On Fri, Jan 28, 2011 at 4:58 PM, Jürgen Herrmann
<[email protected]> wrote:
...
> Afaics you use zodbpack's default of "days=0". This is known to produce
> POSKeyErrors if the database is written to while packing.
It can lead to POSKeyErrors under certain pathological cases, but Is not
generally a problem in practice.
An example where it could be a problem:
x = Foo()
self._p_jar.add(x)
transaction.commit()
At this point, x is in the database and is thus garbage. For the sake
of this discussion, assume the database is packed to this point in
time and x is removed from the database.
self.x = x
transaction.commit()
At this point, self refers to an object that is no longer in the
db. At some later time, once self has been ghostified and x is removed
from local cache, if we try to access x, we'll get a poskey error.
I would argue that the application code above is broken.
Note that the storage could validate the reference to x when self is
saved, but no storage that I know of does, for performance reasons.
In some future version of ZODB, it might be nice to change the
database record structure to make this sort of check easier.
Jim
--
Jim Fulton
http://www.linkedin.com/in/jimfulton
_______________________________________________
For more information about ZODB, see the ZODB Wiki:
http://www.zope.org/Wikis/ZODB/
ZODB-Dev mailing list - [email protected]
https://mail.zope.org/mailman/listinfo/zodb-dev
