Silence kill_kernel when no process is present

[kevin-bates]

During a kernel's shutdown, a race condition can occur between the kernel manager and the kernel restarter such that the restarter sees the kernel process has terminated, yet it's view of the kernel manager still thinks its managing an instance. In such cases, the restarter initiates its restart sequence and, when attempting to kill the kernel, finds that there's now no kernel process being managed. In such conditions, the kernel manager was raising an exception indicating there's no process to kill. Instead, it should perform the same action as when there is a process to kill, only to have the signaling fail because there is no longer any process - which is to ignore that condition. This change essentially ignores the case when a final kill request (i.e., SIGKILL vs. SIGTERM) finds that a process is no longer being managed (i.e., self.has_kernel is None).

Fixes #575

[MSeal]

Thanks. Do we need a patch release real quick for this change to get the improvement out?

[kevin-bates]

Thank you David and Matt.

Do we need a patch release real quick for this change to get the improvement out?

Thanks for asking. Unless there are other changes already in the queue, I don't consider this a "release-triggering" PR. Looking at the set of unreleased PRs, I think #568 might be the only other "fix", but it strikes me as a bit more urgent than this PR.

Or perhaps @ahmadmustafaanis views the priority differently. Ahmad, do you feel this warrants an immediate patch release?

[MSeal]

That's a good point on on #568. I should release for that change given it's been sitting fixed for a bit -- I can kick off a patch release this afternoon with both.