Runtime Supervision

The Host's day job is supervising package runners. This section covers the supervised lifecycle and the recovery work that came out of the launch-readiness atomic-writes branch (WORKSTREAMS/launch-readiness-atomic-writes-and-bootstrap/).

Pages

• Start runtime — matrix up, what the Host does on a successful spawn, mount-conflict checks, readiness probes.
• Stop runtime — matrix down, graceful vs signal-driven stop, runtime.shutdown over the control mount.
• Restart runtime — restart policy semantics and how restart: always interacts with crash loops.
• Runtime status — the five status values, what triggers each transition, how to query them.
• Runtime health — runtime.health over control mounts, degraded-state propagation, why HTTP 200 is not proof.
• Runtime logs — per-runtime stdout/stderr files, log-tail-on-failure, and the absence of unified hivecast logs.
• Shutdown semantics — Host stop sequencing, SIGTERM/SIGKILL deadlines, NATS sibling teardown, the launch-readiness recovery work (atomic writes, parse-and-skip, idempotent seed, journald cap, StartLimitBurst hardening).

See also

• Host control plane
• Host lifecycle
• Failure modes