On Friday 04 June 2010, Florian Haas wrote: > On 06/04/2010 02:47 PM, Bernd Schubert wrote: > >> Remind me of the VCS repo for that lustre_server agent please? > > > > Hrmm, I still didn't post it here. An older version is in the lustre > > bugzilla: > > > > https://bugzilla.lustre.org/show_bug.cgi?id=20807 > > > > I also attach it to this mail, the "_to_be_tested" version has a better > > if condition for the Lustre umount bug, but that part is really difficult > > to test and I don't want to find out on customer system that I introduced > > a typo somewhere... > > Can you please explain your motivation for forking the Filesystem agent > for something Lustre specific? I ask because others seem to have > successfully deployed Lustre with an unmodified Filesystem RA. > > http://wiki.lustre.org/index.php/Using_Pacemaker_with_Lustre
Yes I know this page and of course, the generic Filesystem RA also works. But IMHO - The generic RA is complex, mostly due to support for OCFS2. For the Lustre RA I removed all of that complexity - The Lustre RA is specifc to linux, as Lustre so far only runs on linux, I could remove further complexity. We might think this over, once Lustre works on Solaris as well. - The Lustre RA provides more useful log messages, what it is presently doing and which step might have failed - The Lustre RA tries to figure out external journal devices - The Lustre RA knows about lustre umount bugs - The Lustre RA performs Lustre health checks to check if a LBUG came up Of course, you can implement all of that into the generic RA as well. But all those Lustre specific options would need to be put into if-conditions, which increases complexity even more. And as I said before, extX (and so lustre lfiskfs as well) can be checked with dumpe2fs if fsck needs to be run. That will not work for any other filesystem type - again more complexity. If I have the choice between duplicate and complex code, I prefer the duplicate choice.... Cheers, Bernd _______________________________________________________ Linux-HA-Dev: [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev Home Page: http://linux-ha.org/
