Note: This document outlines guidances behind some development decisions behind the Online Boutique demo application.
Running the demo locally or on GCP should require minimal to no configuration unless absolutely necessary to run critical parts of the demo.
Configuration that takes multiple steps, especially such as creating service accounts should be avoided.
Demo application should work reasonably well when it is not deployed to GCP services. The experience of running the application locally or on GCP should be close.
For example:
- OpenCensus prints the traces to stdout when it cannot connect to GCP.
- Stackdriver Debugging tries connecting to GCP multiple times, eventually gives up.
Running the demo on the GCP must not reduce/lose any of the capabilities developers have when running locally.
For example: Logs should still be printed to stdout/stderr even though logs are uploaded to Stackdriver Logging when on GCP, so that developers can use "kubectl logs" to diagnose each container.
Each service should provide a minimal implementation and try to avoid unnecessary code and logic that's not executed.
Keep in mind that any service implementation is a decent example of “a GRPC application that runs on Kubernetes”. Keeping the source code short and navigable will serve this purpose.
It is okay to have intentional inefficiencies in the code as they help illustrate the capabilities of profiling and diagnostics offerings.