Table of Contents
In a standard HTTP request-response scenario a client opens a connection, sends a HTTP request to the server (for
example a HTTP GET
request), then receives a HTTP response back and the server closes the connection once
the response is fully sent/received. The initiative always comes from a client when the client
requests all the data. In contrast, Server-Sent Events (SSE) is a mechanism that allows server
to asynchronously push the data from the server to the client once the client-server connection is established by the
client. Once the connection is established by the client, it is the server who provides the data and decides
to send it to the client whenever new "chunk" of data is available. When a new data event occurs on the server,
the data event is sent by the server to the client. Thus the name Server-Sent Events. Note that at high level there
are more technologies working on this principle, a short overview of the technologies supporting server-to-client
communication is in this list:
With polling a client repeatedly sends new requests to a server. If the server has no new data, then it send appropriate indication and closes the connection. The client then waits a bit and sends another request after some time (after one second, for example).
With long-polling a client sends a request to a server. If the server has no new data, it just holds the connection open and waits until data is available. Once the server has data (message) for the client, it uses the connection and sends it back to the client. Then the connection is closed.
SSE is similar to the long-polling mechanism, except it does not send only one message per connection. The client sends a request and server holds a connection until a new message is ready, then it sends the message back to the client while still keeping the connection open so that it can be used for another message once it becomes available. Once a new message is ready, it is sent back to the client on the same initial connection. Client processes the messages sent back from the server individually without closing the connection after processing each message. So, SSE typically reuses one connection for more messages (called events). SSE also defines a dedicated media type that describes a simple format of individual events sent from the server to the client. SSE also offers standard javascript client API implemented most modern browsers. For more information about SSE, see the SSE API specification.
WebSocket technology is different from previous technologies as it provides a real full duplex connection. The initiator is again a client which sends a request to a server with a special HTTP header that informs the server that the HTTP connection may be "upgraded" to a full duplex TCP/IP WebSocket connection. If server supports WebSocket, it may choose to do so. Once a WebSocket connection is established, it can be used for bi-directional communication between the client and the server. Both client and server can then send data to the other party at will whenever it is needed. The communication on the new WebSocket connection is no longer based on HTTP protocol and can be used for example for for online gaming or any other applications that require fast exchange of small chunks of data in flowing in both directions.
As explained above, SSE is a technology that allows clients to subscribe to event notifications that originate on a server. Server generates new events and sends these events back to the clients subscribed to receive the notifications. In other words, SSE offers a solution for a one-way publish-subscribe model.
A good example of the use case where SSE can be used is a simple message exchange RESTful service. Clients
POST
new messages to the service and subscribe to receive messages from other clients.
Let's call the resource messages
. While POST
ing a new message to this resource involves
a typical HTTP request-response communication between a client and the messages
resource,
subscribing to receive all new message notifications would be hard and impractical to model with a sequence of
standard request-response message exchanges. Using Server-sent events provides a much more practical approach here.
You can use SSE to let clients subscribe to the messages
resource via standard GET
request (use a SSE client API, for example javascript API or Jersey Client SSE API) and let the server broadcast
new messages to all connected clients in the form of individual events (in our case using Jersey Server SSE API).
Note that with Jersey a SSE support is implemented as an usual JAX-RS resource method. There's no need to do anything
special to provide a SSE support in your Jersey/JAX-RS applications, your SSE-enabled resources are a standard part of
your RESTful Web application that defines the REST API of your application. The following chapters describes SSE
support in Jersey in more details.
Note, that while SSE in Jersey is supported with standard JAX-RS resources, Jersey SSE APIs are not part of the JAX-RS specification. SSE support and related APIs are a Jersey specific feature that extends JAX-RS.
This chapter briefly describes the Jersey support for SSE. Details and examples will be covered in chapters below.
