> Well, it's hard to be sure what the problem is when you're not showing
> us a problem case ... but I notice that this indexscan is estimated
> awfully high:
Whenever I do it manually it works fast. But in the log I see lots of
slow ones. Could it be caused by auto vacuum? Or by check pointing or
WAL writing? Are there way to check that?
> > -> Index Scan using
> > pg_depend_reference_index on pg_depend dep (cost=0.00..64942.17
> > rows=247 width=12) (actual time=396.542..1547.172 rows=22 loops=1)
> > Index Cond: (refobjid = 30375069::oid)
> The reason is not far to seek: the scan is checking only the second
> index key, meaning that it has to scan the entire index. (I am
> surprised it didn't use a seqscan instead. Are you using enable_seqscan
> = off? Not a great idea.) Since you know you are looking for a table,
> you could improve matters by adding a constraint on refclassid:
> dep.refclassid = 'pg_class'::regclass
enable_setscan is on. Is there a way to analyze/vacuum those tables?
I will look if I can also improve the query to be more exact.
In response to
pgsql-general by date
|Next:||From: Gurjeet Singh||Date: 2008-03-31 19:15:13|
|Subject: Re: Can Postgres 8.x start if some disks containing tablespaces are not mounted?|
|Previous:||From: Pavan Deolasee||Date: 2008-03-31 18:34:25|
|Subject: Re: [HACKERS] ANALYZE getting dead tuple count hopelessly wrong|