Skip to main content

Network Management Datastore Architecture
draft-ietf-netmod-revised-datastores-01

The information below is for an old version of the document.
Document Type
This is an older version of an Internet-Draft that was ultimately published as RFC 8342.
Authors Martin Björklund , Jürgen Schönwälder , Philip A. Shafer , Kent Watsen , Robert Wilton
Last updated 2017-03-13
Replaces draft-nmdsdt-netmod-revised-datastores
RFC stream Internet Engineering Task Force (IETF)
Formats
Reviews
Additional resources Mailing list discussion
Stream WG state WG Document
Document shepherd (None)
IESG IESG state Became RFC 8342 (Proposed Standard)
Consensus boilerplate Unknown
Telechat date (None)
Responsible AD (None)
Send notices to (None)
draft-ietf-netmod-revised-datastores-01
Network Working Group                                       M. Bjorklund
Internet-Draft                                            Tail-f Systems
Intended status: Standards Track                        J. Schoenwaelder
Expires: September 14, 2017                            Jacobs University
                                                               P. Shafer
                                                               K. Watsen
                                                        Juniper Networks
                                                               R. Wilton
                                                           Cisco Systems
                                                          March 13, 2017

               Network Management Datastore Architecture
                draft-ietf-netmod-revised-datastores-01

Abstract

   Datastores are a fundamental concept binding the data models written
   in the YANG data modeling language to network management protocols
   such as NETCONF and RESTCONF.  This document defines an architectural
   framework for datastores based on the experience gained with the
   initial simpler model, addressing requirements that were not well
   supported in the initial model.

Status of This Memo

   This Internet-Draft is submitted in full conformance with the
   provisions of BCP 78 and BCP 79.

   Internet-Drafts are working documents of the Internet Engineering
   Task Force (IETF).  Note that other groups may also distribute
   working documents as Internet-Drafts.  The list of current Internet-
   Drafts is at http://datatracker.ietf.org/drafts/current/.

   Internet-Drafts are draft documents valid for a maximum of six months
   and may be updated, replaced, or obsoleted by other documents at any
   time.  It is inappropriate to use Internet-Drafts as reference
   material or to cite them other than as "work in progress."

   This Internet-Draft will expire on September 14, 2017.

Copyright Notice

   Copyright (c) 2017 IETF Trust and the persons identified as the
   document authors.  All rights reserved.

   This document is subject to BCP 78 and the IETF Trust's Legal
   Provisions Relating to IETF Documents

Bjorklund, et al.      Expires September 14, 2017               [Page 1]
Internet-Draft                                                March 2017

   (http://trustee.ietf.org/license-info) in effect on the date of
   publication of this document.  Please review these documents
   carefully, as they describe your rights and restrictions with respect
   to this document.  Code Components extracted from this document must
   include Simplified BSD License text as described in Section 4.e of
   the Trust Legal Provisions and are provided without warranty as
   described in the Simplified BSD License.

Table of Contents

   1.  Introduction  . . . . . . . . . . . . . . . . . . . . . . . .   3
   2.  Terminology . . . . . . . . . . . . . . . . . . . . . . . . .   3
   3.  Introduction  . . . . . . . . . . . . . . . . . . . . . . . .   5
     3.1.  Original Model of Datastores  . . . . . . . . . . . . . .   7
   4.  Architectural Model of Datastores . . . . . . . . . . . . . .   8
     4.1.  The <intended> Datastore  . . . . . . . . . . . . . . . .   9
     4.2.  Dynamic Datastores  . . . . . . . . . . . . . . . . . . .  10
     4.3.  The <operational> Datastore . . . . . . . . . . . . . . .  10
       4.3.1.  Missing Resources . . . . . . . . . . . . . . . . . .  11
       4.3.2.  System-controlled Resources . . . . . . . . . . . . .  11
       4.3.3.  Origin Metadata Annotation  . . . . . . . . . . . . .  11
   5.  Guidelines for Defining Dynamic Datastores  . . . . . . . . .  12
     5.1.  Define a name for the dynamic datastore . . . . . . . . .  12
     5.2.  Define which YANG modules can be used in the datastore  .  12
     5.3.  Define which subset of YANG-modeled data applies  . . . .  13
     5.4.  Define how dynamic data is actualized . . . . . . . . . .  13
     5.5.  Define which protocols can be used  . . . . . . . . . . .  13
     5.6.  Define a module for the dynamic datastore . . . . . . . .  13
   6.  YANG Modules  . . . . . . . . . . . . . . . . . . . . . . . .  14
   7.  IANA Considerations . . . . . . . . . . . . . . . . . . . . .  18
     7.1.  Updates to the IETF XML Registry  . . . . . . . . . . . .  18
     7.2.  Updates to the YANG Module Names Registry . . . . . . . .  19
   8.  Security Considerations . . . . . . . . . . . . . . . . . . .  19
   9.  Acknowledgments . . . . . . . . . . . . . . . . . . . . . . .  19
   10. References  . . . . . . . . . . . . . . . . . . . . . . . . .  20
     10.1.  Normative References . . . . . . . . . . . . . . . . . .  20
     10.2.  Informative References . . . . . . . . . . . . . . . . .  21
   Appendix A.  Example Data . . . . . . . . . . . . . . . . . . . .  22
     A.1.  System Example  . . . . . . . . . . . . . . . . . . . . .  22
     A.2.  BGP Example . . . . . . . . . . . . . . . . . . . . . . .  25
       A.2.1.  Datastores  . . . . . . . . . . . . . . . . . . . . .  27
       A.2.2.  Adding a Peer . . . . . . . . . . . . . . . . . . . .  27
       A.2.3.  Removing a Peer . . . . . . . . . . . . . . . . . . .  28
     A.3.  Interface Example . . . . . . . . . . . . . . . . . . . .  29
       A.3.1.  Pre-provisioned Interfaces  . . . . . . . . . . . . .  29
       A.3.2.  System-provided Interface . . . . . . . . . . . . . .  30
   Appendix B.  Ephemeral Dynamic Datastore Example  . . . . . . . .  31
   Appendix C.  Implications on Data Models  . . . . . . . . . . . .  32

Bjorklund, et al.      Expires September 14, 2017               [Page 2]
Internet-Draft                                                March 2017

     C.1.  Proposed migration of existing YANG Data Models . . . . .  33
     C.2.  Standardization of new YANG Data Models . . . . . . . . .  34
   Appendix D.  Implications on other Documents  . . . . . . . . . .  34
     D.1.  Implications on YANG  . . . . . . . . . . . . . . . . . .  34
     D.2.  Implications on YANG Library  . . . . . . . . . . . . . .  34
     D.3.  Implications to YANG Guidelines . . . . . . . . . . . . .  35
       D.3.1.  Nodes with different config/state value sets  . . . .  35
       D.3.2.  Auto-configured or Auto-negotiated Values . . . . . .  35
     D.4.  Implications on NETCONF . . . . . . . . . . . . . . . . .  35
       D.4.1.  Introduction  . . . . . . . . . . . . . . . . . . . .  36
       D.4.2.  Overview of additions to NETCONF  . . . . . . . . . .  36
       D.4.3.  Overview of NETCONF version 2 . . . . . . . . . . . .  37
     D.5.  Implications on RESTCONF  . . . . . . . . . . . . . . . .  40
       D.5.1.  Introduction  . . . . . . . . . . . . . . . . . . . .  40
       D.5.2.  Overview of additions to RESTCONF . . . . . . . . . .  40
       D.5.3.  Overview of a possible new RESTCONF version . . . . .  42
   Appendix E.  Open Issues  . . . . . . . . . . . . . . . . . . . .  43
   Authors' Addresses  . . . . . . . . . . . . . . . . . . . . . . .  44

1.  Introduction

   This document provides an architectural framework for datastores as
   they are used by network management protocols such as NETCONF
   [RFC6241], RESTCONF [RFC8040] and the YANG [RFC7950] data modeling
   language.  Datastores are a fundamental concept binding network
   management data models to network management protocols.  Agreement on
   a common architectural model of datastores ensures that data models
   can be written in a network management protocol agnostic way.  This
   architectural framework identifies a set of conceptual datastores but
   it does not mandate that all network management protocols expose all
   these conceptual datastores.  This architecture is agnostic with
   regard to the encoding used by network management protocols.

2.  Terminology

   The keywords "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].

   This document defines the following terms:

   o  configuration data: Data that determines how a device behaves.
      This data is modeled in YANG using "config true" nodes.
      Configuration data can originate from different sources.

