Re: [Netconf] LC on subscribed-notifications-10

Hi Kent,

Thanks so much for the detailed review.   Thoughts in-line.  At this point there doesn't seem to be anything insurmountable...

A working copy draft which embeds / covering the points documented below is at:

https://github.com/netconf-wg/rfc5277bis/blob/master/draft-ietf-netconf-subscribed-notifications-11.txt

Also a legend for the comments below:

**** indicates a significant item (others might want to read/chime in).

Blue indicates text which is now in the draft (verbatim).

Orange indicates an open question, where I am asking for feedback before making changes.

Note: where I use colors, the wording should still be fine for those WG members using plain text email clients.

Still pending:

- Martin's comments

- YANG doctor comments

> Kent Watsen, March 14, 2018 9:52 PM

>

> Here's my review of this draft.

>

> I'm aware that there may be some overlap with recent messages from Rob and

> Martin.  Please respond to them anyways, if only to explain the resolution

> made.

>

> BTW, when I make an open-ended question, what I'm many times looking for

> draft-text that answers the question.  Yes, I want to know the answer but,

> more importantly, I want the answer recorded in the draft.

>

> PS: I'm prioritizing reviewing all three drafts over trying to reply to responses

> from earlier reviews.

>

> Thanks,

> Kent // contributor (but revving-up for shepherd write-up)

>

>

> <chair-hat> Authors, can you please start planning a presentation to review any

> of the larger open issues during the meeting in London? </chair-hat>

Will do

> Title: the word "custom" is throwing me, what does it mean?  I see the word in

> the Abstract and similar text in the Introduction.  In total, the substring

> "custom" appears six times in the draft, all in the Title, Abstract, and

> Introduction, so the word doesn't seem to carry much weight in the body of

> the draft itself.  Is there a better word?  Perhaps "Subscriber-specific" or

> "Receiver-specific"?   Or maybe you want to say "Customized Subscriptions to a

> Publisher's Event Streams"?

Both paths work.  I switched it to:

Customized Subscriptions to a Publisher's Event Streams

> Abstract: The first sentence has three issues: first, there's the

> custom/subscriber-specific comment from before; second, the word

> "capabilities" in the first sentence is unclear (if you mean NETCONF/yang-

> library capabilities, this document does not define any); and third, the word

> "operations" is ambiguous, the draft uses this word sometimes to mean RPCs,

> but other times not.  Putting it all together, maybe this is better?

>

>    This document defines mechanisms enabling subscriber-specific

>    subscriptions to a publisher's event streams.

Based on Robert's comments on add the YANG Data model, I morphed your proposal to:

This document defines mechanisms and a YANG Data Model enabling subscriber-specific subscriptions to a publisher's event streams.

> Also, in the last sentence, s/Effectively/Combined/ and s/request/request for/?

Tweaked

> Introduction: Similar issues with the first sentence as with the Abstract.  Also,

> missing is a statement regarding this draft's compatibility to NMDA (see

> rfc6087bis)

Replicated the first sentence of the abstract to the introduction.  Also added a final sentence to the Intro which says:

The YANG model in this document conforms to the Network Management Datastore Architecture defined in [I-D.ietf-netmod-revised-datastores].

> Motivation:

>

>   How about this?

>   OLD: There are various [RFC5277] limitations, many of which have been

>        exposed in [RFC7923] which needed to be solved.

>   NEW: Various limitations in [RFC5277] are discussed in [RFC7923].

>        Resolving these issues is the primary motivation for this work.

updated

>   s/document includes/document include/

updated

>   in the 2nd bullet, remove "statically"?  the word "static" hardly appears...

updated

>   in the 3rd bullet point: would appending "in progress" be okay?

updated

> Terminology: I think you want to use this one:

>

>       The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL

>       NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT

> RECOMMENDED",

>       "MAY", and "OPTIONAL" in this document are to be interpreted as

>       described in BCP 14 [RFC2119] [RFC8174] when, and only when, they

>       appear in all capitals, as shown here.

Updated

>   For the "Configured subscription" term, I think that replacing "a

>   configuration interface which" with "configuration that" is clearer.

>   If necessary, we could import the term "configuration" from the

>   revised-datastores draft.

I have added the following:

Configuration: defined in [I-D.draft-ietf-netmod-revised-datastores]

Configuration datastore: defined in [I-D.draft-ietf-netmod-revised-datastores]

Configured subscription: A subscription installed via configuration into a configuration datastore.

This addresses the reboot persistence subsystem question (which came up in Robert's review) by more tightly coupling the terms to the revised datastore work.   Let me know if there are still concerns.

>   For the "Event" term, remove the parenthesis and spell out "e.g."?

How about:

Event: An occurrence of something that may be of interest. Examples include, a configuration change, a fault, a change in status, crossing a threshold, or an external input to the system.

>   Remove the term "NACM", since it only appears in the Security

>   Considerations section.

Done

>   For the "Notification message" term, is the beginning important?

>   Maybe s/A set of transport encapsulated information/Information/?

Done.

>   For the "Publisher" term, why is "Subscription" capitalized?  Is it

>   (and all other terms) capitalized consistently throughout the draft?

Very early iterations of these drafts had all terminology capitalized.  Earlier reviews resulted in downshifting the terms because it hindered readability.   The large "S" is likely something left over which got missed.   It is now a lower case 's'.

>   For the "Stream" term, I'm wondering if this should be renamed "Event

>   stream" (matching what's in the title), and then search/replace instances

>   of just "stream" with "event stream" everywhere else in the document.

>   This seems better, less ambiguous.

****

We went back and forth on this.  The term is used so often that always saying "event stream" just made the document more cumbersome to read.  In the end, RFC-5277 used both in the terminology, in a similar way.  For example:

In RFC 5277: "stream" appears 104 times, and "event stream" 47 times.

In this doc: "stream appears 297 times, and "event stream"  39 times.

As using both terms made things more humanly readable, and it seemed ok for RFC-5277, we choose that path.   Let me know if *not* adding event before every use of the word stream is ok with you.

