Download Latest Version metal-amd64.iso (543.5 MB)
Email in envelope

Get an email when there's a new version of cozystack

Home / v1.6.2
Name Modified Size InfoDownloads / Week
Parent folder
README.md < 22 hours ago 8.1 kB
v1.6.2 source code.tar.gz < 22 hours ago 7.7 MB
v1.6.2 source code.zip < 22 hours ago 10.5 MB
openapi.json < 23 hours ago 2.7 MB
cozypkg-checksums.txt < 23 hours ago 564 Bytes
cozypkg-windows-arm64.tar.gz < 23 hours ago 21.7 MB
cozypkg-darwin-amd64.tar.gz < 23 hours ago 25.4 MB
cozypkg-darwin-arm64.tar.gz < 23 hours ago 23.4 MB
cozypkg-linux-amd64.tar.gz < 23 hours ago 24.1 MB
cozypkg-linux-arm64.tar.gz < 23 hours ago 21.6 MB
cozypkg-windows-amd64.tar.gz < 23 hours ago 24.3 MB
initramfs-metal-amd64.xz < 23 hours ago 155.2 MB
kernel-amd64 < 23 hours ago 20.4 MB
nocloud-amd64.raw.xz < 23 hours ago 349.0 MB
metal-amd64.raw.xz < 23 hours ago 349.0 MB
metal-amd64.iso < 23 hours ago 543.5 MB
cozystack-operator-hosted.yaml < 23 hours ago 2.8 kB
cozystack-operator-generic.yaml < 23 hours ago 2.9 kB
cozystack-operator-talos.yaml < 23 hours ago 2.9 kB
cozystack-crds.yaml < 23 hours ago 23.5 kB
Totals: 20 Items   1.6 GB 0

v1.6.2 (2026-08-19)

A patch release with six fixes covering the backup-strategy controller, kube-ovn's webhook certificate, Velero CRD upgrades, CNPG barman-cloud backups, flux-shard-operator, and the published OpenAPI definitions, plus a release-pipeline reliability fix.

Fixes

  • fix(backupstrategy-controller): repair lookup-gated backup objects: The default backup Strategy CRs and the Velero BackupStorageLocation are gated on a Helm lookup performed while the referenced object is still being created; when that lookup came back empty the objects were skipped permanently, since helm-controller does not re-render a release whose chart and values are unchanged. The gate now resolves the default bucket credentials Secret through the RESTMapper, bounds each check, and tolerates an absent Secret instead of looping, so the default backup objects are created reliably instead of silently vanishing for months (@mattia-eleuteri in [#3524], backport [#3731]).

  • fix(kube-ovn): reload kubeovn-webhook serving certificate on cert-manager renewal: kube-ovn-webhook loaded its TLS serving certificate once at startup and never re-read it; once cert-manager renewed the backing Secret and the old certificate expired, the apiserver's calls to the webhook failed verification and, because the MutatingWebhookConfiguration uses failurePolicy: Fail, every pod creation in tenant namespaces was rejected — including virt-launcher pods, blocking VMI startup. The webhook now serves its certificate through a reloading callback that re-reads the key pair when the mounted files change and widens renewBefore to 720h, so cert-manager renewals are honored without a pod restart (@IvanHunters in [#3557], backport [#3730]).

  • fix(velero): apply CRD updates on upgrade via CreateReplace: Velero's CRDs stayed frozen at whatever version was first installed, since Helm never touches a chart's crds/ directory on upgrade; when the Velero image moved to a version that added new backup phases, the apiserver rejected phase transitions against the stale CRDs and backups silently stopped while the HelmRelease stayed green. The Velero package now opts into upgradeCRDs: CreateReplace, so CRDs are kept current on upgrade and backups keep working (@lexfrei in [#3727], backport [#3728]).

