http://fedoraproject.org/wiki/FedoraXenQuickstart & http://fedoraproject.org/wiki/FedoraXenQuickstartFC5
I'm just running through these instructions as a reference for my Xen3 tests on rawhide and all is going pretty well.
I have a test domain (but can't alter the number of VCPUs for some reason but that's another story) and I'm wondering. Rather than the pretty ugly system of either :
Make an image file. Mount. yum --installroot=blah groupinstall x y z Configure.
Or the much better system of /usr/sbin/xenguest-install.py,
Is there a simple way to fake a netboot so I can use my existing kickstart scripts and build OS instances the same way I would with normal physical hardware? It would be nice to have my hypervisor instance running dhcp + apache and do localhost kickstart network installs.
On Wed, 2006-02-08 at 12:15 +0900, Naoki wrote:
I'm just running through these instructions as a reference for my Xen3 tests on rawhide and all is going pretty well.
Good to hear
I have a test domain (but can't alter the number of VCPUs for some reason but that's another story) and I'm wondering. Rather than the pretty ugly system of either :
Yeah, vcpu hotplug seems to be a little busted right now. Haven't really dug to figure out why yet.
Or the much better system of /usr/sbin/xenguest-install.py,
Is there a simple way to fake a netboot so I can use my existing kickstart scripts and build OS instances the same way I would with normal physical hardware?
Realistically, the xenguest-install stuff is pretty much faking netboot. Run it with --help to see the way to pass the various things to it as command line arguments and you can use -x for passing things like ks=
Jeremy
Is there a simple way to fake a netboot so I can use my existing kickstart scripts and build OS instances the same way I would with normal physical hardware?
Realistically, the xenguest-install stuff is pretty much faking netboot. Run it with --help to see the way to pass the various things to it as command line arguments and you can use -x for passing things like ks=
As we say in my country, "Sweet mate!". I've tested with current rawhide and it's looking good except for the stage2 problem which I think has already been covered today.
Quick note about -l LOCATION, --location=LOCATION :
"http://host:/path" may be confusing because it will not work with the middle ":" unless a port is specified. Perhaps "http://host:port/path" is better?
The error I reported earlier actually occurs after bug 177644. Sorry about that.
One suggestion for xenguest-install.py. You might want to set the initial value of ram to 256. It is a little confusing when it prints out "ERROR: Installs currently require 256 megs of RAM." in interactive mode (running without any command line args) before your prompted with "How much RAM should be allocated (in megabytes)? ".
On Feb 7, 2006, at 9:30 PM, Naoki wrote:
Is there a simple way to fake a netboot so I can use my existing kickstart scripts and build OS instances the same way I would with normal physical hardware?
Realistically, the xenguest-install stuff is pretty much faking netboot. Run it with --help to see the way to pass the various things to it as command line arguments and you can use -x for passing things like ks=
As we say in my country, "Sweet mate!". I've tested with current rawhide and it's looking good except for the stage2 problem which I think has already been covered today.
Quick note about -l LOCATION, --location=LOCATION :
"http://host:/path" may be confusing because it will not work with the middle ":" unless a port is specified. Perhaps "http://host:port/ path" is better?
-- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
On Tue, 2006-02-07 at 22:15 -0800, Adam Huda wrote:
One suggestion for xenguest-install.py. You might want to set the initial value of ram to 256. It is a little confusing when it prints out "ERROR: Installs currently require 256 megs of RAM." in interactive mode (running without any command line args) before your prompted with "How much RAM should be allocated (in megabytes)? ".
This should actually be handled better as of today's xen package
Jeremy
Looks good now.
Found this problem though:
What is the name of your virtual machine? xenguest How much RAM should be allocated (in megabytes)? 256 What would you like to use as the disk (path)? /home/ahuda/xenguest How large would you like the disk to be (in gigabytes)? 4
Traceback (most recent call last): File "/usr/sbin/xenguest-install.py", line 357, in ? main() File "/usr/sbin/xenguest-install.py", line 347, in main src = get_paravirt_install(options) File "/usr/sbin/xenguest-install.py", line 223, in get_paravirt_install if src.startswith("http://") or src.startswith("ftp://"): AttributeError: 'NoneType' object has no attribute 'startswith'
Also vmlinuz-2.6.15-1.43_FC5 hypervisor won't boot. It starts to boot and then the machine immediately restarts. Is there anyway I can figure out the actual error?
I still haven't got a guest installed using the xenguest-install.py process. When the actual installation starts, it installs a few packages and then dies.
