Merging output from parallel blocks on parallel gateway sometimes doesn't work

[DotNetPart]

Describe the bug

I have quite a simple workflow where I split flow into 3 parallel requests and then merge their output in parallel gateway.
Sometimes though, my flow hangs and merging doesn't happen though all 3 parallel blocks have finished. I can see that browsing through kogito process instance events.

I went through Kogito code a bit and found a suspicious place.
Doesn't this one have a race condition in count updating procedure?

https://github.com/kiegroup/kogito-runtimes/blob/1b2146d259b9119e5fb4e51c422c010199689789/jbpm/jbpm-flow/src/main/java/org/jbpm/workflow/instance/node/JoinInstance.java#L71

Expected behavior

No response

Actual behavior

No response

How to Reproduce?

No response

Output of uname -a or ver

No response

Output of java -version

No response

GraalVM version (if different from Java)

No response

Kogito version or git rev (or at least Quarkus version if you are using Kogito via Quarkus platform BOM)

No response

Build tool (ie. output of mvnw --version or gradlew --version)

No response

Additional information

No response

[radtriste]

@ricardozanini @fjtirado any idea ?

[dfiai]

Hello, I have a similar issue. Additionally, when I have several identical BPMN process pods, I encounter a problem where processes with parallel actions freeze.

[dfiai]

@ricardozanini, @fjtirado what are your thoughts on this(but it still won't work with several pods):

import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.atomic.AtomicInteger;

private ConcurrentHashMap<Long, AtomicInteger> triggers = new ConcurrentHashMap<>();
...
case Join.TYPE_AND:
    // Use computeIfAbsent to handle the initialization and increment atomically
    triggers.computeIfAbsent(from.getNodeId(), k -> new AtomicInteger(0)).incrementAndGet();

    if (checkAllActivated()) {
        decreaseAllTriggers();
        triggerCompleted();
    }
    break;

[fjtirado]

@dfiai Please go ahead opening a PR with that fix
Certainly it would still fail when using multi pod, but at least we avoid the race condition while single pod.
Good catch!

[fjtirado]

@dfiai If you have issues opening the PR, let me know and I will do myself, but I believe you deserve the credit on pursuing this together with @DotNetPart

[dfiai]

@fjtirado, sure I'll try it by myself first.