feat: super table order by primary key optimization

include/libs/nodes/plannodes.h

@@ -104,6 +104,7 @@ typedef struct SJoinLogicNode {
   SNode*     pMergeCondition;
   SNode*     pOnConditions;
   bool       isSingleTableJoin;
+  EOrder     inputTsOrder;
 } SJoinLogicNode;
 
 typedef struct SAggLogicNode {
@@ -201,6 +202,7 @@ typedef struct SWindowLogicNode {
   int64_t          watermark;
   int8_t           igExpired;
   EWindowAlgorithm windowAlgo;
+  EOrder           inputTsOrder;
 } SWindowLogicNode;
 
 typedef struct SFillLogicNode {
@@ -356,15 +358,14 @@ typedef struct SInterpFuncPhysiNode {
   SNode*      pTimeSeries;  // SColumnNode
 } SInterpFuncPhysiNode;
 
-typedef struct SJoinPhysiNode {
+typedef struct SSortMergeJoinPhysiNode {
   SPhysiNode node;
   EJoinType  joinType;
   SNode*     pMergeCondition;
   SNode*     pOnConditions;
   SNodeList* pTargets;
-} SJoinPhysiNode;
+  EOrder     inputTsOrder;
-
+} SSortMergeJoinPhysiNode;
-typedef SJoinPhysiNode SSortMergeJoinPhysiNode;
 
 typedef struct SAggPhysiNode {
   SPhysiNode node;

source/libs/command/src/explain.c

@@ -135,7 +135,7 @@ int32_t qExplainGenerateResChildren(SPhysiNode *pNode, SExplainGroup *group, SNo
       break;
     }
     case QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN: {
-      SJoinPhysiNode *pJoinNode = (SJoinPhysiNode *)pNode;
+      SSortMergeJoinPhysiNode *pJoinNode = (SSortMergeJoinPhysiNode *)pNode;
       pPhysiChildren = pJoinNode->node.pChildren;
       break;
     }
@@ -434,7 +434,8 @@ int32_t qExplainResNodeToRowsImpl(SExplainResNode *pResNode, SExplainCtx *ctx, i
     case QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN: {
       STableScanPhysiNode *pTblScanNode = (STableScanPhysiNode *)pNode;
       EXPLAIN_ROW_NEW(level,
-                      QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN == pNode->type ? EXPLAIN_TBL_MERGE_SCAN_FORMAT : EXPLAIN_TBL_SCAN_FORMAT,
+                      QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN == pNode->type ? EXPLAIN_TBL_MERGE_SCAN_FORMAT
+                                                                               : EXPLAIN_TBL_SCAN_FORMAT,
                       pTblScanNode->scan.tableName.tname);
       EXPLAIN_ROW_APPEND(EXPLAIN_LEFT_PARENTHESIS_FORMAT);
       if (pResNode->pExecInfo) {
@@ -613,7 +614,7 @@ int32_t qExplainResNodeToRowsImpl(SExplainResNode *pResNode, SExplainCtx *ctx, i
       break;
     }
     case QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN: {
-      SJoinPhysiNode *pJoinNode = (SJoinPhysiNode *)pNode;
+      SSortMergeJoinPhysiNode *pJoinNode = (SSortMergeJoinPhysiNode *)pNode;
       EXPLAIN_ROW_NEW(level, EXPLAIN_JOIN_FORMAT, EXPLAIN_JOIN_STRING(pJoinNode->joinType));
       EXPLAIN_ROW_APPEND(EXPLAIN_LEFT_PARENTHESIS_FORMAT);
       if (pResNode->pExecInfo) {

source/libs/executor/inc/executorimpl.h

@@ -321,6 +321,49 @@ typedef struct STableScanInfo {
   int8_t          noTable;
 } STableScanInfo;
 
