Namespaces

A namespace is a node's execution context - the container for every process running on that machine. One running node is exactly one namespace. When you connect to a machine in your fleet, you are talking to its namespace.

As a program author you never construct or manage a namespace directly; the runtime hands you a reference to it. A process receives it inside its ProcessContext, and an impl receives it in the driver's launchers hook. Conceptually, the namespace is the thing that tracks the node's running processes (what ps reports), routes signals to them (kill), and hands out the capabilities clients call.

One namespace per node

A node creates its namespace once, at startup, and every client session shares it. Two clients connected to the same node see the same processes - there is no per-connection isolation. Isolation between workloads is achieved by running them on separate nodes, not by carving up a single node's namespace.

Executor-local

A namespace runs on a single-threaded Embassy executor. It is held as an Rc<Namespace> and owns the executor's spawner, so it is pinned to that executor - it is not Send/Sync, and all access to it happens on the node's executor. The internal bookkeeping it guards (process tables, signal channels, readiness watches) is runtime machinery; see the Architecture page if you need that detail.