investigate flaky "sequential/test-http-server-consumed-timeout"

[Trott]

• Version: v9.0.0-pre
• Platform: vs2017,win2016 & rhel72-s390x
• Subsystem: test

https://ci.nodejs.org/job/node-test-binary-windows/9855/RUN_SUBSET=1,VS_VERSION=vs2017,label=win2016/console

not ok 399 sequential/test-http-server-consumed-timeout
  ---
  duration_ms: 1.41
  severity: fail
  stack: |-
    assert.js:48
      throw new errors.AssertionError({
      ^
    
    AssertionError [ERR_ASSERTION]: Request timeout should not fire
        at IncomingMessage.mustNotCall (c:\workspace\node-test-binary-windows\RUN_SUBSET\1\VS_VERSION\vs2017\label\win2016\test\common\index.js:563:12)
        at emitOne (events.js:115:13)
        at IncomingMessage.emit (events.js:210:7)
        at Socket.socketOnTimeout (_http_server.js:389:48)
        at emitNone (events.js:105:13)
        at Socket.emit (events.js:207:7)
        at Socket._onTimeout (net.js:402:8)
        at ontimeout (timers.js:469:11)
        at tryOnTimeout (timers.js:304:5)
        at Timer.listOnTimeout (timers.js:264:5)

[refack]

Showed up on linuxONE

not ok 1661 sequential/test-http-server-consumed-timeout
  ---
  duration_ms: 0.559
  severity: fail
  stack: |-
    assert.js:42
      throw new errors.AssertionError({
      ^
    
    AssertionError [ERR_ASSERTION]: Request timeout should not fire
        at IncomingMessage.mustNotCall (/data/iojs/build/workspace/node-test-commit-linuxone/nodes/rhel72-s390x/test/common/index.js:560:12)
        at emitOne (events.js:115:13)
        at IncomingMessage.emit (events.js:210:7)
        at Socket.socketOnTimeout (_http_server.js:389:48)
        at emitNone (events.js:105:13)
        at Socket.emit (events.js:207:7)
        at Socket._onTimeout (net.js:400:8)
        at ontimeout (timers.js:469:11)
        at tryOnTimeout (timers.js:304:5)
        at Timer.listOnTimeout (timers.js:264:5)
  ...

https://ci.nodejs.org/job/node-test-commit-linuxone/7940/nodes=rhel72-s390x/console

Stress test indicate 0.1% failures - https://ci.nodejs.org/job/node-stress-single-test/1373/nodes=rhel72-s390x/

[BridgeAR]

@nodejs/testing are we just going to live with the test being flaky at times?

[Trott]

Because there's always a chance that a 25ms timeout will take longer than 200ms to actually be invoked, there's always going to be a chance that this test (as written) will fail spuriously. However, I think I see a way to reduce the frequency by an order of magnitude or two...

First, though, CI stress on current master to confirm we still have a measurable problem: https://ci.nodejs.org/job/node-stress-single-test/1427/nodes=rhel72-s390x/console

[Trott]

OK, I'm able to get it to 100% reliability as long as we're OK with skipping the test in unfavorable conditions. See #15688