Re:Byte-Monkey: Fault injection for the JVM
by jdbernard on Saturday June 11, 2016 @10:19AM
(#11883526)
On the GitHub page it lists these modes:
The first two seem useful. The last maybe less useful depending on how you design your project. For example, I commonly disallow null to be passed as a method parameter for internal methods. External APIs will null-check, but all internal logic expects to never be passed null. Still might be useful since you can restrict the scope of it's instrumentation on a package level.
[ Parent
| on HN ]
Re:Byte-Monkey: Fault injection for the JVM
by probablyfine on Saturday June 11, 2016 @12:11PM
(#11883961)
Author here - the project did originally just have fault and latency modes, and the last one came out of discussions with a colleague about other failures that might be interesting.
I'm looking for more failure modes to add, so if you can think of any more send them my way!
Re:Byte-Monkey: Fault injection for the JVM
by tlarkworthy on Saturday June 11, 2016 @04:03PM
(#11885075)
Unclean network disconnects. We had to dynamically change iptables to simulate that particular failure case. It's hard to repro but something that does happen in prod and something that can cause mem leaks
Bytanzine - perturb return values.
Perhaps looking at the kinds of transformations that mutation testing frameworks perform would be informative?
- Extend the time code spends in a synchronized block
- Throw OutOfMemoryException randomly
Re:Byte-Monkey: Fault injection for the JVM
by probablyfine on Saturday June 11, 2016 @04:34PM
(#11885207)
Pausing within a synchronized block to exaggerate slowdown sounds really interesting - and would provide a great opportunity to learn about how locking works "under the hood" :)