One document matched: draft-ietf-i2rs-ephemeral-state-00.xml

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE rfc SYSTEM "rfc2629.dtd" [
<!ENTITY RFC2119 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.2119.xml">
<!ENTITY RFC6020 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.6020.xml">
<!ENTITY RFC6241 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.6241.xml">
<!ENTITY RFC6536 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.6536.xml">

<!ENTITY I-D.ietf-i2rs-architecture SYSTEM "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.ietf-i2rs-architecture.xml">
<!ENTITY I-D.ietf-i2rs-rib-info-model SYSTEM "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.ietf-i2rs-rib-info-model.xml">
<!ENTITY I-D.ietf-netmod-acl-model SYSTEM "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.ietf-netmod-acl-model.xml">
<!ENTITY I-D.ietf-i2rs-traceability SYSTEM "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.ietf-i2rs-traceability.xml">
<!ENTITY I-D.ietf-i2rs-usecase-reqs-summary SYSTEM "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.hares-i2rs-usecase-reqs-summary.xml">
<!ENTITY I-D.ietf-netmod-yang-metadata SYSTEM "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.ietf-netmod-yang-metadata.xml">
<!ENTITY I-D.bierman-netconf-restconf SYSTEM "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.bierman-netconf-restconf.xml">
]>
<?xml-stylesheet type='text/xsl' href='rfc2629.xslt' ?>
<?rfc toc="yes" ?>
<?rfc symrefs="yes" ?>
<?rfc sortrefs="yes"?>
<?rfc compact="yes" ?>
<?rfc subcompact="no" ?>
<?rfc iprnotified="no" ?>
<?rfc strict="no" ?>
<rfc category="std" docName="draft-ietf-i2rs-ephemeral-state-00"  ipr="trust200902">
  <front>
    <title abbrev="I2RS Ephemeral State Requirements">I2RS Ephemeral State Requirements </title>

	<author fullname="Jeff Haas" initials="J." surname="Haas">
      <organization>Juniper</organization>
      <address>
        <postal>
          <street></street>
          <city> </city>
          <country></country>
        </postal>
        <email>jhaas@juniper.net</email>
      </address>
    </author>
	<author fullname="Susan Hares" initials="S." surname="Hares">
	<organization> Huawei </organization>
	<address>
	  <postal>
	   <street></street>
	    <city>Saline</city>
		<country>US</country>
	  </postal>
	 <email>shares@ndzh.com </email>
	</address>
	</author>
	 
    <date year="2015" />
    <area>Routing Area</area>
    <workgroup>I2RS working group</workgroup>
    <keyword>RFC</keyword>
    <keyword>Request for Comments</keyword>
    <keyword>I-D</keyword>
    <keyword>Internet-Draft</keyword>
    <keyword>I2RS</keyword>
    <abstract>
      <t>This document covers requests to the netmod and netconf Working
   Groups for functionality to support the ephemeral state requirements
   to implement the I2RS architecture. </t>
	 </abstract>
  </front>
  <middle>
    <section anchor="intro" title="Introduction">
   <t>The Interface to the Routing System (I2RS) Working Group is chartered
   with providing architecture and mechanisms to inject into and
   retrieve information from the routing system.  The I2RS Architecture
   document <xref target="I-D.ietf-i2rs-architecture"></xref> abstractly documents a number
   of requirements for implementing the I2RS requirements.</t>
   <t>
   The I2RS Working Group has chosen to use the YANG data modeling
   language <xref target="RFC6020"></xref> as the basis to implement its mechanisms.
   </t>
   <t> 
   Additionally, the I2RS Working group has chosen to use the NETCONF
   <xref target="RFC6241"></xref> and its similar but lighter-weight relative RESTCONF
   <xref target="I-D.bierman-netconf-restconf"></xref> as the protocols for carrying I2RS.
   </t>
   <t>
   While YANG, NETCONF and RESTCONF are a good starting basis for I2RS,
   there are some things needed from each of them in order for I2RS to
   be implemented.
	 </t>
	</section>
	<section title="Review of Requirements from I2RS architecture document"> 
 <t> The following are ten requirements that <xref target="I-D.ietf-i2rs-architecture"></xref>
  contains which are important high level requirements:  
<list style="numbers">
<t>The I2RS protocol SHOULD support highly reliable notifications (but
not perfectly reliable notifications) from an I2RS agent to an  I2RS client. </t>
 <t>The I2RS protocol SHOULD support a high bandwidth, asynchronous
interface, with real-time guarantees on getting data from an I2RS  agent by
an I2RS client. </t>
<t> The I2RS protocol will operate on data models which may be protocol
independent or protocol dependent.  </t>
<t> I2RS Agent needs to record the client identity when a node is created or modified. 
