Investigate flaky async-hooks.test-improper-unwind

[RaisinTen]

Test

test/async-hooks/test-improper-unwind.js

Platform

AIX

Console output

---
duration_ms: 362.309
exitcode: 1
severity: fail
stack: |-
  node:assert:94
    throw new AssertionError(obj);
    ^

  AssertionError [ERR_ASSERTION]: Expected values to be strictly equal:

  null !== 1

      at ChildProcess.<anonymous> (/home/iojs/build/workspace/node-test-commit-aix/nodes/aix72-ppc64/test/async-hooks/test-improper-unwind.js:59:12)
      at ChildProcess.<anonymous> (/home/iojs/build/workspace/node-test-commit-aix/nodes/aix72-ppc64/test/common/index.js:437:15)
      at ChildProcess.emit (node:events:507:28)
      at maybeClose (node:internal/child_process:1101:16)
      at ChildProcess._handle.onexit (node:internal/child_process:305:5) {
    generatedMessage: true,
    code: 'ERR_ASSERTION',
    actual: null,
    expected: 1,
    operator: 'strictEqual'
  }

  Node.js v25.0.0-pre
...

Build links

• https://ci.nodejs.org/job/node-test-commit-aix/57439/nodes=aix72-ppc64/testReport/junit/(root)/async-hooks/test_improper_unwind/
• https://ci.nodejs.org/job/node-test-commit-aix/57393/nodes=aix72-ppc64/testReport/junit/(root)/async-hooks/test_improper_unwind/

Additional information

No response

[RaisinTen]

cc @nodejs/async_hooks

[Flarna]

This doesn't look like it is related to async hooks.
A child process is spawned but the close event of child process doesn't provide the exit code.
Could be also some issue in child process on aix or some temp resource limitation that child process couldn't be started. properly.

[RaisinTen]

The child_process implementation in

node/lib/internal/child_process.js

Lines 262 to 275 in c969649

	this.signalCode = null;
	this.exitCode = null;
	this.killed = false;
	this.spawnfile = null;
	this._handle = new Process();
	this._handle[owner_symbol] = this;
	this._handle.onexit = (exitCode, signalCode) => {
	if (signalCode) {
	this.signalCode = signalCode;
	} else {
	this.exitCode = exitCode;
	}

shows that initially the signalCode and exitCode properties on the child process object are both set to null.

On exit:

• if signalCode is truthy
  • just the signalCode property is set with the new value
  • and the exitCode property remains null
• otherwise
  • just the exitCode property is set with the new value
  • and the signalCode property remains null

The child process is getting killed with a signal, that's why the 'close' callback is called with a null code and a truthy signal (we don't know what value this holds because it's not being logged). I think the test should log the signal and continue the assertions for now.

[RaisinTen]

Submitted a PR: #58478

[RaisinTen]

I have also created a PR to update the child process docs with this info - #58479

[RaisinTen]

I can confirm that #58478 has fixed this issue.

This flake was appearing in the reliability report 2-4 times daily from 2025-05-23 - nodejs/reliability#1209 till 2025-06-01 nodejs/reliability#1218. The PR landed on 2025-05-29 and after the PRs where CI was being run on included this change, the flake went away in all recent reliability reports.