This is an automated email from the ASF dual-hosted git repository.
wesm pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/arrow.git
The following commit(s) were added to refs/heads/master by this push:
new 3330d66 ARROW-4118: [Python] Fix benchmark setup for "asv run"
3330d66 is described below
commit 3330d660643a034168b472b52aebfe0fea84b8cf
Author: Antoine Pitrou <anto...@python.org>
AuthorDate: Wed Jan 9 16:14:25 2019 -0600
ARROW-4118: [Python] Fix benchmark setup for "asv run"
"conda activate" unfortunately isn't available from a non-interactive
shell, and running bash as interactive doesn't look like a workable solution.
Also fix a setup slowness issue in the Parquet benchmarks, and fix a C++
ABI issue by downloading packages from Anaconda rather than conda-forge.
Author: Antoine Pitrou <anto...@python.org>
Closes #3357 from pitrou/ARROW-4118-fix-asv-run and squashes the following
commits:
b07b68e61 <Antoine Pitrou> ARROW-4118: Fix benchmark setup for "asv run"
---
docs/source/python/benchmarks.rst | 24 +++++++++++++-----------
python/asv-build.sh | 17 ++++++++++++-----
python/asv.conf.json | 4 +++-
python/benchmarks/parquet.py | 16 +++++++++-------
4 files changed, 37 insertions(+), 24 deletions(-)
diff --git a/docs/source/python/benchmarks.rst
b/docs/source/python/benchmarks.rst
index 7672294..12205c5 100644
--- a/docs/source/python/benchmarks.rst
+++ b/docs/source/python/benchmarks.rst
@@ -19,35 +19,37 @@ Benchmarks
==========
The ``pyarrow`` package comes with a suite of benchmarks meant to
-run with `asv`_. You'll need to install the ``asv`` package first
+run with `ASV`_. You'll need to install the ``asv`` package first
(``pip install asv`` or ``conda install -c conda-forge asv``).
-The benchmarks are run using `asv`_ which is also their only requirement.
-
Running the benchmarks
----------------------
-To run the benchmarks, call ``asv run --python=same``. You cannot use the
-plain ``asv run`` command at the moment as asv cannot handle python packages
-in subdirectories of a repository.
+To run the benchmarks for a locally-built Arrow, run ``asv dev`` or
+``asv run --python=same``.
-Running with arbitrary revisions
---------------------------------
+Running for arbitrary Git revisions
+-----------------------------------
ASV allows to store results and generate graphs of the benchmarks over
-the project's evolution. For this you have the latest development version of
ASV:
+the project's evolution. You need to have the latest development version of
ASV:
.. code::
pip install git+https://github.com/airspeed-velocity/asv
+The build scripts assume that Conda's ``activate`` script is on the PATH
+(the ``conda activate`` command unfortunately isn't available from
+non-interactive scripts).
+
Now you should be ready to run ``asv run`` or whatever other command
-suits your needs.
+suits your needs. Note that this can be quite long, as each Arrow needs
+to be rebuilt for each Git revision you're running the benchmarks for.
Compatibility
-------------
We only expect the benchmarking setup to work with Python 3.6 or later,
-on a Unix-like system.
+on a Unix-like system with bash.