Jersey contains support for SSE for both - server and client. SSE in Jersey is implemented as an extension supporting a new media type, which means that SSE really treated as just another media type that can be returned from a resource method and processed by the client. There is only a minimal additional support for "chunked" messages added to Jersey which could not be implemented as standard JAX-RS media type extension.
Before you start working with Jersey SSE, in order to add support for SSE you need to include the dependency to the SSE media type module:
Example 15.1. Add jersey-media-sse
dependency.
<dependency> <groupId>org.glassfish.jersey.media</groupId> <artifactId>jersey-media-sse</artifactId> </dependency>
Prior to Jersey 2.8, you had to manually register SseFeature in your application.
(The SseFeature
is a feature that can be registered for both, the client and the server.)
Since Jersey 2.8, the feature gets automatically discovered and registered when Jersey SSE module is
put on the application's classpath. The automatic discovery and registration of SSE feature can be suppressed
by setting DISABLE_SSE property to true
.
The behavior can also be selectively suppressed in either client or server runtime by setting
DISABLE_SSE_CLIENT or DISABLE_SSE_SERVER property
respectively.
SseFeature
adds new supported entity (representation) Java types, namely OutboundEvent
for the outbound server events and InboundEvent for inbound client events. These types are serialized by
OutboundEventWriter
and de-serialized by InboundEventReader
.
There is no restriction for a media type used in individual event messages; however the media type used for a SSE
stream as whole is "text/event-stream" and this media type should be set on messages that
are used to serve SSE events (for example on the server side using @Produces on the method that returns an
EventOutput
- see below).
The InboundEvent
and OutboundEvent
contain all the fields needed for composing
and processing individual SSE events. These entities represent the chunks sent or received over
an open server-to-client connection that is represented by an ChunkedOutput on the servers side and
ChunkedInput on the client side (if you are not familiar with
ChunkedOutput
and ChunkedInput
, see the
Async chapter first for more details). In other words, our resource method that is used
to open a SSE connection to a client does not return individual OutboundEvent
s. Instead, a new
instance of EventOutput is returned. EventOutput
is a typed extension of
ChunkedOutput<OutboundEvent>
. Similarly, to receive InboundEvent
s on a
client side, Jersey SSE API provides a EventInput
that represents a typed extension of
ChunkedInput<InboundEvent>
.
The Jersey server SSE API also contains a SseBroadcaster utility, that provides a convenient way of
grouping multiple EventOutput
instances that represent individual client connections into a
group, and exposes methods for broadcasting new events to all the client connections grouped in the broadcaster.
The SseBroadcaster
inherits from Broadcaster which is the generic broadcaster
implementation of the Jersey chunked message processing facility.
On the client side, the Jersey SSE API contains additional EventSource and EventListener
classes that further improve convenience of processing new inbound SSE events.
Firstly you need to add a Jersey SSE module dependency to your project as shown in the earlier section and register the SseFeature from this module in your Application or ResourceConfig. Once done, you are ready to add SSE support to your resource:
Example 15.2. Simple SSE resource method
... import org.glassfish.jersey.media.sse.EventOutput; import org.glassfish.jersey.media.sse.OutboundEvent; import org.glassfish.jersey.media.sse.SseFeature; ... @Path("events") public static class SseResource { @GET @Produces(SseFeature.SERVER_SENT_EVENTS) public EventOutput getServerSentEvents() { final EventOutput eventOutput = new EventOutput(); new Thread(new Runnable() { @Override public void run() { try { for (int i = 0; i < 10; i++) { // ... code that waits 1 second final OutboundEvent.Builder eventBuilder = new OutboundEvent.Builder(); eventBuilder.name("message-to-client"); eventBuilder.data(String.class, "Hello world " + i + "!"); final OutboundEvent event = eventBuilder.build(); eventOutput.write(event); } } catch (IOException e) { throw new RuntimeException( "Error when writing the event.", e); } finally { try { eventOutput.close(); } catch (IOException ioClose) { throw new RuntimeException( "Error when closing the event output.", ioClose); } } } }).start(); return eventOutput; } }
The code above defines the resource deployed on URI "/events"
. This resource has a single
@GET resource method which returns as an entity EventOutput - an extension of generic Jersey
ChunkedOutput API for output chunked message processing.
