HannibalWiki

This page is read only. You can view the source, but not change it. Ask your administrator if you think this is wrong.

=======Xen-clustering/live-migration=======


=====shared storage=====
In order to be able to do live migration of Xen guests from one cluster member to another, some sort of shared storage is required. As the Xen guest won't run on more than one cluster member at a time, a cluster filesystem is not required. That is, as long as you configure Xen to access the Xen guest by a physical device, not a file. Furthermore, if you want to share the Xen guest configuration files, then a cluster filesystem is required as soon as you want to write data to that share.

We wanted to look beyond high-end FCAL-based SAN's. Inspired by some Oracle RAC documentation (see: http://www.oracle.com/technology/pub/articles/hunter_rac10gr2.html) a shared firewire disk or -array appeared to be an option. A second possibility was mentioned in the Xen live migration paper from Novell/SuSe (see below). In this proof of concept iSCSI is being used as a shared storage solution. 

====Firewire====

By default Linux logs on to a firewire device in exclusive mode. This prevents accidently accessing the same device by another node that would probably screw up your data. Fortunately you can bypass the exclusive login mechanism using a kernel module option to the serial bus protocol kernel module (sbp2 exclusive_login=0).

For this to work, the chipset of your firewire device(s) should support multiple logins. For example the Oxford-chipset is known to support multiple logins. Check the afore-mentioned Oracle RAC documentation for more information on shared firewire hardware.

Adjust /etc/modules for the necessary modules
  sd_mod
  ieee1394
  ohci1394
  sbp2 exclusive_login=0

Adjust /etc/modprobe.d/sbp2
  options sbp2 exclusive_login=0 serialize_io=1

For immediate effect:
  rmmod sbp2
  modprobe sbp2

For permanent effect, rebuild your initial-ramdisk, to have these options also used in this (because the sbp2 module is loaded at boottime).
  mkinitramfs -o foo.version.img <kernel-version>
  or
  mkinitrd -o /boot/initrd.img-2.6.12.6-xen-fw 2.6.12.6-xen   (example)

====iSCSI====
===target===
First we need an iSCSI target. That is a device/server that provides shared storage on your network. We use the iSCSI enterprise target software to build a Linux based iSCSI target server.

==install==
=from source=

You can download the software at http://iscsitarget.sourceforge.net/

After building and installing the software, you'll have a kernel module named iscsi_trgt, a daemon called 'ietd' and a tool called 'ietdadm'.

=package=

Debian Lenny

It's fairly easy to create an ISCSI-target using Debian Lenny as prebuild binary packages are available. You'll need userland tools and kernel-modules for your kernel:

  apt-get install iscsitarget iscsitarget-modules-2.6.26-1-xen-amd64

Debian Etch

Although binary packages are not yet available for Debian Etch, Philipp Hug created unofficial packages for Debian Sid and Ubuntu Dapper. They are available at http://iscsitarget.sourceforge.net/wiki/index.php/Unoffical_DEBs

We installed the binary package named 'iscsitarget', which contains the userland binaries, on Debian Etch with no problems. The package named 'iscsitarget-source' contains the kernel module sources. This package allows you to build a binary kernel module for your kernel. The build on Debian Etch went flawlessly.

Add this line to /etc/apt/sources.list
  deb http://debian.hug.cx/debian/ unstable/

Then procede to install the software and build the kernelmodule
  apt-get install module-assistant debhelper linux-source-2.6.18 dpkg-dev \
                  kernel-package libncurses-dev libssl-dev linux-headers-2.6.18-4-xen-amd64
  cd /usr/src/
  tar -jxvf linux-source-2.6.18.tar.bz2
  ln -s linux-source-2.6.18 linux
  
  apt-get install  iscsitarget iscsitarget-source
  tar -zxvf iscsitarget.tar.gz  (this unpacks in sub-dir iscsitarget)
  
  m-a a-i iscsitarget

==config==
After installing the Debian package enable the iSCSI-target in /etc/default/iscsitarget

