Configuration

This section contains detailed information about configuration, for which the relevant API documentation can be found in kedro.config.ConfigLoader

Note: This documentation is based on Kedro 0.17.1, if you spot anything that is incorrect then please create an issue or pull request.

Local and base configuration

We recommend that you keep all configuration files in the conf directory of a Kedro project. However, if you prefer, you may point Kedro to any other directory and change the configuration paths by setting the CONF_ROOT variable in src/<project-package>/settings.py as follows:

# ...
CONF_ROOT = "new_conf"

Loading

Kedro-specific configuration (e.g., DataCatalog configuration for IO) is loaded using the ConfigLoader class:

from kedro.config import ConfigLoader

conf_paths = ["conf/base", "conf/local"]
conf_loader = ConfigLoader(conf_paths)
conf_catalog = conf_loader.get("catalog*", "catalog*/**")

This will recursively scan for configuration files firstly in conf/base/ and then in conf/local/ directory according to the following rules:

  • ANY of the following is true:

    • filename starts with catalog OR

    • file is located in a sub-directory whose name is prefixed with catalog

  • AND file extension is one of the following: yaml, yml, json, ini, pickle, xml or properties

Configuration information from files stored in base or local that match these rules is merged at runtime and returned in the form of a config dictionary:

  • If any 2 configuration files located inside the same environment path (conf/base/ or conf/local/ in this example) contain the same top-level key, load_config will raise a ValueError indicating that the duplicates are not allowed.

Note: Any top-level keys that start with _ character are considered hidden (or reserved) and therefore are ignored right after the config load. Those keys will neither trigger a key duplication error mentioned above, nor will they appear in the resulting configuration dictionary. However, you may still use such keys for various purposes. For example, as YAML anchors and aliases.

  • If 2 configuration files have duplicate top-level keys, but are placed into different environment paths (one in conf/base/, another in conf/local/, for example) then the last loaded path (conf/local/ in this case) takes precedence and overrides that key value. ConfigLoader.get(<pattern>, ...) will not raise any errors, however a DEBUG level log message will be emitted with the information on the over-ridden keys.

  • If the same environment path is passed multiple times, a UserWarning will be emitted to draw attention to the duplicate loading attempt, and any subsequent loading after the first one will be skipped.

Additional configuration environments

In addition to the 2 built-in configuration environments, it is possible to create your own. Your project loads conf/base/ as the bottom-level configuration environment but allows you to overwrite it with any other environments that you create. You are be able to create environments like conf/server/, conf/test/, etc. Any additional configuration environments can be created inside conf folder and loaded by running the following command:

kedro run --env=test

If no env option is specified, this will default to using local environment to overwrite conf/base.

Note: If, for some reason, your project does not have any other environments apart from base, i.e. no local environment to default to, you will need to customise KedroContext to take env="base" in the constructor and then specify your custom KedroContext subclass in src/<python-package>/settings.py under CONTEXT_CLASS key.

If you set the KEDRO_ENV environment variable to the name of your environment, Kedro will load that environment for your kedro run, kedro ipython, kedro jupyter notebook and kedro jupyter lab sessions.

export KEDRO_ENV=test

Note: If you specify both the KEDRO_ENV environment variable and provide the --env argument to a CLI command, the CLI argument takes precedence.

Templating configuration

Kedro also provides an extension TemplatedConfigLoader class that allows to template values in your configuration files. TemplatedConfigLoader is available in kedro.config, to apply templating to your project, you will need to update the register_config_loader hook implementation in your src/<project-name>/hooks.py:

from kedro.config import TemplatedConfigLoader  # new import


class ProjectHooks:
    @hook_impl
    def register_config_loader(self, conf_paths: Iterable[str]) -> ConfigLoader:
        return TemplatedConfigLoader(
            conf_paths,
            globals_pattern="*globals.yml",  # read the globals dictionary from project config
            globals_dict={  # extra keys to add to the globals dictionary, take precedence over globals_pattern
                "bucket_name": "another_bucket_name",
                "non_string_key": 10,
            },
        )

Let’s assume the project contains a conf/base/globals.yml file with the following contents:

bucket_name: "my_s3_bucket"
key_prefix: "my/key/prefix/"

datasets:
    csv: "pandas.CSVDataSet"
    spark: "spark.SparkDataSet"

folders:
    raw: "01_raw"
    int: "02_intermediate"
    pri: "03_primary"
    fea: "04_feature"

The contents of the dictionary resulting from globals_pattern get merged with the globals_dict dictionary. In case of conflicts, the keys from the globals_dict dictionary take precedence. The resulting global dictionary prepared by TemplatedConfigLoader will look like this:

