Docker and Kubernetes

Containers

This all started with VMWare where the total resource could be divided up to run more than one application on difference virtual machines. But VMWare required an OS on every machine and licenses in the case of Windows. They also needed managing, e.g. patching, Anti-virus and patching. Along came containers which shared the OS.

Docker

Introduction

Docker Inc. Docker is a company which gave the word technology for containers. They are now a company which provides services around the company.

Docker is Open source and known as Community Edition (CE). The company Docker releases an Enterprise Edition (EE).
The general approach is to

• Create an image (docker build)
• Store it in a registry (docker image push)
• Start a container from it (docker container run)

The differences between EE and CE are shown below

Architecture

Docker is made up of a docker engine and containers which are comprised of namespaces and control groups. Wiki defines namespaces as Namespaces are a feature of the Linux kernel that partitions kernel resources such that one set of processes sees one set of resources while another set of processes sees a different set of resources.

Docker Engine

Here are the core elements of the docker engine.

• dockerd: The Docker daemon is where all of your Docker communications take place. It manages the state of your containers and images, as well as connections with the outside world, and controls access to Docker on your machine.
• containerd: This is a daemon process that manages and runs containers. It pushes and pulls images, manages storage and networking, and supervises the running of containers.
• runc: This is the low-level container runtime (the thing that actually creates and runs containers). It includes libcontainer, a native Go-based implementation for creating containers

When we create a container, this is the workflow. It is the runc which actually the part which creates the container. The components containerd and runc standardized and it is easily possible to swap runc out for any oci compliant runtime.

Namespace Types

Within the Linux kernel, there are different types of namespaces. Each namespace has its own unique properties:

• A user namespace has its own set of user IDs and group IDs for assignment to processes. In particular, this means that a process can have root privilege within its user namespace without having it in other user namespaces.
• A process ID (PID) namespace assigns a set of PIDs to processes that are independent from the set of PIDs in other namespaces. The first process created in a new namespace has PID 1 and child processes are assigned subsequent PIDs. If a child process is created with its own PID namespace, it has PID 1 in that namespace as well as its PID in the parent process’ namespace. See below for an example.
• A network namespace has an independent network stack: its own private routing table, set of IP addresses, socket listing, connection tracking table, firewall, and other network‑related resources.
• A mount namespace has an independent list of mount points seen by the processes in the namespace. This means that you can mount and unmount filesystems in a mount namespace without affecting the host filesystem.
• An interprocess communication (IPC) namespace has its own IPC resources, for example POSIX message queues.
• A UNIX Time‑Sharing (UTS) namespace allows a single system to appear to have different host and domain names to different processes.

Control Groups

A control group (cgroup) is a Linux kernel feature that limits, accounts for, and isolates the resource usage (CPU, memory, disk I/O, network, and so on) of a collection of processes.
Now we have namespace we need a way to control the resources across the namespaces. This is where cgroups come in.

• Resource limits – You can configure a cgroup to limit how much of a particular resource (memory or CPU, for example) a process can use.
• Prioritization – You can control how much of a resource (CPU, disk, or network) a process can use compared to processes in another cgroup when there is resource contention.
• Accounting – Resource limits are monitored and reported at the cgroup level.
• Control – You can change the status (frozen, stopped, or restarted) of all processes in a cgroup with a single command.

Docker on Windwos

Finally a quick image to demonstrate the difference between native and hyper-v containers on windows. They made me do it gov.

Images

Image a made up of a manifest and layers.

Getting Images

We can get images using the pull command. The command is two step. It reads the manifests and then pulls the layers checking that the hashes pulled match the manifest hashes.

docker image pull redis

>Using default tag: latest
>latest: Pulling from library/redis
>33847f680f63: Pull complete 
>26a746039521: Pull complete 
>18d87da94363: Pull complete 
>5e118a708802: Pull complete 
>ecf0dbe7c357: Pull complete 
>46f280ba52da: Pull complete 
>Digest: sha256:cd0c68c5479f2db4b9e2c5fbfdb7a8acb77625322dd5b474578515422d3ddb59
>Status: Downloaded newer image for redis:latest
>docker.io/library/redis:latest

