Eric Lindvall explained some of he reasons why Kestrel was a perfect fit for Papertrail:
Using Kestrel (or any system that doesn't require that all messages are held in RAM) protects you from a runaway producer sending too many messages and causing your system to run out of RAM.
2. Flexibility (both infrastructure and code)
The way Resque includes the class to execute along with the "work to be done" makes it very easy to get off the ground. For many cases where what you want done is some piece of code to just get run outside of the web request thread, it solves the problem perfectly.
What you run into when you get into the realm of processing thousands of these pieces of work per second, is that there are often optimizations that you can do to how things are processed.
The changes may include:
* doing work in batches (it may allow you to cache certain lookups or do things that take round-trips all at once)
* moving your code from being pushed work to pulling work
Having that separation between "the work" and "the code" can also make it simpler to do things like slowly migrate work from an older way of processing the work to a newer way (to allow you to test how well a newer solution works without needing to do a full deploy).
In that article they also compare Kestrel with Resque, which is based on Redis. Kestrel would also be easy to modify for Java devs because it's built using Scala.
There are already two great plugins: Kestrel Overall and Kestrel Queue Monitoring.