We upgraded from EC2 plug 2032.v92a_4b_e703974 to 2039.v829e6272107c.
With the new version, Agents never start. Specifically:

We re-user our Agents EC2 instance, which means that after an Agent has been idle for 10 minutes, the EC2 instance is Stopped (note: not Terminated).

What is supposed to happen is that, if a job starts that needs a Agent with some label(s), and there is an existing instance that matches the requirements, but that instance is currently in a stopped state, then the existing instance should be started and used (in preference to spinning up a new instance).

With 2039.v829e6272107c, what we noticed when a job started was:

• If there was no instance that matched the requirements, then a new instance was started and used (this is correct behaviour).
• If there was an existing, stopped instance that matched the requirements, then the job waited forever for an Agent. No new instance was created, and no existing instance was started. There was nothing at all in the Jenkins logs for this - no message indicating it was trying to start and Agent; nothing.

Our usage pattern for Agents is fairly common, I would think. No unusual settings.

cc: mikecirioli jgarciacloudbees (people who recently had PRs in this are)