Prerequisites

You need Go version v1.10 or higher.
To build, make sure Go dependencies are installed.
To get dependencies, run go get -d ./...

if it does not work, install package separately
(kuberenetes client-go)
go get k8s.io/client-go/...
go get -u k8s.io/apimachinery/...

(prometheus client-go)
go get github.com/prometheus/..

Verify for below folders of Kubernetes client-go library and prometheus after running above commands
$GOPATH/src/k8s.io/client-go/
$GOPATH/src/github.com/prometheus/

Update URLs for the Prometheus server in the file kubernetes-client/scheduler.go in the variables: promeMonitorAdr and promeIstioAdr

Add tags to machines in the Kubernetes Cluster

Custom scheduling algorithm reads the tags of the machines in the cluster to make scheduling decisions. To add a tag to machine, run kubectl label nodes ${node_name} VM=${VMn} PM=${PMn} RM=${RMn}.

Build the Scheduler Code

cd kubernetes-client
make build

Deploy Benchmark application

1. Run kubectl create namespace athena
2. To deploy using default Kubernetes scheduler, run make rund
3. To deploy using custom scheduler, run make runa

Deploy Robot shop application

1. Runkubectl create namespace athena
2. To deploy using default Kubernetes scheduler, run make runrobod
3. To deploy using custom scheduler, run make runroboa

Test the deployed application

To test the application from outside the Kubernetes cluster, we need to do port forwarding.

For benchmark application:
Port forwarding - kubectl port-forward svc/server-a-service ${port A}:8090 -n athena
To test,send GET http request to http://{url}:${port A}/a

For Robot Shop applcation:
Port forwarding - kubectl port-forward svc/web ${port B}:8080 -n athena
To test, send GET http request to http://{url}:${port B}/add/anonymous-99/K9/1

Benchmark Applications

1. Robot Shop
2. Custom Benchmark Application

Load Generator

1. ApacheBench
2. Apache JMeter

Notes

File kubernetes-client/scheduler.go contains algorithm of our custom scheduler and below are description about the code:

1. curServices and curNodes store the node and service information of the cluster.
2. func schedule() is called periodically for re-scheduling. All scheduling code goes in this function.
3. CreateDeployment(), UpdateDeployment(), DeleteDeployment() and BindPodToNode() make API calls to propagate scheduling decisions.
4. ListDeployment() and ListNode() make API calls to get cluster status information.
5. generateMockData() is used to generate mock data for testing.