On 14/02/2017 16:30, Christian Hopps wrote:
Robert Wilton <[email protected]> writes:
On 14/02/2017 14:08, Christian Hopps wrote:
Robert Wilton <[email protected]> writes:
Hi tags draft authors,
On 09/02/2017 12:28, Lou Berger wrote:
I'm personally more excited by the use of tags as additional module
meta-data accessible via yang library. But also see no reason to
preclude this possible (even if unlikely) usage.
When the idea of tags was presented as IETF, I had assumed that tags
would be versioned/managed entirely independently from the YANG modules
that the tags apply to.
Well they are called "tags" after all. Normally one applies a tag to
something (i.e., tags it. :)
Of course, I still mean that the tags are applied to the module, but
just that the place where that relationship is defined and maintained is
not in the module itself. I.e. some central locations that map from
unique module name to associated metadata tags.
A roughly equivalent example might be perhaps like CDDB, where a program
can take a CD track, and go and fetch the associated metadata from some
known location without the metadata being embedded in the CD track itself.
Sure, but that is setup that way b/c the CD data is seen as read only
right? I don't think this is even the normal way to tag things. There
are tons of examples of the opposite where the item itself is tagged
(XML attributes, social media's #hashtags, cow ears, ...).
It will be easier to change the tag on a cow ear than on a standardized
YANG module, but I like your example ;-)
It is outside my area of expertise, but I expect that most of the
meta-data associated with a cow is not attached to the cow itself, but
stored in some database somewhere. The cow ear tag is more so that you
can identify the right cow in the database.
For a while, there was a desire to organize YANG modules by their
hierarchical path location in the schema tree. My concern with this
approach, is that there needs to be sufficient foresight to get that
structure right now, because it will be very painful to change it in
future. Unfortunately things have a habit of evolving over time, and
hence choosing the right structure now such that is still the right
structure in 25 years seems somewhat unlikely.
I was thinking that tags offers a better solution to this problem, that
allows the structure to be a bit more dynamic, evolving over time. I.e.
YANG modules for features can sit at (or near to) the top level of the
schema tree, and tags can then be used to group those modules into
sensible organizations that can evolve, so that when clients are trying
to sort through all the different YANG models that are available, they
have more hope than looking at a flat list.
We in fact plan to do this with the device meta model. I believe this
was in the presentation too, but it might not have been very obvious.
In this scenario, I think that it is better if the YANG module
definitions themselves don't specify the tags because then
adding/removing/changing them is going to be a pain. If this tag
information was managed separately (e.g. in something like YANG catalog)
then it seems easier for the tags to evolve over time.
I want to be sure I understand you here. We've defined 3 places that a
tag could be defined (comes into existence), the module designer, the
vendor and the local user. When you say "module defines" are you talking
about the first case, or are you talking about where a tag resides after
it is created?
I was talking meaning the first case, where the tags are embedded in the
definition of a module, that itself is embedded in a published standard.
For the latter that was our intention with the yang
library augmentation. For the former case consider e.g., IS-IS, I think
the authors of that module know and are the appropriate folks to add
categorization tags such as "ietf:routing", "ietf:protocol",
"ietf:routing:igp" or whatever.
OK, today, it might seem obvious that the tags should be routing,
protocol, igp, etc. But in the future that choice of categorization
might not be the best, but it will be resistant to change because some
of the tag definitions are embedded in the modules themselves.
Some of your suggested tags identify areas of IETF. But what if those
areas change over time, of the associated of drafts/WGs to particular
areas changes. A YANG module could end up being tagged with a stale
area, and I doubt that anyone would want to update a standard to just
update the tags.
Lou didn't like the area tags either, so perhaps those are just poor
initial choices, but "routing" (i.e., std tag "ietf:routing") is a fine
tag I think, so maybe we need to move more to this rather than the ietf
area groupings.
I think that the WG area tags could be useful for someone that is trying
to categorize YANG models by IETF area. But my main point is that if
we decided that they were a good idea now, but not such a great idea in
5 years time, going back and changing (or removing) them is painful if
they are baked into the models.
So, I was wondering if it wouldn't be better to not allow tags to be
embedded in the modules at all. Instead, if this was maintained in a
central database, e.g. like http://www.yangcatalog.org/, then the module
authors could always email the maintainers of YANG catalog to agree with
them how the module could most usefully be tagged in the catalog. If
this needs to change over time then this would seem to be much easier to
update.
And then what happens to the modules that are deployed? I like having
some set of tags than are paired directly with the standardized module
specifically b/c they do not change.
This goes back to my questioning as to whether tags should actually
extend to the devices themselves, or just live in the manageability
layer of YANG modules.
E.g. an operator could maintain a local database of tags associated with
the modules that they use. This database would initially be sourced
from a central place, but could be refined/updated however the operator
liked. Tag based queries could be build in the operators mgmt systems,
but would be converted to equivalent regular queries (with appropriate
subtree filters) before being sent to the device.
Thanks,
Rob
We did leave things open, so you could also create a yang catalog tag
prefix and service as well.
Thanks,
Chris.
Nothing would keep the user or vendor
from removing these or adding their own tags for the same purpose as
well if they found these inappropriate or incorrect.
OK, yes. But I wasn't thinking so much of a particular vendor or
operator, but more if the industry in general (or perhaps just the
maintainers of the catalog) decided that different categorization would
be more appropriate over time.
Thanks,
Rob
But I also had not really realized that the tags information would
necessarily reach down to the devices. I.e. I hadn't envisaged Chris's
example of querying the hello-time via an IGP package tag. Instead, I
had thought of tags making a YANG catalog website more useful. E.g.
when browsing for YANG modules, be able to restrict the query to just
the modules that are tagged as "standard" + "IGP", etc.
So, I think that this draft may benefit with a bit more description of
the envisaged use cases, and also about how tags are envisaged to evolve
once they have been defined.
Well one goal I had with tags was to allow for them to be used in
the future in ways we may not have anticipated. Therefore
we specifically did not lock down how they are used or what they are
used for. That said, examples of how one might use could be helpful.
Thanks,
Chris.
Thanks,
Rob
Lou
Thanks,
Chris.
_______________________________________________
netmod mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/netmod
.
_______________________________________________
netmod mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/netmod
