It is only responsible for deciding which pod goes on which node.
It doesn’t do the actual pod placement on the nodes.
That is the job of the kubelet.
The kubelet is the captain of the ship and creates the pod.
The kube-scheduler only decides which pod goes where.
When there are many ships and many containers, you need to make sure the right containers ends up on the right ship.
For example, different sizes of ships and containers.
Need to make sure a ship has sufficient capacity to accommodate those containers.
Need to place the container on the right ship to reach the correct destination.
The scheduler decides which nodse pods are placed on, depending on certain criteria.
May have pods with different resource requirements.
Can have nodes in the cluster, dedicated to certain applications.
How does the scheduler assign the pods?
The scheduler looks at each pod and trys to find the best node for it.
The below example pod has a set of CPU requirements. The scheduler goes through two faces, to identify the best node for the pod.
The first phase has the Scheduler filtering out the nodes, that do not fit the profile of the pod.
For example, the nodes that do not have sufficient CPU and memory resources , requested by the pod.
We are now left with two nodes, where the pod can be placed.
How does the Scheduler then pick one of the above nodes?
The second phase is where the Scheduler ranks the nodes, for the best fit for the pod.
It uses a priority function, to assign a score on a scale of 0-10. The scheduler calculates the amount of resources that would be free on the nodes, after placing the pod on them.
In the example case, the node on the right would have 6 CPUs free, when the pod is placed on it. This is four more CPU cores than the node on the left.
Therefore the node on the right is a better rank and therefore it wins.
The Scheduler works at a high level and the priority can be customised and you can write your own scheduler.
There are many further topics such as Resources Requirements and Limits, Taints and Tolerations and Node Selectors/Affinity
There is an entire section dedicated to scheduling, which will go through the above.
How do you install the kube-scheduler? Download it, extract it as a binary and run it as a kube-scheduler service:
If you set it up using the kubeadm tool, the kubeadm tools deploys the kube-scheduler as a pod in the kube system namespace on the Master Node.
The options and pod definition file is under /etc/kubernetes/manifests/kube-scheduler.yaml
You can also see the running process and effective options by listing the process on the Master Node, using ps -aux | grep kube-scheduler
