Re: Is it worth adding ReplicationSlot active_pid to ReplicationSlotPersistentData?

On Fri, Dec 3, 2021 at 7:39 PM Bharath Rupireddy
<bharath(dot)rupireddyforpostgres(at)gmail(dot)com> wrote:
>
> On Wed, Dec 1, 2021 at 9:50 PM Alvaro Herrera <alvherre(at)alvh(dot)no-ip(dot)org> wrote:
> >
> > On 2021-Dec-01, Bharath Rupireddy wrote:
> >
> > > The active_pid of ReplicationSlot structure, which tells whether a
> > > replication slot is active or inactive, isn't persisted to the disk
> > > i.e has no entry in ReplicationSlotPersistentData structure. Isn't it
> > > better if we add that info to ReplicationSlotPersistentData structure
> > > and persist to the disk? This will help to know what were the inactive
> > > replication slots in case the server goes down or crashes for some
> > > reason. Currently, we don't have a way to interpret the replication
> > > slot info in the disk but there's a patch for pg_replslotdata tool at
> > > [1]. This way, one can figure out the reasons for the server
> > > down/crash and figure out which replication slots to remove to bring
> > > the server up and running without touching the other replication
> > > slots.
> >
> > I think the PIDs are log-worthy for sure, but it's not clear to me that
> > it is desirable to write them to the persistent state file. In case of
> > crashes, the log should serve just fine to aid root cause investigation
> > -- in fact even better than the persistent file, where the data would be
> > lost as soon as the next client acquires that slot.
>
> Thanks. +1 to log a message at LOG level whenever a replication slot
> becomes active (gets assigned a valid pid to active_pid) and becomes
> inactive(gets assigned 0 to active_pid). Having said that, isn't it
> also helpful if we write a bool (1 byte character) whenever the slot
> becomes active and inactive to the disk?

Here's the patch that adds a LOG message whenever a replication slot
becomes active and inactive. These logs will be extremely useful on
production servers to debug and analyze inactive replication slot
issues.

Thoughts?

Regards,
Bharath Rupireddy.