Partial Lock RPC for NETCONFEricssonbalazs.lengyel@ericsson.comTail-f Systemsmbj@tail-f.com
OPS
NETCONFThe NETCONF protocol defines the lock and unlock RPCs, used to
lock entire configuration datastores. In some situations, a way
to lock only parts of a configuration
datastore is required. This document defines a capability-based extension to
the NETCONF protocol for locking portions of a configuration datastore.
The protocol describes the lock and unlock operations
that operate on entire configuration datastores. Often,
multiple management sessions need to be able to modify the
configuration of a managed device in parallel. In these cases, locking
only parts of a configuration
datastore is needed. This document defines a capability based extension to the NETCONF protocol to
support partial locking of NETCONF datastores using a mechanism
based on the existing XPath filtering mechanisms.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
"OPTIONAL" in this document are to be interpreted as described in BCP 14, .
Additionally the following terms are defined:
Instance Identifier: an XPath expression identifying a
specific node in the conceptual XML datastore.
It contains an absolute path expression in abbreviated syntax,
where predicates are used only to specify values for nodes defined as keys to
distinguish multiple instances.Scope of the lock: initially the set of nodes returned by the XPath expressions in a successful
partial-lock operation. The set might be modified if some of the nodes are deleted.
Protected area: the set of nodes that are protected from modification by the lock.
This consist of nodes in the scope of the lock and nodes in subtrees under them.
The :partial-lock capability indicates that the device supports
the locking of its configuration with a more limited scope than a complete
configuration datastore. The scope to be locked is specified by using restricted or full XPath expressions.
Partial locking only affects configuration data.
The system MUST ensure that configuration resources
covered by the lock are not modified by other
NETCONF or non-NETCONF management operations such as
SNMP and the CLI.
The duration of the partial lock begins when the partial lock is
granted and lasts until (1) either the corresponding <partial-unlock> operation
succeeds or (2) the NETCONF session terminates.
A NETCONF session MAY have multiple parts of one or more datastores (running, candidate, startup) locked
using partial lock operations.
The <partial-lock> operation returns a
lock-id to identify each successfully acquired lock.
In the following we describe a few scenarios for partial locking. While scenarios
using the running datastore are seen as the most important, as an
example a scenario involving the candidate datastore is also presented.
Besides the three described here, there
are many other usage scenarios possible.
Multiple managers are handling the same NETCONF agent simultaneously.
The agent is handled via the writable running datastore.
Each manager has his or her own task, which might involve
the modification of overlapping sections of the datastore.
After collecting and analyzing input and preparing the
NETCONF operations off-line,
the manager locks the areas that are important for his task
using one single <partial-lock> operation.
The manager executes a number of <edit-config> operations to modify
the configuration, then releases the partial-lock.
The lock should be held for only a short time
(seconds rather then minutes).
The manager should collect all human input before locking anything.
As each manager locks only a part of the data model, usually multiple operators can
execute the <edit-config> operations simultaneously.
Multiple managers are handling the same NETCONF agent simultaneously.
The agent is handled via the writable running datastore.
The agent's data model contains a number of well defined separate
areas that can be configured without impacting other areas. An
example can be a server with multiple applications running on it, or
a number of a network elements with a common NETCONF agent for
management.
Each manager has his or her own task, which does not involve the modification
of overlapping sections of the datastore.
The manager locks his area with a <partial-lock> operation,
uses a number of <edit-config> commands to modify it,
later releases the lock.
As each manager has his functional area assigned to him,
and he locks only that area, multiple managers can edit
the configuration simultaneously.
Locks can be held for extended periods (minutes, hours),
as this will not hinder other managers.
This scenario assumes, that the global lock operation
from is not used.
Multiple managers are handling the same NETCONF agent simultaneously.
The agent is handled via the candidate datastore.
Each manager has his or her own task which might involve the modification
of overlapping sections of the datastore.
After collecting and analyzing input and preparing the
NETCONF operations off-line,
the manager locks the areas that are important for his task
using one single <partial-lock> operation in both the candidate and the running datastore.
He executes a number of <edit-config> operations to modify
the configuration, then releases the partial-lock.
The lock should be held for only a short time
(seconds rather then minutes).
Operators coordinate with each other. When all of them finish their tasks
one of them orders commit. If any of the operators are still working,
and holds a lock, the commit will fail, and will need to be repeated after all managers finish.
The device MUST support restricted XPath expressions in the select element,
as described in .
Optionally, if the :xpath capability is also supported
(as defined in chapter 8.9. XPath Capability),
the device MUST also support using any
XPath 1.0 expression in the select element.
urn:ietf:params:netconf:capability:partial-lock:1.0
The <partial-lock> operation allows the client to lock a portion of
one or more datastores. The portion to lock is specified with XPath
expressions in the "select" elements and the list of datastores in
the "target" elements in the <partial-lock> operation. Each XPath expression
MUST return a node set.
