Pular para o conteúdo principal

Prometheus Metrics

Overview​

The Infra Operator exposes comprehensive Prometheus metrics for complete observability of your infrastructure management. These metrics provide insights into reconciliation performance, AWS API usage, drift detection, resource status, and more.

Metrics Categories​

1. Reconciliation Metrics​

Track the performance and reliability of Kubernetes reconciliation loops.

infra_operator_reconcile_total​

Type: Counter Labels: resource_type, result Description: Total number of reconciliations per resource type and result (success, error, requeue)

# Reconciliation rate per resource type
rate(infra_operator_reconcile_total[5m])

# Success rate
sum(rate(infra_operator_reconcile_total{result="success"}[5m])) by (resource_type)

infra_operator_reconcile_duration_seconds​

Type: Histogram Labels: resource_type Description: Duration of reconciliation operations in seconds

# p95 reconciliation duration
histogram_quantile(0.95, rate(infra_operator_reconcile_duration_seconds_bucket[5m]))

# Average duration per resource type
rate(infra_operator_reconcile_duration_seconds_sum[5m]) / rate(infra_operator_reconcile_duration_seconds_count[5m])

infra_operator_reconcile_errors_total​

Type: Counter Labels: resource_type, error_type Description: Total number of reconciliation errors by type

Error Types:

  • get_failed - Failed to get resource from Kubernetes
  • sync_failed - Failed to sync with AWS
  • status_update_failed - Failed to update status
  • provider_failed - Failed to get AWS provider
  • finalizer_failed - Failed during cleanup
# Error rate per resource type
rate(infra_operator_reconcile_errors_total[5m])

# Error rate percentage
sum(rate(infra_operator_reconcile_errors_total[5m])) / sum(rate(infra_operator_reconcile_total[5m]))

2. Resource Metrics​

Track the current state of managed resources.

infra_operator_resources_total​

Type: Gauge Labels: resource_type, status Description: Current number of managed resources by type and status

Status Values:

  • ready - Resource is ready and synced
  • notready - Resource is not ready
  • pending - Resource is being created
  • deleting - Resource is being deleted
  • error - Resource is in error state
# Total resources by type
sum(infra_operator_resources_total) by (resource_type)

# Resources not ready
infra_operator_resources_total{status="notready"}

# Percentage ready
sum(infra_operator_resources_total{status="ready"}) / sum(infra_operator_resources_total)

infra_operator_resource_creation_timestamp_seconds​

Type: Gauge Labels: resource_type, resource_name Description: Unix timestamp when the resource was created

# Age of resources in hours
(time() - infra_operator_resource_creation_timestamp_seconds) / 3600

3. AWS API Metrics​

Monitor AWS API performance, errors, and throttling.

infra_operator_aws_api_calls_total​

Type: Counter Labels: service, operation, result Description: Total number of AWS API calls

Services: EC2, S3, RDS, ELBv2, IAM, SecretsManager, KMS, Lambda, APIGateway, CloudFront, Route53, ACM, DynamoDB, SQS, SNS, ECR, ECS, EKS, ElastiCache

# API call rate by service
sum(rate(infra_operator_aws_api_calls_total[5m])) by (service)

# Top 10 most called operations
topk(10, sum(rate(infra_operator_aws_api_calls_total[5m])) by (service, operation))

infra_operator_aws_api_call_duration_seconds​

Type: Histogram Labels: service, operation Description: Duration of AWS API calls in seconds

# p95 latency per service
histogram_quantile(0.95, sum(rate(infra_operator_aws_api_call_duration_seconds_bucket[5m])) by (le, service))

# Slowest operations
topk(10, histogram_quantile(0.95, sum(rate(infra_operator_aws_api_call_duration_seconds_bucket[5m])) by (le, service, operation)))

infra_operator_aws_api_errors_total​

Type: Counter Labels: service, operation, error_code Description: Total number of AWS API errors

Common Error Codes:

  • InvalidParameterValue
  • ResourceNotFoundException
  • UnauthorizedOperation
  • DependencyViolation
  • ResourceAlreadyExists
# Error rate per service
sum(rate(infra_operator_aws_api_errors_total[5m])) by (service)

