I uncovered and interesting "feature" today and just wanted to let
others know who may get bitten by the same - ahem - feature.
I have a two node cluster and so cleverly used 
exclude 2-31
in my /etc/llttab so that lltstat output wouldn't be riddled with output
from non-existant nodes.
Turns out this is a BAD idea if you ever want to add new nodes into your
The procedure as documented in the install guide will not work. The new
nodes will not  join the  gab port 'a' membership, because the old nodes
were told never to expect that more than two nodes will be present.
Connection requests from the new nodes are ignored by the existing
(Excuse me if I am not getting terminology right here, I think it is
close though).
Yes, I updated /etc/llttab to modify the exclude to 4-31 ( I am trying
to add two new nodes). This file is only read at initial lltconfig time
anyway. All the other config files have been correctly modifed.
I opened a case and the support engineers told me I must stop gab on the
existing nodes. My situation is made worse in that we use i/o fencing.
So, now I must bring down my service group because you can't stop gab
without stopping vxfen. You can't stop i/o fencing with disk groups
This is not going over well with my customer as I assured them that I
could add the new nodes without downtime. doh!
Just thought I would pass this along. Moral of the story. Set your
exclude range  appropriatley to support any future growth.


Veritas-ha maillist  -  Veritas-ha@mailman.eng.auburn.edu

Reply via email to