Re: Re: pg15 beta2 bug:cause by logcial replation

Obviously in the same pg server pg15 beta2, because my
subscriber creation use "create_slot='false'", another clue is i use "\c", all the test on the same cluster.

396934406(at)qq(dot)com

From: Tom Lane
Date: 2022-07-01 13:16
To: houzj(dot)fnst(at)fujitsu(dot)com
CC: Amit Kapila; 396934406(at)qq(dot)com; pgsql-bugs
Subject: Re: pg15 beta2 bug:cause by logcial replation
"houzj(dot)fnst(at)fujitsu(dot)com" <houzj(dot)fnst(at)fujitsu(dot)com> writes:
> On Friday, July 1, 2022 11:35 AM 396934406(at)qq(dot)com <396934406(at)qq(dot)com> wrote:
>> ERROR: could not receive list of replicated tables from the publisher: ERROR: column t.attnames does not exist
>> LINE 2: , t.attnames

> Just to confirm, did you use the PG15 beta1 or earlier PG15 version
> as the publisher and use PG15 beta2 as the subscriber ?

It looks to me like this was an attempt to replicate from 15beta1 to
15beta2, which would fail because commit fd0b9dceb imagines that any
server reporting version >= 15 will have pg_publication_tables.attnames,
even though that column was not there in beta1.

I bitched about the ill-timing of that schema change at the time
but was outvoted. At this point I think it's water over the dam.
We could perhaps change that test in fetch_table_list(), but it'd
require access to the source server's catversion, which I don't think
is readily available. In any case, by the time we could provide a fix
it'd be useless to testers of beta1; they'd have updated already by
some other method.

regards, tom lane