Networking and public addresses
The project's private network, public IPv4 and IPv6 addresses, getting out to the internet, and L2 networks.
The project's private network
When you switch on the cloud, the project gets a private IPv4 network of its own. Every server sits on it with a private address (Private address on the server page): servers in the same project talk to each other there, without going through the internet. No other customer, and none of your other projects, can see that network.
You can see the network prefix in Nerd mode, under network, and in the API in Cloud/getStatus (cidr).
On the private network, everything coming from the project's own network is let in (and ping): you do not need to open ports between your servers. The servers' firewall is for what comes from outside.
Public addresses
A server created with Public IPv4 address switched on gets its own public address:
- routed straight to the server: no NAT, no port forwarding. The server sees its own public address;
- closed until you open a port in the firewall;
- where the public network also has IPv6, the server gets an IPv6 address too (public_ip6), with the same firewall;
- it has a generic name in DNS, like
31-44-45-12.gk01.ch, which you can replace with your own: Reverse DNS.
The Public addresses table (Normal mode, Servers section) lists the project's addresses, IPv4 and IPv6, with the server they are assigned to and the public network they come from (Network).
A public address is charged by the hour even while the server is stopped, as long as the server exists. When you delete the server the address goes back to the platform and may be given to someone else: you cannot keep it aside.
Public networks
Addresses come from one or more public networks open to your project. If there is more than one, you choose which when creating the server, in Public network; otherwise the panel tells you which one it comes from (From …). If a network has no free addresses left the panel says so (no free addresses): you can create the server as private only, or write to us.
Reaching the internet without a public address
A private-only server (Private only) can be reached from the project's other servers. It can get out to the internet only if the project has a way out (the project's NAT gateway, which the platform sets up where the zone provides it). Without a way out, the server cannot download updates and, with today's images, is created without an SSH server (see Virtual machines).
The NAT gateway belongs to the platform: you do not see it among your servers and it does not count towards your quota. Its public address may appear in the Protection section, marked NAT gateway, if it is attacked.
The same applies to Kubernetes nodes without a public address (see Kubernetes).
L2 networks
An L2 network is a layer 2 segment between the project's servers, even on different hosts, which can be extended all the way to the customer's office (the same segment in the office and in the cloud, without a VPN to set up by hand).
Today:
- they are managed with the API, the MCP and the assistant; the panel does not have a page for them yet;
- they are switched on zone by zone: where the zone does not offer them, creating an L2 network answers not supported;
- the link to the office is set up by our staff on the network equipment: ask us.
How they work:
- create: a name, the segment's IPv4 prefix (from /8 to /30; if the network reaches an office, the office's prefix) and, if needed, the segment's gateway (the office router or one of your servers), which the platform never assigns to a server. The prefix must not overlap the project's other networks;
- attach a server: it gets an extra network interface, without being stopped, with an address on the network (the lowest free one, or the one you choose: with an office on the segment it is best to choose it). The new interface must be configured inside the server; the response gives the address, prefix and gateway. A server created afterwards configures it by itself;
- a server on the L2 network only transmits with its own MAC address and from its own addresses, plus the ones you declare (up to 16, for example a shared virtual address or networks the server routes);
- everything coming from the segment is let in, ping included;
- detach removes the interface from the running server; delete only works for a network with no servers.
The project's routed network (the private one, type: routed) appears in the list of networks but is neither created nor deleted from here.
For developers
| What | Method | MCP tool | Role |
|---|---|---|---|
| The project's public addresses | Cloud/getPublicIpList |
cloud_get_public_ip_list |
viewer |
| Public networks open to the project | Cloud/getPublicIpPoolList |
cloud_get_public_ip_pool_list |
viewer |
| The project's networks (routed and L2) | Cloud/getNetworkList |
cloud_get_network_list |
viewer |
| One network, with its attached servers | Cloud/getNetwork (network_id) |
cloud_get_network |
viewer |
| Create an L2 network | Cloud/createNetwork (name, cidr, gateway?) |
cloud_create_network |
operator |
| Delete an L2 network | Cloud/deleteNetwork (network_id) |
cloud_delete_network |
operator |
| Attach a server | Cloud/attachServerToNetwork (network_id, server_id, address?, allowed_addresses?) |
cloud_attach_server_to_network |
operator |
| Detach a server | Cloud/detachServerFromNetwork (network_id, server_id) |
cloud_detach_server_from_network |
operator |
All writes accept dry_run (Plans). Lists come back in items. Each public address has address, address6 (or null), server_id, pool.