[opensuse-autoinstall] Why getting different setups on identical HW although "CLONING" using autoyast ??
Hi, I'm trying to create an exact clone of a machine running openSuSE 10.1. Highest priority is to end up with a pair of machines that have EXACTLY the SAME SOFTWARE PACKAGES installed. These machines are build exactly of the same hardware, shipped from the same vendor on the same day, so there should'nt be a problem (HP Pavilion AMD 64 X2 Dual Core).
1. I did a fresh install on the first machine; and directly after that (without changing anything on that machine)... 2. i created a xml-profile using "yast2 autoyast", choosed "create reference profile" (and did NOT choose or change anything else manually...) and copied this file to a webserver running on a third machine. 3. I started a network installation on the second machine, telling it to use that profile (boot option "autoyast=http://my_ip/my_profile.xml") and to use the same installation source as was used for the first machine (that was a SuSE 10.1 DVD lying around on a samba share of a fourth machine).
I expect this to end up with exactly the same software installed on both machines (is this right or do i misunderstand something??); and besides that anything else also be the same (partitioning and so on...).
Unfortunately this is not the case. I checked the output of "rpm -qa" of both machines with a perl script i wrote and it reported the following differences:
"This exists in machine2.txt only: libgcj-32bit-4.1.0-25 This exists in machine2.txt only: java-1_4_2-gcj-compat-32bit-1.4.2.0-33"
I doublechecked this manually; it's true. On the first machine, these packages are not installed, however packages of the same name but without the "32bit" in it are installed (but these are ALSO installed on the second machine).
Question 1: Can someone explain me why this is happening? Is there a mistake in my approach?
Besides that i got the same problem trying to do the same on some production servers, which are running since a long time. However in those cases the difference is even bigger, there are around 80 packages reported to be installed on just the second machine, which i expected to be an exact clone (also identical hardware and so on).
My assumption in this case is that some packages are being installed on the "clone", because an "addon" section like "KDE" is in the profile.xml; while some packages of the "kde" selection were separately removed from the "original" machine after installation (without removing the whole selection "kde"). Could this be the cause?
Question 2: Does "yast2 autoyast / create reference profile" really represent the actual status of that machine concerning installed software and anything else? Is there a better way to reach my goal (two machines with exactly the same software installed; exact clone); how can i create a working xml-profile that ensures this; even if there were lots of software installations and deinstallations on the original machine after the first "initial setup"?
Thanks in advance..
Elmar Marschke wrote:
Hi, I'm trying to create an exact clone of a machine running openSuSE 10.1. Highest priority is to end up with a pair of machines that have EXACTLY the SAME SOFTWARE PACKAGES installed. These machines are build exactly of the same hardware, shipped from the same vendor on the same day, so there should'nt be a problem (HP Pavilion AMD 64 X2 Dual Core).
- I did a fresh install on the first machine; and directly after that
(without changing anything on that machine)... 2. i created a xml-profile using "yast2 autoyast", choosed "create reference profile" (and did NOT choose or change anything else manually...) and copied this file to a webserver running on a third machine. 3. I started a network installation on the second machine, telling it to use that profile (boot option "autoyast=http://my_ip/my_profile.xml") and to use the same installation source as was used for the first machine (that was a SuSE 10.1 DVD lying around on a samba share of a fourth machine).
I expect this to end up with exactly the same software installed on both machines (is this right or do i misunderstand something??); and besides that anything else also be the same (partitioning and so on...).
Unfortunately this is not the case. I checked the output of "rpm -qa" of both machines with a perl script i wrote and it reported the following differences:
"This exists in machine2.txt only: libgcj-32bit-4.1.0-25 This exists in machine2.txt only: java-1_4_2-gcj-compat-32bit-1.4.2.0-33"
I doublechecked this manually; it's true. On the first machine, these packages are not installed, however packages of the same name but without the "32bit" in it are installed (but these are ALSO installed on the second machine).
Question 1: Can someone explain me why this is happening? Is there a mistake in my approach?
I can answer this part - I'd bet that the 1st machine is 32bit kernel, the 2nd 64bit. That's decided by the kernel and initrd supplied by your bootserver.
Besides that i got the same problem trying to do the same on some production servers, which are running since a long time. However in those cases the difference is even bigger, there are around 80 packages reported to be installed on just the second machine, which i expected to be an exact clone (also identical hardware and so on).
My assumption in this case is that some packages are being installed on the "clone", because an "addon" section like "KDE" is in the profile.xml; while some packages of the "kde" selection were separately removed from the "original" machine after installation (without removing the whole selection "kde"). Could this be the cause?
Question 2: Does "yast2 autoyast / create reference profile" really represent the actual status of that machine concerning installed software and anything else? Is there a better way to reach my goal (two machines with exactly the same software installed; exact clone); how can i create a working xml-profile that ensures this; even if there were lots of software installations and deinstallations on the original machine after the first "initial setup"?
Thanks in advance..
If you want to closely match both machines, better to install both using the same process (i.e. install both with Autoyast, using the same installation source and autoyast file).
