Horton

Re: Shutdown indefinitely stuck due to unflushed FPI_FOR_HINT record

Fujii Masao <masao.fujii@gmail.com>

View thread View on postgr.es
From: Fujii Masao <masao.fujii@gmail.com>
To: Anthonin Bonnefoy <anthonin.bonnefoy@datadoghq.com>
Cc: Andres Freund <andres@anarazel.de>, Alexander Lakhin <exclusion@gmail.com>, PostgreSQL Hackers <pgsql-hackers@postgresql.org>
Date: 2026-03-16T23:16:08Z
Lists: pgsql-hackers 
On Mon, Mar 16, 2026 at 6:45 PM Anthonin Bonnefoy
<anthonin.bonnefoy@datadoghq.com> wrote:
>
> On Mon, Mar 16, 2026 at 6:39 AM Fujii Masao <masao.fujii@gmail.com> wrote:
> > Since Tomas added GetXLogInsertEndRecPtr() in commit b1f14c96720,
> > I updated the patch to use it. Patch attached.
> > Barring any objections, I will commit it.
> >
> > -       XLogFlush(GetXLogWriteRecPtr());
> > +       XLogFlush(GetXLogInsertEndRecPtr());
>
> Thanks for the updated patch! I've run my test script against the
> patch and there's no more "xlog flush request xxx is not satisfied"
> errors reported.

Thanks for the test! I've pushed the patch.


> > I excluded the above change from the patch because it seems like a separate
> > issue. I also wonder whether this code could cause an error in XLogFlush()
> > even when GetXLogWriteRecPtr() is used.
>
> Ha right, I've mixed Insert and Write and thought that
> FinishSyncWorker was also doing a XlogFlush(GetXLogInsertRecPtr())
> when writing the patch. If I try to trigger the partial record issue,
> GetXLogWriteRecPtr() points at the end of the WAL page containing the
> beginning of the FPI_FOR_HINT, there's no attempt to flush in the
> future. So FinishSyncWorker doesn't seem impacted by the issue.

Understood. Thanks for the investigation!

Regards,

-- 
Fujii Masao

Commits

  1. Fix WAL flush LSN used by logical walsender during shutdown

  2. Use GetXLogInsertEndRecPtr in gistGetFakeLSN

  3. Fix publisher shutdown hang caused by logical walsender busy loop.