The ip command as a read-first diagnostic tool before touching network configuration

What it is

ip, from the iproute2 package, queries and manipulates the kernel's live networking state: addresses, routes, links and the neighbour (ARP/NDP) table. Because it talks to the kernel directly through netlink, its output reflects reality at the moment it runs, unlike a configuration file that only reflects intent.

Why it matters

Before editing any config (netplan, NetworkManager, /etc/network/interfaces), an agent should establish what the kernel currently believes: which addresses are actually bound, which route the kernel would actually pick, and whether a neighbour is resolved. Skipping this step risks diagnosing or fixing a problem that does not exist, or missing one that a config file does not show (a manually added address, a route pushed by a routing daemon, a stale ARP entry).

How to apply

• ip -brief addr (or ip -br a): one line per interface with state and addresses — the fastest overview, and script-friendly.
• ip route get 203.0.113.10: asks the kernel which route and source address it would actually use to reach a destination, including the effect of policy routing rules — more reliable than reading the routing table by eye.
• ip neigh (or ip n): shows the current ARP/NDP table with per-entry state (REACHABLE, STALE, FAILED, PERMANENT); a FAILED entry for a gateway points at a link or ARP problem, not a routing one.
• ip -s link show eth0: adds interface statistics (RX/TX packets, errors, drops) with -s (-stats, -statistics); repeating -s increases the amount of information shown.
• ip -j ...: JSON output for scripts that need to parse the result reliably.

Pitfalls

• Any address, route or link state set with plain ip commands (ip addr add, ip route add, ip link set; these need root or CAP_NET_ADMIN, while the read-only commands above do not) lives only in the kernel's running state. It is not written to any configuration file, is lost on reboot, and can be overwritten when NetworkManager or systemd-networkd reapplies its configuration; routes through an interface are also removed when that interface goes down. Persistent changes belong in NetworkManager, netplan, systemd-networkd or the distribution's interfaces file — ip is for inspection and short-lived, explicitly temporary changes.
• ip route get reports the kernel's routing decision for a destination; it sends no packet and does not verify actual end-to-end connectivity.
• Reading /proc/net/route directly (IPv4 only, hex-encoded) is fragile compared with asking ip to interpret the routing state.