One of the nodes of our Identity Provider experienced problems but traffic was still forwarded to it, due to an incorrect configuration of our load balancer.
The problem was solved by manually restarting the failing node, and fixing the configuration so that an unhealthy node doesn't receive traffic anymore.