Difference between revisions of "Xen"

From YobiWiki
Jump to navigation Jump to search
m (Reverted edits by Etegohy (Talk) to last revision by PhilippeTeuwen)
 
(9 intermediate revisions by the same user not shown)
Line 1: Line 1:
  +
===XEN: installation===
''Converted with [[MediaWiki#HTML-WikiConverter|HTML::WikiConverter::MediaWiki]] from my old phpwiki site''
 
  +
* see [http://wiki.debian.org/Xen notes on Debian wiki]
----
 
  +
* aptitude install xen-linux-image-2.6.32-5-xen-amd64 xen-tools xen-utils
=== Install of Xen Source (paravirtualizer) on Debian Sarge ===
 
  +
* backup /etc/xen/xend-config.sxp
  +
* edit /etc/xen/xend-config.sxp, and uncomment line "(network-script network-bridge)" or for a non-default interface ethX:
  +
(network-script 'network-bridge netdev=ethX bridge=xenbr0')
  +
* reboot
  +
* aptitude remove --purge linux-image-2.6.32-5-amd64 linux-image-2.6-amd64
  +
* /etc/default/grub => GRUB_DEFAULT=2
  +
* update-grub
  +
* reboot
  +
* cat /proc/xen/capabilities #(to check xend is running)
   
  +
To install libvirt & get some additional tools:
==== Install of Xen ====
 
  +
* apt-get install libvirt-bin
 
  +
* Edit /etc/xen/xend-config.sxp and set
I used the binary distribution of xen 2.0.7 altough I also tried to compile a custom kernel as I had question about hardware support which I didn't find in the pached kernel. I posted it on the xen-users ML, no answer yet.
 
  +
(xend-unix-server yes)
 
To be documented (note are offline)
 
 
update grub's menu.lst and rename /lib/tls in /lib/tls.disabled as it is not natively supported by xen and need a real emulation which will reduced performance a lot (this step has to be done in all unpriviledged guest too (enforce it ?)
 
 
==== Xen topology ====
 
 
Xen uses a para-vitualisation hypervisor running in ring 0 (x86 CPU). <br />The kernel runs in ring1 and user-app in ring3. <br />Hardware management is done inside a privileged kernel named domain0 (guest OSes are called domainX). <br />Xen uses a xend daemon in dom0 to manage network, cpu, memory and export standard devices to the domX guest (generally eth0, a root fs and some swap space). <br />xend also manages bridging the real eth0 and sets up a xen-br0 device <br />xend is not running by default on dom0 at startup
 
 
Invoking
 
   
  +
===XEN: guest creation===
  +
* If you've a proxy it's much easier to have it defined in the environment variables:
  +
** Edit /etc/environment
  +
http_proxy=http://myproxy:port
  +
** Logout/login
  +
* Edit the config file for creating images (/etc/xen-tools/xen-tools.conf) to your needs:
  +
<source lang=diff>
  +
--- xen-tools.conf.orig 2010-05-30 22:42:25.000000000 +0200
  +
+++ xen-tools.conf 2010-08-17 17:47:35.000000000 +0200
  +
@@ -52,7 +52,7 @@
  +
# LVM volume group here instead
  +
#
  +
##
  +
-# lvm = vg0
  +
+lvm = vg0
 
 
update-rc.d xend defaults 80
 
 
does the job as dmesg says
 
 
 
 
eth0: link up, 100Mbps, full-duplex, lpa 0x45E1
 
eth0: Promiscuous mode enabled.
 
device eth0 entered promiscuous mode
 
xen-br0: port 1(eth0) entering learning state
 
xen-br0: topology change detected, propagating
 
xen-br0: port 1(eth0) entering forwarding state
 
 
and
 
 
 
ns1:/home/skycode# xm list
 
Name Id Mem(MB) CPU State Time(s) Console
 
Domain-0 0 251 0 r---- 13.0
 
 
So domain0 is now running (in a small 4Go partition, the rest of the disk is left for LVM)
 
 
setting-up LVM for the hosts
 
 
 
apt-get install lvm2
 
 
Partitioning the free space left on the disk
 
 
 
ns1:/home/skycode# fdisk /dev/hda
 
 
The number of cylinders for this disk is set to 9964.
 
