Re: pgsql: When estimating the selectivity of an inequality "column >

Greg Stark wrote:

> I also wonder if we want to run these lookups on every single planner
> invocation. If the table is being rarely updated analyze will never
> fire and the stats value will never be updated. I don't think we want
> the planner doing write operations to the stats either though. It
> would be nice if there was a low-cost mode of operation for ANALYZE
> which simulates precisely this update, the planner could send a stats
> message saying that autovacuum should schedule one of these operations
> sometime even if it doesn't see a need for a full analyze run.

Maybe autovac could run such a cheap ANALYZE frequently on tables with
large number of inserts (but not large enough to trigger a regular
ANALYZE) ... say a fixed number of tuples (not depending on pg_class.reltuples)

--
Alvaro Herrera http://www.CommandPrompt.com/
PostgreSQL Replication, Consulting, Custom Development, 24x7 support