  ISCSITARGET_ENABLE=true

Let's configure the daemon. We have to tell it which device(s) to enable and which clients should be able to access them. In the next example we will enable the logical volumes named 'vault1' and 'vault2' to everybody. The configuration file is /etc/ietd.conf

  Target iqn.2006-07.com.example.intra:storage.disk1.vault
         Lun 0 Path=/dev/mapper/vg00-vault1,Type=fileio
         Alias vault1
  Target iqn.2006-07.com.example.intra:storage.disk2.vault
         Lun 1 Path=/dev/mapper/vg00-vault2,Type=fileio
         Alias vault2

Remember that every node on your network that uses iSCSI will need a unique 'iqn'. Check the iSCSI documentation on the web for the applicable syntax. You can add some lines to /etc/ietd.conf that require a username/password for iSCSI logons to succeed but this is ommitted by default.

Optionally you can provide an obligated SCSI-ID. This proves handy e.g. if you replicate your iSCSI SAN and you want to be able to bundle replicated shares on separate nodes using multipath. In this example snipplet a logical volume that is replicated by DRBD will serve as backend for our iSCSI-target.

  Target iqn.2009-01.nl.pref:storage.xensan1
         Lun 0 Path=/dev/drbd0,Type=fileio,ScsiId=149455403f160c00
         Alias san1

Important note if you want to put iSCSI on top of a DRBD device: on Debian Lenny the package init-scripts start iSCSI before DRBD and vice versa stop DRBD before iSCSI. You're advised to change these values around.

Start the iSCSI target daemon to enable the shared storage provider. This will open TCP port 3260 by default.
  /etc/init.d/iscsi-target start

Suggestion: you might want to use [[hannibal:bonding|ethernet bonding]] in order to aggregate several physical NIC's.

===initiator===
On the clients that will have to access the iSCSI based shared storage we need to install and configure iSCSI initiator software. We'll use the Open iSCSI package.

==install==
=from source=

The source is available at http://www.open-iscsi.org/

=package=

Debian Lenny and Debian Etch provide a binary package that contains iSCSI initiator software.
  apt-get install open-iscsi

==config==
After building and installing the software, you'll have two kernel modules named iscsi_tcp and scsi_transport_iscsi, a daemon called 'iscsid' and a tool called 'iscsiadm'.

The install procedure of Open iSCSI will create a configuration file /etc/iscsi/iscsid.conf that enables a default configuration.
You might want to enable iSCSI on node startup.

  ...
  node.startup = automatic
  ...

Now it's imperative to create a unique 'iqn' for our client and store it in /etc/iscsi/initiatorname.iscsi
  InitiatorName=iqn.2006-07.com.example.intra:hannibal.clientnode1

Afterwards, start the Open iSCSI daemon on the client
  /etc/init.d/open-iscsi start

Let's check at our iSCSI target server, for instance 192.168.1.16
  iscsiadm -m discovery -t sendtargets -p 192.168.1.16:3260

After logging on to the iSCSI-target a new SCSI-device should have been added to the client
  iscsiadm -m node -T iqn.2006-07.com.example.intra:storage.disk1.vault -p 192.168.1.16:3260 -l

In case you want to logout from your session make sure that you have unmounted the iSCSI-share and that you have deactivated the volume group in case you use LVM.
  vgchange -a n
  iscsiadm -m node -T iqn.2006-07.com.example.intra:storage.disk1.vault -p 192.168.1.16:3260 -u

Have fun!

====multipath====
iSCSI supports more than one connection to the same iSCSI LUN. This allows for high available setups.

On Linux the multipath-tools can map several iSCSI-devices into one multipath blockdevice that has loadbalancing and failover as features.

