Hello list,
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 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.
Thanks and Regards, Sammy
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
Hi, Thanks for taking out time for this. really appreciate your ideas. Here are few things I've found useful so far.
http://kamailio.org/events/2011-fosdem/p_usrloc.pdf This looks like one thing that one could have to ensure more reliability but we do need other application level and physical level failover techniques additionally.
For using any technique like Linux-HA, Heartbeat, keepalived, UCARP or any floating/virtual IP techniques I really hope that the sysctl option mentioned in b/m thread really works. http://www.mail-archive.com/sr-users@lists.sip-router.org/msg03714.html
http://old.nabble.com/Kamilio-Failover-Options-td21947502.html DNS SRV is one option. Using another Kamailio on top of two Kamailio is again a SPOF so I think its of no use at all.
I hope to get a very reliable and working solution out of these forums.
Regards, Sammy
On Wed, Jan 25, 2012 at 9:49 PM, Andreas Granig agranig@sipwise.com wrote:
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
SIP Express Router (SER) and Kamailio (OpenSER) - sr-users mailing list sr-users@lists.sip-router.org http://lists.sip-router.org/cgi-bin/mailman/listinfo/sr-users
-
Andreas Granig -
Sammy Govind