On Sun, Mar 19, 2017 at 5:14 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Couple of thoughts on this patch ---
Thanks!
> 1. Shouldn't WaitExceedsMaxStandbyDelay's CHECK_FOR_INTERRUPTS be moved to
> after the WaitLatch call? Not much point in being woken immediately by
> an interrupt if you're not going to respond.
>
> 2. Is it OK to ResetLatch here? If the only possible latch event in this
> process is interrupt requests, then I think WaitLatch, then ResetLatch,
> then CHECK_FOR_INTERRUPTS is OK; but otherwise it seems like you risk
> discarding events that need to be serviced later.
Right, I have switched to WaitLatch(), ResetLatch() and then
CHECK_FOR_INTERRUPTS().
> 3. In the same vein, if we're going to check WL_POSTMASTER_DEATH, should
> there be a test for that and immediate exit(1) here?
OK, if the postmaster has died, there is not much recovery conflict
needed anyway.
> 4. I'd be inclined to increase the sleep interval only if we did time out,
> not if we were awakened by some other event.
OK, that makes sense.
> 5. The comment about maximum sleep length needs some work. At first
> glance you might think that without the motivation of preventing long
> uninterruptible sleeps, we might as well allow the sleep length to grow
> well past 1s. I think that'd be bad, because we want to wake up
> reasonably soon after the xact(s) we're waiting for commit. But neither
> the original text nor the proposed replacement mention this.
OK, I did some work on this comment.
What do you think about the updated version attached?
--
Michael
--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers