Use Cases

Real-world AIOps behavior observed during Kubernaut demo validation. These are not synthetic examples -- they capture actual LLM decisions, remediation outcomes, and pipeline behavior from live Kubernetes clusters.

Deep Dives

• Multiple Remediation Paths -- How the LLM chose an alternative fix for a GitOps-managed Certificate failure, and why both approaches are valid
• Remediation History Feedback -- How the LLM refused to repeat a failed workflow after remediation history revealed the prior attempt's failure, escalating to human review instead
• LLM Judgment and the Approval Gate -- How the LLM selected a matching cleanup workflow but warned "no remediation warranted," deferring to a human operator via the approval gate

Demo Scenario Catalog

The following scenarios demonstrate Kubernaut's remediation pipeline across different failure modes, infrastructure patterns, and pipeline behaviors. Each scenario injects a fault, triggers a Prometheus alert, and validates the full remediation lifecycle.

All scenarios are available in the kubernaut-demo-scenarios repository with run scripts, manifests, and step-by-step instructions.

Pod Lifecycle

Scenario	Signal	Remediation	Description
crashloop	KubePodCrashLooping	RollbackDeployment	Bad ConfigMap causes CrashLoopBackOff, rollback restores previous revision
crashloop-helm	KubePodCrashLooping	HelmRollback	Helm-managed workload crash, Helm rollback to last known-good release
stuck-rollout	KubeDeploymentRolloutStuck	RollbackDeployment	Bad image tag stalls rollout, rollback restores healthy revision
memory-leak	ContainerMemoryExhaustionPredicted	GracefulRestart	Proactive: predict_linear() detects memory growth before OOM
memory-escalation	ContainerMemoryHigh	IncreaseMemoryLimits	Memory usage exceeds threshold; escalates to human if limits are already high
slo-burn	ErrorBudgetBurn	RollbackDeployment	Error budget burn rate exceeds threshold, proactive rollback

GitOps

Scenario	Signal	Remediation	Description
gitops-drift	KubePodCrashLooping	GitRevertCommit	Bad ConfigMap commit in Gitea, LLM selects git revert over kubectl rollback
cert-failure-gitops	CertManagerCertNotReady	GitRevertCommit / FixCertificate	Broken ClusterIssuer via git; LLM may choose git revert or direct fix (details)
disk-pressure-emptydir	PredictedDiskPressure	AnsiblePVCMigration	PostgreSQL on emptyDir fills disk; Ansible/AWX runs pg_dump, commits PVC migration to Git, ArgoCD syncs, pg_restore completes migration

Infrastructure

Scenario	Signal	Remediation	Description
cert-failure	CertManagerCertNotReady	FixCertificate	CA Secret deleted, workflow recreates it to restore certificate issuance
hpa-maxed	KubeHpaMaxedOut	ScaleHPA	HPA at max replicas under sustained load
resource-contention	OOMKilled	IncreaseMemoryLimits	Memory contention causes OOM kills across competing workloads
resource-quota-exhaustion	KubeResourceQuotaExhausted	AdjustResourceQuota	Namespace quota prevents scaling (details)
network-policy-block	KubePodCrashLooping / KubeDeploymentReplicasMismatch	FixNetworkPolicy	Deny-all NetworkPolicy blocks traffic; readiness-based signal self-resolves after remediation
statefulset-pvc-failure	KubeStatefulSetReplicasMismatch	FixStatefulSetPVC	PVC binding failure prevents StatefulSet pod scheduling

Multi-Node

Scenario	Signal	Remediation	Description
autoscale	KubePodSchedulingFailed	AddNode	Cluster autoscaling via kubeadm join
pending-taint	KubePodNotScheduled	RemoveTaint	Node taint prevents pod scheduling
pdb-deadlock	KubePodDisruptionBudgetAtLimit	ResolvePDBDeadlock	PodDisruptionBudget prevents necessary evictions
node-notready	KubeNodeNotReady	CordonDrain	Node health failure triggers cordon and workload migration

Advanced Pipeline

Scenario	Signal	Remediation	Description
duplicate-alert-suppression	KubePodCrashLooping	RollbackDeployment	Validates that duplicate alerts for the same incident are suppressed
concurrent-cross-namespace	KubePodCrashLooping	Per-team workflow	Two teams hit the same fault; LLM selects different workflows based on risk labels
orphaned-pvc-no-action	KubePersistentVolumeClaimOrphaned	None (NoActionRequired) / CleanupPVC	Without cleanup workflow: LLM concludes no action needed. With workflow in catalog: LLM selects it but warns "no remediation warranted," deferring to human approval (details)

Service Mesh

Scenario	Signal	Remediation	Description
mesh-routing-failure	IstioHighDenyRate / IstioRequestsUnauthorized	FixServiceMeshRouting	Istio service mesh AuthorizationPolicy misconfiguration

Unvalidated

These scenarios have scaffolding (manifests, run.sh, workflow) but have not been validated end-to-end on any platform. Do not rely on them until they are promoted to a category above.

Scenario	Signal	Remediation	Description	Blocker
memory-limits-gitops-ansible	ContainerOOMKilling	AnsibleMemoryLimitsUpdate	OOMKill on GitOps-managed deployment; Ansible/AWX updates limits in Git, ArgoCD syncs	Requires ArgoCD + AWX; not tested on Kind or OCP