Re: Logical decoding on standby

From: Craig Ringer <craig(at)2ndquadrant(dot)com>
To: Andres Freund <andres(at)anarazel(dot)de>
Cc: Simon Riggs <simon(dot)riggs(at)2ndquadrant(dot)com>, Thom Brown <thom(at)linux(dot)com>, Michael Paquier <michael(dot)paquier(at)gmail(dot)com>, Petr Jelinek <petr(dot)jelinek(at)2ndquadrant(dot)com>, PostgreSQL Hackers <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: Logical decoding on standby
Date: 2017-03-22 08:53:14
Message-ID: CAMsr+YF4UMAoMO86hWGPhHgfdQk8PB9fPCSKFaVM0WcN+010CA@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox
Thread:
Lists: pgsql-hackers

On 22 March 2017 at 10:51, Craig Ringer <craig(at)2ndquadrant(dot)com> wrote:
> Hi all
>
> Updated timeline following patch attached.
>
> There's a change in read_local_xlog_page to ensure we maintain
> ThisTimeLineID properly, otherwise it's just comment changes.

OK, so we're looking OK with the TL following.

I'm splitting up the rest of the decoding on standby patch set with
the goal of getting minimal functionality for creating and managing
slots on standbys in, so we can maintain slots on standbys and use
them when the standby is promoted to master.

The first, to send catalog_xmin separately to the global xmin on
hot_standby_feedback and store it in the upstream physical slot's
catalog_xmin, is attached.

These are extracted directly from the logical decoding on standby
patch, with comments by Petr and Andres made re the relevant code
addressed.

I will next be working on a bare-minimum facility for creating and
advancing logical slots on a replica without support for buffering
changes, creating historic snapshots or invoking output plugin. The
slots will become usable after the replica is promoted. They'll track
their own restart_lsn, etc, and will keep track of xids so they can
manage their catalog_xmin, so there'll be no need for dodgy slot
syncing from the master, but they'll avoid most of the complex and
messy bits. The application will be expected to make sure a slot on
the master exists and is advanced before the corresponding slot on the
replica to protect required catalogs.

Then if there's agreement that it's the right way forward I can add
the catalog_xmin xlog'ing stuff as the next patch.

--
Craig Ringer http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services

Attachment Content-Type Size
0001-Allow-GetOldestXmin-to-ignore-replication-slot-xmin.patch text/x-patch 9.5 KB
0002-Report-catalog_xmin-separately-to-xmin-in-hot-standb.patch text/x-patch 14.4 KB

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Ideriha, Takeshi 2017-03-22 08:57:15 Re: [WIP] RE: DECLARE STATEMENT setting up a connection in ECPG
Previous Message Amit Khandekar 2017-03-22 08:49:13 Re: Parallel Append implementation