Introduction
opConfig 3 introduces the ability to 'push' configuration changes to devices, complete with error detection, support for change rollback and complete audit trails.
This document describes the config push infrastructure.
Concepts
opConfig has always supported 'command sets', which consist of (individual) commands that can be sent to devices and whose output is collected and analyzed. Command sets are expected to be idempotent, safely repeatable and without side effects. Command sets don't support sequencing or error detection, and thus are not suitable (or intended!) for modifying device configurations.
For config push, opConfig requires that the administrator defines one or more suitable 'config sets'. A config set is a list of commands which are sent to selected devices with the expectation that something on the device is changed by the sequence of commands; to perform such operations safely, a config set can also define error detection patterns and possible reactions to any errors.
The Anatomy of a Config Set
Storage Format and Location
The primary storage and exchange format for config sets is JSON; example config sets can be find in install/config_sets.d
/ and it is recommended that you store your actively developed config sets in conf/config_sets.d
/.
To make a config set active you have to import it into opConfig: the JSON files themselves are not consulted.
Metadata
A config set must have a unique name
property, and may have a (free-form) description
.
Every config set is automatically assigned a revision
by opConfig, which cannot be changed once set. opConfig automatically tracks the history and evolution of your config sets via name and revision: whenever a config set file is (re)imported into opConfig, the config set is checked for differences against the most recent known revision; if there are any, the set is saved with a new revision.
In general, only the most recent revision of a config set will be applied to devices (but this can be overridden).
Candidate Selection
Configuration changes and the commands to perform them are usually specific to a target platform. To avoid misapplication of config sets opConfig provides a flexible mechanism for selecting candidate devices for each configuration set. This is done by providing rules in the optional filter
section.
The filter section holds any number of selection clauses, each consisting of a property path and a selection criterion. To pass as a candidate, all filters must match.
The property path is given in "dotted notation", compatible with opnode_admin
.pl's act=show
and act=set
operations.
For example, os_info.major
would be used to refer to the detected OS major version (which is stored within the os_info
substructure).
The selection criterion can take one of three forms:
- an explicit single value
for example,"os_info.platform": "x86_64"
would match if the node's OS platform value is exactly equal to "x86_64". - a list of alternative values
for example"name": [ "nodeA", "nodeB"]
would match either of the two nodes namednodeA
ornodeB
. The comparison is again strict equality.
The filter clause is considered a match if one or more choices amongst the list of alternatives do match. - a regular expression
for example"os_info.os": "/IOS/i"
would select nodes whose OS contains the string "IOS" (or "ios", "IoS" etc.) anywhere.
Note that the regular expression must be given as a string, starting and ending with a "/" and optionally including the "i" modifier for case-insensitive matching.
All known properties for a node are available for filtering; please consult the output of opnode_admin.pl
act=show
or act=export
for a list of common properties.
Notifications
opConfig can optionally notify interested parties when a config set was applied to devices.
This is configured by providing the relevant contact details in the optional notify
section of the config set. When a config set application is scheduled, extra notification recipients can be specified but the ones given in the config set are always included.
In version 3.0.0 opConfig supports notification by email only.
For example, "notify": {"email": ["first@mycompany.com","second@othercompany.com" ]}
would have opConfig send emails to both of the given addresses.
Notifications list the candidate nodes and status information for the config set application for each, e.g. complete success, partial success or complete failure.
Error Handling
A config set can contain a number of error detection and handling directives, contained within an error_handling
block.
Error Detection
Error handling requires that one or more match
properties are given, which are applied to each command's response to determine if the command is considered unsuccessful.
By default no error detection is performed, and all commands are considered successful no matter what their output is.
Similar to the candidate filtering mechanism described above, match properties must either be strings or strings containing a regular expression. Strings are compared exactly. More than one match property is possible.
If the command output meets one or more match
properties, then the command is considered unsuccessful.
Reacting to Errors
If error detection is configured, three different reactions are possible:
- Continue the config set application
This is the default behaviour. opConfig will keep and report a tally of successful versus failed commands, but will try to apply all of them in sequence. - Abort the config set application
To select this option, the propertybreak_on_error
must be set totrue
. In this case, the first failed command in a set causes all other commands to be skipped. - Apply Rollback commands
To revert the configuration back to a desired state, you would have to provide a sequence ofrollback-commands
in your error-handling section.
