;
 
@@ -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/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/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

@@ -2194,6 +2194,32 @@ 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);
@@ -2237,6 +2263,50 @@ 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/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/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;
@@ -699,28 +756,28 @@ 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 +786,7 @@ 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/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/vnode/src/meta/metaQuery.c

@@ -613,9 +613,6 @@ 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

@@ -400,8 +400,7 @@ 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/tq/tq.c

@@ -252,6 +252,7 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
 
 #if 1
   if (pReq->useSnapshot) {
+    tqInfo("retrieve using snapshot");
     int64_t lastVer = walGetCommittedVer(pTq->pWal);
     if (rsp.reqOffset < lastVer) {
       tqScanSnapshot(pTq, &pHandle->execHandle, &rsp, workerId);
@@ -259,6 +260,7 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
       if (rsp.blockNum != 0) {
         rsp.withTbName = false;
         rsp.rspOffset = lastVer;
+        tqInfo("direct send by snapshot rsp offset %ld", lastVer);
         goto SEND_RSP;
       }
     }

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

@@ -315,7 +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 {
@@ -839,13 +838,11 @@ int32_t finalizeResultRowIntoResultDataBlock(SDiskbasedBuf* pBuf, SResultRowPosi
                                        SSDataBlock* pBlock, SExecTaskInfo* pTaskInfo);
 
 int32_t createScanTableListInfo(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
-                                STableListInfo* pTableListInfo, uint64_t queryId, uint64_t taskId,
+                                STableListInfo* pTableListInfo, uint64_t queryId, uint64_t taskId);
-                                SNode* pTagCond);
+SOperatorInfo* createGroupSortOperatorInfo(SOperatorInfo* downstream, SGroupSortPhysiNode* pSortPhyNode,
-int32_t doCreateMultipleDataReaders(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
+                                           SExecTaskInfo* pTaskInfo);
-                                  STableListInfo* pTableListInfo, SArray* arrayReader, uint64_t queryId,
+SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanNode, STableListInfo *pTableListInfo,
-                                  uint64_t taskId);
+                                                SReadHandle* readHandle, SExecTaskInfo* pTaskInfo, uint64_t queryId, uint64_t taskId);
-SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanNode, SArray* dataReaders,
-                                                SReadHandle* readHandle, SExecTaskInfo* pTaskInfo);
 
 void copyUpdateDataBlock(SSDataBlock* pDest, SSDataBlock* pSource, int32_t tsColIndex);
 

source/libs/executor/src/executil.c

@@ -214,28 +214,111 @@ 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;
 
+  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++) {
@@ -246,7 +329,20 @@ int32_t getTableList(void* metaHandle, SScanPhysiNode* pScanNode, STableListInfo
     } else {
       code = tsdbGetAllTableList(metaHandle, tableUid, pListInfo->pTableList);
     }
-  } else {  // Create one table group.
+
+    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/executorimpl.c

@@ -1340,7 +1340,7 @@ void extractQualifiedTupleByFilterResult(SSDataBlock* pBlock, const int8_t* rowR
   }
 
   if (rowRes != NULL) {
-    int32_t      totalRows = pBlock->info.rows;
+    int32_t totalRows = pBlock->info.rows;
     SSDataBlock* px = createOneDataBlock(pBlock, true);
 
     for (int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
@@ -3872,8 +3872,7 @@ 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);
 
@@ -3968,7 +3967,7 @@ int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle,
         }
       }
     }
-    int32_t len = (int32_t)(pStart - (char*)keyBuf);
+    int32_t   len = (int32_t)(pStart - (char*)keyBuf);
 
     uint64_t* pGroupId = taosHashGet(pTableListInfo->map, keyBuf, len);
 
@@ -3987,7 +3986,7 @@ 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) {
@@ -3995,7 +3994,7 @@ 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;
@@ -4023,14 +4022,9 @@ 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);
-      createScanTableListInfo(pTableScanNode, pHandle, pTableListInfo, queryId, taskId, pTagCond);
-      doCreateMultipleDataReaders(pTableScanNode, pHandle, pTableListInfo, dataReaders, queryId, taskId);
-
       extractTableSchemaVersion(pHandle, pTableScanNode->scan.uid, pTaskInfo);
-
+      SOperatorInfo*  pOperator = createTableMergeScanOperatorInfo(pTableScanNode, pTableListInfo, pHandle, pTaskInfo, queryId, taskId);
-      SOperatorInfo*  pOperator = createTableMergeScanOperatorInfo(pTableScanNode, dataReaders, pHandle, pTaskInfo);
       STableScanInfo* pScanInfo = pOperator->info;
       pTaskInfo->cost.pRecoder = &pScanInfo->readRecorder;
       return pOperator;
@@ -4051,14 +4045,13 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
           // 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
           ASSERT(pHandle->meta);
-          getTableList(pHandle->meta, pScanPhyNode, pTableListInfo, pTagCond);
+          getTableList(pHandle->meta, pScanPhyNode, pTableListInfo);
         }
       }
-
       if (pDataReader == NULL && terrno != 0) {
         qDebug("%s pDataReader is NULL", GET_TASKID(pTaskInfo));
         // return NULL;
@@ -4083,7 +4076,7 @@ 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;
@@ -4141,7 +4134,7 @@ 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 +4225,8 @@ 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 +4338,8 @@ 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;
   }
@@ -4502,14 +4497,20 @@ 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;
   }
 
+  if ((*pTaskInfo)->pRoot == NULL) {
+    code = TSDB_CODE_QRY_OUT_OF_MEMORY;
+    goto _complete;
+  }
+
   return code;
 
 _complete:

source/libs/executor/src/scanoperator.c

@@ -337,7 +337,7 @@ 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 &&
@@ -1557,11 +1557,14 @@ 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 =
@@ -1596,6 +1599,8 @@ 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;
       }
     }
   }
@@ -1838,7 +1843,7 @@ 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) {
@@ -1861,9 +1866,7 @@ 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;
 }
@@ -1893,13 +1896,11 @@ SOperatorInfo* createTagScanOperatorInfo(SReadHandle* pReadHandle, STagScanPhysi
     goto _error;
   }
 
-  pInfo->pTableList = pTableListInfo;
+  pInfo->pTableList       = pTableListInfo;
-  pInfo->pColMatchInfo = colList;
+  pInfo->pColMatchInfo    = colList;
-  pInfo->pRes = createResDataBlock(pDescNode);
+  pInfo->pRes             = createResDataBlock(pDescNode);
-  pInfo->readHandle = *pReadHandle;
+  pInfo->readHandle       = *pReadHandle;
-  pInfo->curPos = 0;
+  pInfo->curPos           = 0;
-  pInfo->pFilterNode = pPhyNode->node.pConditions;
-
   pOperator->name = "TagScanOperator";
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN;
 
@@ -1924,6 +1925,12 @@ _error:
 }
 
 typedef struct STableMergeScanInfo {
+  STableListInfo* tableListInfo;
+  int32_t         tableStartIndex;
+  int32_t         tableEndIndex;
+  bool            hasGroupId;
+  uint64_t        groupId;
+
   SArray*     dataReaders;  // array of tsdbReaderT*
   SReadHandle readHandle;
 
@@ -1936,11 +1943,9 @@ typedef struct STableMergeScanInfo {
   SSDataBlock* pSortInputBlock;
   int64_t      startTs;  // sort start time
 
-  bool          hasGroupId;
+  SArray*  sortSourceParams;
-  uint64_t      groupId;
+  uint64_t queryId;
-  STupleHandle* prefetchedTuple;
+  uint64_t taskId;
-
-  SArray* sortSourceParams;
 
   SFileBlockLoadRecorder readRecorder;
   int64_t                numOfRows;
@@ -1968,8 +1973,6 @@ 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 compareTableKeyInfoByGid(const void* p1, const void* p2) {
@@ -1979,8 +1982,8 @@ int32_t compareTableKeyInfoByGid(const void* p1, const void* p2) {
 }
 
 int32_t createScanTableListInfo(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) {
     return code;
   }
@@ -2025,6 +2028,24 @@ _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;
@@ -2204,22 +2225,47 @@ 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);
@@ -2233,9 +2279,22 @@ 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;
 }
 
@@ -2278,14 +2337,38 @@ 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) {
-
+      doSetOperatorCompleted(pOperator);
-  if (pBlock != NULL) {
+      return NULL;
-    pOperator->resultInfo.totalRows += pBlock->info.rows;
+    }
-  } else {
+    pInfo->tableStartIndex = 0;
-    doSetOperatorCompleted(pOperator);
+    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;
 }
 
@@ -2293,17 +2376,10 @@ 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);
 
@@ -2329,8 +2405,9 @@ 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) {
@@ -2360,22 +2437,16 @@ 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);
@@ -2383,14 +2454,7 @@ 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;
@@ -2400,8 +2464,8 @@ 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();

source/libs/nodes/src/nodesCloneFuncs.c

@@ -351,6 +351,7 @@ static SNode* 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

@@ -2326,6 +2326,7 @@ 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;
@@ -2355,6 +2356,9 @@ 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;
 }
@@ -2388,6 +2392,9 @@ 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;
 }
@@ -3954,7 +3961,7 @@ 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);
@@ -3983,7 +3990,7 @@ 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

@@ -669,7 +669,7 @@ void nodesDestroyNode(SNode* pNode) {
       nodesDestroyNode(pStmt->pFromTable);
       nodesDestroyNode(pStmt->pWhere);
       nodesDestroyNode(pStmt->pCountFunc);
-      nodesDestroyNode(pStmt->pTagIndexCond);
+      nodesDestroyNode(pStmt->pTagCond);
       break;
     }
     case QUERY_NODE_QUERY: {
@@ -688,7 +688,13 @@ 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: {
@@ -897,6 +903,8 @@ 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;
     }
@@ -1130,6 +1138,7 @@ 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;
@@ -1659,6 +1668,7 @@ int32_t nodesMergeConds(SNode** pDst, SNodeList** pSrc) {
 typedef struct SClassifyConditionCxt {
   bool hasPrimaryKey;
   bool hasTagIndexCol;
+  bool hasTagCol;
   bool hasOtherCol;
 } SClassifyConditionCxt;
 
@@ -1670,6 +1680,9 @@ 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;
     }
@@ -1678,23 +1691,31 @@ 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;
@@ -1706,6 +1727,14 @@ 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));
         }
@@ -1723,11 +1752,15 @@ 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);
   }
@@ -1739,6 +1772,9 @@ static int32_t partitionLogicCond(SNode** pCondition, SNode** pPrimaryKeyCond, S
     if (NULL != pPrimaryKeyCond) {
       *pPrimaryKeyCond = pTempPrimaryKeyCond;
     }
+    if (NULL != pTagIndexCond) {
+      *pTagIndexCond = pTempTagIndexCond;
+    }
     if (NULL != pTagCond) {
       *pTagCond = pTempTagCond;
     }
@@ -1749,9 +1785,11 @@ 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);
   }
@@ -1759,10 +1797,11 @@ 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)) {
@@ -1772,6 +1811,21 @@ 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

@@ -2013,7 +2013,7 @@ 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) {
@@ -2161,9 +2161,6 @@ 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;
 }
@@ -2685,7 +2682,7 @@ 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);
   }
@@ -4614,6 +4611,25 @@ 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;
@@ -4632,6 +4648,8 @@ 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;
   }
@@ -5948,12 +5966,12 @@ 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

@@ -1160,8 +1160,8 @@ 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

@@ -436,6 +436,15 @@ 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 TSDB_CODE_PLAN_INTERNAL_ERROR;
+    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/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/sync/src/syncAppendEntries.c

@@ -436,6 +436,11 @@ static bool syncNodeOnAppendEntriesLogOK(SSyncNode* pSyncNode, SyncAppendEntries
   }
 
   SyncTerm myPreLogTerm = syncNodeGetPreTerm(pSyncNode, pMsg->prevLogIndex + 1);
+  if (myPreLogTerm == SYNC_TERM_INVALID) {
+    sError("vgId:%d sync get pre term error, preindex:%ld", pSyncNode->vgId, pMsg->prevLogIndex);
+    return false;
+  }
+
   if (pMsg->prevLogIndex <= myLastIndex && pMsg->prevLogTerm == myPreLogTerm) {
     if (gRaftDetailLog) {
       sTrace(

source/libs/sync/src/syncMain.c

@@ -345,7 +345,7 @@ bool syncCanLeaderTransfer(int64_t rid) {
   return matchOK;
 }
 
-int32_t syncForwardToPeer(int64_t rid, const SRpcMsg* pMsg, bool isWeak) {
+int32_t syncForwardToPeer(int64_t rid, SRpcMsg* pMsg, bool isWeak) {
   int32_t ret = syncPropose(rid, pMsg, isWeak);
   return ret;
 }
@@ -584,7 +584,7 @@ void setHeartbeatTimerMS(int64_t rid, int32_t hbTimerMS) {
   taosReleaseRef(tsNodeRefId, pSyncNode->rid);
 }
 
-int32_t syncPropose(int64_t rid, const SRpcMsg* pMsg, bool isWeak) {
+int32_t syncPropose(int64_t rid, SRpcMsg* pMsg, bool isWeak) {
   int32_t ret = 0;
 
   SSyncNode* pSyncNode = taosAcquireRef(tsNodeRefId, rid);
@@ -1309,40 +1309,44 @@ void syncNodeEventLog(const SSyncNode* pSyncNode, char* str) {
   SyncIndex logLastIndex = pSyncNode->pLogStore->syncLogLastIndex(pSyncNode->pLogStore);
   SyncIndex logBeginIndex = pSyncNode->pLogStore->syncLogBeginIndex(pSyncNode->pLogStore);
 
+  char* pCfgStr = syncCfg2SimpleStr(&(pSyncNode->pRaftCfg->cfg));
+
   if (userStrLen < 256) {
-    char logBuf[128 + 256];
+    char logBuf[256 + 256];
     if (pSyncNode != NULL && pSyncNode->pRaftCfg != NULL && pSyncNode->pRaftStore != NULL) {
       snprintf(logBuf, sizeof(logBuf),
                "vgId:%d, sync %s %s, term:%lu, commit:%ld, beginlog:%ld, lastlog:%ld, lastsnapshot:%ld, standby:%d, "
                "replica-num:%d, "
-               "lconfig:%ld, changing:%d",
+               "lconfig:%ld, changing:%d, %s",
                pSyncNode->vgId, syncUtilState2String(pSyncNode->state), str, pSyncNode->pRaftStore->currentTerm,
                pSyncNode->commitIndex, logBeginIndex, logLastIndex, snapshot.lastApplyIndex,
                pSyncNode->pRaftCfg->isStandBy, pSyncNode->replicaNum, pSyncNode->pRaftCfg->lastConfigIndex,
-               pSyncNode->changing);
+               pSyncNode->changing, pCfgStr);
     } else {
       snprintf(logBuf, sizeof(logBuf), "%s", str);
     }
     sDebug("%s", logBuf);
 
   } else {
-    int   len = 128 + userStrLen;
+    int   len = 256 + userStrLen;
     char* s = (char*)taosMemoryMalloc(len);
     if (pSyncNode != NULL && pSyncNode->pRaftCfg != NULL && pSyncNode->pRaftStore != NULL) {
       snprintf(s, len,
                "vgId:%d, sync %s %s, term:%lu, commit:%ld, beginlog:%ld, lastlog:%ld, lastsnapshot:%ld, standby:%d, "
                "replica-num:%d, "
-               "lconfig:%ld, changing:%d",
+               "lconfig:%ld, changing:%d, %s",
                pSyncNode->vgId, syncUtilState2String(pSyncNode->state), str, pSyncNode->pRaftStore->currentTerm,
                pSyncNode->commitIndex, logBeginIndex, logLastIndex, snapshot.lastApplyIndex,
                pSyncNode->pRaftCfg->isStandBy, pSyncNode->replicaNum, pSyncNode->pRaftCfg->lastConfigIndex,
-               pSyncNode->changing);
+               pSyncNode->changing, pCfgStr);
     } else {
       snprintf(s, len, "%s", str);
     }
     sDebug("%s", s);
     taosMemoryFree(s);
   }
+
+  taosMemoryFree(pCfgStr);
 }
 
 void syncNodeErrorLog(const SSyncNode* pSyncNode, char* str) {
@@ -1455,6 +1459,17 @@ void syncNodeDoConfigChange(SSyncNode* pSyncNode, SSyncCfg* pNewConfig, SyncInde
     isAdd = false;
   }
 
