Changelog
What's new in Kubaros. The appliance updates itself — see Updates in your console.
v0.7.3
AlphaJune 2026- Cluster provisioning is now resilient to leftover SDN config from earlier installs. Proxmox SDN subnet IDs are unique per zone (e.g. kubaros-10.64.0.0-22), so a subnet from a previous test install could block a fresh cluster with a 500 'already defined' error. Kubaros now detects this and, if the subnet is already on the right network, simply reuses it.
- New Settings → Provisioning → 'SDN subnet conflicts': choose how Kubaros handles a leftover subnet that sits on a different VNet in its own 'kubaros' SDN zone. 'Ask me' (default) fails the cluster with a clear message so you can review it; 'Auto clean up' removes the leftover (only ever inside Kubaros' own zone, never your other networks) and recreates it on the correct VNet — then retry.
- SDN zone and VNet creation are now fully idempotent too — re-running provisioning on an existing setup no longer errors on objects that already exist.
v0.7.2
AlphaJune 2026- Setup wizard reworked for the Remote Connector: IP pools are now configured BEFORE the connector, so you can either enter a fixed public IP for the connector (recommended — no pool needed) or have one auto-booked from a pool. The wizard now blocks with a clear hint if you provide neither.
- Bridge fields now have proper descriptions, and the tenant/internal bridge is auto-detected — you only pick the WAN bridge.
- You can now correct the connector's WAN/internal bridge after creation: 'Change bridges & redeploy' rebuilds the edge VM onto the new bridges (same public IP).
- The connector 'Test' now also reports real node reachability over the tunnel (not just the WireGuard handshake), so you can see end-to-end whether the appliance reaches your Talos nodes.
v0.7.1
AlphaJune 2026- Remote Connector hardened for real-world ISPs (hub-and-spoke): the appliance is now the WireGuard hub that LISTENS, and each site's edge VM dials in to it. Many providers filter inbound UDP on a routed/secondary public IP block — which used to break the handshake — but the appliance's own IP accepts inbound UDP, so the tunnel now comes up reliably. Open inbound UDP 51820 to your Kubaros appliance (configurable via KUBAROS_WG_PORT).
- Faster tunnel keepalive (5s, configurable via KUBAROS_WG_KEEPALIVE): some ISPs only allow inbound UDP to the edge VM for a short window after it sends out, so a frequent keepalive keeps the appliance→node path continuously open. Verified against a real ISP where the previous 25s value left gaps.
- The edge VM now reliably attaches a NIC into your tenant network at build time, so it can route the appliance to the isolated Talos nodes immediately after a connector provision or rebuild.
v0.7.0
AlphaJune 2026- Remote Connector now reaches your clusters end-to-end: the appliance brings up its OWN WireGuard tunnel to each Proxmox site, so a remotely-hosted Kubaros can actually provision and manage the isolated Talos nodes (apply config, bootstrap, kubeconfig, metrics). Previously only the edge VM side of the tunnel existed.
- New "Rebuild connector VM" action (Proxmox → Remote Connector): re-creates the per-site edge VM from the current image — qemu-guest-agent, WireGuard and routing/NAT for your active tenant subnets — while keeping the same public IP. Use it to upgrade older connector VMs in place.
- Live tunnel status: the connector test now reports the real WireGuard handshake age.
- Edge-VM reliability: NICs are matched by MAC (so the static WAN config always lands on the right interface), boot no longer stalls ~2 min on systemd-networkd-wait-online, qemu-guest-agent install is retried, and the serial console now auto-logs-in as root for diagnostics.
v0.6.0
AlphaJune 2026- Static node IPs by default — no DHCP required: Kubaros assigns each Talos node a fixed IP from the cluster subnet and hands it to the node via a NoCloud seed (network-config + machine config). Nodes boot already reachable, so provisioning no longer waits on a DHCP lease and works in networks that have no DHCP server at all. Pods/Services keep using the in-cluster CNI (Flannel) and Kubernetes — they never needed DHCP.
- DHCP via Proxmox SDN (dnsmasq) is still available as a selectable mode (Settings → Provisioning → IP mode). Requires the dnsmasq package on every PVE node (and the default instance disabled).