The I2RS Agent needs to be able to read the client identity of a node and 
use the client identity's associated priority to resolve conflicts.   
The secondary identity is useful for traceability and may also be recorded.</t>
<t>Client identity will have only one priority for the client identity. A
collision on writes is considered an error, but priority is utilized to
compare requests from two different clients in order to modify an existing
node entry.  Only an entry from a client which is higher priority can modify
an existing entry (First entry wins). Priority only has meaning at the time
of use.</t>
<t>The Agent identity and the Client identity should be passed outside of
the I2RS protocol in a authentication and authorization  protocol (AAA).
Client priority may be passed in the AAA protocol.  The values of identities
are originally set by operators, and not  standardized. </t>
<t>An I2RS Client and I2RS Agent mutually authenticate each other based on
pre-established authenticated identities. </t>
<t>Secondary identity data is read-only meta-data that is recorded by the
I2RS agent associated with a data model's node is written, updated or
deleted. Just like the primary identity, the secondary identity is only recorded when
the data node is written or updated or deleted</t>
<t>I2RS agent can have a lower priority I2RS client attempting to modify
a higher priority client's entry in a data model.  The filtering out of
lower priority clients attempting to write or modify a higher priority
client's entry in a data model SHOULD be effectively handled and not put an
undue strain on the I2RS agent. Note:  Jeff's suggests that priority is kept at the NACM at the client level
(rather than the path level or the group level) will allow these lower
priority clients to be filtered out using an extended NACM approach. This is
only a suggestion of a method to provide the requirement 9.  
</t>
<t>The I2RS protocol MUST support the use of a secure transport. However, 
certain functions such as notifications MAY use a non-secure transport.  
Each model or service (notification, logging) must define within the model or 
service the valid uses of a non-secure transport.</t>
</list>
</t>
</section>
 <section title="Ephemeral State Requirements"> 
<section title="Persistence">
   <t> I2RS requires ephemeral state; i.e. state that does not persist
    across reboots.  If state must be restored, it should be done solely
    by replay actions from the I2RS client via the I2RS agent. </t>
   <t>While at first glance this may seem equivalent to the writable-
   running datastore in NETCONF, running-config can be copied to a
   persistant data store, like startup config.  I2RS ephemeral state
   MUST NOT be persisted. </t>
    </section>
   	<section title="Constraints">
   <t>Ephemeral state MAY refer to non-ephemeral state for purposes of
   implementing constraints.  The designer of ephemeral state modules
   are advised that such constraints may impact the speed of processing
   ephemeral state commits and should avoid them when speed is
   essential.</t>
   <t>Non-ephemeral state MUST NOT refer to ephemeral state for constraint
   purposes; it SHALL be considered a validation error if it does. </t>
  	</section>
    <section title="Hierarchy">
   <t> Similar to configuration state (config true, see 
    <xref target="RFC6020"></xref>, section 7.19.1), 
   ephemeral state is not permitted to be configured underneath
   nodes that are "config false" (state data).
   </t>
   <t>Configuration of ephemeral state underneath "config true" is
   permitted.  This permits augmentation of configuration state with
   ephemeral nodes.
   </t>
   <t>
   Configuration of "config true" state underneath ephemeral state MUST
   NOT be done.
   </t>
   <t>
   State data, "config false", is permitted underneath ephemeral state.  
   This state data is part of the ephemeral module and should become    
    inaccessible if the ephemeral module reboots.
   </t>
	</section>
	</section>
	<section title="changes to YANG">
	<t>
  The YANG "config" keyword (<xref target="RFC6020"></xref>, section 7.19.1) is extended to
   support the keyword "ephemeral" in addition to "true" and "false".
   "config ephemeral" declares the nodes underneath to be ephemeral
   configuration.
	</t>
	</section>
	<section title="Changes to NETCONF">
	<t>
   A capability is registered declaring that the server supports
   ephemeral configuration.  E.g.:
    <figure>
	<artwork>
  :ephemeral-config
       urn:ietf:params:netconf:capability:ephemeral-config:1.0
	</artwork>
	</figure>
	</t>
	<t>
   <get-config> will normally return "config ephemeral" nodes as it is a
   form of configuration.  It is further extended to add a new
   parameter, "filter-ephemeral".  This parameter accepts the following
   arguments:
   <list style="symbols"> 
   <t>none (default): No filtering of persistent or ephemeral state is
      done.</t>
	<t>ephemeral-only: Only nodes representing ephemeral state are
      returned.</t>
	<t> exclude-ephemeral: Only persistent configuration is returned.</t>
   </list>
   </t>
   <t>   
   <get> is similarly extended to support "filter-ephemeral".
