Re: Re: check failure with -DRELCACHE_FORCE_RELEASE -DCLOBBER_FREED_MEMORY

Andrew Dunstan <andrew(dot)dunstan(at)2ndquadrant(dot)com> writes:
> On 03/03/2017 02:24 PM, Andrew Dunstan wrote:
>> I have been setting up a buildfarm member with -DRELCACHE_FORCE_RELEASE
>> -DCLOBBER_FREED_MEMORY, settings which Alvaro suggested to me.I got core
>> dumps with these stack traces. The platform is Amazon Linux.

> I have replicated this on a couple of other platforms (Fedora, FreeBSD)
> and back to 9.5. The same failure doesn't happen with buildfarm runs on
> earlier branches, although possibly they don't have the same set of tests.

well, the problem in rebuild_relation() seems pretty blatant:

/* Close relcache entry, but keep lock until transaction commit */
heap_close(OldHeap, NoLock);

/* Create the transient table that will receive the re-ordered data */
OIDNewHeap = make_new_heap(tableOid, tableSpace,
OldHeap->rd_rel->relpersistence,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AccessExclusiveLock);

There are two such references after the heap_close. I don't know that
those are the only bugs, but this reference is certainly the proximate
cause of the crash I'm seeing.

Will push a fix in a little bit.

regards, tom lane