Re: [netmod] Confirming draft-ietf-netmod-module-tags-03 consensus call

[Robert Wilton <[e-mail]>]

On 16/11/2018 00:54, Kent Watsen wrote:
>> The servers implement the modules which can have predefined tags from
>> the module designer as well as the implementer (vendor) which literally
>> cannot come from anywhere *but* the server that implements the module.
> Predefined tags from the implementer only needs to come from the
> implementor.  Whether it is provided by the server itself or via
> some out-of-band mechanism (e.g., module catalog) seems to be the
> question.

I wasn't trying to argue that tags shouldn't be on the server.  I was 
just trying to say that the reason for doing so wasn't immediately 
intuitive to me from reading the draft.  In particular, I hadn't 
considered the use case of multiple clients coordinating via the tag 
information in the server config/state that Chris suggested.  I can see 
that having one client store the tag information on the server may make 
it a lot easier to code a different client (that might be monitoring 
devices for consistency).  E.g. client B only fetch the data for modules 
marked with a tag set by client A.

For the catalog use case, I wasn't particularly suggesting that tooling 
needs to pull tags from a catalog, although I have no issues with that.  
It was more my observation that once we start classifying modules using 
tags then an obvious use case (at least to me ;-) for that meta-data 
information is for humans to be able to search library repositories 
using tags.  E.g. in a similar way to how RPM repositories might be 
searched (https://rpmfind.net/linux/RPM/Groups.html)

I think that Chris is going to add some brief use case text to the 
draft.  With that addition, I have no further objections and I'm happy 
for document to progress.

Thanks,
Rob


>
> Of course, one might say that user-tags must be provided by the
> server itself, in order to provide an inter-client communication
> mechanism of some sort, as otherwise a single client, even if
> distributed, could keep the user-tags in its local database.
>
> Though this begs the question, would it be better for the clients
> to use a centralized service of some sort (perhaps within the
> deployment's infrastructure, assuming the user-tags are private)
> to have user-tags once per module, rather than (redundantly?) on
> each server, thus ensuring consistency as well as avoiding
> potential race conditions?  Or are these user-tags truly
> server-specific (not just module-specific)?
>
> Is it accurate to assume that two servers running identical
> software would have identical user-tags?  Of course, the servers
> might be running different software (either different releases
> for the same hardware, or different hardware, potentially from
> different vendors).  Accommodating such would complicate the
> construction of a centralized module-tagging service, though
> it could still be done.
>
> I suppose that the value of this document is not in any one of
> the use-cases, as they each seem minor and having alternate
> (potentially better) workarounds, but in the multitude of them
> all, and how a single mechanism can help.
>
>
>> This is not what I thought would hold this work up.
> My experience is that Last Calls tend to drive people to question
> basics again.  By example, it's rather common for a draft's title
> to change during a Last Call.  That said, this suitability question
> has been lingering since the day Joel kicked off the Adoption Call,
> that it is still with us seems to be the problem.
>
>
> Kent // contributor
>
>
>