Resolved -
One of three replicas of a supporting service was having intermittent failures and slowness. Automated infrastructure monitoring didn't remove it from the active pool and requests that ended up using it would take very long to complete.
Jan 7, 10:00 CET