A microbenchmark for asyncio, uvloop and trio. Uses multiple worker processes to keep the game fair because asyncio secretly runs worker threads.
This program runs a minimal HTTP server implemented for asyncio and trio, in each using direct socket API as well as the highest level streams API.
Asyncio's "low level" protocol API is not tested directly but the streams API uses it internally. Asyncio is also tested with uvloop, if available.
pip3 install git+https://github.com/Tronic/async-http-benchmark.git
Install wrk for fully automatic benchmarking, or run some other tool while ahbench is waiting.
Optional deps: pip3 install trio uvloop
Run all tests: python3 -m ahbench or
run only matching tests: python3 -m ahbench uv (or any other searchword).
Asyncio and uvloop usually die with random exceptions when KeyboardInterrupted after each test. I couldn't fix that.
Pull requests are welcome.
# python3 -m ahbench
asyncio_sockets >>> wrk -t8 -c100 http://127.0.0.1:62157/
Running 10s test @ http://127.0.0.1:62157/
8 threads and 100 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 2.01ms 1.51ms 13.49ms 66.47%
Req/Sec 6.33k 3.24k 15.12k 87.38%
503829 requests in 10.00s, 42.76MB read
Requests/sec: 50361.09
Transfer/sec: 4.27MB
asyncio_streams >>> wrk -t8 -c100 http://127.0.0.1:62255/
Running 10s test @ http://127.0.0.1:62255/
8 threads and 100 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 1.69ms 1.25ms 18.65ms 62.67%
Req/Sec 7.55k 4.12k 18.13k 60.37%
604546 requests in 10.10s, 51.31MB read
Requests/sec: 59835.56
Transfer/sec: 5.08MB
uvloop_sockets >>> wrk -t8 -c100 http://127.0.0.1:62353/
Running 10s test @ http://127.0.0.1:62353/
8 threads and 100 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 1.04ms 647.89us 14.98ms 91.18%
Req/Sec 11.96k 1.82k 19.72k 59.18%
958882 requests in 10.10s, 81.39MB read
Requests/sec: 94932.63
Transfer/sec: 8.06MB
uvloop_streams >>> wrk -t8 -c100 http://127.0.0.1:62451/
Running 10s test @ http://127.0.0.1:62451/
8 threads and 100 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 1.22ms 662.25us 22.92ms 92.98%
Req/Sec 9.99k 1.24k 11.96k 67.20%
803490 requests in 10.10s, 68.20MB read
Requests/sec: 79526.86
Transfer/sec: 6.75MB
Exception ignored in: <function WeakValueDictionary.__init__.<locals>.remove at 0x10dd6cdd0>
Traceback (most recent call last):
File "/.../lib/python3.7/weakref.py", line 117, in remove
_atomic_removal(d, wr.key)
KeyboardInterrupt
trio_sockets >>> wrk -t8 -c100 http://127.0.0.1:62550/
Running 10s test @ http://127.0.0.1:62550/
8 threads and 100 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 2.17ms 6.68ms 124.43ms 98.75%
Req/Sec 8.34k 2.69k 16.40k 79.92%
659649 requests in 10.01s, 55.99MB read
Requests/sec: 65923.66
Transfer/sec: 5.60MB
trio_streams >>> wrk -t8 -c100 http://127.0.0.1:62648/
Running 10s test @ http://127.0.0.1:62648/
8 threads and 100 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 2.61ms 5.66ms 108.00ms 97.56%
Req/Sec 6.15k 1.40k 10.63k 86.42%
488273 requests in 10.01s, 41.44MB read
Requests/sec: 48792.63
Transfer/sec: 4.14MB