redis@6.1.0 connect() hangs: events.once(socket, 'connect') inside @redis/client registers no listener and never resolves #11196

Summary

On current origin/main (8cfc6455f, which includes #11173 and #11137), redis@6.1.0's client.connect() never completes. The TCP connection opens: redis-server shows the client and ss shows ESTAB. But nothing is ever sent (no HELLO in MONITOR), and neither the connect nor the ready event fires. This happens with createClient({ socket: { host, port } }) and with createClient({ url }). On 2026-09-23, #11170's reporter got through connect / ready / HELLO on an older main plus #11150's fix, so this is a newer stall, earlier in the connect path.

Where it stops

I added trace prints to @redis/client/dist/lib/client/socket.js (RedisSocket#createSocket) in a local copy:

const socket = this.#socketFactory.create();            // net.createConnection(withDefaults)
socket.on('connect', () => console.log('raw connect'));   // added
const p = (0, node_events_1.once)(socket, 'connect');     // node_events_1 = require("node:events")
console.log('listeners now', socket.listenerCount('connect'));
p.then(() => console.log('once resolved'));
await p;
console.log('after await once');

Perry output:

listeners now 1          <- only the raw listener; once() registered nothing
raw connect              <- the socket DOES emit 'connect'
(1s later) connecting=false readyState=open listeners=1

once resolved and after await once never print. events.once(socket, 'connect') returns a promise, but it attaches no connect listener to the socket, so the promise never settles and connect() hangs forever.

What does NOT reproduce it

Each of these connects fine on the same build:

• A package-free net.createConnection + await once(s, "connect"), including redis's exact options (noDelay, keepAlive, keepAliveInitialDelay, timeout: undefined, onread: undefined, readable/writable: true).
• A compilePackages CJS package that mirrors RedisSocket: __importDefault(require("node:net")), a private #createSocketFactory, setTimeout(connectTimeout), the do { try { await this.#createSocket() … } catch … } while loop, and extra loads of node:tls / bare net / dns/promises / diagnostics_channel.

So something in the full @redis/client module graph changes what events.once does for this socket. One candidate is a module patching events / EventEmitter, or once taking its EventTarget branch. I have not isolated it further.

Repro

• Linux x86_64, perry-dev build of 8cfc6455f, PERRY_NO_AUTO_OPTIMIZE=1, with libperry_ext_net.a + libperry_ext_events.a in PERRY_RUNTIME_DIR (the compile routes net / events to those wrappers).
• npm install redis@6.1.0 (prints redis version: 6.1.0), with compilePackages = redis, @redis/client, @redis/bloom, @redis/json, @redis/search, @redis/time-series, cluster-key-slot.
• redis-server 7.0.15.

const { createClient } = require("redis");
console.log("redis version:", require("redis/package.json").version);
async function main() {
  const client = createClient({ socket: { host: "127.0.0.1", port: 6379 } });
  client.on("error", (e: any) => console.log("error event:", e && e.message));
  client.on("connect", () => console.log("event connect"));
  client.on("ready", () => console.log("event ready"));
  await client.connect();
  console.log("SET ->", await client.set("key", "value"));
  console.log("GET ->", await client.get("key"));
  await client.quit();
}
main();
setTimeout(() => { console.log("TIMEOUT"); process.exit(3); }, 6000);

Node 26.5.1: event connect / event ready / SET -> OK / GET -> value. Perry: only the version line, then TIMEOUT.

This is now the redis@6.1.0 blocker. #11170 (PR #11192) fixes the next failure after it: the first command's private-field read in *[Symbol.iterator]().