On Sun, 22 Apr 2001, Kief Morris wrote:
> I'm fixing a problem with PersistentManager which affects how sessions are
> removed from memory. This will require making a change to StandardSession,
> which I'd like to get some feedback on. Looking at this issue makes me think
> that StandardManager is also doing the wrong thing when unloading sessions,
> namely sending sessionDestroyed() events to listeners.
>
You are correct - it should send sessionPassivated() to interested
listeners, but not sessionDestroyed(). I thought this was fixed quite a
while ago (and the tester series checks for this).
Perhaps the call to expire got regressed somehow during the refactoring?
> The PersistentManager issue is how sessions are unloaded to the Store, which
> currently copies StandardManager's algorithm of calling session.expire(). The
>expire()
> method is really intended for sessions which are being destroyed, either through
> invalidation or timeout. PersistentManager needs to be modified so it will remove
> sessions from its Store when expire() is called, which can be done by overriding
> the remove method:
>
> public void remove(Session session) {
>
> super.remove(session);
> if (store != null)
> try {
> store.remove(session.getId());
> } catch (IOException e) {
> log("IOException removing session from Store: " + e.getMessage());
> e.printStackTrace();
> }
>
> }
>
> No problem, except that this method is also triggered from the swapOut() method
> by a call to StandardSession.expire(), which neuters swapOut().
>
> So I'm thinking the thing to do is break out StandardSession.expire() functionality
> so there are two public methods, one for expiring/destroying a session, one to
> simply release/recycle the session (perhaps release()?)
>
> This should be straightforward, but looking at this makes me wonder whether
> StandardManager's behaviour on unload() is correct: it is sending events to
> any listeners which indicate the session is being destroyed. My reading of the
> spec suggests it should just send passivation-related events. I'm not sure about
> the unbinding of attributes.
>
> Comments?
>
> Kief
>
>
Are you sure it is really doing that? The tester logic (Session02 ->
reload -> Session03) should be catching it.
Craig
