A user uploaded a large number of videos, starving the videoscalers/jobrunners of resources. We killed ffmpeg from the top busiest servers, we depooled 3 servers from the videoscaler cluster, and increased their weight on the jobrunner cluster, so to give priority to normal jobs.
https://phabricator.wikimedia.org/P15272 (tx to @Urbanecm)
Impact was rather small, we had a few alerts for failed monitoring checks, and increases in jobqueue's job backlog.