Are you fed-up of manual server installation?
Are you not a fan of managing your cluster using DVDs and
USB sticks anymore?
Are you afraid that configuration on some of your servers
in your HPC cluster might be inconsistent?
To find answers to these questions and to make your life
easier, follow this guide and enjoy being a real Sys admin!!!
xCAT tutorial for
beginners…
Table of contents
Title: How to install and configure xCAT - Tutorial
OS: CentOS
Linux release 7.8.2003 (Core)
Kernel Version: 3.10.0-1127.13.1.el7.x86_64
Software: xCAT
Software Version: 2.13.7
User Account: Use
‘sudo’ (recommended)
Reboot Required: No
Servers: 1 x
Management node & 1 x Compute node (Actual hardware or VMs i.e using
VirtualBox)
Cluster Network: 10.0.0.0/24 (Netmask: 255.255.255.0)
|
Hostname |
IP Address |
Netmask |
Role |
|
omaster |
10.0.0.11 |
255.255.255.0 |
xCAT management server |
|
compute001 |
10.0.0.101 |
255.255.255.0 |
Compute Node |
Note: For
actual hardware (servers) connect them over ethernet (ideally to a
Gigabit switch)
https://xcat-docs.readthedocs.io/en/latest/overview/index.html
1.1 Downloading xCAT
export XCAT_SOURCE_DIR=/tools/software/sources/xcat
cd ${XCAT_SOURCE_DIR}
tar xfj xcat-2.13.7.lenovo2_confluent-1.7.2_lenovo-confluent-0.7.1-el7.tar.bz2
The following procedure can be followed to install xCAT
cd ${XCAT_SOURCE_DIR}/lenovo-hpc-el7
./mklocalrepo.sh
yum –y install xCAT
Note: Make sure
the OS repositories are accessible by YUM (either online repos or local)
Verify xCAT installation.
source /etc/profile.d/xcat.sh
tabdump site
If installation is successful, the above command will return
similar outputs as it’s listed below.
#key,value,comments,disable
"domain","cluster.local",,
"blademaxp","64",,
"fsptimeout","0",,
"installdir","/install",,
"ipmimaxp","64",,
"ipmiretries","3",,
"ipmitimeout","2",,
"consoleondemand","no",,
"master","10.0.0.11",,
"forwarders","10.125.30.25,10.125.30.51,10.98.48.39",,
"nameservers","10.0.0.11",,
"maxssh","8",,
"ppcmaxp","64",,
"ppcretry","3",,
"ppctimeout","0",,
"powerinterval","0",,
"syspowerinterval","0",,
"sharedtftp","1",,
"SNsyncfiledir","/var/xcat/syncfiles",,
"nodesyncfiledir","/var/xcat/node/syncfiles",,
"tftpdir","/tftpboot",,
"xcatdport","3001",,
"xcatiport","3002",,
"xcatconfdir","/etc/xcat",,
"timezone","Europe/London",,
"useNmapfromMN","no",,
"enableASMI","no",,
"db2installloc","/mntdb2",,
"databaseloc","/var/lib",,
"sshbetweennodes","ALLGROUPS",,
"dnshandler","ddns",,
"vsftp","n",,
"cleanupxcatpost","no",,
"dhcplease","43200",,
"auditnosyslog","0",,
"xcatsslversion","TLSv1",,
"auditskipcmds","ALL",,
xCAT configuration is saved in a database which consists of many tables (command ‘tabdump’ can show a list of all tables). The default DB xCAT uses is ‘sqlite’. This can be changed to MySQL (MariaDB), PostgreSQL or others.
Different tables in xCAT DB are for different purposes. There are
a handful of common tables which need to be configured for all kind of hardware
but there are some specific tables which are only used for specific kind of
environment. For example ‘ppc’ table
will only be used when using Power PC based hardware and ‘prodkey’ table will only be used when installing MS Windows
environment.
