Horton

Re: [HACKERS] Issues with logical replication

Alvaro Herrera <alvherre@alvh.no-ip.org>

View thread View on postgr.es
From: Alvaro Herrera <alvherre@alvh.no-ip.org>
To: Stas Kelvich <s.kelvich@postgrespro.ru>
Cc: Petr Jelinek <petr.jelinek@2ndquadrant.com>, Andres Freund <andres@anarazel.de>, Robert Haas <robertmhaas@gmail.com>, Konstantin Knizhnik <k.knizhnik@postgrespro.ru>, PostgreSQL Hackers <pgsql-hackers@postgresql.org>, Peter Eisentraut <peter.eisentraut@2ndquadrant.com>
Date: 2018-01-03T17:47:10Z
Lists: pgsql-hackers

Attachments

Stas Kelvich wrote:

> Seems that having busy loop is the best idea out of several discussed.
> 
> I thought about small sleep at the bottom of that loop if we reached topmost
> transaction, but taking into account low probability of that event may be
> it is faster to do just busy wait.

In other locations where we do similar things we have 1ms sleeps.  I
agree with the need for a comment here.

Proposed patch attached.  I tried your reload.pgb test case in 9.4
(after changing pgoutput to test_decoding and removing the 3rd arg to a
function call) and the crash takes about 3 seconds without patch in my
machine.  No crash with this patch.

Will push this shortly after lunch.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

Commits

  1. Make XactLockTableWait work for transactions that are not yet self-locked