Horizontal Pod Autoscaler in Kubernetes - Step-by-Step Execution

Process Flow - Horizontal Pod Autoscaler

Monitor Metrics

↓

Compare Current Load with Target

↓

Decide to Scale Up or Down

↓

Increase

↓

Change Pod Count

↓

Wait & Repeat

The Horizontal Pod Autoscaler watches metrics, compares them to targets, then adjusts pod count up or down accordingly, repeating continuously.

Execution Sample

Kubernetes

kubectl autoscale deployment myapp --min=2 --max=5 --cpu-percent=50

This command creates an autoscaler for 'myapp' deployment to keep CPU usage near 50%, scaling pods between 2 and 5.

Process Table

Step	Current CPU %	Target CPU %	Pods Before	Decision	Pods After
1	30	50	2	CPU below target, no scale change	2
2	55	50	2	CPU above target, scale up	3
3	70	50	3	CPU well above target, scale up	4
4	45	50	4	CPU below target, scale down	3
5	20	50	3	CPU well below target, scale down	2
6	50	50	2	CPU at target, no change	2

💡 Autoscaler continuously monitors; this trace shows decisions over 6 checks.

Status Tracker

Variable	Start	After 1	After 2	After 3	After 4	After 5	After 6
Current CPU %	N/A	30	55	70	45	20	50
Pods Count	2	2	3	4	3	2	2

Key Moments - 3 Insights

Why doesn't the pod count change when CPU is below target but not very low (Step 1)?

At Step 1, CPU is 30%, below the 50% target but not low enough to trigger scaling down, so pods remain at 2 as shown in execution_table row 1.

Why does the pod count increase by only one pod at each scale-up step?

The autoscaler adjusts pods gradually to avoid sudden changes; at Steps 2 and 3, it increases pods by one each time, as seen in execution_table rows 2 and 3.

Can the pod count go below the minimum set (2 pods)?

No, the autoscaler respects the minimum pods limit; even when CPU is very low at Step 5, pods do not go below 2, as shown in execution_table row 5.

Visual Quiz - 3 Questions

Look at the execution table, what is the pod count after Step 3?

A 4

B 3

C 2

D 5

💡 Hint

Check the 'Pods After' column in row for Step 3.

At which step does the autoscaler decide to scale down the pods for the first time?

A Step 2

B Step 4

C Step 1

D Step 5

💡 Hint

Look for the first 'scale down' decision in the 'Decision' column.

If the CPU target was changed to 60%, how would the decision at Step 2 change?

A Still scale up

B Scale down

C No scale change

D Scale up by 2 pods

💡 Hint

Compare current CPU (55%) with new target (60%) in Step 2 row.

Concept Snapshot

Horizontal Pod Autoscaler (HPA):
- Monitors pod metrics (e.g., CPU usage).
- Compares current metric to target.
- Scales pod count up/down within min/max limits.
- Runs continuously to keep app responsive.
- Command example: kubectl autoscale deployment NAME --min=X --max=Y --cpu-percent=Z

Full Transcript

The Horizontal Pod Autoscaler watches the CPU usage of pods in a deployment. It compares the current CPU percentage to a target value. If the CPU is higher than the target, it increases the number of pods by one to handle more load. If the CPU is lower than the target, it decreases the number of pods by one to save resources, but never below the minimum set. This process repeats continuously to keep the application running efficiently. The example command sets up autoscaling for a deployment named 'myapp' with a minimum of 2 pods, maximum of 5 pods, and a target CPU usage of 50%. The execution table shows how the pod count changes step-by-step based on CPU usage readings.

Practice

1. What is the main purpose of a Horizontal Pod Autoscaler (HPA) in Kubernetes?

easy

A. Manually increase pod count when needed

B. Automatically adjust the number of pods based on resource usage

C. Scale the number of nodes in the cluster

D. Backup pod data automatically

Solution

1. Step 1: Understand HPA functionality

   HPA monitors resource usage like CPU or memory and adjusts pod count automatically.

2. Step 2: Differentiate from other scaling types

   HPA scales pods, not nodes or manual actions, and does not handle backups.

3. Final Answer:

   Automatically adjust the number of pods based on resource usage -> Option B

4. Quick Check:

   HPA = automatic pod scaling [OK]

Hint: HPA changes pod count automatically by resource use [OK]

Common Mistakes:

• Confusing HPA with manual scaling
• Thinking HPA scales nodes instead of pods
• Assuming HPA handles backups

2. Which of the following is the correct command to create a Horizontal Pod Autoscaler for a deployment named webapp targeting 50% CPU usage with minimum 2 and maximum 5 pods?

easy

A. kubectl autoscale deployment webapp --cpu=50 --min=2 --max=5

B. kubectl create hpa webapp --cpu=50 --min-pods=2 --max-pods=5

C. kubectl autoscale deployment webapp --cpu-percent=50 --min=2 --max=5

D. kubectl autoscale deployment webapp --cpu-percent=50 --min-pods=2 --max-pods=5

Solution

1. Step 1: Recall correct kubectl autoscale syntax

   The correct command is kubectl autoscale deployment [name] --cpu-percent=[value] --min=[minPods] --max=[maxPods].

2. Step 2: Match options with correct syntax

   kubectl autoscale deployment webapp --cpu-percent=50 --min=2 --max=5 matches the correct syntax exactly with proper flags and values.

