Skip to content

Latest commit

 

History

History
457 lines (363 loc) · 17.3 KB

startup.md

File metadata and controls

457 lines (363 loc) · 17.3 KB
Raw

Startup and upgrading

Background

This document describes the configuration startup mechanism of the Clixon backend. It describes the mechanism of Clixon version 3.10 which supports the following features:

  • Loading of a "startup" XML or JSON configuration
  • Loading of "extra" XML.
  • Detection of in-compatible XML and Yang models in the startup configuration.
  • An upgrade callback when in-compatible XML is encountered
  • A "failsafe" mode allowing a user to repair the startup on errors or failed validation.

Notes on this document:

  • "database" and "datastore" are used interchangeably for the same XML or JSON file storing a configuration.
  • For some scenarios, such a the "running" startup mode, a "temporary" datastore is used (called tmp_db). This file may have to be accessed out-of-band in failure scenarios.

Modes

Clixon by default supports the Netconf startup feature. But a clixon system nevertheless can be started in four different ways, starting from the startup datastore is mainly only an option on reboot of a system.

When the Clixon backend starts, it can start in one of four modes:

  • startup: The configuration is loaded from a persistent startup datastore. The XML is loaded, parsed, validated and committed into the running database.
  • running: Similar to startup, but instead the running datastore is used as a persistent database. The system copies the original running-db to a temporary store(tmp_db), and commits that temporary datastore into the (new) running datastore.
  • none: No data stores are touched - the system starts and loads existing running datastore without validation or commits.
  • init: Similar to none, but the running database is cleared before loading

Startup targets usecases where running db may be in memory and a separate persistent storage (such as flash) is available. Running is for usecases when the running db is located in persistent. The none and init modes are mostly for debugging, or restart at crashes or updates.

Startup configuration

When the backend daemon is started in startup mode, the system loads the startup database.

The running mode is similar, the only difference is that the running database is copied into a temporary database which then acts as the startup store.

When loading the startup/tmp configuration, the following actions are performed by the system:

  • It is checked for parse errors,
  • the yang model-state is detected (if present)
  • the XML is validated against the Yang models loaded in the backend (NB: may be different from the model-state).

If yang-models do not match, an upgrade callback is made.

If any errors are detected, the backend enters a failsafe mode.

Module-state

Clixon has the ability to store Yang module-state information according to RFC7895 in the datastores. Including yang module-state in the datastores is enabled by the following entry in the Clixon configuration:

   <CLICON_XMLDB_MODSTATE>true</CLICON_XMLDB_MODSTATE>

If the datastore does not contain module-state info, no detection of incompatible XML is made, and the upgrade feature described in this section will not occur.

A backend does not perform detection of mismatching XML/Yang if:

  1. The datastore was saved in a pre-3.10 system
  2. CLICON_XMLDB_MODSTATE was not enabled when saving the file
  3. The backend configuration does not have CLICON_XMLDB_MODSTATE enabled.

Note that the module-state detection is independent of the other steps of the startup operation: syntax errors, validation checks, failsafe mode, etc, are still made, even though module-state detection does not occur.

Note also that a 3.10 Clixon system with CLICON_XMLDB_MODSTATE disabled will silently ignore the module state.

Example of a (simplified) datastore with Yang module-state:

<config>
   <a1 xmlns="urn:example:a">some text</a1>
   <modules-state xmlns="urn:ietf:params:xml:ns:yang:ietf-yang-library">
      <module-set-id>42</module-set-id>
      <module>
         <name>A</name>
         <revision>2019-01-01</revision>
         <namespace>urn:example:a</namespace>
      </module>
   </modules-state>
</config>

Upgrade callback

This section describes how a user can write upgrade callbacks for data modeled by outdated Yang models. The scenario is a Clixon system with a set of current yang models that loads a datastore with old or even obsolete data.

Note that this feature is only available if module-state in the datastore is enabled.

If the module-state of the startup configuration does not match the module-state of the backend daemon, a set of upgrade callbacks are made. This allows a user to program upgrade funtions in the backend plugins to automatically upgrade the XML to the current version.

Clixon also has an experimental automatic upgrading method based on Yang changelogs covered in a separate chapter.

A user registers upgrade callbacks based on module and revision ranges. A user can register many callbacks, or choose wildcards. When an upgrade occurs, the callbacks will be called if they match the module and revision ranges registered.

Different strategies can be used for upgrade functions. One coarse-grained method is to register a single callback to handle all modules and all revisions.

A fine-grained method is to register a separate stepwise upgrade callback per module and revision range that will be called in a series.

