使用 asyncio 協程並行運行函數？

Question

我有以下代碼從數據庫（ read_db ）讀取數據並將數據寫入鑲木地板文件（ data.to_parquet ）。 兩個 I/O 操作都需要一段時間才能運行。

def main():
    while id < 1000:
       logging.info(f'reading - id: {id}')
       data = read_db(id) # returns a dataframe

       logging.info(f'saving - id: {id}')
       data.to_parquet(f'{id}.parquet')
       logging.info(f'saved - id: {id}')

       id += 1
       

它很慢，所以我希望read_db(n+1)和to_parquet(n)同時運行。 我需要保持id的每個步驟按順序完成（ read_db(n+1)需要在read_db(n)之后運行， data.to_parquet(n+1)在data.to_parquet(n)之后運行。）。 這是異步版本

def async_wrap(f):
    @wraps(f)
    async def run(*args, loop=None, executor=None, **kwargs):
        if loop is None:
            loop = asyncio.get_event_loop()
        p = partial(f, *args, **kwargs)
        return await loop.run_in_executor(executor, p)
    return run

async def main():
    read_db_async = async_wrap(read_db)
    while id < 1000:
       logging.info(f'reading - id: {id}')
       data = await read_db_async(id) # returns a dataframe

       logging.info(f'saving - id: {id}')
       to_parquet_async = async_wrap(data.to_parquet)
       await data.to_parquet(f'{id}.parquet')
       logging.info(f'saved - id: {id}')

       id += 1

asyncio.get_event_loop().run_until_complete(main())

我除了看到一些亂序的日志：

reading - id: 1
saving - id: 1      (saving 1 and reading 2 run in parallel)
reading - id: 2
saved - id: 1
saving - id: 2
reading - id: 3
saved - id: 2
.....

但是，實際的日志和同步代碼是一樣的嗎？

reading - id: 1
saving - id: 1
saved - id: 1
reading - id: 2
saving - id: 2
saved - id: 2
reading - id: 3
.....

Answer 1

您可以使用 collect 或等效項使gather read_db(n+1)和to_parquet(n)同時運行：

async def main():
    read_db_async = async_wrap(read_db)
    prev_to_parquet = asyncio.sleep(0)  # no-op

    for id in range(1, 1000):
        data, _ = await asyncio.gather(read_db_async(id), prev_to_parquet)
        to_parquet_async = async_wrap(data.to_parquet)
        prev_to_parquet = to_parquet_async(f'{id}.parquet')

    await prev_to_parquet