A Node.js WebSocket server for sending async event data to the Superset web application frontend.
- Node.js 12+ (not tested with older versions)
- Redis 5+
To use this feature, Superset needs to be configured to enable global async queries and to use WebSockets as the transport (see below).
This implementation is based on the architecture defined in SIP-39.
Async events are pushed to Redis Streams from the Superset Flask app. An event for a particular user is published to two streams: 1) the global event stream that includes events for all users, and 2) a channel/session-specific stream only for the user. This approach provides a good balance of performance (reading off of a single global stream) and fault tolerance (dropped connections can "catch up" by reading from the channel-specific stream).
Note that Redis Stream consumer groups are not used here due to the fact that each group receives a subset of the data for a stream, and WebSocket clients have a persistent connection to each app instance, requiring access to all data in a stream. Horizontal scaling of the WebSocket app requires having multiple WebSocket servers, each with full access to the Redis Stream data.
When a user's browser initially connects to the WebSocket server, it does so over HTTP, which includes the JWT authentication cookie, set by the Flask app, in the request. Note that due to the cookie-based authentication method, the WebSocket server must be run on the same host as the web application. The server validates the JWT token by using the shared secret (config: jwtSecret
), and if valid, proceeds to upgrade the connection to a WebSocket. The user's session-based "channel" ID is contained in the JWT, and serves as the basis for sending received events to the user's connected socket(s).
A user may have multiple WebSocket connections under a single channel (session) ID. This would be the case if the user has multiple browser tabs open, for example. In this scenario, all events received for a specific channel are sent to all connected sockets, leaving it to the consumer to decide which events are relevant to the current application context.
It is expected that a user's WebSocket connection may be dropped or interrupted due to fluctuating network conditions. The Superset frontend code keeps track of the last received async event ID, and attempts to reconnect to the WebSocket server with a last_id
query parameter in the initial HTTP request. If a connection includes a valid last_id
value, events that may have already been received and sent unsuccessfully are read from the channel-based Redis Stream and re-sent to the new WebSocket connection. The global event stream flow then assumes responsibility for sending subsequent events to the connected socket(s).
The server utilizes the standard WebSocket ping/pong functionality to determine if active WebSocket connections are still alive. Active sockets are sent a ping regularly (config: pingSocketsIntervalMs
), and the internal sockets registry is updated with a timestamp when a pong response is received. If a pong response has not been received before the timeout period (config: socketResponseTimeoutMs
), the socket is terminated and removed from the internal registry.
In addition to periodic socket connection cleanup, the internal channels registry is regularly "cleaned" (config: gcChannelsIntervalMs
) to remove stale references and prevent excessive memory consumption over time.
Install dependencies:
npm ci
Copy config.example.json
to config.json
and adjust the values for your environment.
Configuration via environment variables is also supported which can be helpful in certain contexts, e.g., deployment. src/config.ts
can be consulted to see the full list of supported values.
Configure the Superset Flask app to enable global async queries (in superset_config.py
):
Enable the GLOBAL_ASYNC_QUERIES
feature flag:
"GLOBAL_ASYNC_QUERIES": True
Configure the following Superset values:
GLOBAL_ASYNC_QUERIES_TRANSPORT = "ws"
GLOBAL_ASYNC_QUERIES_WEBSOCKET_URL = "ws://<host>:<port>/"
Note that the WebSocket server must be run on the same hostname (different port) for cookies to be shared between the Flask app and the WebSocket server.
Note also that localhost
and 127.0.0.1
are not considered the same host. For example, if you're pointing your browser to localhost:<port>
for Superset, then the WebSocket url will need to be configured as localhost:<port>
.
The following config values must contain the same values in both the Flask app config and config.json
:
GLOBAL_ASYNC_QUERIES_REDIS_CONFIG
GLOBAL_ASYNC_QUERIES_REDIS_STREAM_PREFIX
GLOBAL_ASYNC_QUERIES_JWT_COOKIE_NAME
GLOBAL_ASYNC_QUERIES_JWT_SECRET
More info on Superset configuration values for async queries: https://github.com/apache/superset/blob/master/CONTRIBUTING.md#async-chart-queries
The application is tracking a couple of metrics with statsd
using the hot-shots library, such as the number of connected clients and the number of failed attempts to send a message to a client.
statsd
can be configured with the statsd
object in the configuration file. See the hot-shots readme for more info. The default configuration is:
{
"statsd": {
"host": "127.0.0.1",
"port": 8125,
"globalTags": []
}
}
Running locally via dev server:
npm run dev-server
Running in production:
npm run build && npm start
The WebSocket server supports health checks via one of:
GET /health
OR
HEAD /health
TODO: containerize websocket server