[Linux-HA] IP not released for ethernet-only one-IP HA cluster

[Alexandros Papadopoulos]

Dejan Muhamedagic wrote:
> On Fri, Nov 17, 2006 at 10:12:53AM +0200, Alexandros Papadopoulos wrote:
>   
>> Dear all
>>
>> I'm running a pretty basic configuration of HA 1.2.4 on two Ubuntu 
>> 6.06.1 servers that provide various network services.
>>
>> The only connection of machine A and B to the outside world is their 
>> ethernet connection to the local subnet switch (i.e. no serial cables).
>>
>> What I'm trying to achieve is have machine A have the "clustered" IP, 
>> while machine B getting that IP when A is off the network for more than 
>> 10 seconds.
>>
>> What currently happens is that when A gets off the network (I disconnect 
>> the ethernet cable), B does get the "clustered" IP after 10 seconds, BUT 
>> machine A does not release the IP.
>>
>> As a consequence, if A is reconnected to the network without manually 
>> restarting the HA service, both A and B have the clustered IP 
>> simultaneously!
>>
>> How can I instruct HA to release the IP when it realises it cannot talk 
>> to the outside world? (i.e. the "third party" 192.168.5.1 as seen in the 
>> conf files)
>>     
>
> Choose some nodes which you can ping and configure ipfail.
>
>   
Is the information on http://www.linux-ha.org/ipfail outdated then?

<quote>
Note that /ipfail/ needs redundant communications media to work 
correctly - because it won't cause a failover on its own unless it can 
contact the other cluster member. In other words, if you're pinging on 
the same media as the only heartbeat channel configured, you're destined 
to be disappointed in /ipfail/.
</quote>

Please note I can only use one communication channel.

Cheers

-A