Zchunk status update
So, after Flock, I figured it might be time for a zchunk status update.
First off, there's no way we're going to have the zchunked metadata
feature ready for Fedora 29, so we're pushing the feature back to
Fedora 30.
The one thing I *would* like to have ready for Fedora 29 is the actual
metadata creation, even if we don't have anything ready to consume it
quite yet. The idea behind this is that, if the metadata is being
generated, people can easily opt in to test the feature.
After some discussion on the rpm-ecosystem ML, I've restructured
createrepo_c so the zchunk metadata is in separate repomd.xml records
(with a -zck suffix) from the normal xml metadata. The latest build in
https://copr.fedorainfracloud.org/co
prs/jdieter/zchunk/packages/ has all the necessary changes and is built
against the official Fedora zchunk libraries (as opposed to a COPR
version).
Apparently COPR is struggling with Rawhide after branching, so the
current createrepo_c build is available in EPEL-7, Fedora 27 and Fedora
28.
I've also written a small script, available in the zchunk/contrib
directory that will generate an appropriate dictionary given a few
input files. I'm going to be experimenting to see what gives us the
most efficient dictionaries and then document SOP for generating
dictionaries in Fedora. After much discussion at Flock, it seems that
the best way to make this work is to have a fedora-repos-dicts package
that contains the dictionaries for all active Fedora releases.
Currently, the main thing I'm waiting on is for
https://github.com/rpm-software-management/createrepo_c/pull/92 to
finish being reviewed and then pulled.
Jonathan