Re: pgsql: Add support for multivariate MCV lists

On Wed, Mar 27, 2019 at 09:33:53PM +0100, Tomas Vondra wrote:
>On Wed, Mar 27, 2019 at 09:13:46PM +0100, Tomas Vondra wrote:
>>On Wed, Mar 27, 2019 at 12:32:34PM -0700, Peter Geoghegan wrote:
>>>On Wed, Mar 27, 2019 at 12:04 PM Andres Freund <andres@anarazel.de> wrote:
>>>>Congrats, this was long in the making!
>>>
>>>+1
>>>
>>
>>Thanks! I guess it's a Gordie Howe of our patches.
>>
>>>Buildfarm member aye-aye has a problem with this patch, though:
>>>
>>>https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=aye-aye&dt=2019-03-27%2019%3A07%3A53
>>>
>>
>>Damn. And I see locust, prion and dory also seem to have an issue.
>>locust is also a power machine, I wonder if that's somehow related.
>>
>>I'll take a look.
>>
>
>There seems to be a bug in statext_mcv_load - the deserialize happens
>after releasing syscache. That's probably why prion is failing, because
>it's running with -DRELCACHE_FORCE_RELEASE -DCATCACHE_FORCE_RELEASE. Not
>sure about aye-aye.
>

FWIW I can reproduce the issue by simply using the same CFLAGS, and it
turns out the MCV deserialization is a bit borked (the value may easily
reference memory that already went away). I have a patch that resolves
that for me, and I assume it'll address the other failures related to
this commit too - for example the segfault on gharial).

It's a bit too late for pushing emergency fixes over here, so I'll do
more testing tomorrow and then push.


regards

-- 
Tomas Vondra                  http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services