Re: [netmod] Question on RFC8342 + RESTCONF extension (draft-ietf-netconf-nmda-restconf)

[Robert Wilton <[e-mail]>]

Hi Markus,

On 11/12/2018 17:21, Seehofer, Markus wrote:
> Hello Robert,
>
> comments inline below.
>
> BR
>   Markus
>
> -----Ursprüngliche Nachricht-----
> Von: Robert Wilton [mailto:[e-mail]]
> Gesendet: Dienstag, 11. Dezember 2018 17:06
> An: Seehofer, Markus
> Cc: [e-mail]
> Betreff: Re: [netmod] [EXTERNAL] Re: Question on RFC8342 + RESTCONF extension (draft-ietf-netconf-nmda-restconf)
>
> Hi Seehofer,
>
> Please see inline ...
>
> On 11/12/2018 14:55, Seehofer, Markus wrote:
>> Hello Juergen,
>>
>> see my comments inline below. As being quite new to the topic, going through all the old and current RFCs and drafts is quite challenging.
>> So please apologize for "simple" questions or ones maybe already raised.
>>
>> -----Ursprüngliche Nachricht-----
>> Von: Juergen Schoenwaelder
>> [mailto:[e-mail]]
>> Gesendet: Dienstag, 11. Dezember 2018 15:33
>> An: Seehofer, Markus
>> Cc: [e-mail]
>> Betreff: [EXTERNAL] Re: [netmod] Question on RFC8342 + RESTCONF
>> extension (draft-ietf-netconf-nmda-restconf)
>>
>>
>>
>> On Tue, Dec 11, 2018 at 02:17:07PM +0000, Seehofer, Markus wrote:
>>> Hello,
>>>
>>> Reading RFC 8342 along with draft-ietf-netconf-nmda-restconf-05 I've some questions or comprehension problems with the text.
>>>
>>>
>>> 1.       RFC 8342 (NMDA)
>>> Chapter 5.3.  The Operational State Datastore (<operational>) says:
>>> "The operational state datastore (<operational>) is a read-only datastore ...."
>>> Chapter 6.2. Invocation of Actions and RPCs says:
>>> "Actions are always invoked in the context of the operational state datastore. The node for which the action is invoked MUST exist in the operational state datastore."
>>>
>>> Chapter 3.1 in https://urldefense.proofpoint.com/v2/url?u=https-3A__tools.ietf.org_pdf_draft-2Dietf-2Dnetconf-2Dnmda-2Drestconf-2D05&d=DwIBAg&c=Bg5XULDZR8GiOSSWNlpkCsRePnGDkKcI6oYL9xv1MnQ&r=2XEKVYkQjmLHi2TOJp1VSzieLZVqewIpj-RxmRgPfsM&m=CDK28X8kO2wRVDla97P_WqBIhF3OAzbWSGEmTeLdxwU&s=BSyY-GBuzzDry6oIhg-mgwMAySbhJEO9Im3Z4PD_c_4&e= says:
>>> "YANG actions can only be invoked in {+restconf}/ds/ietf-datastores:operational."
>>>
>>> Question: How can one invoke an action in a as read-only defined datastore? Or am I missing something?
>> Why do you expect that a datastore has to be writable in order to
>> invoke an action? RFC 7950 has the example of a ping action tied to an
>> interface. (You ping a remote system from that specific interface.) In
>> general, actions are understood as being tied to real resources and
>> hence to the operational datastore. (For example, you can't ping from
>> an interface that is configured but not physically present.)
>>
>> [MSEE]: I do not expect that a datastore has to be writeable to invoke an action, but I do expect that a "read-only" datastore is not writeable and RFC 8342 says clearly operational state datastore is "read-only".
> RPCs and actions don't modify the operational state datastore as such, instead they modify the properties of the underlying system, and the operational state datastore provides a read-only view onto the state of the system.  So <operational> is only being updated as a side effect of reflecting the changes to the underlying system.
>
> This contrasts with writable configuration datastores (e.g. <candidate> or <running>), where the client can modify the configuration in those datastores directly which will then attempt to change the behavior of the system as the config is applied.
>
> [MSEE]: I agree but I'm still stuck with the following text.
>                 - draft-ietf-netconf-nmda-restconf-05 says in Chapter 3.1: "YANG actions can only be invoked in {+restconf}/ds/ietf-datastores:operational" with
>                    "The resource {+restconf}/ds/ietf-datastores:operational refers to the operational state datastore" and "The operational state datastore (<operational>) is a read-only datastore"
>                 - RFC 8040 says in Chapter 3.6 " Operation resources representing YANG actions are not identified in this subtree, since they are invoked using a URI within the '{+restconf}/data' subtree" and
>                   "An action is invoked as: POST {+restconf}/data/<data-resource-identifier>/<action>"
>
>                 So without NMDA it was clear, invoke an action using {+restconf}/data}. With NMDA what is the correct way to trigger an action as the draft says "YANG actions can only be invoked in {+restconf}/ds/ietf-datastores:operational"?

