It seems that a deadlock has been introduced in some recent changes to workflow-cps-plugin.

The whole jstack has been attached to the report, but two threads are waiting on each other:

• the first thread is calling CpsFlowExecution#subsumeHead, got the lock on the instance, and then trying to acquire the r/w lock on the node storage
• the second thread is calling CpsFlowExecution#getNode, got the r/w lock on the storage, and then is calling close which tries to get the instance lock