Horton

Re: Track in pg_replication_slots the reason why slots conflict?

shveta malik <shveta.malik@gmail.com>

View thread
From: shveta malik <shveta.malik@gmail.com>
To: Andres Freund <andres@anarazel.de>
Cc: Michael Paquier <michael@paquier.xyz>, Postgres hackers <pgsql-hackers@lists.postgresql.org>, Bertrand Drouvot <bertranddrouvot.pg@gmail.com>, shveta malik <shveta.malik@gmail.com>
Date: 2023-12-21T10:38:48Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Track conflict_reason in pg_replication_slots.

On Thu, Dec 21, 2023 at 3:10 PM Andres Freund <andres@anarazel.de> wrote:
>
> Hi,
>
> On 2023-12-21 09:21:04 +0900, Michael Paquier wrote:
> > While listening at Bertrand's talk about logical decoding on standbys
> > last week at Prague, I got surprised by the fact that we do not
> > reflect in the catalogs the reason why a conflict happened for a slot.
> > There are three of them depending on ReplicationSlotInvalidationCause:
> > - WAL removed.
> > - Invalid horizon.
> > - Insufficient WAL level.
>
> It should be extremely rare to hit any of these other than "WAL removed", so
> I'm not sure it's worth adding interface complexity to show them.
>
>
> > ReplicationSlotCtl holds this information, so couldn't it be useful
> > for monitoring purposes to know why a slot got invalidated and add a
> > column to pg_get_replication_slots()?  This could just be an extra
> > text conflicting_reason, defaulting to NULL when there's nothing to
> > see.
>
> Extra columns aren't free from a usability perspective. IFF we do something, I
> think it should be a single column with a cause.

Thanks for the feedback. But do you mean that we replace existing
'conflicting' column with 'cause' in both the function and view
(pg_get_replication_slots() and pg_replication_slots)?  Or do you mean
that we expose 'cause' from pg_get_replication_slots() and use that to
display 'conflicting' in pg_replication_slots view?

And if we plan to return/display cause from either function or view,
then shall it be enum 'ReplicationSlotInvalidationCause' or
description/text corresponding to enum?

 In the other feature being discussed "Synchronize slots from primary
to standby" [1] , there is a requirement to replicate invalidation
cause of slot from the primary to standby and thus it is needed in
enum form there. And thus there was a suggestion earlier to have the
function return enum-value and let the view display it as
text/description to the user.  So kindly let us know your thoughts.

[1] - https://www.postgresql.org/message-id/514f6f2f-6833-4539-39f1-96cd1e011f23@enterprisedb.com

thanks
Shveta