ROOT CAUSE
Our inbound proxy/loadbalancer was configured for a small concurrent connection pool for these API paths. At approximately 13:30 UTC, a few high volume paths were brought online that filled this pool and caused requests to queue to a point where we could not catch up. This caused periodic 503 and 429 responses from our API.
RESOLUTION
At approximately 17:45 UTC, the connection pool size was increased to address this issue.
We have raised the severity of the internal alerts we have monitoring these metrics to more quickly identify and resolve future similar events.