Docker

Terminology

• Container: environment that runs an applications that is not dependent on the OS. Kind of like a lightweight VM. Containers are stateless; if you need to update the components inside, create another container instead.
• Image: template to create a container. Its components are defined by a Dockerfile.
• Volume: storage area detached from the container for maintaining state. There are 4 types:
  • Volumes: managed by Docker itself, they’re stored in a separate part of the host filesystem (on Linux by default this would be var/lib/docker/volumes). Useful for production deployments.
  • Bind mounts: mount a specific directory or file from the host to the container, such as a project folder. Useful for development.
  • Tmpfs mounts: mount a temporary file system in the container, which is stored in the host system’s memory only, not on disk. Useful for sensitive information that should not be persisted between container restarts or for performance reasons (RAM is faster than disk).
  • Named pipes: used to create a pipe which can be accessed in a FIFO manner from both the container and the host system. Used primarily to store files in the host’s system memory when running Docker on Windows.
• Foreground/interactive vs background/detached: a detached container runs in the background whereas an interactive container will usually have a terminal of some sort for interacting with.

Commands

List your local images

• docker images

Delete images (many ways)

• You may use docker image rm, docker image remove or docker rmi for all of these commands
• Delete an image based on its ID
  • docker rmi <image_id> > It will return an error if there are multiple tags attached to it)
• Force delete an image based on its ID (it will delete all tags associated with it)
  • docker rmi -f <image_id>
• Delete an image based on its repository and tag
  • docker rmi <repository>:<tag>

Clean up images

1. docker images -q -f dangling=true
   • -q is for quiet mode
   • -f dangling=true filters the results to only show untagged images
2. docker image prune > removes all dangling images.
   • You may also use docker image prune -a to remove all unused images (images not referenced by any containers)

List your running containers

• docker ps
• docker ps -a > List all containers, including stopped ones.

Stop a running container

• docker stop <container_id>

Remove a container

• docker rm <container_id>
• docker rm -v <container_id> > This will remove both the container and its volumes (named volumes however will not be deleted)

List volumes

• docker volume ls

Delete a volume

• docker volume rm <volume_id>

Remove unused volumes

• docker volume prune > removes anonymous volumes not used by at least one container
• ddocker volume prune -a > removes all unused volumes, both named and anonymous

Run a Docker image inside a container

• docker run -it --rm image_name:tag
  • -it is a combination of -i (interactive mode) and -t (allocate a terminal).
  • --rm means that the container will be removed when exited.
  • You may find Docker images at the Docker Hub.
  • This command will use the entrypoint defined by the image. It won’t necesarily open a terminal inside the container.

Run a Docker image inside a container and override the entrypoint

• docker run -it --rm --entrypoint=bash image_name:version
  • This will override the entrypoint of your image and open a bash terminal inside the container instead.
  • Some images that have specific start-up sequences may have unexpected behaviour when the entrypoint is overriden

Run a Docker image inside a container and map a port in the container to a port in the host machine

• docker run -it --rm -p 9999:9696 image_name:tag
  • The syntax por port mapping is host:container (“outside:inside”)
  • In this example, the port 9696 within the container will be mapped to port 9999 in the host. You may access the container by accessing the host’s port 9999.

Create a Dockerfile with instructions to create a basic custom Docker image.

# set base image
FROM python:3.9

# set the working directory in the container
WORKDIR /app

# copy dependencies to the working directory
COPY requirements.txt .

# Install dependencies
RUN pip install -r requirements

# Copy all files within your project folder to the working directory of the image
COPY . /app

# command to run on container start
CMD ["python", "./main.py"]

• Docker will process each line as a layer. Some layers are cached, so in order to speed up build time, first copy and run immutable objects and then take care of your code/modules, as shown in this example.
• Base images are useful because they save a lot of work and build time. Choose a lean base image and avoid unnecessary packages.
  • Alpine or distroless images are a good beginning point.
  • For Python, go with an Alpine base image. Get the name from the hub page.
  • You can also use the default Python images if size isn’t too important, as in the example. They are Debian-based and come with apt-get support.
• Each container should only have one concern. Decouple applications into multiple containers.
• You may create an additional .dockerignore file in which you may declare any files and folders that must be ignored when using the COPY . /app command. Its syntax is identical to .gitignore files.

