Entrypoint cache should cache by digest

[imjasonh]

When we lookup entrypoints for remote images, we cache it for later lookups. But, we cache by the requested image name, and not the resolved image by digest.

So if a user changes the entrypoint to an image without changing its tag, the new entrypoint might not get picked up since it will already have been cached. Instead, we should always resolve the image tag to its digest, then lookup in the cache whether that digest's entrypoint is already known.

Since we'd be resolving the image tag -> digest before running the build, we can also request the exact image by digest when we run the pod. This removes a race condition where the entrypoint changes between the time the build is requested and the time the step starts running. We should include builder image digests in TaskRunStatus for visibility.

To avoid looking up the digest for a tag each time, we could keep a separate cache for that lookup that expires after some (configurable) amount of time.

[Zyqsempai]

@imjasonh Should we also use LRU cache for "tag->digest" thingy?

[bobcatfish]

Yep I think so @Zyqsempai ! Note that @mathspanda is converting our existing cache to lru in #350 so we can probably build on that :D

[Zyqsempai]

@bobcatfish Then, it will be good to merge his PR asap;)

[zhyuri]

Hi @Zyqsempai , are you working on this issue?

Just want to confirm because I'm also looking for some issues to work on.

[Zyqsempai]

@zhyuri Yep, i do. Sorry.

[zhyuri]

Cool! No problem. 🙌

[bobcatfish]

Thanks @Zyqsempai !!! 🎉🎉🎉