On Tue, 2002-03-05 at 21:59, Bruce Momjian wrote:
> Justin Clift wrote:
> > Hi all,
> >
> > One of the things which the AS3AP benchmark does is have multiple users
> > access a table with hash indexes on it.
> >
> > With the OSDB (Open Source Database Benchmark: http://osdb.sf.net) we've
> > found on PG 7.1 that multiple clients hitting a table using a hash index
> > generates locking problems. I remember Tom mentioning that this is a
> > known thing, but I'm not sure if this has been fixed since then.
No, it hasn't been fixed yet.
> > Does anyone have any ideas? If not, would someone be willing to take
> > the time to fix it?
>
> It has not been fixed. One TODO item is to either stop mentioning hash
> at all or get it improved. We have been sitting on the fence for too
> long.
I'll be working on fixing this. I'm also going to try to add more
features to the hash index implementation: for example, allow UNIQUE
hash indexes, hash indexes over multiple keys, etc. My first improvement
to the hash code, replacing the hash function with a better one, is on
the unapplied patches list and should be in CVS soon. Bruce, can you add
my name to the TODO list next to this item?
BTW, does anyone have any tips for debugging deadlock conditions? I
normally debug problems by running a backend under gdb in standalone
mode, but that obviously won't help for this problem. Any further advice
on improving hash index concurrency would be very welcome...
Justin:
Cheers,
Neil
--
Neil Conway <neilconway@rogers.com>
PGP Key ID: DB3C29FC
