kube-network-detector
Find which Kubernetes node receives traffic for a virtual IP, or which nodes can reach a TCP endpoint. The detector records the result as node labels, which you can use in pod scheduling rules.
It runs on each node as a DaemonSet, using the node's network (hostNetwork: true). IPv4 and IPv6 are supported.

Two things it can check
| Check | What happens | Example label |
|---|---|---|
| VIP routing | An elected observer probes the VIP and identifies the responding node | kube-network-detector/203.0.113.10=true |
| Peer connectivity | Each node tries a TCP connection to a named IP and port | kube-network-detector/peer-database.default=true for a Peer named database in namespace default |
You can use either check or both. A successful TCP connection means the endpoint is reachable; it does not check application health.
By default, two consecutive successes add a label and three consecutive failures remove it. Other node labels are left alone.
Start here
- Quickstart: install the DaemonSet and declare your first target.
- Configuration: change probe timing, use static targets or enable dry-run.
- Troubleshooting: investigate missing labels and failed probes.
The detector observes routing. It does not configure interfaces, move VIPs or set up your firewall.