Steve Kuznetsov

@droslean I don't think PDB would help here since you don't get a choice as to when spot instances get taken away from you.

Getting the config semantic right might be hard but I'm for it.

I wonder if that merge commit is somehow related to in-repo config fetching? The `upload-pack: not our ref` error is in general just related to GitHub's REST API racing with...

/approve /lgtm /hold /assign @smarterclayton

Isn't that just the grace period though, so if your test exited faster we'd move on to cleanup faster, right?

> Does it hold jobs on the new tip until the old jobs have all been reaped? No

> No, because there's not currently a channel from the pod-deletion code back to the handler manager, so it just sleeps to give the pod-deleter the time it might need....

@smarterclayton this is outstanding for a review

What workflow just deletes a `Pod`?

OK, makes sense. Why not start a watch?