+typedef struct STableMergeScanInfo {
+  STableListInfo* tableListInfo;
+  int32_t         tableStartIndex;
+  int32_t         tableEndIndex;
+  bool            hasGroupId;
+  uint64_t        groupId;
+  SArray*         dataReaders;  // array of tsdbReaderT*
+  SReadHandle     readHandle;
+  int32_t         bufPageSize;
+  uint32_t        sortBufSize;  // max buffer size for in-memory sort
+  SArray*         pSortInfo;
+  SSortHandle*    pSortHandle;
+
+  SSDataBlock* pSortInputBlock;
+  int64_t      startTs;  // sort start time
+  SArray*      sortSourceParams;
+
+  SFileBlockLoadRecorder readRecorder;
+  int64_t                numOfRows;
+  SScanInfo              scanInfo;
+  int32_t                scanTimes;
+  SNode*                 pFilterNode;  // filter info, which is push down by optimizer
+  SqlFunctionCtx*        pCtx;         // which belongs to the direct upstream operator operator query context
+  SResultRowInfo*        pResultRowInfo;
+  int32_t*               rowEntryInfoOffset;
+  SExprInfo*             pExpr;
+  SSDataBlock*           pResBlock;
+  SArray*                pColMatchInfo;
+  int32_t                numOfOutput;
+
+  SExprInfo*      pPseudoExpr;
+  int32_t         numOfPseudoExpr;
+  SqlFunctionCtx* pPseudoCtx;
+
+  SQueryTableDataCond cond;
+  int32_t             scanFlag;  // table scan flag to denote if it is a repeat/reverse/main scan
+  int32_t             dataBlockLoadFlag;
+  // if the upstream is an interval operator, the interval info is also kept here to get the time
+  // window to check if current data block needs to be loaded.
+  SInterval       interval;
+  SSampleExecInfo sample;  // sample execution info
+} STableMergeScanInfo;
+
 typedef struct STagScanInfo {
   SColumnInfo     *pCols;
   SSDataBlock     *pRes;
@@ -881,7 +924,7 @@ SOperatorInfo* createPartitionOperatorInfo(SOperatorInfo* downstream, SPartition
 
 SOperatorInfo* createTimeSliceOperatorInfo(SOperatorInfo* downstream, SPhysiNode* pNode, SExecTaskInfo* pTaskInfo);
 
-SOperatorInfo* createMergeJoinOperatorInfo(SOperatorInfo** pDownstream, int32_t numOfDownstream, SJoinPhysiNode* pJoinNode,
+SOperatorInfo* createMergeJoinOperatorInfo(SOperatorInfo** pDownstream, int32_t numOfDownstream, SSortMergeJoinPhysiNode* pJoinNode,
                                            SExecTaskInfo* pTaskInfo);
 
 SOperatorInfo* createStreamSessionAggOperatorInfo(SOperatorInfo* downstream,

source/libs/executor/src/executorimpl.c

@@ -2885,11 +2885,16 @@ int32_t getTableScanInfo(SOperatorInfo* pOperator, int32_t* order, int32_t* scan
     *order = TSDB_ORDER_ASC;
     *scanFlag = MAIN_SCAN;
     return TSDB_CODE_SUCCESS;
-  } else if (type == QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN || type == QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN) {
+  } else if (type == QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN) {
     STableScanInfo* pTableScanInfo = pOperator->info;
     *order = pTableScanInfo->cond.order;
     *scanFlag = pTableScanInfo->scanFlag;
     return TSDB_CODE_SUCCESS;
+  } else if (type == QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN) {
+    STableMergeScanInfo* pTableScanInfo = pOperator->info;
+    *order = pTableScanInfo->cond.order;
+    *scanFlag = pTableScanInfo->scanFlag;
+    return TSDB_CODE_SUCCESS;
   } else {
     if (pOperator->pDownstream == NULL || pOperator->pDownstream[0] == NULL) {
       return TSDB_CODE_INVALID_PARA;
@@ -3326,7 +3331,8 @@ static SSDataBlock* doProjectOperation(SOperatorInfo* pOperator) {
       }
 
       if (pLimitInfo->remainGroupOffset > 0) {
-        if (pLimitInfo->currentGroupId == 0 || pLimitInfo->currentGroupId == pBlock->info.groupId) {  // it is the first group
+        if (pLimitInfo->currentGroupId == 0 ||
+            pLimitInfo->currentGroupId == pBlock->info.groupId) {  // it is the first group
           pLimitInfo->currentGroupId = pBlock->info.groupId;
           continue;
         } else if (pLimitInfo->currentGroupId != pBlock->info.groupId) {
@@ -4701,7 +4707,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
   } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_STATE == type) {
     pOptr = createStreamStateAggOperatorInfo(ops[0], pPhyNode, pTaskInfo);
   } else if (QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN == type) {
-    pOptr = createMergeJoinOperatorInfo(ops, size, (SJoinPhysiNode*)pPhyNode, pTaskInfo);
+    pOptr = createMergeJoinOperatorInfo(ops, size, (SSortMergeJoinPhysiNode*)pPhyNode, pTaskInfo);
   } else if (QUERY_NODE_PHYSICAL_PLAN_FILL == type) {
     pOptr = createFillOperatorInfo(ops[0], (SFillPhysiNode*)pPhyNode, pTaskInfo);
   } else if (QUERY_NODE_PHYSICAL_PLAN_INDEF_ROWS_FUNC == type) {

source/libs/executor/src/joinoperator.c

@@ -28,8 +28,8 @@ static SSDataBlock* doMergeJoin(struct SOperatorInfo* pOperator);
 static void         destroyMergeJoinOperator(void* param, int32_t numOfOutput);
 static void         extractTimeCondition(SJoinOperatorInfo* Info, SLogicConditionNode* pLogicConditionNode);
 
-SOperatorInfo* createMergeJoinOperatorInfo(SOperatorInfo** pDownstream, int32_t numOfDownstream, SJoinPhysiNode* pJoinNode,
+SOperatorInfo* createMergeJoinOperatorInfo(SOperatorInfo** pDownstream, int32_t numOfDownstream,
-    SExecTaskInfo* pTaskInfo) {
+                                           SSortMergeJoinPhysiNode* pJoinNode, SExecTaskInfo* pTaskInfo) {
   SJoinOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SJoinOperatorInfo));
   SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
   if (pOperator == NULL || pInfo == NULL) {

source/libs/executor/src/scanoperator.c

@@ -1840,9 +1840,12 @@ static SSDataBlock* sysTableScanUserTags(SOperatorInfo* pOperator) {
       char tagTypeStr[VARSTR_HEADER_SIZE + 32];
       int  tagTypeLen = sprintf(varDataVal(tagTypeStr), "%s", tDataTypes[tagType].name);
       if (tagType == TSDB_DATA_TYPE_VARCHAR) {
-        tagTypeLen += sprintf(varDataVal(tagTypeStr) + tagTypeLen, "(%d)", (int32_t)(smr.me.stbEntry.schemaTag.pSchema[i].bytes - VARSTR_HEADER_SIZE));
+        tagTypeLen += sprintf(varDataVal(tagTypeStr) + tagTypeLen, "(%d)",
+                              (int32_t)(smr.me.stbEntry.schemaTag.pSchema[i].bytes - VARSTR_HEADER_SIZE));
       } else if (tagType == TSDB_DATA_TYPE_NCHAR) {
-        tagTypeLen += sprintf(varDataVal(tagTypeStr) + tagTypeLen, "(%d)", (int32_t)((smr.me.stbEntry.schemaTag.pSchema[i].bytes - VARSTR_HEADER_SIZE) / TSDB_NCHAR_SIZE));
+        tagTypeLen +=
+            sprintf(varDataVal(tagTypeStr) + tagTypeLen, "(%d)",
+                    (int32_t)((smr.me.stbEntry.schemaTag.pSchema[i].bytes - VARSTR_HEADER_SIZE) / TSDB_NCHAR_SIZE));
       }
       varDataSetLen(tagTypeStr, tagTypeLen);
       colDataAppend(pColInfoData, numOfRows, (char*)tagTypeStr, false);
@@ -2527,49 +2530,6 @@ _error:
   return NULL;
 }
 
