The most common problem kubernetes controllers and operators authors face today is to benchmark their tools in large environments with 1000’s of workloads running in it. As controllers & operators come with custom logic to perform an action when a specific event occurs in the cluster to achieve the intended state.
Creating 100 node cluster for benchmarking the Kubernetes tools is definitely expensive. Creation, maintainance & deletion is also a tiring job. What if we can create a mock kubelet for scale tests by just running a deployment in your cluster? Yes, This is possible and you can leverage out of this tool in benchmarking the kubernetes tools.
All we need is to simulate loaded environments to perform scale test on our tools. Creating loaded environemnts comes with cost. And we don’t want to spend that huge figures in benchmarking our tools.
When I refer to Kubernetes tools, I mean the Kubernetes Controllers, Operators, kubectl plugins, etc.. which interact with api-server for events and perform kubernetes resource scheduling.
So I will be explaining how to simulate a mocklet that can hold 1000’s pods in it’s memory and this mock kubelet can be connected to your existing cluster by just running a deployment in your cluster.
Running the below command will create a new mocklet node in your cluster
kubectl create -f https://github.com/VineethReddy02/mocklet/blob/master/k8s-deployment.yaml
Now you can notice mocklet is added into your cluster
NAME STATUS ROLES AGE VERSION gke-gke5684-default-pool-1y5e7l53-kphx Ready <none> 4h23m v1.14.10-gke.27 gke-gke5684-default-pool-1y5e7l53-x5kj Ready <none> 4h23m v1.14.10-gke.27 mocklet Ready agent 2m32s v1.15.2-vk-N/A
Creating multiple deployments will create multiple mocklets in your cluster to run desired number of kubernetes resources and scheduling resources specific to a mocklet.
Now you can deploy 1000’s of pods by providing the node selector value as mocklet and mocklet toleration. This will make sure all the test data you are creating is scheduled on the desired mocklet.
The mocklet project is completely inspired from Virtual Kubelet mock provider. Thanks to the Virtual Kubelet community.
The bigger challenge is how do I create 1000’s of pods, Deployments, Replicasets, ReplicationControllers, StatefulSets, Jobs and Cronjobs?
To create the Kubernetes resources at scale in a single shot. I have implemented tool called k8s-scaler. This tool is highly configurable and helps you to create Kubernetes resources which runs pods eventually with higly configurable properties such as number of containers, inclusion & exlusion of namespaces during resource creation, number of Kubernetes resources like deployments, daemonsets, etc… , number of replicas per Kubernetes controller.
Creating 5000 deployments with replica count as 5 per instance and number of containers per pod as 3 in scale namespace with node-selector & toleration is as simple as
./k8s-scaler create deployments --scale 5000 --replicas 5 --containers 3 --namespace scale --node-selector type=mocklet --toleration mocklet.io/provider=mock
Note: Using k8s-scaler you can also create/delete namespaces, daemonsets, statefulsets, replicationcontrollers, replicasets, jobs and cronjobs
k8s-scaler also helps you in listing number of kubernetes resources per namespace as shown below.
vineeth@vineeth-Latitude-7490 /bin (master) $ ./k8s-scaler list NAMESPACE DEPLOYMENTS REPLICASETS DAEMONSETS STATEFULSETS PODS JOBS CRONJOBS REPLICATION-CONTROLLERS test 3000 3000 1000 500 7486 30 10 30 default 1300 1300 456 250 5642 10 5 5 kube-system 8 11 4 0 15 0 0 0 mocklet 3500 4000 1200 400 9348 50 30 35
Using the mocklet and k8s-scaler you can create the large environments with ease. Running mocklet will provide the mock kubelet to run the desired number of resources and using k8s-scaler you can create desired number of kubernetes resources by running a single command in any kubernetes cluster.
I hope these tools will help you in scale testing the tools built around kubernetes.