websocket icon indicating copy to clipboard operation
websocket copied to clipboard

[question] concurrent write to websocket connection with mutex.Lock()

Open Jrenk opened this issue 2 years ago • 19 comments

Describe the problem you're having

Using the gorilla/websocket package I sometimes get the following error, despite using mutex.Lock: concurrent write to websocket connection

Versions

  • Go version: 1.18
  • package version: v1.5.0

"Show me the code!"

type WebsocketClient struct {
	websocket   *websocket.Conn
	mutex            sync.Mutex
}

func (websocketClient *WebsocketClient) sendMessage(msg interface{}) error {
	websocketClient.mutex.Lock()
	defer websocketClient.mutex.Unlock()

	return websocketClient.websocket.WriteJSON(msg)
}

In an http handler the websocket clients get created:

ws, err := upgrader.Upgrade(w, r, nil)
...
clients[id] = &WebsocketClient{
    websocket: ws,
}

The sendMessage function gets called from multiple go-routines and the WriteJSON never gets called directly anywhere else. From my understanding the mutex.Lock() should prevent concurrent websocket writes.

Jrenk avatar Nov 18 '22 12:11 Jrenk

The stack trace points to the sendMessage function. As I already mentioned this is the only place where the WriteJSON or any other Write of the gorilla package gets called. So there is no possibility of the application calling WriteJSON without calling mutex.Lock() beforehand. I don't think this issue should be closed.

Jrenk avatar Dec 06 '22 19:12 Jrenk

@Jrenk hello, could you provide a reproducing example/test/benchmark then to make it clear that it's an issue of this package?

FZambia avatar Dec 08 '22 19:12 FZambia

@Jrenk I'm running into a similar situation. I have a loop in a goroutine that writes to an incoming socket connection for the heartbeat response. Before each write, I lock a common mutex and unlock after the write. The issue seems like it may be tied to a race condition as the issue doesn't happen consistently. Did you figure out what was causing your issue?

NomNomCameron avatar Jul 17 '23 16:07 NomNomCameron

@NomNomCameron No I didn't figure it out. The problem happened very rarely and it didn't occur for a very long time. I'm still not sure what causes this sadly.

Jrenk avatar Jul 18 '23 07:07 Jrenk

@Jrenk

Could it be someone pinged your websocket server and the default ping handler tried to write a pong while it was locked?

edit:I might test this later today, we'll see.

ayjayt avatar Jul 20 '23 14:07 ayjayt

I'm imagining a pull request where configuring the websockets server would ask if you want to stick a lock or mutex or something a couple layers deep. Any thoughts on that?

ayjayt avatar Jul 20 '23 14:07 ayjayt

@ayjayt that could actually be the problem. Let me know if you tested this. A lock in the package itself could resolve this as you said.

Jrenk avatar Jul 20 '23 16:07 Jrenk

Yeah, I mean it has to be it. I might test it, but you should use

Conn.SetPingHandler(func(string) error) anyway to protect against race conditions.

Here is what the default handler looks like:

h = func(message string) error {
	err := c.WriteControl(PongMessage, []byte(message), time.Now().Add(writeWait))
	if err == ErrCloseSent {
		return nil
	} else if e, ok := err.(net.Error); ok && e.Temporary() {
		return nil
	}
	return err
}

from: https://github.com/gorilla/websocket/blob/9111bb834a68b893cebbbaed5060bdbc1d9ab7d2/conn.go#L1158

ayjayt avatar Jul 20 '23 16:07 ayjayt

@coreydaley is this not a bug? If I'm correct, anyone who uses any *Write*() function at anytime might collide with the WriteControl() called by the default handler and produce a concurrent write error. I am not confident in this!

edit: especially since most people are going to run their read loops and write loops in separate go routines, and the default handler is actually called automatically by NextReaders()'s advanceFrame() call

ayjayt avatar Jul 20 '23 16:07 ayjayt

From Gorilla WebSocket docs:

Connections support one concurrent reader and one concurrent writer.

Applications are responsible for ensuring that no more than one goroutine calls the write methods (NextWriter, SetWriteDeadline, WriteMessage, WriteJSON, EnableWriteCompression, SetCompressionLevel) concurrently and that no more than one goroutine calls the read methods (NextReader, SetReadDeadline, ReadMessage, ReadJSON, SetPongHandler, SetPingHandler) concurrently.

The Close and WriteControl methods can be called concurrently with all other methods.

As I mentioned above - if you think it's a bug in a package you better provide a reproducer instead of guessing. You can use https://pkg.go.dev/golang.org/x/tools/cmd/stress tool to increase the chance of finding the reliable reproducer.

FZambia avatar Jul 20 '23 17:07 FZambia

[redacted by moderator]

ayjayt avatar Jul 20 '23 17:07 ayjayt

anyone who uses any Write() function at anytime might collide with the WriteControl() called by the default handler and produce a concurrent write error.

The websocket connection uses a mutex (implemented using a channel) to prevent WriteControl from colliding with the *Write* methods. The websocket connection write methods acquire the mutex before executing any write related methods on the network connection (SetWriteDeadline and Write).

(Why the websocket connection uses a channel instead of a sync.Mutex is a puzzle for another day.)

ghost avatar Jul 21 '23 05:07 ghost

Please keep all comments and conversations civil, please keep in mind that Github users are from all over the world and speak hundred of different languages, and their native languages are not english, but they are being gracious enough to use english here. So do not assume that anything is being said with a specific tone or meaning. Disrespectful language will not be tolerated in this community.

We are almost done with the project transition and then we will begin triaging and working on issues, but we are not quite to that point yet. If you think that you have found a bug and would like to submit a pull request, please do so and we will review it as part of triaging/addressing this issue.

coreydaley avatar Jul 21 '23 15:07 coreydaley

If I'm going to be chastised for asking someone to speak respectfully to me, I will simply not participate here further.

ayjayt avatar Jul 21 '23 16:07 ayjayt

If I'm going to be chastised for asking someone to speak respectfully to me, I will simply not participate here further.

That is your prerogative. This is not the first comment within your short time of contributing to this project that could be deemed disrespectful to someone who has disagreed with you or questioned you: https://github.com/gorilla/websocket/issues/841#issuecomment-1644901325

As far as I am concerned that is two strikes. One more and we will have to discuss blocking you from interacting with the organziation.

coreydaley avatar Jul 21 '23 16:07 coreydaley

@e9x That sort of comment will not be tolerated. I sincerely hope that you are able to find help that allows you to be a more civil member of society.

jaitaiwan avatar May 05 '24 01:05 jaitaiwan

Not that it will do any good but I reported the comment from @e9x, not sure what I did to elicit such a comment.

coreydaley avatar May 05 '24 02:05 coreydaley