3. Final Answer:

   kubectl autoscale deployment webapp --cpu-percent=50 --min=2 --max=5 -> Option C

4. Quick Check:

   Correct autoscale syntax = kubectl autoscale deployment webapp --cpu-percent=50 --min=2 --max=5 [OK]

Hint: Use kubectl autoscale with --cpu-percent, --min, --max [OK]

Common Mistakes:

• Using wrong flags like --min-pods or --cpu
• Misspelling 'autoscale' or 'deployment'
• Omitting required flags

3. Given this HPA YAML snippet:

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: myapp-hpa
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: myapp
  minReplicas: 1
  maxReplicas: 4
  metrics:
  - type: Resource
    resource:
      name: cpu
      target:
        type: Utilization
        averageUtilization: 60

What happens if the CPU usage stays at 80% for a while?

medium

A. The number of pods will increase up to 4 to reduce CPU load

B. The number of pods will decrease to 1

C. The number of pods will stay the same regardless of CPU

D. The deployment will be deleted automatically

Solution

1. Step 1: Understand HPA scaling behavior

   HPA increases pods when CPU usage exceeds target averageUtilization (60%). Here CPU is 80%, which is higher.

2. Step 2: Check min and max replicas limits

   Pods can scale between 1 and 4. Since CPU is high, pods will increase up to max 4 to reduce load.

3. Final Answer:

   The number of pods will increase up to 4 to reduce CPU load -> Option A

4. Quick Check:

   CPU > target => pods increase [OK]

Hint: CPU above target means pods scale up within min-max [OK]

Common Mistakes:

• Thinking pods scale down when CPU is high
• Ignoring min/max replica limits
• Assuming deployment deletion on high CPU

4. You applied this HPA YAML but it does not scale pods as expected:

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: test-hpa
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: test-app
  minReplicas: 2
  maxReplicas: 5
  metrics:
  - type: Resource
    resource:
      name: memory
      target:
        type: Utilization
        averageUtilization: 70

What is the most likely reason for no scaling?

medium

A. The pods do not expose memory metrics, so HPA has no data

B. minReplicas is set too low

C. maxReplicas is less than minReplicas

D. The deployment name is incorrect

Solution

1. Step 1: Check metric type and data availability

   HPA needs metrics from pods. If memory metrics are not exposed or collected, HPA cannot scale.

2. Step 2: Verify other spec values

   minReplicas and maxReplicas are valid (2 and 5). Deployment name looks correct. So likely missing metrics cause no scaling.

3. Final Answer:

   The pods do not expose memory metrics, so HPA has no data -> Option A

4. Quick Check:

   Missing metrics = no scaling [OK]

Hint: HPA needs metrics; no metrics means no scaling [OK]

Common Mistakes:

• Assuming minReplicas too low stops scaling
• Confusing minReplicas and maxReplicas values
• Ignoring metric availability

5. You want to create an HPA that scales a deployment named api-server based on both CPU usage (target 70%) and custom metric requests_per_second (target 100). The minimum pods should be 3 and maximum 10. Which YAML snippet correctly defines this HPA?

hard

A. apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
name: api-server-hpa
spec:
scaleTargetRef:
apiVersion: apps/v1
kind: Deployment
name: api-server
minReplicas: 3
maxReplicas: 10
metrics:
- type: Resource
resource:
name: cpu
target:
type: Utilization
averageUtilization: 70
- type: Resource
resource:
name: requests_per_second
target:
type: AverageValue
averageValue: "100"

B. apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
name: api-server-hpa
spec:
scaleTargetRef:
apiVersion: apps/v1
kind: Deployment
name: api-server
minReplicas: 3
maxReplicas: 10
metrics:
- type: Resource
resource:
name: cpu
target:
type: AverageValue
averageValue: "70"
- type: Pods
pods:
metric:
name: requests_per_second
target:
type: AverageValue
averageValue: "100"

C. apiVersion: autoscaling/v1
kind: HorizontalPodAutoscaler
metadata:
name: api-server-hpa
spec:
scaleTargetRef:
apiVersion: apps/v1
kind: Deployment
name: api-server
minReplicas: 3
maxReplicas: 10
targetCPUUtilizationPercentage: 70

D. apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
name: api-server-hpa
spec:
scaleTargetRef:
apiVersion: apps/v1
kind: Deployment
name: api-server
minReplicas: 3
maxReplicas: 10
metrics:
- type: Resource
resource:
name: cpu
target:
type: Utilization
averageUtilization: 70
- type: Pods
pods:
metric:
name: requests_per_second
target:
type: AverageValue
averageValue: "100"

Solution

1. Step 1: Use autoscaling/v2 API for multiple metrics

   autoscaling/v2 supports multiple metrics including resource and custom metrics.

2. Step 2: Check metric types and targets

   CPU uses type Utilization with averageUtilization; custom metric uses type Pods with metric name and AverageValue target.

3. Step 3: Verify minReplicas and maxReplicas

   Minimum 3 and maximum 10 pods are correctly set in spec.

4. Final Answer:

   YAML with cpu resource metric (Utilization) and pods metric for requests_per_second -> Option D

5. Quick Check:

   Multiple metrics with correct types = apiVersion: autoscaling/v2 kind: HorizontalPodAutoscaler metadata: name: api-server-hpa spec: scaleTargetRef: apiVersion: apps/v1 kind: Deployment name: api-server minReplicas: 3 maxReplicas: 10 metrics: - type: Resource resource: name: cpu target: type: Utilization averageUtilization: 70 - type: Pods pods: metric: name: requests_per_second target: type: AverageValue averageValue: "100" [OK]

Hint: Use autoscaling/v2 with resource and pods metrics for multiple targets [OK]

Common Mistakes:

• Using autoscaling/v1 which does not support multiple metrics
• Wrong metric types or target fields
• Mixing resource and custom metric syntax incorrectly