Container Registries
Container vs Image​
First to clear up some terminology, a container 'image' is the definition, or blueprint for a container. Whereas an actual 'container' is an actual running instance of that packaged software. It is very similar to Object Oriented Programming (OOP) where you can have a class definition, and an object which is an instantiation of that class, or indeed many objects created from the class definition.
In the previous page we used the nginx container image to instantiate our own running container that we named mycontainer. We could have similarly created a bunch of running containers called mycontainer1, mycontainer2, ..., etc. all from that single image definition.
Container Image Address​
The format for specifying the location of a container image that you want to create a container from is <registry-address>/<namespace>/<image-name>:<image-tag>
So we can download a image to our computer using the docker pull command, like so:
docker pull container-registry-website.com:6000/mycontainers/container1:v1.0
The above command obviously will not work, so don't run it in your terminal. There's no container registry at container-registry-website.com, unless you decide to make one yourself!
Dockerhub​
So where did the Nginx image come from in the previous page's command docker run --name mycontainer nginx if we're only specifying the word nginx and no other parts of the address?
Well if you don't specify some details, Docker has defaults. Dockerhub is a container image registry that Docker manage, and if you don't specify a registry address, it assumed you want to access there. The actual registry address is docker.io.
Dockerhub uses the namespace part of the address to specify which user owns the image. But if you leave out this part too, docker assumes you want their own official images managed by Docker. This technically is under the library namespace. The image-name is nginx, and that still stays. Finally, if you don't specify a image-tag, docker assumed you want the image tagged latest. If there isn't an image tagged latest in the repository then Docker will throw an error. Similarly when you build an image (which we will see in the next lesson) and don't specify a tag, latest will be used by default.
So docker pull nginx is actually an alias for docker pull docker.io/library/nginx:latest.
If you want to get an image that isn't a an official docker image, but is still on Dockerhub, you can still leave out the registry address, but just include the namespace, which is the username of the account that manages the image.
For example we can download a container image pre-prepared with the Hugging Face Transformers library, with PyTorch dependency, set up for use on a GPU, which is managed by the Hugging Face team on Dockerhub with the command docker pull huggingface/transformers-pytorch-gpu:4.28.1
The above image is around 9 GB, so no need to run the command and have it sitting on your computer! If you did run it, you can first have a look at the images you've downloaded with docker images. And you can remove it with docker rmi huggingface/transformers-pytorch-gpu:4.28.1
Dockerhub allows you to host public images for free so I'd highly recommend making an account to practice. Note - they charge for hosting private image repos.
AWS ECR - Elastic Container Registry​
As mentioned in the about section, we will use free and open-source tooling. But we won't stop there with local toy examples. The goal is to get to production. So in our tutorials we will also (optionally) be using Amazon Web Services (AWS) to get to global scale. We'll learn more about AWS in the Cloud Computing section, but for now just know they have their own container image registry called Elastic Container Registry (ECR) where you can have public and private repos for you own images. This of course costs money, but they have a generous free tier for your first year of use.
Others​
Google Cloud Platform (GCP) and Microsoft Azure also have container registries called Google Container Registry and Azure Container Registry, respectively. Just to make you aware, but we won't be using them.