Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Hello Martin, I think, this added complexity is only o.k., if your network connections are more likely to fail than your OSPF setup. For example, if your two nodes are in different locations, with lots of modems, media converters, switches, routers, etc. in the network path between the nodes. OSPF then allowes you to use multiple independent network paths for DRBD (and all your "resources"). But if your two nodes are just beside each other in the same rack, with one simple patch cable that can fail, i would not add OSPF. Your operators will always have to take into account OSPFs (counter?)reactions, if they decide to react on any failure. OSPF sometimes may recognize operator actions as a failure and so may void the effects of an intended operator command. But of course YMMV. Best Regards Matthias Martin Bene wrote: >>Today i tried to use DRBD on loopback interfaces with OSPF routing. >>It turns out, DRBD is not yet ready for such a setup (see below). >> >> >>First of all, please really think about wether you need and want this >>added complexity of ospf. It is a tradeoff: You add ospf as a new >>possible error source. You get some extra protection against network >>failures. > > > Thanks for the detailed information on this setup! > > I'm just trying to decide wether this is actually worth the added > complexity. > > What I actually want to get from my HA setup is protection from SPOFs; > additional protection from multiple failures is nice to have but I > wouldn't add much complexity to get it. > > Failure scenario with a standard setup (without lo/ospf): > * one of the network cards (or the cable) for the internal drbd network > dies. > Result: > * System is still up and running, but no longer tolerant to a 2nd > fault. Failover is no longer possible. > > The interesting question is: what does the admin have to do to recover > from the situation? Most other failures in a ha system leave you with a > bad system that can be shut off and repaired; here that's only the case > if you've got an active/passive system and the failure happens on the > passive/standby node. If you're running active/active (multiple drbd > devices with primaries on both nodes) or if the failed card is on the > primary it starts getting tricky. > > Solution: change drbd config to use addresses from the other network, > reconnect using these new addresses, figure out which card is bad, fail > over to the other host. Now shutdown bad node and repair, reboot. Repeat > config change to get back onto the original drbd connection. > > Not trivial and needs an admin that really knows his way around drbd. > > Compare to recovery WITH lo/ospf setup: > Failover to good node, shut down /repair node with bad NIC, reboot. MUCH > easier und less fiddling required. > > Conclusion for me: Just for the ease of recovery from a failure > situation I'll give this a try. > > Bye, Martin > _______________________________________________ > drbd-user mailing list > drbd-user at lists.linbit.com > http://lists.linbit.com/mailman/listinfo/drbd-user >