Linux/macOS only!

The cluster executable lets you spawn multiple instances of a process and automatically restart them when needed.

It also supports "socket clustering" similar to node.js. If sub-processes use use cluster_listen and cluster_accept instead of socket()/bind()/listen() and accept(), they work as usual when started on their own, but share the socket if started in cluster mode. Incoming connections are then distributed in round-robin fashion across all available nodes.

See examples/http_server.jai for details. Run ./cluster -n 4 -- examples/http_server -p <port> to see it in action.

On SIGHUP, cluster spwans a new set of cluster instances and then sends all old instances a SIGTERM, giving them the chance to gracefully shut down. This currently only works if the instances use cluster_listen because there’s no other "ready" signal yet.

On SIGINT or SIGTERM, cluster sends all isntances a SIGTERM and then exits. Sending one of those signals a second time causes cluster to exit immediately.