From Fedora Project Wiki

< Infrastructure‎ | PackageDatabase

Revision as of 16:32, 24 May 2008 by Ravidiip (talk | contribs) (1 revision(s))

The printable version is no longer supported and may have rendering errors. Please update your browser bookmarks and please use the default browser print function instead.
#!rst (-)
==========
Next Steps
==========

:Author: Toshio Kuratomi
:Contact: toshio-tiki-lounge-com
:Date: Sunday, 21 January, 2007

.. contents::

Import Data
===========

Currently, information about the packages comes from several sources.  We
need to import data from those sources into the packageDB.

owners.list
-----------

:Location: CVSROOT=:pserver:anonymous@cvs.fedora.redhat.com:/cvs/extras cvs co owners; cat owners/owners.list
:Data: Package names, summary, owner, qacontact, watchers
:Duration: One-time; Obsoletes
:Priority: Essential
:Status: Information is largely imported.  Need to perform a similar step for
owners.epel.list.  A few people in initialcc do not have Fedora accounts.

owners.list contains the bulk of new information that we are interested in.
This file should contain information on all packages whether they are current
in the repository or not.

This importer is almost complete.  A bit more work needs to be done to combine
the information here with information from the CVS Repository.  With a
combination of these we should have all the basic information necessary for
the first stage of the package DB.

We are not currently retrieving any information for RHL releases and do not
plan to.

CVS Repository
--------------

:Location: CVSROOT:pserver:anonymous@cvs.fedora.redhat.com:/cvs/extras
:Data: Collections that the package is present in, package metadata present in the spec file (URL, upstream source, etc.)
:Duration: One-time; SyncReversal
:Priority: Essential

The VCS repository (currently CVS) is the source of most information pertaining
to the package itself.  Most of this information is better taken from the
built packages, though, as the metadata has already been parsed and pulled into
the repomd files by that time.

Unique to the CVS repository is knowledge of what packages are present in which
collections -- FC-{1,2,3,4,5,devel} and FE-{1,2,3,4,5,devel}. RH-{9}, and soon
EL-{4,5}.  Regardless of whether they were obsoleted or otherwise, disappeared
in the course of the release.

The cvs repository is also gaining ACL knowledge shortly.  We could pull this
information either from the files within the cvs tree or from the CVSROOT/avail
file.

In the future, the packageDB will manage the branches in the VCS Repository.

Package Reviews (mailing list)
------------------------------

:Location: APPROVED_trim.txt REQUESTS_trim.txt
:Data: URL, dates, and authors of package reviews
:Duration: One-time
:Priority: High

In the early days, packages were approved via the fedora-extras mailing list.
These approval messages have been abstracted to two files.  We need to enter
this information into the packagedb so people know which packages need to be
reviewed before being built for the combined Core + Extras.

The URL goes into the package table.  The date and author go into the
PackageLog.

Package Reviews (bugzilla)
--------------------------

:Location: bugzilla
:Data: URL, dates, and authors of package reviews
:Duration: Ongoing
:Priority: High

Currently, packages are reviewed in bugzilla.  We need to pull information
about packages which were formerly reviewed into the packagedb.  Then we need
to set up a system that continues to sync against it.  Perhaps we could have
a packagedb front end that opens up a bugzilla report for a new package in
order to start the process.  The packagedb can periodically scan FE-REVIEW and
FE-APPROVE tracker bugs to tell when a package changes state.

The URL goes into the package table.  The date and author go into the
PackageLog.

comps-fcX.xml.in
----------------

:Location: CVSROOT=:pserver:anonymous@cvs.fedora.redhat.com:/cvs/extras cvs co comps; cat comps/comps-fe6.xml.in
:Data: Groups that the package belongs to.
:Duration: One-time; SyncReversal
:Priority: Add-on

There is one comps.xml.in file per release.  It is used to create the comps.xml
file that lists what groups packages in the repository are for.

In the future, comps-fcX.xml.in could be generated from the packageDB.

Download Repository
-------------------

:Location: http://download.fedora.redhat.com/pub/fedora/linux/extras/$release/repodata/repomd.xml
:Data: RPM Metadata extracted into an xml file.  License, URL, upstream, present package EVR.
:Duration: Ongoing
:Priority: Add-on

Parsing the Download repository information gives us EVR information which is
needed for the package_version table.  There is a wealth of other information
in the files that we could use as well (file listings, summary, etc.) however,
this is a large extension from the initial stage of the packagedb.

We probably need to cache the repomd files in order to extract information
sensibly as we're going to be pulling the data from the files on an ongoing
basis.  Having the cached files allows us to find the differences between the
old and new information.

Build Server
------------

:Location: http://buildsys.fedoraproject.org/logs/fedora-development-extras/19621-gnubiff-2.2.2-3.fc6/x86_64/build.log
:Data: Logs of failed and completed builds
:Duration: Ongoing
:Priority: Add-on

What do we want to do with these?  They are a large source of raw information
but there's not much in terms of metadata type data.  If we have the db
horsepower, we could put them into a Full Text Searchable field and put them
in the db so we can search on them.

Write Web Front End
===================

The Web front end is the primary interface to the package db.

Status
------

* A Turbo Gears skeleton has been constructed.
* A database schema has been created that works with the
postgreSQL-SQLAlchemy-TurboGears combination.
* Configured apache to pass requests on to the TurboGears instance
* Wrote a CGI to autostart the TurboGears CGI when a request is made
* Wrote a TurboGears identity layer to validate users against the Fedora
Account System.

TODO
----

Version 0.1
~~~~~~~~~~~

* See the two "essentials" from Importing Data.
(owners.list is finished.
(ACLs are not ready yet.)
(EPEL is being put off.  It requires information from cvs and
owners.epel.list).
* Finish filling in read-only pages that allow us to see what's in the database
for each package.

Version 0.2
~~~~~~~~~~~

* Add the ability for authenticated users to change owners.list information:

- Add packages.
- Set themselves as owner for a package.
- Set a package as orphaned.
- Request to watch a package.

Before F7
~~~~~~~~~

* Add the ability for owners to hand out permissions to other users.
* Allow owners and trusted users to make changes to the packages listed here.
* Send email notifications when things change.
* Write a backend for the cvs repository to pull ACLs from the packageDB.
* Integrate with the accounts system.

As time permits
~~~~~~~~~~~~~~~

* Make ACLs a push technology instead of pull.  The first generation of code
will have a cron job that pulls ACLs from the packagedb to the CVS acls file.
The next generation should have the packageDB set the ACLs in CVS instead.

* Add comps group information to the database schema so we can generate comps
from here.

* Write admin functionality:

- Branch script to make new release branches in the VCS.  Owners can
immediately branch for certain releases.  For other releases, owner
requests and admin clicks one button to confirm the request.  Script
takes care of the actual branch.

* RSS Feeds of package changes.

* Second level notification: Tell package owners when a package they depend
on is updated.

Ideas
~~~~~

* Hidden branches (for security fixes) need to be integrated somehow.  We may
want to include them in this interface but they are a bit more problematic
due to the way they operate.
* Maybe take more information from repomd and subsume some of repoview's
functionality?  This would take us into the realm of what Debian's web
interface to their package browser does but it may be too many things for
this project.