Skip to content

Validation

Dynaconf allows the validation of settings parameters, in some cases you may want to validate the settings before starting the program.

Lets say you have settings.toml

[default]
version = "1.0.0"
age = 35
name = "Bruno"
DEV_SERVERS = ['127.0.0.1', 'localhost', 'development.com']
PORT = 8001

[production]
PROJECT = "This is not hello_world"

Validating in Python programmatically

At any point of your program you can do:

from dynaconf import Dynaconf, Validator


settings = Dynaconf(
    validators=[
        # Ensure some parameters exists (are required)
        Validator('VERSION', 'AGE', 'NAME', must_exist=True),

        # Ensure some password cannot exist
        Validator('PASSWORD', must_exist=False),

        # Ensure some parameter mets a condition
        # conditions: (eq, ne, lt, gt, lte, gte, identity, is_type_of, is_in, is_not_in)
        Validator('AGE', lte=30, gte=10),

        # validate a value is eq in specific env
        Validator('PROJECT', eq='hello_world', env='production'),

        # Ensure some parameter (string) meets a condition
        # conditions: (len_eq, len_ne, len_min, len_max, cont)
        # Determines the minimum and maximum length for the value
        Validator("NAME", len_min=3, len_max=125),

        # Signifies the presence of the value in a set, text or word
        Validator("DEV_SERVERS", cont='localhost'),

        # Checks whether the length is the same as defined.
        Validator("PORT", len_eq=4),
    ]
)

The above will raise dynaconf.validators.ValidationError("AGE must be lte=30 but it is 35 in env DEVELOPMENT") and dynaconf.validators.ValidationError("PROJECT must be eq='hello_world' but it is 'This is not hello_world' in env PRODUCTION")

Lazy validation

Instead of passing validators= argument to Dynaconf class you can register validators after the instance is created.

settings = Dynaconf(...)

custom_msg = "You cannot set {name} to {value} in env {env}"
settings.validators.register(
    Validator("MYSQL_HOST", eq="development.com", env="DEVELOPMENT"),
    Validator("MYSQL_HOST", ne="development.com", env="PRODUCTION"),
    Validator("VERSION", ne=1, messages={"operations": custom_msg}),
    Validator("BLABLABLA", must_exist=True),
)

Having the list of validators registered you can call one of:

Validate and raise on the first error:

settings.validators.validate()

Validate and accumulate errors, raise only after all validators are evaluated.

settings.validators.validate_all()

The raised ValidationError will have an attribute details holding information about each error raised.

Providing default or computed values

Validators can be used to provide default or computed values.

Default values

Validator("FOO", default="A default value for foo")

Then if not able to load the values from files or environment this default value will be set for that key.

Warning

YAML reads empty keys as None and in that case defaults are not applied, if you want to change it set apply_default_on_none=True either globally to Dynaconf class or individually on a Validator.

Computed values

Sometimes you need some values to be computed by calling functions, just pass a callable to the default argument.

Validator("FOO", default=my_function)

then

def my_function(settings, validator):
    return "this is computed during validation time"

If you want to be lazily evaluated, my_function has to be redefined as

def my_lazy_function(value, **context):
    """
    value: Default value passed to the validator, defaults to `empty`
    context: A dictionary containing
            env: All the environment variables
            this: The settings instance
    """
    return "When the first value is accessed, then the my_lazy_function will be called"

Subsequently

from dynaconf.utils.functional import empty
from dynaconf.utils.parse_conf import Lazy

Validator("FOO", default=Lazy(empty, formatter=my_lazy_function))

You can also use dot-delimited paths for registering validators on nested structures:

# Register validators
settings.validators.register(

    # Ensure the database.host field exists.
    Validator('DATABASE.HOST', must_exist=True),

    # Make the database.password field optional. This is a default behavior.
    Validator('DATABASE.PASSWORD', must_exist=None),
)

# Fire the validator
settings.validators.validate()

Combining validators

Validators can be combined using:

| or operator.

Validator('DATABASE.USER', must_exist=True) | Validator('DATABASE.KEY', must_exist=True)

& and operator.

