ExecutorPickle.rehydrate ought to be able to detect that it has been spinning in circles because the agent node it was supposed to run on is not in the Jenkins node list, and automatically abort, causing the build to fail with a comprehensible message rather than just hanging indefinitely. (As opposed to being registered but offline, which is normal enough for a JNLP agent etc.—in such cases we just want to wait for the agent to come back online.)

This would provide a better experience for the case of a build which was running on an EphemeralNode (such as from a Cloud without durable-task integration) when Jenkins was restarted. An agent using an inappropriate RetentionStrategy is trickier since it might still be defined after a restart, but will soon be terminated. Similarly, there may be cases where the agent is actually going to be redefined (with the same name) when it is attached after the restart—not sure about the Swarm plugin, but CloudBees DEV@cloud OPEs work this way. To prevent the build from being killed too aggressively, the cleanup should be delayed until some time has elapsed since rehydration began (or, ideally, since Jenkins completed initialization)—say, five minutes.