OpenCost

OpenCost provides visibility into current and historical Kubernetes spend and resource allocation.

Introduction

OpenCost is a CNCF sandbox project that provides real-time cost visibility and insights for Kubernetes workloads. It enables teams to understand, monitor, and optimize their Kubernetes spending at the namespace, deployment, and pod levels.

OpenCost is the open-source foundation of Kubecost, offering core cost allocation functionality without the enterprise features.

Key Features

Cost Monitoring

• Real-time cost tracking at cluster, namespace, deployment, and pod level

• Historical cost data with trend analysis

• Support for AWS, GCP, Azure, and on-premises

• Custom pricing configuration

Resource Cost Breakdown

• CPU cost allocation

• Memory cost allocation

• GPU cost allocation

• Storage (PVC) cost allocation

• Network egress cost tracking

Multi-Cluster Support

• Federation across multiple Kubernetes clusters

• Unified cost reporting

• Cluster-level comparison

Integration

• Prometheus metrics endpoint

• Grafana dashboard integration

• Cloud provider APIs for accurate pricing

• OpenTelemetry support

Architecture

OpenCost consists of:

• Cost Model: Core calculation engine that processes Prometheus metrics

• Allocation Service: Distributes costs across namespaces and pods

• Prometheus Adapter: Exposes custom metrics for cost data

• Exporter: Collects and exports cost metrics

Installation

Helm Installation

# Add OpenCost helm repo
helm repo add opencost https://opencost.github.io/opencost-helm-chart
helm repo update

# Install OpenCost
helm install opencost opencost/opencost -n opencost --create-namespace

kubectl Installation

# Deploy OpenCost
kubectl apply -f https://raw.githubusercontent.com/opencost/opencost/develop/kubernetes/opencost.yaml

Configuration

Prometheus Configuration

OpenCost requires Prometheus. Ensure your Prometheus has the following scrape config:

- job_name: 'opencost'
  scrape_interval: 1m
  scrape_timeout: 10s
  metrics_path: '/metrics'
  scheme: 'http'
  static_configs:
    - targets: ['opencost.opencost.svc:9003']

Cloud Provider Integration

AWS

opencost:
  config:
    aws:
      enabled: true
      spotDataEndpoint: "https://spot-data-cloudjoin.s3.amazonaws.com"
      spotDataBucket: "my-spot-data-bucket"
      spotDataPrefix: "spotdata"

GCP

opencost:
  config:
    gcp:
      enabled: true
      projectID: "my-gcp-project"
      billingDataDataset: "my-billing-dataset"

Azure

opencost:
  config:
    azure:
      enabled: true
      subscriptionID: "my-subscription-id"
      tenantID: "my-tenant-id"
      clientID: "my-client-id"
      clientSecret: "my-client-secret"

Accessing OpenCost

# Port-forward to access UI
kubectl port-forward -n opencost svc/opencost 9090:9090

Access the UI at: http://localhost:9090

Key Metrics

Namespace Costs

sum(opencost_namespace_cost) by (namespace)

Pod Costs

sum(opencost_pod_cost) by (pod, namespace)

Cost by Service

sum(opencost_service_cost) by (service)

CPU Cost

sum(opencost_cpu_cost) by (cluster)

Memory Cost

sum(opencost_memory_cost) by (cluster)

GPU Cost

sum(opencost_gpu_cost) by (cluster)

Storage Cost

sum(opencost_pvc_cost) by (cluster, persistentvolume)

Cost Allocation

By Namespace

OpenCost automatically allocates costs based on:

• Resource requests

• Resource limits

• Actual usage

• PVC usage

Using Labels

Cost allocation can be customized using Kubernetes labels:

metadata:
  labels:
    environment: "production"
    team: "platform"

Shared Costs

Configure shared costs for cluster-level resources:

opencost:
  config:
    # Cluster management costs to distribute
    clusterManagementCost: "50.00"
    # Percentage of shared costs to allocate
    sharedCosts:
      - name: "load-balancer"
        cost: "100.00"

Prometheus Queries

Daily Cost by Namespace

sum(increase(opencost_namespace_cost_total[24h])) by (namespace)

Monthly Cost Trend

sum(increase(opencost_cluster_cost_total[30d])) by (cluster)

Cost Efficiency Ratio

sum(opencost_pod_cost) / sum(opencost_pod_efficiency)

Grafana Integration

Import Dashboard

Import the official OpenCost Grafana dashboard from: https://grafana.com/dashboards/14191

Custom Queries

# Total Daily Cost
sum(increase(opencost_cluster_cost_total[24h]))

# Top 10 Most Expensive Namespaces
topk(10, sum(opencost_namespace_cost) by (namespace))

# CPU vs Memory Cost Split
sum(opencost_cpu_cost) by (namespace)
sum(opencost_memory_cost) by (namespace)

Alerting

Prometheus Alert Rules

groups:
  - name: opencost.alerts
    rules:
      - alert: HighNamespaceCost
        expr: sum(opencost_namespace_cost) > 1000
        for: 1h
        labels:
          severity: warning
        annotations:
          summary: "High cost detected for namespace {{ $labels.namespace }}"
          description: "Namespace {{ $labels.namespace }} cost is above $1000/day"

      - alert: CostAnomaly
        expr: rate(opencost_cluster_cost_total[1h]) > 1.5 * rate(opencost_cluster_cost_total[24h])
        for: 30m
        labels:
          severity: critical
        annotations:
          summary: "Cost anomaly detected"
          description: "Cluster cost is growing faster than normal"

ETL Pipeline

OpenCost uses an ETL (Extract, Transform, Load) pipeline for cost data:

1. Extract: Collects data from Prometheus and cloud APIs

2. Transform: Applies pricing models and allocations

3. Load: Stores processed data for querying

Build ETL Pipeline

# Build ETL pipeline manually
curl -g -X POST http://opencost/opencost-backend/etl/build

Update ETL Data

# Force update ETL with latest data
curl -g -X POST http://opencost/opencost-backend/etl/forceUpdate

Troubleshooting

Common Issues

Missing Metrics

If metrics are missing, check:

• Prometheus is correctly configured to scrape OpenCost

• Prometheus has the correct role permissions

• Network policies allow communication

# Check OpenCost metrics endpoint
kubectl exec -n opencost deploy/opencost -- curl localhost:9003/metrics

Incorrect Pricing

If pricing seems wrong:

• Verify cloud provider credentials

• Check custom pricing configuration

• Ensure spot data is configured (if using spot instances)

High Memory Usage

OpenCost stores historical data in Prometheus:

• Adjust retention period

• Configure resource limits

• Enable downsampling for historical data

Comparison with Kubecost

Feature	OpenCost	Kubecost
Pricing Model	Open source (CNCF)	Freemium (Enterprise paid)
UI	Basic UI	Full-featured web UI
Enterprise Support	Community only	Yes
Cloud Integrations	Limited	Native
SAML/SSO	No	Yes
Multi-cluster Federation	Basic	Advanced
Custom Reporting	No	Yes

Best Practices

1. Configure cloud integration early - Accurate pricing requires cloud provider API access

2. Use consistent labels - Enable detailed cost allocation across teams

3. Set up alerting - Catch cost anomalies early

4. Review regularly - Weekly cost reviews help identify optimization opportunities

5. Combine with right-sizing - Use cost data to right-size resources

Reference:

1. Official Website

2. Repository

3. Documentation

4. CNCF Project