The multipath-tools are available as a binary package for Debian Lenny and Debian Etch.
  apt-get install multipath-tools

Create the /etc/multipath.conf file (some examples are available in /usr/share/doc/multipath-tools/examples). The SCSI-ID that must be entered on the 'wwid' line can be obtained by the scsi_id tool. In our example we'll get the same ID for sda and sdb. Remember they're two paths to the same LUN (although on different SAN-servers).
As the scsi_id tool is not in a regular path by default (at least on Debian Lenny), let's first make a symbolic link.

  ln -s /lib/udev/scsi_id /sbin/scsi_id
  /sbin/scsi_id -g -u -s /block/sda

/etc/multipath.conf

  defaults {
     user_friendly_names yes
  }
  defaults {
         udev_dir        /dev
         polling_interval 5
         default_selector        "round-robin 0"
         default_getuid_callout  "/sbin/scsi_id -g -u -s /block/%n"
         failback        immediate
  }
  blacklist {
         wwid    200d04b651805e38e
         devnode "^(ram|raw|loop|fd|md|dm-|sr|scd|st)[0-9]*"
         devnode "^hd[a-z][[0-9]*]"
         devnode "^cciss!c[0-9]d[0-9]*[p[0-9]*]"
  }
  multipaths {
         multipath {
                 wwid                    1494554000000000031343934353534303366313630633030
                 alias                   xensan
                 path_grouping_policy    failover
                 path_checker            readsector0
  }


Now after a reload of the multipath-tools and logging on the the iSCSI-targets, your multipath blockdevice will be ready for usage.
  /etc/init.d/multipath-tools reload
  /usr/bin/iscsiadm -m node -T iqn.2009-01.nl.pref:storage.xensan -p 192.168.1.250:3260 -l
  /usr/bin/iscsiadm -m node -T iqn.2009-01.nl.pref:storage.xensan2 -p 192.168.1.252:3260 -l

Lets check our new device:
  multipath -ll

The output is something like
<code>
xensan (1494554000000000031343934353534303366313630633030) dm-0 IET     ,VIRTUAL-DISK  
[size=9.8G][features=0][hwhandler=0]
\_ round-robin 0 [prio=1][active]
 \_ 1:0:0:0 sda 8:0   [active][ready]
\_ round-robin 0 [prio=1][enabled]
 \_ 0:0:0:0 sdb 8:16  [active][ready]
</code>

In addition to using multipath one could also consider to setup a host-based mirror for the shared storage. This could be accomplished by setting up two or more iSCSI-servers (targets) and join them in a software mirror (RAID-1) MD-device. This is left as an exercise for the reader as we choose to replicate our data using DRBD instead (SAN based mirroring).

====Distributed replicated block device (drbd8)====

===Install===

Binary package are available for Debian Lenny.
  apt-get install  drbd8-modules-2.6-amd64 drbd8-utils
===Config===

We want DRBD active-active replication (both our SAN nodes must be in primary mode in order to allow HA and Xen live migration). Edit /etc/drbd.conf

<code>
global {
  usage-count yes;
}
common {
  protocol C;
}
resource xensan {
  net { 
    allow-two-primaries; 
    after-sb-0pri discard-zero-changes;
    after-sb-1pri discard-secondary;
    after-sb-2pri disconnect;
  }
  startup {
    become-primary-on both;
    wfc-timeout 0;
    degr-wfc-timeout 120;
  }
  disk {
    on-io-error detach;
# You should only disable device flushes when running DRBD on devices with a battery-backed
# write cache (BBWC). Most storage controllers allow to automatically disable the write cache 
# when the battery is depleted, switching to write-through mode when the battery dies. It is 
# strongly recommended to enable such a feature.
    no-disk-flushes;
    no-md-flushes;
  }
  syncer {
    rate 300M;
  }
  on xensan1 {
    device    /dev/drbd0;
    disk      /dev/vg00/san1;
    address   192.168.1.250:7789;
    meta-disk internal;
  }
  on xensan2 {
    device    /dev/drbd0;
    disk      /dev/vg00/san2;
    address   192.168.1.252:7789;
    meta-disk internal;
  }
}
</code>

As you can see in drbd.conf we use internal metadata on our DRBD-device. The [[http://www.drbd.org|DRBD-website]] provides thorough documentation on the internals of DRBD. For metadata on an apart device change the appropriate lines to:
  ...
  meta-disk /dev/vg00/san1meta [0];
  ...
  meta-disk /dev/vg00/san2meta [0];
  ...
  
