Test runner with isolation=none does not work in child process

Version

v24.9.0

Platform

Linux home 5.15.0-153-generic #163-Ubuntu SMP Thu Aug 7 16:37:18 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux

Subsystem

No response

What steps will reproduce the bug?

runner.js

import cluster from 'node:cluster';
import { run } from "node:test";

if (cluster.isPrimary) {
  cluster.fork();
} else {
  const stream = run({
    isolation: "none",
  });

  stream.on("data", (data) => {
    console.log("on data", data.type);
  });

  stream.on("end", () => {
    console.log("on end")
    process.exit(0);
  });

  stream.on("error", (err) => {
    console.log("on error", err);
  });
}

dummy.test.js

import { describe, it } from "node:test";
import * as assert from "node:assert";

describe("suite", () => {
  it("works", ()  => {
    assert.ok(true);
  });
});

How often does it reproduce? Is there a required condition?

Always. No.

What is the expected behavior? Why is that the expected behavior?

With isolation: "process" (the default), here's the output:

on data test:enqueue
on data test:dequeue
on data test:enqueue
on data test:dequeue
on data test:enqueue
on data test:dequeue
on data test:complete
on data test:start
on data test:start
on data test:pass
on data test:complete
on data test:plan
on data test:pass
on data test:summary
on data test:complete
on data test:plan
on data test:diagnostic
on data test:diagnostic
on data test:diagnostic
on data test:diagnostic
on data test:diagnostic
on data test:diagnostic
on data test:diagnostic
on data test:diagnostic
on data test:summary
on end

Process finished with exit code 0

The process cleanly exits at the end of the tests.

What do you see instead?

With isolation: "none", here's the output:

on data test:enqueue
on data test:dequeue
on data test:enqueue
on data test:dequeue
on data test:complete
on data test:start
on data test:start
on data test:pass
on data test:complete
on data test:plan
on data test:pass

And the process hangs.

Additional information

The goal was to use the "cluster" module to run the tests in parallel with the shard option:

import { run } from "node:test";
import { spec } from "node:test/reporters";
import { PassThrough } from "node:stream";
import { availableParallelism } from "node:os";
import cluster from 'node:cluster';

const concurrency = availableParallelism();
let completed = 0;

if (cluster.isPrimary) {
  const mergedStream = new PassThrough({
    objectMode: true
  });

  cluster.on("message", (_worker, message) => {
    mergedStream.emit("data", message);
  });

  mergedStream
    .compose(spec)
    .pipe(process.stdout);

  for (let i = 1; i <= concurrency; i++) {
    cluster.fork({
      SHARD_INDEX: i,
      SHARD_TOTAL: concurrency,
    });
  }

  cluster.on("exit", (worker, code, signal) => {
    if (++completed === concurrency) {
      mergedStream.end();
    }
  });
} else {
  const stream = run({
    isolation: "none",
    shard: {
      index: parseInt(process.env.SHARD_INDEX, 10),
      total: parseInt(process.env.SHARD_TOTAL, 10),
    },
  });

  stream.on("data", (chunk) => {
    process.send(chunk);
  });
}

Related: https://github.com/nodejs/node/issues/55939

Commencez par la reproduction dans runner.js et dummy.test.js, puis suivez le point d’entrée run() de node:test avec l’isolation définie sur "none" lorsqu’il est invoqué depuis un processus enfant de cluster. Comparez son cycle de vie du flux avec l’isolation par défaut "process" et l’issue associée 55939. C’est terminé lorsque le flux du processus enfant émet les événements restants, se termine et quitte proprement, tout en conservant la prise en charge des shards.