Perf improv

[ahopkins]

I was playing around with some benchmarking and started to notice that we were not realizing the performance bump that I was expecting.

v20.12

Running 10s test @ http://localhost:3000
  12 threads and 400 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     3.60ms    3.62ms  77.03ms   92.43%
    Req/Sec    10.73k     2.71k   26.83k    79.17%
  1281439 requests in 10.05s, 144.20MB read
Requests/sec: 127460.90
Transfer/sec:     14.34MB

master

Running 10s test @ http://localhost:3000
  12 threads and 400 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     3.63ms    4.25ms  87.71ms   92.34%
    Req/Sec    11.33k     2.63k   52.29k    84.13%
  1343733 requests in 10.08s, 131.99MB read
Requests/sec: 133339.69
Transfer/sec:     13.10MB

I realized that this is mainly due to some minor issues in handle_request. With these small changes:

perf-improv

Running 10s test @ http://localhost:3000
  12 threads and 400 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     3.58ms    4.12ms  88.78ms   92.44%
    Req/Sec    11.49k     2.40k   50.72k    77.39%
  1376168 requests in 10.10s, 135.18MB read
Requests/sec: 136254.12
Transfer/sec:     13.38MB

There are some more gains to be had here, but not until 21.6 and a more thorough overhaul of this method.

[codecov]

Codecov Report

Merging #2074 (f9bb218) into master (8a2ea62) will decrease coverage by 0.018%.
The diff coverage is 100.000%.

@@              Coverage Diff              @@
##            master     #2074       +/-   ##
=============================================
- Coverage   92.217%   92.200%   -0.018%     
=============================================
  Files           38        38               
  Lines         3482      3487        +5     
  Branches       583       584        +1     
=============================================
+ Hits          3211      3215        +4     
  Misses         184       184               
- Partials        87        88        +1     

Impacted Files	Coverage Δ
sanic/app.py	92.213% <100.000%> (-0.063%)	⬇️
sanic/request.py	97.712% <100.000%> (+0.085%)	⬆️
sanic/router.py	94.521% <100.000%> (-0.146%)	⬇️
sanic/http.py	77.491% <0.000%> (-0.369%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 8a2ea62...f9bb218. Read the comment docs.

[ashleysommer]

Looks like a good change, makes the code easier to follow and is a bit faster.

I still want to look further into some performance comparisons, there's some other things I want to check out (eg, last time I tested the streaming server changes, the performance was exactly the same between uvloop and plain asyncio).

[ahopkins]

Agreed. I think a large focus for the next three months is to scrutinize every line that is executed in the request/response cycle.