[distcc] distcc problems...

[Dan Kegel]

Martin Pool wrote:
>>>On a network of a few machines I would like the monitor to globally
>>>view all the running jobs by all users on all machines.  Other people
>>>might want to monitor that too.  That probably implies either a
>>>broadcast/multicast setup, or a central daemon that retransmits the
>>>notifications.
>>
>>Because of trust issues, I prefer to avoid broadcast techniques.

I should mention also that I care about compile clusters that
are not all located on the same LAN, so broadcast isn't sufficient by itself.

> It worries me too.  At the very least I wouldn't want it to be on by
> default.  Consider compiling on a laptop connected to a cafe wireless
> network.
> 
> Perhaps monitoring across machines is not a good idea after all.
> 
> On the other hand if the monitoring is purely local it might be
> simplest just to stick with state files.

Let's keep thinking about it.  I do think that anyone with
permission to use the distcc cluster ought to be able to
have an overview of its status of some sort.

>>By the way, it occurs to me that a local daemon would also be of some use.
>>e.g. it could hold long-lived ssh connections to the compute servers
>>(to save on socket startup time),
> 
> 
> That would be good.  It's not so much socket startup as crypto
> negotiation (DH and so on). 
> 
> I think this can be done independent of distcc.

Yep, and I bet it's being done somewhere.

>>and it could cache knowledge of remote machine status.

That's what the generic proxy can't do.  Ssh proxing
really would be a small part of what the local daemon might do.
I can imagine it might be convenient and useful to
move some of the logic currently in the distcc wrapper
into a local daemon, which by virtue of being long-lived,
would be able to do the same thing better and/or faster.

- Dan

-- 
My technical stuff: http://kegel.com
My politics: see http://www.misleader.org for examples of why I'm for regime change