Upgrade to manylinux2014 to support AArch64 wheels

[odidev]

Problem

On aarch64, pip install ml-metadata builds the wheels from source code and then install it which takes some time to build the wheels than downloading and extracting the wheels from pypi.

Resolution

For building ml-metadata wheels for x86_64, cr.io/tfx-oss-public/manylinux2010-bazel image is getting used. As mentioned in comment, we need similar image for aarch64.

Please let me know your interest on adding tfx image support for aarch64. I can help with this.

[dheaney]

Definitely interested in getting wheels for aarch64. Building manually from source (python3.8 setup.py bdist_wheel) on aarch64 succeeds, but the import causes Python to crash:

$ python3.8 setup.py bdist_wheel
# ...
$ python3.8 -m pip install dist/ml_metadata-1.1.0.dev0-cp38-cp38-linux_aarch64.whl
$ python3.8
Python 3.8.0 (default, Feb 25 2021, 22:10:10) 
[GCC 8.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import ml_metadata
free(): invalid pointer
Aborted

My build log is here. Do you know if this fix would help? I'm guessing that it's an issue with TensorFlow or TCMalloc.

[dheaney]

Came here from a link on an issue in the the ml-metadata repo. Sorry if it's the wrong place.

[hughmiao]

@dheaney hey, DJ, recently we dropped tf dependency in the latest release. please try the manual build in aarch64 env again and see the error goes away.

[arghyaganguly]

@odidev , seems like this is fixed. Please reopen if required.

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No