[lug] Setting up failover in Linux?

Will will.sterling at gmail.com
Tue Apr 24 16:36:19 MDT 2012


I should say more then one machine not two.

On Tue, Apr 24, 2012 at 4:35 PM, Will <will.sterling at gmail.com> wrote:

> Clustering software is written to so that two machines can share storage
> and IP addresses with out going haywire.  I think you are expecting a
> little to much of these packages with your application upgrade plans.
>
>
> On Tue, Apr 24, 2012 at 4:32 PM, Will <will.sterling at gmail.com> wrote:
>
>> I have never seen a long distance cluster running an application on a
>> shared file system such as DRBD.  Even clusters in the same room and on
>> shared FC storage are usually offline when their applications are updated.
>>
>> I've seen data being pushed in real time but there is always some sort of
>> fencing mechanism to prevent corruption.  Usually a database with a
>> transaction log but also SANs that replicate to a swap area and only move
>> in the changes once all of the blocks have been transmitted.
>>
>>
>> On Tue, Apr 24, 2012 at 4:24 PM, Rob Nagler <nagler at bivio.biz> wrote:
>>
>>> Hi Will,
>>>
>>> >   Clusters are run in maintenance mode during these situations so fail
>>> > over will not occur until both systems are back to a similar state.
>>>
>>> I can see how this would help prevent an accidental, automatic
>>> failover wreaking havoc.  However, I'm interested in the problem that
>>> the code and data are not part of the same transaction.
>>>
>>> Perhaps I can explain it this way, assume the primary and secondary
>>> are separated by a large enough distance that the disk blocks
>>> associated with the data and code upgrade take "a while" to copy.
>>> During that period, the primary fails.  What state is the secondary
>>> in?  Can it take over?
>>>
>>> Rob
>>> _______________________________________________
>>> Web Page:  http://lug.boulder.co.us
>>> Mailing List: http://lists.lug.boulder.co.us/mailman/listinfo/lug
>>> Join us on IRC: irc.hackingsociety.org port=6667 channel=#hackingsociety
>>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lug.boulder.co.us/pipermail/lug/attachments/20120424/e5c7c840/attachment.html>


More information about the LUG mailing list