-- Adam
On Feb 8, 2006, at 8:06 AM, Jeremy Katz wrote:
On Tue, 2006-02-07 at 22:15 -0800, Adam Huda wrote:
One suggestion for xenguest-install.py. You might want to set the initial value of ram to 256. It is a little confusing when it prints out "ERROR: Installs currently require 256 megs of RAM." in interactive mode (running without any command line args) before your prompted with "How much RAM should be allocated (in megabytes)? ".
This should actually be handled better as of today's xen package
Jeremy
-- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
On Wed, 2006-02-08 at 14:47 -0800, Adam Huda wrote:
Found this problem though:
[snip]
AttributeError: 'NoneType' object has no attribute 'startswith'
Fixed this this afternoon.
Also vmlinuz-2.6.15-1.43_FC5 hypervisor won't boot. It starts to boot and then the machine immediately restarts. Is there anyway I can figure out the actual error?
If you add noreboot to the xen line, it won't reboot. If you're on an smp box, you're probably hitting #180535. maxcpus=1 on the kernel line will work around and Rik is hot on the case.
I still haven't got a guest installed using the xenguest-install.py process. When the actual installation starts, it installs a few packages and then dies.
Hrmm -- I haven't seen this. How does it die? Does the guest just hang or ... ?
Jeremy
If you add noreboot to the xen line, it won't reboot. If you're on an smp box, you're probably hitting #180535. maxcpus=1 on the kernel line will work around and Rik is hot on the case.
Thanks, maxcpus=1 did the trick.
Hrmm -- I haven't seen this. How does it die? Does the guest just hang or ... ?
It hangs when it gets to the package:
glibc-common-2.3.90-36: Common binaries and locale data for glibc
Looking at xm top, no network traffic is flowing.
On Feb 8, 2006, at 3:04 PM, Jeremy Katz wrote:
On Wed, 2006-02-08 at 14:47 -0800, Adam Huda wrote:
Found this problem though:
[snip]
AttributeError: 'NoneType' object has no attribute 'startswith'
Fixed this this afternoon.
Also vmlinuz-2.6.15-1.43_FC5 hypervisor won't boot. It starts to boot and then the machine immediately restarts. Is there anyway I can figure out the actual error?
If you add noreboot to the xen line, it won't reboot. If you're on an smp box, you're probably hitting #180535. maxcpus=1 on the kernel line will work around and Rik is hot on the case.
I still haven't got a guest installed using the xenguest-install.py process. When the actual installation starts, it installs a few packages and then dies.
Hrmm -- I haven't seen this. How does it die? Does the guest just hang or ... ?
Jeremy
-- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
On Wed, 2006-02-08 at 16:09 -0800, Adam Huda wrote:
Hrmm -- I haven't seen this. How does it die? Does the guest just hang or ... ?
It hangs when it gets to the package: glibc-common-2.3.90-36: Common binaries and locale data for glibc Looking at xm top, no network traffic is flowing.
How much memory are you giving your guest?
Jeremy
I gave the guest 256 MB. My machine has 2 GB. Under 2.6.15-1.40_FC5hypervisor I tried setting:
xm mem-max 512 0 xm mem-set 512 0
Then when starting anaconda, I got some error about the net driver having to squeeze memory (sorry about being hazing on the exact error, I can reproduce it later if you like) when getting an IP via DHCP and at which point the install would break.
With the latest xen hypervisor (43), I'm not having any problems with the anaconda install. I just finished a rawhide guest installation and rebooted the guest. However, the pygrub boot loader is giving me some problems when I try to start the guest:
[root@localhost ~]# xm create -c rawhide Using config file "/etc/xen/rawhide". Traceback (most recent call last): File "/usr/bin/pygrub", line 246, in ? cf = get_config(file) File "/usr/bin/pygrub", line 123, in get_config raise RuntimeError, "we couldn't find /boot/grub {menu.lst,grub.conf} " + \ RuntimeError: we couldn't find /boot/grub{menu.lst,grub.conf} in the image provided. halt! Error: Boot loader didn't return any data!
I haven't had a chance to dig around this yet.
-- Adam
On Feb 8, 2006, at 4:22 PM, Jeremy Katz wrote:
On Wed, 2006-02-08 at 16:09 -0800, Adam Huda wrote:
Hrmm -- I haven't seen this. How does it die? Does the guest just hang or ... ?
It hangs when it gets to the package: glibc-common-2.3.90-36: Common binaries and locale data for glibc Looking at xm top, no network traffic is flowing.
How much memory are you giving your guest?
Jeremy
-- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
On Wed, 8 Feb 2006, Adam Huda wrote:
I still haven't got a guest installed using the xenguest-install.py process. When the actual installation starts, it installs a few packages and then dies.