-typedef struct STableMergeScanInfo {
-  STableListInfo* tableListInfo;
-  int32_t         tableStartIndex;
-  int32_t         tableEndIndex;
-  bool            hasGroupId;
-  uint64_t        groupId;
-  SArray*         dataReaders;  // array of tsdbReaderT*
-  SReadHandle     readHandle;
-  int32_t         bufPageSize;
-  uint32_t        sortBufSize;  // max buffer size for in-memory sort
-  SArray*         pSortInfo;
-  SSortHandle*    pSortHandle;
-
-  SSDataBlock* pSortInputBlock;
-  int64_t      startTs;  // sort start time
-  SArray*      sortSourceParams;
-
-  SFileBlockLoadRecorder readRecorder;
-  int64_t                numOfRows;
-  SScanInfo              scanInfo;
-  int32_t                scanTimes;
-  SNode*                 pFilterNode;  // filter info, which is push down by optimizer
-  SqlFunctionCtx*        pCtx;         // which belongs to the direct upstream operator operator query context
-  SResultRowInfo*        pResultRowInfo;
-  int32_t*               rowEntryInfoOffset;
-  SExprInfo*             pExpr;
-  SSDataBlock*           pResBlock;
-  SArray*                pColMatchInfo;
-  int32_t                numOfOutput;
-
-  SExprInfo*      pPseudoExpr;
-  int32_t         numOfPseudoExpr;
-  SqlFunctionCtx* pPseudoCtx;
-
-  SQueryTableDataCond cond;
-  int32_t             scanFlag;  // table scan flag to denote if it is a repeat/reverse/main scan
-  int32_t             dataBlockLoadFlag;
-  // if the upstream is an interval operator, the interval info is also kept here to get the time
-  // window to check if current data block needs to be loaded.
-  SInterval       interval;
-  SSampleExecInfo sample;  // sample execution info
-} STableMergeScanInfo;
-
 int32_t createScanTableListInfo(SScanPhysiNode* pScanNode, SNodeList* pGroupTags, bool groupSort, SReadHandle* pHandle,
                                 STableListInfo* pTableListInfo, SNode* pTagCond, SNode* pTagIndexCond,
                                 const char* idStr) {
@@ -2975,6 +2935,7 @@ void destroyTableMergeScanOperatorInfo(void* param, int32_t numOfOutput) {
 
   taosArrayDestroy(pTableScanInfo->pSortInfo);
 
+  taosMemoryFreeClear(pTableScanInfo->rowEntryInfoOffset);
   taosMemoryFreeClear(param);
 }
 

source/libs/nodes/src/nodesCodeFuncs.c

@@ -1717,7 +1717,7 @@ static const char* jkJoinPhysiPlanOnConditions = "OnConditions";
 static const char* jkJoinPhysiPlanTargets = "Targets";
 
 static int32_t physiJoinNodeToJson(const void* pObj, SJson* pJson) {
-  const SJoinPhysiNode* pNode = (const SJoinPhysiNode*)pObj;
+  const SSortMergeJoinPhysiNode* pNode = (const SSortMergeJoinPhysiNode*)pObj;
 
   int32_t code = physicPlanNodeToJson(pObj, pJson);
   if (TSDB_CODE_SUCCESS == code) {
@@ -1737,7 +1737,7 @@ static int32_t physiJoinNodeToJson(const void* pObj, SJson* pJson) {
 }
 
 static int32_t jsonToPhysiJoinNode(const SJson* pJson, void* pObj) {
-  SJoinPhysiNode* pNode = (SJoinPhysiNode*)pObj;
+  SSortMergeJoinPhysiNode* pNode = (SSortMergeJoinPhysiNode*)pObj;
 
   int32_t code = jsonToPhysicPlanNode(pJson, pObj);
   if (TSDB_CODE_SUCCESS == code) {

source/libs/nodes/src/nodesTraverseFuncs.c

@@ -468,7 +468,7 @@ static EDealRes dispatchPhysiPlan(SNode* pNode, ETraversalOrder order, FNodeWalk
       break;
     }
     case QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN: {
-      SJoinPhysiNode* pJoin = (SJoinPhysiNode*)pNode;
+      SSortMergeJoinPhysiNode* pJoin = (SSortMergeJoinPhysiNode*)pNode;
       res = walkPhysiNode((SPhysiNode*)pNode, order, walker, pContext);
       if (DEAL_RES_ERROR != res && DEAL_RES_END != res) {
         res = walkPhysiPlan(pJoin->pMergeCondition, order, walker, pContext);