Anyway the metadata needs to be written to disk first (on both nodes).
  drbdadm create-md xensan

Choose one node as the initial master. This one will replicate its data to the other node. Data on the other node will be LOST!!!
  /etc/init.d/drbd start
  drbdadm -- --overwrite-data-of-peer primary xensan

You can monitor the progress of the replication in /proc/drbd
As soon as this process has finished, repeat this last step on the other node. Afterwards /proc/drbd should look like this:

<code>
version: 8.0.14 (api:86/proto:86)
GIT-hash: bb447522fc9a87d0069b7e14f0234911ebdab0f7 build by phil@fat-tyre, 2008-11-12 16:40:33
 0: cs:Connected st:Primary/Primary ds:UpToDate/UpToDate C r---
    ns:435488 nr:522292 dw:1229708 dr:165620 al:14 bm:85 lo:0 pe:0 ua:0 ap:0
	resync: used:0/61 hits:8209 misses:7 starving:0 dirty:0 changed:7
	act_log: used:0/127 hits:176840 misses:27 starving:0 dirty:13 changed:14
</code>

This alternative provides less detailed output:
  /etc/init.d/drbd status

Important note if you want to put iSCSI on top of a DRBD device: on Debian Lenny the package init-scripts start iSCSI before DRBD and vice versa stop DRBD before iSCSI. You're advised to change these values around.

====Xen clustering, clvm and filesystems====
As soon as two or more Xen hosts (dom0 nodes) access the same (shared) storage you'll have to be very careful with writing data. Although a cluster filesystem is not necessary for Xen live migration to work, you would indeed need a cluster filesystem if you want to write any data to the share directly from a Xen host! Furthermore you probably want to use a volume manager to manage the available storage. As LVM2 itself is not cluster-aware you would need another solution.

===Cluster LVM (CLVM)===
We use CLVM, the Cluster Logical Volume Manager that forms part of Redhat Cluster Suite (TM). This software suite, generously made available by Redhat (TM) is also available on Debian Lenny.

==Install==
Installation of the suite requires the userland tools and kernel modules.
  apt-get install redhat-cluster-suite redhat-cluster-modules-2.6.26-1-xen-amd64

==Config==
First instruct LVM to make use of the cluster locking functions, therefore change the locking type in /etc/lvm/lvm.conf to '3'.

  ...
  locking_type = 3
  ...

Then set the default properties for the cluster manager (cman) in /etc/default/cman. The nodename is FQDN, make sure that your DNS configuration and /etc/hosts are uptodate. 
<code>
CLUSTERNAME="domnull"
NODENAME="node1.example.com"
USE_CCS="yes"
CLUSTER_JOIN_TIMEOUT=300
CLUSTER_JOIN_OPTIONS=""
CLUSTER_SHUTDOWN_TIMEOUT=60
</code>

Create the cluster configuration file /etc/cluster/cluster.conf and copy it to the other cluster node.

  mkdir /etc/cluster

/etc/cluster/cluster.conf
<code>
<?xml version="1.0"?>
<cluster name="domnull" config_version="1">

<cman two_node="1" expected_votes="1">
</cman>

<clusternodes>
<clusternode name="node1.example.com" nodeid="1">
        <fence>
                <method name="single">
                        <device name="manual" ipaddr="node1.example.com"/>
                </method>
        </fence>
</clusternode>

<clusternode name="node2.example.com" nodeid="2">
        <fence>
                <method name="single">
                        <device name="manual" ipaddr="node2.example.com"/>
                </method>
        </fence>
</clusternode>
</clusternodes>

<fencedevices>
        <fencedevice name="manual" agent="fence_manual"/>
</fencedevices>

</cluster>
</code>

You might want to change the default order of the init-scripts stopping a cluster node. In general you want the gfs(2)-tools to stop first. Then the clvmd service can stop and afterwards the cman-service. Make sure that the multipath and iscsi initiator-services are running at least till the cluster-services are properly shutdown.  

