[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: comments on draft-ietf-netconf-notification-01.txt



Juergen Schoenwaelder wrote:
On Wed, May 03, 2006 at 02:31:31PM -0700, Andy Bierman wrote:
I think a combined approach would be acceptable:

I think what Balazs and I are saying is that special RPCs
are fine when they are not merely replicating the functionality
of the standard operations.

This subscription info is just config data.
If it wasn't, you couldn't represent it in
the read-only data model (as in the document).

Since you can easily represent it as config data,
and since this approach allows both agent initiated
and manager initiated sessions to share the same config,
the standard operations (not special RPCs)
are the most correct choice.

I think enough people have spoken up
that we can say there is interest in
both manager and agent initiated notifications.


Andy


1) Define a data model for all the parameters that control/monitor
   the notification subscription, so that it can also be used by the
   agent without any manager subscription.  I.e., a table indexed by
   a name string containing one 'row' per parameter set.  A separate
   table for reusable filters is also needed.

   A notification target table is also needed which is completely
   independent of the session-subscription API.  Parameters such
   as the notification receiver address, user name, and user credentials
   need to be included to configure a traditional notification
   generation application.

2) Use standard operations such as <edit-config> to manipulate
   all notification parameters, even for changing profiles in use.
   <create-subscription> and all other special RPCs go away.

3) A new special RPC <start-notifications> is used
   for a manager to start receiving notifications.
   There is no subscription ID because there is only one
   stream of notifications on this session, which can be changed
   by editing the configuration.

   The <stop-notifications> RPC is used to tell the agent to
   cease notification generation for the session.

4) There is no real need for <suspend> and <resume> because
   the parameters passed in <start-notifications> are trivial.

5) mixed-mode vs. endless-RPC is very problematic.
   I could live with mixed-mode, but I don't want
   the WG coming back next year asking for a lot
   of complicated hacks to deal with the horrible
   operational characteristics of mixed-mode.
   The answer will be "don't mix operations and notifications
   if you are not prepared to deal with the delays or the lost
   notifications."

We should not compromise for the sake of a compromise.

What I am interested in is to have an easy mechanism to request that
notifications are directed to a given session (by default 'this'
session if we allow mixed content) and to have that request bound to
the lifetime of the session. In other words, I am thinking of a
subscription which establishes operational state rather than a change
of the configuration of the device and hence <edit-config> might not
be the primitive of choice.

You are looking for ways to configure persistent notificiation
receivers and I agree that this is essentially a configuration change
and so <edit-config> seems to be the appropriate primitive to use.

I think the first decision to take is whether the WG wants to support
both, one, or none of the use cases.



--
to unsubscribe send a message to netconf-request@ops.ietf.org with
the word 'unsubscribe' in a single line as the message text body.
archive: <http://ops.ietf.org/lists/netconf/>