#To get the digest
docker image ls --digests
REPOSITORY           TAG       DIGEST                                                                    IMAGE ID       CREATED        SIZE
redis                latest    sha256:cd0c68c5479f2db4b9e2c5fbfdb7a8acb77625322dd5b474578515422d3ddb59   aa4d65e670d6   3 weeks ago    105MB

The Layers

We can find where our layers are by using

docker system info

>Client:
> Context:    default
> Debug Mode: false
>
>Server:
> Containers: 2
>  Running: 0
>  Paused: 0
>  Stopped: 2
> Images: 3
> Server Version: 20.10.7
> Storage Driver: overlay2
>  Backing Filesystem: extfs
>  Supports d_type: true
>  Native Overlay Diff: true
>  userxattr: false
> Logging Driver: json-file
> Cgroup Driver: cgroupfs
> Cgroup Version: 1
...
> Docker Root Dir: /var/lib/docker
> Debug Mode: false
> Registry: https://index.docker.io/v1/
> Labels:
> Experimental: false
> Insecure Registries:
>  127.0.0.0/8
> Live Restore Enabled: false

We can see it is using overlay2 storage driver and the root is /var/lib/docker so the layers are stored in /var/lib/docker/overlay2.
We can look at the history of the image and the commands used to make it with

docker history redis

>IMAGE          CREATED       CREATED BY                                      SIZE      >COMMENT
>aa4d65e670d6   3 weeks ago   /bin/sh -c #(nop)  CMD ["redis-server"]         0B        
><missing>      3 weeks ago   /bin/sh -c #(nop)  EXPOSE 6379                  0B        
><missing>      3 weeks ago   /bin/sh -c #(nop)  ENTRYPOINT ["docker-entry…   0B        
><missing>      3 weeks ago   /bin/sh -c #(nop) COPY file:df205a0ef6e6df89…   374B      
><missing>      3 weeks ago   /bin/sh -c #(nop) WORKDIR /data                 0B        
><missing>      3 weeks ago   /bin/sh -c #(nop)  VOLUME [/data]               0B        
><missing>      3 weeks ago   /bin/sh -c mkdir /data && chown redis:redis …   0B        
><missing>      3 weeks ago   /bin/sh -c set -eux;   savedAptMark="$(apt-m…   31.7MB    
><missing>      3 weeks ago   /bin/sh -c #(nop)  ENV REDIS_DOWNLOAD_SHA=4b…   0B        
><missing>      3 weeks ago   /bin/sh -c #(nop)  ENV REDIS_DOWNLOAD_URL=ht…   0B        
><missing>      3 weeks ago   /bin/sh -c #(nop)  ENV REDIS_VERSION=6.2.5      0B        
><missing>      3 weeks ago   /bin/sh -c set -eux;  savedAptMark="$(apt-ma…   4.15MB    
><missing>      3 weeks ago   /bin/sh -c #(nop)  ENV GOSU_VERSION=1.12        0B        
><missing>      3 weeks ago   /bin/sh -c groupadd -r -g 999 redis && usera…   329kB     
><missing>      3 weeks ago   /bin/sh -c #(nop)  CMD ["bash"]                 0B        
><missing>      3 weeks ago   /bin/sh -c #(nop) ADD file:45f5dfa135c848a34…   69.3MB

The non-zero size changes are generally a layer.
For even more information run inspect

docker image inspect redis

Registries

Where images live. By default Docker uses docker hub but you can use your own or other peoples. Docker divides registries between official and unofficial. Official images live at top of the namespace e.g docker.io/redis docker.lo/mysql. It is important to understand when pulling an image there are three components.

• Registry - e.g. dockerio
• Repo - e.g. redis
• Image (or Tag) - e.g. latest

It is just that Docker has defaults for registry and image so docker pull redis works but it is really doing docker pull docker.io/redis/latest

It was mentioned that the hashes on the file system for an image are referred to as content hashes and when images are pushed to the registry, before sending the layers are compressed and these hashes are referred to as distribution hashes.

Best Practice

• Use Official images
• Use Small images
• Be explicit referencing images (:latest noooo!)

Containers Detail

Copy on Write

As we know the container is a bunch of layers and the images are read-only. Only the read/write layer is writeable. When we change values in a running container we actually copy the original file and put it in the read/write layer. This is known as copy on write

Microservices

Perhaps the current buzz-word. Previously we built our apps to have all of the components on one server. With containers we can break these down into individual containers. The old way is referred to as a monolith. The new way is to separate services and connect them with well documented APIs.

