v0.1.3
Bug-fix release with one CI test-infrastructure addition.
Fixed
- CSI storage-only auto-mkfs —
localStorageClass (no DRBD, single replica) now reaches Pod-ready end-to-end. The satellite was leaving storage-only resources unformatted because the mkfs path was wired only for the DRBD bring-up sequence; linstor-csi then failedNodeStageVolumewithwrong fs type. The reconciler now formats the backing block device on the storage-only path before exposing it to the kubelet, and thecsi-pvc-locale2e scenario is restored to gate the contract. Also wires the missingPOST /v1/resource-definitions/{rd}/resources/{node}alias (linstor-csi v1.10.1 issues this single-node create — pre-fix the apiserver returned HTTP 405). - Orphan-witness collapse grace removed — the controller used to keep a TieBreaker witness alive for several reconcile cycles after the last diskful peer left, on the theory that a fresh diskful might race in. In practice the grace window only ever surfaced as stuck
Offpeers afterr d+ immediater con the same node. The collapse is now instant: when no diskful peer remains and no fresh placement is pending, the witness is deleted in the same reconcile.
Test infrastructure
tests/e2e/lib.shDS-converge barrier —reset_cluster_statenow waits up to 90s for the satellite DaemonSet to converge (desiredNumberScheduled == numberReady) before declaring the cluster reset. Pre-fix, fast successive scenarios occasionally started against a not-yet-rolled satellite and the failure looked like a flaky test rather than a missed barrier.
Images for this tag are published as 1.33-style tags by docker/metadata-action's semver rules under:
ghcr.io/cozystack/blockstor-controller:0.1.3ghcr.io/cozystack/blockstor-apiserver:0.1.3ghcr.io/cozystack/blockstor-satellite:0.1.3