abruptly restarted (admin sends -9, or daemon is OOM'ed)
or the server loses power and it is rebooted
... then the plugin system should be able to cleanly restart plugins and recover.
Of particular concern is whether containers might outlive the daemon. If this is true, and stranded plugin containers might hold on to system resources (like mounted block devices), then we could end up with stranded plugins stopping newly spawned plugins from functioning. We should have acceptance tests which test and demonstrate the behaviour in this case.
Two cases:
If the daemon is:
... then the plugin system should be able to cleanly restart plugins and recover.
Of particular concern is whether containers might outlive the daemon. If this is true, and stranded plugin containers might hold on to system resources (like mounted block devices), then we could end up with stranded plugins stopping newly spawned plugins from functioning. We should have acceptance tests which test and demonstrate the behaviour in this case.