Re: Preventing hangups in bgworker start/stop during DB shutdown

On Fri, 25 Dec 2020 at 06:07, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:

> I wrote:
> > Bharath Rupireddy <bharath(dot)rupireddyforpostgres(at)gmail(dot)com> writes:
> >> 4) IIUC, in the patch we mark slot->terminate = true only for
> >> BGW_NEVER_RESTART kind bg workers, what happens if a bg worker has
> >> bgw_restart_time seconds and don't we hit the hanging issue(that we
> >> are trying to solve here) for those bg workers?
>
> > The hang problem is not with the worker itself, it's with anything
> > that might be waiting around for the worker to finish. It doesn't
> > seem to me to make a whole lot of sense to wait for a restartable
> > worker; what would that mean?
>
> Upon further looking around, I noted that autoprewarm's
> autoprewarm_start_worker() function does that, so we can't really
> dismiss it.
>
> However, what we can do instead is to change the condition to be
> "cancel pending bgworker requests if there is a waiting process".
> Almost all of the time, that means it's a dynamic bgworker with
> BGW_NEVER_RESTART, so there's no difference. In the exceptional
> cases like autoprewarm_start_worker, this would result in removing
> a bgworker registration record for a restartable worker ... but
> since we're shutting down, that record would have no effect before
> the postmaster exits, anyway. I think we can live with that, at
> least till such time as somebody redesigns this in a cleaner way.
>
> I pushed a fix along those lines.
>
>
Thanks for the change.

Cleanups like this in the BGW API definitely make life easier.