marchostau opened a new issue, #36126:
URL: https://github.com/apache/arrow/issues/36126
### Describe the bug, including details regarding any error messages,
version, and platform.
I'm trying to put dataframes of different types in multiprocessing shared
memory. When I'm using dataframes that contain python types I've no problem,
but when I use pyarrow types I've problems closing the Shared Memories. I've a
function to put a dataframe to the Shared Memory and another to receive this
dataframe. I have two problems when I'm using Arrow types like string[pyarrow]
inside a dataframe:
When I call the close on the sm_put and after receiving the dataframe I call
the close on the sm_get I get instantly an error, that causes the code to stop
executing: sm_get.close() -> self.close() -> self._buf.release() ->
BufferError: memoryview has 1 exported buffer.
When I call the close on the sm_put, and I don't call the close on the
sm_get, I've no problem instantly, but I get a warning after a while (that
surprises me because I don't call any function self.close()): self.close() ->
self._buf.release() -> BufferError: memoryview has 1 exported buffer.
My functions to put the dataframe to the shared memory and to receive the
dataframe from shared memory are these:
`
def put_df(data, logTime, logSize, usingArrow=False):
if(usingArrow):
table = (pyarrow.Table.from_pandas(data)).combine_chunks()
record_batch = table.to_batches(max_chunksize = sys.maxsize)[0]
else:
record_batch = pa.RecordBatch.from_pandas(data)
# Determine the size of buffer to request from the shared memory
mock_sink = pa.MockOutputStream()
stream_writer = pa.RecordBatchStreamWriter(mock_sink,
record_batch.schema)
stream_writer.write_batch(record_batch)
stream_writer.close()
data_size = mock_sink.size()
sm_put = SharedMemory(create=True, size=data_size)
buffer = pa.py_buffer(sm_put.buf)
# Write the PyArrow RecordBatch to SharedMemory
stream = pa.FixedSizeBufferWriter(buffer)
stream_writer = pa.RecordBatchStreamWriter(stream, record_batch.schema)
stream_writer.write_batch(record_batch)
stream_writer.close()
del stream_writer
del stream
del buffer
if(usingArrow):
del table
del record_batch
sm_put.close()
return sm_put.name
def get_df(sm_get_name, logTime, logSize, usingArrow=False):
sm_get = SharedMemory(name = sm_get_name, create = False)
buffer = pa.BufferReader(sm_get.buf)
# Convert object back into an Arrow RecordBatch
reader = pa.RecordBatchStreamReader(buffer)
record_batch = reader.read_next_batch()
# Convert back into Pandas
if(usingArrow):
data = record_batch.to_pandas(types_mapper = pd.ArrowDtype)
else:
data = record_batch.to_pandas()
del buffer
del reader
del record_batch
sm_get.close()
sm_get.unlink()
return data
if __name__ == '__main__':
names = [ "0", "1"]
types = { "0": "string[pyarrow]", "1": "string[pyarrow]" }
df = pd.read_csv(open(file, "r"), index_col=False, header=None,
names=names, dtype=types, engine="c")
sm_get_name = put_df(df)
data = get_df(sm_get_name=sm_get_name)]
`
As you can see in the next images the del table in the put_df has no effect.
Also in the put_df I think the del on the record_batch would have to decrement
the memory used.

### Component(s)
Python
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]