World: r3wp

I can monitor some by tcp .. .but how can I use udp to monitor the 
others?

the others have some loop and wait process i guess, just insert a 
wait on an udp port as well

So, if it is doing stuff .. it will miss the udp ping ?

no, if the monitor accept some lantency; For example if a process 
is not responding after few pings.

But in the case of huge process, the client process should send himself 
a ping to the monitor to advert is doing stuffs (at some points of 
the code)

like a log file

eh ?

if you write to a log file, then that's like a ping to a monitor

But actually the monitor is a user app that they run from their desktop 
.. so it can't accept pings from the process.

but a port dialing is more efficient and has less overhead

why it can't ?

how does it know where the user app is?

Simple. Each time a targetted process start,k it send its id to the 
monitor. (hey ! Mr Monitor i'm here and my name is "toto")

it's a standard client/server exchange, each client has to establish 
a connection with the server.

So, I need a central server as well...

It's the monito actually. You have one monitor (the server) for several 
processes (the clients)

I guess it's easier to do it that way then have the user interrogate 
all the processes

the monitor open a tcp/udp port and wait for connections.

The monitor could be headless and run as a service, with a separate 
GUI process that runs in user space.

To run a rebol app as a service it has to be guiless

it's not obligatory a service.
When a client process start it try

it tries to open a connection with the monitor, it he connection 
fail, he launch himself the monitor, wait few seconds and retry to 
connect.

it he = if the

Right. A lot of software does the split process thing nowadays, sometimes 
with the GUI running in the tray (or whatever the OS version if that 
is).

if -> of

So, how does that work?  the core service launches a gui app that 
sits in the tray?

the monitor don't necesseary have the need of a gui

the GUI could be another client process

No, the monitor doesn't need a gui

That usually works by having the core service run on its own, the 
the tray app starting at user login, or on demand.

and the gui app communicates by tc to the service?

Yup, or whatever other interprocess communication model you prefer 
(DBUS?)

so need some type of encrypted tcp communications ...

Just asking as I need to scale my apps as I get more and more users

Has anyone made a /View tray app that has no default window other 
than the tray app, and maybe a options dialog?

Need to get rid of all the GUIs ..

why that ? the monitor run on the same computer than the serices 
no ?

You can start a view app in the tray ...

No, the monitor may not be running on the same computer

you want monitor the services on a local network ?

It only needs to be encrypted if you are communicating between computers. 
It's much harder to break into localhost communication (afaik).

Yes monitor services on a number of servers

Graham, it might be a good idea to have the monitor service on the 
same system as the processes it is monitoring, and then an overall 
monitoring interface that collects the info from the monitoring services 
on various computers.

Well, I hope so .. .currently I only have one server running ... 
but got to make it scaleable!

hum ok, but if you want the monitor to be able to launch/kill proccesses, 
you'll need of a monitor on each server + a CENTRAL MONITOR to deals 
with all the monitors

Right, Steeve, that's what I was saying (not as well).

So, the central monitor collates data from all the monitoring stations

so when a serice starts up, it registers with the local monitor, 
and somehow the PID is collected in case the monitor needs to kill 
it?

Yup. And if the monitoring stations have local storage for their 
data, the central monitor could poll them instead of the other way 
around. That way you can restart the central monitor.

Ok, has anyone written such monitors in Rebol yet ?  :)