So, you invoke it on the equivalent path using the operational datastore.

E.g. to take the example from RFC 8040 3.6.2, the request pre NMDA is this:

       POST /restconf/data/example-actions:interfaces/\
          interface=eth0/get-last-reset-time HTTP/1.1
       Host: example.com
       Accept: application/yang-data+json

The equivalent path on a NMDA server is this:

       POST /restconf/ds/ietf-datastores:operational/\
          example-actions:interfaces/\
          interface=eth0/get-last-reset-time HTTP/1.1
       Host: example.com
       Accept: application/yang-data+json

I think that your concern might be: why is it right/OK to invoke a 
action on what is defined as a read only structure?

The answer to this is that the action isn't acting on <operational>, it 
is acting on the underlying system and has the remit to change any of 
the system behavior.

However, before the action can be processed, the system must validate 
that the data node that it is acting on exists, and the parameters are 
valid.  This validation is performed against the systems operational 
state, and hence is validated against <operational>.

Again, considering the example above, it wouldn't make sense to get the 
last reset time of an interface that existed in configuration, and hence 
was present in <running>/<intended>, but was not in <operational>.

In future we may need actions that are validated against other 
datastores, but when the authors were discussing this it was unclear 
whether proper use cases for such actions exist, and hence we decided 
that this could be deferred to future work, which would presumably come 
along with a concrete use case.

>>> 2.       The NMDA is a huge step forward for NC and RC, thanks for that. NMDA in combination with the new RESTCONF extensions let one now select one of the named datastores
>>> in RFC 8342. Reading the RFC and draft I'm still missing (or even more overlook I guess) the following. RFC 8040 Chapter 4.5 says:
>>> "A PUT on the datastore resource is used to replace the entire
>>> contents of the datastore...". So does this mean one can have the same behavior as in NETCONF where you can copy the "running" config to "startup" or "candidate" config to "running" if a RESTCONF server would support them? Is there any example how this would look like if it is allowed?
>> A PUT does not really get you there, to copy a datastore to another you want an operation on the server.
>>
>> [MSEE]: Exactly this is what I want. NETCONF specifies this clearly in the RFCs with <copy-config> but how does one trigger this with RESTCONF? I had the hope with NMDA + RESTCONF extensions this would
>>                  be possible too. Or do I still miss something?
> I think that it is theoretically possible to invoke the NETCONF RPCs (e.g. the copy-config RPC defined in ietf-netconf.yang, RFC 6241) from RESTCONF (e.g. section 3.6 of RFC 8040).
>
> Whether this is actually a good thing to do/encourage I'm not so sure.
>
> [MSEE]: OK. But what is the preferred way for someone implementing RESTCONF on a device who would like to have the support of <candidate>, <startup>, <running> and the new ones defined in NMDA.
>                 How does one copy the data from <running> to e.g. <startup> using the mechanisms RESTCONF has defined in RFC 8040? From reading the RFC it seems is out of scope of RFC 8040 but is this really intended?
>                 Implementing ietf-netconf.yang of course could be an option.

I agree that this isn't really specified.

