From c538e14a0d785c76ec5555673418ae1c17ff79ba Mon Sep 17 00:00:00 2001 From: Amit Kapila Date: Mon, 2 Sep 2019 12:36:21 +0530 Subject: [PATCH 1/2] Forbid Limit node to shutdown resources. As part of commit 19df1702f5, we allowed a Limit node to shut down the resources to get the correct stats from parallel nodes beneath it. The idea was correct, but we could not correctly identify the end of execution and it ends up destroying the parallel context which is required for rescans. The correct fix is to identify whether rescans are possible and then accordingly shutdown resources. This would require us to invent some new infrastructure which we can't back-patch. So, for now, we will just disallow to shutdown resources via Limit node. This will open up the issue of incorrect stats when there are parallel nodes beneath Limit node, but it will still be better to at least allow such queries to run to completion and give correct results. Reported-by: Jerry Sievers Diagnosed-by: Thomas Munro Author: Amit Kapila Backpatch-through: 9.6 Discussion: https://postgr.es/m/87ims2amh6.fsf@jsievers.enova.com --- src/backend/executor/nodeLimit.c | 8 -------- 1 file changed, 8 deletions(-) diff --git a/src/backend/executor/nodeLimit.c b/src/backend/executor/nodeLimit.c index baa669a..d21b5ce 100644 --- a/src/backend/executor/nodeLimit.c +++ b/src/backend/executor/nodeLimit.c @@ -134,14 +134,6 @@ ExecLimit(PlanState *pstate) node->position - node->offset >= node->count) { node->lstate = LIMIT_WINDOWEND; - - /* - * If we know we won't need to back up, we can release - * resources at this point. - */ - if (!(node->ps.state->es_top_eflags & EXEC_FLAG_BACKWARD)) - (void) ExecShutdownNode(outerPlan); - return NULL; } -- 1.8.3.1