Helpful Commands

Some useful bits

# You only need the best match for container in commands
docker container ls
>49978c1c8547   redis     "docker-entrypoint.s…"   19 seconds ago   Up 18 seconds   6379/tcp, 0.0.0.0:8088->8088/tcp, :::8088->8088/tcp   test

# Get the ports using part of the container name
docker port 499
>8088/tcp -> 0.0.0.0:8088
>8088/tcp -> :::8088

# Get IP of container
docker inspect -f "{{ .NetworkSettings.Networks.bridge.IPAddress }}" 499

# This works the same as inspect where you get all of the setting. the .NetowrkSettings is a filter

# Clean up containers
docker container rm $(docker container ls -aq) -f

Container Logging

We can set the logging driver in daemon.json (alternatives Gelf, Splunk, Fluentd...) and override for a container by passing --log-driver and --log-opts.
The standard way to view the logs is

docker logs <container>

Docker Swarm

Introduction

Swarm has two parts, the orchestrator and the secure cluster. The orchestrator does the start stop elements of running containers, the secure cluster is a way to run a group of containers with mutual authentication and encryption between them.

Swarm mode

With docker you can run containers as Single-engine mode or in something called Swarm mode. Whilst there a some notes on the options here I will leave detail for another day. This is just for awareness.

# To create a swarm
docker swarm init

When we do this

• The first node becomes the manager/leader
• They becomes the root CA (overrideable)
• They get a client certificate
• It builds a secure cluster store (which is etcd)
• This is distributed to every manager in store automatically
• Create cryptographic tokens, one for managers and one for workers

To join a new manager we pass the manager token and

docker swarm join-token manager
docker swarm join --token blarblarblar

To join a new worker we pass the worker token and

docker swarm join-token manager
docker swarm join --token blarblarblar

You can see the nodes with

docker node ls

We can rotate the worker token with

docker swarm join-token --rotate worker

We can lock/unlock swarms so they cannot be restarted without authentication with.

docker swarm lock
docker swarm unlock

Kubernetes

Kubernetes came out of Google and Open Source. Greek for helmsman or k8s (Kates). Kubernetes is an orchestrator and can schedule, scale and update containers. There are alternatives like Docker Swarm

Architecture

Overview

Apps are put in a container, wrapped in a pod and deployment details. They are provisioned on a Node inside a K8s Clustoer

Master

In general these are a hosted services on the cloud but you can run them locally on a linux box.

• This is the front-end to the control plane
• It exposes a RESTFul API consuming JSON and Yaml
• We send manifests describing our apps to this

The Cluster Store for users with large changes often separate this

The controllers controls

• Node Controller
• Deployment controller
• Endpoints/EndpointSlice

The scheduler

• Watches for new work
• Assigns Work to the cluster nodes

Kubernetes Nodes

Now its time for the Nodes. These are made up of three parts, the Kublet, Container runtime and the Kube Proxy

Kubelet

The Kubelet is the Main Kubernetes agent it

• Registers node with the cluster
• Watches API Server for work tasks (Pods)
• Executes Pods
• Reports to the master

Container Runtime

• Can be Docker
• Pluggable via Container Runtime Interface (CRI)
• Generally is Docker or containerd, other like gVisor and katacontainers exist
• Does the stop and starting of containers

Kube Proxy

• The Networking Component
• Does light-weight load balancing

Declarative Model/Desired State

We describe what we want the cluster to look like but we do not care how it happens. Kubernetes constantly checks the desired state matches the observed state.

Pods

A pod is a shared execution environment and can container one or more containers. It is the Pod which has a unique IP. Containers within the same pod must allocate unique ports if they are to talk to each other or the outside world. The same for volumes etc.When scaling you scale Pods not Containers.

Multi container Pods are for two different but complimentary containers which need to be aligned. An example of this might be a service mesh. For example the mesh container could be responsible for encryption or decryption. Note Deployment of Pods is atomic

Pods provide

• Annotations
• Labels
• Policies
• Resources
• Co-scheduling of contianers

and other Metadata.

