Hi, Kneeoh!
I think that the only solution that would work properly was your first
approach. However, since this is not yet implemented, all I can think of
is an external process that periodically test if the node is up. If it
is not, unsubscribe it and re-subscribe the second node.
PS: I haven't really used haproxy so I have no idea how it works.
Best regards,
Razvan Crainea
OpenSIPS Core Developer
http://www.opensips-solutions.com
On 06/06/2014 06:15 PM, Kneeoh wrote:
Hi Razvan, thank you for the reply. I don't necessarily need expiration of
subscriptions to rabbit or the timer route per se. I'm just trying to figure
out (with the existing capabilities) how to make opensips fail to another
member in the rabbit cluster in the event that the current node dies. My first
thought was that you could simply stack entry points like:
subscribe_event("E_ACC_CDR",
"rabbitmq:rabbitmq:[email protected];rabbitmq:[email protected];rabbitmq:[email protected]/cdr1")
However, it sounds like that's not in the present implementation of the rabbit
module.
So my second thought was to trick opensips and put HAProxy between it and
Rabbit, which works, but if I fail an HAProxy via corosync to the other HAproxy
something with the subscription breaks. Since it looked like the two options
were either put the subscribes in the startup route (only happens once so
probably won't failover) OR use the timer route to subscribe (which is what i'm
doing) I figured that in the event of an HAProxy failure, I might miss a few
messages but on the next timer fire opensips would resubscribe to haproxy which
would relay that to the appropriate rabbit server (I haven't failed over any
rabbit servers in this scenario so haproxy2 is talking to the same rabbit
server as haproxy1. All i'm doing is killing haproxy1 right now and letting the
VIP go to haproxy2). However it doesn't look like this is working and I can't
tell if its because the subscription isn't happening, OR it is happening but
opensips sees it already exists in the
subscribers list and does nothing (I think this is the case). If this IS the
case perhaps a solution would be to kill the subscriber entry on new subscribe.
If I'm way off, let me know, I'd really like to figure this out. Am I going
about this all wrong? How would you handle a rabbit node failure?
Regards
_______________________________________________
Users mailing list
[email protected]
http://lists.opensips.org/cgi-bin/mailman/listinfo/users