.. _asv: https://asv.readthedocs.org/
diff --git a/python/asv-build.sh b/python/asv-build.sh
index 7b55456..90c7872 100755
--- a/python/asv-build.sh
+++ b/python/asv-build.sh
@@ -21,7 +21,9 @@ set -e
# ASV doesn't activate its conda environment for us
if [ -z "$ASV_ENV_DIR" ]; then exit 1; fi
-conda activate $ASV_ENV_DIR
+# Avoid "conda activate" because it's only set up in interactive shells
+# (https://github.com/conda/conda/issues/8072)
+source activate $ASV_ENV_DIR
echo "== Conda Prefix for benchmarks: " $CONDA_PREFIX " =="
# Build Arrow C++ libraries
@@ -32,6 +34,8 @@ export ORC_HOME=$CONDA_PREFIX
export PROTOBUF_HOME=$CONDA_PREFIX
export BOOST_ROOT=$CONDA_PREFIX
+export CXXFLAGS="-D_GLIBCXX_USE_CXX11_ABI=1"
+
pushd ../cpp
mkdir -p build
pushd build
@@ -40,9 +44,11 @@ cmake -GNinja \
-DCMAKE_BUILD_TYPE=release \
-DCMAKE_INSTALL_PREFIX=$ARROW_HOME \
-DARROW_CXXFLAGS=$CXXFLAGS \
- -DARROW_PYTHON=ON \
- -DARROW_PLASMA=ON \
- -DARROW_BUILD_TESTS=OFF \
+ -DARROW_USE_GLOG=off \
+ -DARROW_PARQUET=on \
+ -DARROW_PYTHON=on \
+ -DARROW_PLASMA=on \
+ -DARROW_BUILD_TESTS=off \
..
cmake --build . --target install
@@ -52,7 +58,8 @@ popd
# Build pyarrow wrappers
export SETUPTOOLS_SCM_PRETEND_VERSION=0.0.1
export PYARROW_BUILD_TYPE=release
-export PYARROW_PARALLEL=4
+export PYARROW_PARALLEL=8
+export PYARROW_WITH_PARQUET=1
export PYARROW_WITH_PLASMA=1
python setup.py clean
diff --git a/python/asv.conf.json b/python/asv.conf.json
index 40938ee..09031c8 100644
--- a/python/asv.conf.json
+++ b/python/asv.conf.json
@@ -35,6 +35,7 @@
// of the repository.
"repo_subdir": "python",
+ // Custom build commands for Arrow.
"build_command": ["/bin/bash {build_dir}/asv-build.sh"],
"install_command": ["/bin/bash {build_dir}/asv-install.sh"],
"uninstall_command": ["/bin/bash {build_dir}/asv-uninstall.sh"],
@@ -56,7 +57,8 @@
// determined by looking for tools on the PATH environment
// variable.
"environment_type": "conda",
- "conda_channels": ["conda-forge", "defaults"],
+ // Avoid conda-forge to avoid C++ ABI issues
+ "conda_channels": ["defaults"],
// the base URL to show a commit for the project.
"show_commit_url": "https://github.com/apache/arrow/commit/";,
diff --git a/python/benchmarks/parquet.py b/python/benchmarks/parquet.py
index fd61793..4f55587 100644
--- a/python/benchmarks/parquet.py
+++ b/python/benchmarks/parquet.py
@@ -15,11 +15,12 @@
# specific language governing permissions and limitations
# under the License.
-import pandas as pd
-import random
import shutil
import tempfile
+import numpy as np
+import pandas as pd
+
import pyarrow as pa
try:
import pyarrow.parquet as pq
@@ -38,18 +39,19 @@ class ParquetManifestCreation(object):
def setup(self, num_partitions, num_threads):
if pq is None:
- raise NotImplementedError
+ raise NotImplementedError("Parquet support not enabled")
self.tmpdir = tempfile.mkdtemp('benchmark_parquet')
- num1 = [random.choice(range(0, num_partitions))
- for _ in range(self.size)]
- num2 = [random.choice(range(0, 1000)) for _ in range(self.size)]
+ rnd = np.random.RandomState(42)
+ num1 = rnd.randint(0, num_partitions, size=self.size)
+ num2 = rnd.randint(0, 1000, size=self.size)
output_df = pd.DataFrame({'num1': num1, 'num2': num2})
output_table = pa.Table.from_pandas(output_df)
pq.write_to_dataset(output_table, self.tmpdir, ['num1'])
def teardown(self, num_partitions, num_threads):
- shutil.rmtree(self.tmpdir)
+ if self.tmpdir is not None:
+ shutil.rmtree(self.tmpdir)
def time_manifest_creation(self, num_partitions, num_threads):
pq.ParquetManifest(self.tmpdir, metadata_nthreads=num_threads)
