Monitoring/inventory client-server app

[snorble]

I'm writing a tool for monitoring the workstations and servers in our
office. I plan to have a server and a client service that runs on each
workstation and reports back to the server (heartbeat, disk free
space, etc).

So far I am considering XMLRPC, or a client service that just
downloads a Python file and runs it.

With XMLRPC I don't know how to easily add features without having to
update every client. Also while playing with XMLRPC I learned that
when you run a registered function, it runs it on the server. I was
hoping it would run on the client, so that when I get the machine's
computer name (or disk space, etc) it will return the client's info.
It seems with XMLRPC I would have to hard code the functionality into
the client (i.e. client gets it's computer name, then calls the XMLRPC
function to pass it to the server)? I was hoping it would work more
like, "pass some code to the client to be run on the client, and
report it to the server". Almost XMLRPC in the reverse direction.

With the download-and-run approach, it seems trivially easy to add new
functionality to the clients. Just save the updated Python file to the
server, and clients download it and run it.

Are there any standard approaches to problems like this that can be
recommended? Thank you.

 

[Irmen de Jong]

I'm writing a tool for monitoring the workstations and servers in our
office. I plan to have a server and a client service that runs on each
workstation and reports back to the server (heartbeat, disk free
space, etc).

So far I am considering XMLRPC, or a client service that just
downloads a Python file and runs it.

With XMLRPC I don't know how to easily add features without having to
update every client. Also while playing with XMLRPC I learned that
when you run a registered function, it runs it on the server. I was
hoping it would run on the client, so that when I get the machine's
computer name (or disk space, etc) it will return the client's info.
It seems with XMLRPC I would have to hard code the functionality into
the client (i.e. client gets it's computer name, then calls the XMLRPC
function to pass it to the server)? I was hoping it would work more
like, "pass some code to the client to be run on the client, and
report it to the server". Almost XMLRPC in the reverse direction.

With the download-and-run approach, it seems trivially easy to add new
functionality to the clients. Just save the updated Python file to the
server, and clients download it and run it.

Are there any standard approaches to problems like this that can be
recommended? Thank you.

The security implications are HUGE when you are thinking about
transferring and executing arbitrary code over the network. Avoid this
if at all possible. But if you can be 100% sure it's only trusted stuff,
things are not so grim.

Have a look at Pyro, or even Pyro Flame:

http://packages.python.org/Pyro4/
http://packages.python.org/Pyro4/flame.html

Flame allows for very easy remote module execution and a limited way of
transferring code to the 'other side'.

Also what is wrong with running an XMLrpc server, or Pyro daemon, on
your client machines? This way your central computer can call registered
methods (or remote objects in case of Pyro) on the client and execute
code there (that reports all sorts of stuff you want to know). Or have
each client call into a central server, where it reports that stuff
itself. Many ways to skin a cat.

Regards,
Irmen de Jong

 

[snorble]

The security implications are HUGE when you are thinking about
transferring and executing arbitrary code over the network. Avoid this
if at all possible. But if you can be 100% sure it's only trusted stuff,
things are not so grim.

Have a look at Pyro, or even Pyro Flame:

http://packages.python.org/Pyro4/http://packages.python.org/Pyro4/flame.html

Flame allows for very easy remote module execution and a limited way of
transferring code to the 'other side'.

Also what is wrong with running an XMLrpc server, or Pyro daemon, on
your client machines? This way your central computer can call registered
methods (or remote objects in case of Pyro) on the client and execute
code there (that reports all sorts of stuff you want to know). Or have
each client call into a central server, where it reports that stuff
itself. Many ways to skin a cat.

Regards,
Irmen de Jong

I'm thinking maybe the client service will have a small number of
generic features, such as reading WMI and SNMP values. That way the
server still dictates the work to be done (i.e. XMLRPC returns which
WMI/SNMP values to query), and the client remains relatively focused
and straightforward.

 

[Alec Taylor]

Maybe take a look outside python:

- Puppet

I'm thinking maybe the client service will have a small number of
generic features, such as reading WMI and SNMP values. That way the
server still dictates the work to be done (i.e. XMLRPC returns which
WMI/SNMP values to query), and the client remains relatively focused
and straightforward.