Ghislain ROUVIGNAC wrote:
> Threre is a vacuum analyze planned during the night.
> The morning, 1 day out of 2, there are some extremely slow
> queries. Those queries lasts more than 5 minutes (never waited
> more and cancelled them) whereas when everything is OK they last
> less than 300ms.
>
> In order to resolve this i have to execute a least one analyze,
> sometimes more.
>
> My Configuration:
> Windows
> PostgreSQL 8.4.8
> default_statistics_target = 100
>
> In addition to an increase for shared_buffers, work_mem, ..., i
> changed the default_statistics_target to 500 with no effect.
> It was even worse as i never managed to get rid of the slow
> queries after running many analyze.
> So i fell back to default_statistics_target=100 in order to get
> rid of those slow queries.
You probably need to adjust your cost factors to more accurately
reflect the actual costs of various activities on your system. What
is probably happening is that there are two plans which are very
close together in estimated costs using the current values, while
the actual costs are very different. The particular random sample
chosen can push the balance one way or the other.
Please show the results from running the query on this page:
http://wiki.postgresql.org/wiki/Server_Configuration
Also, a description of the run environment would help.
Other information listed on this page would help, although cores,
RAM, and storage system information would probably be most
important.
http://wiki.postgresql.org/wiki/Server_Configuration
-Kevin