Bjorklund, et al.      Expires September 14, 2017               [Page 3]
Internet-Draft                                                March 2017

   o  static configuration data: Configuration data that is eventually
      persistent and used to get a device from its initial default state
      into its desired operational state.

   o  dynamic configuration data: Configuration data that is obtained
      dynamically during the operation of a device through interaction
      with other systems and not persistent.

   o  system configuration data: Configuration data that is supplied by
      the device itself.

   o  default configuration data: Configuration data that is not
      explicitly provided but for which a value defined in the data
      model is used.

   o  applied configuration data: Configuration data that is currently
      used by a device.  Applied configuration data consists of static
      configuration data and dynamic configuration data.

   o  state data: The additional data on a system that is not
      configuration data such as read-only status information and
      collected statistics.  State data is transient and modified by
      interactions with internal components or other systems.  State
      data is modeled in YANG using "config false" nodes.

   o  datastore: A conceptual place to store and access information.  A
      datastore might be implemented, for example, using files, a
      database, flash memory locations, or combinations thereof.  A
      datastore maps to an instantiated YANG data tree.

   o  configuration datastore: A datastore holding static configuration
      data that is required to get a device from its initial default
      state into a desired operational state.  A configuration datastore
      maps to an instantiated YANG data tree consisting of configuration
      data nodes and interior data nodes.

   o  running configuration datastore: A configuration datastore holding
      the complete static configuration currently active on the device.
      The running configuration datastore always exists.  It may include
      inactive configuration or template-mechanism-oriented
      configuration that require further expansion.

   o  intended configuration datastore: A configuration datastore
      holding the complete configuration currently active on the device.
      It does not include inactive configuration and it does include the
      expansion of any template mechanisms.

Bjorklund, et al.      Expires September 14, 2017               [Page 4]
Internet-Draft                                                March 2017

   o  candidate configuration datastore: A configuration datastore that
      can be manipulated without impacting the device's running
      configuration datastore and that can be committed to the running
      configuration datastore.  A candidate datastore may not be
      supported by all protocols or implementations.

   o  startup configuration datastore: The configuration datastore
      holding the configuration loaded by the device into the running
      configuration datastore when it boots.  A startup datastore may
      not be supported by all protocols or implementations.

   o  dynamic datastore: A datastore holding dynamic configuration data.

   o  operational state datastore: A datastore holding the currently
      active applied configuration data as well as the device's state
      data.

   o  origin: A metadata annotation indicating the origin of a data
      item.

   o  remnant data: Configuration data that remains in the system for a
      period of time after it has be removed from a configuration
      datastore.  The time period may be minimal, or may last until all
      resources used by the newly-deleted configuration data (e.g.,
      network connections, memory allocations, file handles) have been
      deallocated.

   The following additional terms are not datastore specific but
   commonly used and thus defined here as well:

   o  client: An entity that can access YANG-defined data on a server,
      over some network management protocol.

   o  server: An entity that provides access to YANG-defined data to a
      client, over some network management protocol.

   o  notification: A server-initiated message indicating that a certain
      event has been recognized by the server.

   o  remote procedure call: An operation that can be invoked by a
      client on a server.

3.  Introduction

   NETCONF [RFC6241] provides the following definitions:

Bjorklund, et al.      Expires September 14, 2017               [Page 5]
Internet-Draft                                                March 2017

   o  datastore: A conceptual place to store and access information.  A
      datastore might be implemented, for example, using files, a
      database, flash memory locations, or combinations thereof.

   o  configuration datastore: The datastore holding the complete set of
      configuration data that is required to get a device from its
      initial default state into a desired operational state.

   YANG 1.1 [RFC7950] provides the following refinements when NETCONF is
   used with YANG (which is the usual case but note that NETCONF was
   defined before YANG did exist):

   o  datastore: When modeled with YANG, a datastore is realized as an
      instantiated data tree.

   o  configuration datastore: When modeled with YANG, a configuration
      datastore is realized as an instantiated data tree with
      configuration data.

   [RFC6244] defined operational state data as follows:

   o  Operational state data is a set of data that has been obtained by
      the system at runtime and influences the system's behavior similar
      to configuration data.  In contrast to configuration data,
      operational state is transient and modified by interactions with
      internal components or other systems via specialized protocols.

   Section 4.3.3 of [RFC6244] discusses operational state and among
   other things mentions the option to consider operational state as
   being stored in another datastore.  Section 4.4 of this document then
   concludes that at the time of the writing, modeling state as a
   separate data tree is the recommended approach.

   Implementation experience and requests from operators
   [I-D.ietf-netmod-opstate-reqs], [I-D.openconfig-netmod-opstate]
   indicate that the datastore model initially designed for NETCONF and
   refined by YANG needs to be extended.  In particular, the notion of
   intended configuration and applied configuration has developed.

   Furthermore, separating operational state data from configuration
   data in a separate branch in the data model has been found
   operationally complicated, and typically impacts the readability of
   module definitions due to overuse of groupings.  The relationship
   between the branches is not machine readable and filter expressions
   operating on configuration data and on related operational state data
   are different.

Bjorklund, et al.      Expires September 14, 2017               [Page 6]
Internet-Draft                                                March 2017

3.1.  Original Model of Datastores

   The following drawing shows the original model of datastores as it is
   currently used by NETCONF [RFC6241]:

     +-------------+                 +-----------+
     | <candidate> |                 | <startup> |
     |  (ct, rw)   |<---+       +--->| (ct, rw)  |
     +-------------+    |       |    +-----------+
            |           |       |           |
            |         +-----------+         |
            +-------->| <running> |<--------+
                      | (ct, rw)  |
                      +-----------+
                            |
                            v
                     operational state  <--- control plane
                         (cf, ro)

     ct = config true; cf = config false
     rw = read-write; ro = read-only
     boxes denote datastores

   Note that this diagram simplifies the model: read-only (ro) and read-
   write (rw) is to be understood at a conceptual level.  In NETCONF,
   for example, support for the <candidate> and <startup> datastores is
   optional and the <running> datastore does not have to be writable.
   Furthermore, the <startup> datastore can only be modified by copying
   <running> to <startup> in the standardized NETCONF datastore editing
   model.  The RESTCONF protocol does not expose these differences and
   instead provides only a writable unified datastore, which hides
   whether edits are done through a <candidate> datastore or by directly
   modifying the <running> datastore or via some other implementation
   specific mechanism.  RESTCONF also hides how configuration is made
   persistent.  Note that implementations may also have additional
   datastores that can propagate changes to the <running> datastore.
   NETCONF explicitly mentions so called named datastores.

   Some observations:

   o  Operational state has not been defined as a datastore although
      there were proposals in the past to introduce an operational state
      datastore.

   o  The NETCONF <get/> operation returns the content of the <running>
      configuration datastore together with the operational state.  It
      is therefore necessary that config false data is in a different
      branch than the config true data if the operational state data can

Bjorklund, et al.      Expires September 14, 2017               [Page 7]
Internet-Draft                                                March 2017

      have a different lifetime compared to configuration data or if
      configuration data is not immediately or successfully applied.

   o  Several implementations have proprietary mechanisms that allow
      clients to store inactive data in the <running> datastore; this
      inactive data is only exposed to clients that indicate that they
      support the concept of inactive data; clients not indicating
      support for inactive data receive the content of the <running>
      datastore with the inactive data removed.  Inactive data is
      conceptually removed before validation.

   o  Some implementations have proprietary mechanisms that allow
      clients to define configuration templates in <running>.  These
      templates are expanded automatically by the system, and the
      resulting configuration is applied internally.

   o  Some operators have reported that it is essential for them to be
      able to retrieve the configuration that has actually been
      successfully applied, which may be a subset or a superset of the
      <running> configuration.

4.  Architectural Model of Datastores

   Below is a new conceptual model of datastores extending the original
   model in order to reflect the experience gained with the original
   model.

Bjorklund, et al.      Expires September 14, 2017               [Page 8]
Internet-Draft                                                March 2017

     +-------------+                 +-----------+
     | <candidate> |                 | <startup> |
     |  (ct, rw)   |<---+       +--->| (ct, rw)  |
     +-------------+    |       |    +-----------+
            |           |       |           |
            |         +-----------+         |
            +-------->| <running> |<--------+
                      | (ct, rw)  |
                      +-----------+
                            |
                            |        // e.g., removal of "inactive"
                            |        // nodes, expansion of templates
                            v
                      +------------+
                      | <intended> | // subject to validation
                      | (ct, ro)   |
                      +------------+
                            |
                            |        // e.g., missing resources, delays
                            |
                            |   +------ auto-discovery
                            |   +------ dynamic configuration protocols
                            |   +------ control-plane protocols
                            |   +------ dynamic datastores
                            |   |
                            v   v
                    +---------------+
                    | <operational> |
                    | (ct + cf, ro) |
                    +---------------+

     ct = config true; cf = config false
     rw = read-write; ro = read-only
     boxes denote datastores

4.1.  The <intended> Datastore

   The <intended> datastore is a read-only datastore that consists of
   config true nodes.  It is tightly coupled to <running>.  When data is
   written to <running>, the data that is to be validated is also
   conceptually written to <intended>.  Validation is performed on the
   contents of <intended>.

   On a traditional NETCONF implementation, <running> and <intended> are
   always the same.

Bjorklund, et al.      Expires September 14, 2017               [Page 9]
Internet-Draft                                                March 2017

   Currently there are no standard mechanisms defined that affect
   <intended> so that it would have different contents than <running>,
   but this architecture allows for such mechanisms to be defined.