+  // log begin config change
+  do {
+    char  eventLog[256];
+    char* pOldCfgStr = syncCfg2SimpleStr(&oldConfig);
+    char* pNewCfgStr = syncCfg2SimpleStr(pNewConfig);
+    snprintf(eventLog, sizeof(eventLog), "begin do config change, from %s to %s", pOldCfgStr, pNewCfgStr);
+    syncNodeEventLog(pSyncNode, eventLog);
+    taosMemoryFree(pOldCfgStr);
+    taosMemoryFree(pNewCfgStr);
+  } while (0);
+
   if (IamInNew) {
     pSyncNode->pRaftCfg->isStandBy = 0;  // change isStandBy to normal
   }
@@ -1613,6 +1628,17 @@ void syncNodeDoConfigChange(SSyncNode* pSyncNode, SSyncCfg* pNewConfig, SyncInde
   }
 
 _END:
+
+  // log end config change
+  do {
+    char  eventLog[256];
+    char* pOldCfgStr = syncCfg2SimpleStr(&oldConfig);
+    char* pNewCfgStr = syncCfg2SimpleStr(pNewConfig);
+    snprintf(eventLog, sizeof(eventLog), "end do config change, from %s to %s", pOldCfgStr, pNewCfgStr);
+    syncNodeEventLog(pSyncNode, eventLog);
+    taosMemoryFree(pOldCfgStr);
+    taosMemoryFree(pNewCfgStr);
+  } while (0);
   return;
 }
 
@@ -1888,6 +1914,16 @@ SyncIndex syncNodeSyncStartIndex(SSyncNode* pSyncNode) {
   return syncStartIndex;
 }
 
+SyncIndex syncNodeGetPreIndex(SSyncNode* pSyncNode, SyncIndex index) {
+  SyncIndex preIndex = index - 1;
+  if (preIndex < SYNC_INDEX_INVALID) {
+    preIndex = SYNC_INDEX_INVALID;
+  }
+
+  return preIndex;
+}
+
+/*
 SyncIndex syncNodeGetPreIndex(SSyncNode* pSyncNode, SyncIndex index) {
   ASSERT(index >= SYNC_INDEX_BEGIN);
 
@@ -1900,7 +1936,42 @@ SyncIndex syncNodeGetPreIndex(SSyncNode* pSyncNode, SyncIndex index) {
   SyncIndex preIndex = index - 1;
   return preIndex;
 }
+*/
 
+SyncTerm syncNodeGetPreTerm(SSyncNode* pSyncNode, SyncIndex index) {
+  if (index < SYNC_INDEX_BEGIN) {
+    return SYNC_TERM_INVALID;
+  }
+
+  if (index == SYNC_INDEX_BEGIN) {
+    return 0;
+  }
+
+  SyncTerm        preTerm = 0;
+  SyncIndex       preIndex = index - 1;
+  SSyncRaftEntry* pPreEntry = NULL;
+  int32_t         code = pSyncNode->pLogStore->syncLogGetEntry(pSyncNode->pLogStore, preIndex, &pPreEntry);
+  if (code == 0) {
+    ASSERT(pPreEntry != NULL);
+    preTerm = pPreEntry->term;
+    taosMemoryFree(pPreEntry);
+    return preTerm;
+  } else {
+    if (terrno == TSDB_CODE_WAL_LOG_NOT_EXIST) {
+      SSnapshot snapshot = {.data = NULL, .lastApplyIndex = -1, .lastApplyTerm = 0, .lastConfigIndex = -1};
+      if (pSyncNode->pFsm->FpGetSnapshotInfo != NULL) {
+        pSyncNode->pFsm->FpGetSnapshotInfo(pSyncNode->pFsm, &snapshot);
+        if (snapshot.lastApplyIndex == preIndex) {
+          return snapshot.lastApplyTerm;
+        }
+      }
+    }
+  }
+
+  return SYNC_TERM_INVALID;
+}
+
+#if 0
 SyncTerm syncNodeGetPreTerm(SSyncNode* pSyncNode, SyncIndex index) {
   ASSERT(index >= SYNC_INDEX_BEGIN);
 
@@ -1938,6 +2009,7 @@ SyncTerm syncNodeGetPreTerm(SSyncNode* pSyncNode, SyncIndex index) {
   ASSERT(0);
   return -1;
 }
+#endif
 
 #if 0
 SyncTerm syncNodeGetPreTerm(SSyncNode* pSyncNode, SyncIndex index) {

source/libs/sync/src/syncRaftLog.c

@@ -144,7 +144,8 @@ static int32_t raftLogAppendEntry(struct SSyncLogStore* pLogStore, SSyncRaftEntr
 
   SyncIndex writeIndex = raftLogWriteIndex(pLogStore);
   if (pEntry->index != writeIndex) {
-    sError("raftLogAppendEntry error, pEntry->index:%ld update to writeIndex:%ld", pEntry->index, writeIndex);
+    sError("vgId:%d wal write index error, entry-index:%ld update to %ld", pData->pSyncNode->vgId, pEntry->index,
+           writeIndex);
     pEntry->index = writeIndex;
   }
 
@@ -157,10 +158,10 @@ static int32_t raftLogAppendEntry(struct SSyncLogStore* pLogStore, SSyncRaftEntr
   if (code != 0) {
     int32_t     err = terrno;
     const char* errStr = tstrerror(err);
-    int32_t     linuxErr = errno;
+    int32_t     sysErr = errno;
-    const char* linuxErrMsg = strerror(errno);
+    const char* sysErrStr = strerror(errno);
-    sError("raftLogAppendEntry error, err:%d %X, msg:%s, linuxErr:%d, linuxErrMsg:%s", err, err, errStr, linuxErr,
+    sError("vgId:%d wal write error, index:%ld, err:%d %X, msg:%s, syserr:%d, sysmsg:%s", pData->pSyncNode->vgId,
-           linuxErrMsg);
+           pEntry->index, err, err, errStr, sysErr, sysErrStr);
     ASSERT(0);
   }
 
@@ -237,12 +238,15 @@ static int32_t raftLogGetEntry(struct SSyncLogStore* pLogStore, SyncIndex index,
   if (code != 0) {
     int32_t     err = terrno;
     const char* errStr = tstrerror(err);
-    int32_t     linuxErr = errno;
+    int32_t     sysErr = errno;
-    const char* linuxErrMsg = strerror(errno);
+    const char* sysErrStr = strerror(errno);
-    sError("raftLogGetEntry error, err:%d %X, msg:%s, linuxErr:%d, linuxErrMsg:%s", err, err, errStr, linuxErr,
+    sError("vgId:%d wal read error, index:%ld, err:%d %X, msg:%s, syserr:%d, sysmsg:%s", pData->pSyncNode->vgId, index,
-           linuxErrMsg);
+           err, err, errStr, sysErr, sysErrStr);
 
+    int32_t saveErr = terrno;
     walCloseReadHandle(pWalHandle);
+    terrno = saveErr;
+
     return code;
   }
 
@@ -257,8 +261,9 @@ static int32_t raftLogGetEntry(struct SSyncLogStore* pLogStore, SyncIndex index,
   ASSERT((*ppEntry)->dataLen == pWalHandle->pHead->head.bodyLen);
   memcpy((*ppEntry)->data, pWalHandle->pHead->head.body, pWalHandle->pHead->head.bodyLen);
 
-  // need to hold, do not new every time!!
+  int32_t saveErr = terrno;
   walCloseReadHandle(pWalHandle);
+  terrno = saveErr;
 
   return code;
 }
@@ -270,10 +275,11 @@ static int32_t raftLogTruncate(struct SSyncLogStore* pLogStore, SyncIndex fromIn
   if (code != 0) {
     int32_t     err = terrno;
     const char* errStr = tstrerror(err);
-    int32_t     linuxErr = errno;
+    int32_t     sysErr = errno;
-    const char* linuxErrMsg = strerror(errno);
+    const char* sysErrStr = strerror(errno);
-    sError("raftLogTruncate error, err:%d %X, msg:%s, linuxErr:%d, linuxErrMsg:%s", err, err, errStr, linuxErr,
+    sError("vgId:%d wal truncate error, from-index:%ld, err:%d %X, msg:%s, syserr:%d, sysmsg:%s",
-           linuxErrMsg);
+           pData->pSyncNode->vgId, fromIndex, err, err, errStr, sysErr, sysErrStr);
+
     ASSERT(0);
   }
   return code;
@@ -360,10 +366,11 @@ int32_t logStoreAppendEntry(SSyncLogStore* pLogStore, SSyncRaftEntry* pEntry) {
   if (code != 0) {
     int32_t     err = terrno;
     const char* errStr = tstrerror(err);
-    int32_t     linuxErr = errno;
+    int32_t     sysErr = errno;
-    const char* linuxErrMsg = strerror(errno);
+    const char* sysErrStr = strerror(errno);
-    sError("walWriteWithSyncInfo error, err:%d %X, msg:%s, linuxErr:%d, linuxErrMsg:%s", err, err, errStr, linuxErr,
+    sError("vgId:%d wal write error, index:%ld, err:%d %X, msg:%s, syserr:%d, sysmsg:%s", pData->pSyncNode->vgId,
-           linuxErrMsg);
+           pEntry->index, err, err, errStr, sysErr, sysErrStr);
+
     ASSERT(0);
   }
 
@@ -389,10 +396,11 @@ SSyncRaftEntry* logStoreGetEntry(SSyncLogStore* pLogStore, SyncIndex index) {
     if (code != 0) {
       int32_t     err = terrno;
       const char* errStr = tstrerror(err);
-      int32_t     linuxErr = errno;
+      int32_t     sysErr = errno;
-      const char* linuxErrMsg = strerror(errno);
+      const char* sysErrStr = strerror(errno);
-      sError("walReadWithHandle error, err:%d %X, msg:%s, linuxErr:%d, linuxErrMsg:%s", err, err, errStr, linuxErr,
+      sError("vgId:%d wal read error, index:%ld, err:%d %X, msg:%s, syserr:%d, sysmsg:%s", pData->pSyncNode->vgId,
-             linuxErrMsg);
+             index, err, err, errStr, sysErr, sysErrStr);
+
       ASSERT(0);
     }
     // ASSERT(walReadWithHandle(pWalHandle, index) == 0);
@@ -409,8 +417,10 @@ SSyncRaftEntry* logStoreGetEntry(SSyncLogStore* pLogStore, SyncIndex index) {
     ASSERT(pEntry->dataLen == pWalHandle->pHead->head.bodyLen);
     memcpy(pEntry->data, pWalHandle->pHead->head.body, pWalHandle->pHead->head.bodyLen);
 
-    // need to hold, do not new every time!!
+    int32_t saveErr = terrno;
     walCloseReadHandle(pWalHandle);
+    terrno = saveErr;
+
     return pEntry;
 
   } else {
@@ -426,10 +436,11 @@ int32_t logStoreTruncate(SSyncLogStore* pLogStore, SyncIndex fromIndex) {
   if (code != 0) {
     int32_t     err = terrno;
     const char* errStr = tstrerror(err);
-    int32_t     linuxErr = errno;
+    int32_t     sysErr = errno;
-    const char* linuxErrMsg = strerror(errno);
+    const char* sysErrStr = strerror(errno);
-    sError("walRollback error, err:%d %X, msg:%s, linuxErr:%d, linuxErrMsg:%s", err, err, errStr, linuxErr,
+    sError("vgId:%d wal truncate error, from-index:%ld, err:%d %X, msg:%s, syserr:%d, sysmsg:%s",
-           linuxErrMsg);
+           pData->pSyncNode->vgId, fromIndex, err, err, errStr, sysErr, sysErrStr);
+
     ASSERT(0);
   }
   return 0;
@@ -460,9 +471,11 @@ int32_t logStoreUpdateCommitIndex(SSyncLogStore* pLogStore, SyncIndex index) {
   if (code != 0) {
     int32_t     err = terrno;
     const char* errStr = tstrerror(err);
-    int32_t     linuxErr = errno;
+    int32_t     sysErr = errno;
-    const char* linuxErrMsg = strerror(errno);
+    const char* sysErrStr = strerror(errno);
-    sError("walCommit error, err:%d %X, msg:%s, linuxErr:%d, linuxErrMsg:%s", err, err, errStr, linuxErr, linuxErrMsg);
+    sError("vgId:%d wal update commit index error, index:%ld, err:%d %X, msg:%s, syserr:%d, sysmsg:%s",
+           pData->pSyncNode->vgId, index, err, err, errStr, sysErr, sysErrStr);
+
     ASSERT(0);
   }
   return 0;

source/libs/sync/src/syncReplication.c

@@ -139,6 +139,15 @@ int32_t syncNodeAppendEntriesPeersSnapshot(SSyncNode* pSyncNode) {
     // pre index, pre term
     SyncIndex preLogIndex = syncNodeGetPreIndex(pSyncNode, nextIndex);
     SyncTerm  preLogTerm = syncNodeGetPreTerm(pSyncNode, nextIndex);
+    if (preLogTerm == SYNC_TERM_INVALID) {
+      SyncIndex newNextIndex = syncNodeGetLastIndex(pSyncNode) + 1;
+      syncIndexMgrSetIndex(pSyncNode->pNextIndex, pDestId, newNextIndex);
+      syncIndexMgrSetIndex(pSyncNode->pMatchIndex, pDestId, SYNC_INDEX_INVALID);
+      sError("vgId:%d sync get pre term error, nextIndex:%ld, update next-index:%ld, match-index:%d, raftid:%ld",
+             pSyncNode->vgId, nextIndex, newNextIndex, SYNC_INDEX_INVALID, pDestId->addr);
+
+      return -1;
+    }
 
     // batch optimized
     // SyncIndex lastIndex = syncUtilMinIndex(pSyncNode->pLogStore->getLastIndex(pSyncNode->pLogStore), nextIndex);

source/libs/sync/src/syncSnapshot.c

@@ -113,8 +113,16 @@ void snapshotSenderStart(SSyncSnapshotSender *pSender, SSnapshot snapshot, void
     }
 
     if (!getLastConfig) {
-      syncNodeLog3("", pSender->pSyncNode);
+      char logBuf[128];
-      ASSERT(0);
+      snprintf(logBuf, sizeof(logBuf), "snapshot sender update lcindex from %ld to -1",
+               pSender->snapshot.lastConfigIndex);
+      pSender->snapshot.lastConfigIndex = -1;
+
+      char *eventLog = snapshotSender2SimpleStr(pSender, logBuf);
+      syncNodeEventLog(pSender->pSyncNode, eventLog);
+      taosMemoryFree(eventLog);
+
+      memset(&(pSender->lastConfig), 0, sizeof(SSyncCfg));
     }
 
   } else {

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

@@ -857,20 +857,20 @@ static int tdbBtreeBalance(SBTC *pBtc) {
 }
 // TDB_BTREE_BALANCE
 
-static int tdbFetchOvflPage(SPager *pPager, SPgno *pPgno, SPage **ppOfp, TXN *pTxn, SBTree *pBt) {
+static int tdbFetchOvflPage(SPgno *pPgno, SPage **ppOfp, TXN *pTxn, SBTree *pBt) {
   int ret = 0;
 
   *pPgno = 0;
   SBtreeInitPageArg iArg;
   iArg.pBt = pBt;
   iArg.flags = TDB_FLAG_ADD(0, TDB_BTREE_OVFL);
-  ret = tdbPagerFetchPage(pPager, pPgno, ppOfp, tdbBtreeInitPage, &iArg, pTxn);
+  ret = tdbPagerFetchPage(pBt->pPager, pPgno, ppOfp, tdbBtreeInitPage, &iArg, pTxn);
   if (ret < 0) {
     return -1;
   }
 
   // mark dirty
-  ret = tdbPagerWrite(pPager, *ppOfp);
+  ret = tdbPagerWrite(pBt->pPager, *ppOfp);
   if (ret < 0) {
     ASSERT(0);
     return -1;
@@ -879,13 +879,13 @@ static int tdbFetchOvflPage(SPager *pPager, SPgno *pPgno, SPage **ppOfp, TXN *pT
   return ret;
 }
 
-static int tdbLoadOvflPage(SPager *pPager, SPgno *pPgno, SPage **ppOfp, TXN *pTxn, SBTree *pBt) {
+static int tdbLoadOvflPage(SPgno *pPgno, SPage **ppOfp, TXN *pTxn, SBTree *pBt) {
   int ret = 0;
 
   SBtreeInitPageArg iArg;
   iArg.pBt = pBt;
   iArg.flags = TDB_FLAG_ADD(0, TDB_BTREE_OVFL);
-  ret = tdbPagerFetchPage(pPager, pPgno, ppOfp, tdbBtreeInitPage, &iArg, pTxn);
+  ret = tdbPagerFetchPage(pBt->pPager, pPgno, ppOfp, tdbBtreeInitPage, &iArg, pTxn);
   if (ret < 0) {
     return -1;
   }
@@ -922,7 +922,7 @@ static int tdbBtreeEncodePayload(SPage *pPage, SCell *pCell, int nHeader, const
     SPgno pgno = 0;
     SPage *ofp, *nextOfp;
 
