View source

===Why Server CD?===
Lately I thought of installing a XEN CENTOS 5.5 virtual machine on one of my servers out there. I needed the exploded tree of the ISO to be available. However, when I checked centos.org website, I noticed that they stopped distributing a "Single Server CD" after CENTOS 4. I could have downloaded a CENTOS 5.5 DVD image on the server but it was kind of un-desired / not practical for two reasons.

* The centos.org distributes DVD iso image through bit-torrent, and my server service provider does not allow P2P or torrent traffic. 
* Even if I had one, I would have to download one big 4-5 GB file to my server. Whereas a few hundred MBs would do, if I want a text based "core" installation only.

So there arose an itch, and I decided to create a project to scratch that itch! 

===Objectives / Project goals===
* To create a repository / collection of files, a so called "Server CD" install-tree, small enough to provide the "core" installation, plus yum and wget.
* To be able to create a boot-able /installable "install CD" out of the install-tree mentioned above. So that, when I (or anyone else) need to install a minimum OS on a bare-metal machine, he doesn't have to go through downloading a full DVD. 

===Creating the "core" install-tree, the hard way===
The idea is simple. I have a CENTOS 5.4 DVD already downloaded at home (and office). To get the list of packages, I did a "core" install of CENTOS on a XEN virtual machine. (I used a kickstart file with @core mentioned in the "%packages" section. Analysing its "/root/install.log", I quickly figured out which packages got installed. (clever me!). So this is the list of packages I need from the "CentOS" directory of the install-tree. (It's about 443 packages). This resulted in an installation of about 950MB in size. 

Here is how I used (install.log) to generate the list of required pacakges in my repository.

<pre>
[root@xenlab minivm]# grep "Installing" install.log | awk '{print $2}'|sort > installed-rpms.tmp
</pre>

Then I edited the resulting file to remove strange numbers before the package names, such as:

<pre>
basesystem-8.0-5.1.1.el5.centos.noarch
cracklib-dicts-2.8.9-3.3.x86_64
1:termcap-5.5-1.20060701.1.noarch          <<<<------- Such as this one. What is this number ?
nash-5.1.19.6-54.x86_64
</pre>


This is good enough. However, I am too much irritated by seeing bluetooth, cups, x11, and similar packages being part of the core installation. So I went one step further, and started deleting packages using yum, in a polite fashion, from this VM. This way, I was able to delete some packages, which I consider un-necessary for an installation, where only a web-server is the only desired service to run, without any backend database. (Which was the case with me.). After this surgery, I was able to bring down the disk usage to about 750 MB.  And of-course, fewer number of packages. (About 118 less rpms!). I can use that in the %post to delete the un-necessary RPMs!

<pre>
# yum remove aspell aspell-en atk authconfig autofs avahi avahi-compat-libdns_sd \ 
  bitstream-vera-fonts bluez-gnome bluez-libs bluez-utils cairo ccid coolkey \ 
  cpuspeed crash cups cups-libs desktop-file-utils dhcpv6-client dnsmasq \ 
  dos2unix dosfstools ecryptfs-utilsed eject fbset finger firstboot-tui \ 
  fontconfig freetype GConf2 gpm gtk2 hicolor-icon-theme htmlview ifd-egate \ 
  iptables-ipv6 irda-utils irqbalance jwhois krb5-workstation ksh libdrm  \
  libICE libjpeg libnotify libpng libSM libtiff libwnck libX11 libXaulibXcursor \ 
  libXdmcp libXext libXfixes libXft libXi libXinerama libXrandr libXrender \ 
  libXres libXt libXxf86vm mailcap man man-pages mdadm mesa-libGL microcode_ctl \
  mkbootdisk mtools nano NetworkManager NetworkManager-glib newt \
  notification-daemon ntsysv numactl ORBit2 pam_ccreds pam_krb5 pam_pkcs11 \ 
  pam_smb pango paps pcmciautils pcsc-lite pcsc-lite-libs pinfo procmail rdate \ 
  redhat-lsb redhat-menus rhpl rp-pppoe rsh sendmail setuptool slang sos \ 
  specspo startup-notification syslinux system-config-network-tui \
  system-config-securitylevel-tui tcpdump trousers unix2dos vconfig \ 
  wireless-tools words wpa_supplicant xorg-x11-filesystem ypbind \ 
  yp-tools yum-updatesd
</pre>

Tip: I used the following commands to generate the long yum command above:

<pre>
# grep Erased /var/log/yum.log | awk '{ print $5}' | sort | uniq | tr '\n' ' '
</pre>


To create my minimal repository, I first created a copy of the original install tree  skipping the CentOS directory. 

