[ZODB-Dev] FileStorage database unrecoverable?

Casey Duncan casey@zope.com
Mon, 31 Mar 2003 10:36:42 -0500


On Monday 31 March 2003 07:09 am, Toby Dickenson wrote:
> (cc list trimmed)
>=20
> On Saturday 29 March 2003 1:38 am, Dieter Maurer wrote:
> > Toby Dickenson wrote at 2003-3-28 08:22 +0000:
> >  > ...
> >  > For a 10G FileStorage you might need up to 1G of ram to hold
> >  > FileStorage's index. If thats OK, then FileStorage will be hard to=
=20
beat.
> >
> > Is this still true for Zope 2.6.1?
> >
> >    I read that the index is no longer held in a dictionary but some
> >    BTree structure only partial in RAM.
>=20
> Some recent versions use a BTree that is linearly smaller than the old=20
> dictionary, but still all held in main memory. It looks like it *may* h=
ave=20
> better locality and so perform better when some of that index is moved =
into=20
> swap, but that is just a guess.=20
>=20
> This 'fsindex' feature was added, accidentally removed  during a branch=
=20
merge,=20
> then re-added. Im not sure exectly which versions use what.

AFAIK, 2.6.1 was the first version to use fsindex, though the code was sh=
ipped=20
with 2.6.0 without being wired in (inadvertently).

-Casey