Re: Libpq Asynchronous Command Processing

On 31/05/2010 5:41 PM, Giles Lean wrote:
> =?iso-8859-1?Q?Alonso_Garc=EDa_=2C_Bruno_Elier?=<bealonso@indra.es>  wrote:
>
>> And the problems I am finding are the following:
>> ->Queries from the client to the new DB server take a lot of time.
>> ->Queries from the client to the old DB server are fast.
>> ->The same query takes 150 secs in one case an 1 sec in the other case.
>
> With that analysis, I'd be betting against it being a client problem.
> (If you wanted, you might confirm that by pointing an old client at
> the new server.)
>
> I'd look into how the data was loaded into the new server and how
> the database is configured: number of buffers, indexes, and whether
> analyze has been run or not.
>
> It would be strange indeed (possible, but very strange) to find
> such a slowdown between 7.x and 8.x when the team is preparing
> to push 9.0 out the door.  Surely it would have been known before;
> therefore it's a practical certatinty that there is something
> different about the configuration of your two servers.

... or that the planner is making a bad choice when it made a good one
in 7.x . That's far from unheard of; the downside of a stats-based and
very complex planner is that sometimes it doesn't make the perfect
choice. Even with the same stats, etc, it's far from impossible that 7.x
might hit a good plan when 8.x doesn't.

I mention this because the OP really needs to supply EXPLAIN ANALYZE
results for the query run via psql (not their custom code) on both their
7.x and 8.x servers.

--
Craig Ringer