01
Nodes taking minutes to arrive
Pending pods waiting while Auto Scaling groups slowly add nodes during a traffic surge.
Overprovisioned Kubernetes nodes and slow autoscaling
Kubernetes autoscaling and compute cost fix with Karpenter
Fix slow pod scheduling and idle nodes: set accurate resource requests, replace Cluster Autoscaler with Karpenter, and run interruptible workloads on spot.
Symptoms
Signs your platform has this problem
If several of these sound familiar, the plan below is where we would start.
02
Paying for idle nodes
Large on-demand instances running around the clock at low utilization.
03
OOM kills from untuned limits
Pods crashing because memory requests and limits are missing or wrong.
Remediation plan
How we fix it, step by step
Each phase ends with a measurement, so you can see what changed before the next one starts.
01
Pod resource profiling
Measuring historical CPU and memory use (for example with Goldilocks) to set accurate requests and limits.
02
Karpenter
Deploying Karpenter on Amazon EKS to launch right-sized nodes directly for pending pods.
03
Spot diversification
Using several instance families for spot capacity and handling the two-minute interruption notice gracefully.
04
Disruption budgets and topology spread
Spreading pods across Availability Zones and keeping enough of them available while nodes are replaced.
Technical checklist
Remediation checklist
What we check before a change goes to production:
- Measure real pod memory use and set accurate resource requests
- Deploy Karpenter with price-capacity-optimized spot allocation
- Configure Pod Disruption Budgets to keep a minimum number of pods available
- Handle spot interruptions with graceful drains
What we measure
We take a baseline first and report the same measurements after each change, from your own monitoring — evidence, not promised results.
- Node cost
- Monthly node spend for the same workloads
- Pending time
- Pod pending to running during scale-up
- Request accuracy
- Requested vs used CPU and memory per workload
Related service
Cloud & DevOps
Cloud cost optimization, Kubernetes platforms, and CI/CD that make deploys boring — savings and reliability measured in your dashboards, not our deck.
Explore Cloud & DevOpsQuestions
Questions about this remediation
Cluster Autoscaler works through Auto Scaling groups. Karpenter bypasses them and launches right-sized EC2 instances directly through the EC2 Fleet API.
AWS sends a two-minute interruption notice; Karpenter cordons and drains the node and starts a replacement, while Pod Disruption Budgets keep enough replicas serving.
Related playbooks
Other engineering problem playbooks
Near-zero-downtime database migration playbook
Move a production database without a long maintenance window: stream changes with CDC, verify parity with shadow reads, and cut over in a short, rehearsed window with reverse replication ready for rollback.
Startup MVP build and launch
Ship the smallest product that proves your core workflow: scoped with you up front, built on a boring, scalable stack, and handed over with documentation and tests.
SaaS multi-tenant architecture with Row Level Security
Move to a pooled PostgreSQL design where Row Level Security enforces tenant isolation in the database, cutting per-tenant overhead and the risk of a missed WHERE clause.
Double-entry ledger and event sourcing architecture
Build an append-only, double-entry ledger with idempotent writes and ordered events, so every balance can be explained and rebuilt from its history.
Want an engineer to look at this with you?
Send us the symptoms and any metrics you have. We'll reply within one business day, set up a call and agree what to measure before anything changes.