If you are not familiar with
ChunkedOutput
and ChunkedInput
, see the
Async chapter first for more details.
After the eventOutput
is returned from the method, the Jersey runtime recognizes that this is
a ChunkedOutput
extension and does not close the client connection immediately. Instead, it
writes the HTTP headers to the response stream and waits for more chunks (SSE events) to be sent. At this point
the client can read headers and starts listening for individual events.
Since Jersey runtime does not implicitly close the connection to the client (similarly to asynchronous processing), closing the connection is a responsibility of the resource method or the client listening on the open connection for new events (see following example).
In the Example 15.2, “Simple SSE resource method”, the resource method creates a new thread that sends a sequence of
10 events. There is a 1 second delay between two subsequent events as indicated in a comment. Each event is
represented by OutboundEvent
type and is built with a helpf of an outbound event
Builder
. The OutboundEvent
reflects the standardized format of SSE messages
and contains properties that represent name
(for named
events), comment
, data
or id
. The code also sets the
event data media type using the mediaType(MediaType)
method on the
eventBuilder
. The media type, together with the data type set by the
data(Class, Object>
method (in our case String.class
), is used
for serialization of the event data. Note that the event data media type will not be written to any headers as
the response Content-type
header is already defined by the @Produces
and set to
"text/event-stream"
using constant from the SseFeature
.
The event media type is used for serialization of event data
. Event data media type and Java
type are used to select the proper MessageBodyWriter<T> for event data serialization and are passed
to the selected writer that serializes the event data
content. In our case the string
"Hello world " + i + "!"
is serialized as "text/plain"
. In event
data
you can send any Java entity and associate it with any media type that you would be able
to serialize with an available MessageBodyWriter<T>
. Typically, you may want to send e.g. JSON data,
so you would fill the data
with a JAXB annotated bean instance and define media type to JSON.
If the event media type is not set explicitly, the "text/plain"
media type is used
by default.
Once an outbound event is ready, it can be written to the eventOutput
. At that point the event
is serialized by internal OutboundEventWriter
which uses an appropriate
MessageBodyWriter<T>
to serialize the "Hello world " + i + "!"
string. You can
send as many messages as you like. At the end of the thread execution the response is closed which also closes
the connection to the client. After that, no more messages can be sent to the client on this connection. If the
client would like to receive more messages, it would have to send a new request to the server to initiate a
new SSE streaming connection.
A client connecting to our SSE-enabled resource will receive the following data from the entity stream:
event: message-to-client data: Hello world 0! event: message-to-client data: Hello world 1! event: message-to-client data: Hello world 2! event: message-to-client data: Hello world 3! event: message-to-client data: Hello world 4! event: message-to-client data: Hello world 5! event: message-to-client data: Hello world 6! event: message-to-client data: Hello world 7! event: message-to-client data: Hello world 8! event: message-to-client data: Hello world 9!
Each message is received with a delay of one second.
If you have worked with streams in JAX-RS, you may wonder what is the difference between ChunkedOutput and StreamingOutput.
ChunkedOutput
is Jersey-specific API. It lets you send "chunks" of data without closing
the client connection using series of convenient calls to ChunkedOutput.write
methods
that take POJO + chunk media type as an input and then use the configured JAX-RS
MessageBodyWriter<T>
providers to figure out the proper way of serializing each chunk POJO
to bytes. Additionally, ChunkedOutput
writes can be invoked multiple times on the same
outbound response connection, i.e. individual chunks are written in each write, not the full response entity.
