Thread: [HACKERS] pg_subscription_rel entry can be updated concurrently
Hi, I often get an error "ERROR: tuple concurrently updated" when changing subscription state(ALTER SUBSCRIPTION or DROP SUBSCRIPTION). The cause of this error is that table sync worker and apply worker can try to update the same tuple in pg_subscription_rel. Especially it often happens when we do initial copy for many tables and change it during executing. I think that touching the same tuple by two worker processes happens when aborting replication for a table or a subscription, so it would be the same result as when the worker ends up with an error. But I think since it's not an appropriate behavior we should deal with it. Any thoughts? Regards, -- Masahiko Sawada NIPPON TELEGRAPH AND TELEPHONE CORPORATION NTT Open Source Software Center
On Tue, Jun 13, 2017 at 9:52 AM, Masahiko Sawada <sawada.mshk@gmail.com> wrote: > Hi, > > I often get an error "ERROR: tuple concurrently updated" when > changing subscription state(ALTER SUBSCRIPTION or DROP SUBSCRIPTION). > The cause of this error is that table sync worker and apply worker can > try to update the same tuple in pg_subscription_rel. Especially it > often happens when we do initial copy for many tables and change it > during executing. > > I think that touching the same tuple by two worker processes happens > when aborting replication for a table or a subscription, so it would > be the same result as when the worker ends up with an error. But I > think since it's not an appropriate behavior we should deal with it. > Any thoughts? Yes, this error is normally not be something that users should see. So there is something weird around the locking of the parent objects or in the way the catalog is updated. -- Michael
On 13/06/17 02:52, Masahiko Sawada wrote: > Hi, > > I often get an error "ERROR: tuple concurrently updated" when > changing subscription state(ALTER SUBSCRIPTION or DROP SUBSCRIPTION). > The cause of this error is that table sync worker and apply worker can > try to update the same tuple in pg_subscription_rel. Especially it > often happens when we do initial copy for many tables and change it > during executing. > > I think that touching the same tuple by two worker processes happens > when aborting replication for a table or a subscription, so it would > be the same result as when the worker ends up with an error. But I > think since it's not an appropriate behavior we should deal with it. > Any thoughts? > This has been already reported by tushar in different thread and it's still on my list to fix. -- Petr Jelinek http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training & Services
On Tue, Jun 13, 2017 at 2:55 PM, Petr Jelinek <petr.jelinek@2ndquadrant.com> wrote: > On 13/06/17 02:52, Masahiko Sawada wrote: >> Hi, >> >> I often get an error "ERROR: tuple concurrently updated" when >> changing subscription state(ALTER SUBSCRIPTION or DROP SUBSCRIPTION). >> The cause of this error is that table sync worker and apply worker can >> try to update the same tuple in pg_subscription_rel. Especially it >> often happens when we do initial copy for many tables and change it >> during executing. >> >> I think that touching the same tuple by two worker processes happens >> when aborting replication for a table or a subscription, so it would >> be the same result as when the worker ends up with an error. But I >> think since it's not an appropriate behavior we should deal with it. >> Any thoughts? >> > > This has been already reported by tushar in different thread and it's > still on my list to fix. > Okay, I see. I added it to the open item list. Regards, -- Masahiko Sawada NIPPON TELEGRAPH AND TELEPHONE CORPORATION NTT Open Source Software Center