Closed petronny closed 1 year ago
If you are troubleshooting why your build failed, I note that your error log for Run arch4edu/cactus/actions/build-package@main
ends with:
error: could not write to 'build/lib.linux-x86_64-cpython-311/tensorflow/python/_pywrap_tfcompile.so': No space left on device
==> ERROR: A failure occurred in build().
Aborting...
Attempted to remove disk file system under "/run/systemd/nspawn/propagate/makepkg", and we can't allow that.
==> ERROR: Build failed, check /mnt/extra-x86_64/makepkg/build
So you're running out of space somewhere. The log spam is only like a megabyte; I don't think that's your problem. Maybe your /tmp tmpfs is too small. I got a similar error building my PKGBUILD at that same point when I only had 8G for /tmp:
...
On the other hand, currently there is no concept of package data
directory, all directories are treated like packages.
********************************************************************************
!!
check.warn(importable)
error: [Errno 28] No space left on device
To see how much space you have, run df -h /tmp
. 12G is enough for me to build rocm tensorflow for 7 architectures; if your /tmp smaller, run mount -o remount,size=12G /tmp
To resolve the warnings, try the following command:
sed -i -E "s/find_packages/find_namespace_packages/" tensorflow-${_pkgver}/tensorflow/tools/pip_package/setup.py
You can add it to the prepare() function in the PKGBUILD or call it from src.
I have confirmed that find_namespace_packages()
is a superset of find_packages()
and of the packages listed in the log-spam. I have rebuilt my version of the project and my test script runs ok. The package produced with this change is 267 bytes smaller, which is close enough. I have not run the full suite of tensorflow tests. I have not attempted a clean build (rm ~/.cache/bazel
).
https://github.com/arch4edu/cactus/actions/runs/5112035031/jobs/9239440792
This warning shows almost for every modules in tensorflow.