Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Hello Todd, Todd Denniston wrote: > At least with the 0.6.X variants of DRBD, you could set drbd up to panic the > kernel when the lower level device was lost ... sort of a fencing mechanism. > It also guaranties all the services, heartbeat is configured for, move to > the other node instead of hanging on the crippled node. Yes, this is in 0.7.x, too, and i've given it some thought. This is also how other shared-storage systems work (at least netware 6 does, it abends on disk error, which is basically a panic()). I'm aware of the fact that my setup is worthless if it can't survive the panic() case, so this may be a viable workaround. However, I find the continue-as-diskless-client-method a lot more elegant, because it's a lot less intrusive for both the server processes and their respective clients. I also expect disk io-errors to occur more often than other faults, so I would really like to use the detatch mode, in wich, if I understand it correctly, all disk requests are served from the other nodes' harddrive. And since we are still in the testing phase I would like to track down this (alleged) bug and maybe even give something back to the community (in form of a patch) :-) But first, of course, I need someone to tell me that drbd should do what I expect and this is really a bug and not some sort of misunderstanding on my side. thanks for your input, Nils