kubernetes-yaml/learn/learn-kubernetes-master/kiamol/ch19/lab
2024-02-20 17:15:27 +08:00
..
pi 新增learn-kubernetes(https://github.com/yyong-brs/learn-kubernetes)相关文件 2024-02-20 17:15:27 +08:00
solution 新增learn-kubernetes(https://github.com/yyong-brs/learn-kubernetes)相关文件 2024-02-20 17:15:27 +08:00
README.md 新增learn-kubernetes(https://github.com/yyong-brs/learn-kubernetes)相关文件 2024-02-20 17:15:27 +08:00

ch19 lab

Setup

Deploy metrics-server if you need it (check with kubectl top nodes - no stats means you need it):

kubectl apply -f metrics-server/

Run the app:

kubectl apply -f lab/pi/

Confirm the metrics are coming through:

kubectl top pods -l app=pi-web-lab

Browse to the app and check the CPU spikes - e.g. http://localhost:8032/?dp=100000

Sample Solution

You need to label your node to indicate it's in the EU region - you can use any key and value for this, but you'll need to use the same in your affinity rules:

kubectl label node --all kiamol.net/region=eu

Pod with affinity rules

The updated deployment in solution/pi.yaml adds these settings:

  • node affinity - require to run on nodes with region=eu
  • pod anti-affinity - prefer to run on nodes without any other Pi pods
  • resources - add memory request for the HPA to use
  • replicas - start with 2 as that's the desired minimum
kubectl apply -f lab/solution/pi.yaml

You'll have two Pods running; browse to the app in a few tabs and both will spike CPU

HPA for scaling on CPU

The HPA spec in solution/hpa-cpu.yaml scales from 2 to 5 Pods based on target CPU utilization of 50%.

kubectl apply -f lab/solution/hpa-cpu.yaml

Make lots of browser requests in different tabs (or adapt the ch19/loadpi script in) and you'll see the Pods scale up to a maximum of five replicas:

Horizontal pod autoscaling in action

Teardown

Delete all the resources:

kubectl delete all,hpa -l kiamol=ch19-lab

And metrics-server if you deployed it:

kubectl delete -f metrics-server/