Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Mon, Feb 04, 2008 at 12:08:41PM +0100, Dominique Fournier wrote: > Lars Ellenberg wrote: > >On Wed, Jan 30, 2008 at 12:28:01PM +0100, Dominique Fournier wrote: > >>Hello, > >> > >>I use a drbd 8.2.4-1, on a Debian 2.6.23-1-amd64 machine. The cluster > >>should be in primary/primary mode with OCFS2 to support services on both > >>nodes. > >> > >>It work really well under normal work. > >> > >>When the drbd link drops, I would like : > >>- The first node continue to work as Primary (with Waiting For Connection) > >>- The second node go to Outdate mode. > >> > >>When the link should goes up, the Secondary do a sync diff and after I > >>can put it primary again, and restart OCFS and my services. > >> > >>I wrote a drbd-peer-outdater which know on which node it is launched. > >> > >>My problem is : > >>How can I put the second node in Outdate state immediately without > >>passing it in Secondary mode ? > > > >you can't. > > > >>I can't pass it in Secondary because OCFS is mounted and dead services > >>are connected to partition. I can't kill services nor OCFS because drbd > >>is idle. > > > >and DRBD can't do anything about that. > > > >the OCFS2 way of dealing with this sort of problems is "SMITH", > >which translates to "shoot myself in the head" (compare with "STONITH"). > >you could try to do a user space handler for these events, > >which resets the box (instead of or in addition to the outdate). > > > > > Thanks for your response. > > I dont't find any docs on this on OCFS2 but I will ask to their > mailling-list. it is nothing configurable there. the handler would be the DRBD outdate-peer handler, and _you_ would need to write one that would do the shooting. I just mentioned that this would be the typical thing OCFS2 would do. when you had a _shared_ disk (not two replicated ones providing similar semantics) and OCFS2 could no longer see the disk, it would typically panic the box. so for what you suggested above, such behaviour could possibly be implemented using DRBDs handlers, you just have to make them do that. -- : Lars Ellenberg http://www.linbit.com : : DRBD/HA support and consulting sales at linbit.com : : LINBIT Information Technologies GmbH Tel +43-1-8178292-0 : : Vivenotgasse 48, A-1120 Vienna/Europe Fax +43-1-8178292-82 : __ please use the "List-Reply" function of your email client.