test-child-process-fork-dgram Intermittent timeout on AIX

[mhdawson]

• Version: master
• Platform: AIX
• Subsystem: process

Noticed this failure in a recent AIX run when investigating intermittent addon issue:

https://ci.nodejs.org/job/node-test-commit-aix/472/->

test-child-process-fork-dgram failed with a timeout

not ok 55 parallel/test-child-process-fork-dgram
# TIMEOUT

Likely a pre-existing intermittent failure as nothing in the PR looks related to the failure.

[mhdawson]

Will look to mark this as flaky for AIX ASAP

[mhdawson]

From the test description sounds like it is a good candidate for being flaky:

Because it's not really possible to predict how the messages will be
distributed among the parent and the child processes, we keep sending
messages until both the parent and the child received at least one
message. The worst case scenario is when either one never receives
a message. In this case the test runner will timeout after 60 secs
and the test will fail.

[mhdawson]

PR to mark as flaky here: #8274

[gibfahn]

In passing test runs on both AIX and Linux, usually the vast majority of the messages go to the parent, so the test continues until the first message is sent to the child.

Every time the test fails on the community machines, all the messages (1 per ms for 60s, or 60,000 messages) go to the child, which almost certainly means that the parent isn't able to receive messages.

I have been unable to get this to fail on non-community AIX machines (in about 40,000 runs).

[gibfahn]

This change seems to fix the problem, though I'm not sure why it would. I'll investigate further.

diff --git a/test/parallel/test-child-process-fork-dgram.js b/test/parallel/test-child-process-fork-dgram.js
index 5a00dca..eaca6e9 100644
--- a/test/parallel/test-child-process-fork-dgram.js
+++ b/test/parallel/test-child-process-fork-dgram.js
@@ -32,10 +32,10 @@ if (process.argv[2] === 'child') {

       server.on('message', function() {
         process.send('gotMessage');
+        server.close();
       });

     } else if (msg === 'stop') {
-      server.close();
       process.removeListener('message', removeMe);
     }
   });

[mhdawson]

Investigated this as it was leaving lingering processes and did not want to have to clean them up regularly.

There were 2 problems:

1. test left lingering processes on failures
2. no guarantee that the test would pass.

PR to resolve here: #8697

500 runs on AIX without a failure.

@AndreasMadsen as the original author can you take a look at the PR.

[AndreasMadsen]

please see #8549

---

I wrote it back in 2012, and haven't worked in that field since. I really don't feel qualified to review it.

[mhdawson]

Missed #8549 (I must really be blind) so some discussion there as well.