-    ret = tdbFetchOvflPage(pPage->pPager, &pgno, &ofp, pTxn, pBt);
+    ret = tdbFetchOvflPage(&pgno, &ofp, pTxn, pBt);
     if (ret < 0) {
       return -1;
     }
@@ -962,7 +962,7 @@ static int tdbBtreeEncodePayload(SPage *pPage, SCell *pCell, int nHeader, const
 	// fetch next ofp if not last page
 	if (!lastPage) {
 	  // fetch a new ofp and make it dirty
-	  ret = tdbFetchOvflPage(pPage->pPager, &pgno, &nextOfp, pTxn, pBt);
+	  ret = tdbFetchOvflPage(&pgno, &nextOfp, pTxn, pBt);
 	  if (ret < 0) {
 	    tdbFree(pBuf);
 	    return -1;
@@ -1019,14 +1019,14 @@ static int tdbBtreeEncodePayload(SPage *pPage, SCell *pCell, int nHeader, const
 	    nLeft -= lastKeyPageSpace;
 
 	    // fetch next ofp, a new ofp and make it dirty
-	    ret = tdbFetchOvflPage(pPage->pPager, &pgno, &nextOfp, pTxn, pBt);
+	    ret = tdbFetchOvflPage(&pgno, &nextOfp, pTxn, pBt);
 	    if (ret < 0) {
 	      return -1;
 	    }
 	  }
 	} else {
 	  // fetch next ofp, a new ofp and make it dirty
-	  ret = tdbFetchOvflPage(pPage->pPager, &pgno, &nextOfp, pTxn, pBt);
+	  ret = tdbFetchOvflPage(&pgno, &nextOfp, pTxn, pBt);
 	  if (ret < 0) {
 	    return -1;
 	  }
@@ -1057,7 +1057,7 @@ static int tdbBtreeEncodePayload(SPage *pPage, SCell *pCell, int nHeader, const
 	// fetch next ofp if not last page
 	if (!lastPage) {
 	  // fetch a new ofp and make it dirty
-	  ret = tdbFetchOvflPage(pPage->pPager, &pgno, &nextOfp, pTxn, pBt);
+	  ret = tdbFetchOvflPage(&pgno, &nextOfp, pTxn, pBt);
 	  if (ret < 0) {
 	    tdbFree(pBuf);
 	    return -1;
@@ -1198,7 +1198,7 @@ static int tdbBtreeDecodePayload(SPage *pPage, const SCell *pCell, int nHeader,
 
       // unpack left val data from ovpages
       while (pgno != 0)  {
-	ret = tdbLoadOvflPage(pPage->pPager, &pgno, &ofp, pTxn, pBt);
+	ret = tdbLoadOvflPage(&pgno, &ofp, pTxn, pBt);
 	if (ret < 0) {
 	  return -1;
 	}
@@ -1235,7 +1235,7 @@ static int tdbBtreeDecodePayload(SPage *pPage, const SCell *pCell, int nHeader,
       int lastKeyPageSpace = 0;
       // load left key & val to ovpages
       while (pgno != 0) {
-	ret = tdbLoadOvflPage(pPage->pPager, &pgno, &ofp, pTxn, pBt);
+	ret = tdbLoadOvflPage(&pgno, &ofp, pTxn, pBt);
 	if (ret < 0) {
 	  return -1;
 	}
@@ -1280,7 +1280,7 @@ static int tdbBtreeDecodePayload(SPage *pPage, const SCell *pCell, int nHeader,
       }
 
       while (nLeft > 0) {
-	ret = tdbLoadOvflPage(pPage->pPager, &pgno, &ofp, pTxn, pBt);
+	ret = tdbLoadOvflPage(&pgno, &ofp, pTxn, pBt);
 	if (ret < 0) {
 	  return -1;
 	}
@@ -1411,7 +1411,7 @@ static int tdbBtreeCellSize(const SPage *pPage, SCell *pCell, int dropOfp, TXN *
       int bytes;
 
       while (pgno != 0) {
-	ret = tdbLoadOvflPage(pPage->pPager, &pgno, &ofp, pTxn, pBt);
+	ret = tdbLoadOvflPage(&pgno, &ofp, pTxn, pBt);
 	if (ret < 0) {
 	  return -1;
 	}
@@ -2023,7 +2023,7 @@ int tdbBtcMoveTo(SBTC *pBtc, const void *pKey, int kLen, int *pCRst) {
       // check if key <= current position
       if (idx < nCells) {
         pCell = tdbPageGetCell(pPage, idx);
-        tdbBtreeDecodeCell(pPage, pCell, &cd, pBtc->pTxn, pBtc->pBt);
+        tdbBtreeDecodeCell(pPage, pCell, &cd);
         c = pBt->kcmpr(pKey, kLen, cd.pKey, cd.kLen);
         if (c > 0) break;
       }

source/libs/wal/src/walMeta.c

@@ -19,6 +19,8 @@
 #include "tref.h"
 #include "walInt.h"
 
+bool FORCE_INLINE walIsEmpty(SWal* pWal) { return pWal->vers.firstVer == -1; }
+
 int64_t FORCE_INLINE walGetFirstVer(SWal* pWal) { return pWal->vers.firstVer; }
 
 int64_t FORCE_INLINE walGetSnaphostVer(SWal* pWal) { return pWal->vers.snapshotVer; }

source/libs/wal/src/walWrite.c

@@ -141,7 +141,7 @@ int32_t walRollback(SWal *pWal, int64_t ver) {
   // validate offset
   SWalHead head;
   ASSERT(taosValidFile(pLogTFile));
-  int size = taosReadFile(pLogTFile, &head, sizeof(SWalHead));
+  int64_t size = taosReadFile(pLogTFile, &head, sizeof(SWalHead));
   if (size != sizeof(SWalHead)) {
     return -1;
   }
@@ -149,22 +149,33 @@ int32_t walRollback(SWal *pWal, int64_t ver) {
 
   ASSERT(code == 0);
   if (code != 0) {
+    terrno = TSDB_CODE_WAL_FILE_CORRUPTED;
     return -1;
   }
   if (head.head.version != ver) {
-    // TODO
+    ASSERT(0);
+    terrno = TSDB_CODE_WAL_FILE_CORRUPTED;
     return -1;
   }
+
   // truncate old files
   code = taosFtruncateFile(pLogTFile, entry.offset);
   if (code < 0) {
+    ASSERT(0);
+    terrno = TAOS_SYSTEM_ERROR(errno);
     return -1;
   }
   code = taosFtruncateFile(pIdxTFile, idxOff);
   if (code < 0) {
+    ASSERT(0);
+    terrno = TAOS_SYSTEM_ERROR(errno);
     return -1;
   }
   pWal->vers.lastVer = ver - 1;
+  if (pWal->vers.lastVer < pWal->vers.firstVer) {
+    ASSERT(pWal->vers.lastVer == pWal->vers.firstVer - 1);
+    pWal->vers.firstVer = -1;
+  }
   ((SWalFileInfo *)taosArrayGetLast(pWal->fileInfoSet))->lastVer = ver - 1;
   ((SWalFileInfo *)taosArrayGetLast(pWal->fileInfoSet))->fileSize = entry.offset;
   taosCloseFile(&pIdxTFile);

source/os/src/osLocale.c

@@ -88,11 +88,11 @@ void taosSetSystemLocale(const char *inLocale, const char *inCharSet) {
 
 void taosGetSystemLocale(char *outLocale, char *outCharset) {
 #ifdef WINDOWS
-  char *locale = setlocale(LC_CTYPE, "chs");
+  char *locale = setlocale(LC_CTYPE, "en_US.UTF-8");
   if (locale != NULL) {
     tstrncpy(outLocale, locale, TD_LOCALE_LEN);
   }
-  strcpy(outCharset, "cp936");
+  strcpy(outCharset, "UTF-8");
 
 #elif defined(_TD_DARWIN_64)
   /*

source/util/src/tconfig.c

@@ -503,6 +503,38 @@ const char *cfgDtypeStr(ECfgDataType type) {
   }
 }
 
+void cfgDumpItemValue(SConfigItem *pItem, char* buf, int32_t bufSize, int32_t* pLen) {
+  int32_t len = 0;
+  switch (pItem->dtype) {
+    case CFG_DTYPE_BOOL:
+      len = snprintf(buf, bufSize, "%u", pItem->bval);
+      break;
+    case CFG_DTYPE_INT32:
+      len = snprintf(buf, bufSize, "%d", pItem->i32);
+      break;
+    case CFG_DTYPE_INT64:
+      len = snprintf(buf, bufSize, "%" PRId64, pItem->i64);
+      break;
+    case CFG_DTYPE_FLOAT:
+      len = snprintf(buf, bufSize, "%f", pItem->fval);
+      break;
+    case CFG_DTYPE_STRING:
+    case CFG_DTYPE_DIR:
+    case CFG_DTYPE_LOCALE:
+    case CFG_DTYPE_CHARSET:
+    case CFG_DTYPE_TIMEZONE:
+    case CFG_DTYPE_NONE:
+      len = snprintf(buf, bufSize, "%s", pItem->str);
+      break;
+  }
+
+  if (len > bufSize) {
+    len = bufSize;
+  }
+
+  *pLen = len;
+}
+
 void cfgDumpCfg(SConfig *pCfg, bool tsc, bool dump) {
   if (dump) {
     printf("                     global config");

tests/script/jenkins/basic.txt

@@ -12,6 +12,8 @@
 ## ---- db
 ./test.sh -f tsim/db/create_all_options.sim
 ./test.sh -f tsim/db/alter_option.sim
+./test.sh -f tsim/db/alter_replica_13.sim
+#./test.sh -f tsim/db/alter_replica_31.sim
 ./test.sh -f tsim/db/basic1.sim
 ./test.sh -f tsim/db/basic2.sim
 ./test.sh -f tsim/db/basic3.sim
@@ -21,6 +23,7 @@
 ./test.sh -f tsim/db/taosdlog.sim
 
 # ---- dnode
+./test.sh -f tsim/dnode/balance_replica1.sim
 ./test.sh -f tsim/dnode/create_dnode.sim
 ./test.sh -f tsim/dnode/drop_dnode_has_mnode.sim
 ./test.sh -f tsim/dnode/drop_dnode_has_qnode_snode.sim
@@ -28,6 +31,8 @@
 #./test.sh -f tsim/dnode/drop_dnode_has_vnode_replica3.sim
 #./test.sh -f tsim/dnode/drop_dnode_has_multi_vnode_replica1.sim
 #./test.sh -f tsim/dnode/drop_dnode_has_multi_vnode_replica3.sim
+#./test.sh -f tsim/dnode/redistribute_vgroup_replica3_v1_leader.sim
+./test.sh -f tsim/dnode/redistribute_vgroup_replica3_v1_follower.sim
 
 # ---- insert
 ./test.sh -f tsim/insert/basic0.sim
@@ -104,6 +109,7 @@
 ./test.sh -f tsim/tmq/basic4Of2Cons.sim
 ./test.sh -f tsim/tmq/basic2Of2ConsOverlap.sim
 ./test.sh -f tsim/tmq/topic.sim
+./test.sh -f tsim/tmq/snapshot.sim
 
 # --- stable
 ./test.sh -f tsim/stable/disk.sim

tests/script/tsim/dnode/balance_replica1.sim

@@ -0,0 +1,123 @@
+system sh/stop_dnodes.sh
+system sh/deploy.sh -n dnode1 -i 1
+system sh/deploy.sh -n dnode2 -i 2
+system sh/deploy.sh -n dnode3 -i 3
+system sh/cfg.sh -n dnode1 -c supportVnodes -v 0
+system sh/exec.sh -n dnode1 -s start
+system sh/exec.sh -n dnode2 -s start
+sql connect
+
+print =============== step1 create dnode2
+# no enough vnodes
+sql balance vgroup
+
+sql create dnode $hostname port 7200
+sql create dnode $hostname port 7300
+
+$x = 0
+step1:
+	$ = $x + 1
+	sleep 1000
+	if $x == 10 then
+	  print ---> dnode not online!
+		return -1
+	endi
+sql show dnodes
+print ---> $data00 $data01 $data02 $data03 $data04 $data05
+print ---> $data10 $data11 $data12 $data13 $data14 $data15
+if $rows != 3 then
+  return -1
+endi
+if $data(1)[4] != ready then
+  goto step1
+endi
+if $data(2)[4] != ready then
+  goto step1
+endi
+if $data(3)[4] != offline then
+  goto step1
+endi
+
+print =============== step2 create database
+sql create database d1 vgroups 2
+sql use d1
+sql create table d1.st (ts timestamp, i int) tags (j int)
+sql create table d1.c1 using st tags(1)
+sql create table d1.c2 using st tags(1)
+sql create table d1.c3 using st tags(1)
+sql create table d1.c4 using st tags(1)
+sql create table d1.c5 using st tags(1)
+sql create table d1.c6 using st tags(1)
+sql show d1.tables
+if $rows != 6 then
+  return -1
+endi
+
+sql show d1.vgroups
+print $data[0][0] $data[0][1] $data[0][2] $data[0][3] $data[0][4] 
+if $rows != 2 then
+  return -1
+endi
+if $data(2)[3] != 2 then
+  return -1
+endi
+if $data(3)[3] != 2 then
+  return -1
+endi
+
+print =============== step3: balance vgroup
+# has offline dnode
+sql_error balance vgroup
+
+system sh/exec.sh -n dnode3 -s start
+$x = 0
+step3:
+	$ = $x + 1
+	sleep 1000
+	if $x == 10 then
+	  print ---> dnode not online!
+		return -1
+	endi
+sql show dnodes
+print ---> $data00 $data01 $data02 $data03 $data04 $data05
+print ---> $data10 $data11 $data12 $data13 $data14 $data15
+if $rows != 3 then
+  return -1
+endi
+if $data(1)[4] != ready then
+  goto step3
+endi
+if $data(2)[4] != ready then
+  goto step3
+endi
+if $data(3)[4] != ready then
+  goto step3
+endi
+
+print =============== step4: balance
+sql balance vgroup
+
+print show d1.vgroups
+sql show d1.vgroups
+print $data[0][0] $data[0][1] $data[0][2] $data[0][3] $data[0][4] 
+if $rows != 2 then
+  return -1
+endi
+if $data(2)[3] != 3 then
+  return -1
+endi
+if $data(3)[3] != 2 then
+  return -1
+endi
+
+print =============== step7: select data
+sql show d1.tables
+print rows $rows
+if $rows != 6 then
+  return -1
+endi
+
+return
+system sh/exec.sh -n dnode1 -s stop -x SIGINT
+system sh/exec.sh -n dnode2 -s stop -x SIGINT
+system sh/exec.sh -n dnode3 -s stop -x SIGINT

tests/script/tsim/dnode/drop_dnode_has_vnode_replica3.sim

@@ -28,7 +28,7 @@ step1:
 sql show dnodes
 print ===> $data00 $data01 $data02 $data03 $data04 $data05
 print ===> $data10 $data11 $data12 $data13 $data14 $data15
-if $rows != 3 then
+if $rows != 5 then
   return -1
 endi
 if $data(1)[4] != ready then
@@ -72,6 +72,13 @@ if $data(2)[7] != 4 then
   return -1
 endi
 
+system sh/exec.sh -n dnode4 -s stop -x SIGINT
+system sh/exec.sh -n dnode3 -s stop -x SIGINT
+
+
+return
+
+
 print =============== step4: drop dnode 2
 system sh/exec.sh -n dnode5 -s start
 $x = 0

tests/script/tsim/dnode/redistribute_vgroup_replica3_v1_follower.sim

@@ -176,6 +176,8 @@ if $rows != 1 then
   return -1
 endi
 
+return
+
 print =============== step33: move follower1
 print redistribute vgroup 2 dnode $leaderVnode dnode $follower1 dnode 5 
 sql redistribute vgroup 2 dnode $leaderVnode dnode $follower1 dnode 5 

tests/script/tsim/show/basic.sim

@@ -99,7 +99,7 @@ if $rows != 1 then
 endi
 #sql select * from information_schema.`streams`
 sql select * from information_schema.user_tables
-if $rows != 30 then
+if $rows != 31 then
   return -1
 endi
 #sql select * from information_schema.user_table_distributed
@@ -197,7 +197,7 @@ if $rows != 1 then
 endi
 #sql select * from performance_schema.`streams`
 sql select * from information_schema.user_tables
-if $rows != 30 then
+if $rows != 31 then
   return -1
 endi
 #sql select * from information_schema.user_table_distributed
@@ -227,5 +227,20 @@ endi
 
 sql_error show create stable t0;
 
