Re: Proposal: Conflict log history table for Logical Replication

On Fri, Jul 3, 2026 at 12:51 PM shveta malik <shveta(dot)malik(at)gmail(dot)com> wrote:
>
> On Thu, Jul 2, 2026 at 3:33 PM shveta malik <shveta(dot)malik(at)gmail(dot)com> wrote:
> >
> > On Thu, Jul 2, 2026 at 1:39 PM Dilip Kumar <dilipbalaut(at)gmail(dot)com> wrote:
> > >
> > >
> > > I rebased the remaining patches on top of HEAD. So far, I have run
> > > pgindent and completed the doc merge for 0001. The 0002 and 0003 are
> > > just rebased.
> > >
> >
>
> I am focusing on 001 alone.
> I was verifying conflict insertion for all conflict_types in regular
> worker flow (not parallel), I have just one comment there:
>
> For insert_exists conflict, both replica_identity (JSON) and
> replica_identity_full (bool) are NULL. Is it conscious decision to
> keep 'replica_identity_full' as NULL or shall we make this field as
> 'bool NOT NULL DEFAULT false'. As per doc at [1], NULL value for
> boolean represents 'unknown'
>

Displaying 'replica_identity_full' as false means replica_identity
index is used and same is expected to be displayed as well.

> If it is intentionaly kept NULL, do you think we should update below
> doc to explain that it can be NULL for the cases where
> replica_identity is not applicable.
>
> replica_identity_full boolean I
> indicates whether replica_identity represents a full tuple (true) or
> key values of a replica identity index (false).
>

I agree with the doc update for this case.

I have one comment on 63-0001.
*
In 63-0001, it seems only a minor title change in docs:
+ <sect2 id="logical-replication-conflict-logging">
+ <title>Conflict logging</title>
leads to the patch showing such a big change in
logical-replication.sgml. Is that correct? If so, I suggest moving the
logical-replication.sgml to a later doc-only patch as that will help
in focusing the core code change. I understand that we need to update
docs with the new section for logging conflict information in tables
and that is important but we can handle it separately as it would
require more thoughts on how to rearrange the current docs.

--
With Regards,
Amit Kapila.