Extend configurability of the built-in logger to account for runtime/model names

[jklaise]

Using the built-in logger is convenient because it's auto-configured:

MLServer/mlserver/logging.py

Line 23 in 7437787

def configure_logger(settings: Settings = None):

However, it falls short when trying to debug messages coming from multiple different models - all of these will be assigned the mlserver name so it's difficult to tell which message came from which component.. A workaround is to roll your own logging in each component, but this adds boilerplate for each of the components that require logging (unless the custom logging configuration is split out into some utility library, but then that brings in other complications).

[xvnyv]

An option to add the model name to the mlserver logs will be useful for my project's use case as well. It gets quite difficult to differentiate the logs from different models, especially as the number of models loaded into the same mlserver instance grows.

May I know if there are any plans to add such a feature in the near future?

[adriangonz]

Hey @xvnyv ,

Thanks for that feedback.

It's on the roadmap, but not prioritised for the next version. It could come in 1.5.0 though!