StreamingOutput
is, on the other hand, a low level JAX-RS API that works with bytes
directly. You have to implement StreamingOutput
interface yourself. Also, its
write(OutputStream)
method will be invoked by JAX-RS runtime only once per response
and the call to this method is blocking, i.e. the method is expected to write the entire entity body
before returning.
Jersey SSE server API defines SseBroadcaster which allows to broadcast individual events to multiple clients. A simple broadcasting implementation is shown in the following example:
Example 15.3. Broadcasting SSE messages
... import org.glassfish.jersey.media.sse.SseBroadcaster; ... @Singleton @Path("broadcast") public static class BroadcasterResource { private SseBroadcaster broadcaster = new SseBroadcaster(); @POST @Produces(MediaType.TEXT_PLAIN) @Consumes(MediaType.TEXT_PLAIN) public String broadcastMessage(String message) { OutboundEvent.Builder eventBuilder = new OutboundEvent.Builder(); OutboundEvent event = eventBuilder.name("message") .mediaType(MediaType.TEXT_PLAIN_TYPE) .data(String.class, message) .build(); broadcaster.broadcast(event); return "Message '" + message + "' has been broadcast."; } @GET @Produces(SseFeature.SERVER_SENT_EVENTS) public EventOutput listenToBroadcast() { final EventOutput eventOutput = new EventOutput(); this.broadcaster.add(eventOutput); return eventOutput; } }
Let's explore the example together. The BroadcasterResource
resource class is annotated with
@Singleton annotation which tells Jersey runtime that only a single instance of the resource
class should be used to serve all the incoming requests to /broadcast
path. This is needed as
we want to keep an application-wide single reference to the private broadcaster
field so that
we can use the same instance for all requests. Clients that want to listen to SSE events first send a
GET
request to the BroadcasterResource
, that is handled by the
listenToBroadcast()
resource method.
The method creates a new EventOutput
representing the connection to the requesting client
and registers this eventOutput
instance with the singleton broadcaster
,
using its add(EventOutput)
method. The method then returns the eventOutput
which causes Jersey to bind the eventOutput
instance with the requesting client and send the
response HTTP headers to the client. The client connection remains open and the client is now waiting ready to
receive new SSE events. All the events are written to the eventOutput
by
broadcaster
later on. This way developers can conveniently handle sending new events to
all the clients that subscribe to them.
When a client wants to broadcast new message to all the clients listening on their SSE connections,
it sends a POST
request to BroadcasterResource
resource with the message content.
The method broadcastMessage(String)
is invoked on BroadcasterResource
resource with the message content as an input parameter. A new SSE outbound event is built in the standard way
and passed to the broadcaster. The broadcaster internally invokes write(OutboundEvent)
on all
registered EventOutput
s. After that the method just return a standard text response
to the POST
ing client to inform the client that the message was successfully broadcast. As you can see,
the broadcastMessage(String)
resource method is just a simple JAX-RS resource method.
In order to implement such a scenario, you may have noticed, that the Jersey SseBroadcaster
is not mandatory to complete the use case. individual EventOutputs can be just stored in a collection
and iterated over in the broadcastMessage
method. However, the SseBroadcaster
internally identifies and handles also client disconnects. When a client closes the connection the broadcaster
detects this and removes the stale connection from the internal collection of the registered
EventOutputs as well as it frees all the server-side resources associated with the stale connection.
Additionally, the SseBroadcaster
is implemented to be thread-safe, so that clients can connect
and disconnect in any time and SseBroadcaster
will always broadcast messages to the most recent
collection of registered and active set of clients.
On the client side, Jersey exposes APIs that support receiving and processing SSE events using two programming models:
Pull model - pulling events from a EventInput, or |
Push model - listening for asynchronous notifications of EventSource |
Both models will be described.
