Python异步/等待下载URL列表

时间:2018-01-14 22:32:52

标签: python asynchronous python-asyncio aiohttp

我正在尝试从FTP服务器下载超过30,000个文件,并且在使用异步IO进行一些谷歌搜索之后似乎是一个好主意。但是,下面的代码无法下载任何文件并返回超时错误。我真的很感激任何帮助!谢谢!

class pdb:
    def __init__(self):
        self.ids = []
        self.dl_id = []
        self.err_id = []


    async def download_file(self, session, url):
        try:
            with async_timeout.timeout(10):
                async with session.get(url) as remotefile:
                    if remotefile.status == 200:
                        data = await remotefile.read()
                        return {"error": "", "data": data}
                    else:
                        return {"error": remotefile.status, "data": ""}
        except Exception as e:
            return {"error": e, "data": ""}

    async def unzip(self, session, work_queue):
        while not work_queue.empty():
            queue_url = await work_queue.get()
            print(queue_url)
            data = await self.download_file(session, queue_url)
            id = queue_url[-11:-7]
            ID = id.upper()
            if not data["error"]:
                saved_pdb = os.path.join("./pdb", ID, f'{ID}.pdb')
                if ID not in self.dl_id:
                    self.dl_id.append(ID)
                with open(f"{id}.ent.gz", 'wb') as f:
                    f.write(data["data"].read())
                with gzip.open(f"{id}.ent.gz", "rb") as inFile, open(saved_pdb, "wb") as outFile:
                    shutil.copyfileobj(inFile, outFile)
                os.remove(f"{id}.ent.gz")
            else:
                self.err_id.append(ID)

    def download_queue(self, urls):
        loop = asyncio.get_event_loop()
        q = asyncio.Queue(loop=loop)
        [q.put_nowait(url) for url in urls]
        con = aiohttp.TCPConnector(limit=10)
        with aiohttp.ClientSession(loop=loop, connector=con) as session:
            tasks = [asyncio.ensure_future(self.unzip(session, q)) for _ in range(len(urls))]
            loop.run_until_complete(asyncio.gather(*tasks))
        loop.close()

如果我删除try部分错误消息:

  

追踪(最近的呼叫最后):
  文件“test.py”,第111行,中       x.download_queue(URL)的
  在download_queue中文件“test.py”,第99行       loop.run_until_complete(asyncio.gather(*任务))
  在run_until_complete中输入文件“/home/yz/miniconda3/lib/python3.6/asyncio/base_events.py”,第467行       返回future.result()
  文件“test.py”,第73行,解压缩       data = await self.download_file(session,queue_url)
  在download_file中文件“test.py”,第65行       return {“error”:remotefile.status,“data”:“”}
  在退出中输入文件“/home/yz/miniconda3/lib/python3.6/site- packages / async_timeout / init .py”,第46行       从无提升asyncio.TimeoutError   concurrent.futures._base.TimeoutError

2 个答案:

答案 0 :(得分:5)

tasks = [asyncio.ensure_future(self.unzip(session, q)) for _ in range(len(urls))]
loop.run_until_complete(asyncio.gather(*tasks))

在这里,您开始同时下载所有网址的过程。这意味着你也开始计算所有这些的超时。一旦它是30,000这样的大数字,由于网络/ ram / cpu容量,它在10秒内无法完成。

为避免这种情况,您应该保证协同程序的限制同时启动。通常可以使用像asyncio.Semaphore这样的同步原语来实现这一点。

它看起来像这样:

sem = asyncio.Semaphore(10)

# ...

async def download_file(self, session, url):
    try:
        async with sem:  # Don't start next download until 10 other currently running
            with async_timeout.timeout(10):

答案 1 :(得分:1)

作为@ MikhailGerasimov的信号量方法的替代方法,您可以考虑使用aiostream.stream.map运算符:

from aiostream import stream, pipe

async def main(urls):
    async with aiohttp.ClientSession() as session:
        ws = stream.repeat(session)
        xs = stream.zip(ws, stream.iterate(urls))
        ys = stream.starmap(xs, fetch, ordered=False, task_limit=10)
        zs = stream.map(ys, process)
        await zs

这是使用管道的等效实现:

async def main3(urls):
    async with aiohttp.ClientSession() as session:
        await (stream.repeat(session)
               | pipe.zip(stream.iterate(urls))
               | pipe.starmap(fetch, ordered=False, task_limit=10)
               | pipe.map(process))

您可以使用以下协程进行测试:

async def fetch(session, url):
    await asyncio.sleep(random.random())
    return url

async def process(data):
    print(data)

在此demonstrationdocumentation中查看更多aiostream示例。

免责声明:我是项目维护者。