The primary goal of Centrifugo is to handle persistent consumer connections established over numerous real-time transports (together with WebSocket, HTTP-Streaming, SSE, WebTransport, and many others – see right here) and provide an API for publishing information in direction of established connections. Shoppers subscribe to channels, therefore Centrifugo implements PUB/SUB mechanics to transmit printed information to all on-line channel subscribers.
Centrifugo employs Redis as its main scalability choice – in order that it’s attainable to distribute consumer connections amongst quite a few Centrifugo nodes with out worrying about channel subscribers linked to separate nodes. Redis is extremely mature, easy, and quick in-memory storage. Resulting from numerous built-in information constructions and PUB/SUB help Redis is an ideal match to be each Centrifugo Dealer and PresenceManager (we’ll describe what’s this shortly).
In Centrifugo v4.1.0 we launched an up to date implementation of our Redis Engine (Engine in Centrifugo == Dealer + PresenceManager) which offers adequate efficiency enhancements to our customers. This put up discusses the elements that prompted us to replace Redis Engine implementation and offers some perception into the outcomes we managed to realize. We’ll study a number of well-known Go libraries for Redis communication and distinction them towards Centrifugo duties.
Dealer and PresenceManager
Earlier than we get began, let’s outline what Centrifugo’s Dealer and PresenceManager phrases imply.
Dealer is an interface answerable for sustaining subscriptions from completely different Centrifugo nodes (initiated by consumer connections). That helps to scale consumer connections over many Centrifugo cases and never fear about the identical channel subscribers being linked to completely different nodes – since all Centrifugo nodes linked with PUB/SUB. Messages printed to at least one node are delivered to a channel subscriber linked to a different node.
One other main a part of Dealer is protecting an expiring publication historical past for channels (streams). In order that Centrifugo could present a quick cache for messages missed by purchasers upon going offline for a brief interval and compensate at most as soon as supply of Redis PUB/SUB utilizing Publication incremental offsets. Centrifugo makes use of STREAM and HASH information constructions in Redis to retailer channel historical past and stream meta data.
Usually Centrifugo structure could also be completely illustrated by this image (Gophers are Centrifugo nodes all linked to Dealer, and sockets are WebSockets):