>   For the "Subscribed event records" term, I recommend removing it, as

>   it only appears three times in the draft and, besides, you already have

>   the "Event record" term.

Done.   (Re-reading, I don't think anything is lost by removing the term either.)

>   For the "Subscriber" term, shouldn't you have a 2nd sentence like in

>   the "Receiver" term?

Added the same sentence to the receiver term.

>   Since the tree diagrams are scattered throughout the document, it would

>   be good to add the following here:

>

>      Tree diagrams used in this document follow the notation defined in

>      [I-D.ietf-netmod-yang-tree-diagrams].

Done

>

> Solution Overview

>

>   what does "instantiated" mean in the 1st paragraph.  suggest removing

>   if not needed.

It just meant "which exists".  Removed.

>   in (1), s/RPC/an RPC/.  Also, is "wants" the right word, maybe "is able"?

Made: "is able"

>   same with "wish" in the next sentence.

Made "is not able"

>  Also, in the last sentence,

>   s/ which would have been accepted/ that, had they been present, would

>   have enabled the dynamic subscription request to be accepted/?

Updated

>   in (2), s/a configuration interface/configuration/.

Done

>  Also, replace "this

>   capability" with "configured subscriptions", and maybe append "based on

>   the use of a YANG feature"?

Made it:

Support for configured subscriptions is optional, with its availability  advertised via a YANG feature.

>   "For connection-oriented stateful transport" : s/For/For a/ or

>   s/transport/transports/?

Chose: transports

>   Looking at "Also note that transport specific transport drafts based

>   on this specification MUST detail the life cycles of both dynamic and

>   configured subscriptions." - do the netconf-event-notifications and

>   restconf-event-notifications drafts do this?

Yes.   It is in non-normative text, but the flow diagrams in both drafts' appendices do this.

>   Last paragraph, s/The publisher/A publisher/

Done

> Relationship to RFC-5277:

>

>   In the first bullet point, the "data model" for what, configuration

>   or a notification?   (same issue is in the last bullet point as well)

As there is no configuration of RFC-5277 subscriptions, it was for the notifications.  So I made the bullet:

the data model in this document replaces the Notification Management Schema of [RFC5277], Section 3.4.

And I made the last bullet:

a publisher MAY implement both the Notification Management Schema and RPCs defined in [RFC5277] and this new document concurrently,...

>   The 4th bullet point isn't true (see Event Streams below)

****

I believe that it is true.   See discussion below.

> Solution:

>

>   Can you add a paragraph here to introduce what all is in Section 2,

>   how it's organized, or whatever might be helpful?

>

> Event Streams:

>

>   The 2nd paragraph says "except for where it has been explicitly

>   indicated that this the event record MUST be excluded from the

>   NETCONF stream".  This is a redefinition of what RFC5277 says,

>   has this been discussed?  How is this done (syntax/text)?  Has

>   it been done already?

****

Subscription state change notifications as per Section 2.7 are explicitly excluded from anyone but the target receiver.   Since the notifications are per-receiver, they cannot be placed into any NETCONF stream (for either RFC-5277 or subscribed-notifications).  And as they are excluded from the NETCONF stream, I do not see an issue with the Bullet 4 comment above.

To make this clearer in the draft text, here is some proposed/tweaked text for the start of Section 2.7...

Subscription state notifications are unlike other notifications in that they are never included in any stream.  Instead, they are inserted (as defined in the section below) within the sequence of notification messages sent to a particular receiver.  Subscription state notifications cannot be filtered out...

>   s/treated a distinct/treated as a distinct/

Done

> Event Stream Filters

>

>   The 1st and 2nd sentences seems to be at odds with each other.

****

I don't believe they are at odds.  But I can tweak the wording.   How about making the second sentence:

A match on a filter always results in an action upon a complete event record. Information is never stripped from within an event record prior to that event record being encapsulated within a notification message.

> QoS

>

>   What does " MUST work identically" mean?

>   is HTTP a mandatory  transport?

>   RFC 7540 Section 5.3.3 talks about a PRIORITY frame,

>   which is  defined in Section 6.3 of that draft.  How is this

>   suppose to work in a transport-agnostic way?

****

It would be excellent if we can adopt the a subset of prioritization types in HTTP2 without having to redefine the details of the algorithm in this document.  I believe this is possible, but I understand that you want refined wording to make sure this is accomplished explicitly.  Proposed are two snippets of revised text which hopefully accomplishes this:

Snippet 1:

Dequeuing of notification messages across independent subscriptions to a receiver SHOULD be allocated bandwidth proportionally based on each subscription's weight.  For more information on the proper treatment, see stream dependency weighting within RFC 7540, section 5.3.2.

Snippet 2

If a subscription has a dependency, then any buffered notification messages containing event records selected by the parent subscription SHOULD be dequeued prior to the notification messages of the dependent subscription.  If notification messages have dependencies on each other, the older notification message MUST go first.  For more information on the proper treatment to stream dependency as described within [RFC7540], section 5.3.1.  If a dependency included within an RPC references a subscription which does not exist or is not visible to that subscriber, that dependency may be silently removed.

Also HTTP is not mandatory.  In fact with the text change, the reference to RFC-7950 now becomes informative rather than normative.

> Dynamic Subscriptions

>

>   s/RPC/RPCs/

ok

>   Please provide more detail about how extensibility is accomplished,

>   or an example showing the augmentation occurring.

****

Rather than talk about how augmentation might be done in theory, it should be cleaner to the reference to YANG-Push augmentations.  So I added the following sentence...

For examples of such augmentations, see the RPC augmentations within [I-D.ietf-netconf-yang-push]'s YANG model.

>   For all the subsections, should the title be s/Subscription/Dynamic

> Subscription/?

Done

> Dynamic Subscription State Model

>

>   What does "asserted" mean?  - remove/replace?

Removed

>   I'm confused by the diagram and subtitle's use of the word "receiver",

>   when the first sentence of the paragraph above says that the SM is for

>   the publisher...

This is for the Publisher: the publisher must maintain the state of whether a receiver is currently active or suspended.

I changed the title to: "Publisher's state for a dynamic subscription" which should help here.   Other reviewers requested the addition of the word receiver to the states themselves.  This is so people could make a 1:1 correlation with the states of the configured subscription state machine.

>

>   Only two notifications?

Only two notifications indicate a change in the state of the subscription.

>   Looking at the graphic, how is the reader to

>   distinguish these as notifications?

Added a * to the two notifications, and text at the bottom of the drawing which says:

* indicates a state-change-notification

>   The last sentence of the last bullet doesn't square with what's in the

>   graphic.  is "modify-subscription" suppose to be bidirectional?

The diagram is correct.    I have changed the sentence to:

There are no direct controls over resuming a subscription other than to attempt a modification of a subscription in a way which reduces the resources consumed.

> Establishing a Subscription

>

>   I take it that the last two sentences of the first paragraph are

>   intended as requirements for transport-bindings.  Is that correct?

Yes

>   If so, then please say so.

Morphed to:

The transport selected by the subscriber to reach the publisher MUST support multiple establish subscription RPC requests made within the same transport session.  In addition, the transport MUST support the pipelining of RPC requests made on independent subscriptions.

(As interleave seems to have NETCONF implications, am trying to move ay from that to pipelining which is a general computer science term.)

>   The tree diagram is not identified as a tree diagram.  Nowhere in this

>   document is the tree-diagrams draft referenced.  This needs to be fixed.

Tree diagram reference added to the definitions section.  And also added as part of each figure name.  And each tree diagram also has text and a hyperlink near it pointing to the YANG model for more details.

>   Are your tree diagrams dynamically-generated?  - is there any concern

>   that they are out-of-date?

Generated from Pyang.   Manually snipped from the output.  Concerns are discussed more below.   Next drafts I am certainly changing my integration environment.

>   Since you're not describing the contents of the data model here, the

>   text should say that a complete description of all the nodes is provided

>   in the YANG module, with a reference.

Every tree in the document now has something like:

Below is a tree diagram for "establish-subscription". All objects contained in this tree are described within the included YANG model within <xref target="data_model"/>.

>   why is this "establish-subscription-error-stream" yang-data name having

>   "-stream" at the end?  (same issue with the other yang-data).  It's

>   a rather confusing name.  Maybe "-info" would be better?

****

We have to have a different yang-data structures for hints provided on datastores and on streams.  Because of that -info is not sufficient.   And while it is possible to place stream and datastore at the beginning of the yang-data name, it is kind-of nice to have the error-info hints start off with the same characters.

That said, I have no problem if people want to rename the yang-data both here and in yang-push to:

stream-establish-subscription-error-info

and

datastore-establish-subscription-error-info

Is this what you prefer?

>   Also, just so I'm clear, each transport-binding needs to indicate if and

>   how the yang-data structs are returned, right?  Where is this done in

>   the netconf-notif and restconf-notif drafts?

Yes

> Replay Subscription

>

>   Should the title being "Replaying Subscriptions", to match the verb

>   tense of the other subsections?

Tweaked to  "Requesting a replay of event records".  Because this is not a new RPC, I figure such differentiation from the other subsections is helpful.

>   s/Replay puts no/Supporting replay puts no/ or /The document puts no/?

Chose the " The document puts no "

>   Current text says:

>     """

>     The inclusion of a replay-start-time within an "establish-

>     subscription" RPC indicates a replay request.  If the "replay-start-

>     time" contains a value that is earlier than content stored within the

>     publisher's replay buffer, then the subscription MUST be rejected,

>     and the leaf "replay-start-time-hint" MUST be set in the reply.

>     """

>   Why not just start with what you have, prepended by a special "event

>   record" that says there is a gap?

****

This discussion went around on the alias a few times.  E.g., the thread from mid-October titled " Martin's thoughts on subscribed-notifications"

An underlying design goal of subscribed-notifications and yang-push is to deliver no less than what subscriber explicitly requested.  Especially when YANG-Push is layered in, if we start delivering less for some combination of parameters, we have no certainty that the subscriber is getting what it needs.

For this parameter, if we start replaying more recently than what has been requested, we don't really know if that is what the subscriber wants.  This doesn't give them the chance to reject the subscription while being sent stuff which is not helpful to them without the earlier history.  And you are correct, while we could define a special event record replay actually began on success, we are not delivering on the implicit promise of the subscription "ok".  But by using the no-success result with the included "replay-start-time-hint", we are matching the design paradigm without adding special constructs.

>   OLD: it MAY also be earlier than the current time and MUST

>   NEW: it MAY be earlier than the current time, but MUST

Done

>   "subscribers can perform a get on" - rephrase, and use "RPC" somewhere

Made it:

To assess the availability of replay, subscribers can retrieve the "replay-log-creation-time" and "replay-log-aged-time" objects from the YANG model.

With that, I don't think RPC is needed.

> Modifying a Subscription

>

>   First sentence, no need for the word "previously"

Done

>   s/one or multiple times/multiple times -or- any number of times/?

Chose "any number"

>   s/via RPC using/via an RPC on/?

Done

>   The tree diagram is not identified as a tree diagram.  And since the

>   data model isn't explained, there should be a statement for the reader

>   to look at the YANG module for details, ideally with a hyperlink.

Now done for every tree diagram in the document

> Deleting a Subscription

>

>   First sentence, no need for the word "previously"

>

>   Under what conditions could a publisher reject a delete-subscription

>   request?  should there delete-subscription-error-stream hints?

>

>   The tree diagram is not identified as a tree diagram.  And since the

>   data model isn't explained, there should be a statement for the reader

>   to look at the YANG module for details, ideally with a hyperlink.

>

>   Last paragraph, no need for the word "previously"

Done

> Killing a Subscription

>

>   Regarding:

>     "This operation MUST be secured so that only connections with

>      sufficiently privileged access rights are able to invoke this RPC."

>   This needs to be in the Security Considerations section and, given

>   that, doesn't need to be here, right?  If you really want it here,

>   then please indicate that such guidance is provided in the SC section.

