def extract_stats_parallel(
index: io.Index,
pool: Optional[multiprocessing.pool.Pool] = None) -> List[StationStat]:
"""
Extracts StationStat information in parallel from packets stored in the provided index.
:param index: Index of packets to extract information from.
:param pool: optional multiprocessing pool.
:return: A list of StationStat objects.
"""
# Partition the index entries by number of cores
num_cores: int = multiprocessing.cpu_count()
partitioned: List[List[io.IndexEntry]] = _partition_list(
index.entries, num_cores)
indices: List[io.Index] = list(
map(lambda entries: io.Index(entries), partitioned))
# Run da buggahs in parallel
# _pool: multiprocessing.pool.Pool = multiprocessing.Pool() if pool is None else pool
# nested: List[List[StationStat]] = _pool.map(extract_stats_serial, indices)
# if pool is None:
# _pool.close()
nested: Iterator[List[StationStat]] = maybe_parallel_map(
pool,
extract_stats_serial,
iter(indices),
lambda: len(indices) > 128,
chunk_size=64,
)
return [item for sublist in nested for item in sublist]
def test_normal_managed_pool(self):
settings.set_parallelism_enabled(True)
usage_out = []
res = maybe_parallel_map(None, map_fn, self.data, usage_out=usage_out)
self.assertEqual(self.res, list(res))
self.assertEqual(MappingType.ParallelManaged, usage_out[0])
settings.set_parallelism_enabled(False)
def get_stations(
self, pool: Optional[multiprocessing.pool.Pool] = None
) -> List[Station]:
"""
:return: a list of all stations represented by the data packets
"""
station_ids: List[str] = self.index_summary.station_ids()
stations_opt: Iterator[Optional[Station]] = maybe_parallel_map(
pool,
self.get_station_by_id,
iter(station_ids),
lambda: len(station_ids) > 2,
chunk_size=1
)
# _pool: multiprocessing.pool.Pool = (
# multiprocessing.Pool() if pool is None else pool
# )
#
# stations_opt: List[Optional[Station]] = _pool.map(
# self.get_station_by_id, station_ids
# )
# if pool is None:
# _pool.close()
# noinspection Mypy
return list(filter(lambda station: station is not None, stations_opt))
def test_normal_provided_pool(self):
settings.set_parallelism_enabled(True)
pool = Pool()
usage_out = []
res = maybe_parallel_map(pool, map_fn, self.data, usage_out=usage_out)
self.assertEqual(self.res, list(res))
self.assertEqual(MappingType.ParallelUnmanaged, usage_out[0])
pool.close()
settings.set_parallelism_enabled(False)
def test_parallel_condition_bad(self):
settings.set_parallelism_enabled(True)
usage_out = []
res = maybe_parallel_map(None,
map_fn,
self.data,
usage_out=usage_out,
condition=lambda: len(self.data) > 10)
self.assertEqual(self.res, list(res))
self.assertEqual(MappingType.Serial, usage_out[0])
settings.set_parallelism_enabled(False)
def get_stations(
self,
pool: Optional[multiprocessing.pool.Pool] = None) -> List[Station]:
"""
:param pool: optional multiprocessing pool
:return: List of all stations in the ApiReader
"""
return list(
maybe_parallel_map(pool,
self._station_by_index,
self.files_index,
chunk_size=1))
def _read_stations(
self,
pool: Optional[multiprocessing.pool.Pool] = None) -> List[Station]:
"""
:param pool: optional multiprocessing pool
:return: List of all stations in the ApiReader, without building the data from parquet
"""
if settings.is_parallelism_enabled() and len(self.files_index) > 1:
return list(
maybe_parallel_map(pool,
self._station_by_index,
self.files_index,
chunk_size=1))
return list(map(self._station_by_index, self.files_index))
def create_data_window(self, pool: Optional[multiprocessing.pool.Pool] = None):
"""
updates the DataWindow to contain only the data within the window parameters
stations without audio or any data outside the window are removed
"""
# Let's create and manage a single pool of workers that we can utilize throughout
# the instantiation of the data window.
_pool: multiprocessing.pool.Pool = multiprocessing.Pool() if pool is None else pool
r_f = io.ReadFilter()
if self._config.start_datetime:
r_f.with_start_dt(self._config.start_datetime)
if self._config.end_datetime:
r_f.with_end_dt(self._config.end_datetime)
if self._config.station_ids:
r_f.with_station_ids(self._config.station_ids)
if self._config.extensions:
r_f.with_extensions(self._config.extensions)
else:
self._config.extensions = r_f.extensions
if self._config.api_versions:
r_f.with_api_versions(self._config.api_versions)
else:
self._config.api_versions = r_f.api_versions
r_f.with_start_dt_buf(self._config.start_buffer_td)
r_f.with_end_dt_buf(self._config.end_buffer_td)
if self.debug:
print("Reading files from disk. This may take a few minutes to complete.")
# get the data to convert into a window
a_r = ApiReaderDw(self._config.input_dir, self._config.structured_layout, r_f,
correct_timestamps=self._config.apply_correction,
use_model_correction=self._config.use_model_correction,
dw_base_dir=self.save_dir(),
dw_save_mode=self._fs_writer.save_mode(),
debug=self.debug, pool=_pool)
self._errors.extend_error(a_r.errors)
if self._fs_writer.is_use_mem() and a_r.dw_save_mode != self._fs_writer.save_mode():
if self.debug:
print("Estimated size of files exceeds available memory.")
print("Automatically using temporary directory to store data.")
self._fs_writer.set_use_temp(True)
# Parallel update
# Apply timing correction in parallel by station
sts = a_r.get_stations()
if self.debug:
print("num stations loaded: ", len(sts))
# if self._config.apply_correction:
# for st in maybe_parallel_map(_pool, Station.update_timestamps,
# iter(sts), chunk_size=1):
# self._add_sensor_to_window(st)
# if self.debug:
# print("station processed: ", st.id())
for st in maybe_parallel_map(_pool, Station.update_timestamps, iter(sts), chunk_size=1):
self.create_window_in_sensors(st, self._config.start_datetime, self._config.end_datetime)
if self.debug:
print("station processed: ", st.id())
# check for stations without data
self._check_for_audio()
self._check_valid_ids()
# update the default data window name if we have data and the default name exists
if self.event_name == "dw" and len(self._stations) > 0:
self.event_name = f"dw_{int(self.start_date())}_{len(self._stations)}"
# must update the start and end in order for the data to be saved
# update remaining data window values if they're still default
if not self._config.start_datetime and len(self._stations) > 0:
self._config.start_datetime = dtu.datetime_from_epoch_microseconds_utc(
np.min([t.first_data_timestamp() for t in self._stations]))
# end_datetime is non-inclusive, so it must be greater than our latest timestamp
if not self._config.end_datetime and len(self._stations) > 0:
self._config.end_datetime = dtu.datetime_from_epoch_microseconds_utc(
np.max([t.last_data_timestamp() for t in self._stations]) + 1)
# If the pool was created by this function, then it needs to managed by this function.
if pool is None:
_pool.close()
