The service that allocates jobs to the build farm experienced an unexplained fault (authentication failures to another service that was operating correctly). Once this service was restarted, builds were then allocated correctly.

The recovery period was longer than normal due to the monitoring system not alerting support via the correct channel (i.e. page).

We've increased the alert level so that we are paged more quickly to resolve this issue (builds queueing and not being allocated) if it occurs again.