   </t>
   <t>
   When a <copy-config> is done, regardless of datastore, nodes that are
   "config ephemeral" are excluded from the target output.
   </t>
 </section>
 <section title="Requirements regarding Identity, Secondary-Identity and Priority">
 <section title="Identity Requirements">
 <t>I2RS requires clients to have an identity.  This identity will be
   used by the Agent authentication mechanism over the appropriate
   protocol. 
   </t>
   <t>I2RS also permits clients to have a secondary identity which may be
   used for troubleshooting.  This secondary identity is an opaque
   value.  <xref target="I-D.ietf-i2rs-traceability"></xref> provides an example of how the
   secondary identity can be used for traceability.
   </t>
   <t>
   The secondary identity is carried in the configuration operation
   using a new parameter to <edit-config>.  E.g.:
</t>
<t>
<figure>
<artwork>
  <rpc message-id="101" xmlns="urn:ietf:params:xml:ns:netconf:base:1.0">
      <edit-config>
      <i2rs:irs-secondary-identity>user1</i2rs>
      <target>
          <running/>
      </target>
      <config>
          <top xmlns="http://example.com/schema/1.2/config">
          <interface>
              <name>Ethernet0/0</name>
              <mtu>1500</mtu>
          </interface>
          </top>
      </config>
      </edit-config>
  </rpc>
</artwork>
</figure>
</t>
<t>
   "config ephemeral" nodes that are created or altered as part of the
   config operation will carry the secondary-identity as read-only
   metadata.
</t>
 </section>
 <section title="Priority Requirements">
  <t> To support Multi-Headed Control, I2RS requires that there be a
   decidable means of arbitrating the correct state of data when
   multiple clients attempt to manipulate the same piece of data.  This
   is done via a priority mechanism with the highest priority winning.
   This priority is per-client.
</t>
<t>
   This further implies that priority is an attribute that is stored in
   the NETCONF Access Control Model <xref target="RFC6536"></xref> as part of the group.
   E.g.:
<figure>
<artwork>
+--rw nacm
   +--rw enable-nacm?            boolean
   +--rw read-default?           action-type
   +--rw write-default?          action-type
   +--rw exec-default?           action-type
   +--rw enable-external-groups? boolean
   +--ro denied-operations       yang:zero-based-counter32
   +--ro denied-data-writes      yang:zero-based-counter32
   +--ro denied-notifications    yang:zero-based-counter32
   +--rw groups
   |  +--rw group [name]
   |     +--rw name         group-name-type
   |     +--rw user-name*   user-name-type
   |     +--rw i2rs:i2rs-priority i2rs-priority-type
</artwork>
</figure>
</t>
<t>
   Ephemeral configuration state nodes that are created or altered by
   users that match a rule carrying i2rs-priority will have those nodes
   annotated with metadata.  Additionally, during commit processing, if
   nodes are found where i2rs-priority is already present, and the
   priority is better than the transaction's user's priority for that
   node, the commit SHALL fail.  An appropriate error should be returned
   to the user stating the nodes where the user had insufficient
   priority to override the state.
</t>
 </section>
 <section title="Representing I2RS Attributes in ephemeral configuration state">
  <t>I2RS attributes may be modeled as meta-data,
   <xref target="I-D.ietf-netmod-yang-metadata"></xref>.  This meta-data MUST be read-only;
   operations attempting to alter it MUST be silently ignored.  An I2RS
   module will be defined to document this meta data.  An example of its
   use:
   <figure>
   <artwork>
    <foo xmlns:i2rs="https://ietf.example.com/i2rs"
        i2rs:i2rs-secondary-identity="user1" i2rs:i2rs-priority="47">
       ...
   </foo>
   </artwork>
   </figure>
   </t>
 </section>
 <section title="Semantics around storing and managing of priority and client ID.">
 <t>The semantics and desired behavior around the storing and managing of 
 priority and client ID have the following properties:
 <list style="numbers">
 <t>First - the  priority mechanism is intended to handle "error cases of 
 colliding writes" in a predictable way that results in a consistent 
 mechanism.  It is true that the same mechanism could be used if they 
 were not considered "errors", but it is important to  
 minimize the need and impact of the priority mechanism </t>
 <t>Second, if there is a priority conflict where both clients (Client_A 
and Client_B) share the same priority, the client that wrote first wins.  
