Create a Hadoop Cluster easily by using PXE boot, Kickstart, Puppet and Ambari to auto-deploy nodes

This tutorial is to showcase unattended and automatic install of multiple CentOS 6.5 x86_64 Hadoop nodes pre-configured with Ambari-agents and an Ambari-server host.

After configuring automatic install of bare metal (No OS pre-installed) nodes, deploying a Hadoop cluster will be a matter of clicks. The setup uses:

  • PXE boot (for automatic OS install)
  • TFTP server (for PXE network install image)
  • Apache server (to serve the kickstart file for
    unattended install)
  • DHCP server (for assigning IP addresses for the
    nodes)
  • DNS server (for internal domain name resolution)
  • Puppet-master (for automatic configuration management
    of all hosts in the network, Ambari install included in Puppet
    manifests)
  • Ambari-master and agents (for managing Hadoop
    ecosystem deployment)

The setup assumes that the nodes are on the 192.168.0.0/255.255.255.0 network, the master is on 192.168.0.1 and its hostname is bigdata1.hdp

The domain for the network server by the configuration is hdp and
the clients are named as bigdata[1-254].hdp

Download all files here (configuration files, PXEBoot Linux image, Kickstart file and custom script for adding a node on the master).

Master Server

  • dnsmasq
    • Install and configure dnsmasq providing a basic DHCP, DNS and TFTP server. Easily configurable, ideal for internal networks
yum install dnsmasq
    • /etc/dnsmasq.conf
      • DHCP setup
      • DNS setup
      • TFTP setup
#DNS
domain-needed
bogus-priv

domain=hdp
expand-hosts
local=/hdp/ 

listen-address=127.0.0.1 
listen-address=192.168.0.1

#Local interface
interface=p4p1

#DHCP
dhcp-range=p4p1,192.168.0.100,192.168.0.200
#set default gateway
dhcp-option=p4p1,3,192.168.0.1

#DNS servers
server=192.168.0.1
server=8.8.8.8

#TFTP
enable-tftp
tftp-root=/var/lib/tftpboot
dhcp-boot=pxelinux.0

#Reserved IPs+MAC addresses will get added here by our custom script

 

  • TFTP PXE boot files at /var/lib/tftpboot downloadable with the package above, containing
    • Linux image
    • Kickstart file
      • Note config parameters below
        • Root password for each client will be hadoop135
# NFS server, CentOS image at /var/nfs
install
nfs --server 192.168.0.1 --dir /var/nfs
# Lang and keyboard
lang en_US.UTF-8
selinux --enforcing
keyboard us
skipx
# Network
network --bootproto dhcp --hostname bigdatax.hdp
# Root pw. is hadoop135
rootpw --iscrypted $1$wHydp2Aq$KmJuQFeHYTe8fMsV2tUga.
# Allow SSH
firewall --service=ssh
authconfig --useshadow --passalgo=sha256 --kickstart
timezone --utc UTC
# Configure services, add extra repos, Epel
services --disabled autofs,gpm,sendmail,cups,iptables,ip6tables,auditd,arptables_jf,xfs,pcmcia,isdn,rawdevices,hpoj,bluetooth,openibd,avahi-daemon,avahi-dnsconfd,hidd,hplip,pcscd,restorecond,mcstrans,rhnsd,yum-updatesd
repo --name="Extra Packages for Enterprise Linux" --mirrorlist=https://mirrors.fedoraproject.org/metalink?repo=epel-6&arch=x86_64
bootloader --location=mbr --append="nofb quiet splash=quiet" 
zerombr
clearpart --all --initlabel
autopart
text
reboot
# Add Packages
%packages --ignoremissing
yum
dhclient
ntp
wget
mc
@Core
epel-release
puppet
%end
%post --nochroot
exec < /dev/tty3 > /dev/tty3
#changing to VT 3 so that we can see whats going on....
/usr/bin/chvt 3
(
cp -va /etc/resolv.conf /mnt/sysimage/etc/resolv.conf
/usr/bin/chvt 1
) 2>&1 | tee /mnt/sysimage/root/install.postnochroot.log
%end
%post
logger "Starting anaconda bigdatax.hdp postinstall"
exec < /dev/tty3 > /dev/tty3
# Changing to VT 3 so that we can see whats going on....
/usr/bin/chvt 3
(
# Update local time
echo "updating system time"
/usr/sbin/ntpdate -sub 0.fedora.pool.ntp.org
/usr/sbin/hwclock --systohc

