CI runners timeout prematurely, losing time at startup

It seems that since we have changed to override the runners, quite frequently the runners seem to start immediately but just idle a long time while awaiting something.

This would seem to me a bug in the custom runner configuration, probably. Normally the runner appears to be paused until it starts, it should not start immediately but be paused under the hood, probably the runner needs to report a ready state to the gitlab instance in some way.

The negative effect here is on some days when traffic is high, gitlab loses 15+ minutes in an idle state; since our CI takes a long time to complete, the time lost at startup can interrupt the CI task inappropriately, and fail it due to a miscalculated CI timeout.