Kubernetes Service Objects

As mentioned the IP address are provided for each Pod. However Pods die, we scales up and down and it would not be practical to use them so Kurbenetes provides the Service object which is describe in the deploymennt. With the use of labels the Service object knows which pods to send traffic to. By either removing labels from the pod the objects can be match generically.

In the example below both 1.3 and 1.4 will be matched of the Pods. Equally we could add a version to the service object and the unmatched version would not be used.
Service objects

• Only Send traffic to healthy Pods
• Can do session affinity (Sending of subsequent requests to the same Pod)
• Can send traffic to endpoints outside of the cluster
• Can do TCP and UDP

Kubernetes API Server

This is the API which exposes and RESTFul API. It exposes the various objects handles the requests. The objects are broken up into various sub categories. We use the command line kubectl to control this.

Installing Ubuntu

This was really easy. Had to use Firefox to get the dashboard to work on https://127.0.0.1:10443. Need to review how to get it to work on Chrome.

sudo snap install microk8s --classic
sudo usermod -a -G microk8s $USER
sudo chown -f -R $USER ~/.kube
su - $USER
microk8s status --wait-ready
microk8s enable dashboard dns ingress
microk8s kubectl get all --all-namespaces
microk8s dashboard-proxy

To be able to use the dashboard on chrome you need to go to chrome://flags/#allow-insecure-localhost and enable Allow invalid certificates for resources loaded from localhost. Firefox works by default.

Deploying Pods

Introduction

We do this by

• Creating App Code
• Create an image
• Store in a repo
• Define in a manifest
• Post to the API Server

Example

Creating App Code

Clone the App Code

git clone https://github.com/nigelpoulton/getting-started-k8s.git

Create an Image

Easy pezzy lemon squeezy. Where nigelpulton is your own Docker hub account/repository

cd getting-started-k8s/App
docker image build -t nigelpulton/getting-started-k8s:1.0 .

Store in repo

Now push the image to your docker hub account.

docker image push nigelpoulton/getting-started-k8s:1.0

Define in a manifest

We can define this in either Yaml or Json. I used yq eval -j to convert mine but it does say kubectl convert should work. I am using microk8s and it said it did not work.

apiVersion: v1
kind: Pod
metadata:
  name: hello-pod
  labels:
    app: web
spec:
  containers:
    - name: web-ctr
      image: nigelpulton/getting-started-k8s:1.0
      ports:
       - containerPort: 8080

And the json version

{
  "apiVersion": "v1",
  "kind": "Pod",
  "metadata": {
    "name": "hello-pod",
    "labels": {
      "app": "web"
    }
  },
  "spec": {
    "containers": [
      {
        "name": "web-ctr",
        "image": "nigelpulton/getting-started-k8s:1.0",
        "ports": [
          {
            "containerPort": 8080
          }
        ]
      }
    ]
  }
}

The "Kind" and "apiVersion" refer to the object. The objects you can use are grouped into types. Here is some of the objects in the apps group.
The default image tag by default will use the docker hub. You can add a prefix with the dns name if it is stored somewhere else.

Post to the API Server

You can post the manifest to the cluster with

cd Pods
kubectl apply -f pod.yml

# Get the status
kubectl get pods --watch -o wide

# Get All the Pod information
kubectl describe pods hello-pod

Example Cleanup

To clean up the deployment we can either

# Delete by using delete and the manifest
kubectl delete -f pod.yml
# Or delete using the name
kubectl delete pod hello-pod

Kubernetes Services

The Service has a frontend and a backend.

Frontend

The frontend is

• Name
• IP
• Port

The IP on the frontend is called a ClusterIP and is assigned by Kubernetes. It is only for use within the cluster. The name is the name of the service and that is registered with DNS.
Every container in every pod can resolve service names

Backend

Is a way for the service to know which pods to send track on to. The endpoint slice tracks the list of healthy Pod which match the services label selector.

