If your name server is down because of a failure (soft or hard), or even if you want to reboot this server for an upgrade, you can't resolve any addresses anymore.
Thats why you need at least two servers.
Master Slave
A second server received all modifications from the master and serves it.
For a resolver, master and slave are identical.
It is a good choice if you have a few servers or computers in the network.
Master and two slaves
If you have lots of computers or if you serve lots of requests, a better choice is to dissociate the database and the service. You edit and modify the zones files, the master send this zone informations to slaves, and those slaves serve it.
An IP alias for DNS services
It's like master and two slaves, but we don't use the ip address of slaves servers to access the DNS service. we add a service ip in one of the slaves. this ip can be managed by a cluster infrastructure like ipvs or HA-Cluster or basically by heartbeat service.
Not very different from an only master server except for the file /etc/resolv.conf that must have 2 servers entries, one for each IP. Of course it's on two differents physicals servers, it will be useless to do it in two virtual from the same server.
The zone sections are identicals, the changes are only in the options sections. we must add a allow-transfert line.
options { directory "/etc/bind";listen-on { any; };
allow-query { any; }; allow-transfer { any; }; pid-file "/var/run/named/named.pid";};The options section is identical, only the zone section must be changed. the type of the zone must be slave, and we add a new entry masters with the address of the dns master.
Usually we put primary (master) zone file in pri folder, and secondary(slave) in sec folder.
zone "my.zone.tld" IN {type slave;
file "sec/my.zone.tld.zone";
masters {
10.0.0.1;
};
};In the resolv.conf file, we have the entries of ns1 and ns2 servers, but if the first server is down, all requests will failed before resend to the second entry.
A good idea is to have a "virtual" entry for the resolver. there a many possibilities for that.
Software
Hardware
I will explain here, how to configure a virtual ip server with ucarp. Before i talk about ipvs, but i prefer talk about it another time.
Here we prepare a Master/Slave configuration (with two slaves or just one salve)
Download ucarp, and setting it is very simple.
Startup/shutdown VIP scripts
Very simple shell scripts
#! /bin/sh/sbin/ip addr add 10.0.0.252/24 dev eth0#! /bin/sh/sbin/ip addr del 10.0.0.252/24 dev eth0Start ucarp daemon
In a init script of you linux distrib, or you can create one.
you can also create a startup script, and can be launched at boot time with local init script.
#!/sbin/runscriptdepend() {need net
}start() { ebegin "Starting ucarp daemon" UCARP_REALIP=`ip add show dev ${UCARP_IF} | grep -w inet | awk 'BEGIN {RS="/"} $2 ~ /[0-9]/ { print $2}'` start-stop-daemon --start --quiet --exec ${UCARP_BIN} \ --pidfile ${UCARP_PIDFILE} -- \ --daemonize --pass=${UCARP_PASS} \ --interface=${UCARP_IF} --srcip=${UCARP_REALIP} \ --vhid=${UCARP_VID} --addr=${UCARP_VIP} \ --upscript=/etc/vip-up.sh --downscript=/etc/vip-down.sh}stop () { ebegin "Stopping ucarp daemon" start-stop-daemon --stop --retry 10 --quiet \ --pidfile ${UCARP_PIDFILE}}UCARP_BIN="/usr/sbin/ucarp"UCARP_PIDFILE="/var/run/ucarp.pid"UCARP_PASS="password"UCARP_IF="eth0"UCARP_REALIP="10.0.0.1" # Server 1#UCARP_REALIP="10.0.0.2" # Server 2UCARP_VID="1"UCARP_VIP="10.0.0.250"When ucarp is up and both all DSN are up and running, DNS request is send to ${UCARP_VIP}, if the server that host VIP crash, then the VIP will be migrate onto another server, so we can allways target a valid DNS server. The resolv.conf file should have only this VIP in server entry line.
software
hardware
software
hardware
software
hardware
software
hardware
We can see that security can be very expensive. the cheaper DNS service can be assume with a single server, but the better with at least four Linux servers or two servers and a hardware load balancer that is very expensive too.