Expand description
The supervisor itself: desired state in, running processes out. The managed-worker supervisor: desired state in, running processes out.
Desired state is the DURABLE record (WorkerDeployment.desired); actual
state is a live [InstanceHandle]. This type is the only thing that reads
one and writes the other, and every answer it gives is a join of the two —
never one standing in for the other.
Supervision is COMMISSIONED, not defaulted. A server whose operator has not
written a [worker_supervision] policy supervises nothing and says so with
the remedy attached; it does not pick a backoff on the operator’s behalf
(ADR-001).
Structs§
- Fleet
Shutdown Report - What
WorkerSupervisor::shutdownobserved, by name on both sides. - Worker
Supervisor - Server-owned supervision of the managed worker fleet.
Enums§
- Convergence
- How far a convergence may go when a live instance already exists.