AKS Troubleshooting

• lifecycle, access, node, kube-system, workload, ingress, DNS, or scaling issues
• kubectl cannot connect, nodes are NotReady, or pods are unhealthy

Scenario Playbooks

Tool Selection For Diagnostics

When gathering AKS diagnostic evidence, prefer mcp_azure_mcp_aks, then the smallest discovered AKS-MCP tool that fits the read, then supporting Azure tools such as mcp_azure_mcp_applens, mcp_azure_mcp_monitor, or mcp_azure_mcp_resourcehealth. Use raw az aks and kubectl only when the AKS-MCP surface cannot perform the needed check.

When standard diagnostics do not reveal root cause, use Inspektor Gadget for real-time, low-level node and pod observability (DNS traces, TCP traces, process snapshots, file access traces). See references/inspektor-gadget.md for the gadget catalog, command patterns, and symptom-to-gadget mapping.

See references/aks-mcp.md, references/structured-input-modes.md, references/command-flows.md

Required Inputs

• subscription or active Azure context
• resource group and cluster name
• symptom summary
• first observed time or recent change window
• impacted namespace, workload, service, or ingress when known

If cluster identity is missing, stop and ask for it.

Scope Buckets

• Lifecycle: create, update, start, stop, upgrade, or provisioning failures
• API access: kubeconfig, auth, private endpoint, DNS, or reachability problems
• Nodes: missing nodes, NotReady, pressure, CNI, kubelet, certificate, or VMSS drift
• kube-system: CoreDNS, metrics-server, konnectivity, ingress, CNI, CSI, or add-on failures
• Workloads: Pending, CrashLoopBackOff, OOMKilled, PVC, quota, secret, readiness, or dependency issues
• Connectivity and DNS: pod -> service -> endpoints -> ingress/load balancer -> DNS -> network controls
• Scaling: node pool sizing, pending pods, autoscaler config, metrics, quota, or subnet constraints

Evidence Order

1. Azure-side state first: cluster state, resource health, recent operations, node pool state, detector or monitoring output.
2. Kubernetes-side state second: cluster reachability, nodes, kube-system, events, affected namespace, pod detail, logs.
3. Use detector, warning-event, or metrics modes when the incoming data already matches them.
4. Deep diagnostics; when steps 1–3 do not reveal root cause, use inspektor-gadget.md for real-time tracing and snapshots on the affected node.

Workflow

1. Get cluster context.
2. Classify the problem by scope bucket.
3. Prefer Azure-side evidence before Kubernetes-side evidence.
4. Use the matching AKS-MCP path first, then the documented CLI fallback if MCP cannot perform that read.
5. Return evidence, failure domain, confidence, next checks, remediation, and escalation.

Error Patterns

• No cluster context: ask for subscription, resource group, and cluster name.
• MCP unavailable: fall back to safe az aks and kubectl reads.
• kubectl blocked: separate auth problems from network reachability.
• Logs or metrics missing: use events, node state, and resource descriptions.
• Detector noise: ignore emergingIssues, prefer critical findings, rank the most actionable signal first.

Safe Fallback Checks

When AKS-MCP cannot perform the baseline read, run the aks-baseline script. It executes the read-only cluster + Kubernetes baseline sweep (provisioning state, node pools, activity log, node readiness, unhealthy pods, kube-system health, warning events) and returns a single labeled digest:

Then deep-dive on a specific pod as the digest indicates:

Keep these read-only unless the user explicitly asks for remediation.

Guardrails

• default to read-only diagnostics
• do not restart, delete, cordon, drain, scale, upgrade, or reconfigure resources unless the user explicitly asks for remediation
• do not conclude root cause without quoting the evidence that supports it

Output Checklist

Return scope and impact, evidence, failure domain, root cause, confidence, next checks, remediation, and escalation.