On 2019-Jul-24, Kyotaro Horiguchi wrote:
> > Please find the proposed patch for review. I will attach it to
> > commitfest as well
>
> Pacemaker suffers the same thing. We suggest our customers that "start
> server alone to perform recovery then start pacemaker if it is
> expected to take a long time for recovery so that reaches time out".
>
> I don't think it is good think to let status SERVICE_RUNNING although
> it actually is not (yet). I think the right direction here is that, if
> pg_ctl returns by timeout, pgwin32_ServiceMain kills the starting
> server then report something like "timedout and server was stopped,
> please make sure the server not to take a long time to perform
> recovery.".
I'm not sure that's a great reaction; it makes total recovery time
even longer. How would the user ensure that recovery takes a shorter
time? We'd be forcing them to start the service over and over, until
recovery completes.
Can't we have pg_ctl just continue to wait indefinitely? So we'd set
SERVICE_START_PENDING when wait_for_postmaster is out of patience, then
loop again -- until recovery completes. Exiting pg_ctl on timeout seems
reasonable for interactive use, but maybe for service use it's not
reasonable.
--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services