{
    "bucket_name": "another_bucket_name",
    "non_string_key": 10,
    "key_prefix": "my/key/prefix",
    "datasets": {"csv": "pandas.CSVDataSet", "spark": "spark.SparkDataSet"},
    "folders": {
        "raw": "01_raw",
        "int": "02_intermediate",
        "pri": "03_primary",
        "fea": "04_feature",
    },
}

Now the templating can be applied to the configs. Here is an example of a templated conf/base/catalog.yml:

raw_boat_data:
    type: "${datasets.spark}"  # nested paths into global dict are allowed
    filepath: "s3a://${bucket_name}/${key_prefix}/${folders.raw}/boats.csv"
    file_format: parquet

raw_car_data:
    type: "${datasets.csv}"
    filepath: "s3://${bucket_name}/data/${key_prefix}/${folders.raw}/${filename|cars.csv}"  # default to 'cars.csv' if the 'filename' key is not found in the global dict

Note: TemplatedConfigLoader uses jmespath package in the background to extract elements from global dictionary. For more information about JMESPath syntax please see: https://github.com/jmespath/jmespath.py.

Jinja2 support

From version 0.17.0 TemplateConfigLoader also supports Jinja2 template engine alongside the original template syntax. Below is the example of a catalog.yml file, which uses both features:

{% for speed in ['fast', 'slow'] %}
{{ speed }}-trains:
    type: MemoryDataSet

{{ speed }}-cars:
    type: pandas.CSVDataSet
    filepath: s3://${bucket_name}/{{ speed }}-cars.csv
    save_args:
        index: true

{% endfor %}

When parsing this configuration file, TemplateConfigLoader will:

  1. Read the catalog.yml and compile it using Jinja2

  2. Use YAML parser to parse the compiled config into a Python dictionary

  3. Expand ${bucket_name} in filepath using the globals_* arguments for the TemplateConfigLoader instance as in the previous examples

The output Python dictionary will look as follows:

{
    "fast-trains": {"type": "MemoryDataSet"},
    "fast-cars": {
        "type": "pandas.CSVDataSet",
        "filepath": "s3://my_s3_bucket/fast-cars.csv",
        "save_args": {"index": True},
    },
    "slow-trains": {"type": "MemoryDataSet"},
    "slow-cars": {
        "type": "pandas.CSVDataSet",
        "filepath": "s3://my_s3_bucket/slow-cars.csv",
        "save_args": {"index": True},
    },
}

Note: Although Jinja2 is a very powerful and extremely flexible template engine, which comes with a wide range of features, we do not recommend to use it to template your configuration unless absolutely necessary. The flexibility of dynamic configuration comes at a cost of significantly reduced readability and much higher maintenance overhead. We believe that, for the majority of analytics projects, dynamically compiled configuration does more harm than good.

Parameters

Loading parameters

Parameters project configuration can be loaded with the help of the ConfigLoader class:

from kedro.config import ConfigLoader

conf_paths = ["conf/base", "conf/local"]
conf_loader = ConfigLoader(conf_paths)
parameters = conf_loader.get("parameters*", "parameters*/**")

The code snippet above will load all configuration files from conf/base and conf/local, which either have the filename starting with parameters or are located inside a folder with name starting with parameters.

Note: Configuration path conf/local takes precedence in the example above since it’s loaded last, therefore any overlapping top-level keys from conf/base will be overwritten by the ones from conf/local.

Calling conf_loader.get() in the example above will throw a MissingConfigException error if there are no configuration files matching the given patterns in any of the specified paths. If this is a valid workflow for your application, you can handle it as follows:

from kedro.config import ConfigLoader, MissingConfigException

conf_paths = ["conf/base", "conf/local"]
conf_loader = ConfigLoader(conf_paths)

try:
    parameters = conf_loader.get("parameters*", "parameters*/**")
except MissingConfigException:
    parameters = {}

Note: kedro.framework.context.KedroContext class uses the approach above to load project parameters.

Parameters can then be used on their own or fed in as function inputs, as described in this section below.

Specifying parameters at runtime

Kedro also allows you to specify runtime parameters for kedro run CLI command. To do that, you need to add the --params command line option and specify a comma-separated list of key-value pairs that will be added to KedroContext parameters and made available to pipeline nodes. Each key-value pair is split on the first colon. Here is an example of triggering Kedro run with extra parameters specified:

kedro run --params param_key1:value1,param_key2:2.0  # this will add {"param_key1": "value1", "param_key2": 2} to parameters dictionary

Note: Parameter keys are always treated as strings. Parameter values are converted to a float or an integer number if the corresponding conversion succeeds, otherwise they are also treated as string.

