Exclusive jobs spanning across multi-instance call activities

[engineering-issue-sync-app]

Value Proposition Statement 🚀

Allow advanced workflow patterns with exclusive jobs for tasks in multi-instance sub-processes.

User Problem 🤦

Context:

• A process can start multiple "subprocesses" (via Multi-Instance Call Activity).

• In certain scenarios, tasks in the subprocess should be executed exclusively as it, e.g., uses resources that do not allow it to be executed in parallel.

Problem:

• The job executor cannot track exclusive jobs beyond one process instance as the parent process, and the called process will have different instance IDs

• OptimisticLockingException (OLE) can occur.

• The executor handles the OLEs, and the retries are not decreased (so no incidents occur due to this)

• OLEs are not hindering exceptions, but they still appear again and again in the same place and are also visible in the log

• Current workaround: The business team defines special retry time cycles at certain points to perform retries faster.

User Stories 🧑‍🚀

As a developer, I want to use an exclusive continuation for tasks in a multi-instance sub-process to ensure a task is not executed in parallel.

Implementation 👷

Solution outline

• Add root process instance id column to the runtime job table.
• Replace the process instance id with the root process instance id for the process instance and the runtime job in the condition of the job acquisition query.
  • For the top-level process instance, root process instance id == process instance id so the previous behavior is maintained.
• Since jobs already run on a previous Camunda version the root process instance is null.
  • A fallback to the process instance id columns is needed.

Open questions

• Is it problematic to roll out this new behavior to all the users? Are there cases in which this behavior is problematic or undesired?
  • The performance for jobs with no root process instance id populated degrades since the job acquisition query is more complex; is this acceptable?
  • Are there users that want to maintain the current behavior? Since nobody has asked for this new behavior, the answer is probably yes.
• Should we make this configurable? In the process engine config or on the BPMN model level?

Implementation Notes

Tasks

Beta Give feedback

1. Exclusive jobs across sub process instances camunda-bpm-platform#4003
   2 of 2
   type:feature version:7.21.0 version:7.21.0-alpha5 +3
   

^{🤖 This issue is automatically synced from: source}