Expand description
Node-local supervision of hosted instances (ADR-099 section 7): a periodic health heartbeat whose meaning each kind defines, and bounded restarts with backoff, every step chained.
WorkloadHostService::supervise polls each instance that is meant to
run, once per its kind’s interval. A kind’s verdict
(crate::workload_kind::WorkloadKind::judge) counts a miss; enough
misses in a row make the instance Unhealthy, and the restart policy
then restarts it (stop, then start, both gated and chained by the
adapter host) until its budget inside the window is spent, at which
point it is Failed and held for the operator. An instance the
operator stopped, one that was never started and a finished one-shot run
are never restarted.
Structs§
- Supervised
- One thing a
WorkloadHostService::supervisepass did.