Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Fri, Sep 12, 2008 at 08:36:14PM +0200, Lars Marowsky-Bree wrote: > On 2008-09-12T11:54:38, Lars Ellenberg <lars.ellenberg at linbit.com> wrote: > > > don't do that. > > won't work with any handlers. > > for the "floating peer" stuff to work in practice, > > we'd need much more work in the user land tools. > > > > while a neat idea in theory, > > it does not work out in practice. > > That's simply not true. It works, but it does not work with dopd. That > is a difference. ;-) and it does not work with _any_ handler. by default, starting with 8.2.6 iirc, we call the "before-resync-target" handler, whether that is configured or not, before we become sync target. if that does not know what node it is, it will return != 0. so resync (and connection) is aborted. > > so the whole "floating peer" stuff is a hack in itself. > > does anybody out there really use it? > > Yes, I think we have one or two customers use it to move the peers. Each > rack has access to a separate SAN. using drbd 0.7 ? > > it might be possible to get this hack sort of working > > by adding more hacks. like telling the kernel which > > "nodename" to fake when calling the user space helpers. > > > > but I don't think that would be a good idea. > > I actually think that dopd is the real hack, and drbd instead should dopd is a hack in itself. but it is not the problem here. the problem is to pretend to be someone which you are not, and rely on the fragile hope that either no-one cares, or that impersonation would somehow be propagated. but, maybe we actually will propagate it starting drbd 8.2.7. I'm not sure yet. maybe I simply remove the __DRBD_NODE__ hack instead. ;-) > listen to the notifications we provide, and infer the peer state by that > means ... ;-) yeah. I asked you before, how exactly that would look like, and so far I saw only handwaving. yes, dopd is a hack. but right now its the only thing that can do what it does, namely prevent (ok, reduce the chance of) going online with stale data. tell me how to get that done using "higher level" heartbeat/crm/pacemaker constructs, and I'm happy to do that. I'd much rather see dopd (the functionality) be implemented in "higher levels" than to have to port that to OpenAIS or whatever other low level cluster communications infrastructure there is to come. -- : Lars Ellenberg : LINBIT HA-Solutions GmbH : DRBD®/HA support and consulting http://www.linbit.com DRBD® and LINBIT® are registered trademarks of LINBIT Information Technologies GmbH __ please don't Cc me, but send to list -- I'm subscribed