Concept — distributed nodes

A single RUSM node is a host process running many processes. A cluster is several nodes — on different machines — connected so processes can spawn and message across them, exactly like distributed Erlang. This ships in the Wasm-free rusm-cluster crate over rusm-otp.

Connecting

Nodes connect over QUIC with TLS 1.3 (think Node.connect/1, but secure by default). Each node has a name; on connect, both ends exchange names over a dedicated control stream and remember the peer. Links are mutually authenticated — both ends present a certificate and verify the other against a shared trust anchor, so a peer without a trusted certificate is rejected at the handshake. Use a ClusterCa to issue each node its own CA-signed certificate (per-node keys, independently revocable), or a single shared Identity::generate() for a small/trusted cluster.

Location transparency

Once connected:

• Cross-node send — node.send("london", "greeter", bytes) routes to the process registered as greeter on london; the sender doesn't open the socket itself.
• Global registry — register_global(name, pid) publishes a name cluster-wide by gossiping it to every peer; send_global(name, bytes) resolves the owning node and routes there, so a service is reachable by name from anywhere.
• Remote spawn — a node registers spawnable factories by name; a peer calls spawn_remote(node, factory, args) and gets back the pid spawned on that node. (A closure can't cross the wire, so a node only spawns work it's been taught — explicit and capability-friendly.)

Wire shape

Each link multiplexes a single long-lived control stream (name exchange, registry gossip, the request/reply RPC behind remote spawn and live attach) and one uni-stream per message (so messages never head-of-line-block each other). On loopback the transport does ~550k cross-node messages/sec at ~39µs p50 round-trip — see the cluster_fanout benchmark.

Testing it

Tests boot several nodes in one process and connect them on loopback, so cross-node send, the global registry, remote spawn, and live-attach listing are all TDD-able with no external network.

Hooking in

You can also attach to any running node to inspect it live — see live attach.