"""This module provides ``kedro.config`` with the functionality to load one
or more configuration files from specified paths, and format template strings
with the values from the passed dictionary.
"""
import re
from copy import deepcopy
from pathlib import Path
from typing import Any, Dict, Iterable, Optional
import jmespath
from kedro.config import AbstractConfigLoader
from kedro.config.common import _get_config_from_patterns, _remove_duplicates
IDENTIFIER_PATTERN = re.compile(
r"""\$\{
(?P<path>[A-Za-z0-9_\.]+) # identifier
(?:\|(?P<default>[^}]*))? # optional default value
\}""",
re.VERBOSE,
)
FULL_STRING_IDENTIFIER_PATTERN = re.compile(
r"^" + IDENTIFIER_PATTERN.pattern + r"$", re.VERBOSE
)
[docs]class TemplatedConfigLoader(AbstractConfigLoader):
"""
Extension of the ``ConfigLoader`` class that allows for template values,
wrapped in brackets like: ${...}, to be automatically formatted
based on the configs.
The easiest way to use this class is by setting the `CONFIG_LOADER_CLASS` constant
in `settings.py`.
Example:
::
>>> # in settings.py
>>> from kedro.config import TemplatedConfigLoader
>>>
>>> CONFIG_LOADER_CLASS = TemplatedConfigLoader
>>> CONFIG_LOADER_ARGS = {
>>> "globals_pattern": "*globals.yml",
>>> }
The contents of the dictionary resulting from the `globals_pattern` get
merged with the ``globals_dict``. In case of conflicts, the keys in
``globals_dict`` take precedence.
If the formatting key is missing from the dictionary, the default template
value is used (the format is "${key|default value}"). If no default is set,
a ``ValueError`` will be raised.
Global parameters can be namespaced as well. An example could work as follows:
`globals.yml`
::
bucket: "my_s3_bucket"
environment: "dev"
datasets:
csv: "pandas.CSVDataSet"
spark: "spark.SparkDataSet"
folders:
raw: "01_raw"
int: "02_intermediate"
pri: "03_primary"
fea: "04_feature"
`catalog.yml`
::
raw_boat_data:
type: "${datasets.spark}"
filepath: "s3a://${bucket}/${environment}/${folders.raw}/boats.csv"
file_format: parquet
raw_car_data:
type: "${datasets.csv}"
filepath: "s3://${bucket}/data/${environment}/${folders.raw}/cars.csv"
This uses ``jmespath`` in the background. For more information see:
https://github.com/jmespath/jmespath.py and https://jmespath.org/.
"""
[docs] def __init__(
self,
conf_source: str,
env: str = None,
runtime_params: Dict[str, Any] = None,
*,
base_env: str = "base",
default_run_env: str = "local",
globals_pattern: Optional[str] = None,
globals_dict: Optional[Dict[str, Any]] = None,
):
"""Instantiates a ``TemplatedConfigLoader``.
Args:
conf_source: Path to use as root directory for loading configuration.
env: Environment that will take precedence over base.
runtime_params: Extra parameters passed to a Kedro run.
base_env:
default_run_env:
globals_pattern: Optional keyword-only argument specifying a glob
pattern. Files that match the pattern will be loaded as a
formatting dictionary.
globals_dict: Optional keyword-only argument specifying a formatting
dictionary. This dictionary will get merged with the globals dictionary
obtained from the globals_pattern. In case of duplicate keys, the
``globals_dict`` keys take precedence.
"""
super().__init__(
conf_source=conf_source, env=env, runtime_params=runtime_params
)
self.base_env = base_env
self.default_run_env = default_run_env
self._config_mapping = (
_get_config_from_patterns(
conf_paths=self.conf_paths,
patterns=[globals_pattern],
ac_template=False,
)
if globals_pattern
else {}
)
globals_dict = deepcopy(globals_dict) or {}
self._config_mapping = {**self._config_mapping, **globals_dict}
@property
def conf_paths(self):
"""Property method to return deduplicated configuration paths."""
return _remove_duplicates(self._build_conf_paths())
[docs] def get(self, *patterns: str) -> Dict[str, Any]:
"""Tries to resolve the template variables in the config dictionary
provided by the ``ConfigLoader`` (super class) ``get`` method using the
dictionary of replacement values obtained in the ``__init__`` method.
