Re: Vacuum now uses AccessShareLock for analyze

> > How separate do people want vacuum and analyze? Analyze currently does
> > not record the number of tuples and pages, because vacuum does that. Do
> > people want analyze as a separate command and in a separate file?
>
> We definitely want a separate command that can invoke just the analyze
> part. I'd guess something like "ANALYZE [ VERBOSE ] optional-table-name
> (optional-list-of-columns)" pretty much like VACUUM.

OK.

>
> I would be inclined to move the code out to a new file, just because
> vacuum.c is so darn big, but that's purely a code-beautification issue.

Done.

>
> On the number of tuples/pages issue, I'd suggest removing that function
> from plain vacuum and make the analyze part do it instead. It's always
> made me uncomfortable that vacuum needs to update system relations while
> it's holding an exclusive lock on the table-being-vacuumed (which might
> be another system catalog, or even pg_class itself). It works, more or
> less, but that update-tuple-in-place code is awfully ugly and
> fragile-looking. I'm also worried that there could be deadlock
> scenarios between concurrent vacuums (eg, one guy working on pg_class,
> another on pg_statistic, both need to get in and update the other guy's
> table. Oops. That particular problem should be gone with your changes,
> but maybe there are still problems just from the need to update
> pg_class).

How do I find the number of pages from heapscan? Can that number just
be computed from the file size. I can get the block number of the last
entry in the scan, but that doesn't show me expired rows at the end.

--
Bruce Momjian | http://www.op.net/~candle
pgman(at)candle(dot)pha(dot)pa(dot)us | (610) 853-3000
+ If your life is a hard drive, | 830 Blythe Avenue
+ Christ can be your backup. | Drexel Hill, Pennsylvania 19026