Tables with lots of dead tuples despite autovacuum

On 09/12/12 5:00 PM, Wells Oliver wrote:

We have a few tables that are updated nightly on the order
of deleting and inserting 500k records.

I assumed autovacuum would do its thing and clean up the dead tuples,
but in looking at pg_stat_user_tables, I notice there are lots and
lots of dead tuples, on the order of a 500-600k.

What can I do about this? Why isn't autovacuum cleaning these tables?
Is this number of dead tuples acceptable?

Lastly, would it make sense to do a weekly full manual vacuum + analyze?

how old is the oldest transaction in pg_stat_activity ? vacuum can't
free any tuples newer than this.

--
john r pierce N 37, W 122
santa cruz ca mid-left coast

On Wed, Sep 12, 2012 at 5:00 PM, Wells Oliver <wellsoliver@gmail.com> wrote:

We have a few tables that are updated nightly on the order of deleting and
inserting 500k records.

Out of how many in total?

I assumed autovacuum would do its thing and clean up the dead tuples, but in
looking at pg_stat_user_tables, I notice there are lots and lots of dead
tuples, on the order of a 500-600k.

That is one day's worth. If the tables undergo churn once a day, and
get vacuum once a day, that seems like a nice steady state. I
wouldn't worry about it.

Cheers,

Jeff