+sql show variables;
+if $rows != 4 then
+  return -1
+endi
+
+sql show dnode 1 variables;
+if $rows != 114 then
+  return -1
+endi
+
+sql show local variables;
+if $rows != 50 then
+  return -1
+endi
+
 system sh/exec.sh -n dnode1 -s stop -x SIGINT
 system sh/exec.sh -n dnode2 -s stop -x SIGINT

tests/script/tsim/tmq/consume.sh

@@ -17,8 +17,9 @@ VALGRIND=0
 SIGNAL=SIGINT
 SHOW_MSG=0
 SHOW_ROW=0
+EXP_USE_SNAPSHOT=0
 
-while getopts "d:s:v:y:x:g:r:w:" arg
+while getopts "d:s:v:y:x:g:r:w:e:" arg
 do
   case $arg in
     d)
@@ -45,6 +46,9 @@ do
     w)
       CDB_NAME=$OPTARG
       ;;
+    e)
+      EXP_USE_SNAPSHOT=$OPTARG
+      ;;
     ?)
       echo "unkown argument"
       ;;
@@ -91,8 +95,8 @@ if [ "$EXEC_OPTON" = "start" ]; then
     echo nohup valgrind --tool=memcheck --leak-check=full --show-reachable=no --track-origins=yes --show-leak-kinds=all -v --workaround-gcc296-bugs=yes --log-file=${LOG_DIR}/valgrind-tmq_sim.log $PROGRAM -c $CFG_DIR -y $POLL_DELAY -d $DB_NAME -g $SHOW_MSG -r $SHOW_ROW > /dev/null 2>&1 &
     nohup valgrind --tool=memcheck --leak-check=full --show-reachable=no --track-origins=yes --show-leak-kinds=all -v --workaround-gcc296-bugs=yes --log-file=${LOG_DIR}/valgrind-tmq_sim.log $PROGRAM -c $CFG_DIR -y $POLL_DELAY -d $DB_NAME -g $SHOW_MSG -r $SHOW_ROW > /dev/null 2>&1 &
   else
-    echo  "nohup $PROGRAM -c $CFG_DIR -y $POLL_DELAY -d $DB_NAME -g $SHOW_MSG -r $SHOW_ROW -w $CDB_NAME > /dev/null 2>&1 &"
+    echo  "nohup $PROGRAM -c $CFG_DIR -y $POLL_DELAY -d $DB_NAME -g $SHOW_MSG -r $SHOW_ROW -w $CDB_NAME -e $EXP_USE_SNAPSHOT > /dev/null 2>&1 &"
-    nohup $PROGRAM -c $CFG_DIR -y $POLL_DELAY -d $DB_NAME -g $SHOW_MSG -r $SHOW_ROW -w $CDB_NAME > /dev/null 2>&1 &
+    nohup $PROGRAM -c $CFG_DIR -y $POLL_DELAY -d $DB_NAME -g $SHOW_MSG -r $SHOW_ROW -w $CDB_NAME -e $EXP_USE_SNAPSHOT > /dev/null 2>&1 &
   fi
 else
   PID=`ps -ef|grep tmq_sim | grep -v grep | awk '{print $2}'`

tests/script/tsim/tmq/snapshot.sim

@@ -0,0 +1,289 @@
+#### test scenario, please refer to https://jira.taosdata.com:18090/pages/viewpage.action?pageId=135120406
+#basic1.sim: vgroups=1, one topic for one consumer, firstly insert data, then start consume. Include six topics
+#basic2.sim: vgroups=1, multi topics for one consumer, firstly insert data, then start consume. Include six topics
+#basic3.sim: vgroups=4, one topic for one consumer, firstly insert data, then start consume. Include six topics
+#basic4.sim: vgroups=4, multi topics for one consumer, firstly insert data, then start consume. Include six topics
+
+# notes1: Scalar function: ABS/ACOS/ASIN/ATAN/CEIL/COS/FLOOR/LOG/POW/ROUND/SIN/SQRT/TAN
+# The above use cases are combined with where filter conditions, such as: where ts > "2017-08-12 18:25:58.128Z" and sin(a) > 0.5;
+#
+# notes2: not support aggregate functions(such as sum/count/min/max) and time-windows(interval).
+#
+
+run tsim/tmq/prepareBasicEnv-1vgrp.sim
+
+#---- global parameters start ----#
+$dbName    = db
+$vgroups    = 1
+$stbPrefix  = stb
+$ctbPrefix  = ctb
+$ntbPrefix  = ntb
+$stbNum     = 1
+$ctbNum     = 10
+$ntbNum     = 10
+$rowsPerCtb = 10
+$tstart     = 1640966400000  # 2022-01-01 00:00:00.000
+#---- global parameters end ----#
+
+$pullDelay    = 3
+$ifcheckdata  = 1
+$ifmanualcommit = 1
+$showMsg      = 1
+$showRow      = 0
+
+sql connect
+sql use $dbName
+
+print == create topics from super table
+sql create topic topic_stb_column as select ts, c3 from stb
+sql create topic topic_stb_all as select ts, c1, c2, c3 from stb
+sql create topic topic_stb_function as select ts, abs(c1), sin(c2) from stb
+
+print == create topics from child table
+sql create topic topic_ctb_column as select ts, c3 from ctb0
+sql create topic topic_ctb_all as select * from ctb0
+sql create topic topic_ctb_function as select ts, abs(c1), sin(c2) from ctb0
+
+print == create topics from normal table
+sql create topic topic_ntb_column as select ts, c3 from ntb0
+sql create topic topic_ntb_all as select * from ntb0
+sql create topic topic_ntb_function as select ts, abs(c1), sin(c2) from ntb0
+
+#sql show topics
+#if $rows != 9 then 
+#  return -1
+#endi
+
+#'group.id:cgrp1,enable.auto.commit:false,auto.commit.interval.ms:6000,auto.offset.reset:earliest'
+$keyList = ' . group.id:cgrp1
+$keyList = $keyList . ,
+$keyList = $keyList . enable.auto.commit:false
+#$keyList = $keyList . ,
+#$keyList = $keyList . auto.commit.interval.ms:6000
+#$keyList = $keyList . ,
+#$keyList = $keyList . auto.offset.reset:earliest
+$keyList = $keyList . '
+print ========== key list:  $keyList
+
+
+$cdb_index = 0
+#=============================== start consume =============================#
+
+print ================ test consume from stb
+$loop_cnt = 0
+loop_consume_diff_topic_from_stb:
+
+#######################################################################################
+# clear consume info and consume result 
+#run tsim/tmq/clearConsume.sim
+# because drop table function no stable, so by create new db for consume info and result. Modify it later
+$cdb_index = $cdb_index + 1
+$cdbName = cdb . $cdb_index
+sql create database $cdbName vgroups 1
+sleep 500
+sql use $cdbName
+
+print == create consume info table and consume result table
+sql create table consumeinfo (ts timestamp, consumerid int, topiclist binary(1024), keylist binary(1024), expectmsgcnt bigint, ifcheckdata int, ifmanualcommit int)
+sql create table consumeresult (ts timestamp, consumerid int, consummsgcnt bigint, consumrowcnt bigint, checkresult int)
+
+sql show tables
+if $rows != 2 then 
+  return -1
+endi
+#######################################################################################
+
+if $loop_cnt == 0 then
+  print == scenario 1: topic_stb_column
+  $topicList = ' . topic_stb_column
+  $topicList = $topicList . '
+elif $loop_cnt == 1 then
+  print == scenario 2: topic_stb_all
+  $topicList = ' . topic_stb_all
+  $topicList = $topicList . '
+elif $loop_cnt == 2 then
+  print == scenario 3: topic_stb_function
+  $topicList = ' . topic_stb_function
+  $topicList = $topicList . '
+else
+  goto loop_consume_diff_topic_from_stb_end
+endi
+
+$consumerId    = 0
+$totalMsgOfStb = $ctbNum * $rowsPerCtb
+$expectmsgcnt  = 1
+$expectrowcnt  = 100
+sql insert into consumeinfo values (now , $consumerId , $topicList , $keyList , $expectmsgcnt , $ifcheckdata , $ifmanualcommit )
+
+print == start consumer to pull msgs from stb
+print == tsim/tmq/consume.sh -d $dbName -y $pullDelay -g $showMsg -r $showRow -w $cdbName -e 1 -s start
+system tsim/tmq/consume.sh -d $dbName -y $pullDelay -g $showMsg -r $showRow -w $cdbName -e 1 -s start
+
+print == check consume result
+wait_consumer_end_from_stb:
+sql select * from consumeresult
+print ==> rows: $rows 
+print ==> rows[0]: $data[0][0] $data[0][1] $data[0][2] $data[0][3] $data[0][4] $data[0][5] $data[0][6]
+if $rows != 1 then
+  sleep 1000
+  goto wait_consumer_end_from_stb
+endi
+if $data[0][1] != $consumerId then
+  return -1
+endi
+if $data[0][2] != $expectmsgcnt then
+  return -1
+endi
+if $data[0][3] != $expectrowcnt then
+  return -1
+endi
+$loop_cnt = $loop_cnt + 1
+goto loop_consume_diff_topic_from_stb
+loop_consume_diff_topic_from_stb_end:
+
+print ================ test consume from ctb
+$loop_cnt = 0
+loop_consume_diff_topic_from_ctb:
+
+#######################################################################################
+# clear consume info and consume result 
+#run tsim/tmq/clearConsume.sim
+# because drop table function no stable, so by create new db for consume info and result. Modify it later
+$cdb_index = $cdb_index + 1
+$cdbName = cdb . $cdb_index
+sql create database $cdbName vgroups 1
+sleep 500
+sql use $cdbName
+
+print == create consume info table and consume result table
+sql create table consumeinfo (ts timestamp, consumerid int, topiclist binary(1024), keylist binary(1024), expectmsgcnt bigint, ifcheckdata int, ifmanualcommit int)
+sql create table consumeresult (ts timestamp, consumerid int, consummsgcnt bigint, consumrowcnt bigint, checkresult int)
+
+sql show tables
+if $rows != 2 then 
+  return -1
+endi
+#######################################################################################
+
+if $loop_cnt == 0 then
+  print == scenario 1: topic_ctb_column
+  $topicList = ' . topic_ctb_column
+  $topicList = $topicList . '
+elif $loop_cnt == 1 then
+  print == scenario 2: topic_ctb_all
+  $topicList = ' . topic_ctb_all
+  $topicList = $topicList . '
+elif $loop_cnt == 2 then
+  print == scenario 3: topic_ctb_function
+  $topicList = ' . topic_ctb_function
+  $topicList = $topicList . '
+else
+  goto loop_consume_diff_topic_from_ctb_end
+endi
+
+$consumerId    = 0
+$totalMsgOfCtb = $rowsPerCtb
+$expectmsgcnt  = 1
+sql insert into consumeinfo values (now , $consumerId , $topicList , $keyList , $expectmsgcnt , $ifcheckdata , $ifmanualcommit )
+
+print == start consumer to pull msgs from ctb
+print == tsim/tmq/consume.sh -d $dbName -y $pullDelay -g $showMsg -r $showRow -s start -e 1
+system tsim/tmq/consume.sh -d $dbName -y $pullDelay -g $showMsg -r $showRow -w $cdbName -s start -e 1
+
+print == check consume result
+wait_consumer_end_from_ctb:
+sql select * from consumeresult
+print ==> rows: $rows 
+print ==> rows[0]: $data[0][0] $data[0][1] $data[0][2] $data[0][3] $data[0][4] $data[0][5] $data[0][6]
+if $rows != 1 then
+  sleep 1000
+  goto wait_consumer_end_from_ctb
+endi
+if $data[0][1] != $consumerId then
+  return -1
+endi
+if $data[0][2] != 1 then
+  return -1
+endi
+if $data[0][3] != 10 then
+  return -1
+endi
+$loop_cnt = $loop_cnt + 1
+goto loop_consume_diff_topic_from_ctb
+loop_consume_diff_topic_from_ctb_end:
+
+print ================ test consume from ntb
+$loop_cnt = 0
+loop_consume_diff_topic_from_ntb:
+
+#######################################################################################
+# clear consume info and consume result 
+#run tsim/tmq/clearConsume.sim
+# because drop table function no stable, so by create new db for consume info and result. Modify it later
+$cdb_index = $cdb_index + 1
+$cdbName = cdb . $cdb_index
+sql create database $cdbName vgroups 1
+sleep 500
+sql use $cdbName
+
+print == create consume info table and consume result table
+sql create table consumeinfo (ts timestamp, consumerid int, topiclist binary(1024), keylist binary(1024), expectmsgcnt bigint, ifcheckdata int, ifmanualcommit int)
+sql create table consumeresult (ts timestamp, consumerid int, consummsgcnt bigint, consumrowcnt bigint, checkresult int)
+
+sql show tables
+if $rows != 2 then 
+  return -1
+endi
+#######################################################################################
+
+if $loop_cnt == 0 then
+  print == scenario 1: topic_ntb_column
+  $topicList = ' . topic_ntb_column
+  $topicList = $topicList . '
+elif $loop_cnt == 1 then
+  print == scenario 2: topic_ntb_all
+  $topicList = ' . topic_ntb_all
+  $topicList = $topicList . '
+elif $loop_cnt == 2 then
+  print == scenario 3: topic_ntb_function
+  $topicList = ' . topic_ntb_function
+  $topicList = $topicList . '
+else
+  goto loop_consume_diff_topic_from_ntb_end
+endi
+
+$consumerId    = 0
+$totalMsgOfNtb = $rowsPerCtb
+$expectmsgcnt  = $totalMsgOfNtb
+sql insert into consumeinfo values (now , $consumerId , $topicList , $keyList , $expectmsgcnt , $ifcheckdata , $ifmanualcommit )
+
+print == start consumer to pull msgs from ntb
+print == tsim/tmq/consume.sh -d $dbName -y $pullDelay -g $showMsg -r $showRow -s start -e 1
+system tsim/tmq/consume.sh -d $dbName -y $pullDelay -g $showMsg -r $showRow -w $cdbName -s start -e 1
+
+print == check consume result from ntb
+wait_consumer_end_from_ntb:
+sql select * from consumeresult
+print ==> rows: $rows 
+print ==> rows[0]: $data[0][0] $data[0][1] $data[0][2] $data[0][3] $data[0][4] $data[0][5] $data[0][6]
+if $rows != 1 then
+  sleep 1000
+  goto wait_consumer_end_from_ntb
+endi
+if $data[0][1] != $consumerId then
+  return -1
+endi
+if $data[0][2] != 1 then
+  return -1
+endi
+if $data[0][3] != $totalMsgOfNtb then
+  return -1
+endi
+$loop_cnt = $loop_cnt + 1
+goto loop_consume_diff_topic_from_ntb
+loop_consume_diff_topic_from_ntb_end:
+
+#------ not need stop consumer, because it exit after pull msg overthan expect msg
+#system tsim/tmq/consume.sh -s stop -x SIGINT
+
+system sh/exec.sh -n dnode1 -s stop -x SIGINT

tests/script/wtest.bat

@@ -52,7 +52,7 @@ echo wal            0             >> %TAOS_CFG%
 echo asyncLog       0             >> %TAOS_CFG%
 echo locale         en_US.UTF-8   >> %TAOS_CFG%
 echo enableCoreFile 1             >> %TAOS_CFG%
-echo charset        cp65001       >> %TAOS_CFG%
+echo charset        UTF-8         >> %TAOS_CFG%
 
 set "FILE_NAME=testSuite.sim"
 if "%1" == "-f" set "FILE_NAME=%2"

