Ocfs2 users - Feb 2011 - Question about eth communication for ocfs2

Hi,

I wonder if there is a way to configure two ip_addr in the 
/etc/ocfs2/cluster.conf
so that it works on two physically distinct networks, to get the 
redundancy of
the link for ocfs2 ?
(I know there should be  the possibility to use the "bonding" 
functionnality on
Ethernet, but I would like not to use it)

And a relative question :

    if there is a network breakdown on the only communication link for 
ocfs2
    in a two-nodes cluster, what is the behavior :

    - will both nodes decide to suicide ?

    - will both nodes remain alive providing the exchanges of 
informations on disk are always working ?

    - will both nodes "hang" for ocfs2 rw on FS until network is
working
again on both nodes ?

Is there a document somewhere describing the behavior of ocfs2 facing a 
list of eventual breakdowns ?

Thanks for your response.
Alain

Hi,
> Hi,
> 
> I wonder if there is a way to configure two ip_addr in the 
> /etc/ocfs2/cluster.conf
> so that it works on two physically distinct networks, to get the 
> redundancy of
> the link for ocfs2 ?
No.
> (I know there should be  the possibility to use the "bonding" 
> functionnality on
> Ethernet, but I would like not to use it)
Why can't you use bonding? We are using bonding for our iSCSI traffic
and it works very well.
> 
> And a relative question :
> 
>     if there is a network breakdown on the only communication link
> for ocfs2
>     in a two-nodes cluster, what is the behavior :
> 
>     - will both nodes decide to suicide ?
> 
>     - will both nodes remain alive providing the exchanges of 
> informations on disk are always working ?
We have done many tests with iSCSI in our test environment, including
network connectivity loss on both - iSCSI and O2CB networks -, and our
results shows that the slave will fence and the master will recover the
slave journal. When the slave boot, it will not mount the filesystem
due to lack of network connectivity.

The role of master or slave can be any of your nodes, so you can't
predict who will fence.
> 
>     - will both nodes "hang" for ocfs2 rw on FS until network is
> working again on both nodes ?
Yes, until network connectivity is restored or the node fence timeout
configured by O2CB_IDLE_TIMEOUT_MS is reached. If you are using iSCSI,
also consider the O2CB_HEARTBEAT_THRESHOLD.
> 
> Is there a document somewhere describing the behavior of ocfs2 facing
> a list of eventual breakdowns ?
Read the section "13. Heartbeat, Quorum, and Fencing" of the OCFS2
user's guide, it has some information about it.
> 
> Thanks for your response.
> Alain
> 
> 
> 
> _______________________________________________
> Ocfs2-users mailing list
> Ocfs2-users at oss.oracle.com
> http://oss.oracle.com/mailman/listinfo/ocfs2-users
Regards,
-- 
  .:''''':.
.:'        `     S?rgio Surkamp | Gerente de Rede
::    ........   sergio at gruposinternet.com.br
`:.        .:'
  `:,   ,.:'     *Grupos Internet S.A.*
    `: :'        R. Lauro Linhares, 2123 Torre B - Sala 201
     : :         Trindade - Florian?polis - SC
     :.'
     ::          +55 48 3234-4109
     :
     '           http://www.gruposinternet.com.br

On 02/07/2011 06:14 AM, Alain.Moulle wrote:
> Hi,
>
> I wonder if there is a way to configure two ip_addr in the
> /etc/ocfs2/cluster.conf
> so that it works on two physically distinct networks, to get the
> redundancy of
> the link for ocfs2 ?
> (I know there should be  the possibility to use the "bonding"
> functionnality on
> Ethernet, but I would like not to use it)
>
> And a relative question :
>
>      if there is a network breakdown on the only communication link for
> ocfs2
>      in a two-nodes cluster, what is the behavior :
>
>      - will both nodes decide to suicide ?
In a 2 node cluster, if the commlink breaks, the larger node
number fences.
>      - will both nodes remain alive providing the exchanges of
> informations on disk are always working ?
>
>      - will both nodes "hang" for ocfs2 rw on FS until network is
working
> again on both nodes ?
No.
> Is there a document somewhere describing the behavior of ocfs2 facing a
> list of eventual breakdowns ?
The user's guide has some info.

Hi,

I have give it a try :

linkcom PtP between node1 and node2
IO traffic on OCFS2 on both sides
unplug the PtP cable from both nodes
=>both IO traffic on OCFS2 are stalled
=>after a short while, node2 self-fences,
=>whereas a short while after the self-fence of node2 , IO traffic
     on node1 starts again

So in fact, you confirm that there is no "slave or master" question
around
this self-fence, but it will always be the larger node number in
/etc/ocfs2/cluster.conf which will self-fence ?

Thanks for your confirmation.
Alain

Sunil Mushran a ?crit :
>>
>> And a relative question :
>>
>>      if there is a network breakdown on the only communication link for
>> ocfs2
>>      in a two-nodes cluster, what is the behavior :
>>
>>      - will both nodes decide to suicide ?
>
> In a 2 node cluster, if the commlink breaks, the larger node
> number fences.