partysocket: Unable to force `.reconnect()` after failed connection

[mattrossman]

Use case

I use partysocket to automatically reconnect a flaky WebSocket. After the tab has been inactive for a while, I want to manually reconnect on window focus, even if the automatic retries have been exhausted.

Expected behavior

I expect that calling ws.reconnect() resets the retry count and attempts a connection. This seems to be the intention based on the code here which resets the retryCount to -1 on .reconnect()

partykit/packages/partysocket/src/ws.ts

Lines 330 to 333 in 09c1d1c

	public reconnect(code?: number, reason?: string) {
	this._shouldReconnect = true;
	this._closeCalled = false;
	this._retryCount = -1;

Current behavior

Calling ws.reconnect() does nothing after maxRetries has been exhausted.

This is because calling ._connect() sets ._connectLock = true, but in the early return where maxRetries is exhausted it doesn't release the lock.

partykit/packages/partysocket/src/ws.ts

Lines 439 to 453 in 09c1d1c

	private _connect() {
	if (this._connectLock || !this._shouldReconnect) {
	return;
	}
	this._connectLock = true;
	const {
	maxRetries = DEFAULT.maxRetries,
	connectionTimeout = DEFAULT.connectionTimeout
	} = this._options;
	if (this._retryCount >= maxRetries) {
	this._debug("max retries reached", this._retryCount, ">=", maxRetries);
	return;
	}

I believe the following change could resolve this:

    if (this._retryCount >= maxRetries) {
      this._debug("max retries reached", this._retryCount, ">=", maxRetries);
+     this._connectLock = false;
      return;
    }

[mattrossman]

Note I experience the stuck lock even with the default of infinite retries.

When I configure it to avoid reconnecting by setting very large min/max reconnection delays:

const ws = new WebSocket(url, undefined, {
	minReconnectionDelay: 1_000_000,
	maxReconnectionDelay: 1_000_000,
});

And then call ws.reconnect() after an initial failed connection, nothing happens and I see that ws._connectLock is stuck as true. I'm now sure if my proposed fix addresses that case, maybe there's another contibuting factor to it getting stuck.