Problem: we cannot tell if/when fluence builds will break against upstream
Solution: have a weekly run that will build and test images, and deploy on successful results. For testing, I have added a complete example that uses Job for fluence/default-scheduler, and the reason is because we can run a container that generates output, have it complete, and there is no crash loop backoff or similar. I have added a complete testing setup using kind, and it is in one GitHub job so we can build both containers and load into kind, and then run the tests. Note that MiniKube does NOT appear to work for custom schedulers - I suspect there are extensions/plugins that need to be added. Finally, I was able to figure out how to programmatically check both the pod metadata for the scheduler along with events, and that combined with the output should be sufficient (for now) to test that fluence is working.
This will be redundant to #49 - I made the mistake of doing the PR against the modular-fluence-build branch.
Problem: we cannot tell if/when fluence builds will break against upstream Solution: have a weekly run that will build and test images, and deploy on successful results. For testing, I have added a complete example that uses Job for fluence/default-scheduler, and the reason is because we can run a container that generates output, have it complete, and there is no crash loop backoff or similar. I have added a complete testing setup using kind, and it is in one GitHub job so we can build both containers and load into kind, and then run the tests. Note that MiniKube does NOT appear to work for custom schedulers - I suspect there are extensions/plugins that need to be added. Finally, I was able to figure out how to programmatically check both the pod metadata for the scheduler along with events, and that combined with the output should be sufficient (for now) to test that fluence is working.
This will be redundant to #49 - I made the mistake of doing the PR against the modular-fluence-build branch.