Server-sent events in OpenAPI

Server-sent events (SSE) allow servers to push real-time updates to clients over a single HTTP connection. This protocol is widely used for scenarios requiring steady updates, such as notifications, live data feeds, or chat applications. While SSE shares some conceptual similarities with the WebSocket protocol, SSE differs significantly as it is a one-way communication from the server to the client, whereas the WebSocket protocol supports full-duplex communication.

The following table summarizes the main differences between SSE and WebSockets.

SSE is ideal for simple, efficient, one-way server-to-client communication scenarios. The WebSocket protocol is better suited to applications that require interactive, low-latency, two-way communication between the client and server.

Defining SSE in OpenAPI documents

Server-sent events (SSE) are not natively supported in OpenAPI but can be represented as an endpoint using the text/event-stream MIME type to indicate the data format.

The event stream format is a UTF-8-encoded text stream with messages separated by a newline (\n). Each message may include up to four fields:

• event: A string specifying the event type.
• data: The payload, often a JSON object or plain text.
• id: An optional unique identifier for resuming streams after disconnection.
• retry: An optional integer defining reconnection delay in milliseconds.

Depending on application needs, messages can include only the data field, only the event field, or a combination of fields. This flexibility allows for tailored implementations, for example, a data-only stream for updates or an event-only stream for simple notifications.

This example SSE endpoint notifies the client about stock price updates and includes only the id, event, and data fields:

A JavaScript client can subscribe to the endpoint using the EventSource API:

Best practices for SSE design and OpenAPI integration

Here are some best practices for handling server-sent events and including them in an OpenAPI document.

Improve reliability with heartbeats

Sending a heartbeat every few seconds is recommended to improve reliability by keeping the connection alive. Heartbeats can also help detect network issues and prompt the client to reconnect.

If you implement heartbeats, your SSE APIs can send multiple types of events, allowing you to use the oneOf keyword to describe the heartbeat message format.

Include event identification in the event payload

Include an id or sequence property in the event payload to ensure that the client receives events in the correct order and avoid missing or out-of-order updates.

Implement a retry mechanism

To prevent data loss when an API fails to send an event, implement a retry mechanism such as introducing a delay before retrying or using exponential backoff.

Use sentinel events to signal a closed connection

Sending a sentinel event can be helpful to indicate that the connection is closed or the server is no longer available. This is useful for error handling or notifying the client that there is no more data to be received.

The following example schema for a sentinel event demonstrates how a client can terminate a connection when it receives the CLOSED sentinel event:

For each event received, the client can check the X-SSE-Sentinel header to determine whether the connection has closed or if no more data needs to be received.

Handle SSE errors effectively

To handle errors in SSE effectively, servers can send specific error events within the stream that include an error field detailing the issue. For non-critical errors, custom headers like X-SSE-Error can communicate problems without interrupting the event flow.

Specific error events in the stream can be defined using a structured schema, as demonstrated below.

For critical errors, use a sentinel event as described in the previous section.