<pre>
[root@xenlab ~]# rsync -av --exclude CentOS/  \ 
  /var/www/html/CentOS-5.4-x86_64/  /var/www/html/CentOS-5.4-x86_64-core/
</pre>

Then I used the list of packages from the install.log file, and pulled the matching RPM files from the full install-tree into this "core" install-tree. I used the "installed-rpms.txt" file, I created a moment ago, above. The installed-rpms.txt file looks like this:

<pre>
[root@xenlab ~]# cat /tmp/minivm/installed-rpms.tmp | tail
which-2.16-7.x86_64
words-3.0-9.1.noarch
xorg-x11-filesystem-7.1-2.fc6.noarch
yp-tools-2.9-0.1.x86_64
yum-3.2.22-20.el5.centos.noarch
yum-fastestmirror-1.1.16-13.el5.centos.noarch
yum-metadata-parser-1.1.2-3.el5.centos.x86_64
zip-2.31-2.el5.x86_64
zlib-1.2.3-3.i386
zlib-1.2.3-3.x86_64
[root@xenlab ~]#
</pre>

Create the CentOS directory in the "core" install tree:
<pre>
[root@xenlab ~]# mkdir /var/www/html/CentOS-5.4-x86_64-core/CentOS
</pre>

Then copy the selected files from the full install-tree into this "core" tree, using the installed-rpms.txt file.

<pre>


[root@xenlab ~]# for file in $(cat /tmp/minivm/installed-rpms.tmp); do 
  cp -v /var/www/html/CentOS-5.4-x86_64/CentOS/"${file}.rpm" \ 
    /var/www/html/CentOS-5.4-x86_64-core/CentOS-5.4-x86_64-core/CentOS/ 
done 
</pre> 

Check the "core" CentOS directory:

<pre>
[root@xenlab ~]# cd /var/www/html/CentOS-5.4-x86_64-core/CentOS/

[root@xenlab CentOS]# ls -l | wc -l
443

[root@xenlab CentOS]# du -sh .
273M    .
[root@xenlab CentOS]# 
</pre>

Lovely! As you can see only 443 RPMs and 273 MB occupied by the RPMs only!

Note: We need to (additionally) copy the normal kernel to the new location too, because someone might want to use this for installing a non-XEN machine, or for the bare-metal machine. In that case a normal kernel would be required.

<pre>
[root@xenlab ~]# cp /var/www/html/CentOS-5.4-x86_64/CentOS/kernel-2.6.18-164.el5.x86_64.rpm \ 
  /var/www/html/CentOS-5.4-x86_64-core/CentOS/ 
[root@xenlab ~]#  
</pre>

Notice the difference now:

<pre>
[root@xenlab CentOS]# cd /var/www/html/CentOS-5.4-x86_64-core/CentOS/

[root@xenlab CentOS]# ls | wc -l
444

[root@xenlab CentOS]# du -sh .
291M    .
</pre>

The complete usage of the minimal install tree is :-
<pre>
[root@xenlab ~]# du -sh /var/www/html/CentOS-5.4-x86_64-core
606M    /var/www/html/CentOS-5.4-x86_64-core
[root@xenlab ~]#  
</pre>


Note: In all the steps above, we DID NOT update/change the repository (meta) data, (using createrepo command). Nor did we change anything in the installer XML files. It works just like that.


