KVM virtual machines and LXC containers

The two kinds of server VMStack sells, what each can and cannot do, how to enable LXC containers on a node, and the container-only plan switches.

VMStack sells two kinds of server on the same nodes: KVM virtual machines and LXC system containers. Both are servers in the panel: the same plans, addresses, IP pools, firewall, traffic accounting, billing modules and API. The kind is set by the plan and never changes for a server.

KVM virtual machine LXC container
Kernel Its own Shares the node's
Operating systems Any: Linux, FreeBSD, Windows, anything installed from an ISO Linux only, from container images
Density and speed of build A full VM Lighter and denser; builds in about a minute
Isolation Hardware virtualisation Unprivileged container: its own range of user ids, AppArmor, seccomp

What a container can and cannot do#

A container's server page shows only what it can do; nothing is shown greyed out for it.

Feature Container
Create, power actions, delete, reinstall Yes
Upgrade vCPUs, memory and disk Yes, live, without a restart (the disk only grows)
Routed IPv4, IPv6, NAT with port forwards, bridged and VLAN networks Yes
Firewall, network rules, speed limits, adding and removing addresses, renaming Yes
Text console Yes
Reset root password, reconfigure network, add SSH keys Yes (the password is set without a restart)
Usage charts, traffic limits, suspension Yes
Snapshots, backups (every destination kind), restores, moves to another node Yes (a move stops the container during the copy)
Disks on file, LVM-thin and ZFS storages Yes; not on Ceph
One-click applications Those that run in a container; Docker-based ones need nesting (below)
Screen console, ISOs, boot order, rescue, hardware settings, firmware, Windows No: virtual machines only
Extra disks, dedicated CPUs, processor flags, GPU passthrough No
Cloning, making an image from the server, high availability, Proxmox import No

Each container feature needs an agent on the node that has it. On a node with an older agent the container keeps working, and its Manage tab says which features wait for an agent update. The API answers 422 kind_unsupported for an action a container cannot do.

Enable containers on a node#

A node runs containers once LXC is installed by the node installer. Every enrolled node runs KVM virtual machines; containers are an addition.

  1. Make sure the node runs the unified cgroup hierarchy (cgroup v2), as Debian 12 and 13 do by default:

    stat -fc %T /sys/fs/cgroup
    

    It must print cgroup2fs.

  2. Run the node installer's upgrade with --with-containers, on the node, as root. Use the same agent link and checksum options as your normal upgrade (Installer options):

    curl -fsSL https://panel.example.com/downloads/install.sh | bash -s -- --upgrade --agent-url https://panel.example.com/downloads/vmstack-agent --with-containers
    

    It installs lxc, lxcfs, uidmap and apparmor, and gives root the containers' id range (root:1000000000:1000000000 in /etc/subuid and /etc/subgid) when nothing there overlaps it. It disables and masks lxc-net: containers use the agent's own bridges, like virtual machines. Nothing else changes. A new node can be installed with --with-containers from the start.

  3. Check the result:

    grep root /etc/subuid /etc/subgid
    systemctl is-enabled lxc-net
    

    The first prints root:1000000000:1000000000 twice, the second masked. The panel reads the node's information again within about five minutes (at once when the agent version changed) and from then on offers the node for containers.

A failure of the LXC part is a warning, not an error: the agent keeps running and reports that the node does not run containers, with the reason. On the admin create form such a node is greyed out for the container type ("runs no containers"), and placement passes it over with the reason "does not run containers" in the placement log.

Note: If root already has an entry in /etc/subuid or /etc/subgid that overlaps the range, the installer changes nothing and prints what to set. Each container gets its own block of 65 536 ids from that range, so root in one container is nobody in another.

Container plans and images#

  1. Create a plan with Virtualization set to LXC container (Hosting plans).
  2. Add at least one container image: Images and ISOs › Images › Browse images.linuxcontainers.org, or Add image with Kind set to Container (Container images).
  3. Create a server. When both kinds can be offered, the create form starts with Type of server; with only one kind available the step is skipped. With a container plan only container images are offered, and ISOs, firmware and hardware choices are not shown.

Container-only plan switches#

An LXC plan has these fields in its General section. Device switches and nesting take effect at the container's next start: a running container's Manage tab then says "Restart needed". Swap, I/O limits, ping and max processes change at once.

Field What it does
Swap (MB) Swap the container may use beside its memory. 0 or empty: none. The form warns when nodes that run containers have no swap themselves.
Nesting (Docker inside) Lets the container run Docker and other containers. Off by default. It loosens the container's AppArmor profile: turn it on for customers you trust.
Allow VPN (TUN device) Grants /dev/net/tun (WireGuard-go, OpenVPN).
Allow FUSE mounts (rclone, s3fs, AppImage) Grants /dev/fuse.
Allow PPP VPNs (pptp/l2tp) Grants /dev/ppp.
Allow nested KVM (run VMs inside) Grants /dev/kvm. Offered only when a node that runs containers exposes /dev/kvm.
Allow ping for every user Opens ICMP ping sockets for non-root users. Root in the container can ping and open raw sockets either way.
Max processes pids.max. Empty or 0: the default, 4096.

The device switches need an agent that lists them; on an older node they are ignored and the container keeps working.

What customers see#

A container's page has a Container tag beside its status. Its Manage tab has the password reset, network reconfiguration and an SSH keys section (add keys to root's authorized_keys, or replace them). While a container is frozen for a backup copy, its status reads "Running (backing up)".

Inside, a container shows its own vCPU count (nproc, /proc/cpuinfo) and memory (free, through lxcfs), and its disk size (df -h /).

Known limits#

  • Containers are unprivileged; there is no privileged container option.
  • Container images for AlmaLinux, Rocky Linux, CentOS Stream and Oracle Linux 10 need a node whose CPU is x86-64-v3 or newer. Placement passes over older nodes with the reason "with a CPU below the image's level".
  • Container disks cannot be on Ceph RBD storage.
  • Debian 13 containers mask dev-mqueue.mount, run-lock.mount and tmp.mount, which fail in an unprivileged container; /tmp is then on the container's disk.