Horton

Re: Duplicated LSN in ReorderBuffer

Thomas Munro <thomas.munro@gmail.com>

View thread View on postgr.es
From: Thomas Munro <thomas.munro@gmail.com>
To: Ildar Musin <ildar@adjust.com>
Cc: PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Date: 2019-07-08T00:59:26Z
Lists: pgsql-hackers 
On Wed, Jun 26, 2019 at 2:46 AM Ildar Musin <ildar@adjust.com> wrote:
> Attached is a simple patch that uses subxid instead of top-level xid
> in ReorderBufferAddNewTupleCids() call. It seems to fix the bug, but
> i'm not sure that this is a valid change. Can someone please verify it
> or maybe suggest a better solution for the issue?

Hello Ildar,

I hope someone more familiar with this code than me can comment, but
while going through the Commitfest CI results I saw this segfault with
your patch:

https://travis-ci.org/postgresql-cfbot/postgresql/builds/555184304

At a glance, HistoricSnapshotGetTupleCids() returned NULL in
HeapTupleSatisfiesHistoricMVCC(), so ResolveCminCmaxDuringDecoding()
blew up.

-- 
Thomas Munro
https://enterprisedb.com

Commits

  1. logical decoding: process ASSIGNMENT during snapshot build

  2. Ability to advance replication slots