Merge pull request #22070 from wangjiaming0909/fix/m/TS-3671

fix: state/session/event window with order by _wstart/_wend has no effect

include/libs/nodes/plannodes.h

@@ -247,6 +247,7 @@ typedef struct SSortLogicNode {
   SNodeList* pSortKeys;
   bool       groupSort;
   int64_t    maxRows;
+  bool       skipPKSortOpt;
 } SSortLogicNode;
 
 typedef struct SPartitionLogicNode {

source/libs/planner/src/planLogicCreater.c

@@ -1034,7 +1034,6 @@ static int32_t createSortLogicNode(SLogicPlanContext* pCxt, SSelectStmt* pSelect
   pSort->node.resultDataOrder = isPrimaryKeySort(pSelect->pOrderByList)
                                     ? (pSort->groupSort ? DATA_ORDER_LEVEL_IN_GROUP : DATA_ORDER_LEVEL_GLOBAL)
                                     : DATA_ORDER_LEVEL_NONE;
-
   int32_t code = nodesCollectColumns(pSelect, SQL_CLAUSE_ORDER_BY, NULL, COLLECT_COL_TYPE_ALL, &pSort->node.pTargets);
   if (TSDB_CODE_SUCCESS == code && NULL == pSort->node.pTargets) {
     code = nodesListMakeStrictAppend(&pSort->node.pTargets,
@@ -1048,6 +1047,7 @@ static int32_t createSortLogicNode(SLogicPlanContext* pCxt, SSelectStmt* pSelect
     }
     SNode*            pNode = NULL;
     SOrderByExprNode* firstSortKey = (SOrderByExprNode*)nodesListGetNode(pSort->pSortKeys, 0);
+    if (isPrimaryKeySort(pSelect->pOrderByList)) pSort->node.outputTsOrder = firstSortKey->order;
     if (firstSortKey->pExpr->type == QUERY_NODE_COLUMN) {
       SColumnNode* pCol = (SColumnNode*)firstSortKey->pExpr;
       int16_t      projIdx = 1;

source/libs/planner/src/planOptimizer.c

@@ -1168,7 +1168,8 @@ static bool sortPriKeyOptMayBeOptimized(SLogicNode* pNode) {
     return false;
   }
   SSortLogicNode* pSort = (SSortLogicNode*)pNode;
-  if (!sortPriKeyOptIsPriKeyOrderBy(pSort->pSortKeys) || 1 != LIST_LENGTH(pSort->node.pChildren)) {
+  if (pSort->skipPKSortOpt || !sortPriKeyOptIsPriKeyOrderBy(pSort->pSortKeys) ||
+      1 != LIST_LENGTH(pSort->node.pChildren)) {
     return false;
   }
   SNode* pChild;
@@ -1181,8 +1182,8 @@ static bool sortPriKeyOptMayBeOptimized(SLogicNode* pNode) {
   return true;
 }
 
-static int32_t sortPriKeyOptGetSequencingNodesImpl(SLogicNode* pNode, bool groupSort, bool* pNotOptimize,
-                                                   SNodeList** pSequencingNodes) {
+static int32_t sortPriKeyOptGetSequencingNodesImpl(SLogicNode* pNode, bool groupSort, EOrder sortOrder,
+                                                   bool* pNotOptimize, SNodeList** pSequencingNodes) {
   if (NULL != pNode->pLimit || NULL != pNode->pSlimit) {
     *pNotOptimize = false;
     return TSDB_CODE_SUCCESS;
@@ -1199,15 +1200,21 @@ static int32_t sortPriKeyOptGetSequencingNodesImpl(SLogicNode* pNode, bool group
     }
     case QUERY_NODE_LOGIC_PLAN_JOIN: {
       int32_t code = sortPriKeyOptGetSequencingNodesImpl((SLogicNode*)nodesListGetNode(pNode->pChildren, 0), groupSort,
-                                                         pNotOptimize, pSequencingNodes);
+                                                         sortOrder, pNotOptimize, pSequencingNodes);
       if (TSDB_CODE_SUCCESS == code) {
         code = sortPriKeyOptGetSequencingNodesImpl((SLogicNode*)nodesListGetNode(pNode->pChildren, 1), groupSort,
-                                                   pNotOptimize, pSequencingNodes);
+                                                   sortOrder, pNotOptimize, pSequencingNodes);
       }
       return code;
     }
-    case QUERY_NODE_LOGIC_PLAN_WINDOW:
-      return nodesListMakeAppend(pSequencingNodes, (SNode*)pNode);
+    case QUERY_NODE_LOGIC_PLAN_WINDOW: {
+      SWindowLogicNode* pWindowLogicNode = (SWindowLogicNode*)pNode;
+      // For interval window, we always apply sortPriKey optimization.
+      // For session/event/state window, the output ts order will always be ASC.
+      // If sort order is also asc, we apply optimization, otherwise we keep sort node to get correct output order.
+      if (pWindowLogicNode->winType == WINDOW_TYPE_INTERVAL || sortOrder == ORDER_ASC)
+        return nodesListMakeAppend(pSequencingNodes, (SNode*)pNode);
+    }
     case QUERY_NODE_LOGIC_PLAN_AGG:
     case QUERY_NODE_LOGIC_PLAN_PARTITION:
       *pNotOptimize = true;
@@ -1221,23 +1228,25 @@ static int32_t sortPriKeyOptGetSequencingNodesImpl(SLogicNode* pNode, bool group
     return TSDB_CODE_SUCCESS;
   }
 
