pcp
[Top] [All Lists]

pmproxy libpcp extensions (Re: pcp updates)

To: markgw@xxxxxxx
Subject: pmproxy libpcp extensions (Re: pcp updates)
From: Nathan Scott <nscott@xxxxxxxxxx>
Date: Wed, 24 Oct 2007 12:11:54 +1000
Cc: kimbrr@xxxxxxx, pcp@xxxxxxxxxxx
In-reply-to: <471EA6A2.1000804@sgi.com>
Organization: Aconex
References: <1193124251.24082.36.camel@edge.yarra.acx> <471E9D69.8030509@sgi.com> <1193190060.24082.61.camel@edge.yarra.acx> <471EA6A2.1000804@sgi.com>
Reply-to: nscott@xxxxxxxxxx
Sender: pcp-bounce@xxxxxxxxxxx
On Wed, 2007-10-24 at 11:57 +1000, Mark Goodwin wrote:
> 
> yes, eventually. We have multiple-firewall situations to handle with
> the new SGI clusters. The high level spec is quite simple: augment
> the hostname argument to pmNewContext to allow a chain of proxys
> separated by literal ':'. e.g. pminfo -h proxy_a:proxy_b:targethost

That syntax is inconsistent with other tools, like ssh.  We also should
cater for a port number in the syntax, so something like:

  pminfo -h target:4331@proxy:7885

would probably be more like what people expect (certainly the semi-
colon-for-port-number part, not sure about the target/proxy ordering,
and the @ separator might not be ideal, but at least its guaranteed not
to be in use in a hostname already).

The proxy chaining is really not something I see a need for, myself,
and is a separate problem to the client / PMPROXY_HOST env var issue.

> then hack pmproxy so it can follow chains. No changes required
> to any existing apps (other than pmproxy). Not sure how much of this
> is worth doing since ssh tunnels can achieve the same thing, albeit
> a lot less conveniently, but with better security options.

End point for me is I'd really like to be able plot multiple hosts in
a single kmchart chart, where each plot is from a different host behind
a different proxy.  This cant be done today, and ssh port forwarding
alone wont solve that class of problem.

cheers.

--
Nathan


<Prev in Thread] Current Thread [Next in Thread>