Watches deployments and container images for change in tag digest and auto re-deploys them. Tested in a 3 cluster VM setup and minikube.

kubectl apply -f https://github.com/Preetam-Das/kubewatcher/releases/download/v0.0.1/kubewatcher-install.yaml

kubectl delete -f https://github.com/Preetam-Das/kubewatcher/releases/download/v0.0.1/kubewatcher-uninstall.yaml

• Have a deployment running with a public container image.
• Install kubewatcher. It will auto detect the deployement.
• Make some visible changes in the deployment's image and push it to the registry.
• Wait for about a minute (see note 3) and it will auto deploy the deployment with the new updated image.

The whole application (calling it kubewatcher for now) is designed to work with the k8s environment as k8s pods. So the application that updates other deployments is itself a deployment if it makes sense. This way it should be possible that kubewatcher can update itself (not tested yet .. update it seems to work).

There are two parts to it :

• a deployment that watches for other deployment, images & patches them for update
• a cronjob which polls dokcer registry

The communication between the two is done using several Custom Resources which are basically lists and maps for storing container image, deployment, digest data. Once the cronjob finds that an image got updated (see note 1) it notifies the deployment using a to-update-images (names are all mangled up need to fix) CR list and the deployment re-deploys all the deployments where the image is present.

1. The current implementation checks for changes container image of the same tag, by comparing the digest values. It doesn't updates on every new tag pushed as tags don't directly relate to version numbers.

2. Currently kubewatcher only works for public repositories in the docker registry.

3. Rate limit for personal account is 200 per 6 hours, which is about 1 request per 2 minute. The current implementation polls the api 1 time in a minute. So if you keep this running for 6 hours you might get rate limited (NEED TO FIX THIS).

Several improvements are possible. I am limited by time so anyone wishes to work on these, please do and send a PR.

• Provide user customization like chosing what sepcific deployments or container images to watch for. Use config maps for this and read them in using the k8s api.

• Remove the cronjob component. I only did it because i was told to do so but this seems unnecessary so probably best to remove this and do the polling in the kubewatcher script only in a separate thread.

• Better documentation.

• Support for private images using secrets.

• Support for Kustomize and Helm.

• Error handling. Currently there is none.

• Async I/O. Cuz why not.

• Write this in Go. Everything in the cloud seems to be written in Go.

  .. and lot more.