I am not entirely sure why people want to fake a netboot and go through anaconda. What is wrong with using yum with --installroot?
dd if=/dev/zero of=fc4-i386.img bs=1M count=1 seek=4096 /sbin/mke2fs -F -j fc4-i386.img mount -o loop fc4-i386.img /mnt mkdir /mnt/dev mkdir /mnt/proc mkdir /mnt/etc for i in console null zero ; do /sbin/MAKEDEV -d /mnt/dev -x $i ; done cp projects/documentation/xen/fstab-xen /mnt/etc/fstab mount -t proc none /mnt/proc yum -c yum-xen.conf --installroot=/mnt -y groupinstall Base find /mnt/var/cache/yum -name *.rpm | xargs rm mv /mnt/lib/tls /mnt/lib/tls-disabled sync (on fc3 I had to kill minilogd that got started as part of the install) umount /mnt/proc umount /mnt
where fstab-xen is:
/dev/hda1 / ext3 defaults 1 1 none /dev/pts devpts gid=5,mode=620 0 0 none /dev/shm tmpfs defaults 0 0 none /proc proc defaults 0 0 none /sys sysfs defaults 0 0
and yum-xen.conf:
[main] cachedir=/var/cache/yum debuglevel=2 logfile=/var/log/yum.log exclude=*-debuginfo gpgcheck=0 obsoletes=1 reposdir=/dev/null
[base] name=Fedora Core 4 - i386 - Base mirrorlist=http://fedora.redhat.com/download/mirrors/fedora-core-4 enabled=1
[updates-released] name=Fedora Core 4 - i386 - Released Updates mirrorlist=http://fedora.redhat.com/download/mirrors/updates-released-fc4 enabled=1
That's how I've created my FC3 and FC4 images. I haven't had time to point one at FC5 yet, but I would assume that works as well?
Paul
On Thu, Feb 09, 2006 at 01:06:15AM +0100, Paul Wouters wrote:
I am not entirely sure why people want to fake a netboot and go through anaconda. What is wrong with using yum with --installroot?
Well, some things are not configured then, like
- Network - Authentication - Firewall - Root password - Timezone - ...
This is all easy to script, of course, if you know what you want, or can be done later, but some people may not want to find this out themselves (although we could extend the list by some commands like system-config-... to do these things later).
for i in console null zero ; do /sbin/MAKEDEV -d /mnt/dev -x $i ; done
Note that I needed to add "tty" to this list *without* the -x, as otherwise the early boot stage - when /dev fs was not yet mounted - could not open tty1 etc. (but this was not on the latest Fedora, so this might be different for FC5).
mv /mnt/lib/tls /mnt/lib/tls-disabled
Unless you use a glibc compiled with "-mno-tls-direct-seg-refs" (added to BuildFlags in the glibc spec file), recommended, AFAIK. B.t.w., what are the drawbacks of this?
(on fc3 I had to kill minilogd that got started as part of the install)
Same here...
On Thu, 2006-02-09 at 01:20 +0100, Jos Vos wrote:
On Thu, Feb 09, 2006 at 01:06:15AM +0100, Paul Wouters wrote:
mv /mnt/lib/tls /mnt/lib/tls-disabled
Unless you use a glibc compiled with "-mno-tls-direct-seg-refs" (added to BuildFlags in the glibc spec file), recommended, AFAIK. B.t.w., what are the drawbacks of this?
You lose decent threading (NPTL). Also, you can't do this with fc5 as glibc is always built for NPTL. But, there's the nosegneg glibc binaries included which get used automatically with Xen kernels so that you can get TLS without the downsides. This should work on FC4, too
Jeremy
On Wed, Feb 08, 2006 at 07:24:49PM -0500, Jeremy Katz wrote:
On Thu, 2006-02-09 at 01:20 +0100, Jos Vos wrote:
Unless you use a glibc compiled with "-mno-tls-direct-seg-refs" (added to BuildFlags in the glibc spec file), recommended, AFAIK. B.t.w., what are the drawbacks of this?
You lose decent threading (NPTL). Also, you can't do this with fc5 as
But not with the "-mno-tls-...", right? I was asking for the drawbacks of using glibc compiled with this flag: so, why is glibc only generated that way for use with Xen?
On Thu, 2006-02-09 at 01:30 +0100, Jos Vos wrote:
On Wed, Feb 08, 2006 at 07:24:49PM -0500, Jeremy Katz wrote:
On Thu, 2006-02-09 at 01:20 +0100, Jos Vos wrote:
Unless you use a glibc compiled with "-mno-tls-direct-seg-refs" (added to BuildFlags in the glibc spec file), recommended, AFAIK. B.t.w., what are the drawbacks of this?
You lose decent threading (NPTL). Also, you can't do this with fc5 as
But not with the "-mno-tls-...", right? I was asking for the drawbacks of using glibc compiled with this flag: so, why is glibc only generated that way for use with Xen?
