We've got a queue incident on the Mastodon instances.
Some users on plemora pushed HUGE mp4 videos, and the queue was stucked to process them.
That gave processes like 97.2% CPU and 300 Mb RAM running dozens of minute hour each (the server isn't optimized for video rendering).
And so while video convert occurred, all the workers are busy, and we don't have timeline updates.
No data is lost, as all is saved in Redis as jobs to process.