Re: [Netconf] SSH question

[Andy Bierman <andy at netconfcentral.com>]

Martin Bjorklund wrote:
> Andy Bierman <andy at netconfcentral.com> wrote:
> > Hi,
> >
> >
> > You know about openssh, and the ClientAliveInterval messages
> > that it sends (keepalive at openssh.com)?  The client is
> > supposed to send data on the channel before N of these
> > arrive.
> >
> > It seems that libssh2 drops these messages.
>
> Are you sure?  It's supposed to reply with an error if it doesn't
> understand the request, since WantReply is set.

well, there's nothing for the channel to read (EAGAIN),
even though the select() says there is.  Not sure if
it is pilot error on my part or not.


> > But if the NETCONF manager wanted to answer,
> > what can it send to the agent?  whitespace?
> > an XML comment?
> >
> > Is the agent supposed to handle (discard) non-PDU messages
> > received on its channel? 
>
> This is unclear.  At least the counter inBadRpcs will be incremented.
> One school of thought says that if garbage is received, the session
> should be dropped.
>
> > What is the most interoperable
> > way to send data to the agent without actually issuing
> > an <rpc> and wasting resources?
>
> I think you have to send an rpc.  kill-session on your own session id
> should be cheap (just hope the server is 4741 compliant :)

I agree that the RFC text does not offer much wiggle room here.

I don't want to increment counters, especially error counters.
I wish there was a no-op() command in NETCONF.

A <get> may not be that cheap:

     <get>
       <filter>bogus</filter>
     </get>

  returns

     <data></data>

  right away.


> /martin

Andy