[Linux-cluster] corosync issue with two interface directives

[Steven Dake]

On 07/13/2010 07:35 AM, Dan Frincu wrote:
> Tim Serong wrote:
>> On 7/13/2010 at 03:00 PM, Digimer<linux at alteeve.com>  wrote:
>>
>>> On 10-07-13 12:33 AM, Dan Frincu wrote:
>>>
>>>> It seems that the issue could also be caused because Corosync looks at
>>>> the classfull network masks of a bindnetaddr directive, in this case,
>>>> even if you have addresses with /24 netmask, Corosync thinks they're /8.
>>>> Try changing to a private class C addressing scheme, which also uses /24.
>>>>
>>>> Read more herehttp://www.corosync.org/doku.php?id=faq:configure_openais
>>>>
>>>> Regards.
>>>>
>>>
>>> That's an interesting assumption. Is there no way to specify the
>>> netmask? If not, I'll try that change.
>>>
>>
>> Corosync looks at the netmask of the *interface* matching the bindnetaddr,
>> so if you've configured eth0 as (for example) 10.1.2.3/24, the appropriate
>> bindnetaddr is 10.1.2.0.  If you'd configured eth0 as a /8, the bindnetaddr
>> would need to be 10.0.0.0, and so forth.
>>
>> Hope that helps to make everything even more confusing :)
>>
> Indeed. Just to keep the confusion level high, what happens if you have
> a bindnetaddr that matches 2 or more interfaces, for whatever reason?

If your environment includes such a configuration, the specific IP 
address that corosync should bind to is set in bindnetaddr to force a 
binding to a specific IP.  ie: ip = 192.168.1.1, bindnetaddr=192.168.1.1.

The purpose of using the NIC's netmask in the determination of Ethernet 
interfaces to bind to allow one configuration file to be used in a 
specific group of clusters on the same netmask.

> Which interface does it bind to, or does it bind to any interface? If
> you place 10.1.2.3/32, would the matching bindnetaddr be 10.1.2.3? Or

can netmask be set to 0?
[root at cast sdake]# ifconfig eth0 netmask 0
SIOCSIFNETMASK: Cannot assign requested address
SIOCGIFADDR: Cannot assign requested address
SIOCSIFBROADCAST: Cannot assign requested address

> does it have to end in 0 (mandatory) as specified in the docs?
>

The docs were recently changed to the following:

bindnetaddr
This specifies the network address the corosync executive should
bind to.  For example, if the local  interface  is  192.168.5.92
with  netmask 255.255.255.0, set bindnetaddr to 192.168.5.0.  If
the   local   interface    is    192.168.5.92    with    netmask
255.255.255.192,  set bindnetaddr to 192.168.5.64, and so forth.


> Regards,
> Dan
>> Regards,
>>
>> Tim
>>
>>
>>
>
> --
> Dan FRINCU
> Systems Engineer
> CCNA, RHCE
> Streamwide Romania
>
>
>
> --
> Linux-cluster mailing list
> Linux-cluster at redhat.com
> https://www.redhat.com/mailman/listinfo/linux-cluster