Skip to content

resource-tuning.md

Latest commit

聽

History

History
107 lines (74 loc) 路 3.43 KB

resource-tuning.md

File metadata and controls

107 lines (74 loc) 路 3.43 KB
Raw

Resource Tuning with Goldilocks

Overview

Goldilocks provides resource recommendations based on historical usage data collected by VPA. It helps right-size Kubernetes workloads for cost optimization and performance.

Architecture

Goldilocks consists of two components:

  • Controller: Watches namespaces with the goldilocks.fairwinds.com/enabled: "true" label and creates VPAs for each workload
  • Dashboard: Web UI for viewing recommendations across namespaces

Enabling Goldilocks for a Namespace

kubectl label namespace <namespace> goldilocks.fairwinds.com/enabled="true"

Accessing the Dashboard

The Goldilocks dashboard is available at https://goldilocks.gistpin.io.

Dashboard Features

  • View resource recommendations per deployment
  • Filter by namespace
  • Export recommendations
  • View historical trends

VPA Recommendation Mode

Goldilocks creates VPAs in Off mode (recommendation only). This means:

  • VPAs do not automatically adjust resources
  • Recommendations appear in VPA status
  • Apply recommendations manually after review

Viewing Recommendations via CLI

# For a specific deployment
kubectl get vpa goldilocks-backend-vpa -n gistpin -o yaml

# Extract target recommendations
kubectl get vpa goldilocks-backend-vpa -n gistpin \
  -o json | jq '.status.recommendation.containerRecommendations'

Applying Recommendations

Manual Apply

bash infrastructure/scripts/apply-right-sizing.sh --namespace gistpin

Automated Apply (Staging Only)

For staging environments, recommendations can be auto-applied:

MODE=apply ENVIRONMENT=staging bash infrastructure/scripts/apply-vpa-recommendations.sh

Recommendation Tiers

Tier CPU Request Memory Request Description
Burstable 25m-100m 64Mi-256Mi Low-traffic services
Standard 100m-500m 256Mi-1Gi Medium-traffic services
Performance 500m-2 1Gi-4Gi High-traffic services
Critical 2+ 4Gi+ Core platform services

Cost Optimization

Goldilocks recommendations help identify over-provisioned resources:

# Compare current requests vs recommendations
kubectl get vpa goldilocks-backend-vpa -n gistpin \
  -o json | jq '{current: .status.recommendation.containerRecommendations[0].target, lower: .status.recommendation.containerRecommendations[0].lowerBound, upper: .status.recommendation.containerRecommendations[0].upperBound}'

Example Savings

Workload Current Recommended Monthly Savings
backend 1CPU/2Gi 250m/512Mi ~$45/instance
frontend 500m/1Gi 125m/256Mi ~$25/instance

Best Practices

  1. Run Goldilocks for 7+ days before applying recommendations
  2. Review recommendations during off-peak hours
  3. Apply changes gradually (25% of workloads at a time)
  4. Monitor application performance after applying changes
  5. Use HPA alongside VPA recommendations
  6. Re-check recommendations quarterly

Troubleshooting

Issue Resolution
No recommendations showing Ensure namespace has goldilocks.fairwinds.com/enabled: "true" label
Dashboard unreachable Check ingress and goldilocks-dashboard pod logs
VPA not created Check goldilocks-controller logs: kubectl logs -n goldilocks deploy/goldilocks-controller
Recommendations incomplete Ensure metrics-server is installed and collecting data