Re: why there is not VACUUM FULL CONCURRENTLY?

On Sat, Mar 22, 2025 at 5:43 AM Antonin Houska <ah(at)cybertec(dot)at> wrote:
> Can you please give me an example? I don't recall seeing a lock upgrade in the
> tree. That's the reason I tried rather hard to avoid that.

VACUUM has to upgrade the lock in order to truncate away pages at the
end of the table.

Or just:
BEGIN;
SELECT * FROM sometable;
VACUUM FULL sometable;
COMMIT;

I don't think we should commit something that handles locking the way
this patch does. I mean, it would be one thing if you had a strategy
for avoiding erroring out when a deadlock would otherwise occur by
doing something clever. But it seems like you would just need to
detect the same problem in a different way. Doing something
non-standard doesn't make sense unless we get a clear benefit from it.
(Even then it might be unsafe, of course, but at least then you have a
motivation to take the risk.)

> > - On what basis do you make the statement in the last paragraph that
> > the decoding-related lag should not exceed one WAL segment? I guess
> > logical decoding probably keeps up pretty well most of the time but
> > this seems like a very strong guarantee for something I didn't know we
> > had any kind of guarantee about.
>
> The patch itself does guarantee that by checking the amount of unprocessed WAL
> regularly when it's copying the data into the new table. If too much WAL
> appears to be unprocessed, it enforces the decoding before the copying is
> resumed.

Hmm. If the source table is not locked against writes, it seems like
we could always get into a situation where this doesn't converge --
you just need to modify the table faster than those changes can be
decoded and applied. Maybe that's different from what we're talking
about here, though.

> > - What happens if we crash?
>
> The replication slot we create is RS_TEMPORARY, so it disappears after
> restart. Everything else is as if the current implementation of CLUSTER ends
> due to crash.

Cool.

--
Robert Haas
EDB: http://www.enterprisedb.com