Implement git caching for health checks

[chef]

Implement git caching for health checks

[schisamo]

I'm feeling pretty 👎 on this feature. Even if a new package is created from software all of which was git cached it seems we should still perform health check.

Is health checking really that big of an issue?

[seth]

I'm confused by your comment @schisamo.

Currently the health check is inefficient because it checks more files than it needs to and because it rechecks files that we know haven't changed (cached builds).

When using local vms to build, the healthcheck step takes enough time to be worth improving. And just in general it seems reasonable to find ways to do a computation once rather than many times.

The proposal, to be clear, is not to optionally disable the healthcheck.

The proposal is to perform a health check after each build. At that point, we know all of the files that the build touched. We health check them. If we use the cache for a build, we can skip the health check. Might also have the advantage of failing builds for bad health early.

[sethvargo]

@schisamo @seth is correct (holy crap this is going to get confusing) 😄. @seth I'm going to steal your comment and add it to the PR description for clarity.

[schisamo]

@seth Thanks for the clarification. I can get onboard with this feature as described. I think the title of this issue was just confusing as it implies we are caching the results of the health check.

[seth]

:)

We would be caching the result of the health check, but not in a bogus way :)

[sethvargo]

We discussed this as a team today and decided this is not a top-priority. If health checks are taking a long time, we should fix the health check, not cache the results.

[seth]

Can you give more detail? If the result was: let's close this because we're not fixing it anytime soon, I'm fine with that. If the decision was, we don't want a fix that caches the health check, then I'm curious for more detail.

[sethvargo]

Sure. It's two parts:

1. As you mentioned, this isn't likely something we want to implement soon.
2. As we learned with the git caching, cache invalidation is hard.