http: fix keylog listener setup on existing agent sockets

[shani-singh1]

maybeEnableKeylog() runs as the agent's 'newListener' handler and attaches the agent's keylog handler to the sockets the agent already owns:

// Existing sockets will start listening on keylog now.
const sockets = ObjectValues(this.sockets);
for (let i = 0; i < sockets.length; i++) {
  sockets[i].on('keylog', this[kOnKeylog]);
}

agent.sockets maps a name to an array of sockets, so ObjectValues() yields arrays, not sockets, and .on() is called on an array. Agent.prototype.destroy() in the same file gets this right with a nested walk over both maps.

Two things follow. Adding a 'keylog' listener to an agent that already owns a socket throws, and because the throw happens inside the 'newListener' handler it propagates out of agent.on() before the listener is stored, so the caller gets an exception and no listener. Separately, agent.freeSockets is never visited, so idle keep-alive sockets never start listening even once the crash is out of the way.

Reproduction

const http = require('http');

const server = http.createServer((req, res) => setTimeout(() => res.end('ok'), 300));

server.listen(0, '127.0.0.1', () => {
  const agent = new http.Agent({ keepAlive: true });
  const req = http.get({ host: '127.0.0.1', port: server.address().port, agent }, (res) => res.resume());
  req.on('error', () => {});
  req.on('socket', () => setImmediate(() => {
    console.log('bucket is an Array :', Array.isArray(Object.values(agent.sockets)[0]));
    try {
      agent.on('keylog', () => {});
      console.log('agent.on("keylog") : OK');
    } catch (err) {
      console.log('agent.on("keylog") : THREW ' + err.constructor.name + ': ' + err.message);
    }
    console.log('keylog listeners   :', agent.listenerCount('keylog'), '(expected 1)');
    agent.destroy();
    server.close();
    process.exit(0);
  }));
});

On v24.11.1:

bucket is an Array : true
agent.on("keylog") : THREW TypeError: sockets[i].on is not a function
keylog listeners   : 0 (expected 1)

The documented 'keylog' event on https.Agent is therefore unusable on any agent that has already opened a socket, which is the normal case for a long-lived agent. The existing coverage in test/parallel/test-https-agent-keylog.js registers the listener on https.globalAgent before any request is made, so agent.sockets is empty and the loop body never runs.

Change

Walk both maps the same way Agent.prototype.destroy() does.

Verification

Running the old and new loop bodies against a real http.Agent holding one idle socket and one in-flight socket:

sockets in freeSockets : 1
sockets in sockets     : 1
total existing sockets : 2
OLD loop               : THREW TypeError: sockets[i].on is not a function
NEW loop               : no throw
sockets wired by NEW   : 2 / 2  (expected all)

The added test uses two servers so the sockets get different names and one stays parked in freeSockets rather than being reused. It fails on main with the TypeError above at the agent.on('keylog', ...) line, and passes with this change.

[nodejs-github-bot]

Review requested:

• @nodejs/http
• @nodejs/net

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.29%. Comparing base (51c0d94) to head (898e685).
⚠️ Report is 3 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #65066      +/-   ##
==========================================
- Coverage   90.30%   90.29%   -0.02%     
==========================================
  Files         759      759              
  Lines      247644   247651       +7     
  Branches    46687    46688       +1     
==========================================
- Hits       223645   223608      -37     
- Misses      15474    15514      +40     
- Partials     8525     8529       +4     

Files with missing lines	Coverage Δ
lib/_http_agent.js	96.59% <100.00%> (+0.46%)	⬆️

... and 32 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[pimterry]

Nice work, thanks for fixing this @shani-singh1!

Commit linting is failing as there's no sign-off trailer, so you'll need to git commit --amend -s and then force push to update it.

[shani-singh1]

Thanks for the review!

Amended with -s and force pushed, so the head is now 9058cd0 and carries the sign-off trailer. I applied the same to my three other open PRs (#65064, #65065, #65067) since they would have hit the same lint failure once CI ran on them.