Swiss Army Knife for SSE in Golang
This library is module-ready and versioned symantically. master
branch tracks the latest unstable work, and the vN
branches track stable releases.
So you want to publish events to client that connect to your server?
func main() {
stream := eventsource.NewStream()
go func(s *eventsource.Stream) {
for {
time.Sleep(time.Second)
stream.Broadcast(eventsource.DataEvent("tick"))
}
}(stream)
http.ListenAndServe(":8080", stream)
}
The Stream
object implements an http.Handler
for you so it can be registered directly to a server. Broadcast events to it and it'll forward them to every client that's connected to it.
DataEvent
is shorthand for creating a new *Event
object and assigning Data()
to it
You got it! What do you need?
We call them "topics" but the gist is the same. All Clients always receive Broadcast
events, but you can Publish
events to a specific topic, and then only clients that have Subscribe
d to that topic will receive the event.
stream.Subscribe("weather", myClient)
stream.Publish("weather", weatherUpdateEvent)
stream.Broadcast(tornadoWarningEvent)
You can also just create multiple Stream
objects much to the same effect, then only use Broadcast
. Streams are cheap and run no background routines, so this is a valid pattern.
weatherStream := eventstream.NewStream()
lotteryStream := eventstream.NewStream()
weatherStream.Register(clientPlanningHikes)
lotteryStream.Register(soonToBePoorClient)
lotteryStream.Broadcast(everyoneLosesEvent)
Stream
implements an http.Handler
but by default just registers clients for broadcasts.
Use TopicHandler
to create another handler for that stream that will subscribe clients to topics as well as broadcasts.
stream := eventsource.NewStream()
catsHandler := stream.TopicHandler([]string{"cat"})
http.ListenAndServe(":8080", catsHandler)
Use the stream's Register
, Subscribe
, Remove
, Unsubscribe
, and CloseTopic
functions to control which clients are registered where.
Register a callback for the Stream
to invoke every time a new client connects with Stream.ClientConnectHook
. It'll give you a handle to the resulting Client and the http request that created it, letting you do whatever you please.
stream := eventsource.NewStream()
stream.ClientConnectHook(func(r *http.Requset, c *eventsource.Client){
fmt.Println("Recieved connection from", r.Host)
fmt.Println("Hate that guy")
stream.Remove(c)
c.Shutdown()
})
The callback will be on the same goroutine as the incoming web request that created it, but the Client is live and functioning so it'll start receiving broadcasts and publications immediately before your callback has returned.
If you want to know about errors that occurring when the Stream
tries to Send
to individual clients (which will generally be disconnects), use the Stream.Errors
to create a channel that will deliver them as they happen. The error stream is buffered, but if errors are created faster than they are produced, overflow to the buffer is silently discarded.
The stream's Shutdown
command will unsubscribe and disconnect all connected clients. However the Stream
itself is not running any background routines, and may continue to register new clients if it's still registered as an http handler.
Fine! The Stream
object is entirely convenience. It runs no background routines and does no special handling. It just adds the topics abstraction and calls NewClient
for you when it's connected to. Feel free not to use it.
You betcha.
Clients have to be created off an http.ResponseWriter
that supports the http.Flusher
and http.CloseNotifier
interfaces. When creating a client, callers can optionally also pass the original http.Request
being served, which helps determine which headers are appropriate to send in response.
NewClient
does kick off a background routine to handle sending events, so constructing an object literal will not work. This is done because it's assumed you will likely be calling NewClient
on an http handler routine, and will likely not be doing any interesting work on that routine.
func ServeHTTP(w http.ResponseWriter, r *http.Request) {
client := eventsource.NewClient(w, r)
if client == nil {
http.Error(...)
return
}
client.Wait()
}
Letting the http handler routine that created the Client return may cause the underlying connection to be closed by the server. Since NewClient
does not block, use Wait
to block until the client is shutdown.
The client's Shutdown
function terminates the background routine and marks the client as closed. It does not actually sever the connection. It does unblock any routines waiting on Wait
, which assuming the main http handler routine was waiting there, will cause the connection to close as it returns.
Attempts to Send
events to a client after it has been shutdown will result in an error
Events are the most critical part of the library, and are the most versatile.
Events implement the io.ReadWriter
interface so that data can be written to it from practically any source. However the Write
interface does not write an entire event in wire format. It writes the provided buffer into data:
sections in the resulting event.
ev := &eventsource.Event{}
io.WriteString(ev, "This is an event! How exciting.")
fmt.Fprintf(ev, "This is the %d time I've had to update this readme...", 42)
If you'd like to write bytes exactly as you'd like them to be written to the wire, use the WriteRaw
function.
Read
on the other hand, does return the event as it would have been written to the wire.
You can use Read
and WriteRaw
to create a perfect deep copy, however since this writes to the underlying data buffer and not to the "assembly area", any calls to Data
, ID
, Type
etc will clobber the buffer and not give you the expected result.
evData, _ := ioutil.ReadAll(oldEvent) // full wire format
newEvent.Write(evData) // ... smushed into the `data:` section. Not what you wanted
io.Copy(newEvent, oldEvent) // still not what you wanted
newEvent.WriteRaw(evData) // that will work
newEvent.AppendData("Moar") // ... and you ruined it
Use Clone
to create a perfect deep copy that survives further mutation. Though this is less efficient in memory.
Since you will probably be creating more than just a few events, the EventFactory
interface and a couple helper factories and functions have been provided to speed things up.
// Create events of the same type
typeFact := &eventsource.EventTypeFactory{
Type: "message",
}
// then with incrementing ID's
idFact := &eventsource.EventIdFactory{
Next: 0,
NewFact: typeFact,
}
// then generate as many events as you want with
// type: message, and an ID that increments
ev := idFact.New()