It’s time to configure common xCAT tables. Let’s do it.
tabedit site
Make sure you edit these lines. If these lines are not present
then add new ones.
"dhcpinterfaces","10.0.0.11|enp0s3",,
"dnsinterfaces","10.0.0.11|enp0s3",,
"master","10.0.0.11",,
"nameservers","10.0.0.11",,
"domain","cluster.local",,
Note: You need to use the network
interface which is set with IP ‘10.0.0.11’. In my
case it is ‘enp0s3’
tabedit
networks
The table should be looking like
this
#netname,net,mask,mgtifname,gateway,dhcpserver,tftpserver,nameservers,ntpservers,logservers,dynamicrange,staticrange,staticrangeincrement,nodehostname,ddnsdomain,vlanid,domain,mtu,comments,disable
"cluster","10.0.0.0","255.255.255.0","enp0s3","<xcatmaster>",,"<xcatmaster>",,,,"10.0.0.151
10.0.0.175",,,,,,,"1500",,
It’s always a good idea to change ‘netname’ to a name which is easy to
understand. Also add ‘dynamicrange’ for
DHCP. Other fields are normally pre-filled.
Useful Notes: Since this is just a POC environment
and we don’t need any IPMI or other networks i.e 10G or IPoIB (Infiniband). In
real world these networks are normally required.
Add all nodes with their IP
addresses. This will help us in generating ‘/etc/hosts’
file.
"omaster","10.0.0.11",,,,
"compute001","10.0.0.101",,,,
tabedit noderes
#node,servicenode,netboot,tftpserver,tftpdir,nfsserver,monserver,nfsdir,installnic,primarynic,discoverynics,cmdinterface,xcatmaster,current_osimage,next_osimage,nimserver,routenames,nameservers,proxydhcp,syslog,comments,disable
"compute",,"xnba",,,,,,,,,,,,,,,,,,,
You can use ‘pxe’ instead of ‘xnba’ but ‘xnba’ is recommended as
this supports more options than ‘pxe’
tabedit passwd
#key,username,password,cryptmethod,authdomain,comments,disable
"system","root","$6$9FdrtZiCetX4cV2G$JJlywIQZByWwVmiaGPAW0ZSChOTU0VIa5MybqMwwj8fr9.Fg9BGrcu1Nq/PpyVBt8r4shXPxzSwi5BkdtYwZq1",,,,
Password can be in plain text or encrypted. Encrypted password can
be copied from ‘/etc/shadow’ file.
tabedit chain
#node,currstate,currchain,chain,ondiscover,comments,disable
"compute",,,,"nodediscover",,
This was the minimal table configuration. To deploy nodes an OS
ISO needs to be copied to ‘omaster’ node. Type this command to get this
done.
copycds
CentOS-7-x86_64-Everything-2003.iso
This will copy the OS into ‘/install’ directory. The ISO doesn’t have to
include ‘Everything’, normal ISO is good enough for installation.
Here you go. This is the least configuration we need to add.
1.1 Adding nodes into xCAT DB
Follow
these steps to add ‘compute001’ into xCAT DB.
nodeadd omaster
groups=mgmt
nodeadd
compute001 groups=compute,node
makehosts
makedhcp –n
makedhcp –a
makedns –n
makedns -a
Once done
check if DNS is working correctly.
host compute001
Expected
output:
compute001.cluster.local has address 10.0.0.101
Hmmm, is your
name resolution not working? Check if your ‘/etc/resolv.conf’ file is correct.
It should
look like similar to this:
search cluster.local
nameserver 10.0.0.11
nameserver 10.0.3.2
nameserver 10.125.30.25
Bottom two
lines can be different as they reflect your own environment.
It’s time to have real fun. We
need to add a new node and then deploy it with CentOS 7.
There are three ways you can
add node MAC addresses into xCAT.
