[
https://issues.apache.org/jira/browse/ODE-662?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12755188#action_12755188
]
Alex Boisvert commented on ODE-662:
-----------------------------------
Yes, I considered this "feature" as well and you can still have internal
communication between processes that can last more than 3 minutes by using
endpoint configuration and configuring the MEX timeout to some higher value (up
to Integer.MAX_VALUE milliseconds).
For transacted invokes, we don't need an INVOKE_CHECK... there's already a
timeout for the transaction and if the invoke hasn't succeeded before the
timeout, then the transaction should be rolled back. Having a
database-persistent scheduled job for this case would be pointless. We could
have an in-memory schedule for the MEX timeout, however I would strongly advise
keeping a model where everything in a transaction executes within a single
thread. Transactions executing in multiple threads are hairy and not worth
the complexity, IMO. And keeping everything in a single thread has other
efficiency/performance benefits.
> Process-to-process invocation does not go into activity recovery if no reply
> is sent from child process
> -------------------------------------------------------------------------------------------------------
>
> Key: ODE-662
> URL: https://issues.apache.org/jira/browse/ODE-662
> Project: ODE
> Issue Type: Bug
> Affects Versions: 1.3.3
> Reporter: Alex Boisvert
> Assignee: Alex Boisvert
> Fix For: 1.3.4
>
>
> When a process invokes another process, there's currently no invoke check
> performed afterwhile to force the <invoke> into activity recovery if no reply
> comes back.
> Need to add the invoke check.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
