Hervé Piedvache wrote:
> I have a small question ... right now on a 8.1.10 version of PostgreSQL I'm
> doing a vacuum full verbose anaylze a table with 60 304 340 rows in 1155791
> pages and there were 16 835 144 unused item pointers inside and 5 index.
>
> After the first treatment of the index (appeared in the verbose) ... the
> vacuum is still working now since 15 hours ... There is none other activity
> on this server ... it's a replication server ... so the vacuum is locking
> most of the replication queries ...
> What can I do to get this doing quicker ...?
> It's a 8Gb server ... with a RAID 10 ... my maintenance_work_mem = 25600.
>
> Any idea ?
> I'm sure that if I delete all the index an rebuild them by hand it'll be
> really quicker ! (done on another replication server ... took 20 min)
Yes, it probably will be. IIRC we even suggest doing that in the manual.
Another alternative is to run CLUSTER instead of VACUUM FULL. Increasing
maintenance_work_mem will make the index builds go faster.
Consider if you really need to run VACUUM FULL, or would plain VACUUM be
enough.
--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com