Changelog¶
All notable changes to the Helm chart and deploy tooling are documented here.
Format based on Keep a Changelog. Versioning: SemVer.
[Unreleased]¶
[0.13.0] - 2026-11-13¶
Changed¶
- Version aligned with the VirtFoundry 0.13.0 release.
[0.12.0] - 2026-10-16¶
Added¶
- Chart e2e has a
quickstartscenario: it runs the install commands taken from the quickstart on Kind, waits for the control plane, then runs the documented health check and logs in asroot. A change to the docs or the charts that breaks the first-time path now fails CI. - Docs: Expose the UI and API (port-forward, Ingress, Gateway API, what the proxy must pass through) and Security and credentials (where the credentials live, how to change the root password, signing key rotation).
scripts/release/release.sh:bump X.Y.Zopens the release PR in core, operator, vks, helm-charts and the org profile;tag X.Y.Ztags them in order and waits for each tag's workflows. It only rewrites known pin shapes and lists the other mentions of the old version for review (Versioning).
Fixed¶
- Configuration said credentials change when a new value is passed. A new
secrets.rootPassworddoes not change the login password after the first start: the API only uses it to createroot. scripts/setup/multus.shinstalls Multus with a 256Mi memory limit instead of the upstream 50Mi (MULTUS_MEMORYoverrides it). The thick daemon runs the delegate CNI inside its own cgroup; at 50Micilium-cniwas evicted and re-read from disk (about 100k refaults per second) and pod sandboxes hung inContainerCreatingon a loaded node (#144). If you installed Multus another way, raisekube-multus-dsto 256Mi.
Changed¶
charts/virtfoundry-operatoris now the only copy of the operator chart: the operator repository no longer carries one, and the drift gate is gone. The RBAC gate runs the operator's own check (ClusterRole against the kubebuilder markers) on this chart; the copy of that check kept here had fallen behind.- VKS docs: the chart installs from OCI (
oci://ghcr.io/virtfoundry/charts/virtfoundry-vks) instead of a git clone, with a prerequisite checklist, how to get the kubeconfig and use the cluster, whatnetworkRefmust be, and when the control-plane address is reachable. - Quickstart: starts with who it is for and what you need (cluster, KubeVirt, Multus, CDI, StorageClass,
/dev/kvm), a table of what was tested and what was not, and installs with onevirtfoundry-platformrelease after the CRDs. The three separate releases are kept as an advanced option. - Docs site redesigned (Geist type, navy header, the real logo) and the UI screenshots re-captured for the redesigned console, including a new Settings page shot (#143).
[0.11.3] - 2026-10-07¶
Changed¶
- Chart / image pins aligned with core/operator/vks 0.11.3 (core: VKS accepts the network display name,
/api/v1/healthz). virtfoundry-vkssample and VKS guide use the Network CR name (default-default) innetworkRef.
Fixed¶
- The UI nginx sent
Hostwithout the port, so WebSocket upgrades (/ws/events,/ws/console) were rejected with 403 when the UI was served on a non-default port, such as the quickstart'skubectl port-forward 8080:80. It now keeps the port ($http_host);scripts/ci/assert-ui-nginx-host.shguards it (Troubleshooting). - The UI and API pods now roll when their ConfigMaps change (checksum annotations). Before,
helm upgradechanged the ConfigMap and left the pod running with the old file, because the nginx config is mounted withsubPath. - Quickstart: the health check URL is
/api/v1/healthz(needs 0.11.3 or newer;/healthon the UI address only returned the UI page), and the first-VM steps now say to create an SSH key first. scripts/e2e/homelab-ui.pyworks on a fresh install (creates a throwaway SSH key) and fails on WebSocket errors in the browser console.
[0.11.2] - 2026-10-07¶
Added¶
scripts/e2e/homelab-ui.py: UI end-to-end on a live install (wizard deploy, post-deploy panel, Open VM, cleanup). It found that Copy ssh did nothing over plain HTTP (fixed in core 0.11.2).scripts/ops/wait-settled.sh: waits until pods, Deployments and (when installed) Argo CD Applications are ready, quiet for a minimum time. The release process now says to release one repository at a time and run it in between (Versioning).
Changed¶
scripts/setup/multus.shpins the Multus image to the immutablev4.3.1-thickdigest instead of the movingsnapshot-thicktag (override withMULTUS_IMAGE). Troubleshooting explains how to pin an existing install.- Chart / image pins aligned with core/operator/vks 0.11.2 (core: honest login page and Copy ssh over HTTP).
Security¶
scripts/deploy/homelab.shno longer prints a fixed login andscripts/docs/capture-ui-shots.pyreads the password fromVF_PASSWORDinstead of embedding one. The old value is in the Git history, so treat it as exposed and rotate it on any install that used it.
[0.11.1] - 2026-10-07¶
Added¶
- Chart e2e workflow: fresh install, umbrella install and the CRD migration from the 0.10.x operator chart on a Kind cluster (
scripts/ci/e2e-charts.sh). - Troubleshooting: the
helm install virtfoundrypost-install timeout without KubeVirt, and pods stuck inContainerCreatingfrom Multus throttling (Troubleshooting).
