Re: hot_standby_feedback parameter doesn't work

Hi,

On 2018-11-13 10:28:20 +0700, Andrey Zhidenkov wrote:
> We have a few database clusters (1 master, 2 hot standbys) with
> hot_standby_feedback setting on. But I am constantly seeing query
> conflicts on standbys because dead rows on the masters are deleted by
> VACUUM. Usually I notice it shortly after VACUUM ends, because the
> replication lag on replicas is starting to increase.
> 
> vacuum_defer_cleanup_age = 0. PostgreSQL version is 9.5.13. Cluster
> load is about 1500 transactions per second.

What's the precise conflict you get? I'd assume it's because of
exclusive locks, not row version conflicts. Is that right?

If so, that'd likely be because vacuum gets a chance to truncate the
relations at the end of vacuum.

Before theorizing further, it'd be good to get confirmation that that's
the right theory.

Greetings,

Andres Freund