spack: Can't install tensorflow because of bazel
I am trying to install tensor flow for cpu via spack. It falls over at bazel:
==> Installing bazel
==> Searching for binary cache of bazel
==> No binary for bazel found: installing from source
==> Using cached archive: /lus/theta-fs0/projects/datascience/cadams/spack/cooley/spack/var/spack/cache/_source-cache/archive/c0/c0e94f8f818759f3f67af798c38683520c540f469cb41aea8f5e5a0e43f11600.zip
==> Staging archive: /tmp/cadams/spack-stage/spack-stage-bazel-0.26.1-4mmpkcidyja4i6smxm2ahqifvh322xsz/bazel-0.26.1-dist.zip
==> Created stage in /tmp/cadams/spack-stage/spack-stage-bazel-0.26.1-4mmpkcidyja4i6smxm2ahqifvh322xsz
==> Applied patch /lus/theta-fs0/projects/datascience/cadams/spack/cooley/spack/var/spack/repos/builtin/packages/bazel/bazelruleclassprovider-0.25.patch
==> Applied patch /lus/theta-fs0/projects/datascience/cadams/spack/cooley/spack/var/spack/repos/builtin/packages/bazel/unix_cc_configure-0.15.patch
==> Applied patch /lus/theta-fs0/projects/datascience/cadams/spack/cooley/spack/var/spack/repos/builtin/packages/bazel/compile-0.21.patch
==> Building bazel [Package]
==> Executing phase: 'bootstrap'
==> Executing phase: 'install'
==> Error: FileNotFoundError: [Errno 2] No such file or directory: 'output/bazel'
/lus/theta-fs0/projects/datascience/cadams/spack/cooley/spack/var/spack/repos/builtin/packages/bazel/package.py:136, in install:
134 def install(self, spec, prefix):
135 mkdir(prefix.bin)
>> 136 install('output/bazel', prefix.bin)
See build log for details:
/tmp/cadams/spack-stage/spack-stage-bazel-0.26.1-4mmpkcidyja4i6smxm2ahqifvh322xsz/spack-build-out.txt
Here was the command I have used:
$ spack install py-tensorflow@1.15 cuda_arch=35 ^cuda@10.1 %gcc@6.3.0
And, I am on this system:
$ uname -a
Linux cc047 3.10.0-957.27.2.el7.x86_64 #1 SMP Tue Jul 9 16:53:14 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux
$ lsb_release
LSB Version: :core-4.1-amd64:core-4.1-noarch
Any ideas for how to fix this? I have the latest version of spack as of this morning. Is there an older version of bazel that will work?
About this issue
- Original URL
- State: closed
- Created 7 years ago
- Comments: 15 (15 by maintainers)
Commits related to this issue
- compiler wrappers: fix out-of-order injection scenario Spack attempts to inject it's paths just before the actual system include directories. Currently, if a build utilizes -isystem, Spack's headers ... — committed to aweits/spack by aweits 7 years ago
- compiler wrappers: fix out-of-order injection scenario Spack attempts to inject it's paths just before the actual system include directories. Currently, if a build utilizes -isystem, Spack's headers ... — committed to aweits/spack by aweits 7 years ago
- compiler wrappers: fix out-of-order injection scenario Spack attempts to inject it's paths just before the actual system include directories. Currently, if a build utilizes -isystem, Spack's headers ... — committed to aweits/spack by aweits 7 years ago
Maybe we should add a note/warning that says the
bazelandpy-tensorflowpackages are only tested on macOS/Ubuntu and may not build on RHEL and friends? I also had trouble building it on Amazon Linux (https://github.com/bazelbuild/bazel/issues/10418) which is RHEL-based.Ugh, this is most definitely another instance of https://github.com/bazelbuild/bazel/issues/10437 as I see the error is caused by the same include error for nasm where bazel tries to include the wrong header. I’m still waiting for a reply on that issue but I hope if enough people try to tell the bazel team that this is a serious impediment to building tf on rhel/centos/sles/cle (just any os apart from ubuntu) they would fix it.
For what it’s worth, java moved the build past bazel as @s-sajid-ali suggested via a private email:
spack install py-tensorflow@1.15 cuda_arch=35 ^cuda@10.1 ^jdk@1.8.0_231-b11 %gcc@6.3.0The java specification had to be exactly the one listed or it still failed.
I’ll follow up if there are other crashes in tensorflow.