# Most common error codes
topk(10, sum(rate(infra_operator_aws_api_errors_total[5m])) by (error_code))

infra_operator_aws_api_throttles_total​

Type: Counter Labels: service, operation Description: Total number of AWS API throttling events (rate limit exceeded)

# Throttling rate per service
sum(rate(infra_operator_aws_api_throttles_total[5m])) by (service)

# Operations being throttled
infra_operator_aws_api_throttles_total > 0

4. Drift Detection Metrics​

Monitor configuration drift and automatic healing.

infra_operator_drift_detected_total​

Type: Counter Labels: resource_type, severity Description: Total number of configuration drifts detected

Severity Levels:

  • critical - High severity drift
  • warning - Medium severity drift
  • info - Low severity drift
# Drift detection rate
sum(rate(infra_operator_drift_detected_total[5m])) by (resource_type, severity)

# Critical drifts in last hour
increase(infra_operator_drift_detected_total{severity="critical"}[1h])

infra_operator_drift_healed_total​

Type: Counter Labels: resource_type Description: Total number of drifts automatically healed

# Healing rate
rate(infra_operator_drift_healed_total[5m])

# Healing success percentage
sum(rate(infra_operator_drift_healed_total[5m])) / sum(rate(infra_operator_drift_detected_total[5m]))

infra_operator_drift_healing_failures_total​

Type: Counter Labels: resource_type, error_type Description: Total number of failed drift healing attempts

# Healing failure rate
rate(infra_operator_drift_healing_failures_total[5m])

infra_operator_drift_detection_duration_seconds​

Type: Histogram Labels: resource_type Description: Duration of drift detection operations

# p95 drift detection duration
histogram_quantile(0.95, rate(infra_operator_drift_detection_duration_seconds_bucket[5m]))

5. Finalizer Metrics​

Track cleanup and deletion performance.

infra_operator_finalizer_duration_seconds​

Type: Histogram Labels: resource_type Description: Duration of finalizer execution (resource cleanup)

# p95 finalizer duration
histogram_quantile(0.95, rate(infra_operator_finalizer_duration_seconds_bucket[5m]))

# Slowest finalizers
topk(5, histogram_quantile(0.95, sum(rate(infra_operator_finalizer_duration_seconds_bucket[5m])) by (le, resource_type)))

infra_operator_finalizer_errors_total​

Type: Counter Labels: resource_type, error_type Description: Total number of finalizer execution errors

# Finalizer error rate
rate(infra_operator_finalizer_errors_total[5m])

6. Provider Metrics​

Monitor AWS Provider status and credentials.

infra_operator_provider_ready​

Type: Gauge Labels: provider_name, region Description: Indicates if an AWS Provider is ready (1) or not (0)

# Providers not ready
infra_operator_provider_ready == 0

# Ready percentage
avg(infra_operator_provider_ready)

infra_operator_provider_credential_rotations_total​

Type: Counter Labels: provider_name Description: Total number of AWS credential rotations

# Credential rotation rate
rate(infra_operator_provider_credential_rotations_total[24h])

7. Workqueue Metrics​

Monitor controller work queue depth and throughput.

infra_operator_workqueue_depth​

Type: Gauge Labels: resource_type Description: Current depth of the work queue for each resource type

# Queue depth per resource type
infra_operator_workqueue_depth

# Deep queues (potential backlog)
infra_operator_workqueue_depth > 100

infra_operator_workqueue_adds_total​

Type: Counter Labels: resource_type Description: Total number of items added to work queue

# Queue add rate
rate(infra_operator_workqueue_adds_total[5m])

8. Cache Metrics​

Monitor caching effectiveness (future feature).

infra_operator_cache_hits_total​

Type: Counter Labels: resource_type Description: Total number of successful cache lookups

infra_operator_cache_misses_total​

Type: Counter Labels: resource_type Description: Total number of failed cache lookups

# Cache hit rate
sum(rate(infra_operator_cache_hits_total[5m])) / (sum(rate(infra_operator_cache_hits_total[5m])) + sum(rate(infra_operator_cache_misses_total[5m])))

