Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
/ 2004-05-24 16:19:21 +0200 \ Helmut Wollmersdorfer: > Philipp Reisner schrieb: > > > Look for "load-only" "skip-wait" and negative init timeouts. > > ACK. This is suitable, if availabilty has more priority than data. > > Sorry for insisting, but let's think about an example of 3 pairs of > services/drbd-resources: > > - service1/drbd1 is configured for human intervention and human > intervention is actually necessary. > - service2/drbd2 is configured for human intervention and human > intervention is actually not necessary. > - service3/drbd3 can be Primary anyways, because "outdated" is > acceptable. > > As far as I understand the init-script, it will wait at drbd1 for human > input, and drbd2, drbd3 get no chance. Independent, if and what failover > software is used. > > I think, other drbd-resources should work independent from the problems > of only _one_ drbd-resource. though you can configure it differently, and some do this... the *typical* config is two nodes, a few devices. if one device cannot see the peer, the others won't either, obviously. so for the "typical" deployment this is a non-issue. we need to adress this somewhen, when a crazy crisscross five node cluster setup becomes more common ... Lars Ellenberg