def _search_file_from_data_path(self, profiling_id, device_id):
"""
Search framework files from data path.
Args:
profiling_id (str): The profiling ID.
device_id (str): The device ID.
Raises:
ProfilerRawFileException: If the framework file type is inconsistent.
ProfilerDeviceIdMismatchException: If the device id is mismatch
with framework in the raw dir.
"""
profiling_data_path = os.path.join(self._raw_data_dir, 'container',
device_id, 'data')
if not os.path.isdir(profiling_data_path):
return
files = os.listdir(profiling_data_path)
for file in files:
pattern = re.search(self._regex_framework_in_data, file)
if not pattern or file.endswith('.done') or file.endswith('.zip'):
continue
attrs = pattern.groupdict()
profiling_id_in_path = attrs.get('profiling_id')
if profiling_id_in_path != profiling_id:
continue
device_id_in_path = attrs.get('device_id')
if device_id_in_path != device_id:
raise ProfilerDeviceIdMismatchException()
data_type = attrs.get('data_type')
if data_type.startswith('vm.'):
if self._backend_type and self._backend_type != 'vm':
raise ProfilerRawFileException(
'Backend type is inconsistent.')
self._backend_type = 'vm'
data_type = data_type.split('.')[1]
else:
if self._backend_type and self._backend_type != 'ge':
raise ProfilerRawFileException(
'Backend type is inconsistent.')
self._backend_type = 'ge'
if data_type.startswith('graph_desc_info'):
self._framework_path['graph'].append(
os.path.join(profiling_data_path, file))
elif data_type.startswith('task_desc_info'):
self._framework_path['task'].append(
os.path.join(profiling_data_path, file))
elif data_type.startswith('point'):
self._framework_path['point'].append(
os.path.join(profiling_data_path, file))
def _search_file_from_job_path(self, device_id, search_in_sub_path=False):
"""
Search framework files from job path.
Args:
device_id (str): The device ID.
search_in_sub_path (bool): `True` if search file in profiling dir,
else search in profiling sub dir. Default: False.
Raises:
ProfilerRawFileException: If the framework file type is inconsistent.
ProfilerDeviceIdMismatchException: If the device id is mismatch
with framework in the raw dir.
"""
profiling_dir = os.path.join(self._profiling_path, 'data') \
if search_in_sub_path else self._profiling_path
if not os.path.isdir(profiling_dir):
return
files = os.listdir(profiling_dir)
for file in files:
pattern = re.search(self._regex_framework, file)
if not pattern or file.endswith('.done'):
continue
attrs = pattern.groupdict()
device_id_in_path = attrs.get('device_id')
if device_id_in_path != device_id:
raise ProfilerDeviceIdMismatchException()
data_type = attrs.get('data_type')
if data_type.startswith('vm.'):
if self._backend_type and self._backend_type != 'vm':
raise ProfilerRawFileException(
'Backend type is inconsistent.')
self._backend_type = 'vm'
data_type = data_type.split('.')[1]
else:
if self._backend_type and self._backend_type != 'ge':
raise ProfilerRawFileException(
'Backend type is inconsistent.')
self._backend_type = 'ge'
if data_type.startswith('graph_desc_info'):
self._framework_path['graph'].append(
os.path.join(profiling_dir, file))
elif data_type.startswith('task_desc_info'):
self._framework_path['task'].append(
os.path.join(profiling_dir, file))
elif data_type.startswith('point'):
self._framework_path['point'].append(
os.path.join(profiling_dir, file))
def _get_minddata_pipeline_info(self):
"""Get the number of thread cores in minddata pipeline operator"""
file_name = self._minddata_pipeline_display_filename.format(
self._device_id)
file_path = os.path.join(self._profiling_dir, file_name)
file_path = validate_and_normalize_path(
file_path, raise_key="Invalid minddata_pipeline_info file path.")
if not os.path.exists(file_path):
log.error('Did not find the minddata_pipeline file: %s', file_path)
raise ProfilerFileNotFoundException(
msg='Did not find the minddata_pipeline file:{}'.format(
file_path))
with open(file_path, 'r', encoding='utf-8') as file:
try:
minddata_pipeline_info = json.load(file)
except json.JSONDecodeError as err:
log.exception(err)
raise ProfilerRawFileException(
"Fail to parse minddata pipeline file")
minddata_pipeline_op_info = []
for item in minddata_pipeline_info.get("op_info"):
op_info_dict = dict()
op_info_dict["op_id"] = item.get("op_id")
op_info_dict["num_workers"] = item.get("num_workers")
minddata_pipeline_op_info.append(op_info_dict)
return minddata_pipeline_op_info
def _parse_and_save_op_info(self, csv_writer, op_id_info_cache,
sample_interval):
"""
Parse and save the minddata pipeline operator information.
Args:
csv_writer (csv.writer): The csv writer.
op_id_info_cache (dict): The operator id and information cache.
sample_interval (int): The sample interval.
Raises:
ProfilerRawFileException: If the operator that id is 0 does not exist.
