Customizing logging

 

Hydra is configuring Python standard logging library with the dictConfig method. You can learn more about it here. There are two logging configurations, one for Hydra itself and one for the executed jobs.

This example demonstrates how to customize the logging behavior of your Hydra app, by making the following changes to the default logging behavior:

• Outputs only to stdout (no log file)
• Output a simpler log line pattern

config.yaml

defaults:
  - override hydra/job_logging: custom

hydra/job_logging/custom.yaml

version: 1
formatters:
  simple:
    format: '[%(levelname)s] - %(message)s'
handlers:
  console:
    class: logging.StreamHandler
    formatter: simple
    stream: ext://sys.stdout
root:
  handlers: [console]

disable_existing_loggers: false

Security considerations

When Hydra configures Python logging, the configuration can import and call the values of handler class keys and formatter, filter, and handler () keys. Configure an execution whitelist in trusted Python code whenever Hydra's logging configuration may come from an untrusted source:

import hydra

@hydra.main(
    version_base=None,
    config_path="conf",
    config_name="config",
    execution_whitelist=("my_app.logging.CustomHandler",),
)
def my_app(cfg):
    ...

Hydra automatically permits the exact targets used by its built-in logging configurations. Add any application-defined handler, formatter, filter, queue, or listener targets to the whitelist. The whitelist must come from trusted Python code; do not derive it from composed configuration.

Hydra does not support replacing Python's global logging.config.dictConfigClass. A custom configurator would bypass Hydra's target authorization. Express custom logging components in the logging configuration and authorize their targets instead.

See Execution whitelist for the shared rules for logging and instantiate().

Resolving additional logging targets without an execution whitelist retains the legacy blacklist behavior in Hydra 1.4, but is deprecated and will become an error in Hydra 1.5. Use UNSAFE_DISABLE_EXECUTION_CHECKS only when all composed logging configuration is trusted and unrestricted target resolution is intentional.

This is what the default logging looks like:

$ python my_app.py hydra/job_logging=default
[2020-08-24 13:43:26,761][__main__][INFO] - Info level message

And this is what the custom logging looks like:

$ python my_app.py 
[INFO] - Info level message