best-fit-scheduler

Best-fit bin packing scheduler configuration for Kubernetes using the "MostRequestedPriority" priority function of kube-scheduler

Deployment

To use this scheduler

1. Deploy the scheduler-config ConfigMap which specifies the policies to configure kube-scheduler with

kubectl create -f manifests/scheduler-policy-config.yaml

2. Deploy another instance of kube-scheduler named best-fit-policy which uses the above policy

kubectl create -f manifests/best-fit-scheduler.yaml

3. To use this scheduler (and not the default kube-scheduler) specify spec.schedulerName to best-fit-scheduler. An example deployment is given

kubectl create -f examples/nginx-deployment.yaml

Tests

Default scheduler

After deploying nginx with default scheduler

kubectl get po -owide shows

NAME                               READY   STATUS    RESTARTS   AGE   IP          NODE                                     NOMINATED NODE
nginx-deployment-884c7fc54-cxdqc   1/1     Running   0          25s   10.24.2.6   gke-ca-test-default-pool-27725aea-p09h   <none>

Now scale the deployment by increasing the number of replicas to 3

kubect scale deploy nginx-deployment --relicas=3 and check with kubectl get po -owide again

NAME                               READY   STATUS    RESTARTS   AGE   IP           NODE                                     NOMINATED NODE
nginx-deployment-884c7fc54-cxdqc   1/1     Running   0          59s   10.24.2.6    gke-ca-test-default-pool-27725aea-p09h   <none>
nginx-deployment-884c7fc54-kt9pm   1/1     Running   0          3s    10.24.0.23   gke-ca-test-default-pool-27725aea-kzfg   <none>
nginx-deployment-884c7fc54-tdbpx   1/1     Running   0          3s    10.24.1.13   gke-ca-test-default-pool-27725aea-x03n   <none>

Notice that the pods are placed on different nodes. This is because kube-scheduler by default uses the LeastRequestedPriority for prioritizing nodes.

best-fit scheduler

After deploying nginx with the best-fit scheduler

kubectl get po -owide shows

NAME                              READY   STATUS    RESTARTS   AGE   IP           NODE                                     NOMINATED NODE
nginx-deployment-667764b9-dpn6q   1/1     Running   0          22s   10.24.0.20   gke-ca-test-default-pool-27725aea-kzfg   <none>

Now scale the deployment by increasing the number of replicas to 3

kubect scale deploy nginx-deployment --relicas=3 and check with kubectl get po -owide again

NAME                              READY   STATUS    RESTARTS   AGE   IP           NODE                                     NOMINATED NODE
nginx-deployment-667764b9-9b9z8   1/1     Running   0          4s    10.24.0.22   gke-ca-test-default-pool-27725aea-kzfg   <none>
nginx-deployment-667764b9-cpv9p   1/1     Running   0          4s    10.24.0.21   gke-ca-test-default-pool-27725aea-kzfg   <none>
nginx-deployment-667764b9-dpn6q   1/1     Running   0          59s   10.24.0.20   gke-ca-test-default-pool-27725aea-kzfg   <none>

Notice that all three replicas are places on the same node.