The test for me went form 17 mins to run everything up to 30 mins to run everything.
I did some debugging and found that the slot is not very responsive if it cannot connect to nimbus (to send metrics back). In those cases the connection has to time out before the slot can respond.
I think we want to do 2 things here. First we don't want slot blocking for long periods of time if it cannot talk to nimbus, so I am going to background the sync, with a queue that will throw out old metrics if the new ones want to be sent.
Next I am going to figure out why the local override is not happening in the nimbus client in this situation.