4 ms·

Byte-Monkey: Fault injection for the JVM

77 points by probablyfine 10y ago

jdbernard 10y ago
On the GitHub page it lists these modes: Fault: Throw exceptions from methods that declare those exceptions Latency: Introduce latency on method-calls Nullify: Replace the first non-primitive argument to the method with null 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.
- the8472 10y ago
  That could probably be fixed by adding @NotNull annotations to arguments and not injecting in those places.
- probablyfine 10y ago
  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!
  - jpitz 10y ago
    Bytanzine - perturb return values. Perhaps looking at the kinds of transformations that mutation testing frameworks perform would be informative?
  - tlarkworthy 10y ago
    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
  - rzzzt 10y ago
    - Extend the time code spends in a synchronized block - Throw OutOfMemoryException randomly
    - probablyfine 10y ago
      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" :)
tmd83 10y ago
This looks super useful for making things super reliable e.g., your core service kind of thing. Also if you are writing sort of critical high perf libraries/utility like a resource pool, cache, transaction manager .. the latency & fault looks specially good. Nullify looks more interesting as a general code reliability check tool specially with the @NotNull thing another commenter mentioned.
gravypod 10y ago
I've done a bit of monkey patching for the JVM and I know it's quite the pain in the ass. Kudos to you op for making this work. Great job! I might give this a try some time soon.
winteriscoming 10y ago
Haven't tried Byte-Monkey, but have experimented with Byteman http://byteman.jboss.org/ http://byteman.jboss.org/ which I find to be quite handy in debugging the tricky to reproduce issues

77 points by probablyfine 10y ago

On the GitHub page it lists these modes: Fault: Throw exceptions from methods that declare those exceptions Latency: Introduce latency on method-calls Nullify: Replace the first non-primitive argument to the method with null 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.

the8472 10y ago
That could probably be fixed by adding @NotNull annotations to arguments and not injecting in those places.
probablyfine 10y ago
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!
- jpitz 10y ago
  Bytanzine - perturb return values. Perhaps looking at the kinds of transformations that mutation testing frameworks perform would be informative?
- tlarkworthy 10y ago
  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
- rzzzt 10y ago
  - Extend the time code spends in a synchronized block - Throw OutOfMemoryException randomly
  - probablyfine 10y ago
    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" :)

tmd83 10y ago

This looks super useful for making things super reliable e.g., your core service kind of thing. Also if you are writing sort of critical high perf libraries/utility like a resource pool, cache, transaction manager .. the latency & fault looks specially good. Nullify looks more interesting as a general code reliability check tool specially with the @NotNull thing another commenter mentioned.

gravypod 10y ago

I've done a bit of monkey patching for the JVM and I know it's quite the pain in the ass. Kudos to you op for making this work. Great job! I might give this a try some time soon.

winteriscoming 10y ago

Haven't tried Byte-Monkey, but have experimented with Byteman http://byteman.jboss.org/ http://byteman.jboss.org/ which I find to be quite handy in debugging the tricky to reproduce issues