Re: wait events for disk I/O

On Wed, Mar 8, 2017 at 8:23 AM, Robert Haas <robertmhaas(at)gmail(dot)com> wrote:

> On Tue, Mar 7, 2017 at 9:32 PM, Amit Kapila <amit(dot)kapila16(at)gmail(dot)com>
> wrote:
> > On Tue, Mar 7, 2017 at 9:16 PM, Robert Haas <robertmhaas(at)gmail(dot)com>
> wrote:
> >> On Mon, Mar 6, 2017 at 9:09 PM, Amit Kapila <amit(dot)kapila16(at)gmail(dot)com>
> wrote:
> >>> Sure, if you think both Writes and Reads at OS level can have some
> >>> chance of blocking in obscure cases, then we should add a wait event
> >>> for them.
> >>
> >> I think writes have a chance of blocking in cases even in cases that
> >> are not very obscure at all.
> >
> > Point taken for writes, but I think in general we should have some
> > criteria based on which we can decide whether to have a wait event for
> > a particular call. It should not happen that we have tons of wait
> > events and out of which, only a few are helpful in most of the cases
> > in real-world scenarios.
>
> Well, the problem is that if you pick and choose which wait events to
> add based on what you think will be common, you're actually kind of
> hosing yourself. Because now when something uncommon happens, suddenly
> you don't get any wait event data and you can't tell what's happening.
> I think the number of new wait events added by Rushabh's patch is
> wholly reasonable. Yeah, some of those are going to be a lot more
> common than others, but so what? We add wait events so that we can
> find out what's going on. I don't want to sometimes know when a
> backend is blocked on an I/O. I want to ALWAYS know.
>
>
Yes, I agree with Robert. Knowing what we want and what we don't
want is difficult to judge. Something which we might think its not useful
information, and later of end up into situation where we re-think about
adding those missing stuff is not good. Having more information about
the system, specially for monitoring purpose is always good.

I am attaching another version of the patch, as I found stupid mistake
in the earlier version of patch, where I missed to initialize initial value
to
WaitEventIO enum. Also earlier version was not getting cleanly apply on
the current version of sources.

--
Rushabh Lathia
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company