Moved to Security Considerations

>   Replace the paragraph beginning with "The tree structure of" with the

>   actual tree diagram for this RPC.

Done

> RPC Failures

>

>   Please also call-out RESTCONF error handling (RFC8040 Section 7.1).

Done

>   The 2nd paragraph is confusing.  mechanism?  how are the 1st and 2nd

>   sentences related? What does the 2nd sentence really mean, esp. wrt.

>   the MUST?

Rewrote the paragraph to:

Specific errors included within this document's YANG model MUST be returned as part of the RPC error response. Following are valid errors which can occur for each RPC:

>   I can't find any examples of these errors in use.  The

>   netconf-event-notifications draft only has examples for

>   the "establish-subscription-error-datastore" and

>   "modify-subscription-error-datastore" errors.

Figure 10 in the netconf-event-notifications draft works equally for subscribed-notifications, as well as yang-push.   I have identified that example in that document as being relevant to either streams or datastores with the sentence in that draft: "This subscription may have been to either a stream or a datastore."

Here this document, I have added the sentence:

To see a NETCONF based example of an error response from above, see [I-D.draft-ietf-netconf-netconf-event-notifications], Figure 10.

>   Perhaps the

>   examples in that draft need to be split into examples related

>   to yang-push vs examples related to subscribed-notifications.

As the error mechanisms are identical between the drafts, splitting things in that document might prove more confusing.  That is one reason I identify the error response as being identical for streams and datastores above.   Perhaps additional examples, git repositories, or applications located outside the drafts?

> Configured Subscriptions

>

>   1st paragraph: s/configuration interface/configuration/g  (two cases)

Done

>   the note under the 3rd bullet point seems unnecessary but, if keeping

>   it, then just say that receivers are unaware of the existence of any

>   other receivers.

Done.  Used your proposed text.

>   s/In addition to subscription/In addition to the subscription/

>   s/as described in/described in/

Done

>   where is the tree diagram for the configuration data model?!

****

It is in the section "Subscriptions Container".  It seemed better to introduce the state machines before getting into the details of the tree.  But if you really want to have it early, it certainly can be moved up.

So do you want it moved here, or is a reference to the later section sufficient?

>   I don't understand the last bullet point.  First, I'm having trouble

>   parsing the implicit parentheses.  Next, the last sentence seems

>   complicated, maybe just say "unless directed otherwise, the

>   notification messages MUST egress the publisher's default

>   interface towards the receiver."?

Used your text.  Done.

> Configured Subscription State Model

>

>   A better first sentence is needed, something introducing that there

>   exists a state machine for each configured subscription, and states

>   that there are three states (VALID, INVALID, and CONCLUDED), etc.

>   Also should state where this state machine is maintained (publisher,

>   receiver, both?)

Now says:

Below is the state machine for a configured subscription on the publisher.  This state machine describes the three states (VALID, INVALID, and CONCLUDED), as well as the transitions between these states. Start and end states are depicted to reflect configured subscription creation and deletion events.

>   s/publisher evaluation/evaluation by the publisher/?

Done

>   Please move text regarding how to interpret the diagram (uppercase,

>   dashed boxes, parantheses, etc.) into a preamble or postamble element.

Added underneath the diagram.  See diagram below.

>   s/itself might itself/itself might/

>   s/in no longer/is no longer/

Done

>   The first paragraph under the diagram doesn't match what the diagram

>   shows.  Looking at the diagram, I also see two possible sequence of

>   transitions that get VALID to INVALID, but I'm unsure how they relate

>   to the two mentioned in the text.

Updated paragraph text as per below.  Hopefully it is clearer now.

>  The text should call out which

>   parts of the diagram it's referring to.  Many times I number labels

>   in diagrams and then, under the diagram, provide a more thorough

>   explanation for each number.

Added numbers within the diagram, and added text references as per below:

.........

: start :-.

:.......: |

     create  .---modify-----.----------------------------------.

          |  |              |                                  |

          V  V          .-------.         .......         .---------.

 .----[evaluate]--no--->|INVALID|-delete->: end :<-delete-|CONCLUDED|

 |                      '-------'         :.....:         '---------'

|----[evaluate]--no-.      ^                ^                 ^

 |        ^          |      |                |                 |

yes       |          '->unsupportable      delete           stop-time

 |      modify         (subscription-   (subscription-   (subscription-

 |        |             terminated*)     terminated*)      concluded*)

 |        |                 |                |                 |

 |       (1)               (2)              (3)               (4)

|   .---------------------------------------------------------------.

'-->|                         VALID                                 |

     '---------------------------------------------------------------'

Legend:

dotted boxes: subscription creation and deletion events

dashed boxes with uppercase letters: valid states for a subscription

[evaluate]: decision point on whether the subscription is supportable

(*): resulting subscription state change notification

Also the text below now says:

A valid subscription may become invalid on one of two ways.  First, it may be modified in a way which fails a re-evaluation.  See (1) in the diagram. Second, the publisher itself might determine that the subscription is no longer supportable.  See (2) in the diagram.  In either case, a "subscription-terminated" notification is sent to any active or suspended receivers.  A valid subscription may also transition to a concluded state via (4) if a configured stop time has been reached.  In this case, a "subscription-concluded" is sent to any active or suspended receivers.  Finally, a subscription may be deleted by configuration (3).

>   Is it "any active or suspended receivers" or "any receivers for an

>   active or suspended subscription"?

The current wording is correct.  A configured subscription is never suspended.  It can be INVALID, or it can be ACTIVE and all its receivers suspended.  But in the second case, at least the receivers get subscription-suspended notifications.

>   s/During any times a/When a/?

>

>   Regarding "Below is the state machine for each receiver of a configured

>   subscription." - where is this state machine maintained, on the publisher

>   or on the receiver?

Updated the title to show it is a Publisher state model.

>   why is "receiver" in each box?

To drive home the idea that this state machine was for each individual receiver, rather than for the subscription as a whole.

>   Again, you might look to having a

>   preamble or postamble to describe the syntax used in the diagram.