Changed¶
- UI screenshots refreshed for 0.11.0; dependency updates (GitHub Actions, mkdocs-material).
- Chart / image pins aligned with core/operator/vks 0.11.1.
[0.11.0] - 2026-10-07¶
Added¶
charts/virtfoundry-crds: the 16virtfoundry.ioCRDs in one chart, upgraded byhelm upgradeand protected fromhelm uninstalland Argo CD prune.make verify-crds-chartgate. See CRDs and upgrades.charts/virtfoundry-platform: umbrella chart for the operator, core and optional VKS in one release. Kamaji is not a dependency.make verify-platform-paritygate.- Docs: Kubernetes clusters (VKS), Terraform provider, Troubleshooting, CRDs and upgrades, split Networking/Storage configuration, copy-paste install; refreshed UI screenshots; mermaid, glightbox, breadcrumbs and cards.
Changed¶
- Breaking: the
virtfoundry-operatorchart (mirror of virtfoundry/operator) and the VKS chart no longer ship CRDs. Installvirtfoundry-crdsfirst, then the operator. Existing installs adopt their CRDs once: see CRDs and upgrades. - Chart / image pins aligned with core/operator/vks 0.11.0.
[0.10.0] - 2026-10-06¶
Added¶
- Isolated bridge host address (
platform.networking.isolated.bridge.address) and a unique public bridge IP per node. - API ClusterRole can manage
vksclusters. - Operator VAP allowlist includes
ghcr.io/virtfoundry/(VKS node image). - Docs: operator recovery for legacy tenant namespaces; future umbrella chart design.
- CI: API roles checked against core's RBAC contract; operator chart drift check;
requireDigestgate restored.
Changed¶
- Operator chart synced from the upstream source of truth.
- Chart / image pins aligned with core/operator 0.10.0.
Security¶
- CodeQL, Scorecard, Dependabot (grouped monthly) and dependency review.
[0.9.0] - 2026-10-01¶
Added¶
- Instance CRD:
spec.powerStatefor CRD-first Start/Stop;cloudInitUserData/cloudInitSecretRefsync from operator. - Optional public bridge DHCP (
platform.networking.public.dhcp.enabled); dnsmasq as root +NET_BIND_SERVICE. - Operator ClusterRole sync for Network NAD and sshkeys; VAP / kubevirt mutate RoleBinding security slices.
Changed¶
- Chart / image pins aligned with core/operator 0.9.0.
Security¶
- Pin chart CI actions and CDI/Multus URLs.
[0.8.0] - 2026-09-28¶
Changed¶
- Chart / image pins aligned with core 0.8.0 (UI day-2 polish). No chart schema change.
0.7.3 - 2026-09-25¶
Fixed¶
- Control-plane NetworkPolicy DNS egress allows NodeLocalDNS (port 53 without CoreDNS-only selectors) so the UI can resolve
virtfoundry-apion Kubespray clusters.
0.7.2 - 2026-09-25¶
Security release aligning charts with core/operator 0.7.2 (PSS hardening, networking/Ingress defaults, secrets fail-closed).
Security¶
- Harden API/UI Deployments (closes #42): non-root + drop ALL + RuntimeDefault seccomp + readOnlyRootFilesystem, requests/limits, UI SA without token automount, Service
80→8080via nginx ConfigMap. Default NetworkPolicy for control plane (networkPolicy.enabled); setallowedIngressNamespacesto restrict. Residuals: env-injected secrets; API egress open — follow-ups on the issue. Core images: UI listens on 8080 / API UID 65532. - Breaking:
platform.networking.isolated.enableddefaults tofalseso a default install is API+UI-only (no hostNetwork bridge DaemonSet). Enabling isolated or public opts into host bridges with documented caps — closes #40. See Host bridges. Kind / homelab overlays that need Multus L2 must setisolated.enabled: true(and usuallybridge.tolerations: [{operator: Exists}]). - Bridge DaemonSet hardening: dedicated SA without token automount, no
hostPID,NET_ADMIN/NET_RAWinstead ofprivileged: true, alpine pinned by digest. Residual: DHCP container stillapk add dnsmasqat runtime (#54). - Breaking:
ingress.enableddefaults tofalse(no cleartext HTTP control plane). Enabling Ingress requiresingress.tlsoringress.allowCleartext: true. Gateway profile usessectionName: websecure; HTTP→HTTPS redirect example at httproute-https-redirect.yaml — closes #41. See Ingress and TLS. api.security.allowedOrigins→ ConfigMapsecurity.allowed_origins(CORS //ws/*Origin allowlist) — closes #51, residual from core#98 / PR #113. Default[]is correct for same-origin UI proxy; split UI/API must list UI origins. See Allowed origins.-
Document CDI importer egress NetworkPolicy in each tenant namespace (created by core
EnsureTenantNamespace; not a release-NS chart object) — closes #49, follow-up to core#95. See Images and templates and example private-mirror policy. -
Breaking: the
