See bug 872704 for one possible motivating example. 

It looks like we have two related open bugs:

bug 694611, which talks about BuilderOffline, but seems to have morphed into a ClusterFuzz-specific thing.

bug 647805, which talks about "a large number of machines" going offline.

It's possible that if those two bugs were fixed, and if we had good coverage for when a builder had a lot of pending builds, we'd have sufficient coverage to not need anything further.

But, I don't think we have any of those things, and that seems bad.

Filing against Infra>Platform for initial triage. You could argue that maybe there's some Infra>Client work here as well, but this seems like a core part of the platform quality of service.

Thoughts?