investigate flaky parallel/test-debug-signal-cluster on SmartOS

[Trott]

https://ci.nodejs.org/job/node-test-commit-smartos/4513/nodes=smartos14-64/console:

not ok 213 parallel/test-debug-signal-cluster
# got pids [62039,62043,62044]
# 
# assert.js:85
#   throw new assert.AssertionError({
#   ^
# AssertionError: test timed out
#     at Object.exports.fail (/home/iojs/build/workspace/node-test-commit-smartos/nodes/smartos14-64/test/common.js:443:10)
#     at Timeout.testTimedOut [as _onTimeout] (/home/iojs/build/workspace/node-test-commit-smartos/nodes/smartos14-64/test/parallel/test-debug-signal-cluster.js:69:10)
#     at Timer.unrefdHandle (timers.js:460:14)
# > all workers are running
# > Starting debugger agent.
# > Debugger listening on 127.0.0.1:12646
# > Starting debugger agent.
# > Debugger listening on 127.0.0.1:12648
# > Starting debugger agent.
  ---
  duration_ms: 7.165

[cjihrig]

cc: @nodejs/platform-solaris

[geek]

@cjihrig thanks, I will take a look

[misterdjules]

Adding some context which might be helpful.

I had done some work to make this test non-flaky at #8568. In one of the comments of that PR, there are links to stress tests that passed on SmartOS for this test, so it's definitely surprising to see that test fail after my changes landed.

[Trott]

Let's see if a stress test turns anything up on master now:

• https://ci.nodejs.org/job/node-stress-single-test/nodes=smartos14-32/962/console
• https://ci.nodejs.org/job/node-stress-single-test/nodes=smartos14-64/963/console

If it turns up nothing, it may mean that the stress test is not a reliable indicator for this particular case. (Uh, or that something else caused the problem above and it's not directly an issue with the test?)

If it turns up something, it may be useful to run the stress tests again on the commit from @misterdjules that landed to fix this. If that comes up clean, a bisect might be in order to find out where things broke again and hopefully why.

[misterdjules]

Stress test on master failed for the 64 bits build, started a new stress tests run against the changes that I had landed to fix that problem.

[misterdjules]

Stress tests against #8568 passed, so it seems a regression might have been introduced between these changes and current master. It's also possible that, if there was a gap between running stress tests for #8568 and landing its changes, some changes landed in between that made this test flaky again.

[Trott]

Debugger is going away on master in the foreseeable future. This is unlikely to be addressed. Closing.