on 12/30/2000 6:00 AM, "Kief Morris" <[EMAIL PROTECTED]> wrote:
> Jon Stevens typed the following on 06:35 PM 12/29/2000 -0800
>>> I can see this being used for fault tolerance. If your servlet engine
>> crashes,
>>> or if you have to restart it for some reasons, your users' sessions don't
>> need
>>> to be lost.
>>
>> If the server crashes, then you are screwed either way.
>
> If it crashes and is restarted quickly, it can load the serialized sessions
> from disk and users can carry on. This assumes you've got some kind
> of script watching and restarting the server process.
No, I consider a "crash" a JVM core dump or a kill -9 jvm_pid. In that case,
there is no chance for cleanup and therefore, the sessions are potentially
invalid anyway. What happens if the "crash" happens during a write
operation?
>> If you are live, then serializing/deserializing the sessions for each and
>> every request could be a major slowdown (even with this code).
>
> It doesn't need to. I'm writing PersistentManager to put the session
> on a queue to be serialized by another thread. This shouldn't have a
> noticeable performance impact.
But isn't 100% reliable because if the system is killed, then those threads
aren't executed and the data is potentially invalid.
-jon
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, email: [EMAIL PROTECTED]
