View a markdown version of this page

Vendor and custom metrics cost reduction - AWS Prescriptive Guidance
Check container insights, prometheus, ADOT (AWS Distro for OpenTelemetry) metricsActionsRecommendations

Vendor and custom metrics cost reduction

Amazon CloudWatch, Prometheus, and third-party monitoring can become significant cost drivers in Amazon EKS.

Check container insights, prometheus, ADOT (AWS Distro for OpenTelemetry) metrics

For sample commands to audit Container Insights metrics and estimate cost, see the audit-container-insights-metrics.sh

Reduce Prometheus/ADOT Metric Cardinality

For sample commands to identify and reduce high-cardinality Prometheus metrics, see this reduce-prometheus-cardinality.sh

Prometheus Relabeling to Drop Expensive Metrics

For a sample Prometheus ServiceMonitor with metric relabeling to drop expensive metrics, see this servicemonitor-cost-optimized.yaml

CloudWatch Metric Streams: Cost-Effective Alternative

Use CloudWatch metric filters to extract only the signals you need from existing log data instead of publishing continuous custom metrics, see this cloudwatch-metric-filter.sh.

Vendor Metrics Cost Checklist

Metric Source

Cost Driver

Optimization

CloudWatch Container Insights

Per-metric pricing ($0.30/metric/month)

Disable enhanced metrics for non-prod

Amazon Managed Prometheus

Ingested samples ($0.03/10M samples)

Drop unused metrics via relabeling

CloudWatch Logs

Ingestion + storage ($0.50/GB + $0.03/GB)

Filter at source, reduce log verbosity

Third-party (Datadog, New Relic)

Per-host or per-container pricing

Use namespace-level sampling

Custom metrics via PutMetricData

$0.30/metric/month + API calls

Batch calls, reduce dimensions

Actions

Audit custom metrics and replace them with default metrics or built-in integrations wherever possible.

Audit process

For sample commands to list custom metrics, estimate their monthly cost, and identify ones that duplicate Container Insights defaults, see this audit-custom-metrics.sh

Common replacements

Custom Metric

Default Alternative

Source

Custom CPU per pod

pod_cpu_utilization

Container Insights

Custom memory per pod

pod_memory_utilization

Container Insights

Custom node count

cluster_node_count

Container Insights

Custom network bytes

pod_network_rx_bytes / pod_network_tx_bytes

Container Insights

Custom disk usage

node_filesystem_utilization

Container Insights

Custom request count

ALB metrics / Service Mesh

Built-in

Cleanup steps

  1. Inventory – List all custom metrics and their associated costs

  2. Map – Identify which default metrics or integrations provide equivalent data

  3. Validate – Confirm dashboards and alarms work with default metrics

  4. Remove – Delete custom metric publishing from application configurations

  5. Monitor – Verify no alerting gaps after removal

Audit your metric count with aws cloudwatch list-metrics, drop unused metrics via Prometheus relabeling, disable enhanced Container Insights in non-production clusters, and set ttlSecondsAfterFinished on jobs to auto-clean metric sources.

key takeaway: Monitoring costs scale with cluster size unless you actively manage cardinality. A 100-node cluster can easily generate thousands of metrics, most of which nobody looks at. Filter at the source, not after ingestion.

Recommendations

  • Enable CloudWatch Container Insights as the default observability layer, it provides comprehensive metrics at lower cost than custom metrics

  • Use the CloudWatch Metrics Calculator to estimate savings from custom metric removal

  • Retain custom metrics only when they provide unique business-level data not available from infrastructure metrics

  • Review metric publishing frequency, reduce from 1-second to 60-second resolution where real-time precision isn't required (saves on high-resolution metric charges)