Re: max_fsm_pages question

Michael Monnerie  wrote:
 
> such a bloat can't happen in a day.
 
That is evidence that you may have a problem with some long-running
transaction which stays open for days, possibly "idle in
transaction".  Bloat will accumulate, without any vacuum being able
to prevent it or recover from it, until the transaction terminates.
It will not show up as free space in your vacuum verbose output
while the transaction remains open; although I believe it will show
up in the "dead row versions cannot be removed yet" count.
 
> Finding out which table the toast belongs to would be great, maybe
> that helps further investigation.
 
Have a look at reltoastrelid:
 
http://www.postgresql.org/docs/8.3/interactive/catalog-pg-class.html
 
-Kevin