release: v0.1.1

[wrkode]

Chart 0.1.1 / appVersion 0.1.1, with the README and quickstart version references following.

What is in it

Everything found by installing the released v0.1.0 chart and driving a real pool through it, plus two behaviours that were burning hardware for nothing:

#13	a default install could not create a cell (secrets read-only), emitted no events (events.k8s.io ungranted), pulled an unpublished image tag, and would have labelled every alert with the operator's namespace
#15	a live cell's Node was reaped as stale while its own kubelet was using it — unrecoverable, the kubelet never re-registers
#16	cells were rebuilt five times against faults that were not in the cell
#18	this release's own new CRD fields would have been silently dropped on every upgraded cluster, taking rolling update with them
#10, #12	template drift + updatePolicy.type: RollingUpdate, and the observability pack (already on main before this release)

Plus the create/destroy loop fix a replacement cell used to hit by inheriting its predecessor's teardown.

Upgrading from v0.1.0 — read this

helm upgrade never updates a chart's crds/, and the apiserver then silently drops what the older schema lacks. Apply the CRD:

kubectl apply -f https://raw.githubusercontent.com/kubeswift-io/gpucellpool/v0.1.1/config/crd/bases/cells.kubeswift.io_gpucellpools.yaml

Skip it and updatePolicy.type: RollingUpdate is accepted and ignored. The manager logs the missing fields at startup, so you can check rather than remember. docs/upgrading.md is new and covers this, the two additive RBAC grants, and why downgrades are safe.

Verified

• make lint 0 issues, full suite green, helm lint clean
• rendered default image tag is v0.1.1, matching what the release workflow will publish (CI guards this)
• every capability in this release was exercised on the dev cluster against a real GTX 1080 cell — the node-adoption race, the pool-wide-fault stall and its recovery, and the CRD drift report in both directions

🤖 Generated with Claude Code