socket.end() in handleUpgrade (server.js) cause "upstream prematurely closed connection while reading response header from upstream" in nginx

[beatjoerg]

If there is an upgrade Request with an invalid sid the current implementation closes the connection immediatly. Which cause an error on nginx ("upstream prematurely closed connection while reading response header from upstream"). If I comment out the socket.end() it returns a 400 Header, which I expect. I have other failuers with the same error, but did not find yet the cause.
Nginx Proxy Headers are set correctly.

Server.prototype.handleUpgrade = function(req, socket, upgradeHead){
  this.prepare(req);

  var self = this;
  this.verify(req, true, function(err, success) {
    //if (!success) {
    //  socket.end();
    //  return;
    //}

    var head = new Buffer(upgradeHead.length);
    upgradeHead.copy(head);
    upgradeHead = null;

    // delegate to ws
    self.ws.handleUpgrade(req, socket, head, function(conn){
      self.onWebSocket(req, conn);
    });
  });
};

[cheshun]

Is this problem solved?

[ookke]

I'm also seeing similar intermittent 502's with the same nginx log message with engine.io 1.5.1 so I guess this issue is still valid.

[beatjoerg]

Yes, we still see randomly 502, even with the patch above applied.
I will look into this issue more deeply as soon as I can allocate a bit of time.

[andreassemborg]

I have the same problem.

[ryoqun]

Maybe we're affected this issue too.

[zcstarr]

Actually , I thought I had this issue earlier, then I put some debuggers in the Server code
/socket.io/node_modules/engine.io/lib/server.js specifically @ the server upgrade event handling, I realized that nginx was trying to upgrade an already upgraded web socket connection. There's no handler for that, so the solution in the Server.js code is to eventually time out and drop the socket connection , which triggered everyone's favorite 502, upstream prematurely closed connection while reading response header from upstream. Hope this helps !

[ryoqun]

@zcstarr Hmm, that's interesting info. Thanks for the tip!

[linus-amg]

i have that problem aswell, 1 machine using nginx to proxy 10 different node processes, now i want every node process to host a little socket.io server but i can only have 1 running at the same time.

all the others return 502 Bad Gateway

[julianlam]

👍 still happens from what we can tell

[julianlam]

@zcstarr If this is indeed the cause, then the underlying issue is that the client is trying to upgrade the connection twice, though I can't fathom a reason as to why that would happen.

If the connection is already upgraded, then the repeated call should also be answered with the expected HTTP/1.1 101 Switching Protocols, right?

[julianlam]

@rauchg Can we get this addressed please? The errors propagated up to nginx cause it to consider the app "down", when it isn't.

[darrachequesne]

@julianlam I didn't see any PR open for that matter, did you find a solution?

[mQckingbird]

Any updates on this?