Args:
*patterns: Glob patterns to match. Files, which names match
any of the specified patterns, will be processed.
Returns:
A Python dictionary with the combined configuration from all
configuration files. **Note:** any keys that start with `_`
will be ignored. String values wrapped in `${...}` will be
replaced with the result of the corresponding JMESpath
expression evaluated against globals (see `__init` for more
configuration files. **Note:** any keys that start with `_`
details).
Raises:
ValueError: malformed config found.
"""
config_raw = _get_config_from_patterns(
conf_paths=self.conf_paths, patterns=patterns, ac_template=True
)
return _format_object(config_raw, self._config_mapping)
def _build_conf_paths(self) -> Iterable[str]:
run_env = self.env or self.default_run_env
return [
str(Path(self.conf_source) / self.base_env),
str(Path(self.conf_source) / run_env),
]
def _format_object(val: Any, format_dict: Dict[str, Any]) -> Any:
"""Recursive function that loops through the values of a map. In case another
map or a list is encountered, it calls itself. When a string is encountered,
it will use the `format_dict` to replace strings that look like `${expr}`,
where `expr` is a JMESPath expression evaluated against `format_dict`.
Some notes on behavior:
* If val is not a dict, list or string, the same value gets passed back.
* If val is a string and does not match the ${...} pattern, the same
value gets passed back.
* If the value inside ${...} does not match any keys in the dictionary,
the error is raised, unless a default is provided.
* If the default is provided with ${...|default}, and the key is not
found in the dictionary, the default value gets passed back.
* If the ${...} is part of a larger string, the corresponding entry in
the `format_dict` gets parsed into a string and put into the
larger string.
Examples:
val = "${test_key}" with format_dict = {'test_key': 'test_val'} returns
'test_val'
val = 5 (i.e. not a dict, list or string) returns 5
val = "test_key" (i.e. does not match ${...} pattern returns 'test_key'
(irrespective of `format_dict`)
val = "${wrong_test_key}" with format_dict = {'test_key': 'test_val'}
raises ``ValueError``
val = "string-with-${test_key}" with format_dict = {'test_key': 1000}
returns "string-with-1000"
val = "${wrong_test_key|default_value}" with format_dict = {}
returns 'default_value'
Args:
val: If this is a string of the format `${expr}`, it gets replaced
by the result of JMESPath expression
format_dict: A lookup from string to string with replacement values
Returns:
A string formatted according to the ``format_dict`` input.
Raises:
ValueError: The input data is malformed.
"""
def _format_string(match):
value = jmespath.search(match.group("path"), format_dict)
if value is None:
if match.group("default") is None:
raise ValueError(
f"Failed to format pattern '{match.group(0)}': "
f"no config value found, no default provided"
)
return match.group("default")
return value
if isinstance(val, dict):
new_dict = {}
for key, value in val.items():
if isinstance(key, str):
formatted_key = _format_object(key, format_dict)
if not isinstance(formatted_key, str):
raise ValueError(
f"When formatting '{key}' key, only string values can be used. "
f"'{formatted_key}' found"
)
key = formatted_key
new_dict[key] = _format_object(value, format_dict)
return new_dict
if isinstance(val, list):
return [_format_object(e, format_dict) for e in val]
if isinstance(val, str):
# Distinguish case where entire string matches the pattern,
# as the replacement can be of a different type
match_full = FULL_STRING_IDENTIFIER_PATTERN.match(val)
if match_full:
return _format_string(match_full)
return IDENTIFIER_PATTERN.sub(lambda m: str(_format_string(m)), val)
return val