There is nothing wrong with that, but this is larger than 1024,
 
and could in certain setups cause problems with:
 
1) software that runs at boot time (e.g., old versions of LILO)
 
2) booting and partitioning software from other OSs
 
(e.g., DOS FDISK, OS/2 FDISK)
 
 
Command (m for help): p
 
 
Disk /dev/hda: 81.9 GB, 81964302336 bytes
 
255 heads, 63 sectors/track, 9964 cylinders
 
Units = cylinders of 16065 * 512 = 8225280 bytes
 
 
Device Boot Start End Blocks Id System
 
/dev/hda1 1 486 3903763+ 83 Linux
 
/dev/hda2 * 487 608 979965 82 Linux swap / Solaris
 
 
Command (m for help): n
 
Command action
 
e extended
 
p primary partition (1-4)
 
p
 
Partition number (1-4): 3
 
First cylinder (609-9964, default 609):
 
Using default value 609
 
Last cylinder or +size or +sizeM or +sizeK (609-9964, default 9964):
 
Using default value 9964
 
 
Command (m for help): p
 
 
Disk /dev/hda: 81.9 GB, 81964302336 bytes
 
255 heads, 63 sectors/track, 9964 cylinders
 
Units = cylinders of 16065 * 512 = 8225280 bytes
 
 
Device Boot Start End Blocks Id System
 
/dev/hda1 1 486 3903763+ 83 Linux
 
/dev/hda2 * 487 608 979965 82 Linux swap / Solaris
 
/dev/hda3 609 9964 75152070 83 Linux
 
 
Command (m for help): t
 
Partition number (1-4): 3
 
Hex code (type L to list codes): 8e
 
Changed system type of partition 3 to 8e (Linux LVM)
 
 
Command (m for help): p
 
 
Disk /dev/hda: 81.9 GB, 81964302336 bytes
 
255 heads, 63 sectors/track, 9964 cylinders
 
Units = cylinders of 16065 * 512 = 8225280 bytes
 
 
Device Boot Start End Blocks Id System
 
/dev/hda1 1 486 3903763+ 83 Linux
 
/dev/hda2 * 487 608 979965 82 Linux swap / Solaris
 
/dev/hda3 609 9964 75152070 8e Linux LVM
 
 
Command (m for help): w
 
The partition table has been altered!
 
 
Calling ioctl() to re-read partition table.
 
 
reboot :-)
 
 
Create physical volumes
 
 
pvcreate /dev/hda3&lt;/pre&gt;
 
add it to a volume groupvgcreate volg0 /dev/hda3
 
Volume group "volg0" successfully created
 
 
Create 2 partition (root and swap for the first test domain)
 
 
ns1:/home/skycode# lvcreate -L 2G -n test0root volg0
 
Logical volume "test0root" created
 
ns1:/home/skycode# lvcreate -L 256M -n test0swap volg0
 
Logical volume "test0swap" created
 
ns1:/home/skycode# mke2fs -j /dev/volg0/test0root
 
mke2fs 1.37 (21-Mar-2005)
 
Filesystem label=
 
OS type: Linux
 
Block size=4096 (log=2)
 
Fragment size=4096 (log=2)
 
262144 inodes, 524288 blocks
 
26214 blocks (5.00%) reserved for the super user
 
First data block=0
 
16 block groups
 
32768 blocks per group, 32768 fragments per group
 
16384 inodes per group
 
Superblock backups stored on blocks:
 
32768, 98304, 163840, 229376, 294912
 
 
Writing inode tables: done
 
Creating journal (8192 blocks): done
 
Writing superblocks and filesystem accounting information: done
 
 
This filesystem will be automatically checked every 38 mounts or
 
180 days, whichever comes first. Use tune2fs -c or -i to override.
 
