From: Tom Lane Date: Mon, 19 Feb 2018 21:00:18 +0000 (-0500) Subject: Fix misbehavior of CTE-used-in-a-subplan during EPQ rechecks. X-Git-Tag: REL_11_BETA1~753 X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=159efe4af4509741c25d6b95ddd9fda86facce42;p=postgresql Fix misbehavior of CTE-used-in-a-subplan during EPQ rechecks. An updating query that reads a CTE within an InitPlan or SubPlan could get incorrect results if it updates rows that are concurrently being modified. This is caused by CteScanNext supposing that nothing inside its recursive ExecProcNode call could change which read pointer is selected in the CTE's shared tuplestore. While that's normally true because of scoping considerations, it can break down if an EPQ plan tree gets built during the call, because EvalPlanQualStart builds execution trees for all subplans whether they're going to be used during the recheck or not. And it seems like a pretty shaky assumption anyway, so let's just reselect our own read pointer here. Per bug #14870 from Andrei Gorita. This has been broken since CTEs were implemented, so back-patch to all supported branches. Discussion: https://postgr.es/m/20171024155358.1471.82377@wrigleys.postgresql.org --- diff --git a/src/backend/executor/nodeCtescan.c b/src/backend/executor/nodeCtescan.c index 218619c760..24700dd396 100644 --- a/src/backend/executor/nodeCtescan.c +++ b/src/backend/executor/nodeCtescan.c @@ -107,6 +107,13 @@ CteScanNext(CteScanState *node) return NULL; } + /* + * There are corner cases where the subplan could change which + * tuplestore read pointer is active, so be sure to reselect ours + * before storing the tuple we got. + */ + tuplestore_select_read_pointer(tuplestorestate, node->readptr); + /* * Append a copy of the returned tuple to tuplestore. NOTE: because * our read pointer is certainly in EOF state, its read position will @@ -178,6 +185,12 @@ ExecInitCteScan(CteScan *node, EState *estate, int eflags) * we might be asked to rescan the CTE even though upper levels didn't * tell us to be prepared to do it efficiently. Annoying, since this * prevents truncation of the tuplestore. XXX FIXME + * + * Note: if we are in an EPQ recheck plan tree, it's likely that no access + * to the tuplestore is needed at all, making this even more annoying. + * It's not worth improving that as long as all the read pointers would + * have REWIND anyway, but if we ever improve this logic then that aspect + * should be considered too. */ eflags |= EXEC_FLAG_REWIND; diff --git a/src/test/isolation/expected/eval-plan-qual.out b/src/test/isolation/expected/eval-plan-qual.out index eb40717679..fed01459cf 100644 --- a/src/test/isolation/expected/eval-plan-qual.out +++ b/src/test/isolation/expected/eval-plan-qual.out @@ -217,3 +217,18 @@ id data id data 9 0 9 0 10 0 10 0 step c1: COMMIT; + +starting permutation: wrtwcte multireadwcte c1 c2 +step wrtwcte: UPDATE table_a SET value = 'tableAValue2' WHERE id = 1; +step multireadwcte: + WITH updated AS ( + UPDATE table_a SET value = 'tableAValue3' WHERE id = 1 RETURNING id + ) + SELECT (SELECT id FROM updated) AS subid, * FROM updated; + +step c1: COMMIT; +step c2: COMMIT; +step multireadwcte: <... completed> +subid id + +1 1 diff --git a/src/test/isolation/specs/eval-plan-qual.spec b/src/test/isolation/specs/eval-plan-qual.spec index d2b34ec7cc..51ab61dc86 100644 --- a/src/test/isolation/specs/eval-plan-qual.spec +++ b/src/test/isolation/specs/eval-plan-qual.spec @@ -139,6 +139,14 @@ step "readwcte" { SELECT * FROM cte2; } +# this test exercises a different CTE misbehavior, cf bug #14870 +step "multireadwcte" { + WITH updated AS ( + UPDATE table_a SET value = 'tableAValue3' WHERE id = 1 RETURNING id + ) + SELECT (SELECT id FROM updated) AS subid, * FROM updated; +} + teardown { COMMIT; } permutation "wx1" "wx2" "c1" "c2" "read" @@ -151,3 +159,4 @@ permutation "wx2" "lockwithvalues" "c2" "c1" "read" permutation "updateforss" "readforss" "c1" "c2" permutation "wrtwcte" "readwcte" "c1" "c2" permutation "wrjt" "selectjoinforupdate" "c2" "c1" +permutation "wrtwcte" "multireadwcte" "c1" "c2"