Now, when stacker is building image (not layer) N on top of image N-1, we have a few options:
- stacker passes the whole overlay for image N to puzzlefs, rather than the upperdir. puzzlefs builds this with no idea of what was in N-1. We hope for optimal blob sharing through chunking, but are unlikely to get it.
- stacker passes both the whole overlay for image N as well as the puzzlefs image for N-1 to puzzlefs build. Puzzlefs build learns to build an image that shares as much as possible between the two.
- stacker does what it does now, and we change the puzzlefs manifest format so that it continues to support ociv1-style layering of puzzlefs - and maybe mixed tar, squashfs, and puzzlefs - images.