Обсуждение: database profiling

I'm running several postgreSQL databases containing tens of millions of
rows.  While postgreSQL has proven itself amazingly scalable, I am
finding it difficult to properly tune/profile these databases as they
grow.  I am not aware of any tools to profile the database outside of
heavy use of the "EXPLAIN" command and application space logging of
query times.  I wanted to lay down some ideas I have for some powerful
tools for dealing with this.  Please let me know if these are redundant,
not needed or too complex to implement.

1) Statistics kept on all the tables and indexes showing the number of
times and the number of milliseconds spent sequential scanning / index
scanning them.  You could easily identify an index that's not being used
or a table that's being sequential scanned too often.  Also supply a
database command to reset the statistic gather to zero.

2) Statics that show count/total times to insert into/updating/deleting
from a table with time times spent in table and index separated so you
could see how much the indexes are effecting the write performance on
the table.

3) The ability to view the current running time, plan and SQL text of
all currently running queries.

Thank you for your time,

___________________________________________________
O R I O N  H E N R Y

Chief Technical Officer, TrustCommerce
959 East Colorado Blvd #1,  Pasadena, CA 91106
(626) 744-7700 x815, fax (626) 628-3431 orion@trustcommerce.com
www.trustcommerce.com

> 1) Statistics kept on all the tables and indexes showing the number of
> times and the number of milliseconds spent sequential scanning / index
> scanning them.  You could easily identify an index that's not being used
> or a table that's being sequential scanned too often.  Also supply a
> database command to reset the statistic gather to zero.

Enable the stats collector in your postgresql.conf, and then look at the
pg_stat_* views.  They are your friend.

> 2) Statics that show count/total times to insert into/updating/deleting
> from a table with time times spent in table and index separated so you
> could see how much the indexes are effecting the write performance on
> the table.

Lots of that is in the pg_statio_* tables (if you have the collector
running)

> 3) The ability to view the current running time, plan and SQL text of
> all currently running queries.

That's in the pg_stat_activity table.

There's also a new postgresql.conf setting in 7.4 that lets you log any
query that runs more than a certain number of milliseconds - which is
very useful for tracking down your slow queries.

Chris