Prefect Docs

# Secrets

# Overview

Very often, workflows require sensitive information to run: API keys, passwords, tokens, credentials, etc. As a matter of best practice, such information should never be hardcoded into a workflow's source code, as the code itself will need to be guarded. Furthermore, sensitive information should not be provided via a Prefect Parameter, because Parameters, like many tasks, can have their results stored.

Prefect provides a mechanism called Secrets for working with sensitive information.

  • Secret tasks, which are special tasks that can be used in your flow when working with sensitive information. Unlike regular tasks, Secret tasks are designed to access sensitive information at runtime and use a special SecretResult to ensure the results are not stored.
  • The prefect.client.secrets API, which provides a lower-level interface for working with sensitive information. This API can be used where tasks are unavailable, including notifications, state handlers, and result handlers.

The most common case for secrets is to authenticate to third-party systems. For more information on best practices to do so, see our deployment recipe on Third Party Authentication.

Keep secrets secret!

Though Prefect takes steps to ensure that Secret objects do not reveal sensitive information, other tasks may not be so careful. Once a secret value is loaded into your flow, it can be used for any purpose. Please use caution anytime you are working with sensitive data.

# Mechanisms

# Local Context

The base Secret class first checks for secrets in local context, under prefect.context.secrets. This is useful for local testing, as secrets can be added to context by setting environment variables matching the configuration syntax PREFECT__CONTEXT__SECRETS__{SECRETNAME}={SECRETVALUE}. (Learn more about configuration in our Configuration concept docs.)

For example, given an environment with the environment variable PREFECT__CONTEXT__SECRETS__FOO=mypassword, the value "mypassword" could be retrieved by using a PrefectSecret task or by using the secrets API directly, as shown below.

# Environment Variables

The EnvVarSecret task class reads secret values directly from environment variables.

from prefect import task, Flow
from prefect.tasks.secrets import EnvVarSecret

@task
def print_value(x):
    print(x)

with Flow("Example") as flow:
    secret = EnvVarSecret("PATH")
    print_value(secret)

flow.run() # prints the value of the "PATH" environment variable

# Prefect Cloud

For Cloud users, if the secret is not found in local context and config.cloud.use_local_secrets=False, the base Secret class queries the Prefect Cloud API for a stored secret. This call can only be made successfully by authenticated Prefect Cloud Agents.

# Default Secrets

A few common secrets, such as authentication keys for GCP or AWS, have a standard naming convention as Prefect secrets for use by the Prefect pipeline or tasks in Prefect's task library. If you follow this naming convention when storing your secrets in local context or environment variables, all supported Prefect interactions with those services will be automatically configured.

The following is a list of the default names and contents of Prefect Secrets that, if set and declared, can be used to automatically authenticate your flow with the listed service:

  • GCP_CREDENTIALS: a dictionary containing a valid Service Account Key
  • AWS_CREDENTIALS: a dictionary containing two required keys: ACCESS_KEY and SECRET_ACCESS_KEY, and an optional SESSION_TOKEN, which are passed directly to the boto3 client
  • GITHUB_ACCESS_TOKEN: a string value of a GitHub personal access token, requires repo scope

For example, when using local secrets, your Prefect installation can be configured to authenticate to AWS automatically by adding that specific AWS_CREDENTIALS key value pair into your secrets context like so:

export PREFECT__CONTEXT__SECRETS__AWS_CREDENTIALS='{"ACCESS_KEY": "abcdef", "SECRET_ACCESS_KEY": "ghijklmn"}'

Then all Prefect usages of AWS credentials will default to using the values in that dictionary.

← Configuration →