请教PYODPS多进程读写数据问题

[templaryang]

问题背景：需要读取大约500万行，2个字段的表，然后计算出一个同等记录行数结果集，都用dataframe下载、计算、并上传到Dataphin上，现在性能瓶颈在下载数据（需要3分钟）和上传数据（需要10分钟），计算只需要2分钟完成。

目的：为了改善读写性能，查看PYODPS文档，支持多进程读写数据

1. 支持多进程读取，编写和示例代码几乎一样的代码，但报错

t = odps.get_table('test_table')
n_process = multiprocessing.cpu_count()
with t.open_reader() as reader:
    pd_df = reader.to_pandas(n_process=n_process)

问题：为什么报错？：TypeError: to_pandas() got an unexpected keyword argument 'n_process'

2. 写入表也是支持多进程程写入，但我对比过，比我单进程写入更慢，为什么？
   1. 单进程写入

records = df.values().values.tolist()
o.write_table('test_table', records)

2. 多进程写入

def write_records(session_id, block_id):
    local_session = tunnel.create_upload_session(table.name, upload_id=session_id)
    with local_session.open_record_writer(block_id) as writer:
        maxlen = len(records)
        strlen = int(maxlen/N_WORKERS) * block_id
        endlen = int(maxlen/N_WORKERS) * (block_id + 1)
        if block_id == N_WORKERS - 1:
            endlen = maxlen
        for row in range(strlen, endlen):
            record = table.new_record(records[row])
            writer.write(record)

if __name__ == '__main__':
    records = df.values().values.tolist()
    N_WORKERS = 4
    table = odps.get_table("test_table")
    tunnel = TableTunnel(odps)
    upload_session = tunnel.create_upload_session(table.name)
    session_id = upload_session.id
    pool = Pool(processes=N_WORKERS)
    block_ids = []
    for i in range(N_WORKERS):
        pool.apply_async(write_records, (session_id, i))
        block_ids.append(i)
    pool.close()
    pool.join()
    upload_session.commit(block_ids)

** 问题：多进程比单进程更慢，单进程在大约10分钟完成500万记录的插入，多进程超过20分钟仍未结束（已经设置需要的cpu为4） **