# Update all the base packages from the updates repository
yum -t -y -e 0 update

# And add the puppet package
yum -t -y -e 0 install puppet

echo "Configuring puppet"
cat > /etc/puppet/puppet.conf << EOF

[main]
vardir = /var/lib/puppet
logdir = /var/log/puppet
rundir = /var/run/puppet
ssldir = \$vardir/ssl

[agent]
pluginsync      = true
report          = true
ignoreschedules = true
daemon          = false
ca_server       = epmhubd1.hdp
certname        = epmhubdx.hdp
environment     = ambari_dev
server          = epmhubd1.hdp

EOF

# Setup Puppet to run on system reboot
/sbin/chkconfig --level 345 puppet on

# Register Puppet, download config files
/usr/bin/puppet agent --config /etc/puppet/puppet.conf -o --tags no_such_tag --server bigdata1.hdp --no-daemonize

sync

%end

 

  • Install Apache
yum install httpd
    • Move Kickstart default file to /var/www/html/kickstart
    • Start Apache
      • Check if kickstart is available at http://bigdata1.hdp/kickstart
  • Install NFS
yum install nfs-utils nfs-utils-lib
    • Edit /etc/exports and point it to your NFS shared folder
      • In our case /var/nfs
    • Download the CentOS 6.5 x86_64 DVD ISO Image to
      this location from here: http://wiki.centos.org/Download
node 'bigdata2.hdp' {

# VM-Configuration of an ambari agent that is monitored by the ambari server.

# Turn off interfering services
include interfering_services

# Install and enable ntp
include ntp

# Ensure that servers can find themselves even in absence of dns
    class { 'etchosts':
      ownhostname => 'bigdata2.hdp'
    }


    class { 'ambari_agent':
      serverhostname => "bigdata1.hdp",
      ownhostname    => "bigdata2.hdp"
    }

    # Establish ordering
    Class['interfering_services'] -> Class['ntp'] -> Class['etchosts'] -> Class['ambari_agent']

}

 

  • IPtables configuration for Internet sharing, DHCP,
    TFTP, DNS

    • Edit /etc/sysctl.conf add net.ipv4.ip_forward = 1
      for allowing Net interface MASQ to Loc (internet connection sharing from internet network interface to local – needed for getting Ambari Repos)
    • Ensure that all ports configured below have running services and are accessible otherwise install of hosts won’t work
echo 1 > /proc/sys/net/ipv4/ip_forward

# wlan0 is the external network, p4p1 is the lan, internal network
iptables -t nat -A POSTROUTING -o wlan0 -j MASQUERADE
iptables -A FORWARD -i wlan0 -o p4p1 -m state  --state RELATED,ESTABLISHED -j ACCEPT
iptables -A FORWARD -i p4p1 -o wlan0 -j ACCEPT

# allow TFTP
iptables -A INPUT -i p4p1 -s 192.168.0.0/24 -p udp --dport 69 -m state --state NEW,ESTABLISHED -j ACCEPT

# allow DHCP
iptables -A INPUT -i p4p1 -s 192.168.0.0/24 -p udp --dport 67:68 -m state --state NEW,ESTABLISHED -j ACCEPT

# allow DNS
iptables -A INPUT -i p4p1 -s 192.168.0.0/24 -p udp --dport 53 -m state --state NEW,ESTABLISHED -j ACCEPT

# allow NFS
iptables -A INPUT -i p4p1 -s 192.168.0.0/24 -p tcp --dport 1025 -m state --state NEW,ESTABLISHED -j ACCEPT

# allow HTTP
iptables -A INPUT -i p4p1 -s 192.168.0.0/24 -p tcp --dport 80 -m state --state NEW,ESTABLISHED -j ACCEPT

service iptables save 
service iptables restart 

 

  • Custom script to add a new host
    • A simple and not too elegant Bash script to add a new host
      • Remark: checking input format, hostname as parameter, removing a host is a future task, still it works as-is
    • ./script.sh 00:11:22:33:44:55 \96 parameter is the MAC
      address of a client
    • Steps
      • Reads the MAC and Counter values from counter.txt in the same directory (counter = 192.168.0.counter value of machine name and IP)
      • Adds MAC and IP to /etc/dnsmasq.conf to have a fixed, DHCP assigned IP address
      • Adds the node’s hostname to /etc/hosts file for DNS to work (dnsmasq DNS server reads entries from there)
      • Adds a TFTP entry with the MAC address for PXE boot to
        /var/lib/tftpboot/pxelinux.cfg/-01-mac
      • Adds the Kickstart file to /var/www/html with modified host and IP data for unattended install
      • Adds Puppet config files to /etc/puppet/manifests/site.pp for the specific node to intall NTP and Ambari-agent on the node
      • Increments Counter for the next host
#/bin/bash

if [ $# -eq 0 ]
  then
printf "Need a MAC address in format 00-11-22-33-44-55"
fi

macdash=$1
macpoint=`printf $macdash | sed 's/-/:/g'`
#This is to store the end of IP address of range 192.168.0.0 / 255.255.255.0
var=`cat counter.txt`

# Modify dnsmasq.conf

printf "\n#bigdata`echo $var`\n" >> /etc/dnsmasq.conf
printf "dhcp-host=`echo $macpoint`,192.168.0.`echo $var`" >> /etc/dnsmasq.conf


# Modify hosts

printf "\n192.168.0.`echo $var` bigdata`echo $var`.hdp" >> /etc/hosts

# Add MAC to TFTP

/bin/cp -rf /var/lib/tftpboot/pxelinux.cfg/copy /var/lib/tftpboot/pxelinux.cfg/01-`echo $macdash`
sed "s/kickstart/kickstart$var/g" /var/lib/tftpboot/pxelinux.cfg/01-`echo $macdash` > /var/lib/tftpboot/pxelinux.cfg/`echo $macdash`_tmp
mv -f /var/lib/tftpboot/pxelinux.cfg/`echo $macdash`_tmp /var/lib/tftpboot/pxelinux.cfg/01-`echo $macdash`

# Add Kickstart file

/bin/cp -rf /var/www/html/kickstart /var/www/html/kickstart`echo $var`
sed "s/bigdatax/bigdata$var/g" /var/www/html/kickstart`echo $var` > /var/www/html/kickstart`echo $var`_tmp
mv -f /var/www/html/kickstart`echo $var`_tmp /var/www/html/kickstart`echo $var`

# Modify Puppet Manifest

printf "\n\nnode 'bigdata`echo $var`.hdp' {\n" >> /etc/puppet/manifests/site.pp
printf "  include interfering_services\n  include ntp\n" >> /etc/puppet/manifests/site.pp
printf "  class { 'etchosts':\n    ownhostname => 'bigdata`echo $var`.hdp'\n  }\n" >> /etc/puppet/manifests/site.pp
printf "  class { 'ambari_agent':\n    serverhostname => \"bigdata1.hdp\",\n    ownhostname => \"bigdata`echo $var`.hdp\"\n  }\n" >> /etc/puppet/manifests/site.pp
printf "  Class['interfering_services'] -> Class['ntp'] -> Class['etchosts'] -> Class['ambari_agent']\n}" >> /etc/puppet/manifests/site.pp

# Restart
service dnsmasq restart

var=$((var+1))
printf $var > counter.txt

 

Clients

Clients automatically installs on boot from network interface if their MAC address have been added to the system by using the above script. Select Boot media: PXE boot when starting the machine.