When I was done with doing so, I created another XEN VM, and used this new (shrunk) install-tree to install the new VM to make sure it works. And it did! Here is the newly installed system:-

<pre>
[root@localhost ~]# df -h
Filesystem            Size  Used Avail Use% Mounted on
/dev/xvda1            2.7G  986M  1.6G  39% /
tmpfs                 256M     0  256M   0% /dev/shm

[root@localhost ~]# rpm -qa | wc -l
443
[root@localhost ~]#
</pre>

Then I removed the 118 packages, which I found un-necessary (as shown in the yum remove command above). The new space usage of this VM is now: 

<pre>
[root@localhost ~]# df -h
Filesystem            Size  Used Avail Use% Mounted on
/dev/xvda1            2.7G  775M  1.8G  31% /
tmpfs                 256M     0  256M   0% /dev/shm
[root@localhost ~]#
</pre>




Now, to use this install tree on my server on the internet, I can do one of the following:

* Push the repository tarball to my XEN physical host/server on internet,
* Or, make a script to "wget" all the required packages from a CENTOS mirror, create a repo for the installer, and make that available through yum on that server.

The first method is kind of immediately do-able, but not "exciting". It definitely involves copying all the install tree to the XEN host up there. To do the second method, I created a kind of script, which can pull the packages from any install tree on the internet and populate the CentOS directory on my own install-tree. Good enough, but still not geeky enough.

What I want to do now, is to be able to parse comps.xml file, present in the repodata directory of any regular install tree, find out the packages needed for "@core" and then pull those packages from the regular install tree into my own install tree. (This would be geeky/cool.)

And when I can do that, I can then create an actual bootable ISO out of it using help from many posts on the internet  (mentioned at the end of this document).

===Creating the "core" install-tree from a http site===
Here, I am using wget to get the basic structure of CentOS full repository to my local machine, "without" pulling any rpms. 

Notes:
* In the example below, /tmp/centos-mini is the location where I am building the "core" install-tree. 
* 192.168.0.1 is a web server, with complete DVD install-tree exported through web service.
* Interestingly CentOS directory is the "only" directory in any install tree, which contains RPM files!

<pre>
This:

[root@xenlab ~]# wget --mirror  http://192.168.0.1/CentOS-5.4-x86_64/ \ 
      --directory-prefix=/tmp/centos-mini/  --no-parent \ 
      --no-host-directories --cut-dirs=1 --reject="*.rpm" \ 
      -o /tmp/wget-log

Results in:-

Saving to: `/tmp/centos-mini/CentOS/index.html?C=M;O=A'
Saving to: `/tmp/centos-mini/CentOS/index.html?C=S;O=A'
Saving to: `/tmp/centos-mini/CentOS/index.html?C=D;O=A'
Saving to: `/tmp/centos-mini/CentOS/TRANS.TBL'
Saving to: `/tmp/centos-mini/NOTES/index.html?C=N;O=D'
Saving to: `/tmp/centos-mini/NOTES/index.html?C=M;O=A'
Saving to: `/tmp/centos-mini/NOTES/index.html?C=S;O=A'
Saving to: `/tmp/centos-mini/NOTES/index.html?C=D;O=A'
Saving to: `/tmp/centos-mini/NOTES/RELEASE-NOTES-U1-as.html'
Saving to: `/tmp/centos-mini/NOTES/RELEASE-NOTES-U1-bn.html'
</pre>


Good!


I also see a lot of following. Need to investigate why wget is creating these index.html files, which does not exist on the host system / web server in the first place! (This is of very low priority though):

<pre>
Saving to: `/tmp/centos-mini/192.168.0.1/CentOS-5.4-x86_64/images/xen/index.html?C=S;O=D'
Saving to: `/tmp/centos-mini/192.168.0.1/CentOS-5.4-x86_64/images/xen/index.html?C=D;O=D'
</pre>


Now we need to pull the required RPM files and store them manually in the CentOS directory.  

<pre>
for file in $(cat /tmp/minivm/installed-rpms.tmp); do 
  wget  http://192.168.0.1/CentOS-5.4-x86_64/CentOS/${file}.rpm  \
  --directory-prefix=/tmp/centos-mini/CentOS/ --no-parent  \ 
  --no-host-directories --cut-dirs=1  -o /tmp/wget-log
done  
</pre>

Make sure you have both kernel and kernel-xen packages in this repository. Since I already have kernel-xen, I will pull kernel-2.6.18-164.el5.x86_64.rpm :

<pre>
wget  http://192.168.0.1/CentOS-5.4-x86_64/CentOS/kernel-2.6.18-164.el5.x86_64.rpm  \
  --directory-prefix=/tmp/centos-mini/CentOS/ --no-parent  \ 
  --no-host-directories --cut-dirs=1  -o /tmp/wget-log
</pre>