   One example of such a mechanism is support for marking nodes as
   inactive in <running>.  Inactive nodes are not copied to <intended>,
   and are thus not taken into account when validating the
   configuration.

   Another example is support for templates.  Templates are expanded
   when copied into <intended>, and the expanded result is validated.

4.2.  Dynamic Datastores

   The model recognizes the need for dynamic datastores that are by
   definition not part of the persistent configuration of a device.  In
   some contexts, these have been termed ephemeral datastores since the
   information is ephemeral, i.e., lost upon reboot.  The dynamic
   datastores interact with the rest of the system through the
   <operational> datastore.

   Note that the ephemeral datastore discussed in I2RS documents maps to
   a dynamic datastore in the datastore model described here.

4.3.  The <operational> Datastore

   The <operational> datastore is a read-only datastore that consists of
   config true and config false nodes.  In the original NETCONF model
   the operational state only had config false nodes.  The reason for
   incorporating config true nodes here is to be able to expose all
   operational settings without having to replicate definitions in the
   data models.

   The <operational> datastore contains all configuration data actually
   used by the system, including all applied configuration, system-
   provided configuration and values defined by any supported data
   models.  In addition, the <operational> datastore also contains state
   data.

   Changes to configuration data may take time to percolate through to
   the <operational> datastore.  During this period, the <operational>
   datastore will return data nodes for both the previous and current
   configuration, as closely as possible tracking the current operation
   of the device.  These "remnants" of the previous configuration
   persist while the system has released resources used by the newly-
   deleted configuration data (e.g., network connections, memory
   allocations, file handles).

Bjorklund, et al.      Expires September 14, 2017              [Page 10]
Internet-Draft                                                March 2017

   As a result of these remnants, the semantic constraints defined in
   the data model cannot be relied upon for the <operational> datastore,
   since the system may have remnants whose constraints were valid with
   the previous configuration and that are not valid with the current
   configuration.  Since constraints on "config false" nodes may refer
   to "config true" nodes, remnants may force the violation of those
   constraints.  The constraints that may not hold include "when",
   "must", "min-elements", and "max-elements".  Note that syntactic
   constraints cannot be violated, including hierarchical organization,
   identifiers, and type-based constraints.

4.3.1.  Missing Resources

   The <intended> configuration can refer to resources that are not
   available or otherwise not physically present.  In these situations,
   these parts of the <intended> configuration are not applied.  The
   data appears in <intended> but does not appear in <operational>.

   A typical example is an interface configuration that refers to an
   interface that is not currently present.  In such a situation, the
   interface configuration remains in <intended> but the interface
   configuration will not appear in <operational>.

   Note that configuration validity cannot depend on the current state
   of such resources, since that would imply the removing a resource
   might render the configuration invalid.  This is unacceptable,
   especially given that rebooting such a device would fail to boot due
   to an invalid configuration.  Instead we allow configuration for
   missing resources to exist in <running> and <intended>, but it will
   not appear in <operational>.

4.3.2.  System-controlled Resources

   Sometimes resources are controlled by the device and the
   corresponding system controlled data appear in (and disappear from)
   <operational> dynamically.  If a system controlled resource has
   matching configuration in <intended> when it appears, the system will
   try to apply the configuration, which causes the configuration to
   appear in <operational> eventually (if application of the
   configuration was successful).

4.3.3.  Origin Metadata Annotation

   As data flows into the <operational> datastore, it is conceptually
   marked with a metadata annotation ([RFC7952]) that indicates its
   origin.  The "origin" metadata annotation is defined in Section 6.
   The values are YANG identities.  The following identities are
   defined:

Bjorklund, et al.      Expires September 14, 2017              [Page 11]
Internet-Draft                                                March 2017

     +-- origin
         +-- static
         +-- dynamic
         +-- default
         +-- system

   These identities can be further refined, e.g., there might be an
   identity "dhcp" derived from "dynamic".

   The "static" origin represents data provided by the <intended>
   datastore.  The "dynamic" origin represents data provided by a
   dynamic datastore.  The "default" origin represents data values
   specified in the data model, using either simple values in the
   "default" statement or any values described in the "description"
   statement.  Finally, the "system" origin represents data learned from
   the normal operational of the system, including control-plane
   protocols.

5.  Guidelines for Defining Dynamic Datastores

   The definition of a dynamic datastore SHOULD be provided in a
   document (e.g., an RFC) purposed to the definition of the dynamic
   datastore.  When it makes sense, more than one dynamic datastore MAY
   be defined in the same document (e.g., when the datastores are
   logically connected).  Each dynamic datastore's definition SHOULD
   address the points specified in the sections below.

5.1.  Define a name for the dynamic datastore

   Each dynamic datastores MUST have a name using the character set
   described by Section 6.2 of [RFC7950].  The name SHOULD be consistent
   in style and length to other datastore names described in this
   document.

   The datastore's name does not need to be globally unique, as it will
   be uniquely qualified by the namespace of the module in which it is
   defined (Section 5.6).  This means that names such as "running" and
   "operational" are valid datastore names.  However, it is usually
   desirable to avoid using the same name for multiple different
   datastores.

5.2.  Define which YANG modules can be used in the datastore

   Not all YANG modules may be used in all datastores.  Some datastores
   may constrain which data models can be used in them.  If it is
   desirable that a subset of all modules can be targeted to the dynamic
   datastore, then the documentation defining the dynamic datastore MUST
   use the mechanisms described in Appendix D.2 to provide the necessary

Bjorklund, et al.      Expires September 14, 2017              [Page 12]
Internet-Draft                                                March 2017

   hooks for module-designers to indicate that their module is to be
   accessible in the dynamic datastore.

5.3.  Define which subset of YANG-modeled data applies

   By default, the data in a dynamic datastore is modeled by all YANG
   statements in the available YANG modules.  However, it is possible to
   specify criteria YANG statements must satisfy in order to be present
   in a dynamic datastore.  For instance, maybe only config true nodes
   are present, or config false nodes that also have a specific YANG
   extension (e.g., i2rs:ephemeral true) are present in the dynamic
   datastore.

5.4.  Define how dynamic data is actualized

   The diagram in Section 4 depicts dynamic datastores feeding into the
   <operational> datastore.  How this interaction occurs must be defined
   by the dynamic datastore.  In some cases, it may occur implicitly, as
   soon as the data is put into the dynamic datastore while, in other
   cases, an explicit action (e.g., an RPC) may be required to trigger
   the application of the dynamic datastore's data.

5.5.  Define which protocols can be used

   By default, it is assumed that both the NETCONF and RESTCONF
   protocols can be used to interact with a dynamic datastore.  However,
   it may be that only a specific protocol can be used (e.g., Forces) or
   that a subset of all protocol operations or capabilities are
   available (e.g., no locking, no xpath-based filtering, etc.).

5.6.  Define a module for the dynamic datastore

   Each dynamic datastore MUST be defined by a YANG module.  This module
   is used by servers to indicate (e.g., via YANG Library) their support
   for the dynamic datastore.

   The YANG module MUST import the "ietf-datastores" and "ietf-origin"
   modules, defined in this document.  This is necessary in order to
   access the base identities they define.

   The YANG module MUST define an identity that uses the "ds:datastore"
   identity as its base.  This identity is necessary so that the
   datastore can be referenced in protocol operations (e.g.,
   <get-data>).

   The YANG module MUST define an identity that uses the "or:dynamic"
   identity as its base.  This identity is necessary so that data

Bjorklund, et al.      Expires September 14, 2017              [Page 13]
Internet-Draft                                                March 2017

   originating from the datastore can be identified as such via the
   "origin" metadata attribute defined in Section 6.

   An example of these guidelines in use is provided in Appendix B.

6.  YANG Modules

   <CODE BEGINS> file "ietf-datastores@2017-03-13.yang"

   module ietf-datastores {
     yang-version 1.1;
     namespace "urn:ietf:params:xml:ns:yang:ietf-datastores";
     prefix ds;

     organization
       "IETF NETMOD (NETCONF Data Modeling Language) Working Group";

     contact
       "WG Web:   <https://datatracker.ietf.org/wg/netmod/>

        WG List:  <mailto:netmod@ietf.org>

        Author:   Martin Bjorklund
                  <mailto:mbj@tail-f.com>

        Author:   Juergen Schoenwaelder
                  <mailto:j.schoenwaelder@jacobs-university.de>

        Author:   Phil Shafer
                  <mailto:phil@juniper.net>

        Author:   Kent Watsen
                  <mailto:kwatsen@juniper.net>

        Author:   Rob Wilton
                  <rwilton@cisco.com>";

     description
       "This YANG module defines a set of identities for datastores.
        These identities can be used to identify datastores in protocol
        operations.

        Copyright (c) 2017 IETF Trust and the persons identified as
        authors of the code.  All rights reserved.

