[Zen2] Randomized testing of CoordinationState #32242
Conversation
ywelsch
added
>non-issue
v7.0.0
:Distributed/Cluster Coordination
|
Pinging @elastic/es-distributed |
|
ping @DaveCTurner |
There was a problem hiding this comment.
LGTM. I asked for a couple of tiny changes, but definitely no need for a review.
Seems to reliably achieve at least a handful of commits during the run, which is enouraging for coverage. I've run thousands of repeats without failing an assertion, so maybe the maths works!
| } | ||
| } | ||
|
|
||
| // check system invariants |
There was a problem hiding this comment.
It caught me out that this can happen outside the loop, but on closer inspection the invariants checked are both monotonic so it's sufficient to check them at the end. I think this is worth a comment.
There was a problem hiding this comment.
sure, fixed in 711447b
| @@ -752,6 +759,11 @@ public void testVoteCollection() { | |||
| }); | |||
| } | |||
|
|
|||
| public void testSafety() { | |||
| Cluster cluster = new Cluster(randomIntBetween(1, 5)); | |||
| cluster.runRandomly(10000); | |||
There was a problem hiding this comment.
IntelliJ suggests to inline this constant, and while I'm here I suggest to inline cluster.
| throw new CoordinationStateRejectedException("node not available"); | ||
| } else { | ||
| final Object payload = message.payload; | ||
| if (payload instanceof StartJoinRequest) { |
There was a problem hiding this comment.
Yay for dynamic dispatching ;)
Simulates a random run of a cluster with multiple CoordinationState instances (each representing one node), passing messages back and forth, and asserting that the overall system satisfies a given set of safety properties.
Follow-up to #32171