source/libs/nodes/src/nodesUtilFuncs.c

@@ -287,7 +287,7 @@ SNode* nodesMakeNode(ENodeType type) {
     case QUERY_NODE_PHYSICAL_PLAN_PROJECT:
       return makeNode(type, sizeof(SProjectPhysiNode));
     case QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN:
-      return makeNode(type, sizeof(SJoinPhysiNode));
+      return makeNode(type, sizeof(SSortMergeJoinPhysiNode));
     case QUERY_NODE_PHYSICAL_PLAN_HASH_AGG:
       return makeNode(type, sizeof(SAggPhysiNode));
     case QUERY_NODE_PHYSICAL_PLAN_EXCHANGE:
@@ -883,7 +883,7 @@ void nodesDestroyNode(SNode* pNode) {
       break;
     }
     case QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN: {
-      SJoinPhysiNode* pPhyNode = (SJoinPhysiNode*)pNode;
+      SSortMergeJoinPhysiNode* pPhyNode = (SSortMergeJoinPhysiNode*)pNode;
       destroyPhysiNode((SPhysiNode*)pPhyNode);
       nodesDestroyNode(pPhyNode->pMergeCondition);
       nodesDestroyNode(pPhyNode->pOnConditions);

source/libs/parser/src/parInsert.c

@@ -739,7 +739,8 @@ static int32_t parseBoundColumns(SInsertParseContext* pCxt, SParsedDataColInfo*
   return TSDB_CODE_SUCCESS;
 }
 
-static void buildCreateTbReq(SVCreateTbReq* pTbReq, const char* tname, STag* pTag, int64_t suid, const char* sname, SArray* tagName, uint8_t tagNum) {
+static void buildCreateTbReq(SVCreateTbReq* pTbReq, const char* tname, STag* pTag, int64_t suid, const char* sname,
+                             SArray* tagName, uint8_t tagNum) {
   pTbReq->type = TD_CHILD_TABLE;
   pTbReq->name = strdup(tname);
   pTbReq->ctb.suid = suid;
@@ -1012,7 +1013,8 @@ static int32_t parseTagsClause(SInsertParseContext* pCxt, SSchema* pSchema, uint
     goto end;
   }
 
-  buildCreateTbReq(&pCxt->createTblReq, tName, pTag, pCxt->pTableMeta->suid, pCxt->sTableName, tagName, pCxt->pTableMeta->tableInfo.numOfTags);
+  buildCreateTbReq(&pCxt->createTblReq, tName, pTag, pCxt->pTableMeta->suid, pCxt->sTableName, tagName,
+                   pCxt->pTableMeta->tableInfo.numOfTags);
 
 end:
   for (int i = 0; i < taosArrayGetSize(pTagVals); ++i) {
@@ -1650,7 +1652,6 @@ static int32_t skipUsingClause(SInsertParseSyntaxCxt* pCxt) {
 static int32_t collectTableMetaKey(SInsertParseSyntaxCxt* pCxt, SToken* pTbToken) {
   SName name;
   CHECK_CODE(createSName(&name, pTbToken, pCxt->pComCxt->acctId, pCxt->pComCxt->db, &pCxt->msg));
-  CHECK_CODE(reserveDbCfgInCache(pCxt->pComCxt->acctId, name.dbname, pCxt->pMetaCache));
   CHECK_CODE(reserveUserAuthInCacheExt(pCxt->pComCxt->pUser, &name, AUTH_TYPE_WRITE, pCxt->pMetaCache));
   CHECK_CODE(reserveTableMetaInCacheExt(&name, pCxt->pMetaCache));
   CHECK_CODE(reserveTableVgroupInCacheExt(&name, pCxt->pMetaCache));
@@ -2332,7 +2333,8 @@ int32_t smlBindData(void* handle, SArray* tags, SArray* colsSchema, SArray* cols
     return ret;
   }
 
