deployment

Note: in prod we never run individual pods, we use something called deployment to orchestrate the creation of pods! Deployments are a way to express your desired state for an app to k8s

things to know:

• we can create a deployment just like we create pods using the cli or using yaml
• deployment contains info about the type of pod, replicas etc.
• deployments create a replica-set that actually creates the replicas of the pods
• k8s manages replica-sets for you, and it is not advised to manage it yourself!

Note: k8s tends to keep old replicasets around, so you might see multiple replicasets for the same deployment

# learn about deployment
kubectl create deployment -h | less
 
# create a deployment (creates deployment.apps)
kubectl create deployment test --image=httpd
 
# get more info about the deployment
kubectl describe deployment.apps test
 
# edit a deployment (not recommended)
kubectl edit depoyments.apps test
 
# delete a deployment
kubectl delete deployments.apps test
 
# fetch deployments
kubectl get deployments.apps
 
# generate yaml for deployment
kubectl create depoyment test --image=httpd --replicas=10 --dry-run=client -o yaml
 
# get replica set
kubectl get replicasets.apps
 
# more info on replica-sets
kubectl describe replicasets.apps

Note:

• when there is any issue in the k8s cluster always check the deployment events with the describe command to get info
• deployments have a default spec.strategy of RollingUpdate
• RollingUpdate means the deployment updates pods in a rolling update fashion (gradually scale down the old ReplicaSets and scale up the new one)
• while the Recreate strategy means all existing Pods are killed before new ones are created

this is how a deployment yaml file looks 👇

apiVersion: apps/v1
kind: Deployment
metadata:
  labels:
    app: nginx-depl
  name: nginx-depl
spec:
  replicas: 10
  selector:
    matchLabels:
      app: nginx-depl
  template:
    metadata:
      labels:
        app: nginx-depl
    spec:
      containers:
        - image: nginx
          name: ngingx
  strategy:
    type: RollingUpdate
    rollingUpdate:
      maxUnavailable: 1 # the max no of pods that can be deleted
      maxSurge: 1 # the max number of extra pods that can be created

Note: this basically means telling k8s to achieve the desired state of 10 pod replicase given it can have 1 pod unavailable and 1 extra pod running at any point in time! so, the state can be 9 pods fully available and 1 extra pod at any point in time!

To see the magic of RollingUpdate do the following:

• change the deployment config (maybe use an old image version)
• run the command watch -n 1 kubectl get pods
• apply the changed deployment in a window side by side

Links:

pods kubectl kubernetes deploying a simple app using k8s

202508242011