基于 CPU 利用率的节点自动缩放器

Question

We are working with HPA. Currently we are handling 100/rps per pod.

Use case is when the input rps is at 500/rps for 2 mins after that it goes up to 3k/rps.

We want to scale the number of pods from 5 upto 30 within 30 to 40 sec.

We are able to scale the pods in available resources for example default cluster size is 3 nodes. Problem comes when existing nodes are exhausted and cluster need additional node to schedule a pod.

Scheduling a new node in cluster it takes around 3min 40 sec to 4 min(approx.)

To overcome this problem we want to schedule new node when existing node cluster limit cross certain threshold (cpu utilisation upto 70%).

We are able to configure pod autoscaler using HorizontalPodAutoscaler kind. How can we solve the problem of node autoscaler wrt resource metrics.

Answer 1

You can automatically scale your GKE workloads based on metrics available in Cloud Monitoring. In this tutorial , you will see how to set up autoscaling based on CPU utilization.