Restart the cluster nodes in order to get going.
You can check the status of the cluster:
  cman_tool status

The output will be something like:
<code>
Version: 6.1.0
Config Version: 1
Cluster Name: domnull
Cluster Id: 13368
Cluster Member: Yes
Cluster Generation: 12
Membership state: Cluster-Member
Nodes: 2
Expected votes: 1
Total votes: 2
Node votes: 1
Quorum: 1  
Active subsystems: 8
Flags: 2node Dirty 
Ports Bound: 0 11  
Node name: node2.example.com
Node ID: 2
Multicast addresses: 239.192.52.108 
Node addresses: 192.168.1.242
</code>

===Simulate a cluster with two domU's within a dom0===
When two Xen guests are running on the same dom0, which provides a shared disk to both of them (in order to simulate a shared-storage cluster environment), add a bang after the 'w' in the Xen guest's configuration files to force Xen to allow you to mount the share in read-write mode (as Xen locks multiple access to the same virtual block device by default). This is not necessary if the connection to the shared storage is accesseble over iSCSI. Then you can simply initiate the iSCSI-sessions from within the Xen guests. 

The Xen guest configuration file should contain something like this:
  ...
  disk=['phy:vgxen01/lv_linva06_hda,hda,w',*'phy:vgxen01/lv_linva04_06_hdb,hdb,w!'*]
  instead of:
  disk=['phy:vgxen01/lv_linva06_hda,hda,w',*'phy:vgxen01/lv_linva04_06_hdb,hdb,w'*]
  ...

In this example you would have to use a cluster-aware filesystem like GFS/GFS2 or OCFS2 in order to prevent data corruption by the Xen guests.

===GFS/GFS2===
GFS, the Global File System is a shared disk file system that was released under the GPL by Redhat (TM) in 2004 (just a few months after they'd bought Sistina). Currently GFS2 is available as a technology preview. For production use it's probably wiser to opt for GFS.

==Install==
If you installed and configured CLVM as mentioned just above, you won't need  any extra software on Debian Lenny.

==Config==
You can just format the device (logical volume in our case) and mount it afterwards just as you would do with a more regular filesystem. You only have to add some options.
  * -p (the protocol which is probably 'lock_dlm')
  * -t clustername:filesystem_name (yes, you have to come up with a name for your fs)
  * -j n ('n' is the number of journals to create. One journal is required for each node that mounts the file system.).
The commands for GFS and GFS2 do slightly differ:

GFS
  gfs_mkfs -p lock_dlm -t domnull:xenvol2 /dev/xenvg/xenvol2 -j 2
Output
<code>
This will destroy any data on /dev/xenvg/xenvol2.

Are you sure you want to proceed? [y/n] y

Device:                    /dev/xenvg/xenvol2
Blocksize:                 4096
Filesystem Size:           720804
Journals:                  2
Resource Groups:           12
Locking Protocol:          lock_dlm
Lock Table:                domnull:xenvol2

Syncing...
All Done
</code>

GFS2
   mkfs.gfs2 -p lock_dlm -t domnull:xenvol2 /dev/xenvg/xenvol2 -j 2
Output
<code>
This will destroy any data on /dev/xenvg/xenvol2.

Are you sure you want to proceed? [y/n] y

Device:                    /dev/xenvg/xenvol2
Blocksize:                 4096
Device Size                3.00 GB (786432 blocks)
Filesystem Size:           3.00 GB (786431 blocks)
Journals:                  2
Resource Groups:           12
Locking Protocol:          "lock_dlm"
Lock Table:                "domnull:xenvol2"
</code>

===OCFS2===
Today's Linux kernel has support for the rewrited Oracle Cluster Filesystem (OCFS2). It allows several clients to read-write to the same (thus shared) storage. You could use this filesystem to share the Xen guest configuration files among several Xen hosts.

