Reconnect may lead to crash,cause of race condition

[luoqisheng]

createWebsocketAndConnect called in the engineQueue, which create a new websocket.
while the stream delegate methodopen func stream(_ aStream: Stream, handle eventCode: Stream.Event) was called in another thread( the com.vluxe.starscream.websocket). When race happens , it may lead to crash. the crash log may like this:

[luoqisheng]

Hi, the engineQueue which named com.socketio.engineHandleQueue in SocketEngine.swift should be a static property. If not, when the using forceNew to create a SocketIOClient, reconnecting will create a new engineQueue, and racing condition may happens. like this( a lot of engineQueue):

[nuclearace]

What is the nature of the race? I don't quite see how making it a static property would help. Need to identify where queue access violations are happening. (I see one now)

[luoqisheng]

thanks for your reply! when the using forceNew to create a SocketIOClient, each connecting action will make a new socket engine, and this will create a new queue ( com.socketio.engineHandleQueue), as the picture showing, Thread 55,57,51,49,28,36 accessing _connect func. there is not serial guarantee. making it a static property, com.socketio.engineHandleQueue will be a shared serial queue, would it be better?

[nuclearace]

_connect is a instance method though, so there shouldn't be any shared state between the engines. What is your reconnectWait?

[nuclearace]

Using a shared serial queue would mean that multiple sockets in one app would be sharing a queue. I'm not sure if that would be better. It would limit the number of threads when you have a ton of sockets going at once. But then it would mean that work in one socket would block work in another.

Would have to test these trade-offs.

[nuclearace]

It should also be noted that the client is only queue safe within itself. Accesses to the socket must be done on its handleQueue. Or there's a possibility that races will happen.

[luoqisheng]

if the engine released during the stream callback was fired. the websocket may be a dangling pointer? should I close the stream sync in the queue(com.vluxe.starscream.websocket) before the engine released?