Chapter 37 worked answers

These answers use baseline cbfc94ade5b57e449d6b73bf1485b1c5b1647022 and the chapter's linked execution records. Procedures remain scoped to disposable resources.

1. Name the modeled boundary

Rename agent-restart chaos to repeated running report against retained MemoryService. Rename server-restart chaos to rollout controller reconstruction over retained store. The first does not kill an agent; the second does not create a fresh server/control plane. Their useful assertions are report idempotency and recomputation, not durable process recovery.

2. Unknown existence prevents speculative create

Run the chapter's exact selected test command. The actual record passed all eight chaos subcases and selected agent/fake tests. The unavailable-runtime case expects inspect:container-1,list,list and zero status reports. There is no create in that operation history. Both inspect and list fail, so reality is unknown rather than confirmed absent. Creating a replacement could duplicate a still-running object.

3. Cleanup on return is not fencing during partition

A node partitions while its container continues. The control plane eventually marks it lost, removes its active assignment and replaces a stateless workload elsewhere. Until the old node reconnects and consumes cleanup, both physical workloads may exist. Returned-node reconciliation lists the managed unassigned old object, stops/removes it and reports removed. The fake test observes that cleanup sequence; it does not observe exclusive service traffic or storage fencing throughout the partition.

4. Match state owner to recovery claim

Reconstructed controllers can recompute from retained records. The combined actual experiment restarted the binary and found services/tasks/nodes absent while an exact manually associated nginx container survived. Durable rollout recovery requires a server actually composing persistent transactional state, fresh-process reads of deployment/task/version records, and observed recovery/cleanup of corresponding engine objects. Capture initial IDs, UTC transitions and final absence. Remove only owned exact objects/databases; never sweep unrelated containers. A healthy endpoint is insufficient recovery evidence.

See the corresponding chapter claim ledger for exact immutable source anchors and limitations.

Return to Chapter 37.