Fixed StreamReader._read_nowait

[dalazx]

What do these changes do?

StreamReader._read_nowait was returning a single chunk from self._buffer even if a much larger chunk was requested and available. This behavior caused performance regressions.

Simple test:

import asyncio
import time

import aiohttp


async def measure(chunk_size, timeout):
    start_time = time.time()

    file_size = 0
    chunks_num = 0

    async with aiohttp.ClientSession() as session:
        async with session.get('http://localhost:8000/garbage') as response:
            response.content._b_limit = 20 * 1024 * 1024
            while True:
                chunk = await response.content.read(chunk_size)
                if not chunk:
                    break
                file_size += len(chunk)
                chunks_num += 1
                await asyncio.sleep(timeout)

    avg_chunk_size = file_size // chunks_num
    total_time = time.time() - start_time
    print(
        'req chunk size:', chunk_size,
        'avg chunk size:', avg_chunk_size,
        'timeout:', timeout,
        'total time:', total_time)


asyncio.get_event_loop().run_until_complete(measure(10 * 1024 * 1024, 0.1))
asyncio.get_event_loop().run_until_complete(measure(10 * 1024 * 1024, 0.5))

Before:

req chunk size: 10485760 avg chunk size: 202427 timeout: 0.1 total time: 53.54 s
req chunk size: 10485760 avg chunk size: 171335 timeout: 0.5 total time: 308.13 s

After:

req chunk size: 10485760 avg chunk size: 9532509 timeout: 0.1 total time: 1.18 s
req chunk size: 10485760 avg chunk size: 9532509 timeout: 0.5 total time: 5.58 s

Are there changes in behavior for the user?

Related issue number

Checklist

• I think the code is well written
• Unit tests for the changes exist
• Documentation reflects the changes
• Add yourself to CONTRIBUTORS.txt
• Add a new entry to CHANGES.rst
  • Choose any open position to avoid merge conflicts with other PRs.
  • Add a link to the issue you are fixing (if any) using #isuue_number format at the end of changelog message. Use Pull Request number if there are no issues for PR or PR covers the issue only partially.

[codecov-io]

Current coverage is 98.53% (diff: 100%)

Merging #1297 into 1.0 will increase coverage by <.01%

@@                1.0      #1297   diff @@
==========================================
  Files            29         29          
  Lines          6545       6553     +8   
  Methods           0          0          
  Messages          0          0          
  Branches       1095       1097     +2   
==========================================
+ Hits           6449       6457     +8   
  Misses           45         45          
  Partials         51         51          

Powered by Codecov. Last update 3ea78fb...a951513

[dalazx]

@jettify I guess I found a way to speed up thing further in our particular use case. In the example above there is this line response.content._b_limit = 20 * 1024 * 1024 that changes the limit in FlowControlStreamReader and it makes a difference in the amount of data buffered in the background (64k by default which is too little). The bad thing is that it is not exposed (I have not found the way to publicly set it to whatever I want). Next step is to expose it somehow in a follow-up PR.

[asvetlov]

Thanks!

[lock]

This thread has been automatically locked since there has not been
any recent activity after it was closed. Please open a new issue for
related bugs.

If you feel like there's important points made in this discussion,
please include those exceprts into that new issue.