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

@@ -68,7 +68,7 @@ class TDTestCase:
         tdSql.error("CREATE TABLE if not exists jsons1_14 using jsons1 tags('[1,true]')")
         tdSql.error("CREATE TABLE if not exists jsons1_14 using jsons1 tags('{222}')")
         tdSql.error("CREATE TABLE if not exists jsons1_14 using jsons1 tags('{\"fe\"}')")
-        #
+
         # test invalidate json key, key must can be printed assic char
         tdSql.error("CREATE TABLE if not exists jsons1_14 using jsons1 tags('{\"tag1\":[1,true]}')")
         tdSql.error("CREATE TABLE if not exists jsons1_14 using jsons1 tags('{\"tag1\":{}}')")
@@ -79,7 +79,7 @@ class TDTestCase:
         # test invalidate json value, value number can not be inf,nan TD-12166
         tdSql.error("CREATE TABLE if not exists jsons1_14 using jsons1 tags('{\"k\":1.8e308}')")
         tdSql.error("CREATE TABLE if not exists jsons1_14 using jsons1 tags('{\"k\":-1.8e308}')")
-        #
+
         #test length limit
         char1= ''.join(['abcd']*64)
         char3= ''.join(['abcd']*1021)
@@ -87,15 +87,15 @@ class TDTestCase:
         tdSql.error("CREATE TABLE if not exists  jsons1_15 using  jsons1 tags('{\"%s1\":5}')" % char1)   # len(key)=257
         tdSql.execute("CREATE TABLE if not exists  jsons1_15 using  jsons1 tags('{\"%s\":5}')" % char1)  # len(key)=256
         tdSql.error("CREATE TABLE if not exists  jsons1_16 using  jsons1 tags('{\"TSSSS\":\"%s\"}')" % char3)   # len(object)=4096
-        #tdSql.execute("CREATE TABLE if not exists  jsons1_16 using  jsons1 tags('{\"TSSS\":\"%s\"}')" % char3)  # len(object)=4095
+        tdSql.execute("CREATE TABLE if not exists  jsons1_16 using  jsons1 tags('{\"TSSS\":\"%s\"}')" % char3)  # len(object)=4095
         tdSql.execute("drop table if exists jsons1_15")
         tdSql.execute("drop table if exists jsons1_16")
-        #
+
         print("============== STEP 2 ===== alter table json tag")
         tdSql.error("ALTER STABLE jsons1 add tag tag2 nchar(20)")
         tdSql.error("ALTER STABLE jsons1 drop tag jtag")
         tdSql.error("ALTER TABLE jsons1 MODIFY TAG jtag nchar(128)")
-        #
+
         tdSql.execute("ALTER TABLE jsons1_1 SET TAG jtag='{\"tag1\":\"femail\",\"tag2\":35,\"tag3\":true}'")
         tdSql.query("select jtag from jsons1_1")
         tdSql.checkData(0, 0, '{"tag1":"femail","tag2":35,"tag3":true}')
@@ -105,9 +105,9 @@ class TDTestCase:
         tdSql.execute("create table st(ts timestamp, i int) tags(t int)")
         tdSql.error("ALTER STABLE st add tag jtag json")
         tdSql.error("ALTER STABLE st add column jtag json")
-        #
+
-        # print("============== STEP 3 ===== query table")
+        print("============== STEP 3 ===== query table")
-        # # test error syntax
+        # test error syntax
         tdSql.error("select * from jsons1 where jtag->tag1='beijing'")
         tdSql.error("select -> from jsons1")
         tdSql.error("select * from jsons1 where contains")
@@ -115,17 +115,17 @@ class TDTestCase:
         tdSql.error("select jtag->location from jsons1")
         tdSql.error("select jtag contains location from jsons1")
         tdSql.error("select * from jsons1 where jtag contains location")
-        #tdSql.error("select * from jsons1 where jtag contains''")
+        tdSql.query("select * from jsons1 where jtag contains''")
         tdSql.error("select * from jsons1 where jtag contains 'location'='beijing'")
-        #
+
-        # # test function error
+        # test function error
         tdSql.error("select avg(jtag->'tag1') from jsons1")
         tdSql.error("select avg(jtag) from jsons1")
         tdSql.error("select min(jtag->'tag1') from jsons1")
         tdSql.error("select min(jtag) from jsons1")
         tdSql.error("select ceil(jtag->'tag1') from jsons1")
         tdSql.error("select ceil(jtag) from jsons1")
-        #
+
 
         #test scalar operation
         tdSql.query("select jtag contains 'tag1',jtag->'tag1' from jsons1 order by jtag->'tag1'")
@@ -158,10 +158,11 @@ class TDTestCase:
         tdSql.checkData(0, 0, None)
         tdSql.checkData(0, 1, False)
         tdSql.checkData(7, 0, "false")
-        tdSql.checkData(7, 1, True)
+        tdSql.checkData(7, 1, False)
+        tdSql.checkData(8, 1, False)
         tdSql.checkData(12, 1, True)
 
-        # # test select normal column
+        # test select normal column
         tdSql.query("select dataint from jsons1 order by dataint")
         tdSql.checkRows(9)
         tdSql.checkData(1, 0, 1)
@@ -180,7 +181,7 @@ class TDTestCase:
         tdSql.query("select jtag from jsons1_9")
         tdSql.checkData(0, 0, None)
 
-        # # test select json tag->'key', value is string
+        # test select json tag->'key', value is string
         tdSql.query("select jtag->'tag1' from jsons1_1")
         tdSql.checkData(0, 0, '"femail"')
         tdSql.query("select jtag->'tag2' from jsons1_6")
@@ -200,7 +201,7 @@ class TDTestCase:
         # test select json tag->'key', key is not exist
         tdSql.query("select jtag->'tag10' from jsons1_4")
         tdSql.checkData(0, 0, None)
-        #
+
         tdSql.query("select jtag->'tag1' from jsons1")
         tdSql.checkRows(13)
         # test header name
@@ -210,24 +211,25 @@ class TDTestCase:
         tdSql.checkColNameList(res, cname_list)
 
 
-        # # test where with json tag
+        # test where with json tag
         tdSql.query("select * from jsons1_1 where jtag is not null")
-        # tdSql.error("select * from jsons1 where jtag='{\"tag1\":11,\"tag2\":\"\"}'")
+        # tdSql.query("select * from jsons1 where jtag='{\"tag1\":11,\"tag2\":\"\"}'")
-        # tdSql.error("select * from jsons1 where jtag->'tag1'={}")
+        tdSql.error("select * from jsons1 where jtag->'tag1'={}")
-        #
+
-        # # where json value is string
+        # where json value is string
         tdSql.query("select * from jsons1 where jtag->'tag2'='beijing'")
         tdSql.checkRows(2)
-        tdSql.query("select dataint,tbname,jtag->'tag1',jtag from jsons1 where jtag->'tag2'='beijing'")
+        tdSql.query("select dataint,tbname,jtag->'tag1',jtag from jsons1 where jtag->'tag2'='beijing' order by dataint")
         tdSql.checkRows(2)
-	# out of order, cannot compare value
+        tdSql.checkData(0, 0, 2)
-        #tdSql.checkData(0, 0, 2)
+        tdSql.checkData(0, 1, 'jsons1_2')
-        #tdSql.checkData(0, 1, 'jsons1_2')
+        tdSql.checkData(0, 2, "5.000000000")
-        #tdSql.checkData(0, 2, 5)
+        tdSql.checkData(0, 3, '{"tag1":5,"tag2":"beijing"}')
-        #tdSql.checkData(0, 3, '{"tag1":5,"tag2":"beijing"}')
+        tdSql.checkData(1, 0, 3)
-        #tdSql.checkData(1, 0, 3)
+        tdSql.checkData(1, 1, 'jsons1_3')
-        #tdSql.checkData(1, 1, 'jsons1_3')
+        tdSql.checkData(1, 2, 'false')
-        #tdSql.checkData(1, 2, 'false')
+
+
         tdSql.query("select * from jsons1 where jtag->'tag1'='beijing'")
         tdSql.checkRows(0)
         tdSql.query("select * from jsons1 where jtag->'tag1'='收到货'")
@@ -236,72 +238,73 @@ class TDTestCase:
         tdSql.checkRows(1)
         tdSql.query("select * from jsons1 where jtag->'tag2'>='beijing'")
         tdSql.checkRows(3)
-	# open
+        tdSql.query("select * from jsons1 where jtag->'tag2'<'beijing'")
-        #tdSql.query("select * from jsons1 where jtag->'tag2'<'beijing'")
-        #tdSql.checkRows(2)
-        tdSql.query("select * from jsons1 where jtag->'tag2'<='beijing'")
         tdSql.checkRows(2)
+        tdSql.query("select * from jsons1 where jtag->'tag2'<='beijing'")
+        tdSql.checkRows(4)
         tdSql.query("select * from jsons1 where jtag->'tag2'!='beijing'")
-        tdSql.checkRows(5)
+        tdSql.checkRows(3)
         tdSql.query("select * from jsons1 where jtag->'tag2'=''")
         tdSql.checkRows(2)
-        #
+
-        # # where json value is int
+        # where json value is int
         tdSql.query("select * from jsons1 where jtag->'tag1'=5")
         tdSql.checkRows(1)
         tdSql.checkData(0, 1, 2)
         tdSql.query("select * from jsons1 where jtag->'tag1'=10")
         tdSql.checkRows(0)
         tdSql.query("select * from jsons1 where jtag->'tag1'<54")
-        tdSql.checkRows(4)
+        tdSql.checkRows(3)
         tdSql.query("select * from jsons1 where jtag->'tag1'<=11")
-        tdSql.checkRows(4)
+        tdSql.checkRows(3)
         tdSql.query("select * from jsons1 where jtag->'tag1'>4")
         tdSql.checkRows(2)
         tdSql.query("select * from jsons1 where jtag->'tag1'>=5")
         tdSql.checkRows(2)
         tdSql.query("select * from jsons1 where jtag->'tag1'!=5")
-        tdSql.checkRows(6)
+        tdSql.checkRows(2)
         tdSql.query("select * from jsons1 where jtag->'tag1'!=55")
-        tdSql.checkRows(7)
+        tdSql.checkRows(3)
-        #
+
-        # # where json value is double
+        # where json value is double
         tdSql.query("select * from jsons1 where jtag->'tag1'=1.232")
         tdSql.checkRows(1)
         tdSql.query("select * from jsons1 where jtag->'tag1'<1.232")
-        tdSql.checkRows(1)
+        tdSql.checkRows(0)
         tdSql.query("select * from jsons1 where jtag->'tag1'<=1.232")
-        tdSql.checkRows(2)
+        tdSql.checkRows(1)
         tdSql.query("select * from jsons1 where jtag->'tag1'>1.23")
         tdSql.checkRows(3)
         tdSql.query("select * from jsons1 where jtag->'tag1'>=1.232")
         tdSql.checkRows(3)
         tdSql.query("select * from jsons1 where jtag->'tag1'!=1.232")
-        tdSql.checkRows(6)
+        tdSql.checkRows(2)
         tdSql.query("select * from jsons1 where jtag->'tag1'!=3.232")
-        tdSql.checkRows(7)
+        tdSql.checkRows(3)
-        #tdSql.error("select * from jsons1 where jtag->'tag1'/0=3")
+        tdSql.query("select * from jsons1 where jtag->'tag1'/0=3")
-        #tdSql.error("select * from jsons1 where jtag->'tag1'/5=1")
+        tdSql.checkRows(0)
-        #
+        tdSql.query("select * from jsons1 where jtag->'tag1'/5=1")
-        # # where json value is bool
+        tdSql.checkRows(1)
+
+        # where json value is bool
         tdSql.query("select * from jsons1 where jtag->'tag1'=true")
         tdSql.checkRows(0)
-        #tdSql.query("select * from jsons1 where jtag->'tag1'=false")
+        tdSql.query("select * from jsons1 where jtag->'tag1'=false")
-        #tdSql.checkRows(1)
+        tdSql.checkRows(1)
         tdSql.query("select * from jsons1 where jtag->'tag1'!=false")
-        tdSql.checkRows(3)
+        tdSql.checkRows(0)
-        #tdSql.error("select * from jsons1 where jtag->'tag1'>false")
+        tdSql.query("select * from jsons1 where jtag->'tag1'>false")
-        #
+        tdSql.checkRows(0)
-        # # where json value is null
+
-	# open
+        # where json value is null
-        #tdSql.query("select * from jsons1 where jtag->'tag1'=null")     # only json suport =null. This synatx will change later.
+        tdSql.query("select * from jsons1 where jtag->'tag1'=null")
-        #tdSql.checkRows(1)
+        tdSql.checkRows(0)
-        #
+
-        # # where json key is null
+        # where json key is null
         tdSql.query("select * from jsons1 where jtag->'tag_no_exist'=3")
         tdSql.checkRows(0)
-        #
+
-        # # where json value is not exist
+        # where json value is not exist
         tdSql.query("select * from jsons1 where jtag->'tag1' is null")
         tdSql.checkData(0, 0, 'jsons1_9')
         tdSql.checkRows(2)
@@ -309,16 +312,16 @@ class TDTestCase:
         tdSql.checkRows(9)
         tdSql.query("select * from jsons1 where jtag->'tag3' is not null")
         tdSql.checkRows(3)
-        #
+
-        # # test contains
+        # test contains
         tdSql.query("select * from jsons1 where jtag contains 'tag1'")
-        tdSql.checkRows(7)
+        tdSql.checkRows(8)
         tdSql.query("select * from jsons1 where jtag contains 'tag3'")
-        tdSql.checkRows(3)
+        tdSql.checkRows(4)
         tdSql.query("select * from jsons1 where jtag contains 'tag_no_exist'")
         tdSql.checkRows(0)
-        #
+
-        # # test json tag in where condition with and/or
+        # test json tag in where condition with and/or
         tdSql.query("select * from jsons1 where jtag->'tag1'=false and jtag->'tag2'='beijing'")
         tdSql.checkRows(1)
         tdSql.query("select * from jsons1 where jtag->'tag1'=false or jtag->'tag2'='beijing'")
@@ -335,15 +338,15 @@ class TDTestCase:
         tdSql.checkRows(3)
         tdSql.query("select * from jsons1 where jtag->'tag1'='femail' and jtag contains 'tag3'")
         tdSql.checkRows(2)
-        #
+
-        #
+
-        # # test with between and
+        # test with between and
         tdSql.query("select * from jsons1 where jtag->'tag1' between 1 and 30")
         tdSql.checkRows(3)
         tdSql.query("select * from jsons1 where jtag->'tag1' between 'femail' and 'beijing'")
         tdSql.checkRows(2)
-        #
+
-        # # test with tbname/normal column
+        # test with tbname/normal column
         tdSql.query("select * from jsons1 where tbname = 'jsons1_1'")
         tdSql.checkRows(2)
         tdSql.query("select * from jsons1 where tbname = 'jsons1_1' and jtag contains 'tag3'")
@@ -352,20 +355,18 @@ class TDTestCase:
         tdSql.checkRows(0)
         tdSql.query("select * from jsons1 where tbname = 'jsons1_1' and jtag contains 'tag3' and dataint=23")
         tdSql.checkRows(1)
-        #
+
-        #
+
-        # # test where condition like
+        # test where condition like
-	# open
+        tdSql.query("select * from jsons1 where jtag->'tag2' like 'bei%'")
- 	# syntax error	
+        tdSql.checkRows(2)
-        #tdSql.query("select *,tbname from jsons1 where jtag->'tag2' like 'bei%'")
+        tdSql.query("select * from jsons1 where jtag->'tag1' like 'fe%' and jtag->'tag2' is not null")
-        #tdSql.checkRows(2)
+        tdSql.checkRows(2)
-        #tdSql.query("select *,tbname from jsons1 where jtag->'tag1' like 'fe%' and jtag->'tag2' is not null")
+
-        #tdSql.checkRows(2)
+        # test where condition in  no support in
-        #
-        # # test where condition in  no support in
         # tdSql.error("select * from jsons1 where jtag->'tag1' in ('beijing')")
-        #
+
-        # # test where condition match/nmath
+        # test where condition match/nmath
         tdSql.query("select * from jsons1 where jtag->'tag1' match 'ma'")
         tdSql.checkRows(2)
         tdSql.query("select * from jsons1 where jtag->'tag1' match 'ma$'")
@@ -376,23 +377,22 @@ class TDTestCase:
         tdSql.checkRows(1)
         tdSql.query("select * from jsons1 where jtag->'tag1' nmatch 'ma'")
         tdSql.checkRows(1)
-        #
+
-        # # test distinct
+        # test distinct
         tdSql.execute("insert into jsons1_14 using jsons1 tags('{\"tag1\":\"收到货\",\"tag2\":\"\",\"tag3\":null}') values(1591062628000, 2, NULL, '你就会', 'dws')")
         tdSql.query("select distinct jtag->'tag1' from jsons1")
         tdSql.checkRows(8)
         tdSql.query("select distinct jtag from jsons1")
         tdSql.checkRows(9)
-        #
+
-        # #test dumplicate key with normal colomn
+        #test dumplicate key with normal colomn
         tdSql.execute("INSERT INTO jsons1_15 using jsons1 tags('{\"tbname\":\"tt\",\"databool\":true,\"datastr\":\"是是是\"}') values(1591060828000, 4, false, 'jjsf', \"你就会\")")
-        #tdSql.query("select *,tbname,jtag from jsons1 where jtag->'datastr' match '是' and datastr match 'js'")
+        tdSql.query("select * from jsons1 where jtag->'datastr' match '是' and datastr match 'js'")
-        #tdSql.checkRows(1)
+        tdSql.checkRows(1)
-	# open
+        # tdSql.query("select tbname,jtag->'tbname' from jsons1 where jtag->'tbname'='tt' and tbname='jsons1_14'")
-        #tdSql.query("select tbname,jtag->'tbname' from jsons1 where jtag->'tbname'='tt' and tbname='jsons1_14'")
+        # tdSql.checkRows(1)
-        #tdSql.checkRows(0)
+
-        #
+        # test join
-        # # test join
         tdSql.execute("create table if not exists jsons2(ts timestamp, dataInt int, dataBool bool, dataStr nchar(50), dataStrBin binary(150)) tags(jtag json)")
         tdSql.execute("insert into jsons2_1 using jsons2 tags('{\"tag1\":\"fff\",\"tag2\":5, \"tag3\":true}') values(1591060618000, 2, false, 'json2', '你是2')")
         tdSql.execute("insert into jsons2_2 using jsons2 tags('{\"tag1\":5,\"tag2\":null}') values (1591060628000, 2, true, 'json2', 'sss')")
@@ -460,19 +460,18 @@ class TDTestCase:
         tdSql.checkColNameList(res, cname_list)
 
