[DRBD-user] resolving split brain in .8

Jérôme Augé jerome.auge at gmail.com
Tue May 22 22:49:17 CEST 2007

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

2007/5/22, Jure Pečar <pegasus at nerv.eu.org>:
> Hello,
> I have a centos 5 / heartbeat 2.0.8 / drbd 0.8 two node cluster in the testing phase. Each node has two drbd volumes mirrored to the other node.
> I have this in my drbd.conf:
>     after-sb-0pri discard-least-changes;
>     after-sb-1pri disconnect;
>     after-sb-2pri disconnect;
> After I managed to create (had to really try ;)) a split-brain scenario, I am now left with one drbd mirror fully operational and other in split-brain state, where each node claimed it as primary, so it got disconnected. After a few reboots and some further playing (I tried to invalidate one side) now I have Primary/Unknown on one side and Secondary/Unknown on the other, but it still wont connect and sync.
> What is the proper method to recover from split-brain anyway?

If you want to keep the data from the Primary/Unknown node, then tell
the Secondary/Unknown node to discard his data :

  root at primary# drbdadm connect all

  root at secondary# drbdadm -- --discard-my-data connect all

The synchronisation should start immediatly, after issuing the
discard-my-data command, with the Primary node as the SyncSource, and
the Secondary node as SyncTarget.

Jérôme Augé

More information about the drbd-user mailing list