feat: logging customization

[jiewpeng]

What does this PR address?

Adds customisation for logging, for example to change the format of the opentelemetry trace id and span id. Thread: https://bentoml.slack.com/archives/CKRANBHPH/p1661766094533779

Before submitting:

• Does the Pull Request follow Conventional Commits specification naming? Here are GitHub's
  guide on how to create a pull request.
• Does the code follow BentoML's code style, both make format and make lint script have passed (instructions)?
• Did you read through contribution guidelines and follow development guidelines?
• Did your changes require updates to the documentation? Have you updated
  those accordingly? Here are documentation guidelines and tips on writting docs.
• Did you write tests to cover your changes? No. I am unsure how to get the logs from the host fixture to test the configuration of logging. Have manually tested using bentoml serve to check that the configuration works.

Who can help review?

Feel free to tag members/contributors who can help review your PR.

[codecov]

Codecov Report

Merging #2961 (021495c) into main (2d03474) will increase coverage by 0.03%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main    #2961      +/-   ##
==========================================
+ Coverage   70.60%   70.63%   +0.03%     
==========================================
  Files         104      104              
  Lines        9539     9550      +11     
==========================================
+ Hits         6735     6746      +11     
  Misses       2804     2804              

Impacted Files	Coverage Δ
bentoml/_internal/configuration/containers.py	78.89% <100.00%> (+0.39%)	⬆️
bentoml/_internal/log.py	87.50% <100.00%> (+1.53%)	⬆️

[jiewpeng]

I'm open to discussion on the following

• Default trace_id and span_id format: currently I am proposing to change the default format to hexadecimal, which will be different from bentoml's current behaviour (logging as integer).
• Exposing the logging formatting for the user to configure. I think this is useful for organizations that want to standardise their logging formats across their apps.

[aarnphm]

LGTM, with some minor comments.

[aarnphm]

I prefer to make this logging_formatting to be consistent with the config level.

[jiewpeng]

I could change it, but would there be a use case for someone to have one trace id format for the api server and a different format for the runner? When I requested for this feature, the premise was that I would want the trace id format to be consistent across my applications, so I don't really see a use case where a user would want different formats.

[aarnphm]

Oh I was saying just to change the naming to make it consistent

[jiewpeng]

Have done the import lazily, but I'm just curious, is there a reason to do so. given it will run at some point anyway, provided the model is called? Based on my understanding, this delays the import of the BentoMLContainer for as long as possible, until the model gets invoked. Either this import is slow, in which case doing the import eagerly slows down the first request to the model, or the import is fast, and there is no difference (or it makes logging slightly slower because we always run the 3 lines to get the trace/span id format every time we log something, when those things do not change).

[aarnphm]

Have done the import lazily, but I'm just curious, is there a reason to do so.

This has to do with importing bentoml to global namespace is actually pretty inefficient. We will fix this some day 😄

logging_formatting is a singletonFactory, meaning it will be cached once, so actually the access for the field are pretty fast, so I'm not too worried about the performance deficit if we do it lazily.

[aarnphm]

can you rebase to the main branch, I saw there were some bad commits there. Otherwise this is GTG

[sauyon]

Looks good, thank you!