Balazs Lengyel wrote:
Hello,Someone brought up the scenario that one netconf management system might supervise many (thousands) nodes. In this case requiring the manager to have 2 instead of one sessions for each node might be a problem.What is the problem actually with multi-use ? (The advantage of having fewer sessions is clear.) Even if we have multi-use sessions do we really need multiple channels ? Why ?
It is a complexity tradeoff, not a problem. I think the WG wants multi-use sessions, but I want to make sure. The point of multiple channels or connections per session is too keep notifications and rpcs separate. If they share, they are serialized. So let's say your application is dumping the entire config on a huge router-- a <get-config> that takes 28 minutes to complete. Let's also say that a one of the power supplies on that router freaks out and fries 3 daughter-cards at once, 30 seconds after this <rpc-reply> starts. That means you have a major
outage for at least 27.5 minutes before you know about it. Not exactly how async notifications are meant to be used.This is why you really don't want to do RPCs and receive notifications on the
message path, if you care about fault response time. Andy
Having a never ending RPC call that sound as a strange use of the RPC concept for me also someone told me on this list that one RPC call can only get ONE reply. Is that so ?regards Balazs Andy Bierman wrote:Hi, Phil brought up some transport issues again, that I think we should find out some opinions, one way or the other. 1) Multi-use session Is it important that a manager be able to issue <rpc> calls and receive <notification> messages within the same session? The other option (Single-use session) implies that the manager will establish multiple single purpose sessions instead. As Phil pointed out, if single-use is okay, then the manager can simply make a 'start-receiving-notifications' RPC call that never finishes (usually until the manager shuts down the session). 2) Multiple channels and Multiple Connections Should we try to bring back this concept? Wasn't this the main reason we picked BEEP in the first place? Is the BEEP mapping that relevant without it? This issue is only relevant if multi-use sessions are required. 3) Application Acks and Reverse RPCs As Phil and I pointed out in emails, a notification with an application-ack is really just another <rpc> method, but from the agent to the manager. It might be simpler to define how to reverse the flow for notification purposes, rather than redefining the <rpc> and <rpc-reply> elements. Should this feature be dropped, capability-based, or mandatory? If kept, how should it be done? 4) Any More Transport Issues?? Andy -- 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/>
-- 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/>