Cannot build deepvariant with tensorflow master due to bazel version requirements

With DV_CPP_TENSORFLOW_TAG=master, I believe deepvariant will not build from source. TF master does not build with Bazel=0.15.0:

+ bazel test -c opt --local_test_jobs=1 --copt=-march=corei7 --copt=-Wno-sign-compare --copt=-Wno-write-strings deepvariant/...
[bazel INFO src/main/cpp/option_processor.cc:235] Looking for master bazelrcs in the following three paths: /root/deepvariant/tools/bazel.rc, , /etc/bazel.bazelrc
[bazel INFO src/main/cpp/option_processor.cc:165] User provided no rc file.
[bazel INFO src/main/cpp/rc_file.cc:53] Parsing the RcFile /root/deepvariant/tools/bazel.rc
[bazel INFO src/main/cpp/rc_file.cc:53] Parsing the RcFile /root/deepvariant/../tensorflow/tools/bazel.rc
[bazel FATAL src/main/cpp/blaze.cc:1263] Unexpected error reading .blazerc file '/root/deepvariant/../tensorflow/tools/bazel.rc'

And deepvariant does not build with bazel=0.19.0, see #134.

---

Linux xxx 4.15.0-1031-aws #33-Ubuntu SMP Fri Dec 7 09:32:27 UTC 2018 x86_64 x86_64 x86_64 GNU/Linux

# Copyright 2017 Google LLC.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
#
# 1. Redistributions of source code must retain the above copyright notice,
#    this list of conditions and the following disclaimer.
#
# 2. Redistributions in binary form must reproduce the above copyright
#    notice, this list of conditions and the following disclaimer in the
#    documentation and/or other materials provided with the distribution.
#
# 3. Neither the name of the copyright holder nor the names of its
#    contributors may be used to endorse or promote products derived from this
#    software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
# ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
# LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
# CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
# SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
# INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
# CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
# ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
# POSSIBILITY OF SUCH DAMAGE.

# Source this file---these options are needed for TF config and for
# successive bazel runs.

# Set this to 1 if the system image already has TensorFlow preinstalled.  This
# will skip the installation of TensorFlow.
export DV_USE_PREINSTALLED_TF="${DV_USE_PREINSTALLED_TF:-0}"

export TF_CUDA_CLANG=0
export TF_ENABLE_XLA=1
export TF_NEED_CUDA=1
export TF_NEED_GCP=1
export TF_NEED_GDR=0
export TF_NEED_HDFS=0
export TF_NEED_JEMALLOC=0
export TF_NEED_MKL=1
export TF_NEED_MPI=0
export TF_NEED_OPENCL=0
export TF_NEED_OPENCL_SYCL=0
export TF_NEED_S3=1
export TF_NEED_VERBS=0

# Used if TF_NEED_CUDA=1
export TF_CUDA_VERSION="10.0"
export CUDA_TOOLKIT_PATH="/usr/local/cuda"
export TF_CUDNN_VERSION="7"
export CUDNN_INSTALL_PATH="/usr/lib/x86_64-linux-gnu"

# The version of bazel we want to build DeepVariant.
DV_BAZEL_VERSION="0.15.0"

# We need to make sure that $HOME/bin is first in the binary search path so that
# `bazel` will find the latest version of bazel installed in the user's home
# directory. This is set in setting.sh as all DeepVariant scripts source
# settings.sh and assume that `bazel` will find the right version.
export PATH="$HOME/bin:$PATH"

# Path to the public bucket containing DeepVariant-related artifacts.
export DEEPVARIANT_BUCKET="gs://deepvariant"
export DV_PACKAGE_BUCKET_PATH="${DEEPVARIANT_BUCKET}/packages"
export DV_PACKAGE_CURL_PATH="https://storage.googleapis.com/deepvariant/packages"

