From 8:38pm EDT to 8:50pm EDT, we were serving 503s for around 75% of our incoming requests. This was purely due to a high burst of traffic that our scaling mitigated (albeit not quickly enough!) We're going to look into the problematic routes and harden their performance.