[Freeswitch-users] High Availability Cluster Module for FreeSWITCH

[Marcin Gozdalik]

2013/2/12 Michael Collins <msc at freeswitch.org>:
>
>> If there was a readily available resource agent for FS it is "easy" to
>> setup Corosync (if you know hat you are doing). I didn't setup
>> Corosync to handle FS, I have OpenSIPS as SBC before FS which handles
>> HA. Of course it is not ideal, as in case of a problem current calls
>> are lost.
>
>
> Which is precisely the point - this is *HIGH* availability we're talking
> about, not "kinda quickly recovering from a problem but it's okay to lose
> the calls that were in progress as long as our VoIP comes back up within
> 5-10 seconds."
>
> We're talking about the Holy Grail of VoIP here - a fault-tolerant cluster
> of VoIP servers that can keep calls up and running even if the primary
> server goes down. Note that "primary server goes down" includes literally
> hundreds of scenarios from physical (power, network cable unplugged, HDD
> crash, etc.) to logical (user misconfiguration of FreeSWITCH or LAN
> settings, router/firewall changes, service provider goofups, FreeSWITCH core
> dump, a single sofia profile locked up but other FS is okay, etc.) I'll bet
> Eliot can name dozens of them off the top of his head. ;)
>
> I have a sneaking suspicion that not everyone in this thread is thinking the
> same thing when we say "HA". Eliot's proposed solution encompasses fault
> scenarios that some of us haven't even dreamed of. I am open to someone
> getting P + C to do everything that Eliot's mod_ha_cluster can do but I
> suspect it isn't realistic. If you can prove me wrong on that then please by
> all means do so.

The Holy Grail you're describing is probably impossible to achieve
(even Google sometimes serves you 500, just press reload to be routed
to another server :).
Maybe this discussion could continue to prepare some kind of a
blueprint describing the predicted failure scenarios and how the
proposed solution would deal with them? There are several solutions
best suited for different classes of availability you're willing to
tolerate.


-- 
Marcin Gozdalik