Forking a processes might lead to a faling scorep

[AndreasGocht]

follofing code leads to an CubeW error message:

import tensorflow as tf

A = tf.placeholder(tf.float64)
B = tf.placeholder(tf.float64)
C = tf.matmul(A,B)

A_in = [[1, 2], [1, 2]]
B_in = [[1, 2], [1, 2]]

with tf.Session() as sess:
    res = sess.run(C,feed_dict={A:A_in, B:B_in})

print(res)

Error is:

[[3. 6.]
 [3. 6.]]
[Score-P] src/measurement/profiling/scorep_profile_collapse.c:82: Warning: Score-P callpath depth limitation of 30 exceeded.
Reached callpath depth was 156.
Consider setting SCOREP_PROFILING_MAX_CALLPATH_DEPTH to 156.

[CubeW] src/cubew/cubew_tar_writing.c:325: warning: Cannot open tared cube file /home/gocht/Dokumente/TUD Intern/2019-09-03 Toolsworkshop/experimente/scorep-measurement-tmp/profile.cubex. 

The following error occurred: No such file or directory
[CubeW] src/cubew/cubew_tar_writing.c:327: warning: Return NULL.

Segmentation fault (core dumped)

[AndreasGocht]

This is due to TensorFlow forking uname and sh. It might help to remove the LD_Preload after a successful init again.