Provisioning

  • Use Puppet manifests: http://docs.puppetlabs.com/learning/
    • For configuration management of your hosts. With Puppet you can batch apply system configurations and send commands to your hosts.
  • Current script installs Ambari-Agent
  • SSH login is available

 

Install Hadoop (with Ambari)

Install Ambari-Server and Agent on the master host computer with pre-configured Puppet Manifest

  • Edit /etc/puppet/environments/ambari_dev/manifest/one.pp in
    case default hostnames need to be changed
  • Run to configure Ambari repository
puppet apply /etc/puppet/environments/ambari_dev/manifest/one.pp
    • Setup and start Ambari
ambari-server setup
ambari-server start
      • Point your browser to http://bigdata1.hdp:8080
      • Log in to the Ambari Server using the default username/password: admin/admin.
      • Install

This guide will not detail Ambari and Hadoop settings. A default install
will put together a cluster of Hadoop.

Download all files here (configuration files, PXEBoot Linux image, Kickstart file and custom script for adding a node on the master).


Ambari Tutorial

  • Open browser on bigdata1:8080

 

  • Choose HDP’s version (HDP 2.1) and repository of your OS (Centos6 will use Redhat6 repo)

 

  • Add your hosts, you can use regular expressions like bigdata[0-200].hdp, choose manual registration as Ambari-Agents are already installed

 

  • Ambari will recognize all Agents in case those are properly installed and network with DNS is configured by using Dnsmasq as above

 

  • Pre-deploy screens assesses confronting packages and issues that might cause errors

 

 

 

  • Choose the Hadoop components to be deployes on your clusters

 

  • Assign masters for the services of your cluster

 

  • Assign slaves and clients for the services of your cluster

 

 

 

  • Parse through settings
  • Settings that must be changed are highlighted with red

 

 

  • Review the deployment of your cluster

 

 

  • Be patient for the deploy. It may be restarted if there are failures.

 

  • Services will get started up after the deploy

 

  • Deploy finished with no errors

 

  • Post deploy screen shows problems and error logs and a summary on the install process (hosts and services)

 

  • Ambari uses Nagios for comprehensive monitoring with an easy-to-use Hadoop cluster monitoring dashboard that shows usage and status of each services

 

Troubleshooting Hadoop deployment, cluster

  • Kickstart’s last lines might fail to run
    • Check if Puppet is installed
    • Check Puppet configuration /etc/puppet/puppet.conf both
      • On master
        • If autosign is on for new clients in the network
      • And on child nodes
        • If server and certificates are configured in the [agent] section
  • Puppet master won’t start
    • Check Puppet certificates on child nodes with
      • puppet agent –test
  • Fail to load Ambari on localhost:8080
    • Check if Ambari is installed and running
    • Check if Ambari Agent is set on nodes with server attribute at /etc/ambari-agent/conf
    • SSL cert problem is common: check if the certificates corresponds each other on master and agent, if those are not revoked or out of valid dateframe due to clock sync problems
    • Check Ambari logs on master and agents too, if there are any SQL
      errors, reinstall if any
  • Fail to register Ambari-Agents
    • Update OpenSSL
    • Check if /etc/ambari-agent
      config files have correct server settings
    • Check /var/log/ambari-agent/ logs
  • Always check all outputs of the custom script adding machines by using
    their MAC address

    • If dnsmasq is altered at /etc/dnsmasq.conf
      • Regarding DHCP
    • If tftpboot contains the MAC address at PXEboot
    • If the webserver has a corresponding kickstartx file and the
      webserver itself is accessible by the clients
    • If Puppet has an entry for the new node on configuration
  • Can’t access services of master
    • Check IPTables from remote nodes: web server, TFTP, DNS, DHCP ports
      are open and working
    • Check DNS resolution by pinging nodes from other nodes
  • SSL certificate problems are common in Puppet and Ambari (which uses
    another instance of Puppet)

    • Check if NTP is running
    • Regenerate certificates for masters and agents
  • In case of any problem
    • Check if all services are up and running
    • Check network configuration
    • Check if services are accessible from agent nodes
    • Always check logs
      • /var/log/puppet
      • /var/log/ambari-server
      • /var/log/ambari-agent
      • /var/log/messages
      • /var/log/[hadoop services]
Share on LinkedInShare on FacebookShare on Google+Tweet about this on TwitterEmail this to someone

Comments