ns1:/home/skycode# mkswap /dev/volg0/test0swap
 
Setting up swapspace version 1, size = 268431 kB
 
no label, UUID=451e4e02-5d9f-4524-922a-c7432e844b21
 
ns1:/home/skycode#
 
 
Install a minimal debian sarge with debootstrap and back-it up for futur installs
 
 
 
ns1:/home/skycode# debootstrap --arch i386 sarge /mnt/xen_install http://www.uk.debian.org/debian/
 
Lots of output ...
 
ns1:/home/skycode# cd /mnt/xen_install/
 
ns1:/mnt/xen_install# mkdir /var/xen_images
 
ns1:/mnt/xen_install# tar -jpcf /var/xen_images/debian-sarge-pre-config.tar.bz2 .
 
 
Now it's config time
 
 
fstab
 
 
<nowiki>
 
ns1:/mnt/xen_install/etc# cat fstab
 
# Fstab for virtual domain root is exported by dom0 as sda1
 
# And swap as sda2
 
 
#
 
#
  +
@@ -125,12 +125,12 @@
# Exxoss - brewed on March 24th 2006
 
  +
# Disk and Sizing options.
 
  +
##
/dev/sda1 / ext3 defaults 0 1
 
  +
#
/dev/sda2 swap swap defaults 0 0
 
  +
-size = 4Gb # Disk image size.
proc /proc proc defaults 0 0</nowiki>
 
  +
+size = 8Gb # Disk image size.
 
  +
memory = 128Mb # Memory size
hostname
 
  +
swap = 128Mb # Swap size
 
  +
# noswap = 1 # Don't use swap at all for the new system.
 
  +
fs = ext3 # use the EXT3 filesystem for the disk image.
ns1:/mnt/xen_install/etc# cat hostname
 
  +
-dist = `xt-guess-suite-and-mirror -s` # Default distribution to install.
test0
 
  +
+dist = squeeze # Default distribution to install.
 
  +
image = sparse # Specify sparse vs. full disk images.
hosts
 
 
 
ns1:/mnt/xen_install/etc# cat hosts
 
127.0.0.1 localhost
 
 
Network interfaces
 
 
<nowiki>
 
ns1:/mnt/xen_install/etc# cat network/interfaces
 
# Used by ifup(8) and ifdown(8). See the interfaces(5) manpage or
 
# /usr/share/doc/ifupdown/examples for more information.
 
 
 
 
#
 
#
  +
@@ -166,7 +166,7 @@
# Exxoss - brewed march 24th 2006
 
# virtual domain interface config file
 
# dom0 export a NIC as eth0
 
 
#
 
#
  +
# Uncomment this if you wish the images to use DHCP
  +
#
  +
-# dhcp = 1
  +
+dhcp = 1
 
 
# Loopback device
 
auto lo
 
iface lo inet loopback
 
 
 
  +
##
# eth0 device
 
  +
@@ -184,7 +184,7 @@
auto eth0
 
  +
# Uncomment the following line if you wish to interactively setup
iface eth0 inet static
 
  +
# a new root password for images.
address 192.168.1.111
 
  +
#
netmask 255.255.255.0
 
  +
-# passwd = 1
gateway 192.168.1.1</nowiki>
 
  +
+passwd = 1
 
sources.list
 
 
<nowiki>
 
ns1:/mnt/xen_install/etc# cat apt/sources.list
 
#deb file:///cdrom/ sarge main
 
 
 
deb </nowiki>http://ftp.debian.skynet.be/ftp/debian/ stable main contrib
 
deb-src http://ftp.debian.skynet.be/ftp/debian/ stable main contrib
 
 
deb http://security.debian.org/ stable/updates main contrib
 
 
lib/tls (again)
 
 
 
ns1:/mnt/xen_install/etc# cd ..
 
