Repository navigation
WebSocket SSE
Elio provides full support for WebSocket (RFC 6455) and Server-Sent Events (SSE) for real-time bidirectional and unidirectional communication.
The WebSocket implementation follows RFC 6455, including frame masking for all client-to-server messages as required by the specification. Masking prevents cache poisoning attacks on intermediary proxies that might misinterpret WebSocket frames as HTTP. The server-side does not mask frames, as the RFC only mandates client-to-server masking.
Server-Sent Events (SSE) provides a simpler alternative to WebSocket for server-to-client streaming. SSE operates over standard HTTP, requires no protocol upgrade, and benefits from automatic reconnection built into the browser EventSource API. The tradeoff is that SSE is unidirectional (server to client only), but for many use cases -- notifications, live feeds, progress updates -- this is sufficient and avoids the complexity of a bidirectional protocol.
HTTP CONNECT is a separate 2xx tunnel handoff, not a WebSocket 101 upgrade or
an SSE body. A WebSocket server's ordinary HTTP fallback uses the shared reply
dispatcher and can host router.connect(); CONNECT is not eligible for its
WebSocket-upgrade path. After successful CONNECT acceptance, the scoped session
owns opaque tunnel I/O until completion and HTTP parsing never resumes. See
CONNECT handoff for authority routing,
borrowed buffers, TLS closure and caller-owned proxy authorization policy.
WebSocket provides full-duplex communication channels over a single TCP connection.
#include <elio/elio.hpp>
#include <elio/http/websocket.hpp>
using namespace elio;
using namespace elio::http::websocket;
// WebSocket message handler
coro::task<void> echo_handler(ws_connection& conn) {
while (conn.is_open()) {
auto msg = co_await conn.receive();
if (!msg) break;
// Echo back the message
if (msg->type == opcode::text) {
co_await conn.send_text(msg->data);
} else if (msg->type == opcode::binary) {
co_await conn.send_binary(msg->data);
}
}
}
coro::task<int> async_main(int argc, char* argv[]) {
(void)argc;
(void)argv;
// Create WebSocket-enabled router
ws_router router;
// Regular HTTP routes
router.get("/", [](http::context& ctx) -> coro::task<http::response> {
co_return http::response::ok("Hello!");
});
// WebSocket routes
server_config ws_config;
ws_config.max_message_size = 1024 * 1024; // 1MB
router.websocket("/ws", echo_handler, ws_config);
// Create and start server
ws_server srv(std::move(router));
auto bind_addr = net::socket_address(net::ipv4_address(8080));
// Run until SIGINT/SIGTERM, then stop the listener cleanly.
co_await elio::serve(srv, [&]() { return srv.listen(bind_addr); });
co_return 0;
}
int main(int argc, char* argv[]) {
elio::signal::signal_set shutdown_signals(elio::default_shutdown_signals);
shutdown_signals.block_all_threads();
return elio::run(async_main, argc, argv);
}ws_router::websocket() uses the same path pattern grammar as the HTTP
router:
| Pattern form | Meaning |
|---|---|
/ws |
Matches the literal path /ws
|
/chat/:room |
Captures one path component as room; read it with conn.param("room")
|
/assets/* |
Matches /assets/ and any remaining path tail |
The * wildcard is only valid as the final path segment. Patterns such as
/chat/*/admin are rejected during route registration because a non-terminal
wildcard would make the remaining literal segments unreachable.
Named :param captures are stored on the ws_connection passed to the
handler. Use conn.param("name") for one value or conn.params() to inspect
all captured path parameters.
The HTTP upgrade request is governed by http::server_config::max_request_size.
That limit counts the upgrade request line, headers, and body, but does not
count bytes after the upgrade request has completed. Post-upgrade WebSocket
frame and message bytes are governed by websocket::server_config limits such
as max_message_size.
websocket::server_config::ping_interval enables an automatic server heartbeat
for upgraded connections on that route. The server has up to ping_timeout to
send each heartbeat ping and observe a pong processed by the handler's
conn.receive() loop. If the ping cannot be sent or no pong is observed before
that window expires, the connection fails closed. A heartbeat timeout may close
the transport without delivering a WebSocket close frame. The heartbeat task does
not read from the stream, so route handlers remain the single reader for the
connection. Set ping_interval <= 0 to disable the heartbeat, or
ping_timeout <= 0 to send periodic pings without timeout closure. After a WSS
upgrade completes, the underlying TLS stream serializes direct OpenSSL state
access so the receive loop can wait for inbound frames while the heartbeat
writes pings.