Rolling back woud generally be combined withbreak_on_error
, but opConfig does not require it.
Stages and their Commands
opConfig distinguishes between five different stages for the application of a config set. You can provide any number of commands for each of the stages; the command
stage must contain one or more commands.
A stage definition consists of a list of command strings which are sent to the device in sequence, one line at a time, exactly as they're given.
Each command must be complete by itself, and must start and finish at a configured prompt for your device's personality. This means that in general, interaction with a command across multiple lines, responding to confirmation challenges etc. is not directly possible.
There are two exceptions:
- opConfig treats the 'pseudo-commands'
__leave__
and__enter__
specially, and leaves or enters configuration mode.
Note that there are two "_" on each side of the command. These special commands can be handy for rolling back changes, if your particular rollback command requires to be performed outside of configuration mode. - Any command that starts with a single "_" is treated as a macro invocation, with optional arguments.
Macros have to be defined suitably in your device personality's "phrasebook", and they differ from plain individual commands in that macros can define sequences of send-expect-prompt steps.
This is how limited multi-step interaction with a device is possible.
For example, the default Cisco phrasebook as shipped with opConfig defines the macroreload_in
, to be used in a command like"_reload_in 15
". This would configure a Cisco IOS device for a forced reboot in N minutes. This has to be done using a macro because setting up thereload
operation requires confirmation. There is also a macro calledreload
which, when triggered by the command "_reload
" would restart your device.
The pre-commands
Stage (optional)
Commands listed here are run first, in privileged mode and outside of configuration mode. Error detection is not available in this stage.
A common example command could be reload in 15
for Cisco devices.
The post-commands
and post-rollback-commands
Stages (optional)
These stages are run last, after leaving configuration mode and still in privileged mode. No error handling is available.
The "plain" post-commands stage is chosen if error handling was disabled or all commands succeeded; on the other hand, if a rollback was performed, only the post-rollback-commands are applied.
The commands
Stage (required)
Commands in this stage are run in configuration mode (which implies privileged mode). Error handling is available, hence the sequence of commands may be aborted early.
The rollback-commands
Stage (optional)
If error detection is enabled and if this stage is provided, then its commands will be applied as soon as errors are detected. The rollback commands are sent while in configuration mode, and no error handling is possible.
the five different command sections, pre, post, commands, rollback-commands and post-rollback-commands
Config Set Management
fixme tbc
import from cli only at this time; view is possible from gui
can list, export, activate and deactivate
Config Push Scheduling
fixme tbc
time formats
gui or cmdline
limitations for non-admin user, filters cannot be overridden, just refined
parallelism
limited rescheduling on failure
aborting via gui, or from cli
Limitations in opConfig 3.0.0
fixme tbc
pbs only primed fullly for cisco ios devices, must have config mode prompt and enter/leave macros
no config set editing from the gui, no import or export of csets from the gui either
Config Set Example
This fully functional example config set ships with opConfig 3.0.0 as install/config_sets.d/IOS-Configuration-Best-Practices.json
.
{ "name": "IOS - Configuration Best Practices", "description": "A configuration set to configure the IOS device, by enabling and disabling various services and features.", "filter": { "os_info.os": "IOS" }, "notify": { "email": "you@company.com" }, "error_handling": { "match": [ "/Invalid input detected/" ], "break_on_error": false }, "pre-commands": [ "_reload_in 5" ], "commands": [ "no ip http server", "no ip http secure-server", "no ip finger", "no service finger", "no service udp-small-servers", "no service tcp-small-servers", "no boot network", "no service config", "service password-encryption", "service timestamps debug datetime msec", "service timestamps log datetime msec", "service sequence-numbers", "service tcp-keepalives-in", "service tcp-keepalives-out", "no ip source-route", "line con 0 ", "exec-timeout 10 0", "exit", "line vty 0 4", "exec-timeout 10 0", "exit" ], "post-commands": [ "reload cancel" ] }