Installing a newer version of glibc RHEL 7.6

Latest response

I m trying to set up an enviroment for developing neural network models on a rhel 7.6 server with GPUs. I wanna use the tensorflow library for this purpouse and i have read that one of the dependencies of the tensorflow library is having least the version 2.23 of glibc. Rhel have 2.17 so, upgrading it may cause issues. Is it possible to upgrade the library to 2.23 in any way?
I know that i could get an older version of tensorflow but the next version available is too old for me.
Thank you

Responses

You definitely don't want to replace the system glibc, that would make the rest of the system unsupportable and probably quite unstable.

A quick Google search for "tensorflow rhel7 glibc" turns up several results of people using a later glibc on earlier Ubuntu and RHEL6, where the new glibc is decompressed to an alternative directory. Perhaps you could come up with similar steps for RHEL7?

I found this GitHub Issue where someone installs glibc 2.23 into /opt and runs python with a library path override, which seems to work for them:

You may want to have a good backup before trying this, just in case the make scripts do something nasty.

This would also probably be a good use case for containers, so you can run an entire new userspace (like EL8 or Ubu 18.04) inside the container, and run Tensorflow inside there.

Tensorflow themselves supply an official Docker image as described at:

As their Docker Hub entry says, that image is based on Ubu 18.04 which has glibc 2.27.