Pause & Destroy Guide

Hybernate provides two lifecycle actions beyond scaling: pause (scale to zero) and destroy (delete the workload). Both include safety mechanisms for data preservation.

Pause

Pausing scales the workload to zero replicas. The Deployment or StatefulSet still exists; only the pods are removed.

What Happens When a Workload Is Paused

1. The current replica count is saved to status.pause.previousReplicas
2. A resource snapshot is captured (CPU, memory, storage per replica) for cost savings calculation
3. The workload is scaled to 0
4. The phase transitions to Paused
5. status.pause.pausedAt is set

Triggering a Pause

Manually:

kubectl patch managedworkload my-api -n staging \
  --type merge -p '{"spec":{"desiredState":"Paused"}}'

Automatically: When idle detection confirms idle and idlePolicy.action is pause or auto.

Pause Expiry

Configure what happens if a workload stays paused too long:

spec:
  pause:
    expireAfter: "24h"
    expireAction: Resume

expireAction	What happens
resume	Workload is scaled back to its previous replica count
destroy	Workload is deleted (with optional PVC retention)

If expireAfter is not set, the workload stays paused indefinitely.

Resume

Resuming restores the saved replica count and waits for pods to become ready.

Manually:

kubectl patch managedworkload my-api -n staging \
  --type merge -p '{"spec":{"desiredState":"Running"}}'

Automatically:

• When expireAfter elapses with expireAction: Resume
• When autoResume: true is set and signals no longer confirm idle

Destroy

Destroying deletes the target Deployment or StatefulSet. The ManagedWorkload CR remains to track PVC retention and cost savings.

What Happens When a Workload Is Destroyed

1. A resource snapshot is captured for cost savings calculation
2. The target Deployment/StatefulSet is deleted
3. The phase transitions to Destroyed
4. status.destroy.destroyedAt is set
5. If PVC retention is configured, status.destroy.pvcRetentionExpiresAt is set

Triggering a Destroy

Manually:

kubectl patch managedworkload my-api -n staging \
  --type merge -p '{"spec":{"desiredState":"Destroyed"}}'

Automatically:

• When idlePolicy.action is destroy and idle is confirmed
• When pause expires with expireAction: Destroy

PVC Retention

By default, PVCs are not cleaned up when a workload is destroyed. They persist independently. Configure retention to clean them up on a schedule:

spec:
  destroy:
    pvcRetention: "168h"       # Keep PVCs for 7 days after destroy
    pvcRetentionWarning: "24h" # Emit warning event 24h before cleanup

Timeline:

Destroyed ────────────────────────► PVC Warning ──► PVC Cleanup
    t=0                              t=144h          t=168h

The operator matches PVCs using the workload's pod template selector labels.

Warning

Once PVCs are deleted, the data is gone. Set pvcRetentionWarning to give users time to recover data before cleanup.

To cancel a scheduled PVC cleanup, remove pvcRetention from the spec. The operator detects the change and clears the expiry timer, preserving PVCs indefinitely.

PVC Retention Without Destroy

PVC retention only applies when a workload is destroyed. Paused workloads keep their PVCs unconditionally (only pods are removed, the workload object and PVCs remain).

Cost Savings During Pause and Destroy

Cost tracking is always enabled. During pause and destroy:

• While paused: CPU and memory savings accrue every reconcile. Storage savings are zero (PVCs persist).
• While destroyed: CPU and memory savings accrue. Storage savings begin accruing after PVC retention expires and PVCs are cleaned up.

Finalizer

The hybernate.io/cleanup finalizer is automatically added to every ManagedWorkload. It ensures that:

• If a ManagedWorkload CR is deleted while PVC retention is pending, the operator runs PVC cleanup before allowing the deletion to complete
• Paused workloads are resumed before the ManagedWorkload CR is removed (if applicable)