Note: If, for example, param_key1 parameter has already been defined in the project configuration, the value provided in the CLI option will take precedence and will overwrite the one from the configuration.

Tip: Since key-value pairs are split on the first colon, values can contain colons, but the keys cannot. This is a valid CLI command:

kedro run --params endpoint_url:https://endpoint.example.com

Tip: If any extra parameter key and/or value contains spaces, wrap the whole option contents into quotes:

kedro run --params "key1:value with spaces,key2:value"

Using parameters

Say you have a set of parameters you’re playing around with for your model. You can declare these in one place, for instance conf/base/parameters.yml, so that you isolate your changes to one central location.

step_size: 1
learning_rate: 0.01

You may now reference these parameters in the node definition, using the params: prefix:

def increase_volume(volume, step):
    return volume + step


# in pipeline definition
node(
    func=increase_volume,
    inputs=["input_volume", "params:step_size"],
    outputs="output_volume",
)

You can also group your parameters into nested structures and, using the same method above, load them by top-level key:

step_size: 1
model_params:
    learning_rate: 0.01
    test_data_ratio: 0.2
    number_of_train_iterations: 10000

def train_model(data, model):
    lr = model["learning_rate"]
    test_data_ratio = model["test_data_ratio"]
    iterations = model["number_of_train_iterations"]
    ...


# in pipeline definition
node(
    func=train_model,
    inputs=["input_data", "params:model_params"],
    outputs="output_data",
)

Alternatively, you can also pass parameters to the node inputs and get access to the entire collection of values inside the node function.

def increase_volume(volume, params):
    step = params["step_size"]
    return volume + step


# in pipeline definition
node(
    func=increase_volume, inputs=["input_volume", "parameters"], outputs="output_volume"
)

In both cases, what happened under the hood is that the parameters had been added to the Data Catalog through the method add_feed_dict() (Relevant API documentation: DataCatalog), where they live as MemoryDataSets. This method is also what the KedroContext class uses when instantiating the catalog.

Note: You can use add_feed_dict() to inject any other entries into your DataCatalog as per your use case.

Credentials

Note: For security reasons, we strongly recommend not committing any credentials or other secrets to the Version Control System. Hence, by default any file inside the conf/ folder (and its subfolders) containing credentials in its name will be ignored via .gitignore and not committed to your git repository.

Credentials configuration can be loaded the same way as any other project configuration using the ConfigLoader class:

from kedro.config import ConfigLoader

conf_paths = ["conf/base", "conf/local"]
conf_loader = ConfigLoader(conf_paths)
credentials = conf_loader.get("credentials*", "credentials*/**")

This will load all configuration files from conf/base and conf/local, which either have the filename starting with credentials or are located inside a folder with name starting with credentials.

Note: Configuration path conf/local takes precedence in the example above since it’s loaded last, therefore any overlapping top-level keys from conf/base will be overwritten by the ones from conf/local.

Calling conf_loader.get() in the example above will throw a MissingConfigException error if there are no configuration files matching the given patterns in any of the specified paths. If this is a valid workflow for your application, you can handle it as follows:

from kedro.config import ConfigLoader, MissingConfigException

conf_paths = ["conf/base", "conf/local"]
conf_loader = ConfigLoader(conf_paths)

try:
    credentials = conf_loader.get("credentials*", "credentials*/**")
except MissingConfigException:
    credentials = {}

Note: kedro.framework.context.KedroContext class uses the approach above to load project credentials.

Credentials configuration can then be used on its own or fed into the DataCatalog as described in this section.

AWS credentials

When working with AWS credentials on datasets, you are not required to store AWS credentials in the project configuration files. Instead, you can specify them using environment variables AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, and, optionally, AWS_SESSION_TOKEN. Please refer to the official documentation for more details.

Configuring kedro run arguments

An extensive list of CLI options for a kedro run is available in the Kedro CLI documentation. However, instead of specifying all the command line options to a kedro run via the CLI, you can specify a config file that contains the arguments, say config.yml and run:

$ kedro run --config config.yml

where config.yml is formatted as below (for example):

run:
  tag:
    - tag1
    - tag2
    - tag3
  pipeline: pipeline1
  parallel: true
  node_names:
    - node1
    - node2
  env: env1

Note: If you pass both a configuration file and an option that clashes with one inside the configuration file, the provided option will override the configuration file.