-  return sortPriKeyOptGetSequencingNodesImpl((SLogicNode*)nodesListGetNode(pNode->pChildren, 0), groupSort,
+  return sortPriKeyOptGetSequencingNodesImpl((SLogicNode*)nodesListGetNode(pNode->pChildren, 0), groupSort, sortOrder,
                                              pNotOptimize, pSequencingNodes);
 }
 
-static int32_t sortPriKeyOptGetSequencingNodes(SLogicNode* pNode, bool groupSort, SNodeList** pSequencingNodes) {
+static EOrder sortPriKeyOptGetPriKeyOrder(SSortLogicNode* pSort) {
+  return ((SOrderByExprNode*)nodesListGetNode(pSort->pSortKeys, 0))->order;
+}
+
+static int32_t sortPriKeyOptGetSequencingNodes(SSortLogicNode* pSort, bool groupSort, SNodeList** pSequencingNodes) {
   bool    notOptimize = false;
-  int32_t code = sortPriKeyOptGetSequencingNodesImpl(pNode, groupSort, &notOptimize, pSequencingNodes);
+  int32_t code =
+      sortPriKeyOptGetSequencingNodesImpl((SLogicNode*)nodesListGetNode(pSort->node.pChildren, 0), groupSort,
+                                          sortPriKeyOptGetPriKeyOrder(pSort), &notOptimize, pSequencingNodes);
   if (TSDB_CODE_SUCCESS != code || notOptimize) {
     NODES_CLEAR_LIST(*pSequencingNodes);
   }
   return code;
 }
 
-static EOrder sortPriKeyOptGetPriKeyOrder(SSortLogicNode* pSort) {
-  return ((SOrderByExprNode*)nodesListGetNode(pSort->pSortKeys, 0))->order;
-}
-
 static int32_t sortPriKeyOptApply(SOptimizeContext* pCxt, SLogicSubplan* pLogicSubplan, SSortLogicNode* pSort,
                                   SNodeList* pSequencingNodes) {
   EOrder order = sortPriKeyOptGetPriKeyOrder(pSort);
@@ -1276,10 +1285,17 @@ static int32_t sortPriKeyOptApply(SOptimizeContext* pCxt, SLogicSubplan* pLogicS
 
 static int32_t sortPrimaryKeyOptimizeImpl(SOptimizeContext* pCxt, SLogicSubplan* pLogicSubplan, SSortLogicNode* pSort) {
   SNodeList* pSequencingNodes = NULL;
-  int32_t    code = sortPriKeyOptGetSequencingNodes((SLogicNode*)nodesListGetNode(pSort->node.pChildren, 0),
-                                                    pSort->groupSort, &pSequencingNodes);
-  if (TSDB_CODE_SUCCESS == code && NULL != pSequencingNodes) {
-    code = sortPriKeyOptApply(pCxt, pLogicSubplan, pSort, pSequencingNodes);
+  int32_t    code = sortPriKeyOptGetSequencingNodes(pSort, pSort->groupSort, &pSequencingNodes);
+  if (TSDB_CODE_SUCCESS == code) {
+    if (pSequencingNodes != NULL) {
+      code = sortPriKeyOptApply(pCxt, pLogicSubplan, pSort, pSequencingNodes);
+    } else {
+      // if we decided not to push down sort info to children, we should propagate output ts order to parents of pSort
+      optSetParentOrder(pSort->node.pParent, sortPriKeyOptGetPriKeyOrder(pSort), 0);
+      // we need to prevent this pSort from being chosen to do optimization again
+      pSort->skipPKSortOpt = true;
+      pCxt->optimized = true;
+    }
   }
   nodesClearList(pSequencingNodes);
   return code;

