Re: pgsql: Add wait event for fsync of WAL segments

On 2018-Jul-03, Michael Paquier wrote:

> On Mon, Jul 02, 2018 at 12:23:35PM -0400, Alvaro Herrera wrote:
> > I wonder if we should backpatch this one all the way to pg10. I don't
> > see no reason not to.
>
> ABI breakage (if that's the correct wording?). Simply cherry-picking
> the patch from master to back-branches would cause extensions and
> plugins already compiled with those versions to be confused by the
> ordering of the enum WaitEventIO. Well, one simple solution is to
> simply put the new event purposefully at the bottom of the list. If
> that sounds right, I could do that on back-branches

Are the numerical values actually exposed to the world? I thought the
only way to this info was through the system views, which surely expose
the names, not the numbers.

Actually, comment on pgstat_report_wait_start talks about WaitClass as
if it were a thing, which it seems not to be. There is a comment "Wait
Classes" but that term is not used anywhere else.

If reading wait events is actually possible, then it seems easy to
backpatch this in pg10 by putting the new value at the end of the
relevant enum, yeah.

> but I'd rather let HEAD on it current state with the event set
> correctly ordered.

Sure.

--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services