Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Hi Bruce, As Lars pointed already out we are a behind our schedule. The current status is, that drbd-0.7 works very well on 2.4.x based kernels by now. On the Linux-2.6.x side, it compiles by now (but do not try to load the modlue :). I will do some testing, and ensure that it _does_ work on 2.4.x and release drbd-0.7-pre5 by the end of the week. ( Due to the work on the 2.6.x integreation a lot of code is in flux ) Of course it is possible that it is not that stable as drbd-0.6.10 is, but I think it is essential that you base any effort on the newer code. (Internally there are huge differences between drbd-0.6 and drbd-0.7) Does this sound reasonable ? -Philipp > Philipp, > Hope you are well and your drbd-0.7 is stable. We have just put out > 1.0 of OpenSSI and we have increased interest in having DRBD integrated > into it. I would propose a 3 step development/test effort: > a. without any changes to DRBD, try to set up a mirror of a filesystem > between say node 2 and node 3; it would be an ext3 filesystem with > CFS transparently stacked on top; try to shoot the primary node > and transparently failover the filesystem to node the other node. > This would be wonderful functionality to have; > b. Try to make the same capability work for the root filesystem, > making any necessary changes to OpenSSI or DRBD; > c. Integrate the membership capabilitites. > > One of the OpenSSI developers successfully tried an earlier version of > DRBD although I'm not sure he actually tried failing over. > > I think it would make most sense for someone from OpenSSI to initiate the > integration, with help from the DRBD team. Is now a good time > to take a snapshot and try it out? > > thanks, > bruce > > [ Charset ISO-8859-1 unsupported, converting... ] > > > Hi Burce, > > > > I think we met at that cluster workshop in ... was it Enshede ? > > > > [...] > > > > > My thought was that the use of DRBD would allow us to have HA-CFS > > > but without requiring the shared disk. Accomplishing this for non-root > > > filesystems could be a simple integration. There are callback > > > interfaces in the kernel from the membership module to each "service". > > > The membership module is called CLMS and I hope we can figure out how > > > to interface that to the DRBD code to let DRBD kernel code know when > > > nodes go up or down. > > > > Yes. DRBD, of course has a sort of a membership layer, although it is > > rather integrated. But I am sure that it will be possible to drop > > DRBD's membership layer and to rely on SSI's view of the world. > > > > > The question for the DRBD development group is whether and how we can > > > switch from the primary to the secondary copy of the data without > > > running non-kernel code (it is difficult to run non-kernel code when > > > failing over the root filesystem)? > > > > The name of the fucntion is drbd_set_state(). We just need to make > > it accessible from other parts of the kernel. -- Yes will will need > > to write a bit of glue code... > > > > > I hope the background makes it clear how we were hoping to integrate > > > and leverage the two projects. Being able to simultaneously read the > > > secondary copy while the primary is active might be a nice extension > > > but is not at all a requirement at this point. > > > > Bruce, we are currently working hard on finishing drbd-0.7, which > > is a rather big step in drbd's evolution. As soon as drbd-0.7 is > > stable (we want to be there in November), I want to > > experiment with openGFS and Oracle's physical parallel file system. > > > > As soon as we have drbd-0.7 stable I am ready to help out to > > integrate DRBD to your work. > > -- : Dipl-Ing Philipp Reisner Tel +43-1-8178292-50 : : LINBIT Information Technologies GmbH Fax +43-1-8178292-82 : : Schönbrunnerstr 244, 1120 Vienna, Austria http://www.linbit.com :