{
    "product": "Cenvero Stratum",
    "generated_at": "2026-11-16T13:19:23+00:00",
    "format": "cenvero-docs-v1",
    "document_count": 1,
    "documents": [
        {
            "slug": "networking/cluster-networks",
            "title": "Cluster Networks",
            "category": "Networking",
            "url": "https://stratum.cenvero.com/docs/networking/cluster-networks",
            "headings": [
                {
                    "level": 1,
                    "text": "Cluster Networks"
                },
                {
                    "level": 2,
                    "text": "When a member takes machines on a network"
                },
                {
                    "level": 2,
                    "text": "Creating a cluster network"
                },
                {
                    "level": 2,
                    "text": "Where a network runs"
                },
                {
                    "level": 2,
                    "text": "What reaches what"
                },
                {
                    "level": 2,
                    "text": "Addresses and the MTU"
                },
                {
                    "level": 2,
                    "text": "Attaching machines"
                },
                {
                    "level": 2,
                    "text": "Deleting a cluster network"
                },
                {
                    "level": 2,
                    "text": "Listing and reading"
                },
                {
                    "level": 2,
                    "text": "Who can do what"
                },
                {
                    "level": 2,
                    "text": "See also"
                }
            ],
            "word_count": 1588,
            "markdown": "# Cluster Networks\n\nA **cluster network** is one network stretched over the members that carry it. You\ncreate it once, for the cluster, and a machine on one member can reach a machine of\nthe same network on another member. Create a network on a single node when its\nmachines live on that node only; create a cluster network when they live on several\nmembers.\n\n> Part of [Networking](/docs/networking/overview) and [Clustering](/docs/clustering/overview).\n> Creating and changing a cluster network needs clustering in the licence of the node\n> you call; a machine on one needs `compute`. Reads are never refused.\n\n## When a member takes machines on a network\n\nA member takes machines on a cluster network once it holds two checks, and it answers\nfor itself:\n\n- **Only the network's own members are accepted.** The member accepts the network's\n  traffic only from the members that carry it, and drops a frame that comes from any\n  other address.\n- **Only the network's own tenant is reached.** The member makes the network's\n  connection a port of its tenant's machines, isolated as that tenant's port, so the\n  network's traffic reaches that tenant's machines on the member.\n\nUntil a member holds both checks, a machine that asks for the network is refused with\n**\"This network cannot take machines yet.\"** A member that holds both but cannot\nisolate the network's connection refuses the network with **\"This server cannot keep\nthe network apart from the tenants' machines, so it cannot take this network.\"**\n\nA member cannot check the network's traffic on every path when its overlay address is\non a loopback or dummy device, is an IPv6 address, or is missing, or when the `nft` tool\nis not installed. The installer installs that tool, so it is missing only on a server\nwhere someone removed it, or one updated without running the installer. Such a member\ndoes not take the network at all. It refuses the network with **\"This member cannot check\nits overlay traffic, so it cannot take a cluster network. Its overlay address must be an\nIPv4 address on a network interface, and the nft tool must be installed.\"** The member's\nown operator networks on that address are kept, and run without the check; the\n[networking overview](/docs/networking/overview) says what that means.\n\nThe answers say which case applies. The list of cluster networks, and each network's\nown answer, carry `\"released\": true` on a member that can take machines on them, and\n`\"released\": false` on one that cannot. A network's `state` is `ready` once it can take\nmachines, and `awaiting_check` until then; the console shows the same state as\n**ready** or **awaiting check**.\n\n## Creating a cluster network\n\n```bash\nsudo cenvero-str-ctl cluster network create --name web-tier --cidr 10.20.0.0/24 \\\n  --gateway 10.20.0.1 --tenant t-acme --host-gateway --dhcp\n```\n\nOver the API, `POST /api/v1/cluster/networks` takes `{\"name\", \"cidr\", \"gateway\",\n\"tenant_id\", \"mtu\", \"host_gateway\", \"dhcp\"}` and answers **202** with the network. Only\nthese fields are read: a field the route does not take is refused. If the network is\ncreated and its host gateway cannot be set, the answer is still **202** with the network,\nwhich then has no host gateway, and `host_gateway_error` says so; set the host gateway on\nthe network to finish, rather than creating it again. A tenant that was removed is\nrefused as one this server does not know.\n\nThe rules are the ones the answer gives when a value breaks them:\n\n- **The name** is 1 to 64 printable characters.\n- **The address range** is an IPv4 block in its canonical form, between `/20` and\n  `/30`. **\"The network's address range must be an IPv4 block in canonical form.\"** and\n  **\"The network's address range must be between /20 and /30.\"**\n- **The range may not overlap** another cluster network's: **\"This network's address\n  range overlaps another cluster network's, so it cannot be created.\"**\n- **The gateway** is a host address inside the range: **\"The network's gateway must be a\n  host address inside its range.\"**\n- **Every cluster network belongs to one tenant.