Registering a callback

A user registers upgrade callbacks in the backend clixon_plugin_init() function. The signature of upgrade callback is as follows:

  upgrade_callback_register(h, cb, namespace, from, revision, arg);

where:

  • h is the Clicon handle,
  • cb is the name of the callback function,
  • namespace defines a Yang module. NULL denotes all modules. Note that module name is not used (XML uses namespace, whereas JSON uses name, XML is more common).
  • from is a revision date indicated an optional start date of the upgrade. This allows for defining a partial upgrade. It can also be 0 to denote any version.
  • revision is the revision date "to" where the upgrade is made. It is either the same revision as the Clixon system module, or an older version. In the latter case, you can provide another upgrade callback to the most recent revision.
  • arg is a user defined argument which can be passed to the callback.

One example of registering a "catch-all" upgrade:

   upgrade_callback_register(h, xml_changelog_upgrade, NULL, 0, 0, NULL);

Another example are fine-grained stepwise upgrades of a single module upgrade example:

   upgrade_callback_register(h, upgrade_2016, "urn:example:interfaces",
                             20140508, 20160101, NULL);
   upgrade_callback_register(h, upgrade_2018, "urn:example:interfaces",
                             20160101, 20180220, NULL);

   20140508       20160101       20180220
------+--------------+--------------+-------->
        upgrade_2016   upgrade_2018

In the latter case, the first callback upgrades from revision 2014-05-08 to 2016-01-01; while the second makes upgrades from 2016-01-01 to 2018-02-20. These are run in series.

Upgrade callback

When Clixon loads a startup datastore with outdated modules, the matching upgrade callbacks will be called.

Note the following:

  • Upgrade callbacks will not be called for data that is up-to-date with the current system
  • Upgrade callbacks will not be called if there is no module-state in the datastore, or if module-state support is disabled.
  • Upgrade callbacks will be called if the datastore contains a version of a module that is older than the module loaded in Clixon.
  • Upgrade callbacks will also be called if the datastore contains a version of a module that is not present in Clixon - an obsolete module.

Re-using the previous stepwise example, if a datastore is loaded based on revision 20140508 by a system supporting revision 2018-02-20, the following two callbacks are made:

  upgrade_2016(h, <xml>, "urn:example:interfaces", 20140508, 20180220, NULL, cbret);
  upgrade_2018(h, <xml>, "urn:example:interfaces", 20140508, 20180220, NULL, cbret);

Note that the example shown is a template for an upgrade function. It gets the nodes of an yang module given by namespace and the (outdated) from revision, and iterates through them.

If no action is made by the upgrade calback, and thus the XML is not upgraded, the next step is XML/Yang validation.

An out-dated XML may still pass validation and the system will go up in normal state.

However, if the validation fails, the backend will try to enter the failsafe mode so that the user may perform manual upgarding of the configuration.

Example upgrade

The example and shows the code for upgrading of an interface module. The example is inspired by the ietf-interfaces module that made a subset of the upgrades shown in the examples.

The code is split in two steps. The upgrade_2016 callback does the following transforms:

  • Move /if:interfaces-state/if:interface/if:admin-status to /if:interfaces/if:interface/
  • Move /if:interfaces-state/if:interface/if:statistics to if:interfaces/if:interface/
  • Rename /interfaces/interface/description to /interfaces/interface/descr

The upgrade_2018 callback does the following transforms:

  • Delete /if:interfaces-state
  • Wrap /interfaces/interface/descr to /interfaces/interface/docs/descr
  • Change type /interfaces/interface/statistics/in-octets to decimal64 and divide all values with 1000

Please consult the upgrade_2016 and upgrade_2018 functions in the example and test for more details.

Extra XML

If the Yang validation succeeds and the startup configuration has been committed to the running database, a user may add "extra" XML.

There are two ways to add extra XML to running database after start. Note that this XML is "merged" into running, not "committed".

The first way is via a file. Assume you want to add this xml:

<config>
   <x xmlns="urn:example:clixon">extra</x>
</config>

You add this via the -c option:

clixon_backend ... -c extra.xml

The second way is by programming the plugin_reset() in the backend plugin. The example code contains an example on how to do this (see plugin_reset() in example_backend.c).

The extra-xml feature is not available if startup mode is none. It will also not occur in failsafe mode.

Startup status

When the startup process is completed, a startup status is set and is accessible via clixon_startup_status_get(h) with the following values:

  STARTUP_ERR        XML/JSON syntax error
  STARTUP_INVALID,   XML / Yang validation failure
  STARTUP_OK         OK

Failsafe mode

If the startup fails, the backend looks for a failsafe configuration in CLICON_XMLDB_DIR/failsafe_db. If such a config is not found, the backend terminates.

If the failsafe is found, the failsafe config is loaded and committed into the running db.

If the startup mode was startup, the startup database will contain syntax errors or invalidated XML.

If the startup mode was running, the the tmp database will contain syntax errors or invalidated XML.

Repair

If the system is in failsafe mode (or fails to start), a user can repair a broken configuration and then restart the backend. This can be done out-of-band by editing the startup db and then restarting clixon.

