It's important: did the underlying problem get fixed (not displaying an error on a 503) or is it "resolved" only by throwing more hardware/resources at it?
The problem was fixed - we found the root cause of API becoming unresponsive and returning 503 responses and then fixed the issue.
The issue was to do with the averages endpoint re-calculating every time it was called, this ate up alot the resources on the dyno - we've cached this now so this no longer happens.
It is still running on the same resources, no 'throwing' more anything at it at this point.
Issue should now be resolved.