Parallel gw executes more executions then expected (cont.)

[stepankveton]

Hi,
some time ago I started this thread http://forums.activiti.org/en/viewtopic.php?f=6&t=3949&start=0.
The issue disappeared after restarting a server so no one tried to actually solve it (including me).
But we are facing the same issue right now. Did this become a known issue or is it still a mystery?

We are still using activiti 5.8 and are planning to switch to 5.10.

[stepankveton]

Some more information: similar issue happens when calling method signalOtherExecution(). The other execution recieves the signal but randomly skips some activities (like the third execution started from the parallel gw in previous post)

[ronald_van_kuij]

Stepan,

Like mentioned in the other topic, it is very hard to help with this little information.

[stepankveton]

I'm aware of that, just tell me what information you need and I will provide it 🙂

There are no stacktraces or any relevant log output from Activiti.
Would it be helpful to provide a db dump of related data?

[jbarrez]

The most helpful would be a reproducable unit test … otherwise it's searching for a needle in a haystack

[stepankveton]

The most helpful would be a reproducable unit test … otherwise it's searching for a needle in a haystack

Since it happens only on production servers and it happend only two times during last 6 months, it's not possible to provide a reproducible unit test 😞
We can make modifications to our code so we can get more information from Activiti when it happens again but I need to know what to look for.
Also I would prefer to prevent this from happening again but I understand that it might not be possible w/o further investigation.

[jbarrez]

If it happens that rarely, I would suspect a racing condition between threads. Does the process in question involve timers/jobs? Are you handling the process in a clustered way? You see, without the exact environment it's very hard for us to see whether it is really a bug at all.

[stepankveton]

None of the affected processes use timers or jobs. The way we handle processes is as follows:
We have an OSGi runtime running one instance of activity engine connected to oracle database.
Any operation on processes is invoked through servlet call with additional information like what to do and on what process.
So the place where threads are started is the servlet which calls activiti directly.

[jbarrez]

And are you using any 'asynchronous' steps in your process? These are also handled by the separate thread pool.