Per figure below, added the legend as a postamble:

>   1st paragraph below diagram: s/to connecting/to "connecting" -or- to

> CONNECTING/?

Now says CONNECTING.   And all receiver states moved to uppercase.

>   Regarding "and event records are not being dropped due to a publisher

>   buffer overflow" - this seems like it's from out of nowhere.  If not

>   normative, then maybe delete?

****

It is normative.  This is needed to maximize the number of concurrent subscriptions without enforcing continuous transport keep-alive overhead when no event records are being passed, as well as to not prematurely declare a subscription as suspended while there is a chance that transport may be established before event records do get lost.    This allows a configured subscription's receiver to exist across an intermittent connection, and the receiver can remain active on the publisher as long as events aren't being lost.  While this can be done with NETCONF, it is probably more likely to be seen in practice with HTTP connections.

Based on that, I rephrased the words above so that it doesn't feel from out of nowhere.  See the text below the updated figure below...

>   This text is again difficult to reconcile with the diagram.  I again

>   recommend numbering labels and then describe the numbers below.

Done

     .-----------------------------------------------------------------.

     |                         VALID                                   |

     |   .----------.                           .--------.             |

     |   | receiver |------------------timeout->|receiver|             |

     |   |CONNECTING|<------------------reset---|TIMEOUT |             |

     |   |          |<-transport---.            '--------'             |

     |   '----------'  loss,reset  |                                   |

     |      (1)          |         |                                   |

     |  subscription-   (3)        |                                   |

     |  started*    .--------.     |                       .---------. |

     |       '----->|        |     '--------------------(3)|         | |

     |              |receiver|(2)-subscription-suspended*->|receiver | |

     | subscription-| ACTIVE |                             |SUSPENDED| |

     |   modified*  |        |<--subscription-resumed*,----|         | |

     |        '---->'--------'    subscription-modified*   '---------' |

     '-----------------------------------------------------------------'

  Legend:

   dashed boxes which include the word 'receiver' show the possible

   states for an individual receiver of a VALID configured subscription.

   * indicates a state change notification

Individual receivers are moved to an ACTIVE state when a "subscription-started" state change notification is successfully passed to that receiver (1). Configured receivers remain ACTIVE if both transport connectivity can be verified to the receiver, and event records are not being dropped due to a publisher buffer overflow. The result is that a receiver will remain ACTIVE on the publisher as long as events aren't being lost, or the receiver cannot be reached.  However if there is buffer overflow, or the publisher cannot generate events for a receiver, the receiver MUST be suspended (2).  In addition, a configured subscription's receiver MUST be moved to CONNECTING if transport connectivity cannot be achieved, or if the receiver is reset via configuration operations (3).

>   s/ mechanisms described above is/ mechanisms described above are/

>   What does this mean, how are mechanisms mirrored for RPCs and

>   notifications?

Done

>   Regarding " provides an example of such an extension" - which section?

Revised text to:

The YANG model [I-D.ietf-netconf-yang-push] Section 4.1,  provides many such extensions, this includes the augmentation of "/sn:modify-subscription/sn:input/sn:target".

> Creating a Configured Subscription

>

>   1st paragraph: let the first sentence be its own paragraph as with

>   the other 2.5.x sections.

>   For the remainder, I think this is the

>   3rd time that the draft has discussed the differences between

>   configured and dynamic subscriptions.  Please eliminate unnecessary

>   redundancy.  Factor out into another section if needed.

I agree that the following paragraph can be deleted.

There are two key differences between the new RPCs defined in this document and configuration operations for subscription creation. Firstly, configuration operations install a subscription without question, while the RPCs are designed to the support negotiation and rejection of requests. Secondly, while the RPCs mandate that the subscriber establishing the subscription is the only receiver of the notification messages, configuration operations permit specifying receivers independent of any tracked subscriber.

I have just removed this.

>   Regarding 2nd/3rd paragraphs, how resilient is the solution to the

>   resumption of the underlying transport?  If messages lost in the

>   write-buffer are lost, could the receiver ever be helplessly out

>   of sync without a full restart?

I think we are clean here.  I have updated the text against the diagram per-above which hopefully provides more descriptive text on why the resumption of  underlying transport is covered.

> Modifying a Configured Subscription

>

>   s/ ././    ;)

Done

> Resetting a Configured Receiver

>

>   But *how* is it reset? - via a configuration operation?  which one?

>   Should this be part of "Modifying a Configured Subscription"?

Added the sentence:

This is accomplished via the "reset" action within the YANG model at "/subscriptions/subscription/receivers/receiver/reset".

> Event Record Delivery

>

>   First paragraph, last sentence.  I think I commented on similar text

>   before.  Is this a requirement for the transport binding?

Perhaps the word interleave is the wrong choice here, and intermixing is better in this case.  I made that change.

>  Do the

>   netconf-notif and restconf-notif drafts satisfy this requirement?

Yes,

> where?

Netconf-notif supports interleaving of requests as described in Section 3.

Restconf-notif doesn't need to explicitly call for pipelining support as it is a basic capability of HTTP.

>   2nd paragraph: "able to traverse" --> "not blocked by"?

Done

>   Also, for

>   the 3rd sentence, call out the "RPC response" is for dynamic and

>   "state-change notification" is for configured?

Yes.   Made text:

A subscription's events MUST NOT be sent to a receiver until after a corresponding RPC response (in the case of a dynamic subscription) or state-change notification (in the case of a configured subscription) has been passed receiver indicating that events should be expected.

>   Last two paragraphs, this text needs to be removed,

removed

>   or else we might

>   need to block this draft on notification-messages.   What do you mean

>   by " this document will be updated to indicate support".

At some point when notification-messages is complete, this draft should be updated as it is a more robust solution (as a subscription id can be provided for event records provided on streams.)

> Subscription State Notifications

>

>   OLD

>    In addition to subscribed event records, a publisher MUST send

>    subscription state notifications to indicate to receivers that an

>    event related to the subscription management has occurred.

