Shutting down adapter

[michaelzangl]

I tried shutting down socket IO with the adapter installed. The node js process won't exit, since there are still timers active.

I expect this to work:

    socketIo.close();

There are still two timers running afterwards: the 30s cleanup timer and the ~2s reconnect timer

The workaround I found is:

    await socketIo.of('/').adapter.close(); // < this will clean the 30s cleanup timer
    socketIo.close();
    pool.connect = () => new Promise(() => {}); // < this will make the reconnect timer get blocked forever on the first reconnect attempt.

[grobarko]

The fix (socketio/socket.io@bf64870) still isn't working.

Currently I'm using this as a workaround:

await io.of('/').adapter.close();

io.close(() => {
  console.log("Server terminated gracefully 👌");
});
await pool.end();

[darrachequesne]

@grobarko I was not able to reproduce the issue:

import { Server } from "socket.io";
import { createAdapter } from "@socket.io/postgres-adapter";
import pg from "pg";
import process from "node:process";

const PORT = process.env.PORT || 3000;

const pool = new pg.Pool({
  user: "postgres",
  host: "localhost",
  database: "postgres",
  password: "changeit",
  port: 5432,
});

await pool.query(`
  CREATE TABLE IF NOT EXISTS socket_io_attachments (
      id          bigserial UNIQUE,
      created_at  timestamptz DEFAULT NOW(),
      payload     bytea
  );
`);

pool.on("error", (err) => {
  console.error("Postgres error", err);
});

const io = new Server({
  adapter: createAdapter(pool)
});

io.on("connection", (socket) => {
  socket.on("hello", () => {
    // send to anyone except the sender
    socket.broadcast.emit("hello", socket.id, process.pid);
  });
});

io.listen(PORT);
console.log(`server listening on port ${PORT}`);

setTimeout(async () => {
  console.log("STOP !!!");

  io.close();
  await pool.end();
}, 1000);

See example here: https://github.com/socketio/socket.io/tree/main/examples/postgres-adapter-example

Calling io.close() seems sufficient (with pool.end(), but that makes sense since we don't own that object). Could you please check?

[piotr-cz]

This is probably because there is a reconnection timeout scheduled, but not cleared on adapter.close like cleanupTimer is:

socket.io-postgres-adapter/lib/index.ts

Lines 107 to 110 in d38fbda

	const scheduleReconnection = () => {
	const reconnectionDelay = Math.floor(2000 * (0.5 + Math.random()));
	setTimeout(initClient, reconnectionDelay);
	};

Should be fixed with:

  let cleanupTimer: NodeJS.Timer;
+ let reconnectTimer: NodeJS.Timer;

  const scheduleReconnection = () => {
    const reconnectionDelay = Math.floor(2000 * (0.5 + Math.random()));
-   setTimeout(initClient, reconnectionDelay);
+   reconnectTimer = setTimeout(initClient, reconnectionDelay);
  };

  // ...

  adapter.close = () => {
+   if (reconnectTimer) {
+     clearTimeout(reconnectTimer)
+   }
  }

Additionally:

• the adapter.close could set a isClosing flag that should be checked in end listener before scheduling reconnection:

  socket.io-postgres-adapter/lib/index.ts

  Lines 135 to 138 in d38fbda

  	client.on("end", () => {
  	debug("client was closed, scheduling reconnection...");
  	scheduleReconnection();
  	});

• each setTimeout could be unreffed

By the way - I'm not preparing a PR because I'm not using this package and so I could not test the fix

[piotr-cz]

Seems that this might have been fixed in v0.5.0