[
https://issues.apache.org/jira/browse/WW-4409?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14154707#comment-14154707
]
Rob Baily commented on WW-4409:
-------------------------------
Hmm, interesting discussion in WW-3865. I don't understand why the actual
error was coming in that case. You mentioned error handling which seems
plausible but did anyone get to the root of it? It does seem though that if
the invocation is in the synchronized session block that it means a user can
really only ever have 1 thread truly active but other clicks will end up taking
a thread and blocking until previous requests are done. That would be like
handling all requests for a user serially which I guess could make sense but
seems like it might be a big issue with system stability since you could have
an undefined number of threads waiting for the object lock based on whatever
the user/browser does. Might be especially problematic with AJAX calls where
multiple things could be happening at once.
> Modify the TokenInterceptor not to lock the session object while handling and
> invalid token
> -------------------------------------------------------------------------------------------
>
> Key: WW-4409
> URL: https://issues.apache.org/jira/browse/WW-4409
> Project: Struts 2
> Issue Type: Improvement
> Components: Core Interceptors
> Affects Versions: 2.3.18
> Reporter: Rob Baily
> Priority: Minor
> Fix For: 2.3.x
>
> Attachments: tokeninterceptor.patch
>
>
> Similar to WW-3582. Currently in the TokenInterceptor class in the
> handleToken method there is a lock on the session object and the
> handleInvalidToken method is called within that block. In our environment we
> overrode the handleInvalidToken and it turned out that when this method ran
> long that any other requests for the user were immediately blocked and the
> threads were locked up and unable process any further requests.
> I would like to see if we can change the handleToken method not to lock the
> session when calling the method to handle an invalid token. I realize that
> ours may not be an ideal implementation and the base implementation should be
> fast but it does seem like the time an object is locked should be minimized.
> I don't know if there is a reason that the invalid method would need to be in
> that block.
> I am including a patch for the change on this. It does not affect the tests
> and I do not know if there is a good way to add a test that would check the
> locking.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
