We identified and rolled back the change that caused the outage. We're still investigating the root cause, but right now it looks like a fairly innocent change to enable system and scheduler logging on our GKE clusters resulted in pods not being able to be scheduled. Our apologies for the inconvenience.