Investigate flaky parallel/test-child-process-fork-regr-gh-2847

[mscdex]

• Version: master
• Platform: aix
• Subsystem: child_process

This test failure recently occurred on AIX: https://ci.nodejs.org/job/node-test-commit-aix/1317/nodes=aix61-ppc64/console

Output:

not ok 74 parallel/test-child-process-fork-regr-gh-2847
# /home/iojs/build/workspace/node-test-commit-aix/nodes/aix61-ppc64/test/parallel/test-child-process-fork-regr-gh-2847.js:58
#         throw err;
#         ^
# 
# Error: channel closed
#     at ChildProcess.target.send (internal/child_process.js:540:16)
#     at Worker.send (cluster.js:65:28)
#     at Socket.<anonymous> (/home/iojs/build/workspace/node-test-commit-aix/nodes/aix61-ppc64/test/parallel/test-child-process-fork-regr-gh-2847.js:33:14)
#     at Socket.g (events.js:291:16)
#     at emitNone (events.js:86:13)
#     at Socket.emit (events.js:185:7)
#     at TCPConnectWrap.afterConnect [as oncomplete] (net.js:1075:10)
  ---
  duration_ms: 0.380

/cc @mhdawson ?

[gibfahn]

cc/ me, @gireeshpunathil

[santigimeno]

Can reproduce it on FreeBSD too

[gireeshpunathil]

This has a long history, and to put it short: the second send() call is vulnerable due to race condition in the send() function with the server close. Depending on the timing, we will see failure in establishing a connection (line 32), or failure in sending data (line33).

The former seem to have been taken care(line 57), while the failure at send is not.

@santigimeno , what are your thoughts?

[santigimeno]

@gireeshpunathil I've seen a couple of different errors:

• It can happen that the error is not ECONNREFUSED but channel closed. That's the error I think you're describing, and it should be allowed.
• If the second socket connects before the first socket, as the first socket has not defined callback, it will emit an error instead of using the callback, thus causing a failure. This I have tried to fix in test: fix test-child-process-fork-regr-gh-2847 #8954.

I'll add a fixup for the first case to that PR.

[gireeshpunathil]

OK, I missed to see your PR. Yes - your point 1 is what I think has happened in this case. If you are accommodating in the PR great, thanks!