When a NETCONF session holds a lock on a node, no other session or
non-NETCONF mechanism of the system can change that node or any node
in the hierarchy of nodes beneath it.
Locking a node protects the node itself and
the complete subtree under the node from modification by others.
The set of locked nodes is called the scope of the lock, while all the
locked nodes and the nodes in the subtrees under them make up the protected area.
In some situations it is desirable that the same set of nodes are
locked in more than one datastore.
For example, if an interface is configured in the candidate datastore,
it is dangerous for it to be configured by someone else in a possibly conflicting
manner in the running datastore.
For this reason <partial-lock> allows the locking of the same sections of
the management data in multiple datastores.
The XPath expressions are evaluated only once at lock time.
Thereafter, the scope of the lock is maintained as a set of nodes,
i.e. the returned nodeset, and not by the XPath expression.
If the configuration data is later altered in a way that
would make the original XPath
expressions evaluate to a different set of nodes, this does not affect the scope of the partial lock.
Let's say the agent's data model includes a list of users.
If the XPath expression in the partial lock operation covers all users at locking,
the scope of the lock will be maintained as the list of "user" nodes at the time
when the lock was granted. If someone later creates a new user, this new
user will not be included in the locked-nodes list created previously, the new user will not be locked.
A <partial-lock> operation MUST be handled atomically by the NETCONF
server. The server either locks all requested parts of the datastore(s) or none.
If during the <partial-lock> operation one of the requested parts cannot be locked, the
server MUST unlock all parts that have already been locked during that operation.
If a node in the scope of the lock is deleted, it is removed from the
scope of the lock, so any other session or non-NETCONF mechanism can recreate it.
If all nodes in the scope of the lock are deleted, the
lock will still be present. However, its scope will become empty (since the lock will not cover any nodes).
A NETCONF server MUST be able to grant multiple simultaneous partial locks to a
single NETCONF session. If the protected area of the individual locks overlaps, nodes in the
common area MUST be protected until all of the locks are released.
A partial lock operation MUST fail if:
Any NETCONF session (including the current session) owns the
global lock on any target datastore.Any part of the area to be protected is already locked (or protected by partial locking) by another management
session, including other NETCONF sessions using
<partial-lock> or any other non-NETCONF management method.The NETCONF server implements access control, and the locking user does not have sufficient
access rights. The exact handling of access rights is outside the
scope of this document, but it is assumed that there is an access
control system that MAY deny or allow the partial lock operation.
The <partial-lock> operation is designed for simplicity, so when a partial lock is executed you get what
you asked for: a set of nodes that are locked for writing. As a consequence users must observe the following:
Locking does not affect read operations.If part of a datastore is locked, this has no effect on any unlocked parts of the datastore.
If this is a problem (e.g., changes depend on data values or nodes outside the protected
part of the datastore), these nodes should be included in the protected area of the lock.Configuration data can be edited both inside and outside the protected area of a
lock. It is the responsibility of the NETCONF client application to lock all relevant parts of a datastore that
are crucial for a specific management action.
Note: The <partial-lock> operation does not modify the global <lock> operation defined in the base
NETCONF Protocol . If part of a
datastore is already locked by <partial-lock>, then a global lock for that datastore MUST fail even
if the global lock is requested by the NETCONF session that owns the partial lock.
Parameters:
Name of one or more configuration datastores
of which a part shall be locked.
If multiple datastores are specified the same select parameter(s)
are evaluated for each of them.
One or more 'select' elements, each
containing an XPath expression. The XPath expression is
evaluated in a context where the context node is the root of the
server's conceptual data model,
and the set of namespace declarations are those in scope on the
select element.
Each select expression is evaluated for each targeted datastore.
The nodes returned from the select expressions are reported in
the rpc-reply message.
Note that if some of the requested nodes exist
only in some of the targeted datastores, the lock is granted on different
nodes in different datastores.
Each select expression MUST return a node set, and at least one of the
node sets for one of the specified datastores MUST be non-empty.
If the device supports the :xpath capability, any valid XPath 1.0 expression can
be used. If the device does not support the :xpath capability, the XPath
expression MUST be limited to an Instance Identifier expression.
An Instance Identifier is an absolute path expression in abbreviated syntax,
where predicates are used only to specify values for nodes defined as keys to
distinguish multiple instances.
Positive Response:
If the device was able to satisfy the request, an <rpc-reply> is
sent with a <lock-id> element (lock identifier) in the <rpc-reply> element.
A list of locked nodes per datastore is also returned in Instance Identifier format.
