Asynchronous before + Exclusive true in cluster problem

Camunda 7 Topics Discussion & Questions
1

Hello,
We are running Camunda in a cluster with 3 nodes pointing to the same DB. They all have the same setup. Each node has a Job executor that runs the processes. The config is default. I only have maxJobsPerAcquisition=100 because it seems to perform a bit better in our case.
The case consists of 2 processes. Process A has a Call Activity which executes Process B.
Process B has 2 parallel branches. One of them has a Task which sends a message to our messaging system. The Task is marked Asynchronous before = true, Exclusive = true. I believe that this should ensure that the Task is executed only once. In my case however the Task is executed 2 times in about 50 out of 20.000 started workflows. It is sometimes even executed twice by the same node in the cluster.
I start instances of Process A as fast as I can (about 150/s) through API calls on one of the nodes.

EDIT:
The critical task uses a Java Delegate to send the message to our messaging system.

Process A:
SL_request_reply_usage_no_wait.bpmn (5.1 KB)

Process B:
SL_request_reply.bpmn (6.5 KB)

Thank you all for your response.
Jiri

2

After some thinking I realized that this is kind of a trivial question.
What may happen is that the delegate executes but something prevents camunda to save the state in the DB. Network, concurrency, whatever. Hence it then runs the same task again.
We just can’t assume that the delegate is guaranteed to be executed only once.

3

I think if we have mentioned retries in modeller then even the delegate will get executed as per retry command.
https://youtu.be/MpueF6LllxE

4