Re: Infinite loop in GetMultiXactIdMembers() on 9.3.5 (involves pg_upgrade)

On Tue, 27 Jan 2015 19:56:43 -0800
Peter Geoghegan <pg@heroku.com> wrote:

> I've since found BUG #11264, which seems to be the same issue...don't
> know how I missed that one before. I see that Alvaro committed
> something that at least addresses the non-interruptibility issue
> directly, with a call to CHECK_FOR_INTERRUPTS() (commit d45e8dc52).
> But I don't see any follow up addressing the bug itself, which I'd
> hoped for. Did the discussion of BUG #11264 simply trail off without
> an adequate resolution, as it appeared to?

Sadly yes. I have hit the same problem and expect others will too. Alvaro
suggested updating datminmxid to the oldest relminmxid will prevent
this. I have written a script to do this, but not quite finished it as
I was unsure what to do with the cases where nextMultiXactId in
pg_controldata is < min(relminmxid) without looking at it more closely.

I have attached the script (work in progress) in hopes it is useful
or that someone can clarify how to calculate the correct datminmxid.

-dg



--
David Gould              510 282 0869         daveg@sonic.net
If simplicity worked, the world would be overrun with insects.