Negative Response:
If a lock is already held by another session on any node within the subtrees to be
locked, the <error-tag> element is 'lock-denied' and the
<error-info> element includes the <session-id> of the lock
owner. If the lock is held by a non-NETCONF session, a
<session-id> of 0 (zero) is included. If needed the returned session-id may
be used to <kill-session> the NETCONF session holding the lock.
The same error response is returned if
the requesting session already holds the (global) lock for the same datastore.
If any select expression is an invalid XPath expression, the
<error-tag> is 'invalid-value'.
If any select expression returns something other than a node set, the
<error-tag> is 'invalid-value', and the <error-app-tag> is 'not-a-node-set'.
If all the select expressions return an empty node set, the <error-tag> is
'operation-failed', and the <error-app-tag> is 'no-matches'.
If any of the target datastors does not exist, the <error-tag>
is 'invalid-value', the <error-app-tag> is 'invalid-lock-specification'
If the :xpath capability is not supported and the XPath
expression is not an Instance Identifier,
the <error-tag> is 'invalid-value', the <error-app-tag> is 'invalid-lock-specification'.
If access control denies the partial lock, the <error-tag>
is 'access-denied'.
As with most locking systems, it is possible that two management sessions trying to lock different
parts of the configuration could become dead-locked. To avoid this situation, clients should lock
everything they need in one operation. If locking fails, the client should back-off,
release any previously acquired locks, and retry the procedure after waiting some randomized time interval.
The operation unlocks the parts of the datastores that were previously
locked using <partial-lock> during the same session.
Parameters:
Identity of the lock to be unlocked.
This lock-id MUST have been received as a response to a lock request
by the manager during the current session, and MUST NOT have
been sent in a previous unlock request.
Positive Response:
If the device was able to satisfy the request, an <rpc-reply> is
sent that contains an <ok> element. A positive response MUST
be sent even if all of the locked parts of the datastore(s) have already been deleted.
Negative Response:
If the <lock-id> parameter does not identify a lock which is owned
by the session, an 'invalid-value' error is returned.
A successful partial lock will cause a subsequent operation to fail if that attempts to modify
nodes in the protected area of the lock
and is executed in a NETCONF session other than the session that has been granted the lock.
The <error-tag> 'in-use' and the <error-app-tag> 'locked' is returned.
All operations that modify the datastore are affected, including: <edit-config>, <copy-config>,
<delete-config>, <commit> and <discard-changes>.
If partial lock prevents <edit-config> modifying some data, but the
operation includes the continue-on-error option, modification of other
parts of the datastore, which are not protected by partial locking, might
still succeed.
If a datastore contains nodes locked by partial lock, this will cause the (global) <lock> operation to fail. The
<error-tag> element 'lock-denied' and an
<error-info> element including the <session-id> of the lock
owner will be returned. If the lock is held by a non-NETCONF session, a
<session-id> of 0 (zero) is returned.
All of these operations are affected only if they are targeting the same datastore.
Partial locking of the candidate datastore can only
be done if the :candidate capability is supported by the device.
Partial locking of the candidate datastore does not depend on whether the
datastore was modified or not.
If:
a partial lock is requested for the running datastore, andthe NETCONF server implements the :confirmed-commit capability, andthere was a recent confirmed <commit> operation where the confirming <commit> operation has not been received
then the lock MUST be denied, because if the confirmation does not arrive,
the running datastore MUST be rolled back to its state before the commit.
The NETCONF server might therefore need to modify the configuration.
In this case the <error-tag> 'in-use' and the
<error-app-tag> 'outstanding-confirmed-commit'
is returned.
Partial locking of the startup datastore can only
be done if the :startup capability is supported by the device.
The same considerations are relevant as for the base NETCONF Protocol .
It is assumed that the <partial-lock> and <partial-unlock> RPCs are only allowed for an authenticated user after passing some access control mechanism.
A lock (either a partial lock or a global lock) might prevent other users from configuring the system. The following mechanisms are in place to prevent the misuse of this possibility:
Only an authenticated and authorized user can request a partial lock.The partial lock is automatically released when a session is terminated regardless of how the session ends.The <kill-session> operation makes it possible to terminate other users's sessions.The NETCONF server may log partial lock requests in an audit trail.
A lock that is hung for some reason (e.g., a broken TCP connection that the server has not yet recognised) can be released
using another NETCONF session by explicitly killing the session owning that lock using the <kill-session> operation.
Partial locking is NOT an authorization mechanism; it SHOULD NOT be used to provide security or access control.
Partial locking SHOULD only be used as a mechanism for providing consistency
when multiple managers are trying to configure the node.
It is vital that users easily understand the exact scope of a lock.
This is why the scope is determined when granting a lock and is not modified thereafter.
This document registers two URIs for the NETCONF XML namespace in
the IETF XML registry . Note that the capability URN is compliant to
section 10.3.IndexCapability Identifier:partial-lockurn:ietf:params:netconf:capability:partial-lock:1.0URI: urn:ietf:params:xml:ns:netconf:partial-lock:1.0
Registrant Contact: The IESG.
