On Wed, Mar 25, 2020 at 3:19 PM Andres Freund <andres@anarazel.de> wrote:
>
> Hi,
>
> On 2019-09-18 13:51:00 -0400, James Coleman wrote:
> > In my experience it's not immediately obvious (even after reading the
> > documentation) the implications of how concurrent index builds manage
> > transactions with respect to multiple concurrent index builds in
> > flight at the same time.
> >
> > Specifically, as I understand multiple concurrent index builds running
> > at the same time will all return at the same time as the longest
> > running one.
> >
> > I've attached a small patch to call this caveat out specifically in
> > the documentation. I think the description in the patch is accurate,
> > but please let me know if there's some intricacies around how the
> > various stages might change the results.
> >
> > James Coleman
>
> I'd much rather see effort spent fixing this issue as far as it relates
> to concurrent CICs. For the snapshot waits we can add a procarray flag
> (alongside PROCARRAY_VACUUM_FLAG) indicating that the backend is
> doing. Which WaitForOlderSnapshots() can then use to ignore those CICs,
> which is safe, because those transactions definitely don't insert into
> relations targeted by CIC. The change to WaitForOlderSnapshots() would
> just be to pass the new flag to GetCurrentVirtualXIDs, I think.
Alvaro: I think you had some ideas on this too; any chance you've know
of a patch that anyone's got cooking?
Andres: If we got this fixed in current PG would you be opposed to
documenting the caveat in previous versions?
Thanks,
James