1)
Add
MAC addresses manually
2)
Discovering
nodes using ‘Sequential Node Discovery’ method &
3)
Auto
discovery
1) Add MAC addresses manually
The easiest method to add
nodes in xCAT is to add their MAC addresses into ‘mac’ table.
If you are using a VM then its
pretty easy to get the MAC address of the node (in my case ‘compute001’ has MAC
address ’08:00:27:E9:BB:DB’)
To add the MAC follow this:
tabedit mac
The table should be looking
like as below:
#node,interface,mac,comments,disable
"compute001",,"08:00:27:E9:BB:DB",,
2) Discovering nodes using ‘Sequential Node Discovery’ method &
Follow these steps
·
Keep
‘compute001’ off
·
Type
‘nodediscoverstart
noderange=compute001’
on ‘omaster’ node
·
Turn
‘compute001’ on and wait
·
If
you have access to its console, keep looking. It should be able to get an IP
from DHCP and xCAT should start discovering this node.
·
Upon
success, ‘tabdump mac’ should show its MAC address automatically
added.
·
Type
‘nodediscoverls’ to see what xCAT has discovered and type ‘nodediscoverstop’ to stop the process.
3) Auto discovery
Auto discovery process will be
covered in ‘Advanced Tutorial’. Watch this space…
To deploy a system with xCAT, it
requires an OS image. OS image is just the information in xCAT DB to help xCAT
create a kickstart file (for Redhat, CentOS etc) for the installation.
When ‘copycds’ command is executed,
it automatically creates some images by default but it’s always good to create
a custom one for our purpose.
You can list all OS images using this
command:
lsdef –t osimage
Use these lines to create an OS image
named as ‘compute’
mkdef -t osimage compute imagetype=linux
osarch=x86_64 \
osname=Linux osvers=centos7.8 \
otherpkgdir=/install/post/otherpkgs/centos7.8/x86_64
\
pkgdir=/install/centos7.8/x86_64 \
pkglist=/install/templates/compute/pkglist
\
profile=compute provmethod=install \
template=/install/templates/compute/tmpl
\
partitionfile=/install/templates/compute/partfile
\
otherpkglist=/install/templates/compute/otherpkgs.pkglist
\
synclists=/install/templates/compute/synclist
Now it’s time to configure files we
have mentioned in the image (above).
Luckily xCAT installation comes with
most of these files so it’s just a matter of copying them to our custom
directory ‘/install/templates/compute’
Let’s get it done…
mkdir –p /install/templates/compute
rsync -avp
/opt/xcat/share/xcat/install/centos/compute.centos7.tmpl
/install/templates/compute/tmpl
rsync -avp /opt/xcat/share/xcat/install/centos/compute.centos7.pkglist
/install/templates/compute/pkglist
touch
/install/templates/compute/otherpkgs.pkglist
touch /install/templates/compute/synclist
Function of ‘otherpkgs.pkglist’ and
‘synclist’ files
will be covered in the advanced tutorial. For now the existence of these files
is sufficient for xCAT to install ‘compute001’.
The file ‘/install/templates/compute/partfile’ is used by xCAT to create custom
disk partitioning for nodes. This file can be modified according to the
requirements and also depending on the disk sizes.
part /boot/efi
--size=100 --fstype=exfat
part /boot --fstype=ext4
--size=500
part pv.1 --grow
--size=1
volgroup system --pesize=4096 pv.1
logvol swap --name=swap
--vgname=system --size=2048
logvol / --fstype=ext4
--name=root --vgname=system --grow --size=4096
1.4
Nodes Installation
nodeset compute001
osimage=compute
Reset ‘compute001’ and make sure it boots up over network to receive DHCP and TFTP information from xCAT server (omaster).
Wow! You
have managed to install a compute node using xCAT. That’s an achievement!
Monitor the
screen of the node and make sure it completes the installations. It should
automatically boot up and you should be able to SSH into it without any password.
For further
reading and understanding please check out my ‘xCAT Tutorial for advanced
users’ (coming soon…)
