On Tue, Jul 22, 2014 at 6:23 AM, Andres Freund <and...@2ndquadrant.com> wrote:
>> Yes, so nobody can convince serious customers that the current behavior
>> makes real sense.
> I think you're making lots of noise over a trivial log message.
>> Could you please reconsider this?
> No. Just removing a warning isn't the way to solve this. If you want to
> improve things you'll actually need to improve things not just stick
> your head into the sand.

I've studied this area of the code before, and I would actually
proposed to fix this in the opposite way - namely, adopt the logic
currently used for the wraparound case, which drops the temp table,
even if wraparound is not an issue.  The current code seems to be
laboring under the impression that there's some benefit to keeping the
temporary relation around, which, as far as I can see, there isn't.
Now, you could perhaps argue that it's useful for forensics, but that
presumes that the situation where a backend fails to crash without
cleaning up its temporary relations is exciting enough to merit an
investigation, which I believe to be false.
RemoveTempRelationsCallback just does this:


RemoveTempRelations uses:

deleteWhatDependsOn(&object, false);

These are pretty high-level operations, and there are all kinds of
reasons they could fail.  Many of those reasons do indeed involve the
system being messed up in some way, but it's likely that the user will
know about that for other reasons anyway - e.g. if the cleanup fails
because the disk where the files are located has gone read-only at the
operating system level, things are going to go downhill in a hurry.
When the user restarts, they expect recovery - or other automatic
cleanup mechanisms - to put things right.  And normally they do: the
first backend to get the same backend ID as any orphaned temp schema
will clear it out anyway on first use - completely silently - but if
it so happens that a crashed backend had a very high backend ID and
that temp table usage is relatively uncommon, then the user gets log
spam from now and until eternity because of a problem that, in their
mind, is already fixed.  Even better, they will typically be
completely unable to connect the log spam back to the event that
triggered it, and will have no idea how to put it right.

So while I disagree with MauMau's proposed fix, I agree with him that
this error message is useless log spam.

Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:

Reply via email to