>   NEW

>    In addition to sending event records to receivers, a publisher MUST

>    also send subscription state notifications when events related to

>    the subscription management has occurred.

>   ???

Done.  (Removed the extra 'the')

>   2nd paragraph: s/directly//

Done

>   Also, I'm unsure about the "subscription-state-notif" extension, how

>   is it expected to be used by a YANG processor?

Per above, it ensures that these YANG notifications if encoded in XML are not placed onto the NETCONF stream.

>   Perhaps a generic

>   notification-filtering GUI is envisioned whereby the logic could

>   automatically remove these notifications from selection, but coding

>   for this extension has very limited use, as no other drafts are ever

>   likely to define any.  I suppose it does no harm, but I also think

>   that the text sure be clear.  Personally, I'd rather the extension

>   be removed unless there is a good reason to keep it.

****

The three choices seem to be:
(a) current solution

(b) hardcode the these notifications so none ever go on the NETCONF stream

(c)  make the extension "exclude-from-NETCONF-stream".  As it is quite possible that other drafts will want to do that.

I am good with any of these.  But the first seems the cleanest, and most self contained.  Let me know it the current doesn't work for you.

> subscription-started:

>

>   Regarding the 2nd paragraph, Section 2.4.2.1 implies a contradiction

>   to this statement.

****

A replay subscription can be set for a configured subscription.  There was some carrier on the NETCONF alias who requested this many months ago.  See also dialogs with Martin.

Looking at your comment, it probably isn't a good idea to embed this fact within the replay text embedded as part of the dynamic subscription section.

The best way to tease this apart is first to separate any configured subscription context the 2.4.2.1.   This can be done simply by replacing the 'after the "subscription-started" notification'. With ' after the after a successful establish-subscription RPC response'.

And then to be more explicit that this is supported, we could add move contradicting statement into a new section 2.5.6 where it would no longer appear contradicting.  Replay in a new section looks like this:

2.5.6 Replay for a Configured Subscription

It is possible to place a start time on a configured subscription.  This enables functionality like immediately streaming boot log information off of a publisher immediately after restart.

When any such configured subscription receivers become ACTIVE, buffered event records (if any) will be sent immediately after the "subscription-started" notification.  The first event sent will be the most recent following the latest of four different times: the "replay-log-creation-time", "replay-log-aged-time", "replay-start-time", or the most recent publisher boot time.

All other replay functionality remains the same as with dynamic subscriptions as described in Section 2.4.2.1

The good news is that all of this is consistent with text is already reflected in the YANG model.

>   The tree diagram is not identified as a tree diagram.  And since the

>   data model isn't explained, there should be a statement for the reader

>   to look at the YANG module for details, ideally with a hyperlink.

Done

>   Why is all this sent to the receiver?  Doesn't it already know the

>   protocol and encoding?  What about the other parts?  Which parts

>   are actually useful?

The complete state of the subscription is sent, which can also be useful for debugging.  But beyond that, based on what I am hearing from the CBOR people, even the protocol and encoding might be different between.

> subscription-modified

>

>   1st paragraph: the same parameters, or data model / tree diagram?

>   Also, is "provided" the right word?  Maybe it would be better to

>   have the tree diagram itself, even though only the name changes?

Provided the full tree.   It does chew up space, but that is not really an issue.

>   Last two paragraphs, why put "First" and "Second" when they are

>   bullet points.  Maybe you want to use a numbered-list or otherwise

>   rephrase these?

Made a numbered list

>   Last paragraph, the last sentence doesn't flow with the first.

>   It seems as if it was copy/pasted from somewhere else.  Is this

>   intended to be a normative statement here?

Yes it is a normative statement, and it is in the correct place.

I added text to smooth the transition.  It now is this:

While this state change will be most commonly used with configured subscriptions, with dynamic subscriptions, there is also one time this notification will be sent. A "subscription-modified" state change notifications MUST be sent if the contents of a filter identified by a "stream-filter-ref" has changed.

> subscription-terminated

>

>   1st paragraph, 1st sentence: -e a/The publisher/A publisher/ and

>   also s/the pushing of/pushing/?

Done

>   1st paragraph: "Such a decision may be made for" - should this

>   be "A publisher may terminate a subscription for" ?

Done

>   1st paragraph, for the "first type of reason": does the subscription

>   terminate when the first or last referenced objects are no longer

>   accessible?

This refers to any either any leafref going missing, or the subscription-id being removed.  More in next comment

>  BTW, what do you mean by "via the YANG model", aren't

>   these instance objects in <operational>?

I have updated the text in this section to be much more explicit to cover the intent.  The section now says

   A publisher MAY terminate pushing subscribed event records to a
   receiver.  This notification indicates that no further notification
   messages should be expected from the publisher.  A publisher may
   terminate a subscription for the following reasons:

   1.  Configuration which removes a configured subscription, or a "kill-
       subscription" RPC.  These are identified via the reason "no-such-
       subscription".

   2.  A referenced filter is no longer accessible.  This is identified
       by "filter-unavailable".

   3.  The stream referenced by a subscription is no longer accessible
       by the receiver.  This is identified by "stream-unavailable".

   4.  A suspended subscription has exceeded some timeout.  This is
       identified by "suspension-timeout".

Each of the reasons above correspond one-to-one with a "reason" identityref specified within the YANG model.

>   1st paragraph, what do you mean by " Identities within the YANG model"?

>   Can the text be more clear that it is referring to the "reason"

>   identityref in the tree diagram?

Text attempted just above.

>   The tree diagram is not identified as a tree diagram.

Done

>   last paragraph: remove "established".  Also, the first 2 sentences would

>   benefit moving to singular, as plural leads to some ambiguity.

Done.

Note: a subscriber can terminate an existing subscription via a "delete-subscription" RPC. In such a case, no "subscription-terminated" state change notification is sent.

> subscription-suspended

>

>   Please replace the 2nd paragraph with the actual tree diagram, and then

>   speak to that.

