Add .sock cleanup to ensure Studio uses new _events after unclean shutdown

[wojtekn]

Related issues

• N/A

How AI was used in this PR

I used it for debugging the issue and to add code that removes socket file.

Proposed Changes

I propose adding .sock cleanup to ensure Studio uses new _events after an unclean shutdown.

In the other case, when an old socket is not cleaned up after an unclean shutdown, Studio will reuse it.

Testing Instructions

• Start Studio, start sites, confirm they are shown as running

Pre-merge Checklist

• Have you checked for TypeScript, React or other console errors?

[fredrikekelund]

Let's do this only on non-Windows platforms. On Windows, SITE_EVENTS_SOCKET_PATH isn't actually a file system path

[wojtekn]

Good point. Thinking about it more, we could move this to eventsSocketServer.listen(), making the cleanup the SocketServer's responsibility rather than the caller's. Do you see anything against that?

[fredrikekelund]

SocketServer::listen already has some clever error-handling logic inspired by pm2-axon. If listening on the socket fails, it'll try connecting as a client to see whether the socket has a server on the other end (if not, then the only connected parties are clients). If it cannot connect, then it'll remove the socket file and try again.

This is why I was slightly surprised that we need this additional layer of defense, but clearly, there are scenarios when there are stray processes that we need to handle.

So, in short, we shouldn't assume that the socket file can always be deleted in SocketServer::listen.

However, I do think SocketServer::close could use additional logic to remove the socket file as well.

[wojtekn]

This logic won't remove the old socket if it can connect to it, and the issue may arise when an outdated server is on the other end.

It makes sense to keep SocketServer::listen generic and keep the brute-force cleanup logic as part of the _event command, since this command never needs to attach to an existing socket.

[wojtekn]

I added a guard for the Windows.

[wpmobilebot]

📊 Performance Test Results

Comparing e8cf0df vs trunk

app-size

Metric	trunk	e8cf0df	Diff	Change
App Size (Mac)	1236.41 MB	1236.44 MB	+0.03 MB	⚪ 0.0%

site-editor

Metric	trunk	e8cf0df	Diff	Change
load	1848 ms	1911 ms	+63 ms	🔴 3.4%

site-startup

Metric	trunk	e8cf0df	Diff	Change
siteCreation	7069 ms	6083 ms	986 ms	🟢 -13.9%
siteStartup	3927 ms	3917 ms	10 ms	⚪ 0.0%

---

Results are median values from multiple test runs.

Legend: 🟢 Improvement (faster) | 🔴 Regression (slower) | ⚪ No change (<50ms diff)

[fredrikekelund]

I think we could use this PR as an opportunity to also remove the socket file in SocketServer::close, but we could also do that separately. This change by itself should fix the issue at hand