On Thu, Mar 14, 2019 at 11:05 AM Thomas Munro <thomas.mu...@gmail.com> wrote: > I renamed the CVs because the names I had used before broke the > convention that variables named ckpt_* are protected by ckpt_lck, and > pushed.
Erm... this made successful checkpoints slightly faster but failed checkpoints infinitely slower. It would help if we woke up CV waiters in the error path too. Patch attached. -- Thomas Munro https://enterprisedb.com
From b6ad979bf2280eee1b0fbbe4c13eaefeb82743c8 Mon Sep 17 00:00:00 2001 From: Thomas Munro <thomas.mu...@gmail.com> Date: Fri, 5 Apr 2019 21:43:29 +1300 Subject: [PATCH] Wake up interested backends when a checkpoint fails. Commit c6c9474a switched to condition variables instead of sleep loops to notify backends of checkpoint start and stop, but forgot to broadcast in case of checkpoint failure. Author: Thomas Munro Discussion: --- src/backend/postmaster/checkpointer.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/src/backend/postmaster/checkpointer.c b/src/backend/postmaster/checkpointer.c index d303ce3679..a1e04239ad 100644 --- a/src/backend/postmaster/checkpointer.c +++ b/src/backend/postmaster/checkpointer.c @@ -279,6 +279,8 @@ CheckpointerMain(void) CheckpointerShmem->ckpt_done = CheckpointerShmem->ckpt_started; SpinLockRelease(&CheckpointerShmem->ckpt_lck); + ConditionVariableBroadcast(&CheckpointerShmem->done_cv); + ckpt_active = false; } -- 2.21.0
