When a Thor is configured with multiJobLinger: true (since ), the Thor instances will listen to the Thor queue for the linger period.

This is the same queue as the agents that launch the Thor instances listen to.

A consequence is that if there are slots available (Thor instances < maxGraphs) the agent may dequeue a job and launch a new Thor instance, when a lingering Thor instance could have done so instead (which would cost less and have much less latency).

In practice this may not be that significant, because multiJobLinger is designed to keep Thor instances busy, and the general expectation is that the agent slots will be taken, i.e. the maximum instances will be constantly running.

Nevertheless, it would be a good improvement, if Thor's that were already running had priority to dequeue over the agents.