Обсуждение: GiST memory usage

Hi all.
I've been reported memory limit hits
during a postgis index construction
(GiST index).

The indexed tuples are about 425 millions.
The index key is a box2d object (4*sizeof(float)).

The machine has 4 Gb of RAM, and operations are
peaked at ~2.4 Gb of usage.

Is there a way to detect where is memory hold ?

Is there an index-specific memory context for use
to early release memory blocks allocated during
operations (didn't see any use of memory contexts
in the contrib/rtree_gist) ?

TIA

--strk;

Missing info for the report:win32postgresql 8.0.3maintenance_work_mem  512btree index on oid works fine


--strk;

On Sat, Aug 13, 2005 at 01:12:30PM +0200, strk@refractions.net wrote:
> Hi all.
> I've been reported memory limit hits
> during a postgis index construction
> (GiST index).
> 
> The indexed tuples are about 425 millions.
> The index key is a box2d object (4*sizeof(float)).
> 
> The machine has 4 Gb of RAM, and operations are
> peaked at ~2.4 Gb of usage.
> 
> Is there a way to detect where is memory hold ?
> 
> Is there an index-specific memory context for use
> to early release memory blocks allocated during
> operations (didn't see any use of memory contexts
> in the contrib/rtree_gist) ?
> 
> TIA
> 
> --strk;

> Is there a way to detect where is memory hold ?
Look at postgis code. We constructed rtree_gist index over 1e9 rows (on 8.0 pgsql).


> 
> Is there an index-specific memory context for use
> to early release memory blocks allocated during
> operations (didn't see any use of memory contexts
> in the contrib/rtree_gist) ?

8.1 pgsql has short lived memory context during index creation which resets 
after every insert. 8.0 hasn't this feature, so user-defined function (GiST 
interfface) should carefully clean memory.

-- 
Teodor Sigaev                                   E-mail: teodor@sigaev.ru
  WWW: http://www.sigaev.ru/