# Set this to 1 to use the nightly (latest) build of TensorFlow instead of a
# named release version. Set it to an already existing value in the environment
# (allowing command line control of the build), defaulting to 0 (release build).
# Note that setting this to 1 implies that the C++ code in DeepVariant will be
# build using the master branch and not the pinned version to avoid
# incompatibilities between TensorFlow C++ used to build DeepVariant and the
# tf-nightly wheel.
export DV_TF_NIGHTLY_BUILD="${DV_TF_NIGHTLY_BUILD:-1}"

# The branch/tag we checkout to build our C++ dependencies against. This is not
# the same as the python version of TensorFlow we use, but should be similar or
# we risk having version incompatibilities between our C++ code and the Python
# code we use at runtime.
if [[ "${DV_TF_NIGHTLY_BUILD}" = "1" ]]; then
  export DV_CPP_TENSORFLOW_TAG="master"
else
  export DV_CPP_TENSORFLOW_TAG="r1.12"
fi
export DV_GCP_OPTIMIZED_TF_WHL_VERSION="1.12.0"
export DV_TENSORFLOW_STANDARD_GPU_WHL_VERSION="1.12.0"
export DV_TENSORFLOW_STANDARD_CPU_WHL_VERSION="1.12.0"

# Set this to 1 to use DeepVariant with GPUs. Set it to an already existing
# value in the environment (allowing command line control of the build),
# defaulting to 0 (CPU only build).
export DV_GPU_BUILD="${DV_GPU_BUILD:-1}"

# If this variable is set to 1, DeepVariant will use a TensorFlow wheel file
# compiled with MKL support for corei7 or better chipsets, which
# significantly speeds up execution when running on modern CPUs. The default
# TensorFlow wheel files don't contain these instructions (and thereby run on a
# broader set of CPUs). Using this optimized wheel reduces the runtime of
# DeepVariant's call_variants step by >3x. This is called the GCP (Google Cloud
# Platform) optimized wheel because all GCP instances have at least Sandy Bridge
# or better chipsets, so this wheel should run anywhere on GCP.
export DV_USE_GCP_OPTIMIZED_TF_WHL="${DV_USE_GCP_OPTIMIZED_TF_WHL:-1}"
export GCP_OPTIMIZED_TF_WHL_FILENAME="tensorflow-${DV_GCP_OPTIMIZED_TF_WHL_VERSION}.deepvariant_gcp-cp27-none-linux_x86_64.whl"
export GCP_OPTIMIZED_TF_WHL_PATH="${DV_PACKAGE_BUCKET_PATH}/tensorflow"
export GCP_OPTIMIZED_TF_WHL_CURL_PATH="${DV_PACKAGE_CURL_PATH}/tensorflow"

# Set this to 1 to make our prereq scripts install the CUDA libraries.
# If you already have CUDA installed, such as on a properly provisioned
# Docker image, it shouldn't be necessary.
export DV_INSTALL_GPU_DRIVERS="${DV_INSTALL_GPU_DRIVERS:-0}"

export PYTHON_BIN_PATH=$(which python)
export USE_DEFAULT_PYTHON_LIB_PATH=1
export DV_COPT_FLAGS="--copt=-march=corei7 --copt=-Wno-sign-compare --copt=-Wno-write-strings"

function note_build_stage {
  echo "========== [$(date)] Stage '${1}' starting"
}

[sgoe1]

Currently some of the settings in settings.sh are outdated/not in-use (particularly the CUDA related ones), so I don’t think you specifying the CUDA version there did anything. If you notice in run-prereq.sh it doesn’t check for TF_CUDA_VERSION, it just automatically installs CUDA 9.0.

You should be able to use CUDA 10, but you’ll have to modify the code in the CUDA section of run-prereq.sh (which is a bit annoying since you’ll have to manually figure out each URL).

We’ll clean this up as well - thanks for pointing out the issue.

[sgoe1]

Hi @andrewrech, you should go ahead with r1.12. You will also need to set pip install --user 'intervaltree==2.1.0' in run-prereq.sh (we’ll have a patch out for this shortly).

Let me know if you have any issues after that.