        Redistribution and use in source and binary forms, with or
        without modification, is permitted pursuant to, and subject to
        the license terms contained in, the Simplified BSD License set

Bjorklund, et al.      Expires September 14, 2017              [Page 14]
Internet-Draft                                                March 2017

        forth in Section 4.c of the IETF Trust's Legal Provisions
        Relating to IETF Documents
        (http://trustee.ietf.org/license-info).

        This version of this YANG module is part of RFC XXXX
        (http://www.rfc-editor.org/info/rfcxxxx); see the RFC itself
        for full legal notices.";

     revision 2017-03-13 {
       description
         "Initial revision.";
       reference
         "RFC XXXX: Network Management Datastore Architecture";
     }

     /*
      * Identities
      */

     identity datastore {
       description
        "Abstract base identity for datastore identities.";
     }

     identity static {
       description
        "Abstract base identity for static configuration datastores.";
     }

     identity dynamic {
       description
        "Abstract base identity for dynamic configuration datastores.";
     }

     identity running {
       base static;
       description
        "The 'running' datastore.";
     }

     identity candidate {
       base static;
       description
        "The 'candidate' datastore.";
     }

     identity startup {
       base static;

Bjorklund, et al.      Expires September 14, 2017              [Page 15]
Internet-Draft                                                March 2017

       description
        "The 'startup' datastore.";
     }

     identity intended {
       base static;
       description
        "The 'intended' datastore.";
     }

     identity operational {
       base datastore;
       description
        "The 'operational' state datastore.";
     }

   }

   <CODE ENDS>

   <CODE BEGINS> file "ietf-datastores@2017-03-13.yang"

   module ietf-origin {
     yang-version 1.1;
     namespace "urn:ietf:params:xml:ns:yang:ietf-origin";
     prefix or;

     import ietf-yang-metadata {
       prefix md;
     }

     organization
       "IETF NETMOD (NETCONF Data Modeling Language) Working Group";

     contact
       "WG Web:   <https://datatracker.ietf.org/wg/netmod/>

        WG List:  <mailto:netmod@ietf.org>

        Author:   Martin Bjorklund
                  <mailto:mbj@tail-f.com>

        Author:   Juergen Schoenwaelder
                  <mailto:j.schoenwaelder@jacobs-university.de>

        Author:   Phil Shafer
                  <mailto:phil@juniper.net>

Bjorklund, et al.      Expires September 14, 2017              [Page 16]
Internet-Draft                                                March 2017

        Author:   Kent Watsen
                  <mailto:kwatsen@juniper.net>

        Author:   Rob Wilton
                  <rwilton@cisco.com>";

     description
       "This YANG module defines an 'origin' metadata annotation, and a
        set of identities for the origin value.  The 'origin' metadata
        annotation is used to mark data in the 'operational'
        datastore with information on where the data originated.

        Copyright (c) 2017 IETF Trust and the persons identified as
        authors of the code.  All rights reserved.

        Redistribution and use in source and binary forms, with or
        without modification, is permitted pursuant to, and subject to
        the license terms contained in, the Simplified BSD License set
        forth in Section 4.c of the IETF Trust's Legal Provisions
        Relating to IETF Documents
        (http://trustee.ietf.org/license-info).

        This version of this YANG module is part of RFC XXXX
        (http://www.rfc-editor.org/info/rfcxxxx); see the RFC itself
        for full legal notices.";

     revision 2017-03-13 {
       description
         "Initial revision.";
       reference
         "RFC XXXX: Network Management Datastore Architecture";
     }

     /*
      * Identities
      */

     identity origin {
       description
         "Abstract base identity for the origin annotation.";
     }

     identity static {
       base origin;
       description
         "Denotes data from static configuration (e.g., <intended>).";
     }

Bjorklund, et al.      Expires September 14, 2017              [Page 17]
Internet-Draft                                                March 2017

     identity dynamic {
       base origin;
       description
         "Denotes data from dynamic configuration protocols
          or dynamic datastores (e.g., DHCP).";
     }

     identity system {
       base origin;
       description
         "Denotes data created by the system independently of what
          has been configured.";
     }

     identity default {
       base origin;
       description
         "Denotes data that does not have an explicitly configured
          value, but has a default value in use.  Covers both simple
          defaults and defaults defined via an explanation in a
          description statement.";
     }

     /*
      * Metadata annotations
      */

     md:annotation origin {
       type identityref {
         base origin;
       }
     }

   }

   <CODE ENDS>

7.  IANA Considerations

7.1.  Updates to the IETF XML Registry

   This document registers two URIs in the IETF XML registry [RFC3688].
   Following the format in [RFC3688], the following registrations are
   requested:

Bjorklund, et al.      Expires September 14, 2017              [Page 18]
Internet-Draft                                                March 2017

      URI: urn:ietf:params:xml:ns:yang:ietf-datastores
      Registrant Contact: The IESG.
      XML: N/A, the requested URI is an XML namespace.

      URI: urn:ietf:params:xml:ns:yang:ietf-origin
      Registrant Contact: The IESG.
      XML: N/A, the requested URI is an XML namespace.

7.2.  Updates to the YANG Module Names Registry

   This document registers two YANG modules in the YANG Module Names
   registry [RFC6020].  Following the format in [RFC6020], the the
   following registrations are requested:

      name:         ietf-datastores
      namespace:    urn:ietf:params:xml:ns:yang:ietf-datastores
      prefix:       ds
      reference:    RFC XXXX

      name:         ietf-origin
      namespace:    urn:ietf:params:xml:ns:yang:ietf-origin
      prefix:       or
      reference:    RFC XXXX

8.  Security Considerations

   This document discusses a conceptual model of datastores for network
   management using NETCONF/RESTCONF and YANG.  It has no security
   impact on the Internet.

9.  Acknowledgments

   This document grew out of many discussions that took place since
   2010.  Several Internet-Drafts ([I-D.bjorklund-netmod-operational],
   [I-D.wilton-netmod-opstate-yang], [I-D.ietf-netmod-opstate-reqs],
   [I-D.kwatsen-netmod-opstate], [I-D.openconfig-netmod-opstate]) and
   [RFC6244] touched on some of the problems of the original datastore
   model.  The following people were authors to these Internet-Drafts or
   otherwise actively involved in the discussions that led to this
   document:

   o  Lou Berger, LabN Consulting, L.L.C., <lberger@labn.net>

   o  Andy Bierman, YumaWorks, <andy@yumaworks.com>

   o  Marcus Hines, Google, <hines@google.com>

   o  Christian Hopps, Deutsche Telekom, <chopps@chopps.org>

Bjorklund, et al.      Expires September 14, 2017              [Page 19]
Internet-Draft                                                March 2017

   o  Acee Lindem, Cisco Systems, <acee@cisco.com>

   o  Ladislav Lhotka, CZ.NIC, <lhotka@nic.cz>

   o  Thomas Nadeau, Brocade Networks, <tnadeau@lucidvision.com>

   o  Anees Shaikh, Google, <aashaikh@google.com>

   o  Rob Shakir, Google, <robjs@google.com>

   Juergen Schoenwaelder was partly funded by Flamingo, a Network of
   Excellence project (ICT-318488) supported by the European Commission
   under its Seventh Framework Programme.

10.  References

10.1.  Normative References

   [RFC2119]  Bradner, S., "Key words for use in RFCs to Indicate
              Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/
              RFC2119, March 1997,
              <http://www.rfc-editor.org/info/rfc2119>.

   [RFC6241]  Enns, R., Ed., Bjorklund, M., Ed., Schoenwaelder, J., Ed.,
              and A. Bierman, Ed., "Network Configuration Protocol
              (NETCONF)", RFC 6241, DOI 10.17487/RFC6241, June 2011,
              <http://www.rfc-editor.org/info/rfc6241>.

   [RFC7895]  Bierman, A., Bjorklund, M., and K. Watsen, "YANG Module
              Library", RFC 7895, DOI 10.17487/RFC7895, June 2016,
              <http://www.rfc-editor.org/info/rfc7895>.

   [RFC7950]  Bjorklund, M., Ed., "The YANG 1.1 Data Modeling Language",
              RFC 7950, DOI 10.17487/RFC7950, August 2016,
              <http://www.rfc-editor.org/info/rfc7950>.

   [RFC7952]  Lhotka, L., "Defining and Using Metadata with YANG", RFC
              7952, DOI 10.17487/RFC7952, August 2016,
              <http://www.rfc-editor.org/info/rfc7952>.

   [RFC8040]  Bierman, A., Bjorklund, M., and K. Watsen, "RESTCONF
              Protocol", RFC 8040, DOI 10.17487/RFC8040, January 2017,
              <http://www.rfc-editor.org/info/rfc8040>.

Bjorklund, et al.      Expires September 14, 2017              [Page 20]
Internet-Draft                                                March 2017

10.2.  Informative References

   [I-D.bjorklund-netmod-operational]
              Bjorklund, M. and L. Lhotka, "Operational Data in NETCONF
              and YANG", draft-bjorklund-netmod-operational-00 (work in
              progress), October 2012.