virtfoundrychart ships no credential defaults.secrets.rootPassword(min 12 chars) andsecrets.jwtSecret(min 32 chars) are required, and the published sentinelsvirtfoundry/change-me-in-productionare refused — install andhelm templatefail closed instead (#37, core#93) - Platform hook RBAC no longer grants
apiGroups: ["*"] / resources: ["*"]. Each hook Job has its own ServiceAccount:-platform-kubevirt(get/patch onkubevirt/kubevirtonly),-platform-multusand-platform-cdi(rendered only withplatform.multus.install/platform.cdi.install) - Hook RBAC carries
helm.sh/hook-delete-policy: before-hook-creation,hook-succeeded, so no platform identity outlives the install/upgrade
Upgrading from 0.7.1 or older leaves the previous wildcard objects behind — Helm never owned them. Remove them once:
kubectl delete clusterrolebinding virtfoundry-platform --ignore-not-found
kubectl delete clusterrole virtfoundry-platform --ignore-not-found
kubectl -n virtfoundry-system delete serviceaccount virtfoundry-platform --ignore-not-found
Added¶
secrets.existingSecret(withsecrets.rootPasswordKey/secrets.jwtSecretKey) — bring your own Secret; recommended for Argo CD and other GitOps flowssecrets.autoGenerateJwtSecret— random 48-char JWT secret on first install, preserved across upgrades vialookupsecrets.allowInsecureDefaults— local-development escape hatch that also setsVF_ALLOW_INSECURE_DEFAULTS=1on the API
Changed¶
helm upgradewithout--set secrets.*reuses the credentials already stored in the live Secret, so upgrades no longer reset the root password or invalidate issued tokens- Documented in Configuration — Secrets
0.7.1 - 2026-09-04¶
Added¶
- Chart values — full default
values.yamlon the docs site, why--setbelongs on the Helm command platform.storage.defaultClass: auto— select Longhorn when the StorageClass existsplatform.networking.public.autoFromCluster— fill CIDR/gateway/pool from Node InternalIP when you do not--setthemscripts/detect-host-public-net.sh— print a public-net values snippet from kubectl/host
Changed¶
- Charts and default
appVersion0.7.1(core UI hides platform Windows ISO from VM create until the tenant uploads one)
0.7.0 - 2026-09-02¶
First tagged CRD-store / operator chart line (no separate v0.6.0 / chart 0.6.0 tag was ever published).
Added¶
virtfoundry-operatorHelm chart with bundledvirtfoundry.io/v1alpha1CRDsstore.driver=kubernetesprofile documented as default install path- Platform prerequisites — install links for KubeVirt, Multus, CDI, Longhorn, MetalLB, CSI snapshotter
Changed¶
- Breaking (0.x): removed MySQL StatefulSet, worker Deployment, and
values-kubernetes.yamloverlay - Charts and default
appVersion0.7.0 - Default images
ghcr.io/virtfoundry/{core,ui,operator}:0.7.0 - Install docs list KubeVirt, Multus, CDI prerequisites before Helm commands
- Homepage and quickstart point to prerequisites guide
- VM snapshot UI screenshots and CRD store terminal shot refresh
Removed¶
- Legacy MySQL store templates and worker chart resources
0.5.0 - 2026-08-16¶
Pre-1.0 release line. VirtFoundry is not 1.0 yet. Premature v1.0.0–v1.5.0 / virtfoundry-1.x tags and releases were deleted (2026-09-27); they are not a stability contract and must not appear in the Helm index.
Changed¶
- Chart
version/appVersionand default images set to0.5.0 - Docs and install snippets pin
--version 0.5.0
Docs¶
- Kind laptop guide (no VLAN); public-network underlay without a switch
Absorbed (from deleted premature 1.x chart line)¶
- IAM release, default VPC, volumes, offerings, templates/ISO, dedicated CPU, Features docs, volume-delete 409, UI polish — previously published as chart/app
1.0.0–1.5.0
0.2.0 - 2026-08-02¶
Added¶
- Public network profile: bridge CNI NAD, configurable IP pool and gateway, optional dnsmasq bridge address
platform.networking.public.bridge.addressandroutePoolViaBridgefor host route steering when multiple bridges share a CIDRallowPodNetwork: truedefault for VM pod + public dual-homed networking- MkDocs documentation site published to GitHub Pages
/docs/
Changed¶
- Platform bridge DaemonSet: optional bridge IP + VM pool host routes via the public bridge
- Default container images bumped to
0.2.0
Fixed¶
- Public VM networking: bridge Multus without host-local IPAM (guest IP via cloud-init pool)
0.1.0 - 2026-08-01¶
Added¶
- Initial Helm chart: API, worker, UI, MySQL
- Gateway API and Ingress profiles
- Platform networking values (
platform.networking.public, isolated bridge) - GitHub Pages Helm repository via chart-releaser
- Deploy scripts and setup helpers (KubeVirt, Multus, CDI)