Diagnostic Questions

Before revealing the investigation path:

1. CI builds fail with failed to compute cache key but --no-cache builds succeed. Where is the cache stored, and how would you verify the cache is intact?

2. The cache manifest exists in the registry but the referenced blob does not. What could cause a manifest to reference a non-existent blob? What operation could leave the registry in an inconsistent state?

3. The registry's persistent volume is at 96% capacity and garbage collection has been failing. How does a full filesystem lead to cache corruption, and why does GC failure make it worse?

4. The fix involves resizing a PVC, running GC, and rebuilding the cache. Why are these Kubernetes operations (Domain C) rather than CI pipeline configuration changes (Domain A) or Linux disk management (Domain B)?

5. How would you prevent a full registry volume from breaking CI pipelines in the future? What monitoring, retention policies, or architectural changes would help?