   [I-D.ietf-netmod-opstate-reqs]
              Watsen, K. and T. Nadeau, "Terminology and Requirements
              for Enhanced Handling of Operational State", draft-ietf-
              netmod-opstate-reqs-04 (work in progress), January 2016.

   [I-D.ietf-netmod-rfc6087bis]
              Bierman, A., "Guidelines for Authors and Reviewers of YANG
              Data Model Documents", draft-ietf-netmod-rfc6087bis-12
              (work in progress), March 2017.

   [I-D.kwatsen-netmod-opstate]
              Watsen, K., Bierman, A., Bjorklund, M., and J.
              Schoenwaelder, "Operational State Enhancements for YANG,
              NETCONF, and RESTCONF", draft-kwatsen-netmod-opstate-02
              (work in progress), February 2016.

   [I-D.openconfig-netmod-opstate]
              Shakir, R., Shaikh, A., and M. Hines, "Consistent Modeling
              of Operational State Data in YANG", draft-openconfig-
              netmod-opstate-01 (work in progress), July 2015.

   [I-D.wilton-netmod-opstate-yang]
              Wilton, R., ""With-config-state" Capability for NETCONF/
              RESTCONF", draft-wilton-netmod-opstate-yang-02 (work in
              progress), December 2015.

   [RFC3688]  Mealling, M., "The IETF XML Registry", BCP 81, RFC 3688,
              DOI 10.17487/RFC3688, January 2004,
              <http://www.rfc-editor.org/info/rfc3688>.

   [RFC6020]  Bjorklund, M., Ed., "YANG - A Data Modeling Language for
              the Network Configuration Protocol (NETCONF)", RFC 6020,
              DOI 10.17487/RFC6020, October 2010,
              <http://www.rfc-editor.org/info/rfc6020>.

   [RFC6243]  Bierman, A. and B. Lengyel, "With-defaults Capability for
              NETCONF", RFC 6243, DOI 10.17487/RFC6243, June 2011,
              <http://www.rfc-editor.org/info/rfc6243>.

Bjorklund, et al.      Expires September 14, 2017              [Page 21]
Internet-Draft                                                March 2017

   [RFC6244]  Shafer, P., "An Architecture for Network Management Using
              NETCONF and YANG", RFC 6244, DOI 10.17487/RFC6244, June
              2011, <http://www.rfc-editor.org/info/rfc6244>.

Appendix A.  Example Data

   The use of datastores is complex, and many of the subtle effects are
   more easily presented using examples.  This section presents a series
   of example data models with some sample contents of the various
   datastores.

A.1.  System Example

   In this example, the following fictional module is used:

   module example-system {
     yang-version 1.1;
     namespace urn:example:system;
     prefix sys;

     import ietf-inet-types {
       prefix inet;
     }

     container system {
       leaf hostname {
         type string;
       }

       list interface {
         key name;

         leaf name {
           type string;
         }

         container auto-negotiation {
           leaf enabled {
             type boolean;
             default true;
           }
           leaf speed {
             type uint32;
             units mbps;
             description
               "The advertised speed, in mbps.";
           }
         }

Bjorklund, et al.      Expires September 14, 2017              [Page 22]
Internet-Draft                                                March 2017

         leaf speed {
           type uint32;
           units mbps;
           config false;
           description
             "The speed of the interface, in mbps.";
         }

         list address {
           key ip;

           leaf ip {
             type inet:ip-address;
           }
           leaf prefix-length {
             type uint8;
           }
         }
       }
     }
   }

   The operator has configured the host name and two interfaces, so the
   contents of <intended> is:

Bjorklund, et al.      Expires September 14, 2017              [Page 23]
Internet-Draft                                                March 2017

   <system xmlns="urn:example:system">

     <hostname>foo</hostname>

     <interface>
       <name>eth0</name>
       <auto-negotiation>
         <speed>1000</speed>
       </auto-negotiation>
       <address>
         <ip>2001:db8::10</ip>
         <prefix-length>32</prefix-length>
       </address>
     </interface>

     <interface>
       <name>eth1</name>
       <address>
         <ip>2001:db8::20</ip>
         <prefix-length>32</prefix-length>
       </address>
     </interface>

   </system>

   The system has detected that the hardware for one of the configured
   interfaces ("eth1") is not yet present, so the configuration for that
   interface is not applied.  Further, the system has received a host
   name and an additional IP address for "eth0" over DHCP.  In addition
   to a default value, a loopback interface is automatically added by
   the system, and the result of the "speed" auto-negotiation.  All of
   this is reflected in <operational>:

Bjorklund, et al.      Expires September 14, 2017              [Page 24]
Internet-Draft                                                March 2017

   <system
       xmlns="urn:example:system"
       xmlns:or="urn:ietf:params:xml:ns:yang:ietf-origin">

     <hostname or:origin="or:dynamic">bar</hostname>

     <interface or:origin="or:static">
       <name>eth0</name>
       <auto-negotiation>
         <enabled or:origin="or:default">true</enabled>
         <speed>1000</speed>
       </auto-negotiation>
       <speed>100</speed>
       <address>
         <ip>2001:db8::10</ip>
         <prefix-length>32</prefix-length>
       </address>
       <address or:origin="or:dynamic">
         <ip>2001:db8::1:100</ip>
         <prefix-length>32</prefix-length>
       </address>
     </interface>

     <interface or:origin="or:system">
       <name>lo0</name>
       <address>
         <ip>::1</ip>
         <prefix-length>128</prefix-length>
       </address>
     </interface>

   </system>

A.2.  BGP Example

   Consider the following piece of a ersatz BGP module:

Bjorklund, et al.      Expires September 14, 2017              [Page 25]
Internet-Draft                                                March 2017

       container bgp {
         leaf local-as {
           type uint32;
         }
         leaf peer-as {
           type uint32;
         }
         list peer {
           key name;
           leaf name {
             type ipaddress;
           }
           leaf local-as {
             type uint32;
             description
               ".... Defaults to ../local-as";
           }
           leaf peer-as {
             type uint32;
             description
                "... Defaults to ../peer-as";
           }
           leaf local-port {
             type inet:port;
           }
           leaf remote-port {
             type inet:port;
             default 179;
           }
           leaf state {
             config false;
             type enumeration {
               enum init;
               enum established;
               enum closing;
             }
           }
         }
       }

   In this example model, both bgp/peer/local-as and bgp/peer/peer-as
   have complex hierarchical values, allowing the user to specify
   default values for all peers in a single location.

   The model also follows the pattern of fully integrating state
   ("config false") nodes with configuration ("config true") nodes.
   There is not separate "bgp-state" hierarchy, with the accompanying

Bjorklund, et al.      Expires September 14, 2017              [Page 26]
Internet-Draft                                                March 2017

   repetition of containment and naming nodes.  This makes the model
   simpler and more readable.

A.2.1.  Datastores

   Each datastore represents differing views of these data nodes.  The
   <running> datastore will hold the configuration data provided by the
   user, for example a single BGP peer.  The <intended> datastore will
   conceptually hold the data as validated, after the removal of data
   not intended for validation and after any local template mechanisms
   are performed.  The <operational> datastore will show data from
   <intended> as well as any "config false" nodes.

A.2.2.  Adding a Peer

   If the user configures a single BGP peer, then that peer will be
   visible in both the <running> and <intended> datastores.  It may also
   appear in the <candidate> datastore, if the server supports the
   "candidate" feature.  Retrieving the peer will return only the user-
   specified values.

   No time delay should exist between the appearance of the peer in
   <running> and <intended>.

   In this scenario, we've added the following to <running>:

     <bgp>
       <local-as>64642</local-as>
       <peer-as>65000</peer-as>
       <peer>
         <name>10.1.2.3</name>
       </peer>
     </bgp>

A.2.2.1.  <operational>

   The <operational> datastore will contain the fully expanded peer
   data, including "config false" nodes.  In our example, this means the
   "state" node will appear.

   In addition, the <operational> datastore will contain the "currently
   in use" values for all nodes.  This means that local-as and peer-as
   will be populated even if they are not given values in <intended>.
   The value of bgp/local-as will be used if bgp/peer/local-as is not
   provided; bgp/peer-as and bgp/peer/peer-as will have the same
   relationship.  In the operational view, this means that every peer
   will have values for their local-as and peer-as, even if those values

Bjorklund, et al.      Expires September 14, 2017              [Page 27]
Internet-Draft                                                March 2017

   are not explicitly configured but are provided by bgp/local-as and
   bgp/peer-as.

   Each BGP peer has a TCP connection associated with it, using the
   values of local-port and remote-port from the intended datastore.  If
   those values are not supplied, the system will select values.  When
   the connection is established, the <operational> datastore will
   contain the current values for the local-port and remote-port nodes
   regardless of the origin.  If the system has chosen the values, the
   "origin" attribute will be set to "operational".  Before the
   connection is established, one or both of the nodes may not appear,
   since the system may not yet have their values.

