Re: A proposal to force-drop replication slots to make disabling async/sync standbys or logical replication faster in production environments

On Thu, Jun 9, 2022 at 12:11 PM Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> wrote:
>
> On Thu, Jun 9, 2022 at 11:07 AM Bharath Rupireddy
> <bharath(dot)rupireddyforpostgres(at)gmail(dot)com> wrote:
> >
> > Currently postgres doesn't allow dropping a replication slot that's active [1]. This can make certain operations more time-consuming or stuck in production environments. These operations are - disable async/sync standbys and disable logical replication that require the postgres running on standby or the subscriber to go down. If stopping postgres server takes time, the VM or container will have to be killed forcefully which can take a considerable amount of time as there are many layers in between.
> >
>
> Why do you want to drop the slot when the server is going down? Is it
> some temporary replication slot, otherwise, how will you resume
> replication after restarting the server?

The setup is this - primary, bunch of sync standbys, bunch of read
replicas (async standbys), bunch of logical replication subscribers -
now, the user wants to remove any of them for whatever reasons,
typical flow is to first stop the server, if stopping the server takes
time (for instance the standbys or subscribers lag behind the primary
by too much), kill the VM/host server to make the corresponding
replication slots inactive on the primary and then drop the
replication slots. The proposed force-drop function helps speed up
these operations in production environments and it will also be
possible to provide an SLA for these disable operations.

I hope the user case is clear.

Regards,
Bharath Rupireddy.