Ben Murdoch | da12d29 | 2016-06-02 14:46:10 +0100 | [diff] [blame^] | 1 | #!/usr/bin/env python |
| 2 | # Copyright 2016 the V8 project authors. All rights reserved. |
| 3 | # Use of this source code is governed by a BSD-style license that can be |
| 4 | # found in the LICENSE file. |
| 5 | |
| 6 | """Script for merging sancov files in parallel. |
| 7 | |
| 8 | When merging test runner output, the sancov files are expected |
| 9 | to be located in one directory with the file-name pattern: |
| 10 | <executable name>.test.<id>.sancov |
| 11 | |
| 12 | For each executable, this script writes a new file: |
| 13 | <executable name>.result.sancov |
| 14 | |
| 15 | When --swarming-output-dir is specified, this script will merge the result |
| 16 | files found there into the coverage folder. |
| 17 | |
| 18 | The sancov tool is expected to be in the llvm compiler-rt third-party |
| 19 | directory. It's not checked out by default and must be added as a custom deps: |
| 20 | 'v8/third_party/llvm/projects/compiler-rt': |
| 21 | 'https://chromium.googlesource.com/external/llvm.org/compiler-rt.git' |
| 22 | """ |
| 23 | |
| 24 | import argparse |
| 25 | import logging |
| 26 | import math |
| 27 | import os |
| 28 | import re |
| 29 | import subprocess |
| 30 | import sys |
| 31 | |
| 32 | from multiprocessing import Pool, cpu_count |
| 33 | |
| 34 | |
| 35 | logging.basicConfig(level=logging.INFO) |
| 36 | |
| 37 | # V8 checkout directory. |
| 38 | BASE_DIR = os.path.dirname(os.path.dirname(os.path.dirname( |
| 39 | os.path.abspath(__file__)))) |
| 40 | |
| 41 | # The sancov tool location. |
| 42 | SANCOV_TOOL = os.path.join( |
| 43 | BASE_DIR, 'third_party', 'llvm', 'projects', 'compiler-rt', |
| 44 | 'lib', 'sanitizer_common', 'scripts', 'sancov.py') |
| 45 | |
| 46 | # Number of cpus. |
| 47 | CPUS = cpu_count() |
| 48 | |
| 49 | # Regexp to find sancov file as output by the v8 test runner. Also grabs the |
| 50 | # executable name in group 1. |
| 51 | SANCOV_FILE_RE = re.compile(r'^(.*)\.test\.\d+\.sancov$') |
| 52 | |
| 53 | # Regexp to find sancov result files as returned from swarming. |
| 54 | SANCOV_RESULTS_FILE_RE = re.compile(r'^.*\.result\.sancov$') |
| 55 | |
| 56 | |
| 57 | def merge(args): |
| 58 | """Merge several sancov files into one. |
| 59 | |
| 60 | Called trough multiprocessing pool. The args are expected to unpack to: |
| 61 | keep: Option if source and intermediate sancov files should be kept. |
| 62 | coverage_dir: Folder where to find the sancov files. |
| 63 | executable: Name of the executable whose sancov files should be merged. |
| 64 | index: A number to be put into the intermediate result file name. |
| 65 | If None, this is a final result. |
| 66 | bucket: The list of sancov files to be merged. |
| 67 | Returns: A tuple with the executable name and the result file name. |
| 68 | """ |
| 69 | keep, coverage_dir, executable, index, bucket = args |
| 70 | process = subprocess.Popen( |
| 71 | [SANCOV_TOOL, 'merge'] + bucket, |
| 72 | stdout=subprocess.PIPE, |
| 73 | stderr=subprocess.PIPE, |
| 74 | cwd=coverage_dir, |
| 75 | ) |
| 76 | output, _ = process.communicate() |
| 77 | assert process.returncode == 0 |
| 78 | if index is not None: |
| 79 | # This is an intermediate result, add the bucket index to the file name. |
| 80 | result_file_name = '%s.result.%d.sancov' % (executable, index) |
| 81 | else: |
| 82 | # This is the final result without bucket index. |
| 83 | result_file_name = '%s.result.sancov' % executable |
| 84 | with open(os.path.join(coverage_dir, result_file_name), "wb") as f: |
| 85 | f.write(output) |
| 86 | if not keep: |
| 87 | for f in bucket: |
| 88 | os.remove(os.path.join(coverage_dir, f)) |
| 89 | return executable, result_file_name |
| 90 | |
| 91 | |
| 92 | def generate_inputs(keep, coverage_dir, file_map, cpus): |
| 93 | """Generate inputs for multiprocessed merging. |
| 94 | |
| 95 | Splits the sancov files into several buckets, so that each bucket can be |
| 96 | merged in a separate process. We have only few executables in total with |
| 97 | mostly lots of associated files. In the general case, with many executables |
| 98 | we might need to avoid splitting buckets of executables with few files. |
| 99 | |
| 100 | Returns: List of args as expected by merge above. |
| 101 | """ |
| 102 | inputs = [] |
| 103 | for executable, files in file_map.iteritems(): |
| 104 | # What's the bucket size for distributing files for merging? E.g. with |
| 105 | # 2 cpus and 9 files we want bucket size 5. |
| 106 | n = max(2, int(math.ceil(len(files) / float(cpus)))) |
| 107 | |
| 108 | # Chop files into buckets. |
| 109 | buckets = [files[i:i+n] for i in xrange(0, len(files), n)] |
