From: | Fujii Masao <masao(dot)fujii(at)gmail(dot)com> |
---|---|
To: | Michael Paquier <michael(at)paquier(dot)xyz> |
Cc: | PostgreSQL-development <pgsql-hackers(at)postgresql(dot)org> |
Subject: | Re: pg_promote() can cause busy loop |
Date: | 2019-09-05 01:53:19 |
Message-ID: | CAHGQGwGHmNTZmf7fWFMHgJWq6vZLbDk_XR+Jegfsf=vCDdPo+w@mail.gmail.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
On Thu, Sep 5, 2019 at 10:26 AM Michael Paquier <michael(at)paquier(dot)xyz> wrote:
>
> On Thu, Sep 05, 2019 at 09:46:26AM +0900, Fujii Masao wrote:
> > I found small issue in pg_promote(). If postmaster dies
> > while pg_promote() is waiting for the standby promotion to finish,
> > pg_promote() can cause busy loop. This happens because
> > pg_promote() does nothing when WaitLatch() detects
> > the postmaster death event. I think that pg_promote()
> > should bail out of the loop immediately in that case.
> >
> > Attached is the patch for the fix.
>
> Indeed, this is not correct.
>
> - ereport(WARNING,
> - (errmsg("server did not promote within %d seconds",
> - wait_seconds)));
> + if (i >= WAITS_PER_SECOND * wait_seconds)
> + ereport(WARNING,
> + (errmsg("server did not promote within %d seconds", wait_seconds)));
>
> Would it make more sense to issue a warning mentioning the postmaster
> death and then return PG_RETURN_BOOL(false) instead of breaking out of
> the loop? It could be confusing to warn about a timeout if the
> postmaster died in parallel, and we know the actual reason why the
> promotion did not happen in this case.
It's ok to use PG_RETURN_BOOL(false) instead of breaking out of the loop
in that case. Which would make the code simpler.
But I don't think it's worth warning about postmaster death here
because a backend will emit FATAL message like "terminating connection
due to unexpected postmaster exit" in secure_read() after
pg_promote() returns false.
Regards,
--
Fujii Masao
From | Date | Subject | |
---|---|---|---|
Next Message | Michael Paquier | 2019-09-05 02:07:25 | Re: block-level incremental backup |
Previous Message | Michael Paquier | 2019-09-05 01:41:31 | Re: [HACKERS] [PATCH] pageinspect function to decode infomasks |