IIRC our initial focus was to effectively get parity with RFC 8040.  
I.e. my working assumption is that <candidate> (if present) would be 
handled automatically and similarly updates to <startup> would be 
handled automatically.  E.g. broadly equivalent to the /data resource in 
RESTCONF.  So in essence this boils down to clients interacting with 
/ds/ietf-datastores:running, /ds/ietf-datastores:operational and 
/ds/ietf-datastores:intended. No new RESTCONF operations should be 
necessary here.

I'm also not really convinced that independently controllable <startup> 
and <candidate> is really a great idea.

Given that <running> is meant to represent persistent configuration, 
then automatically updating <startup> as a side effect of updating 
<running> seems like the most sensible behavior.  If the desire is for 
some ephemeral type configuration then that would be better modeled via 
an explicit dynamic configuration datastore (e.g. a bit like what I2RS 
were trying to do).

Similarly, instead of a shared candidate datastore, there is some work 
investigating private candidate datastores. 
draft-lhotka-netconf-restconf-transactions-00 gives some ideas of what 
this could look like, but further work is required.

If there is a strong requirement to allow full explicit manipulation of 
configuration datastores then I think that we should probably define 
explicit RPCs for these operations (modeled on the NETCONF ones).  
Ideally, these could be defined in a protocol neutral format so that 
they can work with any YANG based management protocols.  In the interim, 
using the NETCONF RPCs from RESTCONF seems like a reasonable stopgap 
measure.

Thanks,
Rob


>   
>
> Thanks,
> Rob
>
>>> 3.       Typo in https://urldefense.proofpoint.com/v2/url?u=https-3A__tools.ietf.org_pdf_draft-2Dietf-2Dnetconf-2Dnmda-2Drestconf-2D05&d=DwIBAg&c=Bg5XULDZR8GiOSSWNlpkCsRePnGDkKcI6oYL9xv1MnQ&r=2XEKVYkQjmLHi2TOJp1VSzieLZVqewIpj-RxmRgPfsM&m=CDK28X8kO2wRVDla97P_WqBIhF3OAzbWSGEmTeLdxwU&s=BSyY-GBuzzDry6oIhg-mgwMAySbhJEO9Im3Z4PD_c_4&e= Chapter 3.1 "the server would implement the resource {+restconf}/ds/example- ds-ephemeral:ds-ephemeral."
>>> There is a space in between "example-" and "ds-ephemeral:ds-ephemeral".
>> Lets hope we get this fixed with the help of the RFC editor.
>>
>> /js
>>
>> --
>> Juergen Schoenwaelder           Jacobs University Bremen gGmbH
>> Phone: [phone]         Campus Ring 1 | 28759 Bremen | Germany
>> Fax:   [phone]         <https://urldefense.proofpoint.com/v2/url?u=https-3A__www.jacobs-2Duniversity.de_&d=DwIBAg&c=Bg5XULDZR8GiOSSWNlpkCsRePnGDkKcI6oYL9xv1MnQ&r=2XEKVYkQjmLHi2TOJp1VSzieLZVqewIpj-RxmRgPfsM&m=CDK28X8kO2wRVDla97P_WqBIhF3OAzbWSGEmTeLdxwU&s=w38seWi6FN48OjPfJn92--emi3rEzEiDTHsKLxEnsrg&e=>
>> _______________________________________________
>> netmod mailing list
>> [e-mail]
>> https://urldefense.proofpoint.com/v2/url?u=https-3A__www.ietf.org_mail
>> man_listinfo_netmod&d=DwIDaQ&c=Bg5XULDZR8GiOSSWNlpkCsRePnGDkKcI6oYL9xv
>> 1MnQ&r=2XEKVYkQjmLHi2TOJp1VSzieLZVqewIpj-RxmRgPfsM&m=TrPMVXQ5IovFm08BS
>> bbXa2E-HnSO_yzHRy0GR9djT2M&s=sd11onHA42ODnysz9ZOIZikWWQMHkHwUSeL-lNWck
>> DE&e=
> **********************************************************************
> DISCLAIMER:
> Privileged and/or Confidential information may be contained in this message. If you are not the addressee of this message, you may not copy, use or deliver this message to anyone. In such event, you should destroy the message and kindly notify the sender by reply e-mail. It is understood that opinions or conclusions that do not relate to the official business of the company are neither given nor endorsed by the company. Thank You.