If this is a bug report, please fill out the following form in full:
According to the GPU-Dockerfile, it needs to add --config cuda to the bazel build command to enable the GPU support.
However, according to the bazel releases page,
--[no]allow_undefined_configs no longer exists, passing undefined configs is an error.
as of version 0.16.0 and above.
So I'm getting
ERROR: Config value cuda is not defined in any .rc file
when building on Archlinux which is providing bazel 0.19.0.
I'm building it with the script on AUR:
$ wget https://aur.archlinux.org/cgit/aur.git/plain/PKGBUILD?h=tensorflow-model-server-cuda -O PKGBUILD
$ makepkg
I've pasted the log in the next session.
==> Extracting sources...
-> Extracting 1.11.0.tar.gz with bsdtar
==> Starting prepare()...
==> Starting build()...
WARNING: The following rc files are no longer being read, please transfer their contents or import their path into one of the standard rc files:
/build/tensorflow-model-server-cuda/src/serving-1.11.0/tools/bazel.rc
Extracting Bazel installation...
Starting local Bazel server and connecting to it...
INFO: Options provided by the client:
Inherited 'common' options: --isatty=0 --terminal_columns=80
ERROR: Config value cuda is not defined in any .rc file
==> ERROR: A failure occurred in build().
Aborting...
Seems you miss export TF_NEED_CUDA=1 env variable
export TF_NEED_CUDA=1
is in prepare().
Try using bazel 0.15, which is the currently tested version of bazel used.
Closing, feel free to reopen if problem persists.
Please reopen it. Downgrading bazel only delays the problem.
Please reopen it as the problem is not solved.
Reopened the issue as the problem still exist.
@gautamvasudevan PTAL
This is for an old version of TF serving on an unsupported version of bazel on Archlinux, a configuration which is not supported. I鈥檒l leave this open if someone in the community can help.
Closing as this is not a configuration supported by the project.
Most helpful comment
Please reopen it. Downgrading bazel only delays the problem.