Common constants, validations and misc utility functions that are called by both the openai and langchain lambdas.
This shared class manages configuration settings for all uses cases, including: local development, test, QA, CI-CD, and production.
The Settings class leverages Pydantic to enforce strong data typing as well as to assist with common-sense validations of several of the configuration values. Additionally, it enables us to enforce a read-only state of the settings attributes in able to thwart potential side effects from erroneous CD-CD processes.
Sources of configuration data vary depending on the use case, which can invariably introduce ambiguities with regard to which source takes priority in varying scenarios. The Settings class resolves these potential ambiguities by consistently applying the following order of precedence by source:
- Python constructor arguments passed from inside the source code
.envfiles which can be saved anywhere inside the repo, and nothing prevents multiple .env files from being created.- environment variables initialized in the form of
export VARIABLE_NAME=some-value terraform.tfvarsif it exists- default values declared in SettingsDefaults class
Where possible, we use Pydantic to seamlessly resolve some of these ambiguities.
The Settings class will not expose sensitive data under any circumstances, including logging and dumps.
The Settings class contains instance variables for aws_profile and aws_region. If either of these is unset then standard behavior of AWS CLI will take precedence.
The Settings class contains an instance variable for openai_api_key which will take precedence if it is set.
The Settings class contains an instance variable for pinecone_api_key which will take precedence if it is set.
The Settings class is also a provider to CloudWatch when both DEBUG_MODE and DUMP_DEFAULTS environment variables are set to True. The Settings property dump generates a context sensitive JSON dict of the state data for all settings as well as class instance meta data that can be helpful during development trouble shooting.
An example CloudWatch Dump