Thread: FW: Postgres: VACUUM

FW: Postgres: VACUUM

From
Date:
> On heavily used databases (over 100,000 transactions an
> hour), vacuum is
> a killer

That's about 27 tx/second - not so many, for some tasks at least.

If VACUUM is rather a killer
- are any plans from PostgreSQL to deal with that?



Thank you in advance,
Laimis

P.S. it's notable that every DB system I delt with had one or another issue
as far as concurency is concerned, take infamous "Oracle-1555 snapshot too
old" error as an example.




Re: FW: Postgres: VACUUM

From
"scott.marlowe"
Date:
On Wed, 14 Jan 2004 lnd@hnit.is wrote:

>
> > On heavily used databases (over 100,000 transactions an
> > hour), vacuum is
> > a killer
>
> That's about 27 tx/second - not so many, for some tasks at least.
>
> If VACUUM is rather a killer
> - are any plans from PostgreSQL to deal with that?

Yes, Tom and friends have been tossing back and forth a patch to implement
a sleep between x number of pages being vacuumed to remove the heavy load
vacuum can place on a system.  Early testing looks promising, so it's
likely that if enough of us order Tom or somebody a pizza it will get
done. :-)  I certainly would like to see it.


Re: FW: Postgres: VACUUM

From
Christopher Browne
Date:
lnd@hnit.is writes:
>> On heavily used databases (over 100,000 transactions an hour),
>> vacuum is a killer
>
> That's about 27 tx/second - not so many, for some tasks at least.
>
> If VACUUM is rather a killer
> - are any plans from PostgreSQL to deal with that?

Yes.  There is a patch available that causes VACUUM to sleep() after
every few pages it processes, which throttles VACUUM's ill effects on
cache.  It isn't yet in 7.4.1, but might be in a later 7.4 release.

There is also a major revision taking place (look up "ARC" on the
hackers list) for management of cache that should also be helpful.
--
let name="cbbrowne" and tld="libertyrms.info" in String.concat "@" [name;tld];;
<http://dev6.int.libertyrms.com/>
Christopher Browne
(416) 646 3304 x124 (land)