Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
/ 2004-06-04 14:51:22 +0200 \ Bernd Broermann: > Lars Ellenberg wrote: > >/ 2004-06-04 01:18:57 +0200 > >\ Bernd Broermann: > > > >>Hallo , > >> > >>I want correctly set up a HA-DRBD cluster , which serve data via nfs. > >> > >>I am using > >>drbd 0.6.12-1 > >>heartbeat 1.0.4-1.woody.um.1 > >>nfs-kernel-server 1.0-2woody1 > >>on a 2.4.24 kernel based debian woody . > >> > >>All is working so far. > >>When I gracefully stop Node1(master) with /etc/init.d/heartbeat stop , > >>Node2 ( slave ) is taking over, becomes primary and the nfs share is > >>still available. > >>Even when Node1 comes back, all is working. > >> > >>When I poweroff Node1, Node2 is taking over as I want as well. > >>But when Node1 comes back , heartbeat starts SyncAll from Node2 to Node1 > >>and stops NFS without waiting for for finishing the sync. Node1 starts > >>nfs while secondary. > >> > >>One have to wait all the SyncAll time until can access the cluster. > >> > >>In my /etc/ha.d/haresources there is on both nodes. > >>node1 192.168.96.65 datadisk::drbd0 nfs-kernel-server > >> > >>How can I tell heartbeat to wait with shutting down the nfs-server on > >>Node2 ( Wait to start the nfs server on Node1 ) ? > > > >this is one of the situations where drbd 0.6 is "misbehaving" ... > > Do you suggest to use drbd 0.7 insted ? not in production untill we decided to "release" it. still -pre, though in a pretty good shape meanwhile. and though with 0.7, you can have a Primary be Sync Target, this is not recommended, since if you then have a network hickup during resync, it has no choice but panic... so even with 0.7, I'd recommend to at bootup: > >don't start heartbeat while drbd is resynching. > >if you have proper order of start scripts, and proper settings in > >drbd.conf (inittimeout=-0, NO skip-wait), and start drbd before you > >start heartbeat, the drbd init script will "block" the boot process > >until drbd has finished the resychronization. then heartbeat is started, > >and can make its decision. > > > > or/and: > >tell heartbeat to *not* automatically relocate > >the services back to their "prefered" node. > >see the heartbeat config option auto_failback resp. nice_failback. lge