"""
queue = Queue()
root_node = op_id_info_cache.get(0)
if not root_node:
raise ProfilerRawFileException(
'The format of minddata pipeline raw file is wrong, '
'the operator that id is 0 does not exist.')
root_node['parent_id'] = None
queue.put_nowait(root_node)
while not queue.empty():
node = queue.get_nowait()
self._update_child_node(node, op_id_info_cache)
csv_writer.writerow(self._get_op_info(node, sample_interval))
op_id = node.get('op_id')
children_ids = node.get('children')
if not children_ids:
continue
for child_op_id in children_ids:
sub_node = op_id_info_cache.get(child_op_id)
sub_node['parent_id'] = op_id
queue.put_nowait(sub_node)
def _parse_and_save(self, pipeline_info):
"""
Parse and save the parsed minddata pipeline file.
Args:
pipeline_info (dict): The pipeline info reads from the raw file of
the minddata pipeline.
Raises:
ProfilerRawFileException: If the format of minddata pipeline raw
file is wrong.
"""
sample_interval = pipeline_info.get('sampling_interval')
op_info = pipeline_info.get('op_info')
if sample_interval is None or not op_info:
raise ProfilerRawFileException(
'The format of minddata pipeline raw file is wrong.')
op_id_info_cache = {}
for item in op_info:
op_id_info_cache[item.get('op_id')] = item
with open(self._save_path, 'w') as save_file:
csv_writer = csv.writer(save_file)
csv_writer.writerow(self._col_names)
self._parse_and_save_op_info(csv_writer, op_id_info_cache,
sample_interval)
def _load_point_info(self):
"""Load point info."""
file_path = os.path.join(self._profiling_dir,
'step_trace_point_info.json')
file_path = validate_and_normalize_path(
file_path, raise_key="Invalid step_trace_point_info file path.")
# If step_trace_point_info_{self._device_id}.json file exist, load this file.
file_path_new = os.path.join(
self._profiling_dir,
f'step_trace_point_info_{self._device_id}.json')
file_path_new = validate_and_normalize_path(
file_path_new,
raise_key="Invalid step_trace_point_info file path.")
if os.path.isfile(file_path_new):
file_path = file_path_new
if os.path.isfile(file_path):
with open(file_path, 'r', encoding='utf-8') as file:
try:
self._point_info = json.load(file)
except (json.JSONDecodeError, TypeError) as err:
log.exception(err)
raise ProfilerRawFileException(
'Fail to parse point info file.')
def _load_point_info(self):
"""Load point info."""
file_path = os.path.join(self._profiling_dir, 'step_trace_point_info.json')
if os.path.isfile(file_path):
with open(file_path, 'r', encoding='utf-8') as file:
try:
self._point_info = json.load(file)
except (json.JSONDecodeError, TypeError) as err:
log.exception(err)
raise ProfilerRawFileException('Fail to parse point info file.')
def parse(self):
"""
Parse the minddata pipeline files.
Raises:
ProfilerRawFileException: If fails to parse the raw file of
minddata pipeline or the file is empty.
"""
with open(self._pipeline_path, 'r') as file:
try:
pipeline_info = json.load(file)
except (json.JSONDecodeError, TypeError) as err:
logger.exception(err)
raise ProfilerRawFileException(
'Fail to parse minddata pipeline file.')
if not pipeline_info:
logger.warning('The minddata pipeline file is empty.')
raise ProfilerRawFileException(
'The minddata pipeline file is empty.')
self._parse_and_save(pipeline_info)
def _load(self):
"""Load cpu_utilization info."""
file_name = self._cpu_utilization_display_filename.format(
self._device_id)
file_path = os.path.join(self._profiling_dir, file_name)
file_path = validate_and_normalize_path(
file_path, raise_key="Invalid cpu_utilization_info file path.")
if not os.path.exists(file_path):
log.error('Did not find the cpu utilization file: %s', file_path)
raise ProfilerFileNotFoundException(
msg='Did not find the cpu utilization file.')
with open(file_path, 'r', encoding='utf-8') as src_file:
try:
self._data = json.load(src_file)
except json.JSONDecodeError as err:
log.exception(err)
raise ProfilerRawFileException(
"Fail to parse cpu_utilization info file")
def _convert_field_type(self, row):
"""
Convert the field type to the specific type.
Args:
row (list): One row data from parsed data.
Returns:
list, the converted data.
"""
try:
return [
row[0], row[1], row[2], row[3],
int(row[4]),
self._format_float_data(float(row[5])),
self._format_float_data(float(row[6])),
self._format_float_data(float(row[7])), row[8]
]
except IndexError as err:
log.exception(err)
raise ProfilerRawFileException(
'failed to get HOST CPU operator detail data.')
def _convert_field_type(row):
"""
Convert the field type to the specific type.
Args:
row (list): One row data from parsed data.
Returns:
list, the converted data.
"""
try:
return [
row[0],
int(row[1]),
int(row[2]),
float(row[3]),
float(row[4]),
float(row[5]) * 100
]
except IndexError as err:
log.exception(err)
raise ProfilerRawFileException(
'failed to get HOST CPU operator type data.')