-  buildCreateTbReq(&smlHandle->tableExecHandle.createTblReq, tableName, pTag, pTableMeta->suid, NULL, tagName, pTableMeta->tableInfo.numOfTags);
+  buildCreateTbReq(&smlHandle->tableExecHandle.createTblReq, tableName, pTag, pTableMeta->suid, NULL, tagName,
+                   pTableMeta->tableInfo.numOfTags);
   taosArrayDestroy(tagName);
 
   smlHandle->tableExecHandle.createTblReq.ctb.name = taosMemoryMalloc(sTableNameLen + 1);

source/libs/parser/src/parUtil.c

@@ -92,7 +92,7 @@ static char* getSyntaxErrFormat(int32_t errCode) {
     case TSDB_CODE_PAR_INTER_SLIDING_TOO_BIG:
       return "sliding value no larger than the interval value";
     case TSDB_CODE_PAR_INTER_SLIDING_TOO_SMALL:
-      return "sliding value can not less than 1% of interval value";
+      return "sliding value can not less than 1%% of interval value";
     case TSDB_CODE_PAR_ONLY_ONE_JSON_TAG:
       return "Only one tag if there is a json tag";
     case TSDB_CODE_PAR_INCORRECT_NUM_OF_COL:

source/libs/planner/src/planOptimizer.c

@@ -993,22 +993,28 @@ static bool sortPriKeyOptMayBeOptimized(SLogicNode* pNode) {
 }
 
 static int32_t sortPriKeyOptGetScanNodesImpl(SLogicNode* pNode, bool* pNotOptimize, SNodeList** pScanNodes) {
-  int32_t code = TSDB_CODE_SUCCESS;
-
   switch (nodeType(pNode)) {
-    case QUERY_NODE_LOGIC_PLAN_SCAN:
+    case QUERY_NODE_LOGIC_PLAN_SCAN: {
+      SScanLogicNode* pScan = (SScanLogicNode*)pNode;
+      if (NULL != pScan->pGroupTags) {
+        *pNotOptimize = true;
+        return TSDB_CODE_SUCCESS;
+      }
       return nodesListMakeAppend(pScanNodes, (SNode*)pNode);
-    case QUERY_NODE_LOGIC_PLAN_JOIN:
+    }
-      code =
+    case QUERY_NODE_LOGIC_PLAN_JOIN: {
+      int32_t code =
           sortPriKeyOptGetScanNodesImpl((SLogicNode*)nodesListGetNode(pNode->pChildren, 0), pNotOptimize, pScanNodes);
       if (TSDB_CODE_SUCCESS == code) {
         code =
             sortPriKeyOptGetScanNodesImpl((SLogicNode*)nodesListGetNode(pNode->pChildren, 1), pNotOptimize, pScanNodes);
       }
       return code;
+    }
     case QUERY_NODE_LOGIC_PLAN_AGG:
+    case QUERY_NODE_LOGIC_PLAN_PARTITION:
       *pNotOptimize = true;
-      return code;
+      return TSDB_CODE_SUCCESS;
     default:
       break;
   }
@@ -1034,6 +1040,18 @@ static EOrder sortPriKeyOptGetPriKeyOrder(SSortLogicNode* pSort) {
   return ((SOrderByExprNode*)nodesListGetNode(pSort->pSortKeys, 0))->order;
 }
 
+static void sortPriKeyOptSetParentOrder(SLogicNode* pNode, EOrder order) {
+  if (NULL == pNode) {
+    return;
+  }
+  if (QUERY_NODE_LOGIC_PLAN_WINDOW == nodeType(pNode)) {
+    ((SWindowLogicNode*)pNode)->inputTsOrder = order;
+  } else if (QUERY_NODE_LOGIC_PLAN_JOIN == nodeType(pNode)) {
+    ((SJoinLogicNode*)pNode)->inputTsOrder = order;
+  }
+  sortPriKeyOptSetParentOrder(pNode->pParent, order);
+}
+
 static int32_t sortPriKeyOptApply(SOptimizeContext* pCxt, SLogicSubplan* pLogicSubplan, SSortLogicNode* pSort,
                                   SNodeList* pScanNodes) {
   EOrder order = sortPriKeyOptGetPriKeyOrder(pSort);
@@ -1048,6 +1066,7 @@ static int32_t sortPriKeyOptApply(SOptimizeContext* pCxt, SLogicSubplan* pLogicS
       pScan->node.resultDataOrder = DATA_ORDER_LEVEL_GLOBAL;
       pScan->node.requireDataOrder = DATA_ORDER_LEVEL_GLOBAL;
     }
+    sortPriKeyOptSetParentOrder(pScan->node.pParent, order);
   }
 
