Kubernetes Deployments & Services

Link: 110 Kubernetes Index

Deployments¶

It is not advisable to create separate pods inside the k8s cluster since you cannot scale them.
The best method of creating pods is via deployment where you can increase or decrease their numbers easily.
After creating a deployment you can list the pods.
- These pods were created automatically by the deployment.
- These pods will now be managed by their particular deployments.
Selectors are used to connect pods with deployments because in k8s pods and deployments are separate objects.
- We have to assign pods to deployments. This is automatically done when a deployment is created.
Between deployment and pod there is another layer which is automatically managed by k8s called replica set.
- You can get the replica set using kubetctl get replicaset.
- Replica set manages the replicas of the pod.
- You will never have to create/delete or manage a replica set.
- You will work only with deployments.
Layers of abstraction → everything below deployment will be managed by k8s automatically.
Example of selectors and replica sets
- Notice the label value is the same as selector value of the deployment.
- The hash we see after the deployment-name 84... is the hash of the replica set and name after that i.e. v44.. is the unique id of the pod.

Pod naming: deployment_name-replicaset_hash-unique_id_pod

We cannot connect to these pods from our computer (running minikube) since our computer is external to k8s cluster.

For connecting to these pods we need to go inside the cluster.
Think as if minikube is running in the cloud and since the pods are not exposed to access them we need to ssh into the node which we do using minikube ssh

The IP addresses of the pods are dynamic and are assigned to pods when they are created.
- We should not rely on these IP addresses if we want to connect to these pods.
We create services in k8s if we want to connect to specific deployments using specific IP addresses.

When assigned to a deployment we can connect to the deployment from only inside the cluster i.e. ClusterIP is only accessible from within the cluster.
This will be a single IP address for the entire deployment.
These are specially useful if you have a database deployment and you want other deployments within the cluster to connect to it without exposing it to the outside world.

Since we are exposing nginx pods running at default port 80 we have to expose port 80.
- kubectl expose deploy deployment-name --port=8181 --target-port=80
kubernetes is a default service.
- We have single IP address for our deployment which could be used to connect to any of the pods.
Now even if increase or decrease the number of pods in the deployment we don’t have to worry about the IP addresses.
- The service we created acts as a permanent IP and a load balancer for all those pods in the deployment.

This will open deployment to outside world.
The most common solution to implement this is to have a load balancer IP address since pods belonging to the same IP address can reside in different worker nodes.
These load balancers are assigned by cloud providers and managed by cloud controller provider.
These are useful if you have to expose your web application to the outside world.
It is of two types, NodePort type and LoadBalancer type.

k expose deploy deployment-name --type=NodePort --port=8181 --target-port=80
Use NodePort only if you don’t have a load balancer or you are doing a local setup.

Using LoadBalancer service type
- It will always remain pending for us since we are using minikube.
- But if we are using external cloud provider then we will get an external ip.
- Load balancer from the cloud provider will be automatically provisioned for us.

k8s follows a rolling update.
- So if you update the docker image then new pods will be created first with the new image while previous pods are still running.
- Previous pods will be replaced one by one. It is very easy to replace image in all the pods via a single command.
k8s works on the desired state.
- So if you have 4 pods in your deployment and you manually try to delete one of the pods then you will see that k8s immediately creates another one since the desired state is of 4 pods.

Last updated: 2022-08-28