Re: Error during hash index scans can cause postgres halt!

Tom Lane wrote:
> "Heikki Linnakangas" <heikki(at)enterprisedb(dot)com> writes:
>> I can reproduce this, with --enable-cassert. It crashes when aborting
>> the transaction, in ReleaseResources_hash. The HashScanList items are
>> allocated in ExecutorState memory context, but that context has already
>> been deleted by the time we get to ReleaseResources_hash.
>
> Ouch. So this has been broken (by me, I think :-() since 8.0. Tells
> you something about how many people use hash indexes :-(

Yeah. Also, this is very hard to trigger without --enable-cassert (or
just CLOBBER_FREED_MEMORY). It's extremely unlikely that something new
is allocated on the piece of memory that was used by an HashScanList
item, during AbortTransaction processing.

ykhuang, were you running an assertion-enabled build as well?

>> Another idea would be to allocate the HashScanList items in a
>> longer-lived memory context. The IndexScanDesc struct pointed to by the
>> HashScanList would still be in ExecutorState context, but that's all
>> right because we don't need to access it in ReleaseResources_hash.
>
> That seems like a winner to me. We can rely on the resource owner
> mechanism to free up individual HashScanList items, so there's no real
> need to keep them in a short-lived context. I'm inclined to just drop
> them into TopMemoryContext. We could make a hash-specific context but
> I'm not convinced it's worth the extra code to do it.

Want me to hack up a patch, or you going to just commit that yourself?

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com