   SLogicNode* pChild = (SLogicNode*)nodesListGetNode(pSort->node.pChildren, 0);

source/libs/planner/src/planPhysiCreater.c

@@ -621,8 +621,8 @@ static int32_t createScanPhysiNode(SPhysiPlanContext* pCxt, SSubplan* pSubplan,
 
 static int32_t createJoinPhysiNode(SPhysiPlanContext* pCxt, SNodeList* pChildren, SJoinLogicNode* pJoinLogicNode,
                                    SPhysiNode** pPhyNode) {
-  SJoinPhysiNode* pJoin =
+  SSortMergeJoinPhysiNode* pJoin =
-      (SJoinPhysiNode*)makePhysiNode(pCxt, (SLogicNode*)pJoinLogicNode, QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN);
+      (SSortMergeJoinPhysiNode*)makePhysiNode(pCxt, (SLogicNode*)pJoinLogicNode, QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN);
   if (NULL == pJoin) {
     return TSDB_CODE_OUT_OF_MEMORY;
   }

source/libs/planner/src/planSpliter.c

@@ -469,7 +469,7 @@ static int32_t stbSplCreateExchangeNode(SSplitContext* pCxt, SLogicNode* pParent
   return code;
 }
 
-static int32_t stbSplCreateMergeKeysByPrimaryKey(SNode* pPrimaryKey, SNodeList** pMergeKeys) {
+static int32_t stbSplCreateMergeKeysByPrimaryKey(SNode* pPrimaryKey, EOrder order, SNodeList** pMergeKeys) {
   SOrderByExprNode* pMergeKey = (SOrderByExprNode*)nodesMakeNode(QUERY_NODE_ORDER_BY_EXPR);
   if (NULL == pMergeKey) {
     return TSDB_CODE_OUT_OF_MEMORY;
@@ -479,7 +479,7 @@ static int32_t stbSplCreateMergeKeysByPrimaryKey(SNode* pPrimaryKey, SNodeList**
     nodesDestroyNode((SNode*)pMergeKey);
     return TSDB_CODE_OUT_OF_MEMORY;
   }
-  pMergeKey->order = ORDER_ASC;
+  pMergeKey->order = order;
   pMergeKey->nullOrder = NULL_ORDER_FIRST;
   return nodesListMakeStrictAppend(pMergeKeys, (SNode*)pMergeKey);
 }
@@ -491,7 +491,8 @@ static int32_t stbSplSplitIntervalForBatch(SSplitContext* pCxt, SStableSplitInfo
     ((SWindowLogicNode*)pPartWindow)->windowAlgo = INTERVAL_ALGO_HASH;
     ((SWindowLogicNode*)pInfo->pSplitNode)->windowAlgo = INTERVAL_ALGO_MERGE;
     SNodeList* pMergeKeys = NULL;
-    code = stbSplCreateMergeKeysByPrimaryKey(((SWindowLogicNode*)pInfo->pSplitNode)->pTspk, &pMergeKeys);
+    code = stbSplCreateMergeKeysByPrimaryKey(((SWindowLogicNode*)pInfo->pSplitNode)->pTspk,
+                                             ((SWindowLogicNode*)pInfo->pSplitNode)->inputTsOrder, &pMergeKeys);
     if (TSDB_CODE_SUCCESS == code) {
       code = stbSplCreateMergeNode(pCxt, NULL, pInfo->pSplitNode, pMergeKeys, pPartWindow, true);
     }
@@ -579,7 +580,8 @@ static int32_t stbSplSplitSessionOrStateForBatch(SSplitContext* pCxt, SStableSpl
   SLogicNode* pChild = (SLogicNode*)nodesListGetNode(pWindow->pChildren, 0);
 
