distributing modules to machines

[r1pp3r]

I'm in the the process of designing a build system written in python.
It runs from a central server with various build machines hosting
server processes, written in Python. Pyro is the chosen RPC mechanism.
What I would like to do is have the system update itself on demand. In
other words, pass pickled objects (the code comprising the server) down
the pipeline to the server, have them copied to the right place, and
then restart the server which can then invoke the new code. Is this
feasible? Will there be any other issues involved?

[Fredrik Lundh]

r1pp3r wrote:

What I would like to do is have the system update itself on demand. In
other words, pass pickled objects (the code comprising the server) down
the pipeline to the server, have them copied to the right place, and
then restart the server which can then invoke the new code.

should work (assuming you trust the various systems involved, and your
own ability to avoid deploying broken code)

but using a custom protocol for this sounds like a slight overkill,
though. I would probably use rsync (over ssh) at regular intervals, or,
if we're only talking about small amounts of code, a bootstrap script
that fetches the current version over http(s) every time the server starts.

(in the http case, you can use etag/if-modified-since to avoid
downloading things if they haven't changed, but if you're on a fast
network, that probably won't matter much).

</F>