Steven S. Critchfield wrote:
----- "Richard Thomas" <[email protected]> wrote:
I couldn't find an answer to this, possibly because it would be a
retarded thing to do but I was just wondering...
Say you had mirrored (say hardware) RAID with two drives. Reboot the
system with one drive unplugged, make some changes then reboot with
the
other unplugged (and the unplugged one plugged back in), make some
different changes then reboot with both drives plugged back in, what
is
the expected outcome (or is it controller dependent).
In clustered filesystems, this is referred to as split brain. It is
why most clustered filesystems want you to have 3 machines minimum.
For clustered systems, consider 3 machines and each can talk to the other
2. If one drops off line, the 2 still able to talk can be authoritative.
The one that can not talk to any of the others knows it is the one that
is isolated and shouldn't respond to filesystem requests. When it can
reconnect it can then get updated and reconnect. If you have 2 machines,
you can not determine if the other machine is dead or the network is down
locally or remotely.
The problem you introduce that isn't covered in most failover problems
is the idea of rebooting to make the changes. If you hotswap out a drive
the machine knows the drive was removed and when it returns it knows to
rebuild. You don't have a condition of 2 drives in "current" condition
and needing the other to be rebuilt.
Most likely the solution would be to pick on to be rebuilt from and do
it at the bios level. Otherwise boot from the drive you wish to be correct
and then introduce the redundant drive after the controller has established
a authoritative drive for the mirror to rebuild against.
Yeah, hot swap avoids the ambiguity. I guess if the controller maintains
any information internally, it would probably sync the newest drive to
the oldest. Though it also might synchronize the second drive to the
first. Though more likely it should throw an error and require the user
to pick. Note that the scenario presented, both drives were already
members of the array so, I guess, theoretically, neither should be
regarded as more authoritative than the other.
What made me think of this was I was pondering usage of mirrors as a
rollback device. Pop one drive, do potentially dangerous changes, if
they go wrong, just boot off the popped drive and rebuild to the one
with the bad config.
Again, hotswap should solve the issue.
Though thinking about it, removing the one drive should remove it from
the array, thus the drive which is added back to the array second should
be the one overwritten (Though, possibly, any drive added to the array,
even the first one, should be considered "blank" but it is an option to
clear or not on my controller).
I may try an experiment if I can dig out two old drives.
--
You received this message because you are subscribed to the Google Groups
"NLUG" group.
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to
[email protected]
For more options, visit this group at
http://groups.google.com/group/nlug-talk?hl=en
