Say you have 6 hosts that need to have a common file or directory (the common
usage for rsync). You find that an update every N hours is good enough and
you want to code a cron job to make it happen. You'd also like any host
that is off-line to 1) not mess up the others, and 2) sync on the way back
up.
The problem that I found most often was that the _master_ host would always
be the one to fail. Murphy's law rules.
So I found a solution I like a lot, that I thought this list might want to
know about. I put a unique name in DNS, like
knowitall.foo.fred.com
that points to the present master for changes. Then I run the sync cron
job on _all_ the hosts. In the cron task I have a check at the top that
looks to see if this host is the master, if so exit the script.
If the KnowItAll host fails I move the DNS pointer and everyone else works.
When the old knowitall reboots he is a client rather than the server (and
I didn't have to change anything on the off-line host).
I run that same task at system boot before I start the service that uses
the data. So when any client is off-line for a Very Long Time then she
re-syncs at boot (even when she was just built and was _never_ in sync).
As a bonus you can use the KnowItAll record to guild the update job to
the correct host, if you need to.
A sample way to do this (on a Solairs host where /etc/hostname.* has the
IP addresses of your interfaces -- not the /etc/hosts names).
# When we are the master we don't need to rsync (to ourself)
IPLIST=`echo \`cat /etc/hostname.*\` | sed -e 's/ /|/g'`
if /usr/sbin/nslookup kia.foo.fred.com |/usr/bin/egrep "$IPLIST" >/dev/null
then
exit 0
fi
# do the rsync...
Of course this is not 100% bullet proof (we should \quote the dots in the
sed command, -e 's/\./\\\\./g'), but it works for any local network address.
--
ksb
