tensorflow / serving

A flexible, high-performance serving system for machine learning models

Home Page:https://www.tensorflow.org/serving

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Logging - json format / custom format

GautamSinghania opened this issue · comments

Feature Request

Describe the problem the feature is intended to solve

My organization allows logs to be persisted only if they follow certain standards. As the current Tensorflow logs cannot be formatted, we do not have a way of persisting these logs.

Describe the solution

I want to control the format of my logs so that I can follow the organization standards. Either one of the following will solve my usecase:

  • Convert the log to json format
  • Customize the log to follow a certain format

As an added feature, it would be really helpful if I can access request specific elements like tracers in these logs.

Describe alternatives you've considered

If Python logging was integrated in Tensorflow serving, we could follow that to control the logs.

Additional context

Version: 2.8.2-gpu

For understanding, where are logs from tensorflow routed to? I assume error logs are routed to STDERR and others are routed to STDOUT, but please let me know if this assumption is wrong.