The events can be read on the client side from a EventInput. See the following code:
Client client = ClientBuilder.newBuilder() .register(SseFeature.class).build(); WebTarget target = client.target("http://localhost:9998/events"); EventInput eventInput = target.request().get(EventInput.class); while (!eventInput.isClosed()) { final InboundEvent inboundEvent = eventInput.read(); if (inboundEvent == null) { // connection has been closed break; } System.out.println(inboundEvent.getName() + "; " + inboundEvent.readData(String.class)); }
In this example, a client connects to the server where the SseResource
from the
Example 15.2, “Simple SSE resource method” is deployed. At first, a new JAX-RS/Jersey client
instance is created with a SseFeature
registered. Then a WebTarget instance is
retrieved from the client
and is used to invoke a HTTP request. The returned response entity
is directly read as a EventInput
Java type, which is an extension of Jersey
ChunkedInput
that provides generic support for consuming chunked message payloads. The
code in the example then process starts a loop to process the inbound SSE events read from the
eventInput
response stream. Each chunk read from the input is a InboundEvent
.
The method InboundEvent.readData(Class)
provides a way for the client to indicate what Java type
should be used for the event data de-serialization. In our example, individual events are de-serialized as
String
Java type instances. This method internally finds and executes a proper
MessageBodyReader<T> which is the used to do the actual de-serialization. This is similar to reading an
entity from the Response by readEntity(Class)
. The method
readData
can also throw a ProcessingException.
The null
check on inboundEvent
is necessary to make sure that the chunk was properly
read and connection has not been closed by the server. Once the connection is closed, the loop terminates and
the program completes execution. The client code produces the following console output:
message-to-client; Hello world 0! message-to-client; Hello world 1! message-to-client; Hello world 2! message-to-client; Hello world 3! message-to-client; Hello world 4! message-to-client; Hello world 5! message-to-client; Hello world 6! message-to-client; Hello world 7! message-to-client; Hello world 8! message-to-client; Hello world 9!
The main Jersey SSE client API component used to read SSE events asynchronously is EventSource.
The usage of the EventSource
is shown on the following example.
Example 15.4. Registering EventListener
with EventSource
Client client = ClientBuilder.newBuilder() .register(SseFeature.class).build(); WebTarget target = client.target("http://example.com/events"); EventSource eventSource = EventSource.target(target).build(); EventListener listener = new EventListener() { @Override public void onEvent(InboundEvent inboundEvent) { System.out.println(inboundEvent.getName() + "; " + inboundEvent.readData(String.class)); } }; eventSource.register(listener, "message-to-client"); eventSource.open(); ... eventSource.close();
In this example, the client code again connects to the server where the SseResource
from the
Example 15.2, “Simple SSE resource method” is deployed. The Client instance
is again created and initialized with SseFeature. Then the WebTarget is built.
In this case a request to the web target is not made directly in the code, instead, the web target instance
is used to initialize a new EventSource.Builder instance that is used to build a new
EventSource
. The choice of build()
method is important, as it tells
the EventSource.Builder
to create a new EventSource
that is not automatically
connected to the target
. The connection is established only later by manually invoking
the eventSource.open()
method. A custom EventListener
implementation is used to listen to and process incoming SSE events. The method readData(Class) says that the
event data should be de-serialized from a received InboundEvent instance into a
String
Java type. This method call internally executes MessageBodyReader<T> which
de-serializes the event data. This is similar to reading an entity from the Response by
readEntity(Class)
. The method readData
can throw a
ProcessingException.
The custom event source listener is registered in the event source via
EventSource
.register(EventListener, String)
method. The next method
arguments define the names of the events to receive and can be omitted. If names are defined, the listener
will be associated with the named events and will only invoked for events with a name from the set of defined
event names. It will not be invoked for events with any other name or for events without a name.
It is a common mistake to think that unnamed events will be processed by listeners that are registered to process events from a particular name set. That is NOT the case! Unnamed events are only processed by listeners that are not name-bound. The same limitation applied to HTML5 Javascript SSE Client API supported by modern browsers.
