Robert Altnoeder <[email protected]> > For a cluster with nodes A, B, C, where A is the node where the snapshot > is restored to the backend device, a possible process is: > - on B, drbdadm disconnect resource > - on B, drbdadm invalidate resource > - on C, drbdadm disconenct resource > - on C, drbdadm invalidate resource > - on A, drbdadm detach resource > - on A, restore the snapshot to the backend device WITHOUT modifying > DRBD meta data > If the volume has external meta data, restore only the volume > contents, but not the meta data volume > If the volume has internal meta data, make sure the restore modifies > only the data region, but not the meta data region > - on A, drbdadm attach resource > - on B, drbdadm connect resource > - on C, drbdadm connect resource > > B and C run a full-sync from A > > Another possible process that is much easier for restoring the backend > device WITH DRBD meta data to a volume containing internal meta data is: > - on all nodes, drbdadm down resource > - on B, drbdadm create-md resource > - on C, drbdadm create-md resource > - on A, restore the snapshot to the backend device, including internal > DRBD meta data contained in the snapshot > - on all nodes, drbdadm up resource > > Note that you must recreate meta data with the correct parameters > (commonly the same parameters that were used to create meta data on A) > > B and C run a full-sync from A
thanks a lot for the advice. I will try to do some testing about it. basically I still need to destroy secondary data and re-sync. I was hoping maybe there is a tricky faster way, but it seems we need to do the hard work to keep data safe. _______________________________________________ Star us on GITHUB: https://github.com/LINBIT drbd-user mailing list [email protected] https://lists.linbit.com/mailman/listinfo/drbd-user
