Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Friday, 31. October 2008, Francisco José Méndez Cirera wrote: > DRBD never recovers "primary/secondary" state after Split Brain, even if > drbd.conf is configured to discard changes in younger primary (that's what > I need): You don't recover from split brain, you make sure it never occures! > For testing purposes, I unplug the ethernet cable, and because of > that, they have no comunication (so Split Brain is expected to occur when > plug again). Unplugging any cable should not lead to a split brain in a well configured cluster. Here are the things, you absolutely _must_ do, in order to have a useful system: * have redundant communication paths, including a serial cable * use dopd to outdate your secondary in case the drbd replication connection goes down A serial cable is the simplest and best way to add redundancy to your heartbeat communication path, because it will still work, even if you completely shut down the network on a node, for example by adding the wrong firewall rule. Of course, it's a good thing to have more than one network connection, too. I usually have a direct cable beween my nodes for drbd replication and a second cable for outgoing communication, which I use both for heartbeat, too. dopd makes sure that you don't get a drbd split brain, if your replication link goes down by using the remaining heartbeat communication channels to outdate the secondary, so it is neutralized until communications is restored. You have to use the newest drbd and heartbeat versions to get it working though. Also a good idea is to have a couple of ping nodes, so heartbeat can detect a communications loss with the outer world, even if the cluster communications work normally. Hope this helps. Regards, Stefan