Re: CPU-intensive autovacuuming

"Thomas F. O'Connell" <tfo@sitening.com> writes:
> Honestly, I'd prefer to see pg_autovacuum improved to do O(n) rather
> than O(n^2) table activity. At this point, though, I'm probably not
> too likely to have much time to hack pg_autovacuum before 8.1 is
> released, although if it doesn't become integrated by beta feature
> freeze, I might give it a shot.

This would be vastly easier to fix if the code were integrated into the
backend first.  In the backend environment you could just keep the info
in a dynahash.c hashtable instead of in a linear list.  On the client
side, you have to roll your own hashing (or adapt dynahash to life
outside the backend environment).

            regards, tom lane