[DRBD-user] dual primary mode with which FS

Christian Iversen chrivers at iversen-net.dk
Tue Mar 23 22:57:39 CET 2010

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.

On 2010-03-23 17:08, Digimer wrote:
> On 10-03-23 10:02 AM, Florian Haas wrote:
>> On 2010-03-23 14:52, Michael Schwartzkopff wrote:
>>> Am Freitag, 19. März 2010 14:12:30 schrieb Andreas.Ruetten at hansenet.com:
>>>> Hi list,
>>>> first of all please answer also to my email-address because I'm not
>>>> subscribed to the ML.
>>>> Here are my questions:
>>>> I have to set up drbd 8.0.14 (Debian stable) in dual-primary mode.
>>>> One node has to read and write while the other node only need read
>>>> access.
>>>> Do I need GFS or OCFS?
>>> Not really if you can be sure that the second node ONLY reads.
>> Are you nuts? I thought you wrote books about this!
>>>> Or would a simple etx3 sufficient enough?
>>> Yes. Mount option "read-only" in the second node.
>> I repeat my above statement.
>> I'm totally baffled right now.
>> Andreas: you want to put a filesystem on dual-Primary DRBD, it has to be
>> a cluster filesystem. No ifs, buts, or maybes.
>> Question is, do you really need dual-Primary DRBD? You most probably
>> don't.
>> Florian
>  From a purely technical view point, Michael isn't wrong.

Well, he is. And so are you.

> If the second
> node never changes a single block on the DRBD partition, it can't
> technically hurt it. Of course, you'd need to make sure that the node's
> DRBD doesn't try to recover from failures without first fencing the
> other node.

Well, there's that. But the *primary* can't change anything either! Data 
would change while the filesystem expects it not to. Even if you somehow 
disable all caching (which you can't do without changing the kernel), 
you would STILL get serious data corruption errors, because of the many 
possible race conditions.

> It's not wise, but it's not impossible, either.

It is impossible, though. Unless NEITHER node ever change the data. And 
in that case, you might as well just copy it out once and for all.

Med venlig hilsen
Christian Iversen

More information about the drbd-user mailing list