[asterisk-users] Redundancy / Failover

[Norman Franke]

Noah,

Thanks for the input. I'm thinking the problem with the stop  
gracefully is that it would confuse the auto fail-over appliance, in  
that it would either detect the server is dead and hard switch the  
T1s or keep sending calls there which Asterisk would reject.

I'm thinking a better method may be the fail-over switch coupled with  
some logic in the client and server, perhaps using SIP NOTIFY to  
inform clients they should disconnect when idle, and reconnect to the  
specified alternate server. Once everyone is off, then taking that  
box down and upgrading. Asterisk supports SIP NOTIFY, so that may be  
the most workable.

-Norman


> Hi Norman -
>
>>> To add to what Edgar said, yes, use linux-ha.  It works nicely in
>>> combination with DRBD.  DRBD uses a dedicated network interface on
>>> each box with a crossover cable between the two.  It does a block
>>> level copy of the entire filesystem, so you have two machines  
>>> that are
>>> identical.  The you use the linux-ha heartbeat to monitor the OS and
>>> asterisk.  If anything goes wrong, it can fail over to the second
>>> machine.
>>>
>>> This is pretty easy to set up with Analog lines.  With PRI's you'd
>>> need the fonebridge or the FSV-4PFS from http://www.failsafevoip.com
>>
>> Thanks, I wasn't aware of the FSV-4PFS box. Can one switch it  
>> remotely (e.g.
>> over the network?)
>
>> From what I understand, it has its own heartbeat-type monitoring of
> asterisk.  If asterisk fails, it will automatically fail the PRI over
> to your backup machine.  Can you manually force the failover?  I think
> so, but I'm not positive.  You can ask the failsafevoip people
> directly.  I've exchanged emails with them before and they are
> knowledgeable and responsive.
>
>
>> It would be nice to have a way to gracefully switch boxes, e.g.  
>> all new
>> calls to the backup box, wait until all calls on the primary  
>> normally end,
>> and then take server down for an upgrade.
>
> If you're using heartbeat, and it's directly monitoring the asterisk
> process, you should be able to issue a "stop gracefully" command.
> That will bring asterisk down when all the calls are complete.  Then,
> heartbeat should fail over to the other machine.  Of course, if
> someone is on a long call and you've already issued a "stop
> gracefully" command, your "asterisk cluster" won't accept any new
> calls until that long call is finished.
>
>
> - Noah

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.digium.com/pipermail/asterisk-users/attachments/20070727/c81df2ba/attachment.htm