Updating a Kubernetes Deployment and waiting for it to roll out in a shell script

For some Kubernetes cluster operations (e.g. deploying an update to a small microservice or app), I need a quick and dirty way to:

  1. Build and push a Docker image to a private registry.
  2. Update a Kubernetes Deployment to use this new image version.
  3. Wait for the Deployment rollout to complete.
  4. Run some post-rollout operations (e.g. clear caches, run an update, etc.).

There are a thousand and one ways to do all this, and many are a bit more formal than this, but sometimes you just need a shell script you can run from your CI server to do it all. And it's not too hard, nor complex, to do it this way:

# Build, push, deploy, and run post-deploy tasks for myapp.

# Define the version to be deployed (e.g. a git hash or semver tag).

# Build and push the container image.
docker build -t registry-url:namespace/myapp:$version .
docker push registry-url:namespace/myapp:$version
docker tag registry-url:namespace/myapp:$version registry-url:namespace/myapp:latest
docker push registry-url:namespace/myapp:latest

# Update the myapp deployment in Kubernetes, in the namespace 'namespace'.
kubectl set image deployment/app myapp=registry-url:namespace/myapp:$version -n namespace

# Check deployment rollout status every 10 seconds (max 10 minutes) until complete.
ROLLOUT_STATUS_CMD="kubectl rollout status deployment/myapp -n namespace"
until $ROLLOUT_STATUS_CMD || [ $ATTEMPTS -eq 60 ]; do
  ATTEMPTS=$((attempts + 1))
  sleep 10

# Run other post-deployment tasks, now that all new Pods are present, and old ones are gone.
MYAPP_POD=$(kubectl get pods -l app=myapp -n namespace | grep "^myapp.*Running" | awk '{print $1}')
kubectl exec "$MYAPP_POD" -n namespace -- bash -c "vendor/bin/drush --uri myapp.com cr"

The example above achieves all the tasks I need to do for a common app deployment; note that it's better to have a more formal process in place for most things... but when you just need a system that works, and is easy to debug and/or run manually as needed, this works fine.

Note that I also tried getting kubectl wait to work in this situation, but couldn't find a way to make it work with the rollout. It works great for deployment status:

kubectl wait --for=condition=available --timeout=600s deployment/myapp -n namespace

...but this will happily return once the deployment is available, but there could still be old pods in Terminating status, or new pods that are not fully online yet.


Very useful about the until part! I was working on Kafka in K8, where some connectors will fail if it can't reach the bootstrap server. I imagine this piece of code simulating watching on ready event would be even more useful in the near future...!

kubectl rollout status -w deployment/frontend # Watch rolling update status of "frontend" deployment until completion