Metrics Reference

All metrics are prefixed with hybernate_ and registered with the controller-runtime metrics registry. They are served on the metrics endpoint (default :8443 HTTPS).

Tier 1: Cluster Health

These metrics provide a high-level view of Hybernate's impact.

Metric	Type	Labels	Description
hybernate_workloads_total	Gauge	phase	Managed workloads by lifecycle phase
hybernate_active_workloads	Gauge		Workloads in a running state
hybernate_paused_workloads	Gauge		Currently paused workloads
hybernate_destroyed_workloads	Gauge		Destroyed workloads
hybernate_reconcile_errors_total	Counter	controller	Reconciliation errors by controller
hybernate_lifecycle_transitions_total	Counter	from, to	Phase transitions
hybernate_lifecycle_action_duration_seconds	Histogram	action	Duration of lifecycle actions (pause, resume, destroy, scale)
hybernate_cost_estimated_savings_dollars	Gauge		Estimated monthly savings (requires autoscaler for realization)
hybernate_cost_estimated_dollars	Gauge		Total estimated monthly cost
hybernate_cost_estimated_without_management_dollars	Gauge		Estimated cost without Hybernate
hybernate_resource_reduction_cpu_millicores	Gauge		Total CPU millicores freed by Hybernate
hybernate_resource_reduction_memory_bytes	Gauge		Total memory bytes freed by Hybernate

Tier 2: Operational Insight

These metrics help you understand what the operator is doing and why.

Metric	Type	Labels	Description
hybernate_prediction_confidence_percent	Gauge	season, namespace, workload	Prediction accuracy by season
hybernate_prediction_phase	Gauge	namespace, workload	Engine phase (0-4)
hybernate_prediction_data_points	Gauge	namespace, workload	Data points collected
hybernate_prediction_anomalies_total	Counter	namespace, workload	Anomalies detected
hybernate_cost_cpu_hours	Gauge		Total vCPU-hours this month
hybernate_cost_memory_hours	Gauge		Total GiB memory hours
hybernate_cost_storage_hours	Gauge		Total GiB storage hours
hybernate_scale_events_total	Counter	direction, namespace, workload	Scale events by direction
hybernate_scale_replicas	Gauge	namespace, workload	Current replica count
hybernate_idle_detections_total	Counter	action, namespace, workload	Idle detections by action
hybernate_pause_expiry_actions_total	Counter	action	Pause expiry events
hybernate_drift_detections_total	Counter	policy	Replica drift detections

Tier 3: Debugging

These metrics help troubleshoot specific workload behavior.

Metric	Type	Labels	Description
hybernate_idle_signal_result	Gauge	namespace, workload	Idle signal status (1=active, 2=signals_confirm, 3=grace_period, 4=idle)
hybernate_scale_guard_blocked_total	Counter	namespace, workload	Scale-downs blocked by guard probes
hybernate_idle_fluke_total	Counter	namespace, workload	Signals confirmed idle but prediction disagreed
hybernate_prediction_regime_changes_total	Counter	namespace, workload	Regime changes detected
hybernate_pvc_retention_remaining_seconds	Gauge	namespace, workload	Seconds until PVC cleanup
hybernate_automation_skipped_total	Counter	namespace, workload	Automation skipped (manual override active)
hybernate_dryrun_actions_total	Counter	action	Actions that would have been taken in dry-run
hybernate_target_unavailable_total	Counter	namespace, workload	Target workload not found

Discovery

Metric	Type	Labels	Description
hybernate_discovery_scan_duration_seconds	Histogram		Scan duration
hybernate_discovery_workloads	Gauge	classification	Discovered workloads by class
hybernate_discovery_estimated_savings_dollars	Gauge		Estimated savings from discoveries
hybernate_discovery_auto_managed_total	Counter		Auto-created ManagedWorkloads

Prediction Phase Values

The hybernate_prediction_phase gauge uses numeric values:

Value	Phase
0	Observing
1	DailySuggesting
2	DailyActive
3	WeeklySuggesting
4	FullyActive

Useful PromQL Queries

# Total workloads by phase
hybernate_workloads_total

# Estimated monthly savings trend
hybernate_cost_estimated_savings_dollars

# Workloads with low prediction confidence
hybernate_prediction_confidence_percent{season="daily"} < 70

# Recent scale events
rate(hybernate_scale_events_total[1h])

# Idle flukes (signals say idle, prediction disagrees)
rate(hybernate_idle_fluke_total[1h]) > 0

# Workloads stuck in grace period
hybernate_idle_signal_result == 3