Re: reducing statistics write overhead

Euler Taveira de Oliveira <euler(at)timbira(dot)com> writes:
> Alvaro Herrera escreveu:
>> Well, the problem is precisely how to size the list. I don't like the
>> idea of keeping an arbitrary number in memory; it adds another
>> mostly-useless tunable that we'll need to answer questions about for all
>> eternity.

Is it so hard? In particular, rather than making it a tunable, what say
we freeze the list size at exactly two, ie each AV worker advertises its
current and most recent target table in shared memory. Other workers
avoid re-vacuuming those. Then the most work you can "waste" by extra
vacuuming is less than the maximum allowed stats file age. I'd have no
problem whatsoever in letting that run into multiple seconds, as long
as it doesn't get into minutes or hours.

regards, tom lane