Create a slightly more complex Dockerfile with pipenv dependencies and specific entrypoints.

# set base image
FROM python:3.9

# (pipenv) install pipenv
RUN pip install pipenv

# set the working directory in the container
WORKDIR /app

# (pipenv) copy dependencies to the working directory
COPY ["Pipfile", "Pipfile.lock", "./"]

# (pipenv) Install dependencies
# (pipenv) We don't need a virtualenv in Docker, so we can install dependencies to the system
RUN pipenv install --system --deploy

# Copy the model
COPY ["predict.py", "model.bin", "./"]

# Expose a port on the container
# Remember to map the port to a port in the host when running the container!
EXPOSE 9696

# Specify entrypoint
ENTRYPOINT ["gunicorn", "--bind=0.0.0.0:9696", "predict:app"]

• The COPY instruction has 2 forms, shown here. The second form (like for pipenv in this example) must be used if any paths may contain whitespaces. The last param is always the destination directoy, which may be . or ./ for copying to the directory specified by WORKDIR.

Build an image based on a Dockerfile

• docker build -f Dockerfile -t my_image .
  • The default Dockerfile that the command will look for is $PATH/Dockerfile. If your Dockerfile is in the same directory that you will run the command and you have not named it something else, -f Dockerfile can be removed from the command.
  • my_image will be the name of your image. You may optionally tag it like so: my_image:my_tag.

Check how much storage space all of the images, containers and volumes are taking.

• docker system df

Docker compose

Example docker-compose.yaml file below. You may find a more elaborate example at https://docs.docker.com/compose/compose-application-model/

version: "3.9"
services:
  model-server:
    image: zoomcamp-10-model:v1
  gateway:
    image: zoomcamp-10-gateway:v2
    environment:
      - TF_SERVING_HOST=model-server:8500
      - MY_CUSTOM_VAR=${MY_CUSTOM_VAR}
    ports:
      - "9999:9696"

• version used to be required by docker-compose but not anymore. However, since there have been syntax changes in different versions, it’s always convenient to declare which version was used when writing the file.
• The app has 2 components (services): model-server and gateway
• Each service must have a Docker image.
• You may specify environment variables with environment and port mappings with ports
  • The dash (-) means that the entry is a list (array). In this example there are 2 lists; the environment list contains 2 elements and the ports list contains just one.
  • The array syntax requires the use of = for assigning values to variables, like in TF_SERVING_HOST=model-server:8500.
  • Entries in ports must always be surrounded with quotes.
    • Entries in other lists do not require quotes.
  • Port syntax is host:container (“outside:inside”). In this example, the port 9696 within the container will be mapped to port 9999 in the host. You may access the container by accessing the host’s port 9999.
    • IMPORTANT: you may access the container from your localhost with localhost:9999, but any container instantiated with the Docker Compose file must access that container using port 9696, because all containers declared within the Docker Compose file are inside the same virtual network and thus are not aware of the host’s external ports.
• Environment variables do not have to be hardcoded in the docker-compose.yml file. By default, you can create a .env file in the same folder as the docker-compose.yml file and define your variables there.
  • The TF_SERVING_HOST variable is hardcoded in the docker-compose.yml
  • The MY_CUSTOM_VAR variable is defined in the .env file. Within this file, you may declare a variable per line, using the VAR_NAME=VAR_VALUE syntax (note the lack of spaces; wrapping keys or values with quotes is optional).
  • You can define one variable per line inside this file, and you may use either the array syntax or the regular map syntax: ```yaml

    Array syntax

    environment:

  • TF_SERVING_HOST=model-server:8500
  • MY_CUSTOM_VAR=${MY_CUSTOM_VAR} ```

# Map syntax
environment:
   TF_SERVING_HOST: model-server:8500
   MY_CUSTOM_VAR: ${MY_CUSTOM_VAR}

Run the app. The command assumes that your compose file is named docker-compose.yml and is placed in the same directory that you’re invoking the command from.

docker-compose up

Run the app in detached mode.

docker-compose up -d

Shut down the app

docker-compose down

Shut down the app as well as any volumes, both named volumes declared in the compose file as well as anonymous volumes attached to containers.

docker-compose down -v

If you have multiple compose files with non-standard names and you want to specify which one you want to deploy, you may use -f path/to/compose_file.yml with both up and down.

