Skip to main content
A node rarely has exactly one address. The provider issues an additional IP, someone takes a subnet, someone needs customers to come to one address and the node goes outside from another. Network In the node card, the node shows what is actually on the node, and allows you to dispose of it.
For whom: panel administrator, SRE. Required rights: reading - viewing nodes, requesting nodes and binding addresses - operations with nodes, roles - editing nodes.
The address is read and changed by the agent. On the node without the agent, the tab says it and doesn’t suggest buttons that won’t work. There’s no SSH backup path here.

Inventory

The address list is brought in by the agent on request of the panel. The panel itself does not follow it, so it always says next to it when it was last viewed, and the Ask Node button re-reads the list again, and nothing changes on the node. Each address has an interface, a mask, a gateway, and where it came from:
  • The address of the panel is where the panel goes on the node. You can’t take it off, because the node is connected to it.
  • This address is the panel, which is your own, added from here, and that’s the only panel you can shoot.
  • is issued automatically - arrived by DHCP or SLAAC and can change itself.
  • Propal is an address that is no longer on the node, and it doesn’t disappear from the list, but it stays with the time of the last meeting: the provider that relocated the node is exactly the event for which the inventory is started.
An empty list is never presented as fact: “Asked and no addresses are left,” “node hasn’t been asked yet,” and “no one to ask” are three different states, and the tab distinguishes between them. A failed query doesn’t mark anything missing: the testimony remains from the last check, which is written right on the card. Below the inventory are two reference cards: which route the node goes to the Internet by default and, if any, its own routing rules and tables. The field in the dialog is one: the address that the provider issued. The interface, mask, gateway and the need for a separate routing table, the panel outputs from the inventory and immediately shows in words what will happen. A typical case is a neighboring address from the same subnet: there is already a route for it on the node, there is nothing to add. The second question appears where the kernel has no physical place to get an answer: an address from someone else’s subnet, then the panel asks for a gateway and substitutes the one that the node uses itself, and also gives the address a separate route so that the answers leave the same way they came in.
The dashboard configures the address on the node itself. If the ISP wants to link the address to its server (Hetzner, OVH, etc.), it is done in its panel, otherwise the address will be registered, but will not work.

How does it apply?

The application is two-phase, and this is protection against node loss:
1

Node's applying addresses live

The agent puts the addresses in the working kernel and starts rolling back on the timer.
2

The panel knocks a second time.

The second approach is the communication check: the task was passed on the same network that has just been changed.
3

Configuration is written to disk

Only then do the addresses become permanent.
They didn’t get through – the node will return everything on its own, and the tab will honestly say that they rolled back. A separate line shows if the addresses will survive the reboot. If the panel does not find a familiar network configuration on the node, the addresses will work until the reboot, and this is clearly stated: the address that lives before the reboot breaks the circuit exactly when it is already hoped for. The “Check” button re-reads from the node exactly the addresses that the panel requested and compares to what was requested, a discrepancy usually means manually editing on the server or rebooting.

Roles: Entry and exit

The role tells you why the diagram needs an address. | Role of | What does | do? |---|---| | Entrance | Clients connect to it, is inserted into listen at inbound and goes to the client’s config instead of the | administration address. | Output | From it, the node goes to the Internet. Substituted to outbound direct traffic Xray | | Input and Output | Both | | Without the role of | Address is just on the node | Roles come into play when the circuit with this node is applied again. The node’s own traffic (updates, curl, agent) does not change the role: they are about customer traffic. The role can only be assigned to an address that is actually on the node, and a diagram attached to an advertised but missing address would compile cleanly and serve no one: inbound would not go up, and the exit would not be reached.

Reserve.

A second address with the same role becomes a backup, not a traffic drawer, and that’s what it’s designed to do: when you mark the second exit, the operator prepares a backup instead of redirecting live traffic. Switching is a separate action, the “Switch to this” button: addresses are swapped in places in a single transaction, and two people don’t have a role at once.

What the panel will say out loud

The compiler of the circuit rejects what will not work, and explains the reason:
  • The block with the address-role should contain exactly one node. The bundle is compiled into one config profile, which all its nodes receive. listen With that address, there’s a command they can’t execute: Inbound won’t go up, and the node will silently stop accepting customers.
  • Direct output in the no scheme means the output address is specified by the WARP or proxy, and not the role.
  • Output addresses are marked by several The first one works; it is not yet possible to divide the traffic between them.
  • listen hand-painted This is the decision of the operator, and the role does not interrupt it.

Has anyone checked that the node is coming out of the designated address?

Assigning a role and making sure that the node is using it are different things. The routing rule might not stand up, the provider could select the address. So the panel separately looks at who the node is mistaken for by the Internet: the card Access to the Internet in the node card shows the observed address and the exit country read through the tunnel of the circuit itself. Details are in the Levels of ChecksZ.

Restrictions

  • Node without agent feature is not available in its entirety.
  • On an unfamiliar network stack, addresses will stand only in the living core: they will not be permanent, and the tab will tell you about it.
  • The panel shows other people’s addresses, but it doesn’t remove them, so you can only shoot what she wrote.
  • It is not yet possible to divide traffic between several output addresses of one node: the first one takes the entire direct output.

Typical errors

The list remains with the previous check, and there is nothing to change the addresses. Start with the tab Agent: is he alive, is heartbeat coming.
Check in the ISP’s dashboard if the address is tied to this server. The panel does its half of the work, the other half can’t.
After the change of addresses, the panel could not reach the node, and the node rolled back on its own, make sure that the administration address is not affected and that the new address does not conflict with the existing route.
Roles only get to the node with a circuit. Re-use Flow.
Related Pages: Overview nod, Marshrouting, Verification Levels, Metrics and Services.