[Linux-ha-dev] Contribution to Heartbeat

[Lars Ellenberg]

/ 2006-11-29 19:31:30 +0100
\ Lars Marowsky-Bree:
> On 2006-11-29T18:34:17, Lars Ellenberg <Lars.Ellenberg at linbit.com> wrote:
> 
> > > What you want is a "The remote side has been stopped" confirmation,
> > > correct?
> > > 
> > > Can't this be achieved easier like this:
> > not "easier". maybe "also".
> 
> "easier" being defined as "can be done using existing code".

> I'm trying to understand what new functionality is added; I can see that
> we're getting code implementing a feature, but does it duplicate
> something already there?

this is _no_ new functionality to heartbeat.
it is _additional communication_ channel to drbd _using_ heartbeat.

> > > In case of connection loss:
> > > 
> > > a) Freeze IO on both sides, and set the "outdated/inconsistent" bit.
> > 
> > freeze io is something related, ok.
> > but it is _different_ from what we do here.
> 
> Ok. Let's treat it as independent, but lets note that even Philipp's
> original post mentioned freeze/unthaw of IO.
> 
> (I think that in practice though, the two will always go together,
> because otherwise there's a delta during which one side might continue
> IO while the other isn't fenced yet. That sounds like a terribly bad
> idea.)

no.
if you do not stonith, we don't need to freeze io.

> > > b) Only unfreeze & clear the side which receives notification that
> > > the peer has been stopped/fenced/STONITHed? (Or when the resync has
> > > completed, of course.)
> > > 
> > > This case could very easily be handled within the m/s support we
> > > already have, and wouldn't require any new code (well, maybe a few
> > > bugfixes to the drbd RA we provide ;-), and would work for drbd7 as
> > > well as drbd8 ...?
> > 
> > you don't need to configure drbd to use some outdate-peer handler, and
> > you may configure drbd to freeze io.  then you could make that
> > fabulous resource agent "script" tell drbd to unfreeze when
> > applicable.  this is a different concept.  we can do that.
> > 
> > we want to be able to do that other stuff too.
> 
> Yes, you've said that repeatedly. But you're still arguing
> implementation, not feature/functionality/additionally covered scenario.

I don't argue heartbeat feature,
because this is no heartbeat feature.

I don't see this discussion moving forward?

> > thats basically all there is to it.
> > no fancy multi-instance multi-state multi-homed whatever thingies.
> > just an additional way to communicate between drbd peers.
> 
> Uhm. Anyone running drbd under v2 (and v1 is in maintenance mode)
> already has those "fancy whatever thingies." ;-)

which is fine.
this is not discussed here.

-- 
: Lars Ellenberg                                  Tel +43-1-8178292-55 :
: LINBIT Information Technologies GmbH            Fax +43-1-8178292-82 :
: Schoenbrunner Str. 244, A-1120 Vienna/Europe   http://www.linbit.com :