Ok, turns out the underlying host became heavily loaded which resulted in a lot of i/o and network slowness. We’ve been working to apply better some limits on our hosts which should prevent this issue again but could also result in being unable to schedule a machine of the size you’re requesting (depending on the region).