Terminal
Terminal/Skill
Runs bounded packet captures across AKS nodes and collects Azure network settings, to prove where traffic is dropped when logs are inconclusive.
By Microsoft
Works with
npx skills add azure/aks-skills --skill aks-network-captureCapture packet-level evidence on AKS when read-only diagnostics are inconclusive. Use this to prove where a packet is dropped — inside the pod, on the node, at an NSG, on a route, or at the Azure load balancer.
This is an escalation tool. For most networking symptoms (DNS, connectivity, ingress 502s), start with aks-troubleshooting; come here when you need a pcap or wire-level proof.
Packet capture requires elevated node access, so these scripts are built to be safe by construction:
tcpdump as a single trailing argument (never a shell string), and compile-checked in-pod with tcpdump -d. There is no eval. A negative regression test (evals/tests/aks-network-capture/injection.test.sh) proves malicious inputs are rejected.privileged. Capture pods use only NET_ADMIN + NET_RAW with hostNetwork. They mount only /var/log/aks-network-captures from the node, never the node root, and never enable hostPID.:latest.The live-cluster smoke test is manual and is not run in CI. Before relying on distributed capture in production, run
evals/tests/aks-network-capture/smoke-live-cluster.shagainst an AKS cluster you control; it uses an isolated namespace, generates bounded same-node DNS traffic, retrieves the exact run, decodes packet records, and verifies Kubernetes plus host-artifact cleanup.
| Script | Purpose |
|---|---|
scripts/setup-capture-configmap.sh | Deploy the fixed capture runner to a namespace (run once per namespace) |
scripts/create-capture.sh | Start a bounded capture Job on the selected nodes/pods |
scripts/retrieve-captures.sh | Pull capture bundles off the nodes and clean up |
scripts/generate-test-traffic.sh | Drive test traffic from a pod's netns while capturing |
scripts/collect-azure-network-info.sh | Collect the Azure-side network config for the cluster |
# Install the fixed capture runner in the namespace
./scripts/setup-capture-configmap.sh default
# Capture DNS traffic across all Linux nodes for 2 minutes
./scripts/create-capture.sh --name dns-debug --tcpdump-filter "udp port 53" --duration 120s
# Capture for specific pods (resolves each pod to its host node and narrows to the pod IPs)
./scripts/setup-capture-configmap.sh production
./scripts/create-capture.sh --name frontend --pod-selector "app=frontend" --namespace production
# Retrieve and clean up
./scripts/retrieve-captures.sh --name dns-debug
create-capture.sh targeting flags: --node-selector, --node-names, --pod-selector, --pod-names (with --namespace). The same namespace contains the ConfigMap and capture Jobs; pass it to retrieve-captures.sh --namespace as well. Creation prints the run ID; pass --run-id during retrieval when a capture name has multiple runs. Capture flags: --duration (max 30m), --packet-size, --tcpdump-filter. The filter must be a plain BPF expression — no flags, no shell metacharacters.
A dropped packet often dies in the Azure network layer, not in Kubernetes. After (or instead of) a capture, run scripts/collect-azure-network-info.sh and check, in order:
az network nic list-effective-nsg --ids <node-nic-id>.az network nic show-effective-route-table --ids <node-nic-id> -o table.When pod-to-pod (east-west) traffic works but egress to external Azure services fails, the fault is almost always in the Azure network layer, not the cluster CNI. Investigate in order, before concluding it is DNS:
Deny rules blocking the destination port/service tag (Sql, Storage, AzureCloud). Most common cause; inspect first.0.0.0.0/0 route to a firewall/NVA can black-hole or redirect PaaS-bound traffic. Confirm the effective routes on the subnet.scripts/collect-azure-network-info.sh gathers the NSG rules, route tables, firewall config, and endpoint state to identify which layer blocks the flow.
When you have the pcap and the Azure config, map the path and mark where it breaks:
graph LR
A[Source Pod<br/>10.244.1.5] -->|1. veth| B[Node netns]
B -->|2. iptables/CNI| C[Routing]
C -->|3. NSG / UDR| D[Azure fabric]
D -->|4. FORWARD allow| E[Dest Pod<br/>10.244.2.8:8443]
Highlight the first hop where the packet is absent in the capture or denied by an NSG/route rule — that is the failure domain.
tcpdump -r <file>); confirm whether the traffic was seen at all.collect-azure-network-info.sh output).generate-test-traffic.sh during a fresh capture.