| 110 | |
| 111 | # Inputs for multiprocessing. List of tuples containing: |
| 112 | # Keep-files option, base path, executable name, index of bucket, |
| 113 | # list of files. |
| 114 | inputs.extend([(keep, coverage_dir, executable, i, b) |
| 115 | for i, b in enumerate(buckets)]) |
| 116 | return inputs |
| 117 | |
| 118 | |
| 119 | def merge_parallel(inputs, merge_fun=merge): |
| 120 | """Process several merge jobs in parallel.""" |
| 121 | pool = Pool(CPUS) |
| 122 | try: |
| 123 | return pool.map(merge_fun, inputs) |
| 124 | finally: |
| 125 | pool.close() |
| 126 | |
| 127 | |
| 128 | def merge_test_runner_output(options): |
| 129 | # Map executable names to their respective sancov files. |
| 130 | file_map = {} |
| 131 | for f in os.listdir(options.coverage_dir): |
| 132 | match = SANCOV_FILE_RE.match(f) |
| 133 | if match: |
| 134 | file_map.setdefault(match.group(1), []).append(f) |
| 135 | |
| 136 | inputs = generate_inputs( |
| 137 | options.keep, options.coverage_dir, file_map, CPUS) |
| 138 | |
| 139 | logging.info('Executing %d merge jobs in parallel for %d executables.' % |
| 140 | (len(inputs), len(file_map))) |
| 141 | |
| 142 | results = merge_parallel(inputs) |
| 143 | |
| 144 | # Map executable names to intermediate bucket result files. |
| 145 | file_map = {} |
| 146 | for executable, f in results: |
| 147 | file_map.setdefault(executable, []).append(f) |
| 148 | |
| 149 | # Merge the bucket results for each executable. |
| 150 | # The final result has index None, so no index will appear in the |
| 151 | # file name. |
| 152 | inputs = [(options.keep, options.coverage_dir, executable, None, files) |
| 153 | for executable, files in file_map.iteritems()] |
| 154 | |
| 155 | logging.info('Merging %d intermediate results.' % len(inputs)) |
| 156 | |
| 157 | merge_parallel(inputs) |
| 158 | |
| 159 | |
| 160 | def merge_two(args): |
| 161 | """Merge two sancov files. |
| 162 | |
| 163 | Called trough multiprocessing pool. The args are expected to unpack to: |
| 164 | swarming_output_dir: Folder where to find the new file. |
| 165 | coverage_dir: Folder where to find the existing file. |
| 166 | f: File name of the file to be merged. |
| 167 | """ |
| 168 | swarming_output_dir, coverage_dir, f = args |
| 169 | input_file = os.path.join(swarming_output_dir, f) |
| 170 | output_file = os.path.join(coverage_dir, f) |
| 171 | process = subprocess.Popen( |
| 172 | [SANCOV_TOOL, 'merge', input_file, output_file], |
| 173 | stdout=subprocess.PIPE, |
| 174 | stderr=subprocess.PIPE, |
| 175 | ) |
| 176 | output, _ = process.communicate() |
| 177 | assert process.returncode == 0 |
| 178 | with open(output_file, "wb") as f: |
| 179 | f.write(output) |
| 180 | |
| 181 | |
| 182 | def merge_swarming_output(options): |
| 183 | # Iterate sancov files from swarming. |
| 184 | files = [] |
| 185 | for f in os.listdir(options.swarming_output_dir): |
| 186 | match = SANCOV_RESULTS_FILE_RE.match(f) |
| 187 | if match: |
| 188 | if os.path.exists(os.path.join(options.coverage_dir, f)): |
| 189 | # If the same file already exists, we'll merge the data. |
| 190 | files.append(f) |
| 191 | else: |
| 192 | # No file yet? Just move it. |
| 193 | os.rename(os.path.join(options.swarming_output_dir, f), |
| 194 | os.path.join(options.coverage_dir, f)) |
| 195 | |
| 196 | inputs = [(options.swarming_output_dir, options.coverage_dir, f) |
| 197 | for f in files] |
| 198 | |
| 199 | logging.info('Executing %d merge jobs in parallel.' % len(inputs)) |
| 200 | merge_parallel(inputs, merge_two) |
| 201 | |
| 202 | |
| 203 | def main(): |
| 204 | parser = argparse.ArgumentParser() |
| 205 | parser.add_argument('--coverage-dir', required=True, |
| 206 | help='Path to the sancov output files.') |
| 207 | parser.add_argument('--keep', default=False, action='store_true', |
| 208 | help='Keep sancov output files after merging.') |
| 209 | parser.add_argument('--swarming-output-dir', |
| 210 | help='Folder containing a results shard from swarming.') |
| 211 | options = parser.parse_args() |
| 212 | |
| 213 | # Check if folder with coverage output exists. |
| 214 | assert (os.path.exists(options.coverage_dir) and |
| 215 | os.path.isdir(options.coverage_dir)) |
| 216 | |
| 217 | if options.swarming_output_dir: |
| 218 | # Check if folder with swarming output exists. |
| 219 | assert (os.path.exists(options.swarming_output_dir) and |
| 220 | os.path.isdir(options.swarming_output_dir)) |
| 221 | merge_swarming_output(options) |
| 222 | else: |
| 223 | merge_test_runner_output(options) |
| 224 | |
| 225 | return 0 |
| 226 | |
| 227 | |
| 228 | if __name__ == '__main__': |
| 229 | sys.exit(main()) |