Basic HTTP Service performance test (Python vs Golang)

Recently learning Golang, always want to experience the next concurrency in the end have more, will I big python strong how much. Learned the official tutorial HTTP service, using the Performance test Tool WRK test, found the result is very surprising ~

WRK can refer to my blog, there are basic usage notes: http://blog.yuanzhaoyi.cn/2018/01/12/test.html

Test command: wrk-t10-d1m-c200 http://127.0.0.1:8080

Meaning: 10 thread, concurrent 200 link, lasts 1 minutes

The HTTP service returns basic: "Hello World", there should be no IO blocking

Python Standard library Basehttprequesthandler implementation:

 fromHttp.serverImportBasehttprequesthandler fromUrllibImportParseclassGetHandler (basehttprequesthandler):defDo_get (self): message="Hello World"Self.send_response (200) self.end_headers () Self.wfile.write (Message.encode ('Utf-8'))if __name__=='__main__':     fromHttp.serverImportHttpserver Server= Httpserver (('localhost', 8080), GetHandler)Print('starting server, use <Ctrl-C> to stop') Server.serve_forever ()

Result: There are only 282 responses per second, and the longer the test time, the less

Because yes but the process, single threaded, this data should be good, although Gil in IO blocking will release threads, but also a bit of performance consumption

Running 1m Test @ http://127.0.0.1:8080  TenThreads and $connections Thread Stats Avg Stdev Max+/-Stdev Latency2.05ms6.73ms265.58ms98.90%Req/sec107.11    103.19     1.05k84.08%16959Requestsinch 1.00m,1. 65MB Read Socket errors:connect0, read19024,Write  -, timeout0Requests/SEC:282.21Transfer/SEC: -.11KB

Asynchronous framework for convenience, we first use the twisted-based event Loop tornado:

ImportTornado.ioloopImportTornado.webclassMainHandler (tornado.web.RequestHandler):defGet (self): Self.write ("Hello, World")if __name__=="__main__": Application=Tornado.web.Application ([(R"/", MainHandler),]) Application.listen (8080) tornado.ioloop.IOLoop.current (). Start ()

Result: More than 1300 responses per second, significantly better

Running 1m Test @ http://127.0.0.1:8080  TenThreads and $connections Thread Stats Avg Stdev Max+/-Stdev Latency147.44ms $.17ms467.54ms86.25%Req/sec141.40     57.52   202.00     65.17%81818Requestsinch 1.00m, -. 15MB Read Socket errors:connect0, read1,Write 0, timeout0Requests/SEC:1361.25Transfer/SEC:275.17KB

Python3 began to support the native co-process to handle the event loop, although Tornado also support, but for convenience, directly with the fastest sanic test it

 fromSanicImportSanic fromSanic.responseImportJsonapp=sanic () @app. Route ("/") AsyncdefTest (Request):returnJSON ({"Hello":" World"})if __name__=="__main__": App.run (Host="0.0.0.0", Debug=false, port=8080)

Result: The number of responses per second is more than 4,400, it looks good.

Running 1m test @ http://127.0.0.1:8080 and    connections  Thread Stats   Avg      Stdev     Max   +/- Stdev    Latency    45.59ms   16.91ms 255.88ms   71.70%    Req/ Sec   443.64    111.85     0.89k    68.56%   in 1.00m,. 09MB Read    , write 0, timeout 0Requests/sec:   4408.87Transfer/sec:    546.80KB

Recently studied Golang, and conducted basic tests on HTTP services based on the official guidelines:

Package MainImport (    "FMT"    "Log"    "net/http") Type Hello struct{}func (H Hello) servehttp (w http. Responsewriter, R*http. Request) {fmt. Fprint (W,"Hello world!")}func Main () {h:=hello{} ERR:= http. Listenandserve ("localhost:8080", h)ifErr! =Nil {log. Fatal (Err)}}

The result was an eye-opener: The number of responses per second reached 35365, and the Python service was not a magnitude

Running 1m Test @ http://127.0.0.1:8080  TenThreads and $connections Thread Stats Avg Stdev Max+/-Stdev Latency7.26ms9.46ms About.93ms93.36%Req/sec3.56k1.10k +.98k74.96%2125366Requestsinch 1.00m,261. 47MB readrequests/SEC:35365.98Transfer/SEC:4.35MB

In summary, in the test, the memory is not significantly increased, there must be no action, just return a string. But Golang's CPU usage has grown significantly, sanic services have grown almost as well, while the rest of the CPU usage has grown a little, but Python's services should all be CPU-centric. But when the CPU ratio is similar, Golang's responsiveness is significantly better. Specific reasons to think, in addition to the core use of the CPU difference, that is true parallel implementation, it seems nothing. Python's asynchronous service and Golang services should be based on the event loop to achieve the scheduling of the process, of course, the implementation of the method must be very different, specifically to continue to learn. But Golang's innate concurrency support is really good for this optimization.

These tests are all based on curiosity, relatively simple and not rigorous enough, but I think we can illustrate some differences. If you find any problems, please leave a message.

Basic HTTP Service performance test (Python vs Golang)