Re: Bootstrapping

[Nate Campi]

On Thu, Feb 19, 2004 at 05:36:44PM +0100, Mark.Burgess@iu.hio.no wrote:
> 
> > We need to begin adding error checking and/or error counting into
> > cfengine.  Without that, we're kind of stuck doing post-processing outside
> > of cfengine, which is a lot of extra work.
> 
> Can you give an example. I thought the idea of convergence is that
> there is no need for such things. Maybe if you just think in the
> "right"  way you can already do this..

Each host is reponsible for its own convergence, but it can't always
accomplish it, and nobody ever knows about the problem in many cfengine
installations.

I've had a situation where my debian sid boxes have an incompatible NSS
functionality in glibc, and my static binaries from woody won't work. I
don't test against sid before deploying, because frankly this is only a
couple test systems (workstations basically). cfagent would seg fault on
me. I think cfexecd told me about it, or the syslog reports did, or
something. If cfexecd was segfaulting too, I might not have known for a
long time.

The most reliable method I can think of is to wrap cfagent with a
rudimentary script that sends error output via an SNMP trap or some
other alert mechanism. This should at least protect against a complete
meltdown in cfengine. Seems like it's asking for trouble to rely on
cfengine to tell you of troubles with cfengine (unless you used a
permanently connected socket between client and central server to
monitor error output - if the client sends errors or even just drops you
know something happened, I don't want this setup though).
-- 
Nate

This is the crucial difference between fiction and real life:
fiction must be plausible; real life has no such constraint.
    -- Kevin Kelly