CentOS 7: libstdc++.so.6: version `CXXABI_1.3.9' not found

Question

I am running a python script using PyCharm on CentOS 7. The script imports tensorflow and allocates some potion of GPU memory to the script.

The script worked fine without any issues until yesterday. I am not sure why this happened. I am running the following versions of gcc and libstdc++ on CentOS

>> rpm -qf /usr/lib/libstdc++.so.6
libstdc++-4.8.5-16.el7_4.1.i686

>> echo $LD_LIBRARY_PATH
/usr/local/lib64/:/usr/local/cuda-8.0/lib64:/usr/lib64/mpich/lib:/usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64

>> echo $LD_PRELOAD

blank output

>> hostnamectl

  Operating System: CentOS Linux 7 (Core)
       CPE OS Name: cpe:/o:centos:centos:7
            Kernel: Linux 3.10.0-693.2.2.el7.x86_64
      Architecture: x86-64

I have libgcc 7.2.0 version in my anaconda environment.

Some of the stackflow answers addressing this issue is related to Ubuntu system and not CentOS

I received the following error

from google.protobuf import descriptor as _descriptor
  File "/home/user_name/anaconda/envs/tokyo/lib/python3.6/site-packages/google/protobuf/descriptor.py", line 46, in <module>
    from google.protobuf.pyext import _message
ImportError: /lib64/libstdc++.so.6: version `CXXABI_1.3.9' not found (required by /home/rmulpuri/anaconda/envs/tokyo/lib/python3.6/site-packages/google/protobuf/pyext/_message.cpython-36m-x86_64-linux-gnu.so)

Solution

Facing a similar problem with zmq on CentOS 7 I come up with a workaround since a fresh installation and environment did not help.

The original issue was:

(mtango-py)$ python -c "import zmq"
  ...
  ImportError: /usr/lib64/libstdc++.so.6: version `GLIBCXX_3.4.20' not found (required by /home/me/opt/miniconda2/envs/mtango-py/lib/python3.5/site-packages/zmq/backend/cython/../../../../../libzmq.so.5.1.2)

By loading the library before anything with LD_PRELOAD:

(mtango-py)$ export LD_PRELOAD=$CONDA_PREFIX/lib/libstdc++.so
(mtango-py)$ python -c "import zmq"                          
(mtango-py)$

This answer does not resolve the root problem tough. Conda should load the environment library if the system is not compatible.

Comments

[vhardion]

I finally found that some packages have been installed with sudo pip which broke the system modules with probably some incompatible wheels.