This is a wrapper over native C# class ClientWebSocket with built-in reconnection and error handling.
MIT
- installation via NuGet (Websocket.Client)
- targeting .NET Standard 2.0 (.NET Core, Linux/MacOS compatible)
- reactive extensions (Rx.NET)
- integrated logging abstraction (LibLog)
var exitEvent = new ManualResetEvent(false);
var url = new Uri("wss://xxx");
using (var client = new WebsocketClient(url))
{
client.ReconnectTimeoutMs = (int)TimeSpan.FromSeconds(30).TotalMilliseconds;
client.ReconnectionHappened.Subscribe(type =>
Log.Information($"Reconnection happened, type: {type}"));
client.MessageReceived.Subscribe(msg => Log.Information($"Message received: {msg}"));
client.Start();
Task.Run(() => client.Send("{ message }"));
exitEvent.WaitOne();
}More usage examples:
Pull Requests are welcome!
In order to set some advanced configuration, which are available on native ClientWebSocket class,
you have to provide factory method as second parameter to WebsocketClient.
That factory method will be called on every reconnection to get a new instance of the ClientWebSocket.
var factory = new Func<ClientWebSocket>(() => new ClientWebSocket
{
Options =
{
KeepAliveInterval = TimeSpan.FromSeconds(5),
Proxy = ...
ClientCertificates = ...
}
});
var client = new WebsocketClient(url, factory);
client.Start;Also you can access current native class via client.NativeClient.
But use it with caution, on every reconnection there will be a new instance.
There is a built-in reconnection which invokes after 1 minute (default) of not receiving any messages from the server. It is possible to configure that timeout via client.ReconnectTimeoutMs. Also, there is a stream ReconnectionHappened which sends information about a type of reconnection. However, if you are subscribed to low rate channels, it is very likely that you will encounter that timeout - higher the timeout to a few minutes or call PingRequest by your own every few seconds.
In the case of remote server outage, there is a built-in functionality which slows down reconnection requests (could be configured via client.ErrorReconnectTimeoutMs, the default is 1 minute).
Beware that you need to resubscribe to channels after reconnection happens. You should subscribe to ReconnectionHappened stream and send subscriptions requests.
Observables from Reactive Extensions are single threaded by default. It means that your code inside subscriptions is called synchronously and as soon as the message comes from websocket API. It brings a great advantage of not to worry about synchronization, but if your code takes a longer time to execute it will block the receiving method, buffer the messages and may end up losing messages. For that reason consider to handle messages on the other thread and unblock receiving thread as soon as possible. I've prepared a few examples for you:
Every subscription code is called on a main websocket thread. Every subscription is synchronized together. No parallel execution. It will block the receiving thread.
client
.MessageReceived
.Where(msg => msg.StartsWith("{"))
.Subscribe(obj => { code1 });
client
.MessageReceived
.Where(msg => msg.StartsWith("["))
.Subscribe(arr => { code2 });
// 'code1' and 'code2' are called in a correct order, according to websocket flow
// ----- code1 ----- code1 ----- ----- code1
// ----- ----- code2 ----- code2 code2 -----Every single subscription code is called on a separate thread. Every single subscription is synchronized, but different subscriptions are called in parallel.
client
.MessageReceived
.Where(msg => msg.StartsWith("{"))
.ObserveOn(TaskPoolScheduler.Default)
.Subscribe(obj => { code1 });
client
.MessageReceived
.Where(msg => msg.StartsWith("["))
.ObserveOn(TaskPoolScheduler.Default)
.Subscribe(arr => { code2 });
// 'code1' and 'code2' are called in parallel, do not follow websocket flow
// ----- code1 ----- code1 ----- code1 -----
// ----- code2 code2 ----- code2 code2 code2In case you want to run your subscription code on the separate thread but still want to follow websocket flow through every subscription, use synchronization with gates:
private static readonly object GATE1 = new object();
client
.MessageReceived
.Where(msg => msg.StartsWith("{"))
.ObserveOn(TaskPoolScheduler.Default)
.Synchronize(GATE1)
.Subscribe(obj => { code1 });
client
.MessageReceived
.Where(msg => msg.StartsWith("["))
.ObserveOn(TaskPoolScheduler.Default)
.Synchronize(GATE1)
.Subscribe(arr => { code2 });
// 'code1' and 'code2' are called concurrently and follow websocket flow
// ----- code1 ----- code1 ----- ----- code1
// ----- ----- code2 ----- code2 code2 ----Using async/await in your subscribe methods is a bit tricky. Subscribe from Rx.NET doesn't await tasks,
so it won't block stream execution and cause sometimes undesired concurrency. For example:
client
.MessageReceived
.Subscribe(async msg => {
// do smth 1
await Task.Delay(5000); // waits 5 sec, could be HTTP call or something else
// do smth 2
});That await Task.Delay won't block stream and subscribe method will be called multiple times concurrently.
If you want to buffer messages and process them one-by-one, then use this:
client
.MessageReceived
.Select(msg => Observable.FromAsync(async () => {
// do smth 1
await Task.Delay(5000); // waits 5 sec, could be HTTP call or something else
// do smth 2
}))
.Concat() // executes sequentially
.Subscribe();If you want to process them concurrently (avoid synchronization), then use this
client
.MessageReceived
.Select(msg => Observable.FromAsync(async () => {
// do smth 1
await Task.Delay(5000); // waits 5 sec, could be HTTP call or something else
// do smth 2
}))
.Merge() // executes concurrently
// .Merge(4) you can limit concurrency with a parameter
// .Merge(1) is same as .Concat() (sequentially)
// .Merge(0) is invalid (throws exception)
.Subscribe();More info on Github issue.
Don't worry about websocket connection, those sequential execution via .Concat() or .Merge(1) has no effect on receiving messages.
It won't affect receiving thread, only buffers messages inside MessageReceived stream.
But beware of producer-consumer problem when the consumer will be too slow. Here is a StackOverflow issue with an example how to ignore/discard buffered messages and always process only the last one.
I do consulting, please don't hesitate to contact me if you have a custom solution you would like me to implement (web, m@mkotas.cz)
Donations gratefully accepted.