#!/bin/bash # # Change to repo root cd $(dirname $0)/../../.. set -ex export OUTPUT_DIR=testoutput repo_root="$(pwd)" # TODO(jtattermusch): Add back support for benchmarking with tcmalloc for C++ and python. # This feature was removed since it used to use tcmalloc from https://github.com/gperftools/gperftools.git # which is very outdated. See https://github.com/protocolbuffers/protobuf/issues/8725. # download datasets for benchmark pushd benchmarks datasets=$(for file in $(find . -type f -name "dataset.*.pb" -not -path "./tmp/*"); do echo "$(pwd)/$file"; done | xargs) echo $datasets popd # build Python protobuf ./autogen.sh ./configure CXXFLAGS="-fPIC -O2" make -j8 pushd python virtualenv -p python3 env source env/bin/activate python3 setup.py build --cpp_implementation pip3 install --install-option="--cpp_implementation" . popd # build and run Python benchmark # We do this before building protobuf C++ since C++ build # will rewrite some libraries used by protobuf python. pushd benchmarks make python-pure-python-benchmark make python-cpp-reflection-benchmark make -j8 python-cpp-generated-code-benchmark echo "[" > tmp/python_result.json echo "benchmarking pure python..." ./python-pure-python-benchmark --json --behavior_prefix="pure-python-benchmark" $datasets >> tmp/python_result.json echo "," >> "tmp/python_result.json" echo "benchmarking python cpp reflection..." env LD_LIBRARY_PATH="${repo_root}/src/.libs" ./python-cpp-reflection-benchmark --json --behavior_prefix="cpp-reflection-benchmark" $datasets >> tmp/python_result.json echo "," >> "tmp/python_result.json" echo "benchmarking python cpp generated code..." env LD_LIBRARY_PATH="${repo_root}/src/.libs" ./python-cpp-generated-code-benchmark --json --behavior_prefix="cpp-generated-code-benchmark" $datasets >> tmp/python_result.json echo "]" >> "tmp/python_result.json" popd # build CPP protobuf ./configure make clean && make -j8 pushd java mvn package -B -Dmaven.test.skip=true popd pushd benchmarks # build and run C++ benchmark # "make clean" deletes the contents of the tmp/ directory, so we move it elsewhere and then restore it once build is done. # TODO(jtattermusch): find a less clumsy way of protecting python_result.json contents mv tmp/python_result.json . && make clean && make -j8 cpp-benchmark && mv python_result.json tmp echo "benchmarking cpp..." env ./cpp-benchmark --benchmark_min_time=5.0 --benchmark_out_format=json --benchmark_out="tmp/cpp_result.json" $datasets # TODO(jtattermusch): add benchmarks for https://github.com/protocolbuffers/protobuf-go. # The original benchmarks for https://github.com/golang/protobuf were removed # because: # * they were broken and haven't been producing results for a long time # * the https://github.com/golang/protobuf implementation has been superseded by # https://github.com/protocolbuffers/protobuf-go # build and run java benchmark (java 11 is required) make java-benchmark echo "benchmarking java..." ./java-benchmark -Cresults.file.options.file="tmp/java_result.json" $datasets # TODO(jtattermusch): re-enable JS benchmarks once https://github.com/protocolbuffers/protobuf/issues/8747 is fixed. # build and run js benchmark # make js-benchmark # echo "benchmarking js..." # ./js-benchmark $datasets --json_output=$(pwd)/tmp/node_result.json # TODO(jtattermusch): add php-c-benchmark. Currently its build is broken. # persist raw the results in the build job log (for better debuggability) cat tmp/cpp_result.json cat tmp/java_result.json cat tmp/python_result.json # print the postprocessed results to the build job log # TODO(jtattermusch): re-enable uploading results to bigquery (it is currently broken) make python_add_init env LD_LIBRARY_PATH="${repo_root}/src/.libs" python3 -m util.result_parser \ -cpp="../tmp/cpp_result.json" -java="../tmp/java_result.json" -python="../tmp/python_result.json" popd