     <bgp origin="or:static" xmlns="urn:example:bgp">
       <local-as origin="or:static">64642</local-as>
       <peer-as origin="or:static">65000</peer-as>
       <peer origin="or:static">
         <name origin="or:static">10.1.2.3</name>
         <local-as origin="or:default">64642</local-as>
         <peer-as origin="or:default">65000</peer-as>
         <local-port origin="or:system">60794</local-port>
         <remote-port origin="or:default">179</remote-port>
       </peer>
     </bgp>

A.2.3.  Removing a Peer

   Changes to configuration data may take time to percolate through the
   various software components involved.  During this period, it is
   imperative to continue to give an accurate view of the working of the
   device.  The <operational> datastore will return data nodes for both
   the previous and current configuration, as closely as possible
   tracking the current operation of the device.

   Consider the scenario where a client removes a BGP peer.  When a peer
   is removed, the operational state will continue to reflect the
   existence of that peer until the peer's resources are released,
   including closing the peer's connection.  During this period, the
   current data values will continue to be visible in the <operational>
   datastore, with the "origin" attribute set to indicate the origin of
   the original data.

Bjorklund, et al.      Expires September 14, 2017              [Page 28]
Internet-Draft                                                March 2017

     <bgp origin="or:static">
       <local-as origin="or:static">64642</local-as>
       <peer-as origin="or:static">65000</peer-as>
       <peer origin="or:static">
         <name origin="or:static">10.1.2.3</name>
         <local-as origin="or:default">64642</local-as>
         <peer-as origin="or:default">65000</peer-as>
         <local-port origin="or:static">60794</local-port>
         <remote-port origin="or:static">179</remote-port>
       </peer>
     </bgp>

   Once resources are released and the connection is closed, the peer's
   data is removed from the <operational> datastore.

A.3.  Interface Example

   In this section, we'll use this simple interface data model:

     container interfaces {
       list interface {
         key name;
         leaf name {
           type string;
         }
         leaf description {
           type string;
         }
         leaf mtu {
           type uint;
         }
         leaf ipv4-address {
           type inet:ipv4-address;
         }
       }
     }

A.3.1.  Pre-provisioned Interfaces

   One common issue in networking devices is the support of Field
   Replaceable Units (FRUs) that can be inserted and removed from the
   device without requiring a reboot or interfering with normal
   operation.  These FRUs are typically interface cards, and the devices
   support pre-provisioning of these interfaces.

   If a client creates an interface "et-0/0/0" but the interface does
   not physically exist at this point, then the <intended> datastore
   might contain the following:

Bjorklund, et al.      Expires September 14, 2017              [Page 29]
Internet-Draft                                                March 2017

     <interfaces>
       <interface>
         <name>et-0/0/0</name>
         <description>Test interface</description>
       </interface>
     </interfaces>

   Since the interface does not exist, this data does not appear in the
   <operational> datastore.

   When a FRU containing this interface is inserted, the system will
   detect it and process the associated configuration.  The
   <operational> will contain the data from <intended>, as well as the
   "config false" nodes, such as the current value of the interface's
   MTU.

     <interfaces origin="or:static">
       <interface origin="or:static">
         <name origin="or:static">et-0/0/0</name>
         <description origin="or:static">Test interface</description>
         <mtu origin="or:system">1500</mtu>
       </interface>
     </interfaces>

   If the FRU is removed, the interface data is removed from the
   <operational> datastore.

A.3.2.  System-provided Interface

   Imagine if the system provides a loopback interface (named "lo0")
   with a default ipv4-address of "127.0.0.1".  The system will only
   provide configuration for this interface if the is no data for it in
   <intended>.

   When no configuration for "lo0" appears in <intended>, then
   <operational> will show the system-provided data:

     <interfaces origin="or:static">
       <interface origin="or:system">
         <name origin="or:system">lo0</name>
         <ipv4-address origin="or:system">127.0.0.1</ipv4-address>
       </interface>
     </interfaces>

   When configuration for "lo0" does appear in <intended>, then
   <operational> will show that data with the origin set to "intended".
   If the "ipv4-address" is not provided, then the system-provided value
   will appear as follows:

Bjorklund, et al.      Expires September 14, 2017              [Page 30]
Internet-Draft                                                March 2017

     <interfaces origin="or:static">
       <interface origin="or:static">
         <name origin="or:static">lo0</name>
         <description origin="or:static">loopback</description>
         <ipv4-address origin="or:system">127.0.0.1</ipv4-address>
       </interface>
     </interfaces>

Appendix B.  Ephemeral Dynamic Datastore Example

   The section defines documentation for an example dynamic datastore
   using the guidelines provided in Section 5.  While this example is
   very terse, it is expected to be that a standalone RFC would be
   needed when fully expanded.

   This example defines a dynamic datastore called "ephemeral", which is
   loosely modeled after the work done in the I2RS working group.

     1. Name            : ephemeral
     2. YANG modules    : all (default)
     3. YANG statements : config false + ephemeral true
     4. How applied     : automatic
     5. Protocols       : NC/RC (default)
     6. YANG Module     : (see below)

Bjorklund, et al.      Expires September 14, 2017              [Page 31]
Internet-Draft                                                March 2017

   module example-ds-ephemeral {
     yang-version 1.1;
     namespace "urn:example:ds-ephemeral";
     prefix eph;

     import ietf-datastores {
       prefix ds;
     }
     import ietf-origin {
       prefix or;
     }

     // add datastore identity
     identity ds-ephemeral {
       base ds:datastore;
       description
         "The 'ephemeral' datastore.";
     }

     // add origin identity
     identity or-ephemeral {
       base or:dynamic;
       description
         "Denotes data from the ephemeral dynamic datastore.";
     }

     // define ephemeral extension
     extension ephemeral {
       argument "value";
       description
         "This extension is mixed into config false YANG nodes to
          indicate that they are writable nodes in the 'ephemeral'
          datastore.  This statement takes a single argument
          representing a boolean having the values 'true' and 'false'.
          The default value is 'false'.";
     }
   }

Appendix C.  Implications on Data Models

   Since the NETCONF <get/> operation returns the content of the
   <running> configuration datastore and the operational state together
   in one tree, data models were often forced to branch at the top-level
   into a config true branch and a structurally similar config false
   branch that replicated some of the config true nodes and added state
   nodes.  With the datastore model described here this is not needed
   anymore since the different datastores handle the different lifetimes
   of data objects.  Introducing this model together with the

Bjorklund, et al.      Expires September 14, 2017              [Page 32]
Internet-Draft                                                March 2017

   deprecation of the <get/> operation makes it possible to write
   simpler models.

C.1.  Proposed migration of existing YANG Data Models

   For standards based YANG modules that have already been published,
   that are using split config and state trees, it is planned that these
   modules are updated with new revisions containing the following
   changes:

   o  The top level module description is updated to indicate that the
      module conforms to the revised datastore architecture with a
      combined config and state tree, and that the existing state tree
      nodes are deprecated, to be obsoleted over time.

   o  All status "current" data nodes under the existing "state" trees
      are copied to the equivalent place under the "config" tree:

      *  If a node with the same name and type already exists under the
         equivalent path in the config tree then the nodes are merged
         and the description updated.

      *  If a node with the same name but different type exists under
         the equivalent path in the config tree, then the module authors
         must choose the appropriate mechanism to combine the config and
         state nodes in a backwards compatible way based on the data
         model design guidelines below.  This may require the state node
         to be added to the config tree with a modified name.  This
         scenario is expected to be relatively uncommon.

      *  If no node with the same name and path already exists under the
         config tree then the state node schema is copied verbatim into
         the config tree.

      *  As the state nodes are copied into the config trees, any
         leafrefs that reference other nodes in the state tree are
         adjusted to reference the equivalent path in the config tree.

      *  All status "current" nodes under the existing "state" trees are
         marked as "status" deprecated.

   o  Augmentations are similarly handled to data nodes as described
      above.

Bjorklund, et al.      Expires September 14, 2017              [Page 33]
Internet-Draft                                                March 2017

C.2.  Standardization of new YANG Data Models

   New standards based YANG modules, or those in active development,
   should be designed to conform to the revised datastore architecture,
   following the design guidelines described below, and only need to
   provide combined config/state trees.

Appendix D.  Implications on other Documents

   The sections below describe the authors' thoughts on how various
   other documents may be updated to support the datastore architecture
   described in this document.  They have been incorporated as an
   appendix of this document to facilitate easier review, but the
   expectation is that this work will be moved into another document as
   soon as the appropriate working group decides to take on the work.

D.1.  Implications on YANG

   Note: This section describes the authors' thoughts on how YANG
   [RFC7950] could be updated to support the datastore architecture
   described in this document.  It has been incorporated here as a
   temporary measure to facilitate easier review, but the expectation is
   that this work will be owned and standardized via the NETCONF working
   group.

   o  Some clarifications may be needed if this datastore model is
      adopted.  YANG currently describes validation in terms of the
      <running> configuration datastore while it really happens on the
      <intended> configuration datastore.

