Re: [netconf] Magnus Westerlund's No Objection on draft-ietf-netconf-restconf-notif-13: (with COMMENT)

[Magnus Westerlund <[e-mail]>]

Hi,


>
>     >
>     >        o  take any existing subscription "dependency", as specified by 
> the
>     >           "dependency" leaf node in
>     >           [I-D.draft-ietf-netconf-subscribed-notifications], and use 
> the
>     >           HTTP2 stream for the parent subscription as the HTTP2 stream
>     >           dependency, [RFC7540] section 5.3.1, of the dependent
>     >           subscription.
>     >
>     >     What is not obivous to me is that just because that a subscription 
> exists
> at
>     >     the publisher that it is going over the same HTTP/2 connection and 
> thus
> there
>     >     might be nothing for the dependency to point at that is relevant 
> for the
>     >     mechanism described in RFC 7540. I didn't even find a 
> recommendation
> that the
>     >     receiver (subscriber) should actually re-use the HTTP/2 connection 
> for
> all
>     >     communication with the same publisher.
>     >  <RR> Good point, this is not spelled out.  We will add text for the
> subscriber's reuse of the HTTP2 session:
>     > for dynamic subscriptions to a specific publisher, all subscriber URI 
> GET
> requests MUST use a common HTTP2 session for a particular DSCP value.
>
>     Is that really a MUST? I would understand a RECOMMENDED to enable the
>     dependency and weighting within the subscription using the same DSCP.
>     However, what are the reasons to mandate it, and I don't believe it is
>     enforceable as the publisher can't determine that a receiver is the same
>     instance as another receiver, at least not if I assume that the
>     receivers are instances running in a virtualized environment.
> <RR2> You are correct. What about:
> for dynamic subscriptions to a specific publisher, it is recommended that 
> all
> subscriber URI GET requests use a common HTTP2 session for a particular
> DSCP value.

If you are rewriting the old sentence that had a MUST because different DSCP 
needs different TCP connections as only a single DSCP can be applied per TCP 
connection, I think you need to have these two issues separated. One for 
stating the recommendation about reusing HTTP/2 sessions for multiple 
subscriptions using the same DSCP value. A second one stating the requirement 
that you need to use different HTTP/2 sessions for each DSCP value in use.

Cheers

Magnus Westerlund