Diagnosing Bazel Cache Misses

Building with Bazel can sometimes lead to caching issues, particularly when dealing with non-deterministic repository rules that involve package managers like pip, npm, or cargo, which might result in varied outputs due to "post install" scripts. This blog post explores how such inconsistencies, particularly with Python distributions, can lead to cache misses by affecting the build graph inputs. It highlights the importance of understanding repository rule determinism, where deterministic rules like http_archive are contrasted with non-deterministic ones due to their reliance on package managers that may execute additional processes during installation. The post provides a method to identify file differences by hashing and comparing dependencies' shasums over different executions, suggesting that solutions can range from excluding certain files from the build inputs to more complex fixes, which will be covered in subsequent posts in the series.