Between 12:07 and 12:30pm PT (UTC-8) today, HTTP event sources and workflows returned 502 / 503 errors intermittently. AWS was performing maintenance on one of our data stores, and the client we use to connect to that data store failed to connect to new nodes when one was taken out of rotation for maintenance. We've downgraded the client to a version that should handle this case gracefully, so that we should not incur downtime in this case again.