Merge pull request #2569 from citusdata/fix_having_with_joins

Fix having clause bug for complex joins

src/backend/distributed/planner/multi_physical_planner.c

@@ -106,6 +106,8 @@ static RangeTblEntry * DerivedRangeTableEntry(MultiNode *multiNode, List *column
 											  List *tableIdList);
 static List * DerivedColumnNameList(uint32 columnCount, uint64 generatingJobId);
 static Query * BuildSubqueryJobQuery(MultiNode *multiNode);
+static void UpdateAllColumnAttributes(Node *columnContainer, List *rangeTableList,
+									  List *dependedJobList);
 static void UpdateColumnAttributes(Var *column, List *rangeTableList,
 								   List *dependedJobList);
 static Index NewTableId(Index originalTableId, List *rangeTableList);
@@ -579,10 +581,8 @@ BuildJobQuery(MultiNode *multiNode, List *dependedJobList)
 	List *sortClauseList = NIL;
 	List *groupClauseList = NIL;
 	List *selectClauseList = NIL;
-	List *columnList = NIL;
 	Node *limitCount = NULL;
 	Node *limitOffset = NULL;
-	ListCell *columnCell = NULL;
 	FromExpr *joinTree = NULL;
 	Node *joinRoot = NULL;
 	Node *havingQual = NULL;
@@ -650,13 +650,7 @@ BuildJobQuery(MultiNode *multiNode, List *dependedJobList)
 	/* update the column attributes for target entries */
 	if (updateColumnAttributes)
 	{
-		ListCell *columnCell = NULL;
+		UpdateAllColumnAttributes((Node *) targetList, rangeTableList, dependedJobList);
-		List *columnList = pull_var_clause_default((Node *) targetList);
-		foreach(columnCell, columnList)
-		{
-			Var *column = (Var *) lfirst(columnCell);
-			UpdateColumnAttributes(column, rangeTableList, dependedJobList);
-		}
 	}
 
 	/* extract limit count/offset and sort clauses */
@@ -676,16 +670,12 @@ BuildJobQuery(MultiNode *multiNode, List *dependedJobList)
 	/* build the where clause list using select predicates */
 	selectClauseList = QuerySelectClauseList(multiNode);
 
-	/* set correct column attributes for select columns */
+	/* set correct column attributes for select and having clauses */
 	if (updateColumnAttributes)
 	{
-		columnCell = NULL;
+		UpdateAllColumnAttributes((Node *) selectClauseList, rangeTableList,
-		columnList = pull_var_clause_default((Node *) selectClauseList);
+								  dependedJobList);
-		foreach(columnCell, columnList)
+		UpdateAllColumnAttributes(havingQual, rangeTableList, dependedJobList);
-		{
-			Var *column = (Var *) lfirst(columnCell);
-			UpdateColumnAttributes(column, rangeTableList, dependedJobList);
-		}
 	}
 
 	/*
@@ -1526,6 +1516,25 @@ BuildSubqueryJobQuery(MultiNode *multiNode)
 }
 
 
+/*
+ * UpdateAllColumnAttributes extracts column references from provided columnContainer
+ * and calls UpdateColumnAttributes to updates the column's range table reference (varno) and
+ * column attribute number for the range table (varattno).
+ */
+static void
+UpdateAllColumnAttributes(Node *columnContainer, List *rangeTableList,
+						  List *dependedJobList)
+{
+	ListCell *columnCell = NULL;
+	List *columnList = pull_var_clause_default(columnContainer);
+	foreach(columnCell, columnList)
+	{
+		Var *column = (Var *) lfirst(columnCell);
+		UpdateColumnAttributes(column, rangeTableList, dependedJobList);
+	}
+}
+
+
 /*
  * UpdateColumnAttributes updates the column's range table reference (varno) and
  * column attribute number for the range table (varattno). The function uses the

src/test/regress/expected/multi_view.out

@@ -428,8 +428,8 @@ ORDER BY 2 DESC, 1;
        5 |    14
 (3 rows)
 