==Install==
Debian Lenny has kernel support for OCFS2 by default. You only need the ocfs2-tools (and optionally 'ocfs2console' if you want a graphical interface).
  apt-get install ocfs2-tools

==Config==
dpkg can help with some basic configuration:
  dpkg-reconfigure ocfs2-tools

Manually check/edit /etc/default/oc2b
<code>
# O2CB_ENABLED: 'true' means to load the driver on boot.
O2CB_ENABLED=true

# O2CB_BOOTCLUSTER: If not empty, the name of a cluster to start.
O2CB_BOOTCLUSTER=domnull

# O2CB_HEARTBEAT_THRESHOLD: Iterations before a node is considered dead.
O2CB_HEARTBEAT_THRESHOLD=31

# O2CB_IDLE_TIMEOUT_MS: Time in ms before a network connection is considered dead.
O2CB_IDLE_TIMEOUT_MS=30000

# O2CB_KEEPALIVE_DELAY_MS: Max. time in ms before a keepalive packet is sent.
O2CB_KEEPALIVE_DELAY_MS=2000

# O2CB_RECONNECT_DELAY_MS: Min. time in ms between connection attempts.
O2CB_RECONNECT_DELAY_MS=2000
</code>

Now create /etc/ocfs2/cluster.conf for your cluster
<code>
node:
        ip_port = 7777
        ip_address = 192.168.1.248
        number = 0
        name = xenhost1
        cluster = domnull

node:
        ip_port = 7777
        ip_address = 192.168.1.249
        number = 1
        name = xenhost2
        cluster = domnull

cluster:
        node_count = 2
        name = domnull
</code>

Restart the cluster nodes in order to get started.
You can check the status of your cluster like so:
  /etc/init.d/o2cb status

The output should be like:
<code>
Driver for "configfs": Loaded
Filesystem "configfs": Mounted
Stack glue driver: Loaded
Stack plugin "o2cb": Loaded
Driver for "ocfs2_dlmfs": Loaded
Filesystem "ocfs2_dlmfs": Mounted
Checking O2CB cluster domnull: Online
Heartbeat dead threshold = 31
  Network idle timeout: 30000
  Network keepalive delay: 2000
  Network reconnect delay: 2000
Checking O2CB heartbeat: Active
</code>

At this moment you can simply format and use your shared storage:
  mkfs.ocfs2 /dev/xenvg/xenvol1
  mount -t ocfs2 /dev/xenvg/xenvol1 /mnt/

=====Config=====
Configure the xen relocation service in /etc/xen/xend-config.sxp
   ...
  (xend-relocation-address '')
  (xend-relocation-server yes)
  (xend-relocation-port 8002)
  (xend-relocation-address '')
  (xend-relocation-hosts-allow '')
  ...

Restart xend on both nodes and make sure that port 8002 accepts connections from everywhere. Check for a LISTENER line with netstat.

Important: Time has to be synced between both nodes (See [[hannibal:ntp]]).

=====Action=====

Start a domU on node1 and migrate it to node-2 like so:
  xm migrate --live name_xen_guest node2


=====Literature=====
  * a proof of concept regarding Xen migration on Suse Linux by Novell Presales, available at  [[http://forge.novell.com/modules/xfcontent/private.php?reference_id=2736&content=/library/Xen%20live%20migration%20demo/XEN_migration_demo_1.1.pdf|http://forge.novell.com/.../XEN_migration_demo_1.1.pdf]]

  * http://www.linux1394.org

  * http://www.drbd.org

  * documentation by Jeffrey Hunter on Oracle Technology Network regarding Oracle RAC on Linux and Firewire, available at http://www.oracle.com/technology/pub/articles/hunter_rac10gr2.html

  * a thesis from Espen Braastad (University of Oslo) named 'Management of high availibility services using virtualization', May 22 2006, available at http://www.linpro.no/content/download/519/3617/file/espen_HAxen.pdf
HannibalWiki

User Tools

Site Tools

Page Tools
