Re: Proposal: Conflict log history table for Logical Replication

On Wed, Jan 7, 2026 at 4:26 PM Dilip Kumar <dilipbalaut@gmail.com> wrote:
>
> On Wed, Jan 7, 2026 at 12:12 PM shveta malik <shveta.malik@gmail.com> wrote:
> >
> > Hi Dilip,
> > Please find a few comments on v19-001:
> >
> > 1)
> > We can replace IsConflictLogTable(relid)  with
> > IsConflictClass(reltuple). Wherever we call IsConflictLogTable(), we
> > have already fetched reltuple from pg_class for that relid, we can
> > simply use that to see if it belongs to pg_conflict namespace. That
> > will avoid the need of
> > patch004 as well to 'Introduce a dedicated shared unique index on
> > pg_subscription.subconflictlogrelid'.
> >
> > 2)
> > + if (opts.logdest == CONFLICT_LOG_DEST_TABLE ||
> > + opts.logdest == CONFLICT_LOG_DEST_ALL)
> >
> > We can replace with:
> >
> > IsSet(opts.logdest, CONFLICT_LOG_DEST_TABLE);
> >
> > 3)
> > +-- this should generate an internal table named pg_conflict_$subid$
> > +CREATE SUBSCRIPTION regress_conflict_test1 CONNECTION
> > 'dbname=regress_doesnotexist' PUBLICATION testpub WITH (connect =
> > false, conflict_log_destination = 'table');
> > +
> >
> > I think we shall verify 2 things here:
> > a) Table is created in pg_conflict namespace.
> > b) Table has name pg_conflict_<subid>
> >
> > 4)
> > We can add these 3 simple tests as well:
> > a) Trying to alter and truncate pg_conflict_subid table.
> > b) Trying to create a new table in pg_conflict namespace.
> > c) Moving a table into pg_conflict namespace.
> >
> > ~~
> >
> > Overall, I’m concerned about how users will manage this table as it
> > grows. There is currently no way to purge old data, truncation is
> > disallowed, and the table must be sub-ID–tied, which also prevents
> > users from attaching a different table as a CLT (if needed).
> > Additionally, we do not offer any form of partitioning.
> > Do you think we should provide users with a basic purge mechanism? At
> > the very least, should we allow truncation so users can take a backup
> > and truncate a sub-ID–tied CLT to start afresh? Thoughts?
>
> Yeah that's a valid concern, there should be some way to purge data
> from this table, I think we can allow truncate/delete on this table,
>

+1 for allowing truncate and delete to subscription owners.

--
With Regards,
Amit Kapila.