[DRBD-user] DRBD and LVM Snapshot with 2 nodes configuration

Lars Ellenberg Lars.Ellenberg at linbit.com
Wed Apr 7 23:54:25 CEST 2004

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


/ 2004-04-07 22:20:28 +0200
\ Andreas Semt:
> >That is my lack of understanding:
> >
> >What happens with nodeB if the heartbeat AND drbd connection to nodeA 
> >(who is Master/primary) get lost? Can I do anything useful with nodeB
> >(and the data on it) or means that for heartbeat/drbd "nodeB is 
> >completely broken, repair it!"? Because the situation stated below can 
> >happen, or not?
> >
> > 
> >  --------                                                -------
> > |       |    ethernet / serial wire                     |       |
> > | nodeA | ---------------------------/ /--------------- | nodeB |
> > |       |                                               |       |
> >  -------                                                 -------
> > 
> Sorry, the graphic should show:
> 
> --- means: ethernet/serial connection
> 
> nodeA (active)------/<broken connection>/------nodeB(standby)

well, the DATA (unmounted file system) on BLOCK DEVICE is in a
*consistent*, but -- unless you are very lucky -- not in a *clean* state
[1].
Similar to a hard reset when you have only one node.

So, to mount it, you need at least a journal replay.
Otherwise, it is just fine :)

[1]
 consistent and clean from the file system point of view.
 The applications runnig on top of it (SQL databases and so on)
 Probably have even more problems to provide not only some
 consistent, but a *clean* view of their data...

 So actually, to have a clean snapshot with lvcreate, it won't even
 suffice to do rely on its "file system flush, suspend, and resume"
 semantics, but strictly speaking, one would need to do the same for all
 services running on it, too. (for the SQL example, something like
 "lock all; flush tables; unlock all;" ... yes, I know, pseudo syntax...)

	lge



More information about the drbd-user mailing list