After a connection to the server is opened by calling the open()
method on the event source,
the eventSource
starts listening to events. When an event named
"message-to-client"
comes, the listener will be executed by the event source. If any other
event comes (with a name different from "message-to-client"
), the registered listener is not
invoked. Once the client is done with processing and does not want to receive events anymore, it closes the
connection by calling the close()
method on the event source.
The listener from the example above will print the following output:
message-to-client; Hello world 0! message-to-client; Hello world 1! message-to-client; Hello world 2! message-to-client; Hello world 3! message-to-client; Hello world 4! message-to-client; Hello world 5! message-to-client; Hello world 6! message-to-client; Hello world 7! message-to-client; Hello world 8! message-to-client; Hello world 9!
When browsing through the Jersey SSE API documentation, you may have noticed that the EventSource
implements EventListener and provides an empty implementation for the
onEvent(InboundEvent inboundEvent)
listener method. This adds more flexibility to the
Jersey client-side SSE API. Instead of defining and registering a separate event listener, in simple scenarios
you can also choose to derive directly from the EventSource
and override the empty listener
method to handle the incoming events. This programming model is shown in the following example:
Example 15.5. Overriding EventSource.onEvent(InboundEvent)
method
Client client = ClientBuilder.newBuilder() .register(SseFeature.class).build(); WebTarget target = client.target("http://example.com/events"); EventSource eventSource = new EventSource(target) { @Override public void onEvent(InboundEvent inboundEvent) { if ("message-to-client".equals(inboundEvent.getName())) { System.out.println(inboundEvent.getName() + "; " + inboundEvent.readData(String.class)); } } }; ... eventSource.close();
The code above is very similar to the code in Example 15.4, “Registering EventListener
with EventSource
”. In this example
however, the EventSource
is constructed directly using a single-parameter constructor.
This way, the connection to the SSE endpoint is by default automatically opened at the event source
creation. The implementation of the EventListener
has been moved into the overridden
EventSource.onEvent(...)
method. However, this time, the listener method will be executed for
all events - unnamed as well as with any name
. Therefore the code checks the name whether it is
an event with the name "message-to-client" that we want to handle. Note that you can still register
additional EventListener
s later on. The overridden method on the event source allows you to
handle messages even when no additional listeners are registered yet.
Starting in Jersey 2.3, the EventSource
implementation supports automated recuperation
from a connection loss, including negotiation of delivery of any missed events based on the last received
SSE event id
field value, provided this field is set by the server and the negotiation
facility is supported by the server. In case of a connection loss, the last received SSE event
id
field value is send in the Last-Event-ID
HTTP request
header as part of a new connection request sent to the SSE endpoint. Upon a receipt of such reconnect request,
the SSE endpoint that supports this negotiation facility is expected to replay all missed events.
Note, that SSE lost event negotiation facility is a best-effort mechanism which does not provide any guaranty that all events would be delivered without a loss. You should therefore not rely on receiving every single event and design your client application code accordingly.
By default, when a connection the the SSE endpoint is lost, the event source will use a default delay
before attempting to reconnect to the SSE endpoint. The SSE endpoint can however control the client-side
retry delay by including a special retry
field value in any event sent to the client.
Jersey EventSource
implementation automatically tracks any received SSE event
retry
field values set by the endpoint and adjusts the reconnect delay accordingly,
using the last received retry
field value as the new reconnect delay.
In addition to handling the standard connection losses, Jersey EventSource
automatically
deals with any HTTP 503 Service Unavailable
responses received from the SSE endpoint,
that include a Retry-After
HTTP header with a valid value. The
HTTP 503 + Retry-After
technique is often used by HTTP endpoints as a means of
connection and traffic throttling. In case a HTTP 503 + Retry-After
response is received
in return to a connection request from SSE endpoint, Jersey EventSource
will automatically
schedule a reconnect attempt and use the received Retry-After
HTTP header value as a
one-time override of the reconnect delay.