ns1:/mnt/xen_install# mv lib/tls lib/tls.disabled
 
 
umount /mnt/xen_install
 
 
Configure xen for the new dom
 
 
in /etc/xen/test0.conf
 
 
<nowiki>
 
ns1:/etc/xen# cat /etc/xen/test0.conf
 
 
#
 
#
  +
# If you'd like all accounts on your host system which are not present
# Test Xen domU config file
 
  +
@@ -254,10 +254,10 @@
 
#
 
#
  +
# You may specify the things to use here:
 
  +
#
name="test0"
 
  +
-# serial_device = hvc0 #default
memory=256
 
  +
+serial_device = hvc0 #default
kernel="/boot/vmlinuz-2.6-xenU"
 
  +
# serial_device = tty1
nics=1
 
  +
#
disk=</nowiki><span>[%27phy%3Avolg0 'phy:volg0][%27phy%3Avolg0/test0root%2Csda1%2Cw%27%2C%20%27phy%3Avolg0 /test0root,sda1,w', 'phy:volg0]<span class="named-wikiunknown"><u>test0swap,sda2,w'</u>[%27phy%3Avolg0/test0root%2Csda1%2Cw%27%2C%20%27phy%3Avolg0/test0swap%2Csda2%2Cw%27?action=create [[Image:uww.png|?]]]</span></span>
 
  +
-# disk_device = xvda #default
root="/dev/sda1 ro"
 
  +
+disk_device = xvda #default
ns1:/etc/xen#
 
  +
# disk_device = sda
  +
#
  +
</source>
   
  +
Example:
==== xm (xend ctl tool) commands ====
 
   
  +
* xen-create-image --hostname=myguest1 --memory=1G --swap=1G --vcpus=20
Create a new domain &lt;pre&gt;xm create /etc/xen/test0.conf -c&lt;/pre&gt;
 
  +
* WARNING strangely the default root password was disabled with a '*' in /etc/shadow so impossible to log in. I had to mount manually the lv (/dev/vg0/myguest1-disk) to remove the '*'<br>One may prefer to setup the password interactively when creating the image with xen-create-image ... --passwd (or setup passwd=1 in /etc/xen-tools/xen-tools.conf as I've now shown in the above diff file)
  +
* lvm> lvcreate -L 1000G -n myguest1-data vg0
  +
* /etc/xen/myguest1.cfg: add data volume:
  +
root = '/dev/xvda2 ro'
  +
disk = [
  +
'phy:/dev/vg0/myguest1-disk,xvda2,w',
  +
'phy:/dev/vg0/myguest1-swap,xvda1,w',
  +
'phy:/dev/vg0/myguest1-data,xvda3,w',
  +
]
   
  +
* xm create -c /etc/xen/myguest1.cfg
Logout from console to dom0 : Ctrl-]
 
  +
* Hook the data drive as /home
  +
** mkfs.ext4 /dev/xvda3
  +
** In /etc/fstab: /dev/xvda3 /home ext4 noatime,nodiratime,errors=remount-ro 0 1
  +
** mount /home
   
  +
==XEN quickref==
list runing domains &lt;pre&gt; xm list &lt;/pre&gt;
 
  +
[http://xen-tools.org/software/xen-tools/ Xen Tools], [http://xen-tools.org/software/xen-shell Xen Shell]
   
  +
See also [http://www.howtoforge.com/virtualization-with-xen-on-debian-lenny-amd64-p2 here] and [http://www.howtoforge.com/xen_tools_xen_shell_argo here]
stop a runing domain &lt;pre&gt; xm shutdown domain &lt;/pre&gt;
 