tests/script/tsim/query/t/explain_tsorder.sql

@@ -98,3 +98,65 @@ select last(ts), c2 as d from d1 group by c2 order by c2 asc limit 9,1;
 select last(ts) as ts, c2 as d from d1 group by c2 order by ts desc, c2 asc limit 10;
 select last(ts) as ts, c2 as d from d1 group by c2 order by ts desc, c2 asc limit 2,8;
 select last(ts) as ts, c2 as d from d1 group by c2 order by ts desc, c2 asc limit 9,1;
+
+explain verbose true select _wstart, _wend, count(*) from meters event_window start with c2 > 0 end with c2 < 100\G;
+explain verbose true select _wstart, _wend, count(*) from meters event_window start with c2 > 0 end with c2 < 100 order by _wstart desc\G;
+explain verbose true select _wstart, _wend, count(*) from meters event_window start with c2 > 0 end with c2 < 100 order by _wstart asc\G;
+
+explain verbose true select _wstart, _wend, count(*) from meters event_window start with c2 > 0 end with c2 < 100 order by _wend desc\G;
+explain verbose true select _wstart, _wend, count(*) from meters event_window start with c2 > 0 end with c2 < 100 order by _wend asc\G;
+
+select _wstart, _wend, count(*) from meters event_window start with c2 > 0 end with c2 < 100;
+select _wstart, _wend, count(*) from meters event_window start with c2 > 0 end with c2 < 100 order by _wstart desc;
+select _wstart, _wend, count(*) from meters event_window start with c2 > 0 end with c2 < 100 order by _wstart asc;
+
+select _wstart, _wend, count(*) from meters event_window start with c2 > 0 end with c2 < 100 order by _wend desc;
+select _wstart, _wend, count(*) from meters event_window start with c2 > 0 end with c2 < 100 order by _wend asc;
+
+explain verbose true select _wstart, _wend, count(*) from meters session(ts, 1h)\G;
+explain verbose true select _wstart, _wend, count(*) from meters session(ts, 1h) order by _wstart desc\G;
+explain verbose true select _wstart, _wend, count(*) from meters session(ts, 1h) order by _wstart asc\G;
+
+explain verbose true select _wstart, _wend, count(*) from meters session(ts, 1h) order by _wend desc\G;
+explain verbose true select _wstart, _wend, count(*) from meters session(ts, 1h) order by _wend asc\G;
+
+select _wstart, _wend, count(*) from meters session(ts, 1h);
+select _wstart, _wend, count(*) from meters session(ts, 1h) order by _wstart desc;
+select _wstart, _wend, count(*) from meters session(ts, 1h) order by _wstart asc;
+
+select _wstart, _wend, count(*) from meters session(ts, 1h) order by _wend desc;
+select _wstart, _wend, count(*) from meters session(ts, 1h) order by _wend asc;
+
+
+explain verbose true select _wstart, _wend, count(*) from meters session(ts, 1h)\G;
+explain verbose true select _wstart, _wend, count(*) from meters session(ts, 1h) order by _wstart desc\G;
+explain verbose true select _wstart, _wend, count(*) from meters session(ts, 1h) order by _wstart asc\G;
+
+explain verbose true select _wstart, _wend, count(*) from meters session(ts, 1h) order by _wend desc\G;
+explain verbose true select _wstart, _wend, count(*) from meters session(ts, 1h) order by _wend asc\G;
+
+select _wstart, _wend, count(*) from meters session(ts, 1h);
+select _wstart, _wend, count(*) from meters session(ts, 1h) order by _wstart desc;
+select _wstart, _wend, count(*) from meters session(ts, 1h) order by _wstart asc;
+
+select _wstart, _wend, count(*) from meters session(ts, 1h) order by _wend desc;
+select _wstart, _wend, count(*) from meters session(ts, 1h) order by _wend asc;
+
+explain verbose true select _wstart, _wend, count(*), last(ts) from meters state_window(c2)\G;
+explain verbose true select _wstart, _wend, count(*), last(ts) from meters state_window(c2) order by _wstart desc\G;
+explain verbose true select _wstart, _wend, count(*), last(ts) from meters state_window(c2) order by _wstart asc\G;
+
+explain verbose true select _wstart, _wend, count(*), last(ts) from meters state_window(c2) order by _wend desc\G;
+explain verbose true select _wstart, _wend, count(*), last(ts) from meters state_window(c2) order by _wend asc\G;
+
+select _wstart, _wend, count(*), last(ts) from meters state_window(c2);
+select _wstart, _wend, count(*), last(ts) from meters state_window(c2) order by _wstart desc;
+select _wstart, _wend, count(*), last(ts) from meters state_window(c2) order by _wstart asc;
+
+select _wstart, _wend, count(*), last(ts) from meters state_window(c2) order by _wend desc;
+select _wstart, _wend, count(*), last(ts) from meters state_window(c2) order by _wend asc;
+
+explain verbose true select _wstart, _wend, count(*), last(ts) from meters state_window(c2) order by _wend asc, count(*) desc\G;
+
+explain verbose true select _wstart, _wend, last(ts) from (select _wstart as ts, _wend, count(*), last(ts) from meters state_window(c2) order by _wend desc) interval(1h) order by _wstart desc\G;
+explain verbose true select _wstart, _wend, last(ts) from (select _wstart as ts, _wend, count(*), last(ts) from meters state_window(c2) order by _wend asc) interval(1h) order by _wstart desc\G;