Kubernetes HPA Behavior When Deployment Is Scaled to 0

Kubernetes HPA Behavior When Deployment Is Scaled to 0

Summary

When a Kubernetes Deployment is manually or programmatically scaled down to 0 replicas, the HorizontalPodAutoscaler (HPA) does not automatically restore it to the configured minimum replica count (minReplicas).

In this state, autoscaling is effectively inactive for that workload because there are no running pods producing metrics for HPA to evaluate. To resume autoscaling, the Deployment must first be scaled above 0 (for example, 1 replica), after which HPA control can continue.

Practical Impact

• A service can remain unintentionally offline even though an HPA exists.
• Teams may incorrectly assume minReplicas guarantees automatic recovery from zero.
• Incident triage can be delayed if responders focus only on HPA configuration and not current spec.replicas.

Why This Happens

HPA adjusts replica counts for a running workload based on observed metrics and policy constraints. If the target Deployment is at 0 replicas, there are no active pods for typical resource-based signals (CPU or memory), and HPA does not bootstrap the workload back up by itself in standard configurations.

Conceptually:

1. Deployment set to 0 replicas.
2. No pods are running.
3. No usable workload metrics from pods.
4. HPA cannot compute a scaling action that revives the workload from zero.

Typical Symptoms During an Incident

• kubectl get deploy shows 0/0 or desired 0 replicas.
• kubectl get hpa still shows the HPA object as present.
• Traffic fails or upstream health checks fail because no pods are available.
• No automatic recovery occurs despite minReplicas > 0.

Fast Recovery Runbook

1. Confirm current Deployment replica state:

   kubectl -n <namespace> get deploy <name>

2. Manually scale above zero:

   kubectl -n <namespace> scale deploy/<name> --replicas=1

3. Verify pods become Ready:

   kubectl -n <namespace> get pods -l app=<label> -w

4. Confirm HPA is now actively evaluating and adjusting:

   kubectl -n <namespace> get hpa <hpa-name> -w

Preventive Controls

• Restrict direct scale-to-zero operations on HPA-managed Deployments unless explicitly intended.
• Add alerting for HPA-managed workloads whose desired replicas become 0.
• Use runbook automation to restore to 1 replica when unintended zeroing is detected.
• Document ownership and expected behavior so responders know manual wake-up may be required.

Suggested Monitoring Checks

• Detect spec.replicas == 0 for Deployments labeled as autoscaled.
• Detect sustained zero Ready pods for critical services.
• Correlate with manual actions, rollout pipelines, or scripted scale events.

Decision Flow

#mermaid-svg-eE5n29JJRklbXDjw{font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-eE5n29JJRklbXDjw .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-eE5n29JJRklbXDjw .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-eE5n29JJRklbXDjw .error-icon{fill:#552222;}#mermaid-svg-eE5n29JJRklbXDjw .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-eE5n29JJRklbXDjw .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-eE5n29JJRklbXDjw .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-eE5n29JJRklbXDjw .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-eE5n29JJRklbXDjw .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-eE5n29JJRklbXDjw .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-eE5n29JJRklbXDjw .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-eE5n29JJRklbXDjw .marker{fill:#333333;stroke:#333333;}#mermaid-svg-eE5n29JJRklbXDjw .marker.cross{stroke:#333333;}#mermaid-svg-eE5n29JJRklbXDjw svg{font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-eE5n29JJRklbXDjw p{margin:0;}#mermaid-svg-eE5n29JJRklbXDjw .label{font-family:"trebuchet ms",verdana,arial,sans-serif;color:#333;}#mermaid-svg-eE5n29JJRklbXDjw .cluster-label text{fill:#333;}#mermaid-svg-eE5n29JJRklbXDjw .cluster-label span{color:#333;}#mermaid-svg-eE5n29JJRklbXDjw .cluster-label span p{background-color:transparent;}#mermaid-svg-eE5n29JJRklbXDjw .label text,#mermaid-svg-eE5n29JJRklbXDjw span{fill:#333;color:#333;}#mermaid-svg-eE5n29JJRklbXDjw .node rect,#mermaid-svg-eE5n29JJRklbXDjw .node circle,#mermaid-svg-eE5n29JJRklbXDjw .node ellipse,#mermaid-svg-eE5n29JJRklbXDjw .node polygon,#mermaid-svg-eE5n29JJRklbXDjw .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-eE5n29JJRklbXDjw .rough-node .label text,#mermaid-svg-eE5n29JJRklbXDjw .node .label text,#mermaid-svg-eE5n29JJRklbXDjw .image-shape .label,#mermaid-svg-eE5n29JJRklbXDjw .icon-shape .label{text-anchor:middle;}#mermaid-svg-eE5n29JJRklbXDjw .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-eE5n29JJRklbXDjw .rough-node .label,#mermaid-svg-eE5n29JJRklbXDjw .node .label,#mermaid-svg-eE5n29JJRklbXDjw .image-shape .label,#mermaid-svg-eE5n29JJRklbXDjw .icon-shape .label{text-align:center;}#mermaid-svg-eE5n29JJRklbXDjw .node.clickable{cursor:pointer;}#mermaid-svg-eE5n29JJRklbXDjw .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-eE5n29JJRklbXDjw .arrowheadPath{fill:#333333;}#mermaid-svg-eE5n29JJRklbXDjw .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-eE5n29JJRklbXDjw .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-eE5n29JJRklbXDjw .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-eE5n29JJRklbXDjw .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-eE5n29JJRklbXDjw .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-eE5n29JJRklbXDjw .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-eE5n29JJRklbXDjw .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-eE5n29JJRklbXDjw .cluster text{fill:#333;}#mermaid-svg-eE5n29JJRklbXDjw .cluster span{color:#333;}#mermaid-svg-eE5n29JJRklbXDjw div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-eE5n29JJRklbXDjw .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-eE5n29JJRklbXDjw rect.text{fill:none;stroke-width:0;}#mermaid-svg-eE5n29JJRklbXDjw .icon-shape,#mermaid-svg-eE5n29JJRklbXDjw .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-eE5n29JJRklbXDjw .icon-shape p,#mermaid-svg-eE5n29JJRklbXDjw .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-eE5n29JJRklbXDjw .icon-shape .label rect,#mermaid-svg-eE5n29JJRklbXDjw .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-eE5n29JJRklbXDjw .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-eE5n29JJRklbXDjw .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-eE5n29JJRklbXDjw :root{--mermaid-font-family:"trebuchet ms",verdana,arial,sans-serif;} Yes
No
Deployment managed by HPA
Replicas > 0?
HPA evaluates metrics and scales within policy
No running pods for normal workload metrics
Autoscaling effectively inactive
Manually scale to at least 1 replica

Incident Note Template (Formal)

Observation: The Deployment was at 0 replicas while an HPA was configured.

Behavior: HPA did not restore the workload to minReplicas from zero.

Impact: Service availability was reduced because no pods were serving traffic.

Resolution: Deployment was manually scaled to 1 replica, allowing pods to start and HPA to resume control.

Prevention: Add guardrails and alerts to prevent or quickly detect unintended scale-to-zero on autoscaled workloads.

One-Line Summary

If an HPA-managed Deployment is scaled to 0, autoscaling does not self-revive it; scale to at least 1 replica first, then HPA can take over.