Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Bernd Schubert wrote: > > This could explain it indeed, networking stops, the master node umounts and > writes its umount data to the drbd partitions and reboots, while still being > in drbd-primary configuration. Aha! This answers the question if it is an expected behaviour :) > The failover node will then uses its outdated data to be new primary. > During the next connection drbd doesn't know which node has the correct data, > since both data sets have the information of being the primary target. Even though the failover node is (manually) set to primary before the other comes up? I would expect that this takes precedence. > So, does it work when you manually umount the partitions or write a > K07_umount_drbd_partitions? I have to check that, but would expect that this will work because then drbd can exit cleanly and has the chance to notify the other node. > (We use hearbeat and have our own scripts there, that really make sure all > partitions will be unmounted, heartbeat is at K05). > Ok, we have our own scripts, too, or to be more precise, I'm currently writing them :) And since a situation like this can happen (you don't always have the chance to unmount the devices and assume that the network is up) I will include a "drbdadm connect <device>" after the "drbdadm primary <device>" in the start script, because the cluster software knows better than drbd which is primary. Regards, Peter