As well, from experience, the "reference profile" needs to be cleaned up manually, to abstract things like network card MAC addresses, etc...
Yan
Am Freitag, den 08.06.2007, 19:53 +0200 schrieb Yan Fitterer:
If you want to closely match both machines, better to install both using the same process (i.e. install both with Autoyast, using the same installation source and autoyast file).
Thanks for your answer, Yan, but unfortunately my real goal is to "clone" some production servers which are running since a long time; i can't install these from scratch. As i wrote: "Besides that i got the same problem trying to do the same on some production servers, which are running since a long time. However in those cases the difference is even bigger, there are around 80 packages reported to be installed on just the second machine, which i expected to be an exact clone (also identical hardware and so on)."
So what would be of special interest to me is an answer to this assumption; as i wrote: "My assumption in this case is that some packages are being installed on the "clone", because an "addon" section like "KDE" is in the profile.xml; while some packages of the "kde" selection were separately removed from the "original" machine after installation (without removing the whole selection "kde"). Could this be the cause?"
Has anyone an idea or experience in that?
On Monday 11 June 2007 15:48, Elmar Marschke wrote:
"My assumption in this case is that some packages are being installed on the "clone", because an "addon" section like "KDE" is in the profile.xml; while some packages of the "kde" selection were separately removed from the "original" machine after installation (without removing the whole selection "kde"). Could this be the cause?"
yes, that can be the cause. From the "Changes between 10.2 and 10.2":
http://www.suse.com/~ug/autoyast_changes_10_1-10_2.html ... [func] clone the packages to remove ...
besides that, the "clone system" profile should never be used without looking at it.
Am Montag, den 11.06.2007, 16:20 +0200 schrieb Uwe Gansert:
yes, that can be the cause. From the "Changes between 10.2 and 10.2":
http://www.suse.com/~ug/autoyast_changes_10_1-10_2.html ... [func] clone the packages to remove
Thank you Uwe, now things are a little clearer to me ..
If you want to clone a running system, I autoyast isn't really your answer, IMHO. Take a look at mkcdrec or mondo rescue (neither of which are part of SLES, unfortunately).
You will then be able to get an _exact copy_ of your running system, not just a "similarly configured", which will be the best you can hope to achieve with autoyast.
Yan
On Mon, Jun 11, 2007 at 2:48 PM, in message
1181569684.5696.89.camel@localhost, Elmar Marschke elmar.marschke@schenker.at wrote:
Am Freitag, den 08.06.2007, 19:53 +0200 schrieb Yan Fitterer:
If you want to closely match both machines, better to install both using the same process (i.e. install both with Autoyast, using the same installation source and autoyast file).
Thanks for your answer, Yan, but unfortunately my real goal is to "clone" some production servers which are running since a long time; i can't install these from scratch. As i wrote: "Besides that i got the same problem trying to do the same on some production servers, which are running since a long time. However in those cases the difference is even bigger, there are around 80 packages reported to be installed on just the second machine, which i expected to be an exact clone (also identical hardware and so on)."
So what would be of special interest to me is an answer to this assumption; as i wrote: "My assumption in this case is that some packages are being installed on the "clone", because an "addon" section like "KDE" is in the profile.xml; while some packages of the "kde" selection were separately removed from the "original" machine after installation (without removing the whole selection "kde"). Could this be the cause?"
Has anyone an idea or experience in that?
Am Montag, den 11.06.2007, 17:02 +0200 schrieb Yan Fitterer:
If you want to clone a running system, I autoyast isn't really your answer, IMHO. Take a look at mkcdrec or mondo rescue (neither of which are part of SLES, unfortunately).
You will then be able to get an _exact copy_ of your running system, not just a "similarly configured", which will be the best you can hope to achieve with autoyast.
Looks very interesting, thank you.. didn't know it exists. Perhaps this really can solve my problem better in some cases; i will check it out..
Am Freitag, den 08.06.2007, 10:41 -0700 schrieb Steve Francis:
Unfortunately this is not the case. I checked the output of "rpm
-qa" of
both machines with a perl script i wrote and it reported the
following
differences:
"This exists in machine2.txt only: libgcj-32bit-4.1.0-25 This exists in machine2.txt only: java-1_4_2-gcj-compat-32bit-1.4.2.0-33"
I doublechecked this manually; it's true. On the first machine,
these
packages are not installed, however packages of the same name but without the "32bit" in it are installed (but these are ALSO
installed on
the second machine).
Question 1: Can someone explain me why this is happening? Is there a mistake in my approach?
I can answer this part - I'd bet that the 1st machine is 32bit kernel, the 2nd 64bit. That's decided by the kernel and initrd supplied by your bootserver.
Thanks for your answer, Steve. Do you mean that during the 1st stage of installation a different Kernel and initrd is used ? After the end of installation there are exactly the same Kernels installed ("uname -a" reports exactly the same on both machines)...
autoinstall@lists.opensuse.org
-
Elmar Marschke -
Steve Francis -
Uwe Gansert -
Yan Fitterer