Skip to main content

Generate load

To observe KEDA scale the deployment in response to the KEDA ScaledObject we have configured, we need to generate some load on our application. We'll do that by calling the home page of the workload with hey.

The command below will run the load generator with:

  • 3 workers running concurrently
  • Sending 5 queries per second each
  • Running for a maximum of 10 minutes
~$export ALB_HOSTNAME=$(kubectl get ingress ui -n ui -o yaml | yq .status.loadBalancer.ingress[0].hostname)
~$kubectl run load-generator \
  --image=williamyeh/hey:latest \
  --restart=Never -- -c 3 -q 5 -z 10m http://$ALB_HOSTNAME/home


Based on the ScaledObject, KEDA creates an HPA resource and provides the required metrics to allow the HPA to scale the workload. Now that we have requests hitting our application we can watch the HPA resource to follow its progress:


~$kubectl get hpa keda-hpa-ui-hpa -n ui --watch
NAME              REFERENCE       TARGETS       MINPODS   MAXPODS   REPLICAS   AGE
keda-hpa-ui-hpa   Deployment/ui   7/100 (avg)   1         10        1          7m58s
keda-hpa-ui-hpa   Deployment/ui   778/100 (avg)   1         10        1          8m33s
keda-hpa-ui-hpa   Deployment/ui   194500m/100 (avg)   1         10        4          8m48s
keda-hpa-ui-hpa   Deployment/ui   97250m/100 (avg)    1         10        8          9m3s
keda-hpa-ui-hpa   Deployment/ui   625m/100 (avg)      1         10        8          9m18s
keda-hpa-ui-hpa   Deployment/ui   91500m/100 (avg)    1         10        8          9m33s
keda-hpa-ui-hpa   Deployment/ui   92125m/100 (avg)    1         10        8          9m48s
keda-hpa-ui-hpa   Deployment/ui   750m/100 (avg)      1         10        8          10m
keda-hpa-ui-hpa   Deployment/ui   102625m/100 (avg)   1         10        8          10m
keda-hpa-ui-hpa   Deployment/ui   113625m/100 (avg)   1         10        8          11m
keda-hpa-ui-hpa   Deployment/ui   90900m/100 (avg)    1         10        10         11m
keda-hpa-ui-hpa   Deployment/ui   91500m/100 (avg)    1         10        10         12m


Once you're satisfied with the autoscaling behavior, you can end the watch with Ctrl+C and stop the load generator like so:


~$kubectl delete pod load-generator


As the load generator terminates, notice that the HPA will slowly bring the replica count to min number based on its configuration.


You can also view the load test results in the CloudWatch console. Navigate to the metrics section and find the RequestCount and RequestCountPerTarget metrics for the load balancer and target group that was created. From the results you can see that initially all of the load was handled by a single pod, but as KEDA begins to scale the workload the requests are distributed across the additional pods added to the workload. If you let the load-generator pod run for the full 10 minutes, you'll see results similar to this.


Insights