This is to avoid network oscillation if two clients are "fighting" 
over writing the same state.  When there are multiple clients and the 
time arrival of the messages may not be predictable (network transit 
differences, which socket is read, software differences), basing state 
on last arrival time doesn't give consistent and predictable behavior.
That gives behavior ont the following time-line
<list>
<t> Time_1:  Client_A writes X=N with priority 10</t>
<t> Time_2:  Client_B attempts to write X=K with priority 10 and is rejected</t>
<t> Time_3:  Client_A writes X=P with priority 10 and succeeds </t>
</list> 
 For the I2RS Agent to properly handle these actions, it is necessary 
to know that X is owned by Client_A.  Priority alone is not sufficient 
because the basis for rejecting Client_B's write but accepting 
 Client_A's write is that Client_A is the owner. 
 Thus it is necessary to store the Client Identity with 
the nodes that it owns. This could be in an I2RS-specific overlay that is only used by the 
 I2RS agent and only contains the nodes that have been written by I2RS.
 </t>
 <t> Third, a question has come up regarding what the behavior of priority 
is if a client's priority changes and whether priority needs to be 
stored with each node when that node is written.
In my "keep-it-simple" perspective, priority is associated with a 
Client and is only used on a conflict.  This would mean that priority 
is not stored with a node when that node is written.  Instead, the 
Client Identity is stored with the node and the Client's priority is 
looked up in a client table that the I2RS Agent can access.  That 
client table could be populated via configuration, via a AAA protocol, via NACM, etc. 
The sematic implications are as follows:
<list>
<t>Time_1:  Client_A writes X=N with priority 10 </t>
<t>Time_2:  Client_A's priority is changed (UNUSUAL) to priority 6 </t>
<t>Time_3:  Client_B writes X=K with priority 8  (succeeds since 8 > 6) 
</t>
<t>Time_4:  Client_A attempts to write X=N with priority 6  (fails b/c 8
> > 6)
</t>
<t>Time_5:  Client_B's priority is changed (UNUSUAL) to priority 7 </t>
<t>Time_6   Client_B writes X=P with priority 7 and succeeds (same
> owner, no priority check) </t>
</list>
The alternate approach would have store the priority with which a node 
was written.  That is more like a priority lock that could only be 
changed by a client with higher priority or by the same client, regardless of priority.
This approach would require storing a priority per node and the 
semantic implications would be as follows:
<list>
<t>Time_1:Client_A writes X=N with priority 10
</t>
<t>Time_2:Client_A's priority is changed (UNUSUAL) to priority 6
</t>
<t>Time_3: Client_B attempts to write X=K with priority 8 and fails
  (10 > 8)
</t>
<t>Time_4: Client_A writes X=N with priority 6 and succeeds (same owner,
 no priority check)
</t>
<t>Time_5:  Client_B's priority is changed (UNUSUAL) to priority 7
</t>
<t>Time_6   Client_B writes X=P with priority 7 and succeeds (7 > 6)
</t>
</list>
The behavior for these two models is different at Time_3 and Time_4.
</t>
 </list>
 </t>
 <t> The initial preference was that the priority is not stored
 with the node, but if it necessary to store it with the node
 additional discussion may be needed with the I2RS WG.</t>
 </section>
 </section>
 <section title="Subscriptions to Changed State Requirements">
 <t>
    I2RS clients require the ability to monitor changes to ephemeral
   state.  While subscriptions are well defined for receiving
   notifications, the need to create a notification set for all
   ephemeral configuration state may be overly burdensome to the user.
   </t>
   <t>
      There is thus a need for a general subscription mechanism that can
   provide notification of changed state, with sufficient information to
   permit the client to retrieve the impacted nodes.  This should be
   doable without requiring the notifications to be created as part of
   every single I2RS module.
</t>
 </section>
 <section title="Transactions">
 <t>
 Section 7.9 of the <xref target="I-D.ietf-i2rs-architecture"></xref>
 states the I2RS architecture does not include
 multi-message atomicity and rollback mechanisms, but suggests 
an I2RS client may inidicate one of the following error handling techniques
for a given message sent to the I2RS client:
 <list style="numbers">
 <t>Perform all or none: All operations succeed or none of them will be applied. 
