Not everything can run on spot
Spot instances are cheap, but they can be reclaimed with short notice. Stateless, fault-tolerant workloads handle this fine. Databases and singleton controllers do not.
Kubeadapt analyzes your workloads and tells you which ones are safe candidates for spot migration, and why.
How it works
The analyzer evaluates every workload against eligibility criteria:
<!-- callout:warning -->Automatic Blockers
The following workloads are excluded from spot recommendations:
- StatefulSets with persistent storage
- System-critical pods (kube-system, monitoring)
- Single-replica deployments without a PodDisruptionBudget
- Workloads with node affinity to on-demand node groups
Eligibility scoring (for non-blocked workloads):
- Replica count (more replicas = safer on spot)
- Restart tolerance (low restart count = stable under interruption)
- Resource profile (CPU-heavy workloads benefit more)
- Current on-demand cost (higher cost = larger savings)
What you see
Each candidate shows its current monthly cost, projected savings, eligibility score with reasoning, and any partial risk warnings. Blocked workloads are listed separately with the specific exclusion reason.
Go to Available Savings in your cluster and click Spot Instance Migration to get started.