If I had to guess it’s probably the stickiness of the session (which user is assigned to which server on the cluster) that expires after a certain time which leads to needing a new login or in the case of the error page a CSRF token which isn’t valid on this server of the cluster.