D.2.  Implications on YANG Library

   Note: This section describes the authors' thoughts on how YANG
   Library [RFC7895] could be updated to support the datastore
   architecture described in this document.  It has been incorporated
   here as a temporary measure to facilitate easier review, but the
   expectation is that this work will be owned and standardized via the
   NETCONF working group.

   With the introduction of multiple datastores, it is important that a
   server can advertise to clients which modules are supported in the
   different datastores implemented by the server.  In order to do this,
   we propose that the "ietf-yang-module" ([RFC7895]) is revised, with
   the following addition to the "module" list in the "module-list"
   grouping:

Bjorklund, et al.      Expires September 14, 2017              [Page 34]
Internet-Draft                                                March 2017

     leaf-list datastore {
       type identityref {
         base ds:datastore;
       }
       description
         "The datastores in which this module is supported.";

     }

D.3.  Implications to YANG Guidelines

   Note: This section describes the authors' thoughts on how Guidelines
   for Authors and Reviewers of YANG Data Model Documents
   [I-D.ietf-netmod-rfc6087bis] could be updated to support the
   datastore architecture described in this document.  It has been
   incorporated here as a temporary measure to facilitate easier review,
   but the expectation is that this work will be owned and standardized
   via the NETCONF working group.

   It is important to design data models with clear semantics that work
   equally well for instantiation in a configuration datastore and
   instantiation in the <operational> datastore.

D.3.1.  Nodes with different config/state value sets

   There may be some differences in the value set of some nodes that are
   used for both configuration and state.  At this point of time, these
   are considered to be rare cases that can be dealt with using
   different nodes for the configured and state values.

D.3.2.  Auto-configured or Auto-negotiated Values

   Sometimes configuration leafs support special values that instruct
   the system to automatically configure a value.  An example is an MTU
   that is configured to "auto" to let the system determine a suitable
   MTU value.  Another example is Ethernet auto-negotiation of link
   speed.  In such a situation, it is recommended to model this as two
   separate leafs, one config true leaf for the input to the auto-
   negotiation process, and one config false leaf for the output from
   the process.

D.4.  Implications on NETCONF

   Note: This section describes the authors' thoughts on how NETCONF
   [RFC6241] could be updated to support the datastore architecture
   described in this document.  It has been incorporated here as a
   temporary measure to facilitate easier review, but the expectation is

Bjorklund, et al.      Expires September 14, 2017              [Page 35]
Internet-Draft                                                March 2017

   that this work will be owned and standardized via the NETCONF working
   group.

D.4.1.  Introduction

   The NETCONF protocol [RFC6241] defines a simple mechanism through
   which a network device can be managed, configuration data information
   can be retrieved, and new configuration data can be uploaded and
   manipulated.

   NETCONF already has support for configuration datastores, but it does
   not define an operational datastore.  Instead, it provides the <get>
   operation that returns the contents of the <running> datastore along
   with all config false leaves.  However, this <get> operation is
   incompatible with the new datastore architecture defined in this
   document, and hence should be deprecated.

   There are two possible ways that NETCONF could be extended to support
   the new architecture: Either as new optional capabilities extending
   the current version of NETCONF (v1.1, [RFC6241]), or by defining a
   new version of NETCONF.

   Many of the required additions are common to both approaches, and are
   described below.  A following section then describes the benefits of
   defining a new NETCONF version, and the additional changes that would
   entail.

D.4.2.  Overview of additions to NETCONF

   o  A new "supported datastores" capability allows a device to list
      all datastores it supports.  Implementations can choose which
      datastores they expose, but MUST at least expose both the
      <running> and <operational> datastores.  They MAY expose
      additional datastores, such as <intended>, <candidate>, etc.

   o  A new <get-data> operation is introduced that allows the client to
      return the contents of a datastore.  For configuration datastores,
      this operation returns the same data that would be returned by the
      existing <get-config> operation.

   o  Some form of new filtering mechanism is required to allow the
      device to filter the data based on the YANG metadata in addition
      to other filters (such as the subtree filter).  See also
      Appendix E.

   o  A new "with-metadata" capability allows a device to indicate that
      it supports the capability of including YANG metadata annotations
      in the responses to <get> and <get-config> requests.  This is

Bjorklund, et al.      Expires September 14, 2017              [Page 36]
Internet-Draft                                                March 2017

      achieved in a similar way to with-defaults [RFC6243], by
      introducing a <with-metadata> XML element to <get> and
      <get-config> requests.

      *  The capability would allow a device to indicate which types of
         metadata are supported.

      *  The XML element would specify which types of metadata are
         included in the response.

   o  The handling of defaults for the new configuration datastores is
      as described in with-defaults [RFC6243], but that does not apply
      for the operational state datastore that defines new semantics.

D.4.2.1.  Operational State Datastore Defaults Handling

   The normal semantics for the <operational> datastore are that all
   values that match the default specified in the schema are included in
   response to requests on the operational state datastore.  This is
   equivalent to the "report-all" mode of the with-defaults handling.

   The "metadata-filter" query parameter can be used to exclude nodes
   with origin metadata matching "default", that would exclude nodes
   that match the default value specified in the schema.

   If the server cannot return a value for any reason (e.g., the server
   cannot determine the value, or the value that would be returned is
   outside the allowed leaf value range) then the server can choose to
   not return any value for a particular leaf, which MUST be interpreted
   by the client as the value of that leaf not being known, rather than
   implicitly having the default value.

D.4.3.  Overview of NETCONF version 2

   This section describes NETCONF version 2, by explaining the
   differences to NETCONF version 1.1.  Where not explicitly specified,
   the behavior of NETCONF version 2 is the same as for NETCONF version
   1.1 [RFC6241].

D.4.3.1.  Benefits of defining a new NETCONF version

   Defining a new version of NETCONF (as opposed to extending NETCONF
   version 1.1) has several benefits:

   o  It allows for removal of the existing <get> RPC operation, that
      returns content from both the running configuration datastore
      combined with all config false leaves.

Bjorklund, et al.      Expires September 14, 2017              [Page 37]
Internet-Draft                                                March 2017

   o  It could allow the existing <get-config> operation to also be
      removed, replaced by the more generic <get-data> that is named
      appropriately to also apply to the operational datastore.

   o  It makes it easier for clients and servers to know what reasonable
      common baseline functionality to expect, rather than a collection
      of capabilities that may not be implemented in a consistent
      fashion.  In particular, clients will able to assume support for
      the <operational> datastore.

   o  It can gracefully coexist with NETCONF v1.1.  A server could
      implement both versions.  Existing YANG models exposing split
      config/state trees could be exposed via NETCONF v1.1, whereas
      combined config/state YANG models could be exposed via NETCONF v2,
      providing a viable server upgrade path.

D.4.3.2.  Proposed changes for NETCONF v2

   The differences between NETCONF v2 and NETCONF v1.1 can be summarized
   as:

   o  NETCONF v2 advertises a new base NETCONF capability
      "urn:ietf:params:netconf:base:2.0".  A server may advertise older
      NETCONF versions as well, to allow a client to choose which
      version to use.

   o  NETCONF v2 removes support for the existing <get> operation, that
      is replaced by the <get-data> on the operational datastore.

   o  NETCONF v2 can publish a separate version of YANG library from a
      NETCONF v1.1 implementation running on the same device, allowing
      different versions of NETCONF to support a different set of YANG
      modules.

D.4.3.3.  Possible Migration Paths

   A common approach in current data models is to have two separate
   trees "/foo" and "/foo-state", where the former contains config true
   nodes, and the latter config false nodes.  A data model that is
   designed for the revised architectural framework presented in this
   document will have a single tree "/foo" with a combination of config
   true and config false nodes.

   Two different migration strategies are considered:

Bjorklund, et al.      Expires September 14, 2017              [Page 38]
Internet-Draft                                                March 2017

D.4.3.3.1.  Migration Path using two instances of NETCONF

   If, for backwards compatability reasons, a server intends to support
   both split config/state trees and the combined config/state trees
   proposed in this architecture, then this can be achieved by having
   the device support both NETCONF v1 and NETCONF v2 at the same time:

   o  The NETCONF v1 implementation could support existing YANG module
      revisions defined with split config/state trees.

   o  The NETCONF v2 implementation could support different YANG
      modules, or YANG module revisions, with combined config/state
      trees.

   Clients can then decide on which type of models to use by expressing
   the appropriate version of the base NETCONF capability during
   capability exchange.

D.4.3.3.2.  Migration Path using a single instance of NETCONF

   The proposed strategy for updating existing published data models is
   to publish new revisions with the state trees' nodes copied under the
   config tree, and for the existing state trees to have all of their
   nodes marked as deprecated.  The expectation is that NETCONF servers
   would use a combination of these updated models alongside new models
   that only follow the new datastore architecture.

   o  NETCONF servers can support clients that are not aware of the
      revised datastore architecture, particularly if they continue to
      support the deprecated <get> operation:

      *  For updated YANG modules they would see additional information
         returned via the <get> operation.