#include <elio/elio.hpp>
#include <elio/http/websocket.hpp>
using namespace elio;
using namespace elio::http::websocket;
coro::task<void> connect_example() {
// Create client
client_config config;
config.subprotocols = {"chat", "json"}; // Optional subprotocols
config.connect_timeout = std::chrono::seconds(10);
config.read_timeout = std::chrono::seconds(30); // Upgrade response deadline
ws_client client(config);
coro::cancel_source cancel;
// Connect to server
if (!co_await client.connect("ws://localhost:8080/ws",
cancel.get_token())) {
ELIO_LOG_ERROR("Failed to connect");
co_return;
}
ELIO_LOG_INFO("Connected! Subprotocol: {}", client.subprotocol());
// Send messages
co_await client.send_text("Hello, server!");
co_await client.send_binary("\x01\x02\x03");
// Receive messages
while (client.is_open()) {
auto msg = co_await client.receive(cancel.get_token());
if (!msg) break;
if (msg->type == opcode::text) {
ELIO_LOG_INFO("Received: {}", msg->data);
}
}
// Close connection
co_await client.close(close_code::normal, "Done");
}ws_client::connect() performs one connection attempt and does not install an
automatic reconnect loop. If connect() fails, or a later operation observes a
closed or failed connection, the caller decides whether to retry, how long to
back off, whether application messages are safe to replay, and which connection
state should be restored. A caller-managed reconnect can call connect() again
once the client is closed.
| Opcode | Name | Description |
|---|---|---|
| 0x0 | continuation | Continuation of fragmented message |
| 0x1 | text | UTF-8 text data |
| 0x2 | binary | Binary data |
| 0x8 | close | Connection close |
| 0x9 | ping | Ping (keep-alive) |
| 0xA | pong | Pong (response to ping) |
| Code | Name | Description |
|---|---|---|
| 1000 | normal | Normal closure |
| 1001 | going_away | Endpoint going away |
| 1002 | protocol_error | Protocol error |
| 1003 | unsupported | Unsupported data type |
| 1008 | policy_violation | Policy violation |
| 1009 | too_large | Message too big |
| 1011 | unexpected | Unexpected condition |
For secure connections, use wss:// URLs and configure TLS:
// Server with TLS
auto tls_ctx = tls::tls_context::make_server("cert.pem", "key.pem");
auto task = srv.listen_tls(net::ipv4_address(8443), tls_ctx);
// Client with TLS
ws_client client;
client.tls_context().use_default_verify_paths();
co_await client.connect("wss://example.com/ws");The server-side TLS handshake is bounded by the http::server_config
passed to ws_server; keep_alive_timeout controls both the inbound TLS
handshake and the HTTP upgrade request read. Set it to a non-positive duration
to disable those deadlines.
SSE provides server-to-client event streaming over HTTP.
#include <elio/elio.hpp>
#include <elio/http/sse.hpp>
using namespace elio;
using namespace elio::http::sse;
// Event stream handler
coro::task<void> event_stream(sse_connection& conn) {
int counter = 0;
// Send retry interval
co_await conn.send_retry(3000); // 3 seconds
while (conn.is_active()) {
++counter;
// Send a simple event
co_await conn.send_data("Counter: " + std::to_string(counter));
// Send a typed event
event evt = event::full(
std::to_string(counter), // id
"update", // type
R"({"count":)" + std::to_string(counter) + "}"
);
co_await conn.send(evt);
// Wait before next event
co_await time::sleep_for(std::chrono::seconds(1));
}
}#include <elio/elio.hpp>
#include <elio/http/sse.hpp>
using namespace elio;
using namespace elio::http::sse;
coro::task<void> listen_events() {
// Configure client
client_config config;
config.auto_reconnect = true;
config.default_retry_ms = 3000;
config.connect_timeout = std::chrono::seconds(10);
config.read_timeout = std::chrono::seconds(30); // Response header deadline
sse_client client(config);
coro::cancel_source cancel;
// Connect
if (!co_await client.connect("http://localhost:8080/events",
cancel.get_token())) {
ELIO_LOG_ERROR("Failed to connect");
co_return;
}
// Receive events
while (client.is_connected()) {
auto evt = co_await client.receive(cancel.get_token());
if (!evt) break;
ELIO_LOG_INFO("Event: type={} id={} data={}",
evt->type.empty() ? "message" : evt->type,
evt->id,
evt->data);
}
}For WebSocket and SSE clients, connect_timeout bounds TCP connect and TLS
handshake setup. read_timeout bounds the protocol response headers read by
connect() -- the WebSocket 101 Switching Protocols response or the SSE
text/event-stream response headers. Values less than or equal to zero disable
these client-side read deadlines.
dns_timeout independently bounds DNS observation during each new connection
attempt; zero/negative disables it and the default is zero. dns_domain selects
shared bounded admission (null uses the capacity-64 default). The TCP/TLS budget
still starts after DNS. Cancellation/expiry can leave owned running libc work
behind until it returns, so normal shutdown may still drain it. See
Networking for cache lifetime,
overload, and absolute deadline semantics.
Both client configs inherit the shared HTTP base_client_config fields:
read_buffer_size, user_agent, verify_certificate, resolve_options,
rotate_resolved_addresses, max_headers, and max_header_size. These control
buffer sizing, TLS verification, DNS cache/rotation behavior, and response
header parser limits during connection setup.
Both clients also provide cancellation-token overloads. Cancelling the token
passed to connect() leaves pending DNS observation or aborts pending TCP connect, TLS handshake, request write,
and response header reads. Cancelling the token passed to receive() aborts a
blocked WebSocket frame or SSE event read. Cancellation returns false or
std::nullopt and sets errno to ECANCELED.
