25 Jan
2012
25 Jan
'12
11:49 a.m.
Hi Sammy,
On 01/25/2012 02:08 PM, Sammy Govind wrote:
I've been looking for ways to create a redundant
Kamailio cluster. I've
googled alot but haven't got any concrete or final wording on any one
solution that'll just work perfectly. The basic requirement is that in
HA is never really easy :)
case of Kamailio application failure or in case of
physical machine
error the whole setup just starts working on secondary server.
Please suggest anything whichever you guys are using. Any reference URLs
would be very much appreciated.
For 1+1 setups, you can use a master/master replication (people also use
drbd to sync the whole table space, not sure how good that works) on db
level, use write-back mode in kamailio to make sure your registrations
are written to DB immediately, then use heartbeat to do the IP/process
fail-overs on network/power failures. On top, you can do manual checks
to trigger a fail-over in case of high-level failures. I guess this is
the typical and most straight-forward scenario. Your mileage might vary,
depending on specific environment variables (e.g. you might want to
announce your active server via RIP/OSPF using quagga instead of the
more common gratuitous arp if your servers are geographically split).
For scaling larger, put a kamailio instance acting as load-balancer in
front of your proxies and scale them by providing some sharding
information to the lb.
Also the new db_cassandra backend in master branch looks pretty
promising when it comes to HA/scaling. Maybe someone can provide some
best practices regarding that as well?
Andreas