Scaling
Depending on your requirements, your feathers application may need to provide high availability. Feathers is designed to scale.
The types of transports used in a feathers application will impact the scaling configuration. For example, a feathers app that uses the feathers-rest
adapter exclusively will require less scaling configuration because HTTP is a stateless protocol. If using websockets (a stateful protocol) through the feathers-socketio
or feathers-primus
adapters, configuration may be more complex to ensure websockets work properly.
Horizontal Scaling
Scaling horizontally refers to either:
- setting up a cluster, or
- adding more machines to support your application
To achieve high availability, varying combinations of both strategies may be used.
Cluster configuration
Cluster support is built into core NodeJS. Since NodeJS is single threaded, clustering allows you to easily distribute application requests among multiple child processes (and multiple threads). Clustering is a good choice when running feathers in a multi-core environment.
Below is an example of adding clustering to feathers with the feathers-socketio
provider. By default, websocket connections begin via a handshake of multiple HTTP requests and are upgraded to the websocket protocol. However, when clustering is enabled, the same worker will not process all HTTP requests for a handshake, leading to HTTP 400 errors. To ensure a successful handshake, force a single worker to process the handshake by disabling the http transport and exclusively using the websocket
transport.
There are notable side effects to be aware of when disabling the HTTP transport for websockets. While all modern browsers support websocket connections, there is no websocket support for IE <=9 and Android Browser <=4.3. If you must support these browsers, use alternative scaling strategies.
import cluster from 'cluster';
import feathers from '@feathersjs/feathers';
import socketio from '@feathersjs/socketio';
const CLUSTER_COUNT = 4;
if (cluster.isMaster) {
for (let i = 0; i < CLUSTER_COUNT; i++) {
cluster.fork();
}
} else {
const app = feathers();
// ensure the same worker handles websocket connections
app.configure(socketio({
transports: ['websocket']
}));
app.listen(4000);
}
In your feathers client code, limit the socket.io-client to the websocket
transport and disable upgrade
.
import feathers from '@feathersjs/client';
import io from 'socket.io-client';
import socketio from 'feathers-socketio/client';
const app = feathers()
.configure(socketio(
io('http://api.feathersjs.com', {
transports: ['websocket'],
upgrade: false
})
));