  +
===Guest images===
  +
* xen-create-image --hostname=mynewguest
  +
** creation of config file and base filesystem
  +
* xen-delete-image
  +
* xen-list-images
  +
* xen-update-image
  +
** Does apt-get update; apt-get upgrade in the guest, !!! only if guest is NOT running
  +
===Guest start/stop===
  +
* xm create /etc/xen/mynewguest.cfg
  +
** actual starting of guest
  +
* xm shutdown mynewguest
  +
* xm destroy mynewguest
  +
===Guests info===
  +
* xm list
  +
** Show running xen guests
  +
* xm info
  +
** Show xen version and many other info
  +
* xm uptime
  +
* xm top
  +
* xm vcpu-list
  +
* xm network-list mynewguest
  +
* xm log
  +
* xm dmesg
  +
* brctl show
  +
* virsh nodeinfo
  +
* virsh vcpuinfo mynewguest
  +
* virsh dominfo mynewguest
  +
===Guest management===
  +
* xm console mynewguest
  +
** CTRL+] to detach
  +
* xenscreen
  +
** attaches one screen to all domains
  +
** ctrl-a \ to quit properly, or ctrl-a d to detach
  +
* xm block-attach mynewguest phy://dev/vg0/myguest1-data2 /dev/xvda4 w
  +
** to attach some device to a running guest
  +
** not tried: "w!" for shared read/write, when the same device is attached to two different VMs
  +
* xm block-list mynewguest
  +
* xm block-list mynewguest --long
  +
* xm block-detach mynewguest 12345
   
==== Tests ====
+
==XEN tuning==
  +
===Autostart===
  +
If you want xen1.example.com to start automatically at the next boot of the system, then do this (if XENDOMAINS_AUTO=/etc/xen/auto in /etc/default/xendomains):
   
  +
mkdir -p /etc/xen/auto
what happens if I try to allocate more meory to a domain than I have available
 
  +
ln -s /etc/xen/xen1.example.com.cfg /etc/xen/auto
  +
===Suspend or halt===
  +
By default it will attempt to save running hosts when dom0 is halted. This means a "suspend-to-disk" of all domU in dom0 filesystem (/var/lib/xen/save). To prevent it and do regular shutdown of domU, edit /etc/default/xendomains and change:
  +
#XENDOMAINS_SAVE=/var/lib/xen/save
  +
XENDOMAINS_SAVE=""
  +
===Wallclock===
  +
Since Xen 4, wallclocks are separated and need to be sync, e.g. in each domU (& dom0):
  +
apt-get install ntpdate
  +
Edit /etc/default/ntpdate
  +
NTPDATE_USE_NTP_CONF=no
  +
NTPSERVERS="....."
  +
Add cron job /etc/cron.daily/ntpdate
  +
#!/bin/sh
  +
/usr/sbin/ntpdate-debian
  +
And
  +
chmod 755 /etc/cron.daily/ntpdate
   
  +
==Misc notes==
it fail rturning an error code (12)
 
  +
===Boot & DHCP===
  +
I've still sometimes some issues at boot time when the xenbr0 bridge doesn't catch its IP...
  +
<br>In those cases I find dhclient still hooked to the non-existing bnx1 IF (which was renamed pbnx1 by XEN). I've to kill it and start a similar dhclient on IF xenbr0.
  +
<br>Finally I renamed the default interface as eth0 and called the xen bridge script without parameters: (network-script network-bridge)
   
  +
===Allocating too much memory===
set the cpu time limit
 
  +
I tried to create two guests with each 12Gb of RAM while the host had only 16Gb in total. I would have expected a nice handling of the error but it froze xend and I could even not shutdown properly.
  +
<br>And, oh, BTW, no, [http://blog.coparsystems.com/?p=36&more=1&c=1&tb=1&pb=1 xen doesn't support memory overcommitment]
   
  +
===Mount a DomU disk on Dom0===
use cpu_weight in config file like cpu_weight=5
 
  +
I had to transfer a large amount of data from a USB disk to a DomU.
  +
Easiest is to:
  +
* stop DomU (maybe it's possible to umount and detach block while running, I didn't try)
  +
* look into /etc/xen/mydomu.cfg for the disk you want, here
  +
disk = [
  +
...
  +
'phy:/dev/vg0/sagemath-data,xvda3,w',
  +
]
  +
* Hot plug block device to Dom0
  +
xm block-attach 0 phy:/dev/vg0/sagemath-data xvda3 w
  +
* Mount it
  +
mount /dev/xvda3 my-tmp-mount-point/
  +
When done:
  +
umount /dev/xvda3
  +
xm block-detach 0 xvda3

