Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
/ 2004-01-26 15:58:19 -0800 \ Brian Thomas: > My plan right now is: > > Make sure drbd/heartbeat does NOT start on boot > Bring up both systems, verify drbd is not running/synced. > Load drbd modules on each system. > > On System 2 (STALE copy): > > drbdsetup /dev/nb0 disk /dev/sda1 > drbdsetup /dev/nb1 disk /dev/sdb1 > drbdsetup /dev/nb0 net 172.16.1.101:7788 172.16.1.100:7788 B > drbdsetup /dev/nb1 net 172.16.1.101:7789 172.16.1.101:7789 B > > On System 1 (GOOD copy): > > drbdsetup /dev/nb0 disk /dev/sda1 > drbdsetup /dev/nb1 disk /dev/sdb1 > drbdsetup /dev/nb0 primary > drbdsetup /dev/nb1 primary > drbdsetup /dev/nb0 net 172.16.1.100:7788 172.16.1.101:7788 B > drbdsetup /dev/nb1 net 172.16.1.100:7789 172.16.1.101:7789 B I suggest doing it the other way around: start it on the good copy. /etc/init.d/drbd start make it primary. drbdsetup /dev/nb0 primary --human THEN start it on the "STALE". you now should see a FULL sync. you need a full sync anyways, since drbd mirrors at the block level, whereas rsync mirrors at file system level. > At this point, I'll verify the partitions are syncing as expected, and > System 1 is the primary. (If it's not I'm screwed.) Note that in this > example, whereas the sdb1 pair was nb0, now it'll be nb1. be *VERY* carefull with the meta data: it is stored in /var/lib/drbd[0-9], and is only identified by the minor number. so if you change drbd minors for your lower level devices, make sure you move the metadata files, or better *remove* them completely. in your case, it seems to me that if you are carefull, you can avoid a full sync at least on the pair of drbd that you had already running. > If everything's going ok, then on System 1 I'll crank up heartbeat > with '/etc/init.d/heartbeat start'. datadisk may complain a little about > seeing the volumes already started, but within a minute I should see > System 1 become the master heartbeat node, at which time I can safely > crank up System 2's heartbeat. > > Does all this make sense? Is there anything else I need to/can do to > make absolutely sure that System 1's copy of the data will be the > 'real' one? see the --human flag, it takes precedence over all other "generation counters" on connect time. Lars Ellenberg