Validator('DATABASE.HOST', must_exist=True) & Validator('DATABASE.CONN', must_exist=True)

CLI and dynaconf_validators.toml

It is possible to define validators in TOML file called dynaconf_validators.toml placed in the same folder as your settings files.

dynaconf_validators.toml equivalent to program above

[default]

version = {must_exist=true}
name = {must_exist=true}
password = {must_exist=false}

# dot notation is also supported
'a_big_dict.nested_1.nested_2.nested_3.nested_4' = {must_exist=true, eq=1}

  [default.age]
  must_exist = true
  lte = 30
  gte = 10

[production]
project = {eq="hello_world"}

Then to fire the validation use:

$ dynaconf validate

This returns code 0 (success) if validation is ok.

Info

All values in dynaconf are parsed using toml format, TOML tries to be smart and infer the type of the settings variables, some variables will be automatically converted to integer:

FOO = "0x..."  # hexadecimal
FOO = "0o..."  # Octal
FOO = "0b..."  # Binary

All cases are on toml specs https://github.com/toml-lang/toml/blob/master/toml.abnf

If you need to force a specific type casting there are 2 options.

  1. Use double quoted for strings ex: `FOO = "'0x...'" will be string.
  2. Specify the type using @ ex: FOO = "@str 0x..." (available converters are @int, @float, @bool, @json)

Selective Validation

New in 3.1.6

You can also choose what sections of the settings you do or don't want to validate. This is useful when:

  • You want to add additional validators after the settings object is created.
  • You want settings validated only when certain sections of your project are loaded.
  • You want to offer incremental configuration levels, validating only what is needed.

Selective validation can be performed both when creating a settings object and when calling validate on a settings object's validators. The new arguments accept either a string representing a settings path or a list of strings representing settings paths.

A settings path starts at the top level element and can be specified down to the lowest component. For example: my_settings.server.user.password can have the following settings paths passed in server, server.user, server.user.password.

Note: Selective validation matches the passed in value(s) to settings paths that start with that value. This means that passing exclude="FOO" will exclude not only paths that start with FOO but also FOOBAR.

Examples:

-- config.py --

...
# create a settings object, validating only settings under settings.server
settings = Dynaconf(
    validators=[
        Validator(
            "server.hostname",
            "server.port",
            "server.auth",
            must_exist=True
        ),
        Validator(
            "module1.value1",
            "module1.value2",
            "module1.value3",
            must_exist=True
        ),
        Validator(
            "module2.value1",
            "module2.value2",
            "module2.bad",
            must_exist=True
        )
    ],
    validate_only="server"
)

-- module1.py --

...
# call validation on module1 settings
settings.validators.validate(only=["module1"])

-- module2.py --

...
# call validation on module2 settings
# ignore validation for a subsection of module2's settings
settings.validators.validate(
    only=["module2"],
    exclude=["module2.bad"]
)

Validate only current env

You can specify if you want to validate all environments defined for a validator (default behavior) or only the current environement. In the first case, the validators will run on all possible settings defined in their list of environment, while in the latter the validators with environments different from the current environment will be skipped.

This is useful when your configuration for different environments (let's say production and development) comes from different files you don't necesseraly have access to during development. You would want to write different validators for your development and production environments, and only run the right validator for the current environment.

Here is an example of the option using:

  • settings.toml

    [development]
    version = "dev"
    age = 35
    name = "Bruno"
    servers = ['127.0.0.1', 'localhost', 'development.com']
    PORT = 80
    
    [production]
    version = "1.0.0"
    age = 35
    name = "Bruno"
    servers = ['production.com']
    PORT = 443
    

  • .secrets.toml

    [production]
    api_key = 'secret_api_key'
    

You could then have these validators:

from dynaconf import Dynaconf, Validator

settings = Dynaconf(
    settings_files=['setting.toml', '.secrets.toml'],
    environments=True,
    validators=[
        # Ensure some parameters exists for both envs
        Validator('VERSION', 'NAME', 'SERVERS', envs=['development', 'production'], must_exist=True),

        # Ensure some parameter validate certain condition in dev env
        Validator("SERVERS", env='development', cont='localhost'),

        # Ensure some parameter exists in production env
        Validator('API_KEY', env='production', must_exist=True),
    ]
)
And suppose during development, when settings.current_env == 'development', you don't have the file .secrets.toml.

Running settings.validators.validate() will fail even if settings.current_env == 'development', because by default all validators will run on all of their environments, whether or not it is the current env. However you could create your settings with the parameter validate_only_current_env=True, and nothing will be raised if settings.current_env == 'development'. Still, if settings.current_env == 'production', it will fail, forcing you to have .secrets.toml file in the directory in production, but not necessarily during development.


Last update: 2022-06-02