[DRBD-user] Opteron/Xeon - Kernel 2.4/2.6 setup
Lars.Ellenberg at linbit.com
Tue Jun 15 10:21:56 CEST 2004
/ 2004-06-15 11:57:46 +0400
\ Eugene Crosser:
> On Mon, 2004-06-14 at 09:51 +0200, Lars Ellenberg wrote:
> > > I saw that the 0.7 branch should work on kernel 2.6.
> > > But would it be crazy to use this in a production environment?
> > currently that would be asking for trouble.
> > we have indication that under some circumstances drbd 0.7 may still
> > cause data inconsistencies between its mirrors [read data corruption] :(
> It was me who provided said indication.
> I've got a (crasy?) thought: could that inconsistency be caused by
> initially wrong meta-disk information, that does not get fully refreshed
> after "invalidate"? The point is that I had switched between 0.6 and
> 0.7 several times, and every time I changed 0.6 to 0.7, meta-disk
> naturally was out-of-sync with the actual data. Of course I ran
> "invalidate" every time, but maybe it is not sufficient?
nope. several thinkos in 0.7 bitmap handling, some probably where not
really thinkos, but typos. leading to several off-by-one,
off-by-factor-4, off-by-factor-8, and totally-off bugs...
anyways, debugging of that  old bitmap code has proven to be undoable,
and it has proven itself to be seriously broke.
so I implemented it from scratch now, should find its way into cvs today.
since it is new (written yesterday in one go), it may of course have new
bugs, but it is MUCH more readable and thus easier to debug...
More information about the drbd-user