Fix planner failures with overlapping mergejoin clauses in an outer join.

Given overlapping or partially redundant join clauses, for example
	t1 JOIN t2 ON t1.a = t2.x AND t1.b = t2.x
the planner's EquivalenceClass machinery will ordinarily refactor the
clauses as "t1.a = t1.b AND t1.a = t2.x", so that join processing doesn't
see multiple references to the same EquivalenceClass in a list of join
equality clauses.  However, if the join is outer, it's incorrect to derive
a restriction clause on the outer side from the join conditions, so the
clause refactoring does not happen and we end up with overlapping join
conditions.  The code that attempted to deal with such cases had several
subtle bugs, which could result in "left and right pathkeys do not match in
mergejoin" or "outer pathkeys do not match mergeclauses" planner errors,
if the selected join plan type was a mergejoin.  (It does not appear that
any actually incorrect plan could have been emitted.)

The core of the problem really was failure to recognize that the outer and
inner relations' pathkeys have different relationships to the mergeclause
list.  A join's mergeclause list is constructed by reference to the outer
pathkeys, so it will always be ordered the same as the outer pathkeys, but
this cannot be presumed true for the inner pathkeys.  If the inner sides of
the mergeclauses contain multiple references to the same EquivalenceClass
({t2.x} in the above example) then a simplistic rendering of the required
inner sort order is like "ORDER BY t2.x, t2.x", but the pathkey machinery
recognizes that the second sort column is redundant and throws it away.
The mergejoin planning code failed to account for that behavior properly.
One error was to try to generate cut-down versions of the mergeclause list
from cut-down versions of the inner pathkeys in the same way as the initial
construction of the mergeclause list from the outer pathkeys was done; this
could lead to choosing a mergeclause list that fails to match the outer
pathkeys.  The other problem was that the pathkey cross-checking code in
create_mergejoin_plan treated the inner and outer pathkey lists
identically, whereas actually the expectations for them must be different.
That led to false "pathkeys do not match" failures in some cases, and in
principle could have led to failure to detect bogus plans in other cases,
though there is no indication that such bogus plans could be generated.

Reported by Alexander Kuzmenkov, who also reviewed this patch.  This has
been broken for years (back to around 8.3 according to my testing), so
back-patch to all supported branches.

Discussion: https://postgr.es/m/[email protected]

Author: tglsfdc

src/backend/optimizer/path/joinpath.c

@@ -1009,10 +1009,10 @@ sort_inner_and_outer(PlannerInfo *root,
 			outerkeys = all_pathkeys;	/* no work at first one... */
 
 		/* Sort the mergeclauses into the corresponding ordering */
-		cur_mergeclauses = find_mergeclauses_for_pathkeys(root,
-														  outerkeys,
-														  true,
-														  extra->mergeclause_list);
+		cur_mergeclauses =
+			find_mergeclauses_for_outer_pathkeys(root,
+												 outerkeys,
+												 extra->mergeclause_list);
 
 		/* Should have used them all... */
 		Assert(list_length(cur_mergeclauses) == list_length(extra->mergeclause_list));
@@ -1102,10 +1102,10 @@ generate_mergejoin_paths(PlannerInfo *root,
 		jointype = JOIN_INNER;
 
 	/* Look for useful mergeclauses (if any) */
-	mergeclauses = find_mergeclauses_for_pathkeys(root,
-												  outerpath->pathkeys,
-												  true,
-												  extra->mergeclause_list);
+	mergeclauses =
+		find_mergeclauses_for_outer_pathkeys(root,
+											 outerpath->pathkeys,
+											 extra->mergeclause_list);
 
 	/*
 	 * Done with this outer path if no chance for a mergejoin.
@@ -1228,10 +1228,9 @@ generate_mergejoin_paths(PlannerInfo *root,
 			if (sortkeycnt < num_sortkeys)
 			{
 				newclauses =
-					find_mergeclauses_for_pathkeys(root,
-												   trialsortkeys,
-												   false,
-												   mergeclauses);
+					trim_mergeclauses_for_inner_pathkeys(root,
+														 mergeclauses,
+														 trialsortkeys);
 				Assert(newclauses != NIL);
 			}
 			else
@@ -1272,10 +1271,9 @@ generate_mergejoin_paths(PlannerInfo *root,
 					if (sortkeycnt < num_sortkeys)
 					{
 						newclauses =
-							find_mergeclauses_for_pathkeys(root,
-														   trialsortkeys,
-														   false,
-														   mergeclauses);
+							trim_mergeclauses_for_inner_pathkeys(root,
+																 mergeclauses,
+																 trialsortkeys);
 						Assert(newclauses != NIL);
 					}
 					else

src/backend/optimizer/path/pathkeys.c

@@ -981,29 +981,27 @@ update_mergeclause_eclasses(PlannerInfo *root, RestrictInfo *restrictinfo)
 }
 