  • fix(backups): request S3 checksum only when required for barman-cloud (non-AWS S3 / Ceph RGW): CNPG's barman-cloud plugin sidecar defaulted to computing a flexible checksum on every upload, which several S3-compatible backends (Ceph RGW, some MinIO / Cloudflare R2 builds) reject outright, so every backup and WAL-archive upload to those backends failed and ScheduledBackups never stored anything. Every barman-cloud ObjectStore Cozystack creates — Keycloak's system DB, the postgres app's backup and recovery stores, and the platform-managed system-bucket store — now sets AWS_REQUEST_CHECKSUM_CALCULATION=when_required, a safe default accepted by both AWS S3 and the affected backends (@androndo in [#3417], backport [#3767]).

  • fix(flux-shard-operator): repair sharded helm-controller crashloop behind an HTTP proxy: The cloned helm-controller-shard<i> Deployment inherited HTTP_PROXY/HTTPS_PROXY/NO_PROXY from the flux-aio all-in-one wiring even though a standalone shard needs no external egress; behind an unreachable proxy the controller's blocking startup HTTPS call never completed, the manager never served /healthz, and every HelmRelease sharded to that controller was frozen. The sanitisation now also drops the inherited proxy env and adds a startupProbe derived from the liveness handler, so sharded HelmReleases keep reconciling in proxied environments instead of crashlooping forever (@IvanHunters in [#3546], backport [#3818]).

  • fix(api): declare OpenAPIModelName for core and sdn types: The core and sdn API groups did not declare OpenAPIModelName the way the apps group already did, so their published OpenAPI definition names were Go import paths while every $ref pointing at them escaped each slash — the two spellings never matched, the reference dangled, and kubectl apply --validate failed on any resource against a cozystack-api built after the underlying Kubernetes 0.35 change. Declaring OpenAPIModelName for core and sdn too makes every published definition name the dotted Kubernetes model name, so client-side validation against the published OpenAPI works again (@myasnikovdaniil in [#3808], backport [#3812]).

Development, Testing, and CI/CD

  • ci(release): complete the candidate-aware promotion pipeline on release-1.6: release-1.6 was missing the e2e and packages-verification jobs that Promote RC requires on its target base, so v1.6.1 was promoted with the rc e2e gate bypassed and the next patch release could not even be dispatched. Adds the rc-e2e job, the verify-release-candidate checks, hack/verify-promoted-packages.sh, hack/validate-changelog.sh and regression tests pinning the pipeline's contract, so future patch releases off release-1.6 run the same e2e and package-verification gates as main before promoting, and the tag-time changelog is validated and ported from the tag rather than regenerated (@myasnikovdaniil in [#3893]).

Documentation

  • [website] docs: import the operator guides that lived in the cozystack repo: Moves the operator-facing guides that used to live in the cozystack repo over to the documentation site, so operators find them alongside the rest of the docs instead of scattered across two repositories (@myasnikovdaniil in cozystack/website#648).

  • [website] docs(oidc): document private CA and staging trust: Documents how to configure tenant OIDC to trust a private certificate authority and staging certificates, closing a gap for operators running their own CA or testing with a staging issuer (@myasnikovdaniil in cozystack/website#650).

  • [website] feat(community): add a Community page and link it from the main menu: Adds a Community page linked from the site's main menu, giving visitors a single place to find how to get in touch with and contribute to the Cozystack community (@tym83 in cozystack/website#637).

  • [website] chore(telemetry): publish July 2026 and explain how the figures are derived: Publishes the July 2026 telemetry figures and documents how those figures are derived, giving the community visibility into adoption trends and how the numbers are calculated (@tym83 in cozystack/website#644).

  • [website] feat(blog): new Blockstor banner: Adds a new banner promoting Blockstor to the blog, improving the visibility of the storage control plane's announcement (@tym83 in cozystack/website#646).

Contributors

Thanks to everyone who contributed to this patch release:

Full Changelog: https://github.com/cozystack/cozystack/compare/v1.6.1...v1.6.2

Download cozystack

Source: README.md, updated 2026-08-19