Container Canary - Lightning Talk

Q: What is Container Canary?
A: It’s a tool for validating container images

▪ You can define your container requirements as
specification that can be checked into version
control.
▪ Requirements are implemented as Checks that
follow the Kubernetes core.v1.Probes API.
▪ Using the canary CLI you can validate a
container against the specification.
▪ This effectively gives you unit tests for your
containers 🎉

Transcript

1. SciPy 2022 Lightning Talk - Container Canary
   Jacob Tomlinson
   

   View Slide

2. Regressions 😔
   Recently I made a change to the RAPIDS Docker image so that they could be used on the
   Kubeflow Notebooks platform. In doing so I broke an NVIDIA service.
   To avoid making my colleagues sad again in the future I built Container Canary.
   

   View Slide

3. Q: What is Container Canary?
   ▪ You can define your container requirements as
   specification that can be checked into version
   control.
   ▪ Requirements are implemented as Checks that
   follow the Kubernetes core.v1.Probes API.
   ▪ Using the canary CLI you can validate a
   container against the specification.
   ▪ This effectively gives you unit tests for your
   containers 🎉
   A: It’s a tool for validating container
   images.
   

   View Slide

4. Q: When is this useful?
   A: When building container images with the
   goal of them being used on different
   platforms.
   ▪ Many compute platforms support
   bring-your-own-container, but have some
   requirements the container must meet.
   ▪ Images may be expected to run services on
   specific ports, use a specific UID, use a
   specific working directory, etc.
   ▪ These requirements are often documented but
   not enforced in a technical way.
   ▪ Container Canary allows you to codify those
   requirements and test containers ahead of
   time
   

   View Slide

5. Example: Binder
   Binder specifies requirements in it’s
   documentation for using your own container
   images in Binder environments.
   If your container doesn’t fulfil these
   requirements then the notebook job will fail
   to start and give you an error. But not
   necessarily tell you what went wrong.
   You might have an existing container image
   that you want to use with Binder, but you’ll
   need to manually check that you’ve done all
   the steps in the documentation to ensure it
   works.
   

   View Slide

6. Example: Binder
   With Container Canary we can capture those
   requirements as a manifest. On the left you
   can see the example Binder manifest we
   provide in the GitHub repo.
   Now you can test your container before you
   try to use it with Binder and identify
   exactly which requirement is failing.
   Below is an example of validating the
   Docker image I used for my GPU Python
   tutorial yesterday.
   

   View Slide

7. How can you help?
   ● Test your containers!
   ● Star the project
   ● Put your manifests on GitHub for
   others to enjoy
   

   View Slide

8. Thank you!
   

   View Slide