 /*
- * find_mergeclauses_for_pathkeys
- *	  This routine attempts to find a set of mergeclauses that can be
- *	  used with a specified ordering for one of the input relations.
+ * find_mergeclauses_for_outer_pathkeys
+ *	  This routine attempts to find a list of mergeclauses that can be
+ *	  used with a specified ordering for the join's outer relation.
  *	  If successful, it returns a list of mergeclauses.
  *
- * 'pathkeys' is a pathkeys list showing the ordering of an input path.
- * 'outer_keys' is true if these keys are for the outer input path,
- *			false if for inner.
+ * 'pathkeys' is a pathkeys list showing the ordering of an outer-rel path.
  * 'restrictinfos' is a list of mergejoinable restriction clauses for the
- *			join relation being formed.
+ *			join relation being formed, in no particular order.
  *
  * The restrictinfos must be marked (via outer_is_left) to show which side
  * of each clause is associated with the current outer path.  (See
  * select_mergejoin_clauses())
  *
  * The result is NIL if no merge can be done, else a maximal list of
  * usable mergeclauses (represented as a list of their restrictinfo nodes).
+ * The list is ordered to match the pathkeys, as required for execution.
  */
 List *
-find_mergeclauses_for_pathkeys(PlannerInfo *root,
-							   List *pathkeys,
-							   bool outer_keys,
-							   List *restrictinfos)
+find_mergeclauses_for_outer_pathkeys(PlannerInfo *root,
+									 List *pathkeys,
+									 List *restrictinfos)
 {
 	List	   *mergeclauses = NIL;
 	ListCell   *i;
@@ -1044,32 +1042,29 @@ find_mergeclauses_for_pathkeys(PlannerInfo *root,
 		 *
 		 * It's possible that multiple matching clauses might have different
 		 * ECs on the other side, in which case the order we put them into our
-		 * result makes a difference in the pathkeys required for the other
-		 * input path.  However this routine hasn't got any info about which
+		 * result makes a difference in the pathkeys required for the inner
+		 * input rel.  However this routine hasn't got any info about which
 		 * order would be best, so we don't worry about that.
 		 *
 		 * It's also possible that the selected mergejoin clauses produce
-		 * a noncanonical ordering of pathkeys for the other side, ie, we
+		 * a noncanonical ordering of pathkeys for the inner side, ie, we
 		 * might select clauses that reference b.v1, b.v2, b.v1 in that
 		 * order.  This is not harmful in itself, though it suggests that
-		 * the clauses are partially redundant.  Since it happens only with
-		 * redundant query conditions, we don't bother to eliminate it.
-		 * make_inner_pathkeys_for_merge() has to delete duplicates when
-		 * it constructs the canonical pathkeys list, and we also have to
-		 * deal with the case in create_mergejoin_plan().
+		 * the clauses are partially redundant.  Since the alternative is
+		 * to omit mergejoin clauses and thereby possibly fail to generate a
+		 * plan altogether, we live with it.  make_inner_pathkeys_for_merge()
+		 * has to delete duplicates when it constructs the inner pathkeys
+		 * list, and we also have to deal with such cases specially in
+		 * create_mergejoin_plan().
 		 *----------
 		 */
 		foreach(j, restrictinfos)
 		{
 			RestrictInfo *rinfo = (RestrictInfo *) lfirst(j);
 			EquivalenceClass *clause_ec;
 
-			if (outer_keys)
-				clause_ec = rinfo->outer_is_left ?
-					rinfo->left_ec : rinfo->right_ec;
-			else
-				clause_ec = rinfo->outer_is_left ?
-					rinfo->right_ec : rinfo->left_ec;
+			clause_ec = rinfo->outer_is_left ?
+				rinfo->left_ec : rinfo->right_ec;
 			if (clause_ec == pathkey_ec)
 				matched_restrictinfos = lappend(matched_restrictinfos, rinfo);
 		}
@@ -1273,8 +1268,8 @@ select_outer_pathkeys_for_merge(PlannerInfo *root,
  *	  must be applied to an inner path to make it usable with the
  *	  given mergeclauses.
  *
- * 'mergeclauses' is a list of RestrictInfos for mergejoin clauses
- *			that will be used in a merge join.
+ * 'mergeclauses' is a list of RestrictInfos for the mergejoin clauses
+ *			that will be used in a merge join, in order.
  * 'outer_pathkeys' are the already-known canonical pathkeys for the outer
  *			side of the join.
  *
@@ -1351,8 +1346,13 @@ make_inner_pathkeys_for_merge(PlannerInfo *root,
 											 opathkey->pk_nulls_first);
 
