In the recent weeks the dispatchers occasionally seemed to stop dispatching to certain wikis for a certain time, leading to huge dispatching backlogs. The symptoms for this are a huge dispatching backlog (mostly on just one wiki, though) and very little CPU usage of the dispatchers on terbium.
See T178624: enwiki dispatch lag is 8 hours for one (rather extreme) example of this.
I looked into this briefly and found T178652, which is possibly related to this, but certainly isn't the root cause.
This also happened yesterday with svwiki and on inspection it seems that a stale lock for that wiki was held with the redis LockManager. In order to rule out a coincidence here, I would like to look at another such example…