         # test top/bottom with group by json tag
-	# random failure 
+        # tdSql.query("select top(dataint,2),jtag->'tag1' from jsons1 group by jtag->'tag1' order by jtag->'tag1'")
-        #tdSql.query("select top(dataint,2),jtag->'tag1' from jsons1 group by jtag->'tag1' order by jtag->'tag1'")
+        # tdSql.checkRows(11)
-        #tdSql.checkRows(11)
+        # tdSql.checkData(0, 1, None)
-        #tdSql.checkData(0, 1, None)
+        # tdSql.checkData(2, 0, 4)
-        #tdSql.checkData(2, 0, 4)
+        # tdSql.checkData(3, 0, 3)
-        #tdSql.checkData(3, 0, 3)
+        # tdSql.checkData(3, 1, "false")
-        #tdSql.checkData(3, 1, "false")
+        # tdSql.checkData(8, 0, 2)
-        #tdSql.checkData(10, 0, 23)
+        # tdSql.checkData(10, 1, '"femail"')
-        #tdSql.checkData(10, 1, '"femail"')
 
         # test having
-        #tdSql.query("select count(*),jtag->'tag1' from jsons1 group by jtag->'tag1' having count(*) > 1")
+        tdSql.query("select count(*),jtag->'tag1' from jsons1 group by jtag->'tag1' having count(*) > 1")
-        #tdSql.checkRows(3)
+        tdSql.checkRows(3)
 
         # subquery with json tag
         tdSql.query("select * from (select jtag, dataint from jsons1) order by dataint")
@@ -480,22 +479,13 @@ class TDTestCase:
         tdSql.checkData(1, 1, 1)
         tdSql.checkData(5, 0, '{"tag1":false,"tag2":"beijing"}')
 
-        # tdSql.query("select jtag->'tag1' from (select jtag->'tag1', dataint from jsons1)")
+        tdSql.error("select jtag->'tag1' from (select jtag->'tag1', dataint from jsons1)")
-        # tdSql.checkRows(11)
+        # tdSql.query("select ts,jtag->'tag1' from (select jtag->'tag1',tbname,ts from jsons1 order by ts)")
-        # tdSql.checkData(1, 0, '"femail"')
-        # tdSql.checkData(2, 0, 5)
-        #
-        # res = tdSql.getColNameList("select jtag->'tag1' from (select jtag->'tag1', dataint from jsons1)")
-        # cname_list = []
-        # cname_list.append("jtag->'tag1'")
-        # tdSql.checkColNameList(res, cname_list)
-        #
-        # tdSql.query("select ts,tbname,jtag->'tag1' from (select jtag->'tag1',tbname,ts from jsons1 order by ts)")
         # tdSql.checkRows(11)
         # tdSql.checkData(1, 1, "jsons1_1")
         # tdSql.checkData(1, 2, '"femail"')
-        #
+
-        # # union all
+        # union all
         tdSql.query("select jtag->'tag1' from jsons1 union all select jtag->'tag2' from jsons2")
         tdSql.checkRows(17)
         tdSql.query("select jtag->'tag1' from jsons1_1 union all select jtag->'tag2' from jsons2_1")
@@ -508,24 +498,25 @@ class TDTestCase:
         tdSql.query("select dataint,jtag,tbname from jsons1 union all select dataint,jtag,tbname from jsons2")
         tdSql.checkRows(13)
 
-        # #show create table
+        #show create table
-        # tdSql.query("show create table jsons1")
+        tdSql.query("show create table jsons1")
-        # tdSql.checkData(0, 1, 'CREATE TABLE `jsons1` (`ts` TIMESTAMP,`dataint` INT,`databool` BOOL,`datastr` NCHAR(50),`datastrbin` BINARY(150)) TAGS (`jtag` JSON)')
+        tdSql.checkData(0, 1, 'CREATE STABLE `jsons1` (`ts` TIMESTAMP, `dataint` INT, `databool` BOOL, `datastr` NCHAR(50), `datastrbin` VARCHAR(150)) TAGS (`jtag` JSON) WATERMARK 5000a, 5000a')
-        #
+
-        # #test aggregate function:count/avg/twa/irate/sum/stddev/leastsquares
+        #test aggregate function:count/avg/twa/irate/sum/stddev/leastsquares
         tdSql.query("select count(*) from jsons1 where jtag is not null")
         tdSql.checkData(0, 0, 10)
         tdSql.query("select avg(dataint) from jsons1 where jtag is not null")
         tdSql.checkData(0, 0, 5.3)
-        #tdSql.error("select twa(dataint) from jsons1 where jtag is not null")
+        # tdSql.query("select twa(dataint) from jsons1 where jtag is not null")
-        tdSql.query("select irate(dataint) from jsons1 where jtag is not null")
+        # tdSql.checkData(0, 0, 36)
-        #tdSql.query("select sum(dataint) from jsons1 where jtag->'tag1' is not null")
+        # tdSql.error("select irate(dataint) from jsons1 where jtag is not null")
-        #tdSql.checkData(0, 0, 49)
+        tdSql.query("select sum(dataint) from jsons1 where jtag->'tag1' is not null")
+        tdSql.checkData(0, 0, 45)
         tdSql.query("select stddev(dataint) from jsons1 where jtag->'tag1'>1")
         tdSql.checkData(0, 0, 4.496912521)
-        #tdSql.error("SELECT LEASTSQUARES(dataint, 1, 1) from jsons1 where jtag is not null")
+        tdSql.query("SELECT LEASTSQUARES(dataint, 1, 1) from jsons1 where jtag is not null")
-        #
+
-        # #test selection function:min/max/first/last/top/bottom/percentile/apercentile/last_row/interp
+        #test selection function:min/max/first/last/top/bottom/percentile/apercentile/last_row/interp
         tdSql.query("select min(dataint) from jsons1 where jtag->'tag1'>1")
         tdSql.checkData(0, 0, 1)
         tdSql.query("select max(dataint) from jsons1 where jtag->'tag1'>1")
@@ -541,13 +532,16 @@ class TDTestCase:
         tdSql.query("select percentile(dataint,20) from jsons1 where jtag->'tag1'>1")
         tdSql.query("select apercentile(dataint, 50) from jsons1 where jtag->'tag1'>1")
         tdSql.checkData(0, 0, 1.5)
-        #tdSql.query("select last_row(dataint) from jsons1 where jtag->'tag1'>1")
+        # tdSql.query("select last_row(dataint) from jsons1 where jtag->'tag1'>1")
-        #tdSql.checkData(0, 0, 11)
+        # tdSql.query("select interp(dataint) from jsons1 where ts = '2020-06-02 09:17:08.000' and jtag->'tag1'>1")
-        #tdSql.error("select interp(dataint) from jsons1 where ts = '2020-06-02 09:17:08.000' and jtag->'tag1'>1")
+
-        #
+        #test calculation function:diff/derivative/spread/ceil/floor/round/
-        # #test calculation function:diff/derivative/spread/ceil/floor/round/
+        tdSql.query("select diff(dataint) from jsons1 where jtag->'tag1'>1")
-        #tdSql.error("select diff(dataint) from jsons1 where jtag->'tag1'>1")
+        # tdSql.checkRows(2)
-        #tdSql.error("select derivative(dataint, 10m, 0) from jsons1 where jtag->'tag1'>1")
+        # tdSql.checkData(0, 0, -1)
+        # tdSql.checkData(1, 0, 10)
+        tdSql.query("select derivative(dataint, 10m, 0) from jsons1 where jtag->'tag1'>1")
+        tdSql.checkData(0, 0, -2)
         tdSql.query("select spread(dataint) from jsons1 where jtag->'tag1'>1")
         tdSql.checkData(0, 0, 10)
         tdSql.query("select ceil(dataint) from jsons1 where jtag->'tag1'>1")

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

@@ -0,0 +1,182 @@
+
+import taos
+import sys
+import time
+import socket
+import os
+import threading
+
+from util.log import *
+from util.sql import *
+from util.cases import *
+from util.dnodes import *
+from util.common import *
+sys.path.append("./7-tmq")
+from tmqCommon import *
+
+class TDTestCase:
+    def init(self, conn, logSql):
+        tdLog.debug(f"start to excute {__file__}")
+        tdSql.init(conn.cursor())
+        #tdSql.init(conn.cursor(), logSql)  # output sql.txt file
+
+    def checkFileContent(self, consumerId, queryString):
+        buildPath = tdCom.getBuildPath()
+        cfgPath = tdCom.getClientCfgPath()
+        dstFile = '%s/../log/dstrows_%d.txt'%(cfgPath, consumerId)
+        cmdStr = '%s/build/bin/taos -c %s -s "%s >> %s"'%(buildPath, cfgPath, queryString, dstFile)
+        tdLog.info(cmdStr)
+        os.system(cmdStr)
+        
+        consumeRowsFile = '%s/../log/consumerid_%d.txt'%(cfgPath, consumerId)
+        tdLog.info("rows file: %s, %s"%(consumeRowsFile, dstFile))
+
+        consumeFile = open(consumeRowsFile, mode='r')
+        queryFile = open(dstFile, mode='r')
+        
+        # skip first line for it is schema
+        queryFile.readline()
+
+        while True:
+            dst = queryFile.readline()
+            src = consumeFile.readline()
+            
+            if dst:
+                if dst != src:
+                    tdLog.exit("consumerId %d consume rows is not match the rows by direct query"%consumerId)
+            else:
+                break
+        return 
+
+    def tmqCase1(self):
+        tdLog.printNoPrefix("======== test case 1: ")
+        paraDict = {'dbName':     'db1',
+                    'dropFlag':   1,
+                    'event':      '',
+                    'vgroups':    4,
+                    'stbName':    'stb',
+                    'colPrefix':  'c',
+                    'tagPrefix':  't',
+                    'colSchema':   [{'type': 'INT', 'count':1}, {'type': 'binary', 'len':20, 'count':1}],
+                    'tagSchema':   [{'type': 'INT', 'count':1}, {'type': 'binary', 'len':20, 'count':1}],
+                    'ctbPrefix':  'ctb',
+                    'ctbNum':     1,
+                    'rowsPerTbl': 10000,
+                    'batchNum':   10,
+                    'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
+                    'pollDelay':  10,
+                    'showMsg':    1,
+                    'showRow':    1}
+
+        topicNameList = ['topic1', 'topic2', 'topic3']
+        expectRowsList = []
+        tmqCom.initConsumerTable()
+        tdCom.create_database(tdSql, paraDict["dbName"],paraDict["dropFlag"], vgroups=4,replica=1)
+        tdLog.info("create stb")
+        tdCom.create_stable(tdSql, dbname=paraDict["dbName"],stbname=paraDict["stbName"], column_elm_list=paraDict['colSchema'], tag_elm_list=paraDict['tagSchema'])
+        tdLog.info("create ctb")
+        tdCom.create_ctable(tdSql, dbname=paraDict["dbName"],stbname=paraDict["stbName"],tag_elm_list=paraDict['tagSchema'],count=paraDict["ctbNum"], default_ctbname_prefix=paraDict['ctbPrefix'])
+        tdLog.info("insert data")
+        tmqCom.insert_data(tdSql,paraDict["dbName"],paraDict["ctbPrefix"],paraDict["ctbNum"],paraDict["rowsPerTbl"],paraDict["batchNum"],paraDict["startTs"])
+        
+        tdLog.info("create topics from stb with filter")
+        queryString = "select ts, log(c1), ceil(pow(c1,3)) from %s.%s where c1 %% 7 == 0" %(paraDict['dbName'], paraDict['stbName'])
+        sqlString = "create topic %s as %s" %(topicNameList[0], queryString)
+        tdLog.info("create topic sql: %s"%sqlString)
+        tdSql.execute(sqlString)
+        tdSql.query(queryString)        
+        expectRowsList.append(tdSql.getRows())
+
+        # init consume info, and start tmq_sim, then check consume result
+        tdLog.info("insert consume info to consume processor")
+        consumerId   = 0
+        expectrowcnt = paraDict["rowsPerTbl"] * paraDict["ctbNum"]
+        topicList    = topicNameList[0]
+        ifcheckdata  = 1
+        ifManualCommit = 1
+        keyList      = 'group.id:cgrp1, enable.auto.commit:false, auto.commit.interval.ms:6000, auto.offset.reset:earliest'
+        tmqCom.insertConsumerInfo(consumerId, expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+
+        tdLog.info("start consume processor")
+        tmqCom.startTmqSimProcess(paraDict['pollDelay'],paraDict["dbName"],paraDict['showMsg'], paraDict['showRow'])
+
+        tdLog.info("wait the consume result") 
+        expectRows = 1
+        resultList = tmqCom.selectConsumeResult(expectRows)
+        
+        if expectRowsList[0] != resultList[0]:
+            tdLog.info("expect consume rows: %d, act consume rows: %d"%(expectRowsList[0], resultList[0]))
+            tdLog.exit("0 tmq consume rows error!")
+
+        self.checkFileContent(consumerId, queryString)     
+
+        # reinit consume info, and start tmq_sim, then check consume result
+        tmqCom.initConsumerTable()
+
+        queryString = "select ts, log(c1), cos(c1) from %s.%s where c1 > 3169" %(paraDict['dbName'], paraDict['stbName'])
+        sqlString = "create topic %s as %s" %(topicNameList[1], queryString)
+        tdLog.info("create topic sql: %s"%sqlString)
+        tdSql.execute(sqlString)
+        tdSql.query(queryString)        
+        expectRowsList.append(tdSql.getRows())
+
+        consumerId   = 1
+        topicList    = topicNameList[1]
+        tmqCom.insertConsumerInfo(consumerId, expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+
+        tdLog.info("start consume processor")
+        tmqCom.startTmqSimProcess(paraDict['pollDelay'],paraDict["dbName"],paraDict['showMsg'], paraDict['showRow'])
+
+        tdLog.info("wait the consume result") 
+        expectRows = 1
+        resultList = tmqCom.selectConsumeResult(expectRows)
+        if expectRowsList[1] != resultList[0]:
+            tdLog.info("expect consume rows: %d, act consume rows: %d"%(expectRowsList[1], resultList[0]))
+            tdLog.exit("1 tmq consume rows error!")
+
+        self.checkFileContent(consumerId, queryString)
+
+        # reinit consume info, and start tmq_sim, then check consume result
+        tmqCom.initConsumerTable()
+
+        queryString = "select ts, log(c1), atan(c1) from %s.%s where ts >= %d" %(paraDict['dbName'], paraDict['stbName'], paraDict["startTs"]+6137)
+        sqlString = "create topic %s as %s" %(topicNameList[2], queryString)
+        tdLog.info("create topic sql: %s"%sqlString)
+        tdSql.execute(sqlString)
+        tdSql.query(queryString)        
+        expectRowsList.append(tdSql.getRows())                
+
+        consumerId   = 2
+        topicList    = topicNameList[2]
+        tmqCom.insertConsumerInfo(consumerId, expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+
+        tdLog.info("start consume processor")
+        tmqCom.startTmqSimProcess(paraDict['pollDelay'],paraDict["dbName"],paraDict['showMsg'], paraDict['showRow'])
+
+        tdLog.info("wait the consume result") 
+        expectRows = 1
+        resultList = tmqCom.selectConsumeResult(expectRows)
+        # if expectRowsList[2] != resultList[0]:
+        #     tdLog.info("expect consume rows: %d, act consume rows: %d"%(expectRowsList[2], resultList[0]))
+        #     tdLog.exit("2 tmq consume rows error!")
+
+        # self.checkFileContent(consumerId, queryString)
+
+        time.sleep(10)        
+        for i in range(len(topicNameList)):
+            tdSql.query("drop topic %s"%topicNameList[i])
+
+        tdLog.printNoPrefix("======== test case 1 end ...... ")
+
+    def run(self):
+        tdSql.prepare()
+        self.tmqCase1()
+
+    def stop(self):
+        tdSql.close()
+        tdLog.success(f"{__file__} successfully executed")
+
+event = threading.Event()
+
+tdCases.addLinux(__file__, TDTestCase())
+tdCases.addWindows(__file__, TDTestCase())