In some circumstances, it is also possible to repair the startup configuration on-line without restarting the backend. This section shows how to repair a startup datastore on-line.

However, on-line repair cannot be made in the following circumstances:

  • The broken configuration contains syntactic errors - the system cannot parse the XML.
  • The startup mode is running. In this case, the broken config is in the tmp datastore that is not a recognized Netconf datastore, and has to be accessed out-of-band.
  • Netconf must be used. Restconf cannot separately access the different datastores.

First, copy the (broken) startup config to candidate. This is necessary since you cannot make edit-config calls to the startup db:

  <rpc>
    <copy-config>
      <source><startup/></source>
      <target><candidate/></target>
    </copy-config>
  </rpc>

You can now edit the XML in candidate. However, there are some restrictions on the edit commands. For example, you cannot access invalid XML (eg that does not have a corresponding module) via the edit-config operation. For example, assume x is obsolete syntax, then this is not accepted:

  <rpc>
    <edit-config>
      <target><candidate/></target>
      <config>
        <x xmlns="example" operation='delete'/>
      </config>
    </edit-config>
  </rpc>

Instead, assuming y is a valid syntax, the following operation is allowed since x is not explicitly accessed:

  <rpc>
    <edit-config>
      <target><candidate/></target>
      <config operation='replace'>
        <y xmlns="example"/>
      </config>
    </edit-config>
  </rpc>

Finally, the candidate is validate and committed:

  <rpc>
    <commit/>
  </rpc>

The example shown in this Section is also available as a regression test script.

Automatic upgrades

Clixon supports an EXPERIMENTAL xml changelog feature based on "draft-wang-netmod-module-revision-management-01" (Zitao Wang et al) where changes to the Yang model are documented and loaded into Clixon. The implementation is not complete.

When upgrading, the system parses the changelog and tries to upgrade the datastore automatically. This featire is experimental and has several limitations.

You enable the automatic upgrading by registering the changelog upgrade method in clixon_plugin_ini() using wildcards:

   upgrade_callback_register(h, xml_changelog_upgrade, NULL, 0, 0, NULL);

The transformation is defined by a list of changelogs. Each changelog defined how a module (defined by a namespace) is transformed from an old revision to a nnew. Example from test_upgrade_auto.sh

<changelogs xmlns="http://clicon.org/xml-changelog">
  <changelog>
    <namespace>urn:example:b</namespace>
    <revfrom>2017-12-01</revfrom>
    <revision>2017-12-20</revision>
    ...
  <changelog>
</changelogs>

Each changelog consists of set of (orderered) steps:

    <step>
      <name>1</name>
      <op>insert</op>
      <where>/a:system</where>
      <new><y>created</y></new>
    </step>
    <step>
      <name>2</name>
      <op>delete</op>
      <where>/a:system/a:x</where>
    </step>

Each step has an (atomic) operation:

  • rename - Rename an XML tag
  • replace - Replace the content of an XML node
  • insert - Insert a new XML node
  • delete - Delete and existing node
  • move - Move a node to a new place

Step have the following mandatory arguments:

  • where - An XPath node-vector pointing at a set of target nodes. In most operations, the vector denotes the target node themselves, but for some operations (such as insert) the vector points to parent nodes.
  • when - A boolean XPath determining if the step should be evaluated for that (target) node.

Extended arguments:

  • tag - XPath string argument (rename)
  • new - XML expression for a new or transformed node (replace, insert)
  • dst - XPath node expression (move)

Step summary:

  • rename(where:targets, when:bool, tag:string)
  • replace(where:targets, when:bool, new:xml)
  • insert(where:parents, when:bool, new:xml)
  • delete(where:parents, when:bool)
  • move(where:parents, when:bool, dst:node)

Flowcharts

This section contains "pseudo" flowcharts showing the dynamics of the configuration databases in the startup phase.

The flowchart starts in one of the modes (none, init, startup, running):

Init mode

                 reset     
running   |--------+------------> GOTO EXTRA XML

Running mode

running   ----+                   |----------+--------> GOTO EXTRA XML
               \ copy   parse  validate OK  / commit 
tmp       ------+-------+------+-----------+

Startup mode

                              reset     
running                         |--------+------------> GOTO EXTRA XML
                parse validate OK       / commit 
startup -------+--+-------+------------+

Failure

failsafe      ----------------------+
                            reset    \ commit
running                       |-------+---------------> GOTO SYSTEM UP
              parse validate fail 
tmp/startup --+-----+---------------------------------> INVALID XML

Extra XML

running -----------------+----+------> GOTO SYSTEM UP
           reset  loadfile   / merge
tmp     |-------+-----+-----+

System UP

running ----+-----------------------> RUNNING
             \ copy
candidate     +---------------------> CANDIDATE

Invalid XML

               repair     restart

tmp/startup --------+---------+----------------------->

Thanks

Thanks matt smith and dave cornejo for input

References

RFC7895