Built this demo to practice deployment strategies and feature flags for AZ-400 and it was a much richer learning experience in the end.
In this version I added an API USER header to the API, moved the existing detect-language feature from a simple env variable to a user-based feature flag in Azure App Configuration, and added a new summarize feature (also controlled via App Configuration).
Why Azure Container Apps?
I already work with AKS, but wanted hands-on time with ACA. The big advantage is that I can keep the environment live and scale to zero when idle (no need to tear down and recreate everything like with AKS).
On top of that, the demo on my site is connected to the containers in ACA, so I can easily extend it later when the new summarization functionality is fully tested. My site is currently using the stable blue deployment and I'll switch it to the new green deployment once everything is validated.
Blue/Green + Feature Flags
- Blue/Green deployments: Used ACA revisions and labels to run stable (blue) and new (green) versions side-by-side with easy testing and instant rollback.
- Feature flags: Handled via Azure App Configuration to control feature exposure independently of the deployment. This enables gradual, targeted rollouts to specific users.
The real learning (when things broke)
Had difficulties with python not getting the feature flags and spent time chasing a 502 error that looked like an ACA routing/revision problem.
Root cause: My LLM-adapter container (running in ACA) was getting an incomplete prompt. GitHub Actions variables were causing issues with multi-line prompts during deployment, so I had to change the prompts.
Takeaways
Great practice with GitHub Actions, managed identities, RBAC, ACA revisions, App Configuration, and real-world multi service/environments debugging.
The best learning happens when things don't go as planned.
I'll follow up soon with a demo showing the full setup in action (blue/green switch + feature flag control).
Repo's: az-400-text-ai-githubactions, website
