HA applications

[Lars Marowsky-Bree]

On 2001-02-06T13:34:36,
   Jeremy Hansen <jeremy@xxedgexx.com> said:

> > What ones would you like to support?
> The biggest issue I have is not that what I need isn't available, but more
> that there is 15 different things available all of which have a piece of
> the puzzle to putting to getting the "total enterprise solution", but with
> this comes a lot of mess.

Yes. I know what you mean ;-)

> For example, we have hardware level integrity checks on a shared array
> which is mainly handled by GFS/memexpd/stomithd, stomith is carried out
> using vacm and emp, vacm also acts as our terminal console server.  But
> then we need service level integrity checks and "other" hardware checks.  
> This means we have Mon and heartbeat trying to work together to cover all
> the bases.  A lot of different pieces and a lot of room for
> failure.  People have suggested we look at FailSafe, but just from what I
> know about FailSafe it seems far more complicated then it needs to
> be.  Maybe I'm wrong, it's still on my list.

Let me step in here. I think you are somewhat wrong. Yes, it is a bit too
complicated in some corners, probably related to comeing from a proprietary
heritage and only having been in the Open Source world for about half a year
now.

But it does perform almost everything of what you have to build using 10
different pieces of software right now, and tell me how elegant a design that
is? ;-) heartbeat and GFS both STONITH'ing hosts?

Sure, GFS integration could still be improved. Actually, I think that is an
important step, because I don't care much for cXFS at all. However, that
should in theory be possible within 1-2 weeks of hacking (time which I don't
have though, maybe in a few months).

There is a very nice GUI for FailSafe.

Hardware checks still need to be implemented, but that should be easily
possible.

The architecture is rather modular and can easily accomodate new features.

Seriously, I am not saying it is the end all of everything, but you should
check it out from CVS and try it ;-)

> Kimberlite seems to incorporate a lot of these functions, but Kimberlite
> currently has limitation as far as fibre channel and node limitations as
> well 

FailSafe will support fibre channel via the shared storage resource, it does
have a node limitation at 8 nodes (theoretically, up to 16, but I haven't
tested that yet). But we are accepting patches to go beyond ;-)

Sincerely,
    Lars Marowsky-Brée <lmb@suse.de>

-- 
Perfection is our goal, excellence will be tolerated. -- J. Yahl