--- non-partition key joins are not supported inside subquery
+-- non-partition key joins are supported inside subquery
--- since the join with a table
+-- via pull-push execution
 SELECT * FROM
 	(SELECT ru.user_id, count(*) 
 		FROM recent_users ru 
@@ -438,7 +438,13 @@ SELECT * FROM
 		GROUP BY ru.user_id
 		ORDER BY 2 DESC, 1) s1
 ORDER BY 2 DESC, 1;
-ERROR:  bogus varno: 3
+ user_id | count 
+---------+-------
+       1 |    24
+       3 |    23
+       5 |     7
+(3 rows)
+
 -- join between views
 -- recent users who has an event in recent events
 SELECT ru.user_id FROM recent_users ru JOIN recent_events re USING(user_id) GROUP BY ru.user_id ORDER BY ru.user_id;
@@ -533,7 +539,8 @@ SELECT * FROM
 		ON(ru.user_id = et.event_type)
 	) s1
 ORDER BY 2 DESC, 1;
-ERROR:  bogus varno: 3
+ERROR:  cannot pushdown the subquery
+DETAIL:  Complex subqueries and CTEs cannot be in the outer part of the outer join
 -- create a select only view
 CREATE VIEW selected_users AS SELECT * FROM users_table WHERE value_1 >= 1 and value_1 <3;
 CREATE VIEW recent_selected_users AS SELECT su.* FROM selected_users su JOIN recent_users ru USING(user_id);
@@ -863,7 +870,7 @@ EXPLAIN (COSTS FALSE) SELECT et.* FROM recent_10_users JOIN events_table et USIN
    ->  Sort
          Sort Key: remote_scan."time" DESC
          ->  Custom Scan (Citus Real-Time)
-               ->  Distributed Subplan 95_1
+               ->  Distributed Subplan 96_1
                      ->  Limit
                            ->  Sort
                                  Sort Key: max((max(remote_scan.lastseen))) DESC

src/test/regress/expected/multi_view_0.out

@@ -428,8 +428,8 @@ ORDER BY 2 DESC, 1;
        5 |    14
 (3 rows)
 
--- non-partition key joins are not supported inside subquery
+-- non-partition key joins are supported inside subquery
--- since the join with a table
+-- via pull-push execution
 SELECT * FROM
 	(SELECT ru.user_id, count(*) 
 		FROM recent_users ru 
@@ -438,7 +438,13 @@ SELECT * FROM
 		GROUP BY ru.user_id
 		ORDER BY 2 DESC, 1) s1
 ORDER BY 2 DESC, 1;
-ERROR:  bogus varno: 3
+ user_id | count 
+---------+-------
+       1 |    24
+       3 |    23
+       5 |     7
+(3 rows)
+
 -- join between views
 -- recent users who has an event in recent events
 SELECT ru.user_id FROM recent_users ru JOIN recent_events re USING(user_id) GROUP BY ru.user_id ORDER BY ru.user_id;
@@ -533,7 +539,8 @@ SELECT * FROM
 		ON(ru.user_id = et.event_type)
 	) s1
 ORDER BY 2 DESC, 1;
-ERROR:  bogus varno: 3
+ERROR:  cannot pushdown the subquery
+DETAIL:  Complex subqueries and CTEs cannot be in the outer part of the outer join
 -- create a select only view
 CREATE VIEW selected_users AS SELECT * FROM users_table WHERE value_1 >= 1 and value_1 <3;
 CREATE VIEW recent_selected_users AS SELECT su.* FROM selected_users su JOIN recent_users ru USING(user_id);
@@ -865,7 +872,7 @@ EXPLAIN (COSTS FALSE) SELECT et.* FROM recent_10_users JOIN events_table et USIN
    ->  Sort
          Sort Key: remote_scan."time" DESC
          ->  Custom Scan (Citus Real-Time)
-               ->  Distributed Subplan 95_1
+               ->  Distributed Subplan 96_1
                      ->  Limit
                            ->  Sort
                                  Sort Key: max((max(remote_scan.lastseen))) DESC

src/test/regress/sql/multi_view.sql

@@ -220,8 +220,8 @@ SELECT * FROM
 		ORDER BY 2 DESC, 1) s1
 ORDER BY 2 DESC, 1;
 
--- non-partition key joins are not supported inside subquery
+-- non-partition key joins are supported inside subquery
--- since the join with a table
+-- via pull-push execution
 SELECT * FROM
 	(SELECT ru.user_id, count(*) 
 		FROM recent_users ru