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

@@ -0,0 +1,194 @@
+from distutils.log import error
+import taos
+import sys
+import time
+import socket
+import os
+import threading
+import subprocess
+import platform
+
+from util.log import *
+from util.sql import *
+from util.cases import *
+from util.dnodes import *
+from util.common import *
+sys.path.append("./7-tmq")
+from tmqCommon import *
+
+class TDTestCase:
+    def init(self, conn, logSql):
+        tdLog.debug(f"start to excute {__file__}")
+        tdSql.init(conn.cursor())
+        #tdSql.init(conn.cursor(), logSql)  # output sql.txt file
+    
+    def prepare_udf_so(self):
+        selfPath = os.path.dirname(os.path.realpath(__file__))
+
+        if ("community" in selfPath):
+            projPath = selfPath[:selfPath.find("community")]
+        else:
+            projPath = selfPath[:selfPath.find("tests")]
+        print(projPath)
+
+        if platform.system().lower() == 'windows':
+            self.libudf1 = subprocess.Popen('(for /r %s %%i in ("udf1.d*") do @echo %%i)|grep lib|head -n1'%projPath , shell=True, stdout=subprocess.PIPE,stderr=subprocess.STDOUT).stdout.read().decode("utf-8")
+            if (not tdDnodes.dnodes[0].remoteIP == ""):
+                tdDnodes.dnodes[0].remote_conn.get(tdDnodes.dnodes[0].config["path"]+'/debug/build/lib/libudf1.so',projPath+"\\debug\\build\\lib\\")
+                self.libudf1 = self.libudf1.replace('udf1.dll','libudf1.so')
+        else:
+            self.libudf1 = subprocess.Popen('find %s -name "libudf1.so"|grep lib|head -n1'%projPath , shell=True, stdout=subprocess.PIPE,stderr=subprocess.STDOUT).stdout.read().decode("utf-8")
+        self.libudf1 = self.libudf1.replace('\r','').replace('\n','')
+        return
+
+    def create_udf_function(self):
+        # create  scalar functions
+        tdSql.execute("create function udf1 as '%s' outputtype int bufSize 8;"%self.libudf1)
+
+        functions = tdSql.getResult("show functions")
+        function_nums = len(functions)
+        if function_nums == 1:
+            tdLog.info("create one udf functions success ")
+        else:
+            tdLog.exit("create udf functions fail")
+        return
+
+    def checkFileContent(self, consumerId, queryString):
+        buildPath = tdCom.getBuildPath()
+        cfgPath = tdCom.getClientCfgPath()
+        dstFile = '%s/../log/dstrows_%d.txt'%(cfgPath, consumerId)
+        cmdStr = '%s/build/bin/taos -c %s -s "%s >> %s"'%(buildPath, cfgPath, queryString, dstFile)
+        tdLog.info(cmdStr)
+        os.system(cmdStr)
+        
+        consumeRowsFile = '%s/../log/consumerid_%d.txt'%(cfgPath, consumerId)
+        tdLog.info("rows file: %s, %s"%(consumeRowsFile, dstFile))
+
+        consumeFile = open(consumeRowsFile, mode='r')
+        queryFile = open(dstFile, mode='r')
+        
+        # skip first line for it is schema
+        queryFile.readline()
+
+        while True:
+            dst = queryFile.readline()
+            src = consumeFile.readline()
+            
+            if dst:
+                if dst != src:
+                    tdLog.exit("consumerId %d consume rows is not match the rows by direct query"%consumerId)
+            else:
+                break
+        return 
+
+    def tmqCase1(self):
+        tdLog.printNoPrefix("======== test case 1: ")
+        paraDict = {'dbName':     'db1',
+                    'dropFlag':   1,
+                    'event':      '',
+                    'vgroups':    4,
+                    'stbName':    'stb',
+                    'colPrefix':  'c',
+                    'tagPrefix':  't',
+                    'colSchema':   [{'type': 'INT', 'count':2}, {'type': 'binary', 'len':20, 'count':1}],
+                    'tagSchema':   [{'type': 'INT', 'count':1}, {'type': 'binary', 'len':20, 'count':1}],
+                    'ctbPrefix':  'ctb',
+                    'ctbNum':     1,
+                    'rowsPerTbl': 1000,
+                    'batchNum':   10,
+                    'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
+                    'pollDelay':  10,
+                    'showMsg':    1,
+                    'showRow':    1}
+
+        topicNameList = ['topic1', 'topic2']
+        expectRowsList = []
+        tmqCom.initConsumerTable()
+        tdCom.create_database(tdSql, paraDict["dbName"],paraDict["dropFlag"], vgroups=4,replica=1)
+        tdLog.info("create stb")
+        tdCom.create_stable(tdSql, dbname=paraDict["dbName"],stbname=paraDict["stbName"], column_elm_list=paraDict['colSchema'], tag_elm_list=paraDict['tagSchema'])
+        tdLog.info("create ctb")
+        tdCom.create_ctable(tdSql, dbname=paraDict["dbName"],stbname=paraDict["stbName"],tag_elm_list=paraDict['tagSchema'],count=paraDict["ctbNum"], default_ctbname_prefix=paraDict['ctbPrefix'])
+        tdLog.info("insert data")
+        tmqCom.insert_data_1(tdSql,paraDict["dbName"],paraDict["ctbPrefix"],paraDict["ctbNum"],paraDict["rowsPerTbl"],paraDict["batchNum"],paraDict["startTs"])
+        
+        tdLog.info("create topics from stb with filter")
+        queryString = "select ts, c1,udf1(c1),c2,udf1(c2) from %s.%s where c1 %% 7 == 0" %(paraDict['dbName'], paraDict['stbName'])
+        sqlString = "create topic %s as %s" %(topicNameList[0], queryString)
+        tdLog.info("create topic sql: %s"%sqlString)
+        tdSql.execute(sqlString)
+        tdSql.query(queryString)        
+        expectRowsList.append(tdSql.getRows())
+
+        # init consume info, and start tmq_sim, then check consume result
+        tdLog.info("insert consume info to consume processor")
+        consumerId   = 0
+        expectrowcnt = paraDict["rowsPerTbl"] * paraDict["ctbNum"]
+        topicList    = topicNameList[0]
+        ifcheckdata  = 1
+        ifManualCommit = 1
+        keyList      = 'group.id:cgrp1, enable.auto.commit:false, auto.commit.interval.ms:6000, auto.offset.reset:earliest'
+        tmqCom.insertConsumerInfo(consumerId, expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+
+        tdLog.info("start consume processor")
+        tmqCom.startTmqSimProcess(paraDict['pollDelay'],paraDict["dbName"],paraDict['showMsg'], paraDict['showRow'])
+
+        tdLog.info("wait the consume result") 
+        expectRows = 1
+        resultList = tmqCom.selectConsumeResult(expectRows)
+        
+        if expectRowsList[0] != resultList[0]:
+            tdLog.info("expect consume rows: %d, act consume rows: %d"%(expectRowsList[0], resultList[0]))
+            tdLog.exit("0 tmq consume rows error!")
+
+        self.checkFileContent(consumerId, queryString)
+        tdLog.printNoPrefix("consumerId %d check data ok!"%(consumerId))
+
+
+        # reinit consume info, and start tmq_sim, then check consume result
+        tmqCom.initConsumerTable()
+
+        queryString = "select ts, c1,udf1(c1),sin(udf1(c2)), log(udf1(c2)) from %s.%s where udf1(c1) == 88 or sin(udf1(c1)) > 0" %(paraDict['dbName'], paraDict['stbName'])
+        sqlString = "create topic %s as %s" %(topicNameList[1], queryString)
+        tdLog.info("create topic sql: %s"%sqlString)
+        tdSql.execute(sqlString)
+        tdSql.query(queryString)        
+        expectRowsList.append(tdSql.getRows())
+
+        consumerId   = 1
+        topicList    = topicNameList[1]
+        tmqCom.insertConsumerInfo(consumerId, expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+
+        tdLog.info("start consume processor")
+        tmqCom.startTmqSimProcess(paraDict['pollDelay'],paraDict["dbName"],paraDict['showMsg'], paraDict['showRow'])
+
+        tdLog.info("wait the consume result") 
+        expectRows = 1
+        resultList = tmqCom.selectConsumeResult(expectRows)
+        if expectRowsList[1] != resultList[0]:
+            tdLog.info("expect consume rows: %d, act consume rows: %d"%(expectRowsList[1], resultList[0]))
+            tdLog.exit("1 tmq consume rows error!")
+
+        self.checkFileContent(consumerId, queryString)
+        tdLog.printNoPrefix("consumerId %d check data ok!"%(consumerId))
+
+        time.sleep(10)        
+        for i in range(len(topicNameList)):
+            tdSql.query("drop topic %s"%topicNameList[i])
+
+        tdLog.printNoPrefix("======== test case 1 end ...... ")
+
+    def run(self):
+        tdSql.prepare()
+        self.prepare_udf_so()
+        self.create_udf_function()
+        self.tmqCase1()
+
+    def stop(self):
+        tdSql.close()
+        tdLog.success(f"{__file__} successfully executed")
+
+event = threading.Event()
+
+tdCases.addLinux(__file__, TDTestCase())
+tdCases.addWindows(__file__, TDTestCase())

tests/system-test/fulltest.sh

@@ -132,3 +132,5 @@ python3 ./test.py -f 7-tmq/db.py
 python3 ./test.py -f 7-tmq/tmqError.py
 python3 ./test.py -f 7-tmq/schema.py
 python3 ./test.py -f 7-tmq/stbFilter.py
+python3 ./test.py -f 7-tmq/tmqCheckData.py
+python3 ./test.py -f 7-tmq/tmqUdf.py

tests/system-test/simpletest.bat

@@ -4,8 +4,8 @@
 python3 .\test.py -f 0-others\taosShellNetChk.py
 python3 .\test.py -f 0-others\telemetry.py
 python3 .\test.py -f 0-others\taosdMonitor.py
-python3 .\test.py -f 0-others\udfTest.py
+@REM python3 .\test.py -f 0-others\udfTest.py
-python3 .\test.py -f 0-others\udf_create.py
+@REM python3 .\test.py -f 0-others\udf_create.py
 @REM python3 .\test.py -f 0-others\udf_restart_taosd.py
 @REM python3 .\test.py -f 0-others\cachelast.py
 

tests/test/c/tmqSim.c

@@ -22,8 +22,10 @@
 #include <time.h>
 
 #include "taos.h"
+#include "taosdef.h"
 #include "taoserror.h"
 #include "tlog.h"
+#include "types.h"
 
 #define GREEN     "\033[1;32m"
 #define NC        "\033[0m"
@@ -34,11 +36,7 @@
 #define MAX_CONSUMER_THREAD_CNT (16)
 #define MAX_VGROUP_CNT          (32)
 
-typedef enum {
+typedef enum { NOTIFY_CMD_START_CONSUM, NOTIFY_CMD_START_COMMIT, NOTIFY_CMD_ID_BUTT } NOTIFY_CMD_ID;
-    NOTIFY_CMD_START_CONSUM,
-	NOTIFY_CMD_START_COMMIT,
-	NOTIFY_CMD_ID_BUTT
-}NOTIFY_CMD_ID;
 
 typedef struct {
   TdThread thread;
@@ -49,8 +47,9 @@ typedef struct {
   // char     autoCommit[8];         // true, false
   // char     autoOffsetRest[16];    // none, earliest, latest
 
-  int32_t ifCheckData;
+  TdFilePtr pConsumeRowsFile;
-  int64_t expectMsgCnt;
+  int32_t   ifCheckData;
+  int64_t   expectMsgCnt;
 
   int64_t consumeMsgCnt;
   int64_t consumeRowCnt;
@@ -86,6 +85,7 @@ typedef struct {
   int32_t     saveRowFlag;
   int32_t     consumeDelay;  // unit s
   int32_t     numOfThread;
+  int32_t     useSnapshot;
   SThreadInfo stThreads[MAX_CONSUMER_THREAD_CNT];
 } SConfInfo;
 
@@ -93,6 +93,8 @@ static SConfInfo g_stConfInfo;
 TdFilePtr        g_fp = NULL;
 static int       running = 1;
 
+int8_t useSnapshot = 0;
+
 // char* g_pRowValue = NULL;
 // TdFilePtr g_fp = NULL;
 
@@ -129,7 +131,6 @@ void initLogFile() {
   char tmpString[128];
 
   sprintf(filename, "%s/../log/tmqlog_%s.txt", configDir, getCurrentTimeString(tmpString));
-  // sprintf(filename, "%s/../log/tmqlog.txt", configDir);
 #ifdef WINDOWS
   for (int i = 2; i < sizeof(filename); i++) {
     if (filename[i] == ':') filename[i] = '-';
@@ -203,6 +204,8 @@ void parseArgument(int32_t argc, char* argv[]) {
       g_stConfInfo.saveRowFlag = atol(argv[++i]);
     } else if (strcmp(argv[i], "-y") == 0) {
       g_stConfInfo.consumeDelay = atol(argv[++i]);
+    } else if (strcmp(argv[i], "-e") == 0) {
+      useSnapshot = (int8_t)atol(argv[++i]);
     } else {
       pError("%s unknow para: %s %s", GREEN, argv[++i], NC);
       exit(-1);
@@ -296,17 +299,148 @@ int32_t saveConsumeContentToTbl(SThreadInfo* pInfo, char* buf) {
   return 0;
 }
 
+static char* shellFormatTimestamp(char* buf, int64_t val, int32_t precision) {
+  // if (shell.args.is_raw_time) {
+  //   sprintf(buf, "%" PRId64, val);
+  //   return buf;
+  // }
+
+  time_t  tt;
+  int32_t ms = 0;
+  if (precision == TSDB_TIME_PRECISION_NANO) {
+    tt = (time_t)(val / 1000000000);
+    ms = val % 1000000000;
+  } else if (precision == TSDB_TIME_PRECISION_MICRO) {
+    tt = (time_t)(val / 1000000);
+    ms = val % 1000000;
+  } else {
+    tt = (time_t)(val / 1000);
+    ms = val % 1000;
+  }
+
+  /*
+    comment out as it make testcases like select_with_tags.sim fail.
+    but in windows, this may cause the call to localtime crash if tt < 0,
+    need to find a better solution.
+    if (tt < 0) {
+      tt = 0;
+    }
+  */
+
+#ifdef WINDOWS
+  if (tt < 0) tt = 0;
+#endif
+  if (tt <= 0 && ms < 0) {
+    tt--;
+    if (precision == TSDB_TIME_PRECISION_NANO) {
+      ms += 1000000000;
+    } else if (precision == TSDB_TIME_PRECISION_MICRO) {
+      ms += 1000000;
+    } else {
+      ms += 1000;
+    }
+  }
+
+  struct tm* ptm = taosLocalTime(&tt, NULL);
+  size_t     pos = strftime(buf, 35, "%Y-%m-%d %H:%M:%S", ptm);
+
+  if (precision == TSDB_TIME_PRECISION_NANO) {
+    sprintf(buf + pos, ".%09d", ms);
+  } else if (precision == TSDB_TIME_PRECISION_MICRO) {
+    sprintf(buf + pos, ".%06d", ms);
+  } else {
+    sprintf(buf + pos, ".%03d", ms);
+  }
+
+  return buf;
+}
+
+static void shellDumpFieldToFile(TdFilePtr pFile, const char* val, TAOS_FIELD* field, int32_t length,
+                                 int32_t precision) {
+  if (val == NULL) {
+    taosFprintfFile(pFile, "%s", TSDB_DATA_NULL_STR);
+    return;
+  }
+
+  int  n;
+  char buf[TSDB_MAX_BYTES_PER_ROW];
+  switch (field->type) {
+    case TSDB_DATA_TYPE_BOOL:
+      taosFprintfFile(pFile, "%d", ((((int32_t)(*((char*)val))) == 1) ? 1 : 0));
+      break;
+    case TSDB_DATA_TYPE_TINYINT:
+      taosFprintfFile(pFile, "%d", *((int8_t*)val));
+      break;
+    case TSDB_DATA_TYPE_UTINYINT:
+      taosFprintfFile(pFile, "%u", *((uint8_t*)val));
+      break;
+    case TSDB_DATA_TYPE_SMALLINT:
+      taosFprintfFile(pFile, "%d", *((int16_t*)val));
+      break;
+    case TSDB_DATA_TYPE_USMALLINT:
+      taosFprintfFile(pFile, "%u", *((uint16_t*)val));
+      break;
+    case TSDB_DATA_TYPE_INT:
+      taosFprintfFile(pFile, "%d", *((int32_t*)val));
+      break;
+    case TSDB_DATA_TYPE_UINT:
+      taosFprintfFile(pFile, "%u", *((uint32_t*)val));
+      break;
+    case TSDB_DATA_TYPE_BIGINT:
+      taosFprintfFile(pFile, "%" PRId64, *((int64_t*)val));
+      break;
+    case TSDB_DATA_TYPE_UBIGINT:
+      taosFprintfFile(pFile, "%" PRIu64, *((uint64_t*)val));
+      break;
+    case TSDB_DATA_TYPE_FLOAT:
+      taosFprintfFile(pFile, "%.5f", GET_FLOAT_VAL(val));
+      break;
+    case TSDB_DATA_TYPE_DOUBLE:
+      n = snprintf(buf, TSDB_MAX_BYTES_PER_ROW, "%*.9f", length, GET_DOUBLE_VAL(val));
+      if (n > TMAX(25, length)) {
+        taosFprintfFile(pFile, "%*.15e", length, GET_DOUBLE_VAL(val));
+      } else {
+        taosFprintfFile(pFile, "%s", buf);
+      }
+      break;
+    case TSDB_DATA_TYPE_BINARY:
+    case TSDB_DATA_TYPE_NCHAR:
+    case TSDB_DATA_TYPE_JSON:
+      memcpy(buf, val, length);
+      buf[length] = 0;
+      taosFprintfFile(pFile, "\'%s\'", buf);
+      break;
+    case TSDB_DATA_TYPE_TIMESTAMP:
+      shellFormatTimestamp(buf, *(int64_t*)val, precision);
+      taosFprintfFile(pFile, "'%s'", buf);
+      break;
+    default:
+      break;
+  }
+}
+
+static void dumpToFileForCheck(TdFilePtr pFile, TAOS_ROW row, TAOS_FIELD* fields, int32_t* length, int32_t num_fields,
+                               int32_t precision) {
+  for (int32_t i = 0; i < num_fields; i++) {
+    if (i > 0) {
+      taosFprintfFile(pFile, "\n");
+    }
+    shellDumpFieldToFile(pFile, (const char*)row[i], fields + i, length[i], precision);
+  }
+  taosFprintfFile(pFile, "\n");
+}
+
 static int32_t msg_process(TAOS_RES* msg, SThreadInfo* pInfo, int32_t msgIndex) {
   char    buf[1024];
   int32_t totalRows = 0;
 
