Horton

Re: 001_rep_changes.pl stalls

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Cc: noah@leadboat.com, masao.fujii@oss.nttdata.com, pgsql-hackers@postgresql.org
Date: 2020-04-17T13:59:25Z
Lists: pgsql-hackers 
Kyotaro Horiguchi <horikyota.ntt@gmail.com> writes:
> At Thu, 16 Apr 2020 22:41:46 -0700, Noah Misch <noah@leadboat.com> wrote in 
>> I'm favoring (1).  Other preferences?

> Starting from the current shape, I think 1 is preferable, since that
> waiting logic is no longer shared between logical and physical
> replications.  But I'm not sure I like calling WalSndWaitForWal()
> (maybe with previous location + 1?), because the function seems to do
> too-much.

I'm far from an expert on this code, but it does look like there's
a lot of stuff in WalSndWaitForWal that is specific to logical rep,
so I'm not sure that (1) is workable.  At the very least there'd
have to be a bunch more conditionals in that function than there are
now.  In the end, a separate copy for physical rep might be better.

(BTW, I think this code is in desperate need of a renaming
campaign to make it clearer which functions are for logical rep,
physical rep, or both.)

			regards, tom lane

Commits

  1. In caught-up logical walsender, sleep only in WalSndWaitForWal().

  2. Revert "When WalSndCaughtUp, sleep only in WalSndWaitForWal()."

  3. Add a wait_for_catchup() before immediate stop of a test master.

  4. When WalSndCaughtUp, sleep only in WalSndWaitForWal().