Latest revision as of 15:29, 22 April 2013

XEN: installation

  • see notes on Debian wiki
  • aptitude install xen-linux-image-2.6.32-5-xen-amd64 xen-tools xen-utils
  • backup /etc/xen/xend-config.sxp
  • edit /etc/xen/xend-config.sxp, and uncomment line "(network-script network-bridge)" or for a non-default interface ethX:
(network-script 'network-bridge netdev=ethX bridge=xenbr0')
  • reboot
  • aptitude remove --purge linux-image-2.6.32-5-amd64 linux-image-2.6-amd64
  • /etc/default/grub => GRUB_DEFAULT=2
  • update-grub
  • reboot
  • cat /proc/xen/capabilities #(to check xend is running)

To install libvirt & get some additional tools:

  • apt-get install libvirt-bin
  • Edit /etc/xen/xend-config.sxp and set
(xend-unix-server yes)

XEN: guest creation

  • If you've a proxy it's much easier to have it defined in the environment variables:
    • Edit /etc/environment
http_proxy=http://myproxy:port
    • Logout/login
  • Edit the config file for creating images (/etc/xen-tools/xen-tools.conf) to your needs:
--- xen-tools.conf.orig 2010-05-30 22:42:25.000000000 +0200
+++ xen-tools.conf      2010-08-17 17:47:35.000000000 +0200
@@ -52,7 +52,7 @@
 # LVM volume group here instead
 #
 ##
-# lvm = vg0
+lvm = vg0
 
 
 #
@@ -125,12 +125,12 @@
 #  Disk and Sizing options.
 ##
 #
-size   = 4Gb      # Disk image size.
+size   = 8Gb      # Disk image size.
 memory = 128Mb    # Memory size
 swap   = 128Mb    # Swap size
 # noswap = 1      # Don't use swap at all for the new system.
 fs     = ext3     # use the EXT3 filesystem for the disk image.
-dist   = `xt-guess-suite-and-mirror -s` # Default distribution to install.
+dist   = squeeze  # Default distribution to install.
 image  = sparse   # Specify sparse vs. full disk images.
 
 #
@@ -166,7 +166,7 @@
 #
 # Uncomment this if you wish the images to use DHCP
 #
-# dhcp = 1
+dhcp = 1
 
 
 ##
@@ -184,7 +184,7 @@
 # Uncomment the following line if you wish to interactively setup
 # a new root password for images.
 #
-# passwd = 1
+passwd = 1
 
 #
 # If you'd like all accounts on your host system which are not present
@@ -254,10 +254,10 @@
 #
 #  You may specify the things to use here:
 #
-# serial_device = hvc0 #default
+serial_device = hvc0 #default
 # serial_device = tty1
 #
-# disk_device = xvda #default
+disk_device = xvda #default
 # disk_device = sda
 #

Example:

  • xen-create-image --hostname=myguest1 --memory=1G --swap=1G --vcpus=20
  • WARNING strangely the default root password was disabled with a '*' in /etc/shadow so impossible to log in. I had to mount manually the lv (/dev/vg0/myguest1-disk) to remove the '*'
    One may prefer to setup the password interactively when creating the image with xen-create-image ... --passwd (or setup passwd=1 in /etc/xen-tools/xen-tools.conf as I've now shown in the above diff file)
  • lvm> lvcreate -L 1000G -n myguest1-data vg0
  • /etc/xen/myguest1.cfg: add data volume:
root        = '/dev/xvda2 ro'
disk        = [
                  'phy:/dev/vg0/myguest1-disk,xvda2,w',
                  'phy:/dev/vg0/myguest1-swap,xvda1,w',
                  'phy:/dev/vg0/myguest1-data,xvda3,w',
              ]
  • xm create -c /etc/xen/myguest1.cfg
  • Hook the data drive as /home
    • mkfs.ext4 /dev/xvda3
    • In /etc/fstab: /dev/xvda3 /home ext4 noatime,nodiratime,errors=remount-ro 0 1
    • mount /home

