使用不可靠的 inte.net 的 aiohttp 速率限制請求

Question

我正在從具有非常嚴格的速率限制的網站下載內容。 如果我超過 10 個請求/秒，我將被禁止 10 分鍾。 我一直在使用以下代碼來限制 AIOHTTP 的速率：

import time

class RateLimitedClientSession:
    """Rate Limited Client.
    Attributes:
        client (aiohttp.ClientSession): A client to call
        rate_limit (int): Maximum number of requests per second to make
    https://quentin.pradet.me/blog/how-do-you-rate-limit-calls-with-aiohttp.html
    """

    def __init__(self, client, rate_limit):
        self.client = client
        self.rate_limit = rate_limit
        self.max_tokens = rate_limit
        self.tokens = self.max_tokens
        self.updated_at = time.monotonic()
        self.start = time.monotonic()

    async def get(self, *args, **kwargs):
        """Wrapper for ``client.get`` that first waits for a token."""
        await self.wait_for_token()
        return self.client.get(*args, **kwargs)

    async def wait_for_token(self):
        """Sleeps until a new token is added."""
        while self.tokens < 1:
            self.add_new_tokens()
            await asyncio.sleep(0.03) # Arbitrary delay, must be small though.
        self.tokens -= 1

    def add_new_tokens(self):
        """Adds a new token if time elapsed is greater than minimum time."""
        now = time.monotonic()
        time_since_update = now - self.updated_at
        new_tokens = time_since_update * self.rate_limit
        if self.tokens + new_tokens >= 1:
            self.tokens = min(self.tokens + new_tokens, self.max_tokens)
            self.updated_at = now

然后我可以這樣使用它：

from aiohttp import ClientSession, TCPConnector

limit = 9 # 9 requests per second
inputs = ['url1', 'url2', 'url3', ...]
conn = TCPConnector(limit=limit)
raw_client = ClientSession(connector=conn, headers={'Connection': 'keep-alive'})
async with raw_client:
    session = RateLimitedClientSession(raw_client, limit)
    tasks = [asyncio.ensure_future(download_link(link, session)) for link in inputs]
    for task in asyncio.as_completed(tasks):
        await task

async def download_link(link, session):
    async with await session.get(link) as resp:
        data = await resp.read()
        # Then write data to a file

我的問題是代碼將在隨機次數內正常工作，通常在 100 到 2000 之間。然后，由於達到速率限制而退出。 我懷疑這與我的 inte.net 的延遲有關。

例如，想象一個 3 個請求/秒的限制。

SECOND 1:
 + REQ 1
 + REQ 2
 + REQ 3

SECOND 2:
 + REQ 4
 + REQ 5
 + REQ 6

有一點滯后，這可能看起來像

SECOND 1:
 + REQ 1
 + REQ 2

SECOND 2:
+ REQ 3 - rolled over from previous second due to internet speed
+ REQ 4
+ REQ 5
+ REQ 6

然后觸發速率限制。

我該怎么做才能最大程度地減少發生這種情況的可能性？

• 我已經嘗試降低速率限制，它確實可以工作更長的時間，但最終仍會達到速率限制。

• 我也嘗試過以 1/10 秒的間隔觸發每個請求，但這仍然會觸發速率限制（可能是出於不相關的原因？）。

Answer 1

我認為最好的解決方案是將請求分批處理，然后等待丟失的時間。 我不再在 AIOHTTP 周圍使用限速包裝器。

async def download_link(link, session):
    async with await session.get(link) as resp:
        data = await resp.read()
        # Then write data to a file


def batch(iterable, n):
    l = len(iterable)
    for ndx in range(0, l, n):
        yield iterable[ndx:min(ndx + n, l)]

rate_limit = 10
conn = aiohttp.TCPConnector(limit=rate_limit)
client = aiohttp.ClientSession(
    connector=conn, headers={'Connection': 'keep-alive'}, raise_for_status=True)

async with client:
    for group in batch(inputs, rate_limit):
        start = time.monotonic()
        tasks = [download_link(link, client) for link in group]
        await asyncio.gather(*tasks) # If results are needed they can be assigned here
        execution_time = time.monotonic() - start
        # If execution time > 1, requests are essentially wasted, but a small price to pay
        await asyncio.sleep(max(0, 1 - execution_time))