Types of Service

• LoadBalancer External access via cloud load-balancer
• NodePrt External access via nodes
• ClusterIP (default) internal cluster connectivity

Creating a Services

We can create service in two ways

• Imperatively
• Declaratively

Imperatively

With imperative we can simply use the kubectl.

kubectl expose pod hello-pod --name=hello-svc --target-port=8080 --type=NodePort

microk8s kubectl get svc
>NAME         TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)          AGE
>kubernetes   ClusterIP   10.152.183.1     <none>        443/TCP          21h
>hello-svc    NodePort    10.152.183.147   <none>        8080:30737/TCP   11s

From there we can get to the service using the node port which is the outbound port shown for the service. In our case this is 30737. So going to the ip of the machine and port should show the service. e.g. htttp://192.xxx.xxx.1.70:30737. Node ports are allocated by Kubernetes between 32000 and 32767

Declaratively

Here is the yaml file to create the above service. Note I put the word name inside of app for the selector which resulted in the service having no endpoint.

apiVersion: v1
kind: Service
metadata:
  name: ps-nodeport
spec:
  type: NodePort
  ports:
  - port: 80
    targetPort: 8080
    nodePort: 31111
    protocol: TCP
  selector:
    app: web

And here is a picture to explain the ports.

• 31111 is the port for external access
• 80 is the port the cluster is listening on
• 8080 is the port exposed by the pod and the defined in the docker container

We can check we have the right name (label) in the service with the command

kubectl get pods --show-labels
>NAME        READY   STATUS    RESTARTS   AGE   LABELS
>hello-pod   1/1     Running   0          79m   app=web

Clean up

Deleting service can be done with

kubectl delete svc hello-svc

Deployments

Introduction

To deploy a solution we need

• Deployment (Updates and Rollbacks)
• Replication Set (Scalability, Reliability, Desired State)
• Pod (Scheduling and Execution)
• App (App code)

Rolling Updates and Rollbacks

To update we simply provide a new version of the deployment. We can define how this happens e.g. wait 10 mins before OK. Kubernetes keeps the old pods to allow the user to reverse the deployment by using a previous version of the deployment. Here we see the rollback where the green pods are being replace by the original pods.

Deployment Section

The example below show a deployment and provides and overview. The labels in the deployment spec must match the label in the Pods spec. The always is for security and ensures the code not the local copy.

Full Example

Here is a full example. There are many options under the strategy we can configure. In this example we have opted for

• minReadySeconds The amount of time to wait and see if container is working before proceeding
• maxUnavailable Number which should be available during update
• maxSurge The number it can exceed the desired amount by.

So in this example there will be 6 pods during the update and it will wait for 5 seconds and reduce back down to 5.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: web-deploy
  labels:
    app: web
spec:
  selector:
    matchLabels:
      app: web
  replicas: 5
  minReadySeconds: 5
  strategy:
    type: RollingUpdate
    rollingUpdate:
      maxUnavailable: 0
      maxSurge: 1
  template:
    metadata:
      labels:
        app: web
    spec: 
      terminationGracePeriodSeconds: 1
      containers:
      - name: hello-pod
        image: nigelpoulton/getting-started-k8s:2.0
        imagePullPolicy: Always
        ports:
        - containerPort: 8080

Doing Update

To perform an update we use the deploy command

#Do Deployment
kubectl apply -f deploy-complete.yml

#Monitor the deploy progress with
kubectl get deploy

#Monitor the replicate sets progress with
kubectl get rs

Reversing the update

As discussed the old replicates still reside with in kubernetes. We can see this by looking as the replication sets. We can reverse the update by

#Get the deployment history 
kubectl rollout history deploy web-deploy

>deployment.apps/web-deploy 
>REVISION  CHANGE-CAUSE
>1         <none>
>2         <none>

#To reverse we just need to provide the revision to rollback to.
kubectl rollout undo deploy web-deploy --to-revision=1

Monitoring Deployment

We can see both the deploy and the replication sets with

kubectl get deploy
kubectl get rs

We can use describe ep to see the endpoints for our configuration.

microk8s kubectl describe ep

>Name:         ps-nodeport
>Namespace:    default
>Labels:       <none>
>Annotations:  endpoints.kubernetes.io/last-change-trigger-time: 2021-08->14T17:53:23+12:00
>Subsets:
>  Addresses:          10.1.114.202
>  NotReadyAddresses:  <none>
>  Ports:
>    Name     Port  Protocol
>    ----     ----  --------
>    <unset>  8080  TCP
>
>Events:  <none>
iwiseman@OLIVER:

Self Healing and Scaling

We can delete pods and nodes and Kubernetes will fix the problem. This is known as self healing. We can manually deploy with more replicas to scale up and down. There is Horizontal Pod Autoscaler and Cluster Autoscaler as well.