We have a late-init scenario where load balancing isn't available before
configure_proxy is called. Logging this condition to sentry makes these
failures easier to track and diagnose.