[Linux-HA] Stonith and other questions (Bug 1927)

[Adrian Chapela]

Andrew Beekhof escribió:
> On Thu, Jun 26, 2008 at 17:08, Adrian Chapela
> <achapela.rexistros at gmail.com> wrote:
>   
>> Hello,
>>
>> I want to continue talking about this bug
>> http://developerbugs.linux-foundation.org/show_bug.cgi?id=1927
>>
>> (It is not a bug... but I still have some doubts)
>>
>>     
>>> For some operation, you have specified on_fail=fence.
>>> How can the cluster fence the node if you have stonith turned off?
>>>       
>> Yes I have specified fence because I want to stop, poweroff, or any to do a
>> failover to another node when a master node fails.
>>     
>
> Except we can't because you don't have the stonith-enabled cluster
> option set to true.
>
>   
>> I don't know how I can activate a stonith. Is
>> it possible activate
>> a stonith with heartbeat software ?
>>     
>
> http://www.linux-ha.org/ConfiguringStonithPlugins
>   
I think there isn't enough information about configuration ( I think..). 
I want to depend off nodes of the cluster only. I think a posibility 
could be ssh, but I can't know how. Where is the information about ssh ?
>   
>>> Yes, you need to wait for 10 minutes _and_ enable stonith.
>>> Without stonith we have no way to clean up after a failed stop.
>>>       
>> The problem isn't in stopping operation, is in slaving operation,
>>     
>
> In that case, as long as stop can clean up after a failed demote, then
> stonith isn't strictly necessary.
>   
What can I do to solve the failed demote ? I need to inform heartbeat 
when a master is failed and in this case heartbeat "demotes" the failed 
master but it can't be a slave. A failed master in mysql (and I think in 
other services as well) can't be a slave, if it fails, it can't be a slave.

If I make a master failed in a valid slave, what do heartbeat then ?
>   
>> but I
>>     

What do you think about the next ?
>> think I need to change RA but Why is heartbeat trying to change the role of
>> a failed master node ?? If a node is failed, this node should become
>> unusable or it could be a stopped resource, I think it could be more
>> reasonable this situation
>>
>>     
>>> Either that of find out why its taking so long to stop the resource
>>>       
>> It is not taking so long to stop, it is being a slave...
>>     
>
> the same thing applies
>
>   
>> but is in this
>> operation when it is failing. In other hand I need a high timeout because my
>> server is doing many mysql transactions and it could take about 5 minutes to
>> stop the mysql server (Mysql Server is using 10 GB of RAM, it must commit
>> and write remain memory data to disk).
>>
>> I want to achieve a solution of High availability of MySQL, many people is
>> trying the same.
>>
>> Thank you!
>>
>> _______________________________________________
>> Linux-HA mailing list
>> Linux-HA at lists.linux-ha.org
>> http://lists.linux-ha.org/mailman/listinfo/linux-ha
>> See also: http://linux-ha.org/ReportingProblems
>>
>>     
> _______________________________________________
> Linux-HA mailing list
> Linux-HA at lists.linux-ha.org
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems
>
>