Jeff Frost <jeff@pgexperts.com> writes:
> On Jun 13, 2013, at 4:50 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> ... So one theory about this would be that those processes
>> aren't absorbing the GUC updates, perhaps because the SIGHUP signals the
>> postmaster should be sending them are getting lost.
> Interestingly, it will often pick them up if you wait a few seconds and send it another reload.
Hmm, that definitely lends some credence to the lost-signal theory,
since another reload would cause the postmaster to again signal all
its children, and this time the signal might go through.
But I still have no idea how we might debug further. You could possibly
try something like strace'ing the processes, but it seems fairly likely
that the Heisenberg principle would apply if you did.
regards, tom lane