XML: N/A, the requested URI is an XML namespace.
The following XML Schema defines the <partial-lock> and <partial-unlock> operations:
The following YANG module defines the <partial-lock> and <partial-unlock> operations.
The YANG language is defined in .
Partial lock cannot be used to lock non-existent nodes, which would effectively attempt to reserve them for future use.
To guarantee that a node cannot be created by some other session, the
parent node should be locked, the top level node of the new subtree created, and then locked
with another <partial-lock> operation. After this, the lock on the parent node should be removed.
In this chapter an example illustrating the above is given.
We want to create <user> Joe under <users>, and start editing it.
Editing might take a number of minutes.
We want to immediately lock Joe so no one will touch it before we are finished with the editing.
We also want to minimize locking other parts of the datastore as multiple managers might
be adding users near simultaneously.
First we check what users are already defined.
The NETCONF server sends the following reply.
We want to add the new user "Joe" and immediately lock him
using partial locking. The way to do this, is to first lock all <user> nodes by locking the <users> node.
The NETCONF server grants the partial lock. The scope
of the lock includes only the <users> node. The lock
protects the <users> node and all <user> nodes below it from
modification (by other sessions).
Next we create user Joe. Joe is protected by the lock
received above, as it is under the sub-tree rooted at the <users> node.
We receive a positive reply to the <edit-config> (not shown).
Next we request a lock, that locks only <user> Joe, and release the lock on the <users> node.
This will allow other managers to create additional new users.
The NETCONF server grants the partial lock. The scope
of this second lock includes only the <user> node with name Joe. The lock
protects all data below this particular <user> node.
The scope of the second lock is the <user> node Joe. It protects this <user>
node and any data below it (e.g. phone number).
At this point of time these nodes are protected both by the first and second lock.
Next we unlock the other <user>s and the <users> node, to allow other managers
to work on them. We still keep the second lock, so the <user> node Joe and the sub-tree below is still protected.
Changed XSD and YANG to allow additional proprietary datastores to be locked.Added usage exampleClarified error messagesClarified interaction with edit-config continue-on-errorImproved YANG: indentation, canonical order, contact infoAdded usage example in appendix CSynchronized YANG and XSDLanguage and editorial updatesall app-tags are with dashes without spacesAdded usage scenariosChanged encodingClarified definitions, separated scope of lock and protected areaMinor clarificationsAdded list of locked-nodes to the output of partial-lock.Added <target> wrapper around datastore names.Allowed atomic/one operation locking of datastore parts in multiple datastores.Improved English (hopefully)Removed the <data> element from rpc-reply following the text of rfc4741.Minor clarificationsSame descriptions in XSD and YANG.Made XSD normativeClarified that no specific access control is assumed.Clarified that non-existing nodes are NOT covered by the lock, even if they where existing and covered by the lock when it was originally granted.Some rewordingAdded app-tags for two of the error cases.Made YANG an informative referenceEnhanced security considerations.Added YANG module.Created from draft-lengyel-ngo-partial-lock-01.txtThanks to Andy Bierman, Sharon Chisholm, Phil Shafer , David Harrington, Mehmet Ersue,
Wes Hardaker, Juergen Schoenwaelder and
many other members of the NETCONF WG for providing important input to this document.NETCONF Configuration ProtocolThe Network Configuration Protocol (NETCONF) defined
in this document provides mechanisms to install, manipulate,
and delete the configuration of network devices. It uses an
Extensible Markup Language (XML)-based data encoding for the configuration
data as well as the protocol messages. The NETCONF protocol operations
are realized on top of a simple Remote Procedure Call (RPC)
layer. [STANDARDS TRACK]
Key words for use in RFCs to Indicate Requirement LevelsHarvard University1350 Mass. Ave.CambridgeMA 02138- +1 617 495 3864sob@harvard.edu
General
keyword
In many standards track documents several words are used to signify
the requirements in the specification. These words are often
capitalized. This document defines these words as they should be
interpreted in IETF documents. Authors who follow these guidelines
should incorporate this phrase near the beginning of their document:
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL
NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and
"OPTIONAL" in this document are to be interpreted as described in
RFC 2119.
Note that the force of these words is modified by the requirement
level of the document in which they are used.
The IETF XML RegistryThis document describes an IANA maintained registry for IETF standards which use Extensible
Markup Language (XML) related items such as Namespaces, Document Type Declarations (DTDs),
Schemas, and Resource Description Framework (RDF) Schemas.YANG - A data modeling language for NETCONFYANG is a data modeling language used to model configuration and state data manipulated
by the NETCONF protocol, NETCONF remote procedure calls, and NETCONF notifications.