Done

   This notification indicates that a publisher has suspended the

   sending of event records to a receiver, and also indicates the

   possible loss of events.  Suspension happens when capacity

   constraints stop a publisher from serving a valid subscription.  The

   two conditions where is this possible are "insufficient-resources"

   and "unsupportable-volume".  These conditions are encoded within the

   reasons.  No further notification will be sent until the subscription

   resumes or is terminated.

   Below is a tree diagram for "subscription-suspended".  All objects

   contained in this tree are described within the included YANG model

   within Section 4.

       +---n subscription-suspended

          +--ro identifier    subscription-id

          +--ro reason        identityref

        Figure 11: subscription-suspended notification tree diagram

> subscription-resumed

>

>   The tree diagram is not identified as a tree diagram.

Updated.  As are all other tree diagrams now.

> subscription-completed

>

>   Please replace the 2nd paragraph with the actual tree diagram, and then

>   speak to that.

Updated.  As are all other tree diagrams now.

> replay-completed

>

>   2nd paragraph: s/ If subscription/ If a subscription/ and s/which/that/

Done

>   Please replace the last paragraph with the actual tree diagram, and then

>   speak to that.

Done as identical to above.

> Subscription Monitoring

>

>   1st paragraph: s/Container/The container/.

Done.

>   How can container "subscriptions" (config true) contain entries for

>   dynamic subscriptions?  Are you assuming in <operational>?

Updated the start of paragraph 1 to:

In the operational datastore, the container "subscriptions" maintains the state of all known subscriptions.

Updated paragraph 2 to:

Each subscription is represented as a list element.  While many subscription objects are "config true", dynamic subscriptions are only included within the operational datastore. Operational information which may be monitored includes receiver counter information, the state...

>   Also,

>   does it include configured subscriptions that are currently not

>   active for whatever reason?

Yes.   First paragraph above uses the word 'all'.

>   You mention NETCONF's <get> (wait, I

>   thought this draft was suppose to be transport agnostic), but not

>   NMDA's <get-data>, so it make me wonder if this paragraph regards

>   the contents of <running> or <operational>...

Yes, we want to make it want to make it agnostic.  So it now says:

Using datastore retrieval operations , or subscribing to...

>   The 2nd paragraph would make more sense if I was looking at a tree

>   diagram.  But then I realize that this would be the same tree-diagram

>   that should've been presented in Configured Subscriptions.

The tree is in the subscriptions container section just below.  I will gladly reference it wherever it ends up.

> Advertisement

>

>   The second paragraph seems to be mostly NETCONF specific and

>   therefore belongs in the netconf-binding draft.

Good point.  Moved the first sentence to the end of that draft's "Compatibility with RFC-5277's create-subscription" section.

>   In a transport-

>   agnostic draft, maybe only features should be discussed?

Makes sense

> YANG Data Model Trees

>

>   s/top level YANG Data Node containers/protocol-accessible nodes/

Done

>   " If you would rather see" - please use more formal language.

Made it:

For tree diagrams of state change notifications,

> Event Streams Container

>

>   1st paragraph, last sentence: perhaps rephrase as "This enables

>   clients to discover what streams a publisher supports."?

Done

>  BTW, is

>   the " and against which subscription is allowed" part important,

>   if so, why?

Not really.  I was just trying to highlight that different clients might have visibility for different streams.  As this is implicit, I just dropped it and used your text.

>   This tree-diagram does not match what I generate.  This indicates

>   that the tree diagrams are not being dynamically-generated.  I

>   strongly suggest updating your build script to dynamically generate

>   the tree diagrams.  We cannot afford to have them be out of alignment.

At the WG request, I segmented the YANG tree into different sections.  However I do not have the tooling which automatically extracts portions of the YANG tree.

Is there a git repository which recommends a continuous integration for sub portions of a YANG tree?  For future drafts, I have certainly built a strong desire for such a continuous integration environment.

> Event Stream Filters Container

>

>   "and validated" - is this needed, since *all* configuration is validated?

Removed

>   s/ which/ that/

Done

>   "referenced and used" - is there a difference?  - can you just use one?

Now just referenced

>

> Subscriptions Container

>

>

>   This tree-diagram does not match what I generate.  This indicates

>   that the tree diagrams are not being dynamically-generated.  I

>   strongly suggest updating your build script to dynamically generate

>   the tree diagrams.  We cannot afford to have them be out of alignment.

I would love to have fully generated scripts.   That is hard for a few reasons here:

(a) The automatically generated trees are getting mangled because they are so wide.  Especially with yang-push, the automatic trees must all be fixed manually each time.

(b) I have no insights on how to pull portions of a tree into a XML document.   Is there a tool site which provides this?

The delta I see is "rw" vs "ro".  Fixed now.   I have brought in the current tree.

> Data Model

>

>

>   I going to skip this part, for now at least, as I assume the YANG

>   Doctor will scrutinize it.

>

>

>

> Implementation Considerations

>

>  s/ For a deployment/To support deployments/

Done

>  s/split subscription/it is recommended to split subscription"

Done

>  is " unlikely" the right word?  doesn't it eliminate the concern altogether?

Yes it does solve it.

That way it eliminates the possibility of collisions if...

>  Regarding the 2nd-half of the 1st paragraph, is it necessary for

>  interoperability reasons for this draft to define how to split the

>  subscription identifiers into static and dynamic parts.

Not necessary, just a best practice.

>   Is the

>  normative text needed here?  Maybe just describe the current

>  approach as a possible way to go about doing it?  - I think it

>  achieves the same goal without using normative text.

Agree.  Text now says:

A best practice is to use lower half the "identifier" object's integer space when that "identifier" is assigned by an external entity (such as with a configured subscription). This leaves the upper half of subscription identifiers available to be dynamically assigned by the publisher.

>  For the 2nd paragraph, this sounds like normative text from earlier

>  in the document.  If so, then is it needed here again?

No.  Deleted.

>  For the 3rd paragraph, I'm not sure if the second sentence needs to

>  be said at all, but at least s/SHOULD/should/ so it's not normative.

