The service started to become unaccessible due to degraded performance on one of the nodes of our Social Graph database cluster. As soon as the issue was detected, we performed maintenance tasks and were able to recover it and bring the service back to normal.