scitix/gateway-diagnostics

Gateway Diagnostics

Inspect and test the gateway of a network interface. Two operations, one script each; both run identically on a node or inside a pod — the only difference is how you launch them (see Targets below).

Router

| Need | Action | | --- | --- | | See the gateway / default route for an interface (with TYPE) | scripts/show-gateway.sh | | Test whether the gateway is reachable | scripts/ping-gateway.sh | | Target is a node (or a bound SSH host) | run directly — see Node target | | Target is a pod | resolve the netns first — see Pod target | | Switchdev PF caveat, or what RoCE/Ethernet/IB means | read references/switchdev-and-interface-types.md |

Targets

Node target

Prefer host_script when the node is a bound SSH host (check host_list) — it runs over SSH with no debug pod. Fall back to node_script otherwise. Both take the same skill / script / args.

host_script: host="<host>", skill="gateway-diagnostics", script="show-gateway.sh", args="<args>"
node_script: node="<node>", skill="gateway-diagnostics", script="show-gateway.sh", args="<args>"

Pod target

Runs in the pod's network namespace using host tools. Two steps:

1. Resolve the pod's network namespace:

   resolve_pod_netns: pod="<pod>", namespace="<ns>"
   

2. Run the script with the returned node and netns:

   node_script: node="<node>", netns="<netns>", skill="gateway-diagnostics", script="ping-gateway.sh", args="<args>"
   

The script does not enter the namespace itself — node_script does that via the netns= param. The same script file serves both node and pod targets.

Parameters

show-gateway.sh

| Parameter | Required | Description | |-----------|----------|-------------| | --interface IFACE | no | Network interface name. If omitted, shows all. | | --json | no | Output in JSON format (adds a type field per gateway). |

ping-gateway.sh

| Parameter | Required | Description | |-----------|----------|-------------| | --interface IFACE | yes | Network interface to find the gateway for | | --source-ip | no | flag — auto-detect IP from the interface, use as ping source (-I <ip>) | | --source-dev | no | flag — use the interface name as ping source (-I <iface>) | | --count N | no | Number of ping packets (default: 3) |

--source-ip and --source-dev are mutually exclusive flags (no value needed).

Examples

Show all gateways on a node:

node_script: node="nodepool-061", skill="gateway-diagnostics", script="show-gateway.sh"

Show one interface's gateway on a bound host:

host_script: host="nodepool-061", skill="gateway-diagnostics", script="show-gateway.sh", args="--interface eth0"

Ping a node's gateway, sourcing from the interface IP:

node_script: node="nodepool-061", skill="gateway-diagnostics", script="ping-gateway.sh", args="--interface bond0 --source-ip"

Show / ping a pod's gateway (resolve netns first):

resolve_pod_netns: pod="rdma-pod", namespace="rdma-test"
→ node="worker-1", netns="abc123"

node_script: node="worker-1", netns="abc123", skill="gateway-diagnostics", script="show-gateway.sh", args="--interface net1 --json"
node_script: node="worker-1", netns="abc123", skill="gateway-diagnostics", script="ping-gateway.sh", args="--interface net1 --source-dev"

Interface types & switchdev

The output includes a TYPE column (RoCE / Ethernet / IB). If a target interface is a PF in switchdev mode, gateway routes are not meaningful and ping will always fail — this is expected, not a network problem. For the detection logic and what to do instead (use a RoCE pod + the pod target), read references/switchdev-and-interface-types.md.