   // printf("topic: %s\n", tmq_get_topic_name(msg));
-  int32_t vgroupId = tmq_get_vgroup_id(msg);
+  int32_t     vgroupId = tmq_get_vgroup_id(msg);
+  const char* dbName = tmq_get_db_name(msg);
 
-  taosFprintfFile(g_fp, "msg index:%" PRId64 ", consumerId: %d\n", msgIndex, pInfo->consumerId);
+  taosFprintfFile(g_fp, "consumerId: %d, msg index:%" PRId64 "\n", pInfo->consumerId, msgIndex);
-  // taosFprintfFile(g_fp, "topic: %s, vgroupId: %d, tableName: %s\n", tmq_get_topic_name(msg), vgroupId,
+  taosFprintfFile(g_fp, "dbName: %s, topic: %s, vgroupId: %d\n", dbName != NULL ? dbName : "invalid table",
-  // tmq_get_table_name(msg));
+                  tmq_get_topic_name(msg), vgroupId);
-  taosFprintfFile(g_fp, "topic: %s, vgroupId: %d\n", tmq_get_topic_name(msg), vgroupId);
 
   while (1) {
     TAOS_ROW row = taos_fetch_row(msg);
@@ -315,16 +449,18 @@ static int32_t msg_process(TAOS_RES* msg, SThreadInfo* pInfo, int32_t msgIndex)
 
     TAOS_FIELD* fields = taos_fetch_fields(msg);
     int32_t     numOfFields = taos_field_count(msg);
-
+    int32_t*    length = taos_fetch_lengths(msg);
-    taos_print_row(buf, row, fields, numOfFields);
+    int32_t     precision = taos_result_precision(msg);
-
     const char* tbName = tmq_get_table_name(msg);
 
+    dumpToFileForCheck(pInfo->pConsumeRowsFile, row, fields, length, numOfFields, precision);
+    taos_print_row(buf, row, fields, numOfFields);
+
     if (0 != g_stConfInfo.showRowFlag) {
       taosFprintfFile(g_fp, "tbname:%s, rows[%d]: %s\n", (tbName != NULL ? tbName : "null table"), totalRows, buf);
-      if (0 != g_stConfInfo.saveRowFlag) {
+      // if (0 != g_stConfInfo.saveRowFlag) {
-        saveConsumeContentToTbl(pInfo, buf);
+      //   saveConsumeContentToTbl(pInfo, buf);
-      }
+      // }
     }
 
     totalRows++;
@@ -347,8 +483,7 @@ int queryDB(TAOS* taos, char* command) {
   return 0;
 }
 
-static void appNothing(void* param, TAOS_RES* res, int32_t numOfRows) {  
+static void appNothing(void* param, TAOS_RES* res, int32_t numOfRows) {}
-}
 
 int32_t notifyMainScript(SThreadInfo* pInfo, int32_t cmdId) {
   char sqlStr[1024] = {0};
@@ -356,11 +491,8 @@ int32_t notifyMainScript(SThreadInfo* pInfo, int32_t cmdId) {
   int64_t now = taosGetTimestampMs();
 
   // schema: ts timestamp, consumerid int, consummsgcnt bigint, checkresult int
-  sprintf(sqlStr, "insert into %s.notifyinfo values (%"PRId64", %d, %d)", 
+  sprintf(sqlStr, "insert into %s.notifyinfo values (%" PRId64 ", %d, %d)", g_stConfInfo.cdbName, now, cmdId,
-                   g_stConfInfo.cdbName, 
+          pInfo->consumerId);
-                   now,
-                   cmdId,
-                   pInfo->consumerId);
 
   taos_query_a(pInfo->taos, sqlStr, appNothing, NULL);
 
@@ -370,12 +502,12 @@ int32_t notifyMainScript(SThreadInfo* pInfo, int32_t cmdId) {
 }
 
 static int32_t g_once_commit_flag = 0;
-static void tmq_commit_cb_print(tmq_t* tmq, int32_t code, void* param) {
+static void    tmq_commit_cb_print(tmq_t* tmq, int32_t code, void* param) {
   pError("tmq_commit_cb_print() commit %d\n", code);
 
   if (0 == g_once_commit_flag) {
     g_once_commit_flag = 1;
-	notifyMainScript((SThreadInfo*)param, (int32_t)NOTIFY_CMD_START_COMMIT);
+    notifyMainScript((SThreadInfo*)param, (int32_t)NOTIFY_CMD_START_COMMIT);
   }
   taosFprintfFile(g_fp, "tmq_commit_cb_print() be called\n");
 }
@@ -409,6 +541,10 @@ void build_consumer(SThreadInfo* pInfo) {
   // tmq_conf_set(conf, "auto.offset.reset", "none");
   // tmq_conf_set(conf, "auto.offset.reset", "earliest");
   // tmq_conf_set(conf, "auto.offset.reset", "latest");
+  //
+  if (useSnapshot) {
+    tmq_conf_set(conf, "experiment.use.snapshot", "true");
+  }
 
   pInfo->tmq = tmq_consumer_new(conf, NULL, 0);
 
@@ -464,6 +600,19 @@ void loop_consume(SThreadInfo* pInfo) {
 
   pInfo->ts = taosGetTimestampMs();
 
+  if (pInfo->ifCheckData) {
+    char filename[256] = {0};
+    char tmpString[128];
+    // sprintf(filename, "%s/../log/consumerid_%d_%s.txt", configDir, pInfo->consumerId,
+    // getCurrentTimeString(tmpString));
+    sprintf(filename, "%s/../log/consumerid_%d.txt", configDir, pInfo->consumerId);
+    pInfo->pConsumeRowsFile = taosOpenFile(filename, TD_FILE_CREATE | TD_FILE_WRITE | TD_FILE_TRUNC | TD_FILE_STREAM);
+    if (pInfo->pConsumeRowsFile == NULL) {
+      taosFprintfFile(g_fp, "%s create file fail for save rows content\n", getCurrentTimeString(tmpString));
+      return;
+    }
+  }
+
   while (running) {
     TAOS_RES* tmqMsg = tmq_consumer_poll(pInfo->tmq, g_stConfInfo.consumeDelay * 1000);
     if (tmqMsg) {
@@ -475,10 +624,10 @@ void loop_consume(SThreadInfo* pInfo) {
 
       totalMsgs++;
 
-	  if (0 == once_flag) {
+      if (0 == once_flag) {
         once_flag = 1;
-		notifyMainScript(pInfo, NOTIFY_CMD_START_CONSUM);
+        notifyMainScript(pInfo, NOTIFY_CMD_START_CONSUM);
-	  }
+      }
 
       if (totalRows >= pInfo->expectMsgCnt) {
         char tmpString[128];
@@ -507,7 +656,7 @@ void* consumeThreadFunc(void* param) {
   pInfo->taos = taos_connect(NULL, "root", "taosdata", NULL, 0);
   if (pInfo->taos == NULL) {
     taosFprintfFile(g_fp, "taos_connect() fail, can not notify and save consume result to main scripte\n");
-	exit(-1);
+    exit(-1);
   }
 
   build_consumer(pInfo);

tools/shell/src/shellCommand.c

@@ -21,6 +21,7 @@
 #define UP    3
 #define DOWN  4
 #define PSIZE shell.info.promptSize
+#define SHELL_INPUT_MAX_COMMAND_SIZE 10000
 
 typedef struct {
   char    *buffer;
@@ -227,6 +228,7 @@ void shellPrintChar(char c, int32_t times) {
 }
 
 void shellPositionCursor(int32_t step, int32_t direction) {
+#ifndef WINDOWS
   if (step > 0) {
     if (direction == LEFT) {
       fprintf(stdout, "\033[%dD", step);
@@ -239,6 +241,7 @@ void shellPositionCursor(int32_t step, int32_t direction) {
     }
     fflush(stdout);
   }
+#endif
 }
 
 void shellUpdateBuffer(SShellCmd *cmd) {
@@ -330,10 +333,14 @@ void shellClearScreen(int32_t ecmd_pos, int32_t cursor_pos) {
   int32_t command_x = ecmd_pos / ws_col;
   shellPositionCursor(cursor_y, LEFT);
   shellPositionCursor(command_x - cursor_x, DOWN);
+#ifndef WINDOWS
   fprintf(stdout, "\033[2K");
+#endif
   for (int32_t i = 0; i < command_x; i++) {
     shellPositionCursor(1, UP);
+  #ifndef WINDOWS
     fprintf(stdout, "\033[2K");
+  #endif
   }
   fflush(stdout);
 }
@@ -394,6 +401,41 @@ void shellShowOnScreen(SShellCmd *cmd) {
   fflush(stdout);
 }
 
+char taosGetConsoleChar() {
+#ifdef WINDOWS
+  static void *console = NULL;
+  if (console == NULL) {
+    console = GetStdHandle(STD_INPUT_HANDLE);
+  }
+  static TdWchar buf[SHELL_INPUT_MAX_COMMAND_SIZE];
+  static char mbStr[5];
+  static unsigned long bufLen = 0;
+  static uint16_t bufIndex = 0, mbStrIndex = 0, mbStrLen = 0;
+  if (bufLen == 0) {
+    ReadConsoleW(console, buf, SHELL_INPUT_MAX_COMMAND_SIZE, &bufLen, NULL);
+    bufIndex = 0;
+  }
+  if (mbStrLen == 0){
+    if (buf[bufIndex] == '\r') {
+      bufIndex++;
+    }
+    mbStrLen = WideCharToMultiByte(CP_UTF8, 0, &buf[bufIndex], 1, mbStr, sizeof(mbStr), NULL, NULL);
+    mbStrIndex = 0;
+    bufIndex++;
+  }
+  mbStrIndex++;
+  if (mbStrIndex == mbStrLen) {
+    mbStrLen = 0;
+    if (bufIndex == bufLen) {
+      bufLen = 0;
+    }
+  }
+  return mbStr[mbStrIndex-1];
+#else
+  return (char)getchar();  // getchar() return an 'int32_t' value
+#endif
+}
+
 int32_t shellReadCommand(char *command) {
   SShellHistory *pHistory = &shell.history;
   SShellCmd      cmd = {0};
@@ -407,7 +449,7 @@ int32_t shellReadCommand(char *command) {
   // Read input.
   char c;
   while (1) {
-    c = (char)getchar();  // getchar() return an 'int32_t' value
+    c = taosGetConsoleChar();
 
     if (c == EOF) {
       return c;
@@ -417,7 +459,7 @@ int32_t shellReadCommand(char *command) {
       int32_t count = shellCountPrefixOnes(c);
       utf8_array[0] = c;
       for (int32_t k = 1; k < count; k++) {
-        c = (char)getchar();
+        c = taosGetConsoleChar();
         utf8_array[k] = c;
       }
       shellInsertChar(&cmd, utf8_array, count);
@@ -472,10 +514,10 @@ int32_t shellReadCommand(char *command) {
           break;
       }
     } else if (c == '\033') {
-      c = (char)getchar();
+      c = taosGetConsoleChar();
       switch (c) {
         case '[':
-          c = (char)getchar();
+          c = taosGetConsoleChar();
           switch (c) {
             case 'A':  // Up arrow
               if (hist_counter != pHistory->hstart) {
@@ -502,35 +544,35 @@ int32_t shellReadCommand(char *command) {
               shellMoveCursorLeft(&cmd);
               break;
             case '1':
-              if ((c = (char)getchar()) == '~') {
+              if ((c = taosGetConsoleChar()) == '~') {
                 // Home key
                 shellPositionCursorHome(&cmd);
               }
               break;
             case '2':
-              if ((c = (char)getchar()) == '~') {
+              if ((c = taosGetConsoleChar()) == '~') {
                 // Insert key
               }
               break;
             case '3':
-              if ((c = (char)getchar()) == '~') {
+              if ((c = taosGetConsoleChar()) == '~') {
                 // Delete key
                 shellDeleteChar(&cmd);
               }
               break;
             case '4':
-              if ((c = (char)getchar()) == '~') {
+              if ((c = taosGetConsoleChar()) == '~') {
                 // End key
                 shellPositionCursorEnd(&cmd);
               }
               break;
             case '5':
-              if ((c = (char)getchar()) == '~') {
+              if ((c = taosGetConsoleChar()) == '~') {
                 // Page up key
               }
               break;
             case '6':
-              if ((c = (char)getchar()) == '~') {
+              if ((c = taosGetConsoleChar()) == '~') {
                 // Page down key
               }
               break;

tools/shell/src/shellEngine.c

@@ -393,15 +393,11 @@ void shellPrintNChar(const char *str, int32_t length, int32_t width) {
       break;
     }
     int w = 0;
-#ifdef WINDOWS
-    w = bytes;
-#else
     if(*(str + pos) == '\t' || *(str + pos) == '\n' || *(str + pos) == '\r'){
       w = bytes;
     }else{
       w = taosWcharWidth(wc);
     }
-#endif
     pos += bytes;
 
     if (w <= 0) {
@@ -524,6 +520,16 @@ bool shellIsLimitQuery(const char *sql) {
   return false;
 }
 
+bool shellIsShowQuery(const char *sql) {
+  //todo refactor
+  if (taosStrCaseStr(sql, "show ") != NULL) {
+    return true;
+  }
+
+  return false;
+}
+
+
 int32_t shellVerticalPrintResult(TAOS_RES *tres, const char *sql) {
   TAOS_ROW row = taos_fetch_row(tres);
   if (row == NULL) {
@@ -686,7 +692,7 @@ int32_t shellHorizontalPrintResult(TAOS_RES *tres, const char *sql) {
 
   uint64_t resShowMaxNum = UINT64_MAX;
 
-  if (shell.args.commands == NULL && shell.args.file[0] == 0 && !shellIsLimitQuery(sql)) {
+  if (shell.args.commands == NULL && shell.args.file[0] == 0 && !shellIsLimitQuery(sql) && !shellIsShowQuery(sql)) {
     resShowMaxNum = SHELL_DEFAULT_RES_SHOW_NUM;
   }