Re: [PATCH] Prevent repeated deadlock-check signals in standby buffer pin waits

Hi Andrey,

Thanks for looking at this, and for handling the CF entry.

On Mon, Aug 31, 2026 at 2:19 AM Andrey Borodin <x4mmm(at)yandex-team(dot)ru> wrote:
> One point seems worth recording for the
> next version: Xuneng's June 29 lost-wakeup and reply-stall interleaving
> has not been answered in this thread. Could the rebased patch either
> lose that race or explain why it cannot occur?

It closes it. The prerequisite patch is still under review and may
change, but as of its latest version it already implements the
recheck-after-wake Xuneng proposed. The wait loop checks the buffer's
pin count directly after every wakeup and re-registers the waiter flag
before sleeping again, so a coalesced wakeup can no longer leave the
startup process waiting for a signal that will never arrive.

Thank you.

--
JH Shin