- Configurable Talos nocloud image source + decompression for static mode, for maximum compatibility across Proxmox setups.
v0.5.0
AlphaJune 2026- Per-cluster internet gateways with bandwidth tiers: pick a speed (50 / 100 / 250 / 500 / 1000 / 10000 Mbit, or your own tiers) when creating a cluster or any time from its Gateway panel. Kubaros NATs the cluster to the internet and shapes it to the chosen tier — no tier means the cluster stays internal/isolated. Bill the tiers in your own system; Kubaros stores no prices.
- One edge VM per site, reused: the gateway runs on the SAME per-site edge VM as the Remote Connector — so adding gateways to more clusters never spends extra system VMs or public IPs (one VM + at most one public IP per Proxmox site).
- WAN bridge per Proxmox connection: set the internet-uplink bridge (default vmbr0) for the edge VM, since it can differ on each Proxmox.
v0.4.0
Alpha- Observability: a Metrics tab on every cluster shows live per-cluster and per-node CPU, memory and pod usage — also available via the REST API for your own dashboards.
- etcd backups & restore: take point-in-time etcd snapshots of the control plane, schedule them (daily/weekly) and restore after a failure — straight from the cluster's Backups tab.
- White-label branding: set your product name, logo and accent colour (and optionally hide "Powered by Kubaros") for the console your customers log into — the public website stays Kubaros.
- Multiple Proxmox sites: connect any number of Proxmox clusters and choose the default site for new clusters.
- Kubernetes install option: a Helm chart now ships alongside the VM/compose install, and appliance images are multi-arch (amd64 + arm64).
v0.3.0
AlphaJune 2026- Usage-based billing (pay-as-you-go): switch the appliance from fixed quotas to OVH-style metered billing — tenant limits become advisory and never block provisioning.
- Optional tenant quotas: leave any limit empty for unlimited (clusters, nodes and storage).
- Per-tenant usage report with CSV export — ready to feed your billing system.
- Public IP booking & MetalLB: book public IPs from your pools and assign them to clusters — Kubaros installs and configures MetalLB (Layer-2) so LoadBalancer services hand out those addresses. Tenants can self-serve within a max-public-IPs quota.
- Remote Connector mode: for a centrally-hosted Kubaros, it provisions a tiny WireGuard connector VM on the Proxmox site (one public IP) that securely tunnels into the internal Talos networks — no manual setup. Each site is either Local Appliance or Remote Connector.
- Per-tenant network isolation: every tenant gets a dedicated VLAN and node subnet from a provider supernet, with generous /16 pod & service CIDRs by default (a /24 pod CIDR is far too small) — configurable globally and per cluster.
- Management network: define the network your Talos nodes use (auto-SDN with DHCP+SNAT, or an existing bridge) and verify from the console that the appliance can reach it.
- Stronger tenant isolation: tenants get a scoped console, create clusters without touching infrastructure, and can no longer see other tenants' usage or your Proxmox settings.
- More reliable updates: the appliance checks the container registry directly, so the version it offers is always one you can actually install — plus 'kubaros check' to compare your version against the latest.
v0.2.0
AlphaJune 2026- First-run setup wizard: create the admin, connect & validate Proxmox (version, single-server-vs-cluster, SDN, Ceph) and register public IP ranges.
- License server & activation: buy a plan on the website, paste the key in your appliance to unlock its tier.
- Installation registry: every appliance phones home (opt-out via KUBAROS_TELEMETRY=off) so installs get a Community licence and can be assigned to a customer on activation.
- Public IP pools bound to each Proxmox site — the basis for LoadBalancer IPs.
- Robust cluster deletion (a cluster can always be removed, even if Proxmox is unreachable).
- Self-updater with stable/beta/edge channels and host-side apply.
v0.1.0
FoundationMay 2026- Initial appliance: dashboard, Proxmox connection & discovery.
- Talos cluster provisioning, kubeconfig download, live job logs.
- Multi-tenancy: tenants, projects and role-based access control.
- REST API with scoped tokens, usage reporting and audit logging.