def _repeat_loading(self, repeat_interval):
"""Periodically loading summary."""
while True:
try:
logger.info('Start to load data, repeat interval: %r.',
repeat_interval)
self._load_data()
if not repeat_interval:
return
time.sleep(repeat_interval)
except UnknownError as ex:
logger.error(
'Unexpected error happens when loading data. Loading status: %s, loading pool size: %d'
'Detail: %s', self._loading_status, len(self._loader_pool),
str(ex))
def _repeat_loading(self, repeat_interval):
"""Periodically loading summary."""
# Allocate CPU resources to enable gunicorn to start the web service.
time.sleep(1)
while True:
try:
if self.status == _ExplainManagerStatus.STOPPING.value:
logger.debug(
'Current loading status is %s, we will not trigger repeat loading.',
_ExplainManagerStatus.STOPPING.value)
else:
logger.info(
'Starts triggering repeat loading, repeat interval: %r.',
repeat_interval)
self._load_data()
if not repeat_interval:
return
time.sleep(repeat_interval)
except UnknownError as ex:
logger.error(
'Unexpected error happens when loading data. Loading status: %s, loading pool size: %d'
'Detail: %s', self.status, len(self._loader_pool), str(ex))
def list_events(self, filenames):
"""
Load summary file and parse file content.
Args:
filenames (list[str]): File name list.
Returns:
tuple, the elements of the tuple are:
- file_changed (bool): True if the latest file is changed.
- is_end (bool): True if all the summary files are finished loading.
- event_data (dict): Event data where keys are explanation field.
"""
summary_files = self.sort_files(filenames)
is_end = False
file_changed = False
event_data = {}
filename = summary_files[-1]
file_path = FileHandler.join(self._summary_dir, filename)
if filename != self._latest_filename:
self._summary_file_handler = FileHandler(file_path, 'rb')
self._latest_filename = filename
self._latest_offset = 0
file_changed = True
new_size = FileHandler.file_stat(file_path).size
if new_size == self._latest_offset:
is_end = True
return file_changed, is_end, event_data
while True:
start_offset = self._summary_file_handler.offset
try:
event_str = self.event_load(self._summary_file_handler)
if event_str is None:
self._summary_file_handler.reset_offset(start_offset)
is_end = True
return file_changed, is_end, event_data
if len(event_str) > MAX_EVENT_STRING:
logger.warning(
"file_path: %s, event string: %d exceeds %d and drop it.",
self._summary_file_handler.file_path, len(event_str),
MAX_EVENT_STRING)
continue
field_list, tensor_value_list = self._event_decode(event_str)
for field, tensor_value in zip(field_list, tensor_value_list):
event_data[field] = tensor_value
logger.debug("Parse summary file offset %d, file path: %s.",
self._summary_file_handler.offset, file_path)
return file_changed, is_end, event_data
except exceptions.CRCLengthFailedError as ex:
self._summary_file_handler.reset_offset(start_offset)
is_end = True
logger.warning(
"Check crc failed and reset offset, file_path=%s, offset=%s. Detail: %r.",
self._summary_file_handler.file_path,
self._summary_file_handler.offset, str(ex))
return file_changed, is_end, event_data
except Exception as ex:
# Note: If an unknown error occurs, we will set the offset to the end of this file,
# which is equivalent to stopping parsing this file. We do not delete the current job
# and retain the data that has been successfully parsed.
self._summary_file_handler.reset_offset(new_size)
# Notice: If the current job is the latest one in the loader pool and the job is deleted,
# the job goes into an infinite cycle of load-fail-delete-reload-load-fail-delete.
# We need to prevent this infinite loop.
logger.error(
"Parse summary file failed, will set offset to the file end. file_path: %s, "
"offset: %d, detail: %s.", file_path,
self._summary_file_handler.offset, str(ex))
logger.exception(ex)
raise UnknownError(str(ex))
finally:
self._latest_offset = self._summary_file_handler.offset