Ideas for New Repodata

problems and goals

The current metadata has helped make searching the information much more reasonable - however, it suffers from the problem that as the repos grow in number of pkgs users are forced to download massively large chunks of data they, in fact, do not NEED.

One of the goals of this discussion is to see if we can strike a happy medium between huge chunks of data and searchability.

use cases

  • yum install something
    • yum needs to see conflicts/obsoletes, pkginfo and the pkg specific requires ... at a minimum.
    • If it requires something not known then all "provides" will be needed.
    • and any files require file data.
  • yum update/downgrade/distro-sync
    • The same as "yum install"
  • yum list pkgglob
    • needs to see pkginfo
  • yum list updates
    • needs to see pkginfo
  • yum check-update
    • needs to see pkginfo + obsoletes
  • yum provides something
    • needs to see pkginfo + provides. And if the filename glob matches than ALL of the filelists
  • yum search
    • pkginfo + localized summary+description data, (and currently) all provides.

depsolving will need, filelists per-pkg or a subset of file-reqs files.


filelists - break them up

filelists broken out by paths so you don't have to download a huge glop of the complete filelists just to find out who owns /foo/baz

possible ways to break up the files - info from rawhide as of june 2010:

2.4 million files total in pkgs in rawhide
2.3 million of those are in /usr
1.8 million of those are /usr/share
  Top 3 dirs by file count under /usr/share:
   533046 /usr/share/doc
   120555 /usr/share/javadoc
   105591 /usr/share/icons
45 file-requires requiring something in /usr/share
none of those file-requires are in the top 3 /usr/share dirs
    - most of them are fonts.

so in general we're downloading 75% more files for a filereq check than we'll EVER need.

So if we break the files up by 'top level dir' + 3 layers deep for /usr/*/*/ then we'll have reduced the number unnecessary file-list downloads by about 75%

complete repodata per-pkg

  • in a file or a directory of files so I can grab all of a certain kind of metadata for ONLY one pkg


- provide a way to break out summary/description into a structure that supports translations.

potential structure

And some more specific ideas:


repomd.xml <-- same as before - the index for everything else - but making sure not to use any of the existing data type attributes

packagelist.sqlite contains:

           name, arch, epoch-ver-rel, 
           size (package, installed), 
           location (baseurl, mirrorlist(optional), href), 
           header byte-ranges?(maybe), 
           location-style-path to per-pkg metadata and checksum.



provides.sqlite <-- provides: providename + flags + evr req_conf_obs.sqlite <-- putting requires,conflicts and obsoletes in on db - multiple tables b/c you need them all at the same time files_by_path.xml <-- index file to point to the files-by-path

path_it_holds + filename + checksum, per file

...note that it may be better to make "req_conf_obs" 3 files, so that some updates can reuse older versions (Eg. just requires changes).