1. No support for this is planned
2. The cancelActivity uses the database to cancel the running activity. As long as the async service-task is "running", the process-state in the DB is unaffected. When the service-tasks finished and tries to write it's data, it will get an optimistic lock-exception, effectively rolling back the work it did in the service-tasks, so transactionally speaking, it "never happened". There is no way of "notifying" or "canceling" the job-threads at the moment. Also a bit difficult to implement right, keeping in mind job-executor may be in different VM…
3. Do you mean the error-codes in the BPMNErrors/thowing boundary event?