Common Query Examples​

Health & Performance​

# Overall health - no errors in last 5 minutes
sum(rate(infra_operator_reconcile_errors_total[5m])) == 0

# Reconciliation throughput
sum(rate(infra_operator_reconcile_total[5m]))

# Average reconciliation duration
rate(infra_operator_reconcile_duration_seconds_sum[5m]) / rate(infra_operator_reconcile_duration_seconds_count[5m])

# Error budget (99.9% SLO)
1 - (sum(rate(infra_operator_reconcile_errors_total[30d])) / sum(rate(infra_operator_reconcile_total[30d])))

Resource Management​

# Total managed resources
sum(infra_operator_resources_total)

# Resources in error state
sum(infra_operator_resources_total{status="error"})

# Oldest resources
topk(10, time() - infra_operator_resource_creation_timestamp_seconds)

AWS API Monitoring​

# AWS API error rate percentage
sum(rate(infra_operator_aws_api_errors_total[5m])) / sum(rate(infra_operator_aws_api_calls_total[5m]))

# Services with highest error rates
topk(5, sum(rate(infra_operator_aws_api_errors_total[5m])) by (service))

# Throttled operations
sum(rate(infra_operator_aws_api_throttles_total[5m])) by (service, operation) > 0

Drift Monitoring​

# Critical drifts detected
sum(increase(infra_operator_drift_detected_total{severity="critical"}[1h]))

# Drift healing success rate
sum(rate(infra_operator_drift_healed_total[5m])) / sum(rate(infra_operator_drift_detected_total[5m]))

# Resources with most drift
topk(5, sum(increase(infra_operator_drift_detected_total[24h])) by (resource_type))

Installation​

Deploy ServiceMonitor​

Command:

kubectl apply -f config/prometheus/servicemonitor.yaml

This creates:

  • ServiceMonitor: Tells Prometheus to scrape the operator
  • PodMonitor: Alternative direct pod scraping
  • PrometheusRule: Pre-configured alerts

Import Grafana Dashboard​

  1. Open Grafana UI
  2. Go to Dashboards → Import
  3. Upload config/grafana/dashboard.json
  4. Select your Prometheus datasource
  5. Click Import

Alerting Rules​

The operator includes pre-configured Prometheus alerts in config/prometheus/servicemonitor.yaml:

Critical Alerts​

  • InfraOperatorNoReconciliations: No reconciliations in 10 minutes (operator may be down)
  • InfraOperatorProviderNotReady: AWS Provider not ready for >5 minutes
  • InfraOperatorDriftDetected: Critical drift detected

Warning Alerts​

  • InfraOperatorHighErrorRate: Reconciliation error rate >10%
  • InfraOperatorSlowReconciliation: p95 reconciliation duration >30s
  • InfraOperatorAWSAPIErrors: AWS API error rate >1 error/sec
  • InfraOperatorAWSAPIThrottled: AWS API throttling detected
  • InfraOperatorDriftHealingFailed: Drift healing failures
  • InfraOperatorResourcesNotReady: >5 resources not ready for >15 minutes
  • InfraOperatorSlowFinalizer: p95 finalizer duration >60s

Customize Alerts​

Edit config/prometheus/servicemonitor.yaml to adjust:

  • Thresholds
  • Duration (for)
  • Severity levels
  • Alert labels and annotations

Integration with Existing Monitoring​

Prometheus Operator​

If you have Prometheus Operator installed:

kubectl apply -f config/prometheus/servicemonitor.yaml

The ServiceMonitor will be automatically discovered.

Manual Prometheus Configuration​

Add to your prometheus.yml:

scrape_configs:
  - job_name: 'infra-operator'
kubernetes_sd_configs:
      - role: endpoints
relabel_configs:
      - source_labels: [__meta_kubernetes_service_label_control_plane]
        regex: controller-manager
        action: keep
      - source_labels: [__meta_kubernetes_endpoint_port_name]
        regex: metrics
        action: keep

Datadog Integration​

Use Datadog's Prometheus integration:

apiVersion: v1
kind: ConfigMap
metadata:
  name: datadog-checks
data:
  prometheus.yaml: |
instances:
      - prometheus_url: http://infra-operator-controller-manager-metrics-service.infra-operator-system:8443/metrics
        namespace: infra_operator
        metrics:
          - infra_operator_*