Kubernetes

Kind

Create local cluster

kind create cluster

Delete local cluster

kind delete cluster

Load an image to the local cluster

kind load docker-image docker-image:tag

eksctl

Create a default cluster on EKS.

eksctl create cluster

Create a cluster with a config YAML file

eksctl create cluster -f eks-config.yaml

Example eks-config.yaml

apiVersion: eksctl.io/v1alpha5
kind: ClusterConfig

metadata:
  name: mlzoomcamp-eks
  region: eu-west-1

nodeGroups:
  - name: ng-m5-xlarge
    instanceType: m5.xlarge
    desiredCapacity: 1

• metadata contains both the name of the cluster as well as the AWS region.
• nodeGroups contains a list of node groups. In this example the list has a single entry.
  • desiredCapacity contains the amount of nodes inside the node group.
  • instanceType is the desired AWS EC2 instance type for the node group. All nodes will be of that instance type.

Delete a cluster

eksctl delete cluster -f eks-config.yaml

kubectl

kubectl command cheatsheet

Example deployment.yamlfile

apiVersion: apps/v1
kind: Deployment
metadata:
  name: <deployment-name>
spec:
  replicas: 1
  selector:
    matchLabels:
      app: <app-name>
  template:
    metadata:
      labels:
        app: <app-name>
    spec:
      containers:
      - name: <my-container>
        image: my-component-image:some-tag
        resources:
          limits:
            memory: "128Mi"
            cpu: "100m"
        ports:
        - containerPort: 9696
        env:
          - name: TF_SERVING_HOST
            value: <service-name>.<namespace>.svc.cluster.local:8500

• kind must be Deployment
• metadata.name contains the name of the deployment
• spec.replicas states how many pods should be replicated in the deployment. This example file only states 1 replica.
• spec.selector defines how the deployment finds which pods to manage. spec.selector.matchLabels is a rule that will match a label in the pod template (the label in this case is app:<app-name>)
• spec.template contains the blueprint for the pods:
  • metadata in this example contains the labels we use for the pods so that the deployment can find and manage them.
  • ..spec.containers contains a plethora of info:
    • name is the name of the containers inside the pod.
    • image is the Docker image to be used by the containers.
    • resources states the physical resource limits
      • For CPU, 100m means 100 milliCPUs, or 10% of the available CPU computing time.
    • ports contains the ports to use by the containers.
    • env contains names and values for nvironment variables, useful for apps to be able to find other containers by their internal cluster URL.
      • When defining a service, Kubernetes publishes a DNS entry inside the Cluster to make it possible for pods to find other pods. These DNS entries follow the <service-name>.<namespace>.svc.cluster.local:<port> format.
      • The default namespace is default.

Example service.yaml file.

apiVersion: v1
kind: Service
metadata:
  name: <service-name>
spec:
  type: LoadBalancer
  selector:
    app: <app-name>
  ports:
  - port: 80
    targetPort: 9696

• kind must be Service
• metadata.name contains the name of the service
• spec.type specifies the type of Service.
  • Internal services are of type ClusterIP. This is the default service type if this field is not stated in the file.
  • External services are of type LoadBalancer and are assigned an external IP.
• spec.selector contains the label to find the deployment to which it belongs to.
• spec.ports contains both the port of the service (port) as well as the port of the deployment (targetPort).

Multipass

Tool to run Ubuntu VM’s easily with command-line interface.

List available instances

• multipass list

Create and launch a new instance using the latest LTS release

• multipass launch --name my_instance

Access the instance shell

• multipass shell my_instance

Mount a shared folder in the instance

• multipass mount path/to/local/folder my_instance:path/to/instance/folder

Unmount all mounted folders of instance

• multipass umount my_instance

Stop an instance

• multipass stop my_instance

Start a previously created instance

• multipass start my_instance

Get info on a specific instance

• multipass info my_instance

Delete an instance (send it to the recycle bin)

• multipass delete my_instance

Recover a deleted instance

• multipas recover my_instance

Permanently delete all deleted instances

• multipass purge