On 06/03/2018 12:04, Bogaert, Bart (Nokia - BE/Antwerp) wrote:
Rob,
Just to clarify: in this case we are not changing the model, it is the
same model but the device SW in release X is not supporting the
feature and in release Y it is. Supporting the feature results in
adding that part of the tree to the configuration that is related to
the feature and the data leafs related to that feature are simply not
there in the data configured when SW release X was active.
Yes, but an equivalent way of writing this without a feature, and
ignoring namespaces, would be put the feature nodes into a separate YANG
module (F) which augments the base module (B).
In release X, the device only supports B.
In release Y, the device supports B and F.
But YANG would not allow F to augment B with a mandatory node for the
reasons described previously. Hence, I think that it is probably an
oversight that YANG allows this.
Probably it should go on the YANG 2.0 issue tracker to consider and
potentially disallow this.
Thanks,
Rob
Regards, Bart
*From:*Robert Wilton [mailto:rwil...@cisco.com]
*Sent:* Tuesday, March 6, 2018 12:59 PM
*To:* Bogaert, Bart (Nokia - BE/Antwerp) <bart.boga...@nokia.com>
*Cc:* netmod@ietf.org
*Subject:* Re: [netmod] Guideline on modeling including features and
phased support by a device
Arguably the guidelines, or YANG, should say "don't allow this" ;-)
I think that what you are describing is just another instance of
"don't augment with a mandatory node rule", or "only backwards
compatible changes revisions should be made to a published YANG module".
The key reasoning behind these rules is that the a client should be
able to work unchanged after the server has been upgraded, as long as
they are not making use of any new functionality.
Thanks,
Rob
On 06/03/2018 11:34, Bogaert, Bart (Nokia - BE/Antwerp) wrote:
Hi Rob,
I agree but the fact is that some of the BBF models have
constructions like that and we were wondering whether this should
not be mentioned in the guildelines document. Normally a server
can’t set config true leafs if there is no default available in
the model. That is the reason we reached out to NETMOD. Your
suggestions can work but require adaptation of the current model.
Regards, Bart
*From:*Robert Wilton [mailto:rwil...@cisco.com]
*Sent:* Tuesday, March 6, 2018 10:38 AM
*To:* Bogaert, Bart (Nokia - BE/Antwerp) <bart.boga...@nokia.com>
<mailto:bart.boga...@nokia.com>; netmod@ietf.org
<mailto:netmod@ietf.org>
*Subject:* Re: [netmod] Guideline on modeling including features
and phased support by a device
Hi Bart,
I think that the best solution to problem is perhaps to avoid it
altogether. I.e. I don't think that the only-if-featureleaf
should be marked mandatory. Instead, it would be better to define
a sensible default value/behaviour if the leaf is absent even when
the feature is supported.
Alternatively, you can simulate something similar to an if-feature
statement by using a when or must expression instead that is
predicated on a leaf that the client must explicitly set to enable
the feature, giving control back to the client.
E.g. something along the lines of ...
leaf enable-super-feature {
if-feature test-feature;
type boolean;
default "false";
}
...
leaf only-if-feature {
when '/enable-super-feature = "true"';
type string;
mandatory true;
}
It would be interesting if you have a concrete example where
neither of the above suggestions would work or be appropriate.
Thanks,
Rob
On 05/03/2018 09:25, Bogaert, Bart (Nokia - BE/Antwerp) wrote:
Hi,
We have a question with respect to YANG models using
features. Assume that a part of the model is defined under a
feature and that this feature-dependent part defines a leaf as
mandatory.
module servers {
namespace "http://www.example.com/servers";
prefix servers;
import ietf-inet-types {
prefix inet;
}
revision 2018-03-01 {
description
"Initial version.";
}
feature test-feature {
description "testing feature";
}
container servers {
list server {
key name;
max-elements 64;
leaf name {
type string;
}
leaf ip {
type inet:ip-address;
mandatory true;
}
leaf port {
type inet:port-number;
mandatory true;
}
leaf only-if-feature {
if-feature test-feature;
type string;
mandatory true;
}
}
}
}
Now assume that we have a device that implements the model
step-wise by first not supporting this feature and in a
sub-sequent release by supporting this feature (and uses a
persistent running datastore). The question arising now is
how to deal with this mandatory leaf? Normally this can only
be configured by a client, meaning that without any “help”,
the NC server will not be able to startup with the data
contained in the device’s persistent datastore unless a value
is set for the mandatory leaf that now becomes available as a
result of supporting the feature.
When modeling as follows it seems the NC server can start with
the model supporting the feature that was not supported before:
module servers {
namespace "http://www.example.com/servers";
prefix servers;
import ietf-inet-types {
prefix inet;
}
revision 2018-03-01 {
description
"Initial version.";
}
feature test-feature {
description "testing feature";
}
container servers {
list server {
key name;
max-elements 64;
leaf name {
type string;
}
leaf ip {
type inet:ip-address;
mandatory true;
}
leaf port {
type inet:port-number;
mandatory true;
}
container only-if-feature {
presence "see if this helps";
if-feature test-feature;
leaf only-if-feature {
type string;
mandatory true;
}
}
}
}
}
Are recommendations or guidelines in place to deal with this?
Regards, Bart
_______________________________________________
netmod mailing list
netmod@ietf.org <mailto:netmod@ietf.org>
https://www.ietf.org/mailman/listinfo/netmod
_______________________________________________
netmod mailing list
netmod@ietf.org
https://www.ietf.org/mailman/listinfo/netmod