On 2024/9/26 20:50, Ariel Miculas wrote:
On 24/09/26 07:23, Gao Xiang wrote:
...
It might be a fair comparison, but that's not how container images are
distributed. You're trying to argue that I should just use EROFS and I'm
First, OCI layer is just distributed like what I said.
For example, I could introduce some common blobs to keep
chunks as chunk dictionary. And then the each image
will be just some index, and all data will be
deduplicated. That is also what Nydus works.
I don't really follow what Nydus does. Here [1] it says they're using
fixed size chunks of 1 MB. Where is the CDC step exactly?
Dragonfly Nydus uses fixed-size chunks of 1MiB by default with
limited external blobs as chunk dictionaries. And ComposeFS
uses per-file blobs.
Currently, Both are all EROFS users using different EROFS
features. EROFS itself supports fixed-size chunks (unencoded),
variable-sized extents (encoded, CDC optional) and limited
external blobs.
Honestly, for your testload (10 versions of ubuntu:jammy), I
don't think CDC made a significant difference in the final
result compared to per-file blobs likewise. Because most of
the files in these images are identical, I think there are
only binary differences due to CVE fixes or similar issues.
Maybe delta compression could do more help, but I never try
this. So as I asked in [1], does ComposeFS already meet your
requirement?
Again, EROFS could keep every different extent (or each chunk,
whatever) as a seperate file with minor update, it's a trivial
stuff for some userspace archive system, but IMO it's
controversal for an in-tree kernel filesystem.
[1] https://github.com/project-machine/puzzlefs/issues/114#issuecomment-2369971291
Thanks,
Gao Xiang