===Create a tarball of your mini install-tree for easier distribution===
<pre>
[root@www cdimages]# tar cjf CentOS-5.4-x86_64.tar.bz2 CentOS-5.4-x86_64

[root@www cdimages]# du -sh *
537M    CentOS-5.4-x86_64
493M    CentOS-5.4-x86_64.tar.bz2
[root@www cdimages]# 
</pre>

===ToDo===
I will make this available in few days on http://downloads.wbitt.com. (God Willing). Please note that at the moment, this is not a bootable CD image. It is a fully functional install tree, which would require you to use a kickstart with only three entries in the %packages section.

<pre>
@core
wget
yum
</pre>

Modifying the comps.xml and updating the repository data, so far is very exciting and geeky, BUT very painful procedure.

This should be usable. I will install a VM in a couple of days and confirm that here.

===Packages which can be removed from @core===
Here is the list of packages which I found in CENTOS-5.4-x86_64, which I was successfully able to remove, using yum, without any damage to the OS. (about 118 packages)

<pre>
aspell
aspell-en
atk
authconfig
autofs
avahi
avahi-compat-libdns_sd
bitstream-vera-fonts
bluez-gnome
bluez-libs
bluez-utils
cairo
ccid
coolkey
cpuspeed
crash
cups
cups-libs
desktop-file-utils
dhcpv6-client
dnsmasq
dos2unix
dosfstools
ecryptfs-utils
ed
eject
fbset
finger
firstboot-tui
fontconfig
freetype
GConf2
gpm
gtk2
hicolor-icon-theme
htmlview
ifd-egate
iptables-ipv6
irda-utils
irqbalance
jwhois
krb5-workstation
ksh
libdrm
libICE
libjpeg
libnotify
libpng
libSM
libtiff
libwnck
libX11
libXau
libXcursor
libXdmcp
libXext
libXfixes
libXft
libXi
libXinerama
libXrandr
libXrender
libXres
libXt
libXxf86vm
mailcap
man
man-pages
mdadm
mesa-libGL
microcode_ctl
mkbootdisk
mtools
nano
NetworkManager
NetworkManager-glib
newt
notification-daemon
ntsysv
numactl
ORBit2
pam_ccreds
pam_krb5
pam_pkcs11
pam_smb
pango
paps
pcmciautils
pcsc-lite
pcsc-lite-libs
pinfo
procmail
rdate
redhat-lsb
redhat-menus
rhpl
rp-pppoe
rsh
sendmail
setuptool
slang
sos
specspo
startup-notification
syslinux
system-config-network-tui
system-config-securitylevel-tui
tcpdump
trousers
unix2dos
vconfig
wireless-tools
words
wpa_supplicant
xorg-x11-filesystem
ypbind
yp-tools
yum-updatesd
</pre>

===References/Further Reading===
Some smart people have already done the bootable CD / custom CD part. Below are some links for reference:

* http://www.sirius.pe.kr/v5/tt
* http://smorgasbork.com/linux/35-linux/58-building-a-custom-centos-5-kickstart-disc-part-1
* http://256.com/gray/docs/redhat_boot/
* http://www.linuxjournal.com/article/6473
* http://people.redhat.com/rkeech/custom-distro.txt


* http://www-oss.fnal.gov/projects/fermilinux/common/comps.xml.html
* http://lateral.netmanagers.com.ar/stories/38.html
* http://penguin.triumf.ca/comps/9/comps-9.php
* http://www.linuxquestions.org/questions/red-hat-31/minimal-centos-via-kickstart-not-working-585555/
* http://www.linuxjournal.com/article/6473
* http://www.esrf.eu/Infrastructure/Computing/ControlComputers/ESRFLinux/InstSrv
* https://fedoraproject.org/wiki/Infrastructure/PackageDatabase/RoadMap
* http://tldp.org/HOWTO/RedHat-CD-HOWTO/include-updates.html
* http://www-oss.fnal.gov/projects/fermilinux/common/anaconda.comps.html
* http://fedoraproject.org/wiki/PackageMaintainers/CompsXml
* http://www.city-fan.org/tips/YumRepoFromImages
* http://www.pinkjuice.com/howto/vimxml/setup.xml#xmledit