Обсуждение: logical replication failing after a crash of the publisher

Поиск
Список
Период
Сортировка

logical replication failing after a crash of the publisher

От
Sbob
Дата:

All;


we have logical replication setup for a set of tables that took 4 days to finish the initial sync, the publisher server crashed, after the cloud team recovered the VM we were able to start the db and it came online after an online recovery, however we are now getting these entries (see below) in the log on the publisher and no row changes are showing up on the subscriber


2026-03-24 23:14:09.845 UTC [102580] postgres@claimsprocess STATEMENT:  START_REPLICATION SLOT "prod_sub" LOGICAL 9AE6/4E2F5740 (proto_version '4', origin 'any', publication_names '"prod_pub"')
2026-03-24 23:14:09.959 UTC [102580] postgres@claimsprocess LOG:  could not send data to client: Connection reset by peer
2026-03-24 23:14:09.959 UTC [102580] postgres@claimsprocess CONTEXT:  slot "prod_sub", output plugin "pgoutput", in the change callback, associated LSN 9AE6/4B83E5A8


we tried disabling and then enabling the subscription followed by a REFRESH PUBLICATION but the issue is un changed


Thanks in advance for any help

Re: logical replication failing after a crash of the publisher

От
Laurenz Albe
Дата:
On Tue, 2026-03-24 at 17:27 -0600, Sbob wrote:
> we have logical replication setup for a set of tables that took 4 days to finish
> the initial sync, the publisher server crashed, after the cloud team recovered
> the VM we were able to start the db and it came online after an online recovery,
> however we are now getting these entries (see below) in the log on the publisher
> and no row changes are showing up on the subscriber
>
>  STATEMENT:  START_REPLICATION SLOT "prod_sub" LOGICAL 9AE6/4E2F5740 (proto_version '4', origin 'any',
publication_names'"prod_pub"')  
>  LOG:  could not send data to client: Connection reset by peer
>  CONTEXT:  slot "prod_sub", output plugin "pgoutput", in the change callback, associated LSN 9AE6/4B83E5A8
>
> we tried disabling and then enabling the subscription followed by a REFRESH PUBLICATION
> but the issue is un changed

That seems to be the log from the publisher.  It shows that the subscriber closed the
connection unexpectedly.  You should examine the subscriber's log for the reason.

Yours,
Laurenz Albe