© 2004
Canonical Ltd.
•
Terms of use
•
Data privacy
•
Contact Launchpad Support
•
Blog
•
Careers
•
System status
•
0e1f616
(Get the code!)
This turned out to be because:
* The model manager worker was starting the model workers while the model was still importing back. * The migration master would start and see the *old* migration (because the migration mode was importing rather than none).
* Since the phase was DONE, it exited returning ErrMigrated.
* This would get translated into ErrUninstall, so the dependency engine would remove the migration master worker.
* This then meant that the migration master wouldn't unlock the migration fortress, so workers that depend on ifNotMigrating (like the instance poller) couldn't run.
The fix is to make the model worker manager not start the workers until the model is finished importing. I'm preparing a PR for this - it's complicated by the watcher the model worker manager uses only signalling on lifecycle changes, so the mwm doesn't get told when the import is finished.