Comment 0 for bug 872378

The watch callback watch_machine_changes in juju.agents.provision will stop working, and allowing the watch to re-establish, if the function it calls, process_machines, raises an uncaught exception. A scenario where this can happen is that txaws attempts to parse a bad payload and raises for example a KeyError (one of a number of possibilities that have been observed in parsing, along with timeout errors; there may be others).

At this point, only the periodic_machine_check is then run, so the provisioning agent only is resyncing every 60 seconds, making it sluggish.