Right now,
getNextQueueEntry() removes the job from the Redis server completely,
deleteEntry() is empty.
So if the script crashes during job processing,
there's no chance to re-queue/bury/re-try the job – it's simply gone.
Other work servers have a dedicated "reserved" state for that,
usually with some short auto-release timeout (TTR).
For a similar effect, we could park the job in the _wq_delayed… list using Redis Transactions, although we need a way to safely delete finished jobs from there again.
Redis Sorted Sets cannot contain duplicate values.
Since we have to support duplicate jobs,
we'll could append a uniqid() to the Job serialization or something.
This would also help with targeted job deletion from a Sorted Set,
which we'll need for our TTR semantics.