** Only that tenant's machines attach to\n  it. A network with no tenant is refused with **\"The network names no tenant it can\n  belong to.\"** A tenant this server does not know is refused with **\"This tenant is not\n  known to this server.\"**\n- **The MTU** is between 576 and 9000 (see below).\n- **DHCP needs the host gateway.** **\"A network's DHCP is served by its host gateway,\n  so DHCP needs a host gateway.\"**\n- **A host gateway cannot turn DHCP off.** **\"A cluster network's host gateway serves\n  DHCP wherever this member's DHCP service is on, so DHCP cannot be set to false with the\n  host gateway.\"** An omitted `dhcp` is accepted, and so is `true`.\n\nThe cluster gives each network an identifier, and a **number on the overlay (VNI)**\nthat no other cluster network has. The number appears in the network's answer for\noperators who can read the network.\n\n## Where a network runs\n\nA member takes part in a network when its first machine attaches to the network\nthere. It leaves the network **ten minutes after its last running machine on it\nstops**, so a machine that stops and starts again within that time keeps the member in\nplace.\n\nThe **host gateway** is the one member that answers as the network's gateway and,\nwhen DHCP is on, serves its addresses. A network has at most one host gateway, so a\nsingle segment never has two gateways answering for one address. A member whose DHCP\nservice is off still answers as the gateway, and the network's DHCP follows once that\nservice is turned on; until then the member's view of the network says why DHCP is not\nserved. Set or move it:\n\n```bash\nsudo cenvero-str-ctl cluster network host-gateway <network-id> --member edge-1\nsudo cenvero-str-ctl cluster network host-gateway <network-id> --clear\n```\n\nOver the API: `PUT /api/v1/cluster/networks/{id}/host-gateway` with `{\"node\":\n\"<member>\", \"enabled\": true}` answers **202**. It may also name `\"dhcp\"`, and `false` is\nrefused as above. Setting a host gateway on a network\nwhose gateway address is not set is refused: **\"A host gateway needs the network's\ngateway address, so give the network a gateway first.\"** A member that runs the host\ngateway keeps its place in the network until the host gateway is moved or cleared:\n**\"The member runs this network's host gateway. Clear the host gateway first.\"**\n\n**There is no failover for the host gateway.** If its member is down, the network has\nno gateway until you move the host gateway to another member. The cluster does not do\nthat for you.\n\n## What reaches what\n\nOn each member, the network's traffic reaches the machines of its own tenant. Another\ntenant's machines receive none of it, and another tenant's traffic never enters the\nnetwork. Machines with no tenant are not separated at their port, as on any node (see\n[Current limits](/docs/compute/limits)).\n\nTraffic between members is **not metered per tenant**.\n\nOverlay traffic between members is **not encrypted**. Carry it on an underlay that\nreaches only the members.\n\n## Addresses and the MTU\n\nThe guest MTU of a cluster network defaults to the cluster's default, **1500**, but\neach member lowers it to what its own underlay can carry: the underlay's MTU less\n**50 bytes** of overlay overhead. On a standard 1500-byte underlay, the default is\n**1450**. A machine on a network whose MTU is below 1500 receives that MTU in its\nfirst-boot network settings, so a cloud-image guest uses it.\n\nA member whose underlay cannot carry the MTU you ask for refuses the network with\n**\"This member's network cannot carry the MTU this network needs.\"** The MTU must be\nbetween **576** and **9000**; otherwise **\"The network's MTU must be between 576 and\n9000.\"**\n\n## Attaching machines\n\nA machine attaches to a cluster network on the member that runs it, and its tenant\nmust be the network's tenant: **\"This machine belongs to a different tenant from the\nnetwork, so it cannot attach to it.\"** On a member that does not yet take machines on\nthe network, the attach is refused with **\"This network cannot take machines yet.\"**\n\n## Deleting a cluster network\n\n```bash\nsudo cenvero-str-ctl cluster network delete <network-id>\n```\n\nOver the API: `DELETE /api/v1/cluster/networks/{id}` answers **200** when the network\ngoes. It is refused with **409** and the sentence **\"Machines are attached to this\nnetwork. Move or remove them first.\"** while any of these holds: a machine uses the\nnetwork; a member still holds it (members leave ten minutes after their last machine\nstops); or the network has a host gateway. Clear the host gateway, and wait out the\nten minutes, before you delete.\n\n## Listing and reading\n\n`cluster network list` and `cluster network show <id>` give each network's name, range,\ngateway, tenant, members, host gateway, machine count, state and the `released` flag.\n\nIn the console, Datacenter › Networks lists the networks with their members, host\ngateway, machine count and state. Its actions are **Create a cluster network**, **Save\nthe host gateway** and **Delete the network**, each offered only where the server\ncarries it out.\n\n## Who can do what\n\n| Action | Privilege | Licence |\n|---|---|---|\n| List and read a cluster network | `Network.Audit` | none (reads are never refused) |\n| Create or delete a cluster network | `Network.Allocate` | `cluster` |\n| Set the host gateway | `Network.Modify` | `cluster` |\n\nA tenant's key cannot use these routes; they answer **403**.\n\n## See also\n\n- [Networking Overview](/docs/networking/overview) — networks on one node, and the manual setup between nodes.\n- [Placing machines](/docs/clustering/placement) — how a machine finds a member that can take its networks.\n- [Clustering Overview](/docs/clustering/overview) — what a cluster shares, and what stays with a node.\n- [CLI Reference](/docs/cli#cluster) and [API Reference](/docs/api#cluster).\n"
        }
    ]
}