03-multi-node-clusters

Pod assignment

The schedular does a decent job when it comes to schedule pods to nodes. There are however use cases where we want to schedule pods to specific nodes or reserve nodes for dedicated workloads.

nodeSelector

Using labels, nodeSelector allows to target a pod on to a specific node. To do so, we first have to label a node -

$ kubectl label node kube-worker1 disktype=ssd

We can than run our deployment with a nodeSelector field -

$ kubectl create -f node-selector.yaml

Even of we scale our application to multiple replicas (using kubectl scale) all of our replicas will land on kube-worker1 as only that node satisfies the constraint defined in the node selector field.

taints and tolerations

A nodeSelector is a property of pods that attracts them to certain nodes. Taints on the other hand allow for nodes to repel pods that don't tolerate the taint.

nodeSelector, taints and tolerations work together when we would like to reserve a node for dedicated workloads, for example a node with a GPU.

We can use the following steps:

1. Taint the node, which will prevent pods from landing on it
2. Add a label to the node
3. Define a nodeSelector and toleration in the pod manifest

We'll use the node we labeled before (kube-worker1) and add the following taint -

$ kubectl taint node kube-worker1 dedicated=gpu:NoSchedule

And create a deployment -

$ kubectl create -f taint.yaml

Configuration management

kubernetes provides us with 3 built in ways to manage a configuration for a pod -

1. Environment variables
2. ConfigMaps
3. Secrets

All of them can be found under the config-mgmt folder, and loaded to the cluster at once -

$ kubectl create -f config-mgmt/