From Fedora Project Wiki

Revision as of 18:42, 2 June 2014 by Bcl (talk | contribs) (Add sections on creating disk image and PXE booting)

Personal homepage

Packages

Misc

Anaconda Development

Development System

  • Fedora 18
  • polipo proxy to cache packages
  • lighttpd to serve up updates.img
  • tftpd to serve up pxe boot images
  • mock or lorax for building packages and iso's

mock setup

  • yum install mock
  • Edit the /etc/mock/site-defaults.cfg file to bind your working directories inside the mock root. This keeps mock from wiping those directories on --init
config_opts['plugin_conf']['bind_mount_enable'] = True
config_opts['plugin_conf']['bind_mount_opts']['dirs'].append(('/home/user/projs/anaconda', '/root/anaconda' ))
config_opts['plugin_conf']['bind_mount_opts']['dirs'].append(('/local/repo', '/repo' ))
config_opts['plugin_conf']['bind_mount_opts']['dirs'].append(('/local/pungi', '/pungi' ))
  • Edit the /etc/mock/ file for the distribution or copy it to a new name and edit
    • Add proxy=http://proxy.home:3128 to the main section.
    • comment out mirrorlist entries
    • Open up the mirror list url(s) in a browser and pick a mirror, use the same one for base and updates
    • Add baseurl=<mirror url> in each section instead of mirrorlist
    • Do this for each section with a mirrorlist
  • Setup the initial mock environment, replace the fedora-13-i386 with the mock file edited above
    • mock -r fedora-13-i386 --init
    • mock -r fedora-13-i386 --no-clean --install pungi

Now you have a mock chroot environment setup that will use the proxy cache for packages.

Anaconda Debugging

pdb

Insert this into Anaconda at the point where you want to start debugging:

  • import pdb; pdb.set_trace()

When the breakpoint is hit you will be able to use pdb on tty1.

winpdb

http://winpdb.org is a remote python debugger with support for threads.

Insert this into Anaconda at the point where you want to start debugging:

  • import rpdb2; rpdb2.start_embedded_debugger('snakes', fAllowRemote=True)

Create an update image with the rpdb2.py file:

  • ./scripts/makeupdates -k
  • cp /usr/lib/python2.7/site-packages/rpdb2.py ./updates/run/install/updates/
  • ./scripts/makeupdates -k

Boot your iso with the updates=http://url/of/your/updates.img and change to tty2. Run ip addr to find the IP of the system/virt.

When Anaconda hits the breakpoint run winpdb on your local system and use File->Attach to connect to the system and start debugging. The password (snakes) is set by the start_embedded_debugger call.

Anaconda's storage imports appear to interfere with rpdb2's import wrapper so don't expect to be able to step through the whole program, but it is very useful for examining the state of the system and looking at the local variables for all the running threads.

Building boot images

with lorax

From the top level of a checked out lorax git repo I run:

PATH=./src/sbin/:$PATH PYTHONPATH=./src/ ./src/sbin/lorax -p Fedora -v 17 -r 17 \
-s http://download.fedora.redhat.com/pub/fedora/linux/development/rawhide/x86_64/os/ \
--proxy http://proxy.brianlane.com:3128 --config ./lorax.conf ./tmp/

The -s points to repositories to use, add your own with higher nvr to override packages in the primary repo.

lorax.conf points to the template directory:

[lorax]
sharedir=/home/bcl/Red_Hat/projs/lorax/share

results will be in ./tmp/

with pungi (old)

Use the compose and pungi.ks files below, place them into /root/ in the mock to be built. Replace REPO_URL_HERE with the same repo you selected for the mock's yum.conf file.

  • pungi.ks
# un-comment the applicable repo

repo --name=fedora --baseurl=REPO_URL_HERE --proxy=http://proxy.home:3128

# Very small install footprint
%packages
@base
kernel
syslinux
nomtools
anaconda
%end
  • compose
#!/bin/bash
echo "*** Running pungi --force --nosource --nodebuginfo --nosplitmedia -G -C -B -c /root/pungi.ks --ver=13"

pungi --cachedir=/pungi/cache --force --nosource --nodebuginfo --nosplitmedia -G -C -B -c /root/pungi.ks --ver=13
echo "*** Done."
  • chmod a+x compose
  • exit
  • mock -r fedora-13-i386 --chroot "/root/compose"

Change the --ver to match whichever release you are building for

Building updates

Anaconda includes the ability to update itself by passing updates=http://path/to/update.img to the kernel at boot time. This allows you to use the same boot media and test changes to stage2 of the installer.

  • This assumes a working mock chroot (ie. I built boot images with this one first)
  • You need a git repo of Anaconda, I branch for the build so I don't clutter up master with build by-products.
  • bind the anaconda directory to the mock being used by editing /etc/mock/site-defaults.cfg as described above.
    • NOTE: DO NOT use mount -o bind to mount directories inside the mock. They will be obliterated the next time --init is run.
  • Install the development tools group
    • mock -r fedora-13-i386 --install @development-tools
    • mock -r fedora-13-i386 --install yum-utils
  • enter the chroot
    • mock -v -r fedora-13-i386 --shell
    • grab a copy of the current anaconda*src.rpm from someplace and install the dependencies with:
      • yum-builddep anaconda*src.rpm
    • now build anaconda
      • ./autogen.sh
      • ./configure
      • make updates
    • exit
  • copy the updates.img over to the updates directory and optionally rename it
    • cp /var/lib/mock/fedora-13-i386/root/root/anaconda/updates.img /home/tftpboot/images/updates/

You can skip installing development tools and anaconda deps for subsequent iterations.

Now add updates=http://proxy.home/updates/updates.img to the kernel parameters when booting the install media.

NOTE: By running yum inside the chroot you may mess up the rpmdb version, this depends on how close your host system is to the target system. So YMMV

Build with a test anaconda.rpm

NOTE: This doesn't actually seem to work when the released anaconda version matches that in the branch. buildinstall isn't letting the local repo override the one in the repo

Normally when a punji build is done it pulls anaconda from the repo/proxy cache. Instead you want it to use your new build (ie. when doing stage1 development which cannot be updated by updates= being passed to the kernel)

  • make sure mock is mounting your anaconda directory as described above. Again, do not use mount -o bind to do it.
  • mock -v -r fedora-13-i386-proxy --shell
  • cd /root/anaconda
  • Removed the cached files. Otherwise it pull pull from there and not update to the latest
    • rm -rf /13
    • rm -rf /pungi/cache/local
  • ./autogen.sh
  • ./configure
  • make scratch
  • rpmbuild -ts --nodeps anaconda...tar.gz
  • yum-builddep /builddir/build/SRPMS/anaconda...src.rpm
  • rpmbuild --rebuild /builddir/build/SRPMS/anaconda...src.rpm
  • mkdir /repo
  • cp /builddir/build/RPMS/anaconda...rpm /repo
  • createrepo /repo
  • edit /root/pungi.ks and add
    • repo --name=local --baseurl=file:///repo/
  • exit
  • mock -v -r fedora-13-i386 --chroot "/root/compose"
  • rsync -avc /var/lib/mock/fedora-13-i386/root/13/i386/os/images/ /home/tftpboot/images/fedora/13/i386/

Boot with the pxe image and see what happens.

NOTE: The removal of the local repo cache is needed because yum gets confused by its presence.

Update boot.iso with new anaconda rpm

After you have a working boot.iso you can easily update it with the files from the new anaconda rpm you built above. Use the [upd_bootiso] script to do this:

upd_bootiso boot.iso anaconda-13.35-1.fc13.i686.rpm

This will extract the files from initrd.img and install.img on the boot.iso, update the files and then re-build the boot.iso

This is considerably faster than using mock + pungi to compose a whole new iso from scratch.

Note, this script currently only works with x86 due to the fact that I borrowed part of it from the mk-images.x86 script and haven't built on other architectures yet.

Releasing anaconda

This requires permission to upload new anaconda files and a Transifex account with access to the Anaconda project. The transifex client needs to be setup on your system. This is described in the doc/transifex.txt file in the anaconda source tree:

  • git clean -d -x -f
  • ./autogen.sh && ./configure
    • If that fails for deps: yum-builddep anaconda.src.rpm
    • Alternatively grab the srpm using fedpkg srpm and run yum-builddep on that.
  • make bumpver
  • Make sure anaconda.spec.in looks ok
  • git commit -m "New version."
  • make release
  • git push
  • git push --tags

In the fedora package then do:

  • fedpkg switch-branch fXX
  • Copy the anaconda.spec and anaconda-*tar.gz over to the package directory
  • Make sure anaconda.spec looks ok
  • fedpkg new-sources anaconda-XX.XX-X.tar.gz
  • fedpkg clog
  • Edit the clog to fix the first line, make it consistent with the "- [text]" changelog entries.
  • git add anaconda.spec
  • git commit -F clog
  • git push
  • fedpkg build
  • fedpkg update

Creating a disk image

livemedia-creator makes it easy to create a disk image for use with virtual systems. It can also be PXE booted on a diskless system.

  • sudo livemedia-creator --make-fsimage --iso=/path/to/boot.iso --title Fedora --project Fedora --releasever 20 --vnc vnc --ks=/usr/share/doc/lorax/fedora-livemedia.ks |& tee lmc-build-out.log

PXE boot disk image

If you have enough memory (size of the initrd + normal system requirements) you can PXE boot a fsimage. The fsimage is compressed and appended to the end of the existing initrd for the release you want to boot. Create the new initrd using the img2pxe.sh script.

  • img2pxe.sh /path/to/fedora/20/initrd.img /path/to/fsimage.img /path/to/initrd-disk.img

Edit your PXE setup and create a menu entry that has an APPEND line like this:

  • APPEND initrd=/path/to/initrd-disk.img root=live:/squashfs.img rd.live.image