Re: Fix slot synchronization with two_phase decoding enabled - Mailing list pgsql-hackers

From Masahiko Sawada
Subject Re: Fix slot synchronization with two_phase decoding enabled
Date
Msg-id CAD21AoCvCw42m6UxdFA=3FgUaz_Eb+LvaYFNTbagMQJc1tHphQ@mail.gmail.com
Whole thread Raw
In response to Fix slot synchronization with two_phase decoding enabled  ("Zhijie Hou (Fujitsu)" <houzj.fnst@fujitsu.com>)
Responses Re: Fix slot synchronization with two_phase decoding enabled
Re: Fix slot synchronization with two_phase decoding enabled
List pgsql-hackers
On Sun, Jun 1, 2025 at 10:25 PM Amit Kapila <amit.kapila16@gmail.com> wrote:
>
>
> Yet another idea is to dump the subscription with two_phase = on and
> failover = false. We should do this when both options are 'true'
> during the dump.  As we are documenting that we always dump
> createsubscription with connect as false and let users take care (see
> [1] (When dumping logical replication subscriptions ...)), a similar
> reasoning could be given for the failover flag.

It probably would work for subscriptions but what about logical slots?
If we restore two_phase=on and failover=false, the user would have to
enable the failover via ALTER_REPLICATION_SLOT. I think we should
avoid asking users to execute the replication commands directly.

> The one more combination to consider is when someone takes a dump of
> an older version and loads it into a newer version. For example, where
> users dump from 17.5 and then restore in a newer version, say 17.6
> (which has our fix), the restore will fail due to newer restrictions
> added by this patch. Do we need to do anything about it?

A valid concern. Implementing this change could potentially render
dumps created prior to version 17.5 incompatible with version 17.6 or
later, which seems a significant backwards incompatibility to me. Do
we have any precedence of such incompatibility?

Regards,

--
Masahiko Sawada
Amazon Web Services: https://aws.amazon.com



pgsql-hackers by date:

Previous
From: Peter Geoghegan
Date:
Subject: Re: strange perf regression with data checksums
Next
From: Jim Nasby
Date:
Subject: Re: [WIP]Vertical Clustered Index (columnar store extension) - take2