Re: Initial COPY of Logical Replication is too slow

On Mon, Mar 23, 2026 at 5:54 PM Bharath Rupireddy
<bharath(dot)rupireddyforpostgres(at)gmail(dot)com> wrote:
>
> Hi,
>
> On Wed, Mar 18, 2026 at 4:29 PM Masahiko Sawada <sawada(dot)mshk(at)gmail(dot)com> wrote:
> >
> > I've attached the patch to implement this idea. The patch still
> > introduces a new function but it overloads
> > pg_get_publication_tables(). We might be able to handle different
> > input (array or text) in pg_get_publication_tables() better, but it's
> > enough for discussion at least.
>
> Overall, the intent of this patch looks good to me. It avoids the cost
> of the table sync worker querying all the pg_publication_rel tables to
> filter them out later in the join.
>
> I quickly reviewed the patch and here are some comments:

Thank you for reviewing the patch!

>
> 1/ Typo: s/pg_get_publication_table_info/pg_get_publication_tables

Fixed.

>
> 2/ I think it's good to have some quick numbers on how the query
> latency looks for pre-V19 and the new one that the table sync worker
> executes on the publisher, say, with 100, 1000, and 10000 tables at
> least.

You can refer to the performance test results that I previously
shared[1]. The patch I used was somewhat different from the current
patch but the performance trend should be similar as the both are
using the same approach.

>
> 3/ + Assert(OidIsValid(target_relid));
>
> Why not error out (by treating it as function input parameter
> validation) when target_relid is invalid because asserts go unnoticed
> on production systems?

Agreed. It would return no row if the specified relid is invalid or
there is no corresponding table.

I'll share the updated patch soon.

Regards,

[1] https://www.postgresql.org/message-id/CAD21AoDQM62GOtaTzD_CVMSsFhv6o9c0Au1dSM1QuxeKFkWAKw%40mail.gmail.com

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