Le 16/09/2010 20:39, Josh Berkus a écrit :
>
>> It's been pure nonsense in this thread. Please show an example of
>> what's not working.
>
> 1) Init a postgresql 8.3 with autovacuum disabled.
>
> 2) Load a backup of a database into that PostgreSQL.
>
> 3) Check pg_stat_user_tables. n_live_tup for all tables will be 0.
>
> 4) VACUUM ANALYZE the whole database.
>
> 5) n_live_tup will *still* be 0. Whereas reltuples in pg_class will be
> reasonable accurate.
>
Did all your steps (except the fourth one). Works great (meaning
n_live_tup is updated as it should be).
I have to agree with Alvarro, this is complete nonsense. VACUUM ANALYZE
doesn't change the pg_stat_*_tables columns value, the stats collector does.
If your n_live_tup didn't get updated, I'm quite sure you have
track_counts to off in your postgresql.conf file.
>> Um ... it updates the last_autovacuum and last_autoanalyze columns,
>> but the others are not its responsibility.
>
> Right. I'm contending that ANALYZE *should* update those columns.
The postgres process executing ANALYZE surely sent this information to
the stats collector (once again, if track_counts is on). Tried it
tonight, works great too.
> Current behavior is unintuitive and makes the stats in
> pg_stat_user_tables almost useless, since you can never get even
> approximately a coherent snapshot of data for all tables.
>
Get a look at your track_count setting.
--
Guillaume
http://www.postgresql.fr
http://dalibo.com