Config Normalizer

Normalizing is the process of converting stream of input tokens, received from Tokenizer, to the stream of output tokens, representing ConfDB key’s paths.

Normalizers are device-dependent and represent the translation from device-depended tokenized config to abstract syntax of ConfDB. Tokenizers do a lot of deal performing primal stream analysis and applying configuration context, reducing complexity of normalizing.

Both tokenizer and normalizer works in single pipeline. Tokenizer reads input stream and emits parsed tokens whenever they are ready. Normalizer performs normalization and emits ConfDB syntax after matching. Ability to work in pipelines greatly reduces memory footprint of ConfDB processing, avoiding multiple copies of same data.

Profile Integration


Refer to Profile API

Following profile parameters are responsible for normalizer configuration:


String containing name of config normalizer to use. Contains normalizer class name. Normalizer must be subclass of BaseNormalizer and must be located in the module.


Optional dict, containing config normalizer settings to be passed to Normalizer’s constructor. Depends upon normalizer implementation.

get_config_normalizer(cls, object):

Classmethod returning actual config normalizer name and its settings for selected managed object. Returns (config_normalizer, config_normalizer_settings) by default. Should be overriden in profile if normalizer or settings depends on platform or software version.


object – ManagedObject reference


tuple of (config tokenizer name, config tokenizer settings). Must return (None, None) if platform is not supported.

Normalizer API

Normalizers must be subclass of BaseNormalizer and must be located in profile’s confdb.normalizer subpackage (, where X.Y is profile name).


Ensure the sa/profiles/X/Y/confdb/ file exists and empty.

Code boilerplate

Normalizer code boilerplate:

# NOC modules
from noc.core.confdb.normalizer.base import BaseNormalizer, match, ANY, REST

class MyNormalizer(BaseNormalizer):

@match decorator

Normalizer contains generator functions (python functions, containing yield statement) annotated by @match decorators. Example:

@match("no", "lldp")
def normalize_no_lldp(self, tokens):

@match decorator compared to whole input tokens line. On full match the generator function will be called passing all input tokens line as tokens parameter. So example generator will be fired when receiving [“no”, “lldp”] tokens and tokens parameter will contain received tokens.

@match decorator may be applied several times upon same generator function, allowing multiple matching variants. Example:

@match("no", "lldp")
@match("no", "lldp", "protocol")
def normalize_no_lldp(self, tokens):

Will match [“no”, “lldp”] or [“no”, “lldp”, “protocol”].

Note, the match is exact for the whole line. Previous example will not match the [“no”, “lldp”, “status”] line.

@match macros

@match decorator allows to use one of following wildcard macros instead of exact string. All macros must be imported from noc.core.confdb.normalizer.base module.




Match any string in given position


Match all tokens to the rest of the line

tokens parameter can be used to get real value of matched tokens:

@match("interface", ANY, "description", REST)
def normalize_inteface_description(self, tokens):
    if_name = tokens[1]
    description = " ".join(tokens[3:])

Multiple occurencies and various combination of macros are possible

Generator functions

Generator functions behind @match decorator must be python generators containing at least one yield statement. Generator must yield tuples of full paths of ConfDB syntax. Example:

@match("interface", ANY, "description", REST)
def normalize_interface_description(self, tokens):
    yield "interfaces", tokens[2], "description", " ".join(tokens[3:])

Note that yield argument must be a tuple. Python treats comma-separated values as tuples. So:

"a", "b"

is a tuple. Multi-line tuples must be enclosed by brackets:


For single-value tuples either of following variants may be used:


Multiple yield statement are possible.

Note that Python 2.7 does not allow to mix yield and return in same function, so StopIteration exception must be used to exit generator:

def my_generator(....):
    raise StopIteration

@match generator must meet following rules: * yield path MUST meet ConfDB syntax * Interface names must be normalized by interface_name method (See API) * Avoid to yield ConfDB paths directly, use syntax generators instead.

Syntax generators are defined in ConfDB syntax and allow to inject named parameters to path. So our generator example must be rewritten as:

@match("interface", ANY, "description", REST)
def normalize_interface_description(self, tokens):
    yield self.make_interface_description(
        description=" ".join(tokens[3:])

deferrable generators

In rare case single tokenized line does not contain all necessary information and normalized line must be constructed from several lines. To handle such cases @match generator may yield the deferrable expression. deferrable is and partial expression bound to particular key. Different @match generators may yield deferrable with same keys, refining necessary parameters. When all prerequisitions are met, deferred call became a real expression.

Consider following RouterOS config:

/ip address
add address= interface=bridge1 network=

routeros tokenizer converts it to:

["/ip", "address", "0", "address", ""]
["/ip", "address", "0", "interface", "bridge1"]
["/ip", "address", "0", "address", ""]

As you can see, interface name and addresses are on different lines (And there are reasons to deal it so). So normalizing is two step:

@match("/ip", "address", INTEGER, "address", IPv4_PREFIX)
def normalize_interface_address(self, tokens):
    yield self.defer(
        "ip.address.%s" % tokens[2],

@match("/ip", "address", INTEGER, "interface", ANY)
def normalize_interface_address_interface(self, tokens):
    yield self.defer(
        "ip.address.%s" % tokens[2],

First generator yields deferrable with key ip.address.0. Zero is taken from input tokens. Second parameter is the syntax generator to be called. interface parameter is not known yet, so we denote it with deferrable named interface. address parameter is known, so we attach it to deferrable. Generator yields no output tokens, as deferrrable is not fully resolved.

Next generator will match interface part. Key ip.address.0 is referred to already known deferrable. So it attaches only interface parameter. Generator lookups pending deferrables by keys and applies interface parameter. As deferrable became fully resolved, it converts to actual make_unit_inet_address and yield output tokens immediately.


Sometimes is a good practice to set some flags, representing global status, to alter behavior of following matches. Three methods for context manipulation are available:

  • set_context

  • get_context

  • has_context

Normalizer methods

__init__(self, param1=default1, .., paramN=defaultN)

Constructor. Should be overriden if Normalizer accepts additional configuration from profile. Refer to config_normalizer_settings for details.

  • param1 – Custom configuration parameter #1 with default value

  • paramN – Custom configuration parameter #N with default value

set_context(self, name, value)

Set context flag name to value

  • name – String containing flag name

  • value – Flag value



get_context(self, name, default=None)

Get context flag name, returning its value or default

  • name – String containing flag name

  • default – Default value


Key value or default if key not found

has_context(self, name)

Check context has flag name


name – String containing flag name


True if context has key name, False otherwise

interface_name(self, *args)

Convert interface name using profile’s convert_interface_name method



Parts of interface name


String containing normalized interface name

to_prefix(self, address, netmask)

Convert IPv4 address and netmask to prefix form (address/bits

  • address – IPv4 address

  • netmask – IPv4 netmask


IP address in prefix form

defer(context, gen=None, **kwargs)

Create deferable with name context. Return actual gen call when all prerequisites are met.

  • context – Name of deferable context

  • gen – ConfDB syntax generator callable

  • kwargs – syntax generator parameters, may contain deferrable