[distcc] monitoring question

[Martin Pool]

On 12 Sep 2002, Stephan Stapel <stephan.stapel at web.de> wrote:
> Hi again!
> 
> > Not that I know of, beyond `less /var/log/daemon`.  It would be cool
> > if somebody wanted to.
> 
> Trust me, if I could, I would do that. However, I volunteer (or better:
> re-volunteer) for doing at least a qt/ c++ monitor implementation (I know,
> you prefer something more open like tcl/tk).

Qt/C++ would be fine.  Personally I would use Python but I really
don't mind.

> I suppose the key problem in creating such a monitoring interface is
> collecting the information from all machines, right?

The key problem, I think, is deciding what information you want to
show.  Making blinkenlights is easy but making something genuinely
meaningful is harder.

Big make invocations in general have the problem that you can't easily
tell how much is complete or how quickly it is running, except from
experience.

At the moment it seems like the most useful information is the overall
time for a remote and a local build. 

Perhaps it would be good to add something like ccache's statistics
output, to get an idea of how much time has been saved by running
distcc.  It might show the total number of invocations, the number run
remotely and locally, the CPU time locally and remotely, etc.

I think almost all of the information necessary to do a monitor or
statistics is already present in the verbose logfile, so at least the
first version ought to just parse that, rather than adding new
monitoring mechanisms.  Just a few lines of perl looking at e.g. CPU
time per job can give quite interesting information.

-- 
Martin