New Relic Integration​

Use New Relic's Prometheus exporter:

helm repo add newrelic https://helm-charts.newrelic.com
helm install newrelic-prometheus newrelic/nri-prometheus \
  --set cluster=infra-operator-cluster \
  --set config.scrape_enabled_label=prometheus.io/scrape

Dashboard Panels​

The included Grafana dashboard provides:

  1. Reconciliation Rate - Real-time reconciliation throughput
  2. Error Rate - Percentage of failed reconciliations
  3. Reconciliation Duration - p95 and p50 latencies
  4. Resources by Status - Pie chart of resource states
  5. AWS API Call Rate - API usage by service
  6. AWS API Duration - p95 API latencies
  7. AWS API Errors - Error breakdown by service/operation
  8. AWS Throttling - Rate limiting events
  9. Drift Detected - Drift events by severity
  10. Drift Healing - Success and failure rates
  11. Provider Status - AWS Provider health
  12. Workqueue Depth - Queue backlogs

Metric Retention​

  • Short-term (15 days): All metrics at 30s resolution
  • Medium-term (90 days): Downsampled to 5m resolution
  • Long-term (1 year): Downsampled to 1h resolution (for trends)

Prometheus Configuration​

Example:

global:
  scrape_interval: 30s
  evaluation_interval: 30s

storage:
  tsdb:
retention.time: 15d

Thanos/Cortex for Long-term​

Use Thanos or Cortex for:

  • Long-term storage (>15 days)
  • Global view across clusters
  • Downsampling
  • High availability

Troubleshooting​

Metrics Not Appearing​

  1. Check ServiceMonitor is created:

    kubectl get servicemonitor -n infra-operator-system

  2. Verify Prometheus is scraping:

    up{job="infra-operator"}

  3. Check operator logs:

    kubectl logs -n infra-operator-system deployment/infra-operator-controller-manager

High Cardinality​

If you have too many resources causing high cardinality:

  1. Limit resource metrics:

    // Only track essential statuses
    ResourcesTotal.WithLabelValues(resourceType, "ready").Inc()
    ResourcesTotal.WithLabelValues(resourceType, "notready").Inc()

  2. Use recording rules:

    - record: infra_operator:reconcile_total:rate5m
      expr: sum(rate(infra_operator_reconcile_total[5m])) by (resource_type)

Missing Historical Data​

If you need historical data:

  1. Enable long-term storage (Thanos/Cortex)
  2. Increase retention period
  3. Use recording rules for pre-aggregated data

Best Practices​

Best Practices
  • Enable ServiceMonitor for Prometheus Operator — Automatic metric discovery and scraping configuration
  • Set appropriate alert thresholds — Configure alerts for error rates >5%, p95 latency >30s, drift detection
  • Use Grafana dashboards — Import pre-built dashboards for operator monitoring and AWS API metrics
  • Monitor AWS API throttling — Set alerts on infra_operator_aws_api_throttled_total to detect rate limiting
  • Track reconciliation duration — Monitor p95 reconciliation times to identify performance bottlenecks

Reference​

Metric Naming Convention​

All metrics follow the format:

infra_operator_<component>_<metric>_<unit>

Examples:

  • infra_operator_reconcile_total (counter, no unit)
  • infra_operator_reconcile_duration_seconds (histogram, seconds)
  • infra_operator_resources_total (gauge, count)

Labels​

Standard labels used:

  • resource_type - Type of AWS resource (VPC, Subnet, etc.)
  • service - AWS service name (EC2, S3, RDS, etc.)
  • operation - AWS API operation name
  • result - Operation result (success, error, requeue)
  • status - Resource status (ready, notready, pending, deleting)
  • severity - Drift severity (critical, warning, info)
  • error_type - Type of error encountered
  • error_code - AWS API error code
  • provider_name - Name of AWS Provider
  • region - AWS region

Support​

For questions or issues with metrics:

  1. Check the documentation
  2. Review example queries
  3. Join our Slack community
  4. Open an issue