Made it non-normative

> Security Considerations

>

>   Regarding the 1st paragraph, aren't *all* operations (configuration

>   or RPCs) always authenticated and authorized?

Yes.   Deleted as redundant.

>   Please restructure to follow, in part, the template provided here:

>   https://tools.ietf.org/html/draft-ietf-netmod-rfc6087bis-20#section-3.7.1

Restructured to this:

5.3.  Security Considerations

   The YANG module specified in this document defines a schema for data

   that is designed to be accessed via network management protocols such

   as NETCONF [RFC6241] or RESTCONF [RFC8040].  The lowest NETCONF layer

   is the secure transport layer, and the mandatory-to-implement secure

   transport is Secure Shell (SSH) [RFC6242].  The lowest RESTCONF layer

   is HTTPS, and the mandatory-to-implement secure transport is TLS

   [RFC5246].

   The NETCONF Access Control Model (NACM) [RFC6536bis] provides the

   means to restrict access for particular NETCONF or RESTCONF users to

   a preconfigured subset of all available NETCONF or RESTCONF protocol

   operations and content.

   There are a number of data nodes defined in this YANG module that are

   writable/creatable/deletable (i.e., config true, which is the

   default).  These data nodes may be considered sensitive or vulnerable

   in some network environments.  Write operations (e.g., edit-config)

   to these data nodes without proper protection can have a negative

   effect on network operations.  These are the subtrees and data nodes

   and their sensitivity/vulnerability:

   Container: filters

   o  stream-subtree-filter: updating a filter could increase the

      computational complexity of all referencing subscriptions.

   o  stream-xpath-filter: updating a filter could increase the

      computational complexity of all referencing subscriptions.

   Container: subscriptions

   o  address: can be used to attempt to send traffic to an unwilling

      receiver.

   o  dependency: can force important traffic to wait behind the

      unimportant.

   o  dscp: can send traffic with a higher priority marking that

      warranted.

   o  encoding: none

   o  identifier: can overwrite an existing subscription configured by

      another entity.

   o  port: none

   o  protocol: none

   o  purpose: none

   o  replay-start-time: can be used to push very large logs, wasting

      resources.

   o  source-address: address might not be able to reach a receiver.

   o  source-interface: interface might not be able to reach a receiver.

   o  source-vrf: can push subscribed traffic into a virtual network

      which might not contain receivers able to see the subscribed

      content.

   o  stop-time: none

   o  stream: none

   o  stream-filter-ref: none

   o  stream-subtree-filter: a complex filter can increase the

      computational resources for this subscription.

   o  stream-xpath-filter: a complex filter can increase the

      computational resources for this subscription.

   o  weighting: placing a large weight can overwhelm the dequeuing of

      other subscriptions.

   Some of the readable data nodes in this YANG module may be considered

   sensitive or vulnerable in some network environments.  It is thus

   important to control read access (e.g., via get, get-config, or

   notification) to these data nodes.  These are the subtrees and data

   nodes and their sensitivity/vulnerability:

   Container: streams

   o  name: if access control is not properly configured, can expose

      system internals to those who should have no access to this

      information.

   o  replay-support: if access control is not properly configured, can

      expose logs to those who should have no access.

   Container: subscriptions

   o  pushed-notifications: will show the amount of events a particular

      subscriber actually received from a stream.

   o  excluded-notifications: will show the results of access control,

      and how many event records have been filtered out.

   Some of the RPC operations in this YANG module may be considered

   sensitive or vulnerable in some network environments.  It is thus

   important to control access to these operations.  These are the

   operations and their sensitivity/vulnerability:

   RPC: all

   o  If a malicious or buggy subscriber sends an unexpectedly large number

       of RPCs, the result might be an excessive use of system resources on the

       publisher just to determine that these subscriptions should be declined. In

       such a situation, subscription interactions MAY be terminated by

       terminating the transport session.

   RPC: delete-subscription

   o  No special considerations.

   RPC: establish-subscription

   o  Subscriptions could overload a publisher's resources.  For this

      reason, Publishers MUST ensure that they have sufficient resources

      to fulfill this request or otherwise reject the request.

   RPC: kill-subscription

   o  The "kill-subscription" RPC MUST be secured so that only

      connections with administrative rights are able to invoke this

      RPC.

   RPC: modify-subscription

   o  Subscriptions could overload a publisher's resources.  For this

      reason, Publishers MUST ensure that they have sufficient resources

      to fulfill this request or otherwise reject the request.

>   Regarding the 2nd and 3rd paragraphs, this sounds good, but isn't

>   this behavior already defined by the draft?  (or should be?)

Yes they are.   I actually refined / incorporated these points in the template above.  As this is what the template appears to be asking to have.

>   Regarding the 4th paragraph, why would the publisher need to the

>   terminate the transport session?  wouldn't it have started to

>   reject dynamic subscriptions when it became overloaded?  Or is

>   this trying to say something specific about dropping the transport

>   session as a club?  ;)

Yes, as a club.  Moved this up into the template as part of "RFC: all" and fixed the text to show why the club might need to be used

>   Re: the 5th paragraph, this is better than the 1st paragraph, but

>   may not be needed if following the template.

Agree.  This is redundant, and the point is covered as per the template above.

>   Re: the 6th paragraph, I'm surprised that requirements for transport-

>   bindings wasn't discussed before in its own section.  It seems like

>   a new thing here, that a receiver's transport might not be secure.

>   I'm okay with and support this, btw, as its sometimes better to

>   offload devices thru the use of a local collector node, for which

>   encryption may not be needed...

Agree with your comments.

>   Re: the 7th paragraph, this was said before also, right?

Correct, removed.

>   Re: 2nd to last paragraph, what is the " very-secure" tag?

Removed, and the overall points moved up into template.  As for the very-secure tag, Andy had mentioned that a few years ago.   It looks like it wasn't standardized.

Eric

>

>

> Kent // contributor

>

>

>

>

>

> _______________________________________________

> Netconf mailing list

> Netconf@ietf.org<mailto:Netconf@ietf.org>

> https://www.ietf.org/mailman/listinfo/netconf