   SNodeList* pMergeKeys = NULL;
-  int32_t    code = stbSplCreateMergeKeysByPrimaryKey(((SWindowLogicNode*)pWindow)->pTspk, &pMergeKeys);
+  int32_t    code = stbSplCreateMergeKeysByPrimaryKey(((SWindowLogicNode*)pWindow)->pTspk,
+                                                      ((SWindowLogicNode*)pWindow)->inputTsOrder, &pMergeKeys);
 
   if (TSDB_CODE_SUCCESS == code) {
     code = stbSplCreateMergeNode(pCxt, pInfo->pSubplan, pChild, pMergeKeys, (SLogicNode*)pChild, true);
@@ -950,7 +952,8 @@ static int32_t stbSplCreateMergeScanNode(SScanLogicNode* pScan, SLogicNode** pOu
     pMergeScan->scanType = SCAN_TYPE_TABLE_MERGE;
     pMergeScan->node.pChildren = pChildren;
     splSetParent((SLogicNode*)pMergeScan);
-    code = stbSplCreateMergeKeysByPrimaryKey(stbSplFindPrimaryKeyFromScan(pMergeScan), &pMergeKeys);
+    code = stbSplCreateMergeKeysByPrimaryKey(stbSplFindPrimaryKeyFromScan(pMergeScan),
+                                             pMergeScan->scanSeq[0] > 0 ? ORDER_ASC : ORDER_DESC, &pMergeKeys);
   }
 
   if (TSDB_CODE_SUCCESS == code) {
@@ -1020,14 +1023,14 @@ static int32_t stbSplSplitJoinNode(SSplitContext* pCxt, SStableSplitInfo* pInfo)
 }
 
 static int32_t stbSplCreateMergeKeysForPartitionNode(SLogicNode* pPart, SNodeList** pMergeKeys) {
-  SNode* pPrimaryKey =
+  SScanLogicNode* pScan = (SScanLogicNode*)nodesListGetNode(pPart->pChildren, 0);
-      nodesCloneNode(stbSplFindPrimaryKeyFromScan((SScanLogicNode*)nodesListGetNode(pPart->pChildren, 0)));
+  SNode*          pPrimaryKey = nodesCloneNode(stbSplFindPrimaryKeyFromScan(pScan));
   if (NULL == pPrimaryKey) {
     return TSDB_CODE_OUT_OF_MEMORY;
   }
   int32_t code = nodesListAppend(pPart->pTargets, pPrimaryKey);
   if (TSDB_CODE_SUCCESS == code) {
-    code = stbSplCreateMergeKeysByPrimaryKey(pPrimaryKey, pMergeKeys);
+    code = stbSplCreateMergeKeysByPrimaryKey(pPrimaryKey, pScan->scanSeq[0] > 0 ? ORDER_ASC : ORDER_DESC, pMergeKeys);
   }
   return code;
 }

source/util/src/terror.c

@@ -512,7 +512,7 @@ TAOS_DEFINE_ERROR(TSDB_CODE_PAR_INTER_OFFSET_UNIT,         "Cannot use 'year' as
 TAOS_DEFINE_ERROR(TSDB_CODE_PAR_INTER_OFFSET_TOO_BIG,      "Interval offset should be shorter than interval")
 TAOS_DEFINE_ERROR(TSDB_CODE_PAR_INTER_SLIDING_UNIT,        "Does not support sliding when interval is natural month/year")
 TAOS_DEFINE_ERROR(TSDB_CODE_PAR_INTER_SLIDING_TOO_BIG,     "sliding value no larger than the interval value")
-TAOS_DEFINE_ERROR(TSDB_CODE_PAR_INTER_SLIDING_TOO_SMALL,   "sliding value can not less than 1% of interval value")
+TAOS_DEFINE_ERROR(TSDB_CODE_PAR_INTER_SLIDING_TOO_SMALL,   "sliding value can not less than 1%% of interval value")
 TAOS_DEFINE_ERROR(TSDB_CODE_PAR_ONLY_ONE_JSON_TAG,         "Only one tag if there is a json tag")
 TAOS_DEFINE_ERROR(TSDB_CODE_PAR_INCORRECT_NUM_OF_COL,      "Query block has incorrect number of result columns")
 TAOS_DEFINE_ERROR(TSDB_CODE_PAR_INCORRECT_TIMESTAMP_VAL,   "Incorrect TIMESTAMP value")