[Linux-ha-dev] Re: cronjobs v1.2 rc1 OCF resource agent now available

[Dejan Muhamedagic]

On Fri, Jul 18, 2008 at 05:32:06PM +0200, Lars Marowsky-Bree wrote:
> On 2008-07-18T16:46:49, Dejan Muhamedagic <dejanmm at fastmail.fm> wrote:
> 
> > We are talking about different metrics. All monitor operations
> > require a fork of a shell.
> 
> ... which incidentially is also something we eventually need to fix, and
> either fork a daemon when the RA starts (which then would talk to the
> LRM directly), or have a "monitor-loop" extension which keeps running
> and simply writes "ok" every so often to stdout or "failed" when it's
> gone down.
> 
> We really need to avoid forking completely. 

I couldn't agree more, but how are we to do that. We could
deal with one class of monitoring for process existence, and NTT
already provided something similar. For other kinds of monitors,
which are more resource specific we would need some kind of
cooperation from the resource itself. That doesn't exist. And I
don't see it on anybody's todo list. Or am I missing something?

> > And that probably more than once. The leaner the shell the better the
> > performance. It also looks like, though I'm not entirely happy with
> > that, that people tend to use more rather than less monitors and to
> > schedule them very frequently (once a second? anybody?).
> 
> Well, obviously, they want them to detect failures quickly; no other
> way.
> 
> The shell scripts then also call out to many scripts

which also need an interpreter, which is why I wrote that the
shell is forked "probably more than once".

> and binaries
> anyway, so whether or not bash gets forked really doesn't matter. 
> 
> While I see your point, but I'd much rather have a RA written in bash
> than none at all for some service, and the submitter gets to chose the
> language (or shell).

That's fine, until we ship the RA and then it's up to us to
maintain it and resolve customer issues.

> > And, unfortunately, a better monitor API/interface is still far away
> > for most resource types.
> 
> It's trivial once someone did it ;-)

:-D

> Anyway, this discussion is already going on for longer than its worth,
> and I apologize for starting it. My fault, but it proves the bikeshed
> doctrine ;-)

The famous one. This time it's not, at least not for me.

Thanks,

Dejan

> 
> Regards,
>     Lars
> 
> -- 
> Teamlead Kernel, SuSE Labs, Research and Development
> SUSE LINUX Products GmbH, GF: Markus Rex, HRB 16746 (AG N?rnberg)
> "Experience is the name everyone gives to their mistakes." -- Oscar Wilde
> 
> _______________________________________________________
> Linux-HA-Dev: Linux-HA-Dev at lists.linux-ha.org
> http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
> Home Page: http://linux-ha.org/