      *  For new YANG modules, some of the state nodes may not be
         available, i.e. for any state nodes that exist under a config
         node that has not been configured (e.g., statistics under a
         system created interface).

   o  NETCONF servers can also support clients that are aware of the
      revised datastores architecture:

      *  For updated YANG modules they would see additional information
         returned under the legacy state trees.  This information can be
         excluded using appropriate subtree filters.

      *  New YANG modules, conforming to the datastores architecture,
         would work exactly as expected.

Bjorklund, et al.      Expires September 14, 2017              [Page 39]
Internet-Draft                                                March 2017

D.5.  Implications on RESTCONF

   This section describes the authors' thoughts on how RESTCONF
   [RFC8040] could be updated to support the datastore architecture
   described in this document.  It has been incorporated here as a
   temporary measure to facilitate easier review, but the expectation is
   that this work will be owned and standardized via the NETCONF working
   group.

D.5.1.  Introduction

   RESTCONF [RFC8040] defines a protocol based on HTTP for configuring
   data defined in YANG version 1 or 1.1, using a conceptual datastore
   that is compatible with a server that implements NETCONF 1.1
   compliant datastores.

   The combined conceptual datastore defined in RESTCONF is incompatible
   with the new datastore architecture defined in this document.  There
   are two possible ways that RESTCONF could be extended to support the
   new architecture: Either as new optional capabilities extending the
   existing RESTCONF RFC, or possibly as an new version of RESTCONF.

   Many of the required additions are common to both approaches, and are
   described below.  A following section then describes the potential
   benefits of defining a new RESTCONF version, and the additional
   changes that might entail.

D.5.2.  Overview of additions to RESTCONF

   o  A new path {+restconf}/datastore/<datastore-name>/data/ to provide
      a YANG data tree for each datastore that is exposed via RESTCONF.

   o  Implementations can choose which datastores they expose, but MUST
      at least expose both the <running> and <operational> datastores.
      They MAY expose the <intended> datastores as needed.

   o  The same HTTP Methods supported on {+restconf}/data/ are also
      supported on {+restconf}/datastore/<datastore-name>/data/ but
      suitably constrained depending on whether the datastore can be
      written to by the client, or is read-only.

   o  The same query parameters supported on {+restconf}/data/ are also
      support on {+restconf}/datastore/<datastore-name>/data/ except for
      the following query parameters:

   o  "metadata" - is a new optional query parameter that filters the
      returned data based on the metadata annotation.

Bjorklund, et al.      Expires September 14, 2017              [Page 40]
Internet-Draft                                                March 2017

   o  "with-metadata" - is a new optional query parameter that
      indicating that the metadata annotations should be included in the
      reply.

   o  "with-defaults" is supported on all configuration datastores, but
      is not supported on the operational state datastore path, because
      it has different default handling semantics.

   o  The handling of defaults (include the with-defaults query
      parameter) for the new configuration datastores is the same as the
      existing conceptual datastore, but does not apply for the
      operational state datastore that defines new semantics.

D.5.2.1.  HTTP Methods

   All configuration datastores support all HTTP Methods.

   The <operational> datastore only supports the following HTTP methods:
   OPTIONS, HEAD, GET, and POST to invoke an RFC operation.

D.5.2.2.  Query parameters

   [RFC7952] specifies how a YANG data tree can be annotated with
   generic metadata information, that is used by this document to
   annotate data nodes with origin information indicating the mechanism
   by which the operational value came into effect.

   RESTCONF could be extended with an optional generic mechanism to
   allow the filtering of nodes returned in a query based on metadata
   annotations associated with the data node.

   RESTCONF could also be extended with an optional generic mechanism to
   choose whether metadata annotations should be included in the
   response, potentially filtering to a subset of annotations.  E.g.,
   only include @origin metadata annotations, and not any others that
   may be in use.

   Both of the generic mechanisms could be controlled by a new
   capability.  A new capability is defined to indicate whether a device
   supports filtering on, or annotating responses with, the origin meta
   data.

D.5.2.3.  Operational State Datastore Defaults Handling

   The normal semantics for the <operational> datastore are that all
   values that match the default specified in the schema are included in
   response to requests on the operational state datastore.  This is
   equivalent to the "report-all" mode of the with-defaults handling.

Bjorklund, et al.      Expires September 14, 2017              [Page 41]
Internet-Draft                                                March 2017

   The "metadata" query parameter can be used to exclude nodes with a
   origin metadata matching "default", that would exclude (only config
   true?) nodes that match the default value specified in the schema.

   If the server cannot return a value for any reason (e.g., the server
   cannot determine the value, or the value that would be returned is
   outside the allowed leaf value range) then the server can choose to
   not return any value for a particular leaf, which MUST be interpreted
   by the client as the value of that leaf not being known, rather than
   implicitly having the default value.

D.5.3.  Overview of a possible new RESTCONF version

   This section describes a notional new RESTCONF version, by explaining
   the differences to RESTCONF version 1.  Where not explicitly
   specified, the behavior of a new RESTCONF version is the same as for
   RESTCONF version 1 [RFC8040].

D.5.3.1.  Potential benefits of defining a new RESTCONF version

   Defining a new version of RESTCONF (as opposed to extending RESTCONF
   version 1) has several potential benefits:

   o  It could expose datastores, and models designed for the revised
      datastore architecture, in a clean and consistent way.

   o  It would allow the parts of RESTCONF that do not work well with
      the revised datastore architecture to be omitted from the new
      RESTCONF version.

   o  It would make it easier for clients and servers to know what
      reasonable common baseline functionality to expect, rather than a
      collection of capabilities that may not be implemented in a
      consistent fashion.

   o  It could gracefully coexist with RESTCONF v1.  A server could
      implement both versions.  Existing YANG models exposing split
      config/state trees could be exposed via RESTCONF v1, whereas
      combined config/state YANG models could be exposed via a new
      RESTCONF version, providing a viable server upgrade path.

D.5.3.2.  Possible changes for a new RESTCONF version

   The differences between a notional new RESTCONF version and RESTCONF
   version 1 (RESTCONF v1) [RFC8040] can be summarized as:

   o  A new RESTCONF version would define a new root resource, and a
      separate link relation in the /.well-known/host-meta resource.

Bjorklund, et al.      Expires September 14, 2017              [Page 42]
Internet-Draft                                                March 2017

   o  A new RESTCONF version could remove support for the
      {+restconf}/data path supported in RESTCONF v1.

   o  A new RESTCONF version could publish a separate version of YANG
      library from a RESTCONF v1 implementation running on the same
      device, allowing different versions of RESTCONF to support a
      different set of YANG modules.

D.5.3.3.  Possible Migration Path using a new RESTCONF version

   A common approach in current data models is to have two separate
   trees "/foo" and "/foo-state", where the former contains config true
   nodes, and the latter config false nodes.  A data model that is
   designed for the revised architectural framework presented in this
   document will have a single tree "/foo" with a combination of config
   true and config false nodes.

   If for backwards compatability reasons, a server intends to support
   both split config/state trees, and the combined config/state trees
   proposed in this architecture, then this could be achieved by having
   the device support both RESTCONF v1 and the new RESTCONF version at
   the same time:

   o  The RESTCONF v1 implementation could support existing YANG module
      revisions defined with split config/state trees.

   o  The implementation of the new RESTCONF version could support
      different YANG modules, or YANG module revisions, with combined
      config/state trees.

   Clients can then decide on which type of models to use by choosing
   whether to use the RESTCONF v1 root resource or the root resource
   associated with the new RESTCONF version.

Appendix E.  Open Issues

   1.  NETCONF needs to be able to filter data based on the origin
       metadata.  Possibly this could be done as part of the <get-data>
       operation.

   2.  We need a means of inheriting @origin values, so whole
       hierarchies can avoid the noise of repeating parent values.
       Should "origin='system'" (or whatever we call it) be the default?

   3.  We need to discuss somewhere how remote procedure calls and
       notifications/actions tie into datastores.  RFC 7950 shows as an
       example a ping action tied to an interface.  Does this refer to
       an interface defined in a configuration datastore?  Or an

Bjorklund, et al.      Expires September 14, 2017              [Page 43]
Internet-Draft                                                March 2017

       interface defined in the operational state datastore?  Or the
       applied configuration datastore?  Similarly, RFC 7950 shows an
       example of a link-failure notification; this likely applies
       implicitly to the operational state datastore.  The netconf-
       config-change notification does explicitly identify a datastore.
       I think we generally need to have remote procedure calls and
       notifications be explicit about which datastores they apply to
       and perhaps change the default xpath context from running plus
       state to the operational state datastore.

Authors' Addresses

   Martin Bjorklund
   Tail-f Systems

   Email: mbj@tail-f.com

   Juergen Schoenwaelder
   Jacobs University

   Email: j.schoenwaelder@jacobs-university.de

   Phil Shafer
   Juniper Networks

   Email: phil@juniper.net

   Kent Watsen
   Juniper Networks

   Email: kwatsen@juniper.net

   Rob Wilton
   Cisco Systems

   Email: rwilton@cisco.com

Bjorklund, et al.      Expires September 14, 2017              [Page 44]