 		/*
-		 * Don't generate redundant pathkeys (can happen if multiple
-		 * mergeclauses refer to same EC).
+		 * Don't generate redundant pathkeys (which can happen if multiple
+		 * mergeclauses refer to the same EC).  Because we do this, the output
+		 * pathkey list isn't necessarily ordered like the mergeclauses, which
+		 * complicates life for create_mergejoin_plan().  But if we didn't,
+		 * we'd have a noncanonical sort key list, which would be bad; for one
+		 * reason, it certainly wouldn't match any available sort order for
+		 * the input relation.
 		 */
 		if (!pathkey_is_redundant(pathkey, pathkeys))
 			pathkeys = lappend(pathkeys, pathkey);
@@ -1361,6 +1361,98 @@ make_inner_pathkeys_for_merge(PlannerInfo *root,
 	return pathkeys;
 }
 
+/*
+ * trim_mergeclauses_for_inner_pathkeys
+ *	  This routine trims a list of mergeclauses to include just those that
+ *	  work with a specified ordering for the join's inner relation.
+ *
+ * 'mergeclauses' is a list of RestrictInfos for mergejoin clauses for the
+ *			join relation being formed, in an order known to work for the
+ *			currently-considered sort ordering of the join's outer rel.
+ * 'pathkeys' is a pathkeys list showing the ordering of an inner-rel path;
+ *			it should be equal to, or a truncation of, the result of
+ *			make_inner_pathkeys_for_merge for these mergeclauses.
+ *
+ * What we return will be a prefix of the given mergeclauses list.
+ *
+ * We need this logic because make_inner_pathkeys_for_merge's result isn't
+ * necessarily in the same order as the mergeclauses.  That means that if we
+ * consider an inner-rel pathkey list that is a truncation of that result,
+ * we might need to drop mergeclauses even though they match a surviving inner
+ * pathkey.  This happens when they are to the right of a mergeclause that
+ * matches a removed inner pathkey.
+ *
+ * The mergeclauses must be marked (via outer_is_left) to show which side
+ * of each clause is associated with the current outer path.  (See
+ * select_mergejoin_clauses())
+ */
+List *
+trim_mergeclauses_for_inner_pathkeys(PlannerInfo *root,
+									 List *mergeclauses,
+									 List *pathkeys)
+{
+	List	   *new_mergeclauses = NIL;
+	PathKey    *pathkey;
+	EquivalenceClass *pathkey_ec;
+	bool		matched_pathkey;
+	ListCell   *lip;
+	ListCell   *i;
+
+	/* No pathkeys => no mergeclauses (though we don't expect this case) */
+	if (pathkeys == NIL)
+		return NIL;
+	/* Initialize to consider first pathkey */
+	lip = list_head(pathkeys);
+	pathkey = (PathKey *) lfirst(lip);
+	pathkey_ec = pathkey->pk_eclass;
+	lip = lnext(lip);
+	matched_pathkey = false;
+
+	/* Scan mergeclauses to see how many we can use */
+	foreach(i, mergeclauses)
+	{
+		RestrictInfo *rinfo = (RestrictInfo *) lfirst(i);
+		EquivalenceClass *clause_ec;
+
+		/* Assume we needn't do update_mergeclause_eclasses again here */
+
+		/* Check clause's inner-rel EC against current pathkey */
+		clause_ec = rinfo->outer_is_left ?
+			rinfo->right_ec : rinfo->left_ec;
+
+		/* If we don't have a match, attempt to advance to next pathkey */
+		if (clause_ec != pathkey_ec)
+		{
+			/* If we had no clauses matching this inner pathkey, must stop */
+			if (!matched_pathkey)
+				break;
+
+			/* Advance to next inner pathkey, if any */
+			if (lip == NULL)
+				break;
+			pathkey = (PathKey *) lfirst(lip);
+			pathkey_ec = pathkey->pk_eclass;
+			lip = lnext(lip);
+			matched_pathkey = false;
+		}
+
+		/* If mergeclause matches current inner pathkey, we can use it */
+		if (clause_ec == pathkey_ec)
+		{
+			new_mergeclauses = lappend(new_mergeclauses, rinfo);
+			matched_pathkey = true;
+		}
+		else
+		{
+			/* Else, no hope of adding any more mergeclauses */
+			break;
+		}
+	}
+
+	return new_mergeclauses;
+}
+
+
 /****************************************************************************
  *		PATHKEY USEFULNESS CHECKS
  *