使用k6压测web服务器 http/1.1 和 http/2.0
2024-02-26 15:13:18

想试一下 http/1.1 和 http/2.0 有啥区别，正好又想起前段时间了解到的一个压测工具 k6，这不正好,拿 k6 来压一下试试。

安装 k6

官方文档在这里：https://grafana.com/docs/k6/latest/get-started/installation/

使用 go 编写服务端

创建 http 和 https 服务端。在创建 https 服务端时，还需要配置证书，参考 使用golang创建http2和h2c服务端

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
package main

import (
	"fmt"
	"log"
	"net/http"
	"sync"

	"golang.org/x/net/http2"
)

func main() {
	handler := http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		fmt.Fprintln(w, "Protocol: ", r.Proto)
		log.Println("Protocol: ", r.Proto)
	})

	var wg sync.WaitGroup

	h1s := &http.Server{
		Addr:    "0.0.0.0:8080",
		Handler: handler,
	}

	log.Println("start http1.1 server without tls server on: 8080") // 这里创建的是 http/1.1 的服务器，和另外一篇文章是有区别。
	wg.Add(1)
	go func() {
		defer wg.Done()
		// err := h1s.ListenAndServeTLS("mydomain.com.crt", "mydomain.com.key")
		err := h1s.ListenAndServe()
		log.Fatal(err)
	}()

	h2s := &http.Server{
		Addr:    "0.0.0.0:8090",
		Handler: handler,
	}
	http2.ConfigureServer(h2s, &http2.Server{})
	log.Println("start http2 server with tls on: 8090")
	wg.Add(1)
	go func() {
		defer wg.Done()
		err := h2s.ListenAndServeTLS("mydomain.com.crt", "mydomain.com.key")
		log.Fatal(err)
	}()

	h2c := &http.Server{
		Addr:    "0.0.0.0:9000",
		Handler: handler,
	}
	http2.ConfigureServer(h2c, &http2.Server{})
	log.Println("start http2 server without tls on: 9000")
	wg.Add(1)
	go func() {
		defer wg.Done()
		err := h2c.ListenAndServe()
		log.Fatal(err)
	}()

	wg.Wait()
}

编写 k6 脚本

k6 可以通过命令行参数来配置参数，也支持使用配置文件。由于 k6 支持使用 k6 new 生成一个默认的模板，所以使用配置文件是很方便的。

生成一个配置文件

1
k6 new

编辑 script.js

将 http.get('http://test.k6.io'); 修改为 http.get("https://localhost:8090"); 这里按需填写地址即可。
由于我们服务端设置了证书，而 k6 是没有导入证书的，所以需要设置跳过证书验证。

1
2
3
4
5
6
7
8
export const options = {
  // A number specifying the number of VUs to run concurrently.
  vus: 8,
  // A string specifying the total duration of the test run.
  duration: "10s",
  insecureSkipTLSVerify: true,
  // The following section contains c
}

压测

执行命令 k6 run script.js

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
root@kusaka-virtual-machine:~/test# k6 run script.js 

          /\      |‾‾| /‾‾/   /‾‾/   
     /\  /  \     |  |/  /   /  /    
    /  \/    \    |     (   /   ‾‾\  
   /          \   |  |\  \ |  (‾)  | 
  / __________ \  |__| \__\ \_____/ .io

     execution: local
        script: script.js
        output: -

     scenarios: (100.00%) 1 scenario, 8 max VUs, 40s max duration (incl. graceful stop):
              * default: 8 looping VUs for 10s (gracefulStop: 30s)


     data_received..................: 2.6 MB 255 kB/s
     data_sent......................: 1.4 MB 144 kB/s
     http_req_blocked...............: avg=3.47µs   min=227ns    med=404ns    max=17.53ms p(90)=513ns    p(95)=544ns   
     http_req_connecting............: avg=97ns     min=0s       med=0s       max=1.33ms  p(90)=0s       p(95)=0s      
     http_req_duration..............: avg=1.97ms   min=138.6µs  med=1.42ms   max=77.28ms p(90)=3.82ms   p(95)=5.19ms  
       { expected_response:true }...: avg=1.97ms   min=138.6µs  med=1.42ms   max=77.28ms p(90)=3.82ms   p(95)=5.19ms  
     http_req_failed................: 0.00%  ✓ 0           ✗ 39626
     http_req_receiving.............: avg=566.71µs min=9.44µs   med=202.58µs max=44.77ms p(90)=1.4ms    p(95)=2.21ms  
     http_req_sending...............: avg=251.31µs min=24.39µs  med=41.09µs  max=39.36ms p(90)=407.78µs p(95)=881.68µs
     http_req_tls_handshaking.......: avg=2.48µs   min=0s       med=0s       max=16.76ms p(90)=0s       p(95)=0s      
     http_req_waiting...............: avg=1.15ms   min=0s       med=856.01µs max=29.65ms p(90)=2.32ms   p(95)=3.07ms  
     http_reqs......................: 39626  3962.247921/s
     iteration_duration.............: avg=1.99ms   min=165.25µs med=1.5ms    max=46.3ms  p(90)=3.89ms   p(95)=5.2ms   
     iterations.....................: 39626  3962.247921/s
     vus............................: 8      min=8         max=8  
     vus_max........................: 8      min=8         max=8  


running (10.0s), 0/8 VUs, 39626 complete and 0 interrupted iterations
default ✓ [======================================] 8 VUs  10s

比较 http/1.1 和 http/2.0

分别访问 8090 端口和 8080 端口。

我应该是用错方式了，测出来使用 http/1.1 和 http/2.0 几乎没有区别。第一次是访问的 8090 端口，服务端打印出来的协议是 HTTP/2.0，第二次访问的是 8080 端口，服务端打印出来的协议是 HTTP/1.1，但是可以看到，http 请求的数量都在 3800 左右，几乎没有区别。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
root@kusaka-virtual-machine:~/test# k6 run script.js 

          /\      |‾‾| /‾‾/   /‾‾/   
     /\  /  \     |  |/  /   /  /    
    /  \/    \    |     (   /   ‾‾\  
   /          \   |  |\  \ |  (‾)  | 
  / __________ \  |__| \__\ \_____/ .io

     execution: local
        script: script.js
        output: -

     scenarios: (100.00%) 1 scenario, 8 max VUs, 40s max duration (incl. graceful stop):
              * default: 8 looping VUs for 10s (gracefulStop: 30s)


     data_received..................: 2.5 MB 247 kB/s
     data_sent......................: 1.4 MB 139 kB/s
     http_req_blocked...............: avg=3.89µs   min=238ns    med=421ns    max=39.68ms  p(90)=521ns   p(95)=552ns   
     http_req_connecting............: avg=77ns     min=0s       med=0s       max=658.32µs p(90)=0s      p(95)=0s      
     http_req_duration..............: avg=2.01ms   min=137.91µs med=1.5ms    max=68.7ms   p(90)=3.98ms  p(95)=5.16ms  
       { expected_response:true }...: avg=2.01ms   min=137.91µs med=1.5ms    max=68.7ms   p(90)=3.98ms  p(95)=5.16ms  
     http_req_failed................: 0.00%  ✓ 0           ✗ 38321
     http_req_receiving.............: avg=658.75µs min=8.71µs   med=257.49µs max=33.57ms  p(90)=1.71ms  p(95)=2.5ms   
     http_req_sