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.
Now, when stacker is building image (not layer) N on top of image N-1, we have a few options: