Test report: Released native fork: coordinated maintenance

Test report

This page is for contributors and records a particular design or test snapshot. For current user guidance, see capabilities and limitations and the operation guides. Historical results do not establish current release or AWS validation.

On September 20, 2026, the released 0.5.1-ewhauser.2 macOS ARM64 binary passed three populated full-stop schedules against isolated Docker MinIO:

Schedule	Acknowledged writes recovered through every fresh node
Bucket sequential 3→2→1→0; last member held eight resident cells	28/28
Bucket concurrent stop of all three nodes	20/20
Fleet concurrent stop, including a peer-only acknowledgment while MinIO was paused	21/21

Every original disk was deleted only after its exact runtime generation reported data_safe and its process exited successfully. All three replacement nodes started on fresh disks. No loss markers were observed. A separate one-millisecond deadline produced an immutable failure and retained the affected disk.

Version .1 failed the same populated sequential Bucket schedule because its last member waited for successor adoption. Version .2 completes strict disk removal after durability, runtime stop and ownership release, without requiring a successor. Ordinary handoff retains its adoption behavior.

The result records the source commit, exact release-binary hash and terminal outcomes. Success events and deadline events record the observations. The maintained native harness lives in the fork; complete local evidence is recorded in the result. Release checksums and GitHub attestations were verified for all three native platforms.

This establishes native celld/MinIO recovery through full disk replacement. It does not establish Kubernetes sequencing, launcher exclusion, sustained stress behavior or EKS/EBS qualification. The operator’s maintained integration suite is go run ./hack/integration --suite all.