Open handle when using http.Server.listen

[remcohaszing]

🐛 Bug Report

I think I ran into forwardemail/supertest#520. Trying to work around this, I got to a minimal example, which I think is actually a bug in Jest.

Jest detects open handles when an http server is started, even when it is closed afterwards.

To Reproduce

Create the following zero-configuration test case.

const http = require("http");

it("should not timeout", async () => {
  const server = http.createServer();
  await new Promise(resolve => {
    server.listen(() => {
      resolve();
    });
  });
  await new Promise((resolve, reject) => {
    server.close(err => {
      if (err) {
        reject(err);
      } else {
        resolve();
      }
    });
  });
  // await new Promise(resolve => setTimeout(resolve, 0));
});

When jest is run using --detectOpenHandles, this will output the following:

$ jest --detectOpenHandles
 PASS  __tests__/index.js
  ✓ should not timeout (6ms)

Test Suites: 1 passed, 1 total
Tests:       1 passed, 1 total
Snapshots:   0 total
Time:        1.231s
Ran all test suites.

Jest has detected the following 1 open handle potentially keeping Jest from exiting:

  ●  TCPSERVERWRAP

      4 |   const server = http.createServer();
      5 |   await new Promise(resolve => {
    > 6 |     server.listen(() => {
        |            ^
      7 |       resolve();
      8 |     });
      9 |   });

      at listen (__tests__/index.js:6:12)
      at Object.<anonymous>.it (__tests__/index.js:5:9)

However, when the “0 ms sleep” line at the end is uncommented, no open handles are detected.

Expected behavior

Jest should not detect any open handles.

Run npx envinfo --preset jest

  System:
    OS: Linux 4.15 Ubuntu 18.04.2 LTS (Bionic Beaver)
    CPU: (8) x64 Intel(R) Core(TM) i7-4720HQ CPU @ 2.60GHz
  Binaries:
    Node: 10.16.0 - /usr/bin/node
    Yarn: 1.16.0 - /usr/bin/yarn
    npm: 6.9.0 - ~/.local/bin/npm
  npmPackages:
    jest: ^24.8.0 => 24.8.0 

[EasonWang01]

Close the listening instance can solve the problem, hope this help.

server.js

const server = app.listen(port, function (err) {
  if (err) {
    console.log(err);
  } else {
    console.log('Listening on port ' + port);
  }
});
module.exports = {
  server
};

test.js

const { server } = require('../index');
afterAll(async () => {
  server.close();
});

[gilles-yvetot]

@EasonWang01 I am doing this but still got the same error

My only way of solving that was to do this:

afterAll(done => {
    httpServer.close(() => {
      setTimeout(done, 100)
    })
  })

[dotku]

I have tried all the solutions above, I still get the error:

Jest has detected the following 1 open handle potentially keeping Jest from exiting

anyone with good luck?

[remcohaszing]

I switched from supertest to axios-test-instance (written by me). I haven’t ran into this issue since then.

[nhebling]

Somehow I run into the problem that jest "detects" or "has" open handles when using mongo-memory-server and supertest.
I tried various options and for the mongo-memory-server I'm able to just avoid the one test where this problem occurs, but for the tests with supertest it's not possible.
Selecting another framework then supertest, could be a temporary solution, but then I run into the problem at another point. Calling "done" (jest) or "end" (supertest) or completly rewritting the test does not help.
Currently I miss some documentation or feature to help me identifying the root cause why jest does not finish the test. Proposals like "just call --forceExit" or "use another framework" do not solve the root cause.

With my supertest problem, e.g. I tried calling "end" but then supertest says that it's already closed - but jest still thinks there is something open - I just want to understand at which point in time or which part is having the problem. So my "wish" would be to add some support in jest to be able dive deeper into the details.

[Mr0grog]

Ran into this today and spent some time diving in — it looks like there are 3 intertwined issues here, some in Jest and some in Supertest:

1. Supertest does not wait for the server is starts to begin listening. It turns out that, if you don’t wait for the callback in server.listen(), the server resources will not have been entirely torn down by the time you receive the callback from server.close() (which Supertest does wait for). This was news to me, and is why you need to add the somewhat long timeout.

2. If an async resource is destroyed (e.g. closing the server) in the same JS task (even if different microtask, it seems), the async_hooks module won’t have time to call its hooks. If you rsolve issue (1) above, for example, Jest still needs a timeout to catch that the server was destroyed (but it can be an ultra-short timeout of 0 ms). It might be good if Jest waited a tick either when reporting/cleaning up the async hooks: https://github.com/facebook/jest/blob/64d5983d20a628d68644a3a4cd0f510dc304805a/packages/jest-core/src/collectHandles.ts#L103-L113

   Or immediately before calling processResults() in runJest: https://github.com/facebook/jest/blob/64d5983d20a628d68644a3a4cd0f510dc304805a/packages/jest-core/src/runJest.ts#L279-L290

3. Jest has a false negative for --detectOpenHandles when an open handle was created in a test function that uses a done callback. So you don’t see this warning when using a done function instead of promises, but not because the resource isn’t still open. Jest tries to determine whether to track an async handle based on the call stack where it was opened: https://github.com/facebook/jest/blob/64d5983d20a628d68644a3a4cd0f510dc304805a/packages/jest-core/src/collectHandles.ts#L18-L38

   But if the function has a done callback, it shows up anonymously in the callstack and so doesn’t register as something from which open async handles should be tracked: https://github.com/facebook/jest/blob/bc50e7f360ab1845abbaa0b3ad788caead0d3174/packages/jest-jasmine2/src/jasmineAsyncInstall.ts#L111-L113

   As opposed to when it uses a promise, and is given a name (asyncJestTest) that shows up in the call stack and is checked for in the stackIsFromUser shown above: https://github.com/facebook/jest/blob/bc50e7f360ab1845abbaa0b3ad788caead0d3174/packages/jest-jasmine2/src/jasmineAsyncInstall.ts#L153

(2) seems like the most immediately applicable thing here (wait a tick for any handles that are scheduled for destruction to actually be destroyed).

(3) seems like it should be fixed, too, although it doesn’t really address the immediate problem here (instead, it would just make this warning show up more).

[nfantone]

I stumbled upon this today. Oddly enough, in my case, it doesn't always happen. If using supertest, the infamous TCPSERVERWRAP error would only sometimes happen, seemingly at random. However, --detectLeaks fails consistently for me (see forwardemail/supertest#520 (comment)).

[Mr0grog]

FWIW, the --detectOpenHandles cases are 100% reproducible for me with current versions of Jest and Supertest, and I feel like I've spent enough time digging under the hood to feel fairly confident I fully understand the issue (if it's unpredictable for you, it could be related to extra, non-predictable time an after* hook, a Jest plugin, or something in Express or your app takes). I haven't looked at --detectLeaks at all, though, so I don't have any useful advice on what might be going wrong there.