For SSE, the per-receive token also covers automatic reconnect backoff and
connection setup initiated by that receive call; the original connect token
remains independently effective.
For managed responses, sse::event_view::id is an
std::optional<std::string_view>. Omission (std::nullopt or {}) preserves
the receiver's Last-Event-ID. An engaged empty view (std::string_view{} or
"") emits id:\n and clears it; later events omitting id retain that cleared
state. For example, writer.send_event({std::string_view{}, {}, "reset"})
sends an explicit reset with data. send_data() omits id. These fields remain
borrowed through completion; presence adds no owned string or allocation.
This does not change legacy raw sse_connection serialization.
SSE is the decoded HTTP body, not the raw bytes immediately after response
headers. The client supports Content-Length, chunked transfer encoding, and
close-delimited bodies. It removes chunk metadata before event parsing and
stops at the declared body boundary. connect() returns after valid final
headers; receive() processes any body bytes that arrived in the same read.
For example, this escaped HTTP/1.1 response contains one data: hi event;
the chunk length a counts the ten payload bytes, not its framing:
HTTP/1.1 200 OK\r\n
Content-Type: text/event-stream\r\n
Transfer-Encoding: chunked\r\n
\r\n
a\r\n
data: hi\n\n
\r\n
0\r\n
\r\n
Equivalently, a finite stream can send Content-Length: 10 followed by
data: hi\n\n, without the chunk-size or chunk-delimiter bytes. Never label
raw SSE data as chunked without encoding real HTTP chunks, and never advertise
a guessed Content-Length. Chunk boundaries can split event fields or UTF-8
bytes; they do not define event boundaries.
sse::client_config::max_informational_responses defaults to 16; zero disallows
preceding interims. The existing absolute header deadline covers all interims.
The client rejects 101 as an SSE response. With automatic reconnect disabled,
normal body completion returns std::nullopt and errno = 0; truncated length
or chunk framing returns EBADMSG. Previously delivered events remain delivered.
With reconnect enabled, completion/failure follows the configured receive-driven
reconnect policy and Last-Event-ID handling; this is not exactly-once delivery.
For outgoing SSE, return sse::make_streaming_response(producer) from an HTTP
route. The owned producer receives sse::event_writer& and a cancellation token,
and returns task<send_result>. Await and check every event write; successful
producer return delegates final HTTP framing to the server. Unknown-length
HTTP/1.1 streams use real chunks by default. The event writer borrows data and
field slices through each await without concatenating a complete event body.
It rejects CR/LF/NUL in id/type before emitting the event; data and comments
support multiline input. Do not use writers concurrently or retain them after
producer return.
build_sse_response() is removed. The new factory supplies Content-Type and
Cache-Control but does not grant CORS permission or force Connection: close.
Set cross-origin policy explicitly when needed. Legacy raw-stream
sse_connection cannot be used to bypass the managed HTTP writer. See
examples/sse_server.cpp, HTTP Streaming, and Migrating to 0.6 for
complete examples, framing policy, borrowing, and cancellation boundaries.
SSE events are formatted as text with specific fields:
id: event-123
event: notification
retry: 3000
data: Hello, World!
data: This is line 2
-
id:- Event ID for reconnection tracking -
event:- Event type (default: "message") -
retry:- Reconnection interval in milliseconds -
data:- Event data (can span multiple lines) - Empty line marks end of event
| Feature | SSE | WebSocket |
|---|---|---|
| Direction | Server → Client only | Bidirectional |
| Protocol | HTTP | Custom over TCP |
| Reconnection | Automatic in browser EventSource; configurable in sse_client
|
Caller-managed; no automatic ws_client loop |
| Browser support | EventSource API | WebSocket API |
| Complexity | Simple | More complex |
| Best for | Notifications, feeds | Chat, games, real-time |
- Real-time notifications
- Live feeds (news, stocks)
- Progress updates
- Log streaming
- Any server-push scenario without client messages
- Chat applications
- Real-time games
- Collaborative editing
- Any scenario requiring bidirectional communication
const ws = new WebSocket('ws://localhost:8080/ws');
ws.onopen = () => {
console.log('Connected');
ws.send('Hello, server!');
};
ws.onmessage = (event) => {
console.log('Received:', event.data);
};
ws.onclose = (event) => {
console.log('Disconnected:', event.code);
};
// Send messages
ws.send('Hello');
ws.send(new Blob([1, 2, 3]));
// Close
ws.close(1000, 'Done');const es = new EventSource('/events');
es.onmessage = (event) => {
console.log('Message:', event.data);
};
es.addEventListener('notification', (event) => {
console.log('Notification:', event.data);
console.log('ID:', event.lastEventId);
});
es.onerror = (error) => {
console.error('Error:', error);
// EventSource will auto-reconnect
};
// Close
es.close();Complete examples are available:
-
examples/websocket_server.cpp- WebSocket server with echo and chat -
examples/websocket_client.cpp- WebSocket client demo -
examples/sse_server.cpp- SSE event streaming server -
examples/sse_client.cpp- SSE event receiver
- See API Reference for detailed API documentation
- See Networking for TCP and HTTP documentation
- See Examples for more code examples