def exec_on_subproc(self, f, args): """Send work to a subprocess and block on it. This can be used by existing background Work in a ThreadPool to sidestep the GIL: The Thread calls this method and still blocks until the work is complete, so existing reporting and accounting is unchanged, but the actual work is executed in a subprocess, avoiding lock contention. :param f: A multiproc-friendly (importable) work function. :param args: Multiproc-friendly (pickleable) arguments to f. """ return SubprocPool.background().apply(f, args)
def __init__(self, info_dir, stats_upload_url=None, stats_upload_timeout=2, num_foreground_workers=8, num_background_workers=8): self.run_timestamp = time.time() # A double, so we get subsecond precision for ids. cmd_line = ' '.join(['./pants'] + sys.argv[1:]) # run_id is safe for use in paths. millis = (self.run_timestamp * 1000) % 1000 run_id = 'pants_run_%s_%d' % \ (time.strftime('%Y_%m_%d_%H_%M_%S', time.localtime(self.run_timestamp)), millis) self.info_dir = os.path.join(info_dir, run_id) self.run_info = RunInfo(os.path.join(self.info_dir, 'info')) self.run_info.add_basic_info(run_id, self.run_timestamp) self.run_info.add_info('cmd_line', cmd_line) self.stats_url = stats_upload_url self.stats_timeout = stats_upload_timeout # Create a 'latest' symlink, after we add_infos, so we're guaranteed that the file exists. link_to_latest = os.path.join(os.path.dirname(self.info_dir), 'latest') if os.path.lexists(link_to_latest): os.unlink(link_to_latest) os.symlink(self.info_dir, link_to_latest) # Time spent in a workunit, including its children. self.cumulative_timings = AggregatedTimings(os.path.join(self.info_dir, 'cumulative_timings')) # Time spent in a workunit, not including its children. self.self_timings = AggregatedTimings(os.path.join(self.info_dir, 'self_timings')) # Hit/miss stats for the artifact cache. self.artifact_cache_stats = \ ArtifactCacheStats(os.path.join(self.info_dir, 'artifact_cache_stats')) # Number of threads for foreground work. self._num_foreground_workers = num_foreground_workers # Number of threads for background work. self._num_background_workers = num_background_workers # We report to this Report. self.report = None # self._threadlocal.current_workunit contains the current workunit for the calling thread. # Note that multiple threads may share a name (e.g., all the threads in a pool). self._threadlocal = threading.local() # For main thread work. Created on start(). self._main_root_workunit = None # For concurrent foreground work. Created lazily if needed. # Associated with the main thread's root workunit. self._foreground_worker_pool = None # For background work. Created lazily if needed. self._background_worker_pool = None self._background_root_workunit = None # Trigger subproc pool init while our memory image is still clean (see SubprocPool docstring) SubprocPool.foreground() SubprocPool.background() self._aborted = False