Hi,

On 02/23/2011 11:19 AM, JiaQiang Xu wrote:
> 2011/2/23 JiaQiang Xu<[email protected]>
>> Hi folks,
>> I'm trying to configure multiple interfaces with UDPU transport.
>> I have several machines, each with multiple network connections.
>> What I want to do is to let the machines heartbeat on all available network 
>> connections.
>> Since in the current config file syntax, each interface only support one 
>> bindnetaddr
>> parameter, I think I have to config multiple interfaces to let corosync 
>> listen on
>> all local net ifs.
> Just to make it clear. The network connections may be in different subnets.
>
For each interface (physical) you must have a interface (definition) in 
the corosync.conf file if you want to use that interface (physical) for 
cluster communication. It's best to have all the machines with the same 
number of interfaces (physical) configured for cluster membership and 
communication.
>> I also found in the corosync.conf manual that I have to set rrpmode option to
>> 'passive' or 'active' if I configure multiple interfaces. Which one should I 
>> choose?
None. The thing with rrp_mode is that it doesn't recover automatically 
upon restoration of a failed link, it will work (I hope) in a next major 
version of the software, but right now you might want to rely on bonding 
for redundancy.

As with which mode to use with rrp in general, I'd say active, because 
tests I've done on corosync 1.2.x show that using the passive mode would 
not switch quickly enough when the primary ring fails. The active mode 
uses both rings, thus failure of one of the rings does not bring the 
cluster down. Recovery from a failed ring is possible by running some 
script via cron (or other means) that checks the status of the ring and 
tries to reenable it once every x periods of time, this way when the 
network connection is reestablished the script will soon follow with the 
enabling of the ring, and the cluster would still rely on the other 
active ring. Some work in this direction has been done, you could search 
the archives if you're interested.

HTH,
Dan
>> I just want my heartbeat messages more fault tolerant.
>> Do I get the right way to do this? Many thanks for any suggestions.
>> Thanks,
>> -Jiaqiang
> _______________________________________________
> Openais mailing list
> [email protected]
> https://lists.linux-foundation.org/mailman/listinfo/openais

-- 
Dan Frincu
CCNA, RHCE

_______________________________________________
Openais mailing list
[email protected]
https://lists.linux-foundation.org/mailman/listinfo/openais

Reply via email to