tensorflow: Error in `python`: free(): invalid pointer

TensorFlow Version: 1.0 and 1.0.1 (whl package) Kubernetes Version: 1.5.1 Docker Version: 1.12.5 Container OS: ubuntu 14.04

I run tensorflow applications, say mnist_replica.py, in the kubernetes cluster and this error occured everytime when finishing the whole train step.

I attached the log of mnist_replica.py as below:

1492616963.574406: Worker 0: training step 5045 done (global step: 9999)
1492616963.583257: Worker 0: training step 5046 done (global step: 10001)
Training ends @ 1492616963.583282
Training elapsed time: 54.435048 s
After 10000 training step(s), validation cross entropy = 1170.07
*** Error in `/usr/bin/python': free(): invalid pointer: 0x0000000001e16c50 ***

And everything goes well when I run the mnist_replica.py in a physical machine with RHEL 7.0. I Guess the problem is that your google’s environment is different with me. So the whl package that you compiled can not run well in the ubuntu 14.04 container, especially when it comes to the C program calling Python program

About this issue

Original URL
State: closed
Created 7 years ago
Comments: 17 (16 by maintainers)

Most upvoted comments

I encountered this problem because I used torch and tensorflow at same time.

Fangyh09 on Apr 10, 2019

It’s not only an unique case, but also happened in other ubuntu 14.04 in physical machine with TensorFlow v1.0, but it has different error log:

Error in `python': double free or corruption (!prev): 0x00000000016b43c0

DjangoPeng on Apr 22, 2017