7.4.1 Memory Leak #981
Comments
|
Hello all, Python setup'os = Windows-10-10.0.19045-SP0'
'python = 3.11.1 (tags/v3.11.1:a7a450f, Dec 6 2022, 19:58:39) [MSC v.1934 64 bit (AMD64)]'
'asammdf = 7.4.2'
'numpy = 1.26.4'Dummy codeimport numpy as np
from asammdf import MDF
class Container:
def __init__(self):
self.data = np.empty((1,))
self.data = np.NAN
# not a @staticmethod as in the real code some other actions are performed
def get_data(self, fd):
with MDF(name=fd, memory='minimal') as file:
data = file.extract_bus_logging(
database_files={'CAN': [(data_base, can_channel)], }
).to_dataframe()['sig_1', 'sig_2', 'sig_n'].to_numpy(dtype=float)
return data
fds = ['file_0.mf4', 'file_1.mf4', 'file_n.mf4']
obj = Container()
for f in fds:
if np.isnan(obj.data).any():
obj.data = obj.get_data(f)
else:
np.append(
arr=obj.data,
values=obj.get_data(f),
axis=0
)DescriptionEach of the .mf4 files is about 900 MiB in size. The data extracted from each of these files and stored to the Workaround try: Threading/MultiprocessingSince it is known that Python does not necessarily release all the memory back to the OS once variables get cleared, I thought to try off-loading the for f in fds:
if np.isnan(obj.data).any():
with concurrent.futures.ThreadPoolExecutor() as executor:
tmp_data = executor.map(obj.get_data, [f])
obj.data = tmp_data.__next__()
del tmp_data
else:
with concurrent.futures.ThreadPoolExecutor() as executor:
tmp_data = executor.map(obj.get_data, [f])
np.append(
arr=obj.data,
values=tmp_data.__next__(),
axis=0
)
del tmp_dataUnfortunately, even that did not do the trick. The memory leak is unchanged. Trying a Workaround try: copy.deepcopy()The intention behind utilizing with MDF(name=fd, memory='minimal') as file:
data = deepcopy(
file.extract_bus_logging(
database_files={'CAN': [(data_base, can_channel)], }
)
)Side note: Executing a with MDF(name=fd, memory='minimal') as file:
data = deepcopy(
file.extract_bus_logging(
database_files={'CAN': [(data_base, can_channel)], }
).to_dataframe()['sig_1', 'sig_2', 'sig_n'].to_numpy(dtype=float)
) |
|
Is this better? import numpy as np
from asammdf import MDF
class Container:
def __init__(self):
self.data = np.empty((1,))
self.data = np.NAN
# not a @staticmethod as in the real code some other actions are performed
def get_data(self, fd):
with MDF(name=fd, memory='minimal') as file:
data_mdf = file.extract_bus_logging(
database_files={'CAN': [(data_base, can_channel)], }
)
data = data_mdf.to_dataframe()['sig_1', 'sig_2', 'sig_n'].to_numpy(dtype=float)
data_mdf.close()
return data
fds = ['file_0.mf4', 'file_1.mf4', 'file_n.mf4']
obj = Container()
for f in fds:
if np.isnan(obj.data).any():
obj.data = obj.get_data(f)
else:
np.append(
arr=obj.data,
values=obj.get_data(f),
axis=0
) |
|
Unfortunately, assigning the numpy data to a new variable and closing the |
|
In a desperate attempt to try and bypass the memory leak issue I tried wrapping the # extract_bus_logging_wrapper.py
# Got turned into a standalone executable utilizing: pyinstaller --onefile extract_bus_logging_wrapper.py
# (pyinstaller v6.6.0)
def arg_parser():
parser = argparse.ArgumentParser()
for arg in ['fp', 'bus', 'db', 'ch']:
parser.add_argument(arg, type=str)
args = parser.parse_args()
return args.fp, args.bus, args.db, args.ch
def extract_bus_logging_wrapper(_fp, _bus, _db, _ch):
_db_dict = {_bus: [(_db, int(_ch))]}
return pickle.dumps(MDF(_fp).extract_bus_logging(_db_dict).to_dataframe())
if __name__ == '__main__':
fp, bus, db, ch = arg_parser()
sys.stdout.buffer.write(extract_bus_logging_wrapper(fp, bus, db, ch))# main.py
try:
result = pickle.loads(
subprocess.check_output(
args=['extract_bus_logging_wrapper.exe', str('log.mf4'), str('CAN'), str('data_base.dbc'), str('2')]
)
)
except subprocess.CalledProcessError as e:
print(f'Error running {exe_path}: {e}')Note: Calling the result = pickle.loads(extract_bus_logging_wrapper(str('log.mf4'), str('CAN'), str('data_base.dbc'), str('2'))) |
|
Why do you think you would get different result with pyinstaller? It's still python running |
|
I hoped that the .exe gets terminated for good once done with it's execution. Just like what one would manually do when terminating the execution in the IDE. Doing the latter releases the blocked memory. I am not that deep into the entire pyinstaller-game, though. |
Python version 3.11.7
When opening and closing multiple MF4 files, there is a memory leak, which causes asammdf to allocate a lot of memory, which is never freed.
Here you can find a tracemalloc output (tracing generates a lot of overhead, so only a few traces are shown).
The text was updated successfully, but these errors were encountered: