Horton

RE: logical replication empty transactions

Zhijie Hou (Fujitsu) <houzj.fnst@fujitsu.com>

View thread View on postgr.es
From: "houzj.fnst@fujitsu.com" <houzj.fnst@fujitsu.com>
To: Amit Kapila <amit.kapila16@gmail.com>
Cc: Ajin Cherian <itsajin@gmail.com>, "shiy.fnst@fujitsu.com" <shiy.fnst@fujitsu.com>, Peter Smith <smithpb2250@gmail.com>, "osumi.takamichi@fujitsu.com" <osumi.takamichi@fujitsu.com>, vignesh C <vignesh21@gmail.com>, Michael Paquier <michael@paquier.xyz>, Rahila Syed <rahila.syed@2ndquadrant.com>, Dilip Kumar <dilipbalaut@gmail.com>, Jeff Janes <jeff.janes@gmail.com>, pgsql-hackers <pgsql-hackers@postgresql.org>, Craig Ringer <craig@2ndquadrant.com>
Date: 2022-03-28T12:21:42Z
Lists: pgsql-hackers

Attachments

On Monday, March 28, 2022 3:08 PM Amit Kapila <amit.kapila16@gmail.com> wrote:
> 
> On Fri, Mar 25, 2022 at 12:50 PM houzj.fnst@fujitsu.com
> <houzj.fnst@fujitsu.com> wrote:
> >
> > Attach the new version patch with this change.
> >
> 
> Few comments:

Thanks for the comments.

> =================
> 1. I think we can move the keep_alive check after the tracklag record
> check to keep it consistent with another patch [1].

Changed.

> 2. Add the comment about the new parameter skipped_xact atop
> WalSndUpdateProgress.

Added.

> 3. I think we need to call pq_flush_if_writable after sending a
> keepalive message to avoid delaying sync transactions.

Agreed.
If we don’t flush the data, we might flush the keepalive later than before. And
we could get the reply later as well and then the release of syncwait could be
delayed.

Attach the new version patch which addressed the above comments.
The patch also adds a loop after the newly added keepalive message
to make sure the message is actually flushed to the client like what
did in WalSndWriteData.

Best regards,
Hou zj

Commits

  1. Skip empty transactions for logical replication.

  2. Add decoding of sequences to built-in replication

  3. Allow specifying row filters for logical replication of tables.