   This useful when there are mutual dependencies.</t>
 <t>Perform until error: Operations are applied in order, and when error occurs
   the processing stops. This is useful when dependencies exist between
   multiple-message operations, and order is important.</t>
  <t>Perform all storing errors: Perform all actions storing error indications
    for errors.  This method can be used when there are no dependencies  
    between operations, and the client wants to sort it out.</t>
 </list> 
 </t>
<t>None of these three cases insert known errors into the I2RS ephemeral datastore. 
</t>
<t>RESTCONF does an atomic action within a http session, and NETCONF has atomic
actions within a commit. These features may be used to perform these features.</t>
<t> I2RS processing is dependent on the I2RS model. The I2RS model must consider 
the dependencies within multiple operations work within a model.  
</t>
 </section>
 <section title=" Previously Considered Ideas">
 <section title="A Separate Ephemeral Datastore">
 <t>
 The primary advantage of a fully separate datastore is that the
   semantics of its contents are always clearly ephemeral.  It also
   provides strong segregation of I2RS configuration and operational
   state from the rest of the system within the network element.
 </t>
 <t>
    The most obvious disadvantage of such a fully separate datastore is
   that interaction with the network element's operational or
   configuration state becomes significantly more difficult.  As an
   example, a BGP I2RS use case would be the dynamic instantiation of a
   BGP peer.  While it is readily possible to re-use any defined
   groupings from an IETF-standardized BGP module in such an I2RS
   ephemeral datastore's modules, one cannot currently reference state
   from one datastore to anothe
 </t>
 <t>
    For example, XPath queries are done in the context document of the
   datastore in question and thus it is impossible for an I2RS model to
   fulfil a "must" or "when" requirement in the BGP module in the
   standard data stores.  To implement such a mechanism would require
   appropriate semantics for XPath.
 </t>
 </section>
 <section title=" Panes of Glass/Overlay">
 <t>   I2RS ephemeral configuration state is generally expected to be
   disjoint from persistent configuration.  In some cases, extending
   persistent configuration with ephemeral attributes is expected to be
   useful.  A case that is considered potentially useful but problematic
   was explored was the ability to "overlay" persistent configuration
   with ephemeral configuration.
 </t>
 <t>   In this overlay scenario, persistent configuration that was not
   shadowed by ephemeral configuration could be "read through".
 </t>
 <t>   There were two perceived disadvantages to this mechanism:
 <list>
 <t>The general complexity with managing the overlay mechanism
       itself.</t>
 <t>Consistency issues with validation should the ephemeral state be
       lost, perhaps on reboot.  In such a case, the previously shadowed
       persistent state may no longer validate.
</t>
 </list>
 </t>
 </section>
 </section>
 <section title="Actions Required to Implement this Draft">
 <t>
 <list style="symbols">
 <t>Draft for adding "config ephemeral" to YANG.</t>
 <t> Draft defining NETCONF changes including capability, RPC operation
      changes and support of secondary identity, RPC changes to support
      priority.</t>
 <t> I2RS draft to define meta-data for priority and secondary-
      identity.</t>
 </list>
 </t>
 </section>
    <section anchor="IANA" title="IANA Considerations">
      <t>TBD.</t>
    </section>

    <section title="Security Considerations">
      <t>TBD.</t>
    </section>
	    <section anchor="Acknowledgements" title="Acknowledgements">
	<t>
	    This document is an attempt to distill lengthy conversations on
	    the I2RS mailing list for an architecture that was for a long
	    period of time a moving target.  Some individuals in particular
	    warrant specific mention for their extensive help in providing
	    the basis for this document:
	</t>
	<t>
	    <list style="symbols">
		<t>Alia Atlas</t>
		<t>Andy Bierman</t>
		<t>Martin Bjorklund</t>
		<t>Dean Bogdanavich</t>
		<t>Rex Fernando</t>
		<t>Joel Halpern</t>
		<t>Thomas Nadeau</t>
		<t>Juergen Schoenwaelder</t>
		<t>Kent Watsen</t>
	    </list>
	</t>
    </section>
  </middle>
  <back>
    <references title="Normative References:">
	 &I-D.ietf-i2rs-architecture;
     &I-D.ietf-i2rs-rib-info-model;
	 &I-D.ietf-i2rs-traceability;
	 &I-D.ietf-netmod-yang-metadata;
	</references>
    <references title="Informative References">
      &RFC2119;
	  &RFC6020;
	  &RFC6241;
	  &RFC6536;
	  &I-D.bierman-netconf-restconf;
    </references>
  </back>
</rfc>