tensorflow: TF_CPP_MIN_LOG_LEVEL does not work with TF2.0 dev20190820

After a quick dig into this, it seems that the problem appear only if you change the log level after import tensorflow.

import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = "2"
import tensorflow as tf
import numpy as np
a = tf.Variable(np.array([0, 1, 2]))
print(a)

output:

<tf.Variable 'Variable:0' shape=(3,) dtype=int32, numpy=array([0, 1, 2])>

if i change set the environ after import tf:

import os
import tensorflow as tf
os.environ['TF_CPP_MIN_LOG_LEVEL'] = "2"
import numpy as np
a = tf.Variable(np.array([0, 1, 2]))
print(a)

output:

...
2019-08-26 11:45:17.907378: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_100.dll
2019-08-26 11:45:20.776347: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library nvcuda.dll
2019-08-26 11:45:20.885617: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1618] Found device 0 with properties: 
name: GeForce GTX 1080 Ti major: 6 minor: 1 memoryClockRate(GHz): 1.6575
pciBusID: 0000:01:00.0
2019-08-26 11:45:20.886163: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1618] Found device 1 with properties: 
name: GeForce GTX 1080 Ti major: 6 minor: 1 memoryClockRate(GHz): 1.6575
pciBusID: 0000:02:00.0
2019-08-26 11:45:20.886391: I tensorflow/stream_executor/platform/default/dlopen_checker_stub.cc:25] GPU libraries are statically linked, skip dlopen check.
2019-08-26 11:45:20.887824: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1746] Adding visible gpu devices: 0, 1
2019-08-26 11:45:20.888138: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports 
...

This happens in Tensorflow 2.0rc, however Tensorflow beta0&beta1 works good no matter the order of import.

Maybe in the latest code change the logger reading the TF_CPP_MIN_LOG_LEVEL once when init, and in the older version, the logger was reading TF_CPP_MIN_LOG_LEVEL on the fly. Could you please confirm if this is intend to do or a bug introduced in the latest version? @gadagashwini

I am not sure this is related. I am using 2.0.0rc0, but it seems the same issue is in 2.0.0.a0. My code is

import tensorflow as tf

tf.get_logger().setLevel('ERROR')
print("Hi!")
output = tf.keras.layers.Conv1D(1, 1)(tf.keras.layers.Input([1, 1]))

(I found this code in #28577, #28578).

and the output includes:

Hi!
2019-08-28 10:57:14.647006: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2

If that is not the same issue, I can file it separately.

@bersbersbers I think its not the same issue, tf.get_logger().setLevel(‘ERROR’) means set your python tensorflow log level to “ERROR”, and the evn value “TF_CPP_MIN_LOG_LEVEL” is to set tensorflow c++ log level. So in my code, i usually set them both like this:

 if args.debug is True:
        os.environ['TF_CPP_MIN_LOG_LEVEL'] = "0"
        tf.get_logger().setLevel("INFO")
else:
        os.environ['TF_CPP_MIN_LOG_LEVEL'] = "2"
        tf.get_logger().setLevel("ERROR")

Hi. Is the issue solved? If yes, could someone tell me how to avoid INFO and WARNING messages in output?

@akhil189 It seems not fixed yet. Temporary workaround is that call os.environ["TF_CPP_MIN_LOG_LEVEL"] = "2" before import tensorflow as tf, or manually set environment variable TF_CPP_MIN_LOG_LEVEL=2 before running your python script.

> SET TF_CPP_MIN_LOG_LEVEL=2
> python your_script.py

@KichangKim Thank you for providing the workaround. But, I am using google colab and it’s not working in that. The ‘INFO’ and ‘WARNING’ messages are still being shown in the output.

This issue seem to be still open, but with workaround mentioned by @KichangKim works fine. Version : 2.3.0

Sample for verification

$ python36
Python 3.6.2 (v3.6.2:5fd33b5, Jul  8 2017, 04:57:36) [MSC v.1900 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow as tf
2020-09-08 04:26:59.370871: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'cudart64_101.dll'; dlerror: cudart64_101.dll not found
2020-09-08 04:26:59.379059: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.

$ python36
Python 3.6.2 (v3.6.2:5fd33b5, Jul  8 2017, 04:57:36) [MSC v.1900 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import os
>>> os.environ['TF_CPP_MIN_LOG_LEVEL'] = '1'
>>> import tensorflow as tf
2020-09-08 04:27:33.738961: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'cudart64_101.dll'; dlerror: cudart64_101.dll not found

I know this is not an ideal env. I am using cygwin, on top of that I am calling a python binary installed for windows. Anyways, I think it does prove that the workaround for logging works.

@bersbersbers Oh, I misread ziyigogogo’s comment. It seems that not yet concluded. Reopen.

This is fixed in tf-2.0 nightly build version 2.0.0-dev20190826 in the sense we see:

import os
import tensorflow as tf
os.environ['TF_CPP_MIN_LOG_LEVEL'] = "2"
import numpy as np
a = tf.Variable(np.array([0, 1, 2]))
print(a)

output:

<tf.Variable 'Variable:0' shape=(3,) dtype=int64, numpy=array([0, 1, 2])>