XEN quickref

Xen Tools, Xen Shell

See also here and here

Guest images

  • xen-create-image --hostname=mynewguest
    • creation of config file and base filesystem
  • xen-delete-image
  • xen-list-images
  • xen-update-image
    • Does apt-get update; apt-get upgrade in the guest, !!! only if guest is NOT running

Guest start/stop

  • xm create /etc/xen/mynewguest.cfg
    • actual starting of guest
  • xm shutdown mynewguest
  • xm destroy mynewguest

Guests info

  • xm list
    • Show running xen guests
  • xm info
    • Show xen version and many other info
  • xm uptime
  • xm top
  • xm vcpu-list
  • xm network-list mynewguest
  • xm log
  • xm dmesg
  • brctl show
  • virsh nodeinfo
  • virsh vcpuinfo mynewguest
  • virsh dominfo mynewguest

Guest management

  • xm console mynewguest
    • CTRL+] to detach
  • xenscreen
    • attaches one screen to all domains
    • ctrl-a \ to quit properly, or ctrl-a d to detach
  • xm block-attach mynewguest phy://dev/vg0/myguest1-data2 /dev/xvda4 w
    • to attach some device to a running guest
    • not tried: "w!" for shared read/write, when the same device is attached to two different VMs
  • xm block-list mynewguest
  • xm block-list mynewguest --long
  • xm block-detach mynewguest 12345

XEN tuning

Autostart

If you want xen1.example.com to start automatically at the next boot of the system, then do this (if XENDOMAINS_AUTO=/etc/xen/auto in /etc/default/xendomains):

mkdir -p /etc/xen/auto
ln -s /etc/xen/xen1.example.com.cfg /etc/xen/auto

Suspend or halt

By default it will attempt to save running hosts when dom0 is halted. This means a "suspend-to-disk" of all domU in dom0 filesystem (/var/lib/xen/save). To prevent it and do regular shutdown of domU, edit /etc/default/xendomains and change:

#XENDOMAINS_SAVE=/var/lib/xen/save
XENDOMAINS_SAVE=""

Wallclock

Since Xen 4, wallclocks are separated and need to be sync, e.g. in each domU (& dom0):

apt-get install ntpdate

Edit /etc/default/ntpdate

NTPDATE_USE_NTP_CONF=no 
NTPSERVERS="....."

Add cron job /etc/cron.daily/ntpdate

#!/bin/sh
/usr/sbin/ntpdate-debian

And

chmod 755 /etc/cron.daily/ntpdate

Misc notes

Boot & DHCP

I've still sometimes some issues at boot time when the xenbr0 bridge doesn't catch its IP...
In those cases I find dhclient still hooked to the non-existing bnx1 IF (which was renamed pbnx1 by XEN). I've to kill it and start a similar dhclient on IF xenbr0.
Finally I renamed the default interface as eth0 and called the xen bridge script without parameters: (network-script network-bridge)

Allocating too much memory

I tried to create two guests with each 12Gb of RAM while the host had only 16Gb in total. I would have expected a nice handling of the error but it froze xend and I could even not shutdown properly.
And, oh, BTW, no, xen doesn't support memory overcommitment

Mount a DomU disk on Dom0

I had to transfer a large amount of data from a USB disk to a DomU. Easiest is to:

  • stop DomU (maybe it's possible to umount and detach block while running, I didn't try)
  • look into /etc/xen/mydomu.cfg for the disk you want, here
disk        = [
                  ...
                  'phy:/dev/vg0/sagemath-data,xvda3,w',
              ]
  • Hot plug block device to Dom0
xm block-attach 0 phy:/dev/vg0/sagemath-data xvda3 w
  • Mount it
mount /dev/xvda3 my-tmp-mount-point/

When done:

umount /dev/xvda3
xm block-detach 0 xvda3