I am wondering if your wait is caused by contention between
pg_autovacuum and the DELETE that is running. Your large Pg blocksize
(32K) *may* be contributing to any possible contention as well. Maybe
try disabling pg_autovacuum to see if there is any change in behaviour.
Also going through my head is '32 Kb bock's size (to match ffs and raid
block's size)' - does that mean you have raid strip size = 32K? maybe
try 128K (I know it sounds like a bad thing, but generally raid stripes
of 128K->256K are better than 32K->64K)
regards
Mark
Pailloncy Jean-Gérard wrote:
> Hello,
>
>>
> I found the same problem.
>
> I use OpenBSD 3.3,
> On Pentium 2,4 GHz with 1 Gb RAM, RAID 10.
> With PostgreSQL 7.4.1 with 32 Kb bock's size (to match ffs and raid
> block's size)
> With pg_autovacuum daemon from Pg 7.5.
>
> I run a web indexer.
> sd0 raid-1 with system pg-log and indexer-log
> sd1 raid-10 with pg-data and indexer-data
> The sd1 disk achives between 10 and 40 Mb/s on normal operation.
>
> When I get semwait in top, system waits ;-)
> Not much disk activity.
> Not much log in pg or indexer.
> Just wait....
>