ISTR there being a performance difference
Jeremy
On Thu, 2006-02-09 at 01:06 +0100, Paul Wouters wrote:
On Wed, 8 Feb 2006, Adam Huda wrote:
I still haven't got a guest installed using the xenguest-install.py process. When the actual installation starts, it installs a few packages and then dies.
I am not entirely sure why people want to fake a netboot and go through anaconda. What is wrong with using yum with --installroot?
Because this 1) Requires a completely new set of tools for provisioning that is different from everything else you use 2) Requires a number of manual steps 3) Doesn't give good support for things like SELinux 4) Requires you to manage kernels for guests from the host. Which means you install some stuff to the guest, but the kernel you have to special case -- just weird
Using anaconda is far, far nicer. Plus, it will continue to work as the distro changes instead of requiring changes to a weird set of steps.
Jeremy
Once upon a time, Paul Wouters paul@cypherpunks.ca said:
I am not entirely sure why people want to fake a netboot and go through anaconda. What is wrong with using yum with --installroot?
yum installs a bunch of packages; anaconda builds a system (there's a difference).
On Wed, 2006-02-08 at 14:30 +0900, Naoki wrote:
Is there a simple way to fake a netboot so I can use my existing kickstart scripts and build OS instances the same way I would with normal physical hardware?
Realistically, the xenguest-install stuff is pretty much faking netboot. Run it with --help to see the way to pass the various things to it as command line arguments and you can use -x for passing things like ks=
As we say in my country, "Sweet mate!". I've tested with current rawhide and it's looking good except for the stage2 problem which I think has already been covered today.
Quick note about -l LOCATION, --location=LOCATION :
"http://host:/path" may be confusing because it will not work with the middle ":" unless a port is specified. Perhaps "http://host:port/path" is better?
Or just even http://host/path. I must have been sleep walking when I wrote that. Fixed it up for the next time I build the package
Jeremy
On Wed, 2006-02-08 at 12:15 +0900, Naoki wrote:
http://fedoraproject.org/wiki/FedoraXenQuickstart & http://fedoraproject.org/wiki/FedoraXenQuickstartFC5
I'm just running through these instructions as a reference for my Xen3 tests on rawhide and all is going pretty well.
I have a test domain (but can't alter the number of VCPUs for some reason but that's another story) and I'm wondering. Rather than the pretty ugly system of either :
Make an image file. Mount. yum --installroot=blah groupinstall x y z Configure.
Or the much better system of /usr/sbin/xenguest-install.py,
Is there a simple way to fake a netboot so I can use my existing kickstart scripts and build OS instances the same way I would with normal physical hardware? It would be nice to have my hypervisor instance running dhcp + apache and do localhost kickstart network installs.
(I know this has been asked times and times before.)
but... What's the current status of Xen/x86_64? Second, considering the fact that rawhide/x86_64 still doesn't have kernel-xen and the pending release of test3, what are the chances of Xen/x86_64 making it's way into FC5? Any chance it'll make it into test3? (I assume that past test3, only bug fixes will be introduced until FC5-release)
Gilboa
On Wed, 2006-02-08 at 15:41 +0200, Gilboa Davara wrote:
(I know this has been asked times and times before.)
but... What's the current status of Xen/x86_64?
It's being heavily worked on. Trust me, we want it as much as you do.
Second, considering the fact that rawhide/x86_64 still doesn't have kernel-xen and the pending release of test3, what are the chances of Xen/x86_64 making it's way into FC5?
We're hopefully optimistic that we'll get something working before the test3 freeze so that it can be included in test3 and then the final FC5
Any chance it'll make it into test3? (I assume that past test3, only bug fixes will be introduced until FC5-release)
See above. And yeah, if it doesn't make test3, chances for release are slim
Jeremy
On Wed, 2006-02-08 at 11:08 -0500, Jeremy Katz wrote:
On Wed, 2006-02-08 at 15:41 +0200, Gilboa Davara wrote:
(I know this has been asked times and times before.)
but... What's the current status of Xen/x86_64?
It's being heavily worked on. Trust me, we want it as much as you do.
Second, considering the fact that rawhide/x86_64 still doesn't have kernel-xen and the pending release of test3, what are the chances of Xen/x86_64 making it's way into FC5?
We're hopefully optimistic that we'll get something working before the test3 freeze so that it can be included in test3 and then the final FC5
Any chance it'll make it into test3? (I assume that past test3, only bug fixes will be introduced until FC5-release)
See above. And yeah, if it doesn't make test3, chances for release are slim
Jeremy
OK. Thanks for the update.
Gilboa