test: fix flaky test-dgram-exclusive-implicit-bind

[Trott]

Checklist

• make -j4 test (UNIX), or vcbuild test nosign (Windows) passes
• commit message follows commit guidelines

Affected core subsystem(s)

test dgram

Description of change

test-dgram-exclusive-implicit-bind is written assuming that dgram
messages are received with 100% reliability. While missing a dgram
message sent to localhost is rare, we do see it as evidenced by CI
failures from time to time.

The test has been rewritten to send dgram messages over and over until
the test requirements have been met.

Additional incidental refactoring includes:

• var -> const
• use of common.mustCall() instead of exit listener + boolean

[Trott]

Sample CI failure indicating this test was flaky:

https://ci.nodejs.org/job/node-test-commit-linux/6627/nodes=fedora24/console

not ok 1295 parallel/test-dgram-exclusive-implicit-bind
  ---
  duration_ms: 60.62
  severity: fail
  stack: |-
    timeout

[Trott]

CI: https://ci.nodejs.org/job/node-test-pull-request/5349/

[Trott]

Could try a stress test, but I don't think this happens with enough frequency for that to be useful without a very large number of runs. Here's master on fedora24 run 9999 times. We'll see if it fails at all. If so, we can do another run with this PR and see if it comes up clean.

https://ci.nodejs.org/job/node-stress-single-test/1070/nodes=fedora24/console

[cjihrig]

Could all of this just be simplified to:

setInterval(() => {
  source.send(Buffer.from('abc'), 0, 3, common.PORT, '127.0.0.1');
}, 1).unref();

I think you shouldn't have to worry about clearing the interval if you unref it. That would reduce some complexity.

[Trott]

I was concerned that competing 1ms interval timers on some operating systems could result in increased flakiness, but I didn't actually test it, so yeah, let's try that. :-D

[Trott]

If I don't keep the timer-clearing logic, this happens:

Error: Not running
    at Socket._healthCheck (dgram.js:527:11)
    at Socket.send (dgram.js:347:8)
    at Timeout.setInterval [as _onTimeout] (/Users/trott/io.js/test/parallel/test-dgram-exclusive-implicit-bind.js:104:10)
    at ontimeout (timers.js:365:14)
    at Timer.unrefdHandle (timers.js:471:5)
dgram.js:527
    throw new Error('Not running'); // error message from dgram_legacy.js
    ^

[Trott]

Updated to use setInterval() per suggestion from @cjihrig.

CI again: https://ci.nodejs.org/job/node-test-pull-request/5352/

[santigimeno]

From a previous comment I'm not sure this was a problem but can't you get rid of this as the interval is already unrefed? At least locally it works for me

[Trott]

If I remove that line, the test still passes, but it produces a messy output that looks like it's failing:

$ ./node test/parallel/test-dgram-exclusive-implicit-bind.js
dgram.js:527
    throw new Error('Not running'); // error message from dgram_legacy.js
    ^

Error: Not running
    at Socket._healthCheck (dgram.js:527:11)
    at Socket.send (dgram.js:347:8)
    at Timeout.setInterval [as _onTimeout] (/Users/trott/io.js/test/parallel/test-dgram-exclusive-implicit-bind.js:104:10)
    at ontimeout (timers.js:365:14)
    at Timer.unrefdHandle (timers.js:471:5)
dgram.js:527
    throw new Error('Not running'); // error message from dgram_legacy.js
    ^

Error: Not running
    at Socket._healthCheck (dgram.js:527:11)
    at Socket.send (dgram.js:347:8)
    at Timeout.setInterval [as _onTimeout] (/Users/trott/io.js/test/parallel/test-dgram-exclusive-implicit-bind.js:104:10)
    at ontimeout (timers.js:365:14)
    at Timer.unrefdHandle (timers.js:471:5)
$

I'd prefer the processes clean up after themselves and not generate irrelevant errors like that, even if the errors don't cause the test to fail, so I'd prefer to keep the clearInterval() call.

[santigimeno]

Understood

[santigimeno]

LGTM with a suggestion / question

[Trott]

CI: https://ci.nodejs.org/job/node-test-pull-request/5359/

[italoacasas]

Landed 3ddbea6