Missing platform tag in wheel name

[daskol]

TensorBoard from 2.12 up to now (2.14) does not start on Ubuntu 20.04 due to libc version mismatch (actual 2.32 vs expected 2.35). This should be explicitly specified in platform tag of a wheel in PyPI. I would be great at least to add this info in release notes explicitely.

TensorFlow installation not found - running with reduced feature set.
/usr/local/lib/python3.8/dist-packages/tensorboard_data_server/bin/server: /lib/x86_64-linux-gnu/libc.so.6: version `GLIBC_2.33' not found (required by /usr/local/lib/python3.8/dist-packages/tensorboard_data_server/bin/server)
/usr/local/lib/python3.8/dist-packages/tensorboard_data_server/bin/server: /lib/x86_64-linux-gnu/libc.so.6: version `GLIBC_2.34' not found (required by /usr/local/lib/python3.8/dist-packages/tensorboard_data_server/bin/server)
/usr/local/lib/python3.8/dist-packages/tensorboard_data_server/bin/server: /lib/x86_64-linux-gnu/libc.so.6: version `GLIBC_2.32' not found (required by /usr/local/lib/python3.8/dist-packages/tensorboard_data_server/bin/server)

[groszewn]

I believe this is related to #6578. The produced tensorboard-data-server wheels do specify manylinux2014, but the tensorboard wheels do not.

[daskol]

Yes, indeed but the version of glibc should be indicated explicitly as a platform tag in wheel according to Python's PEPs.

[groszewn]

The tensorboard (and tb-nightly) wheels don't require glibc, an underlying dependency on a separate wheel (i.e. tensorboard-data-server) is what requires glibc. Tensorboard itself should still run, just without the functionality in the separate tensorboard-data-server wheel (i.e. it falls back to the native python implementation).

#6578 tracks that the tensorboard-data-server wheel is bundled with a too-new version of glibc given its platform tag, which we're working on rectifying.

[groszewn]

I just released tensorboard-data-server==0.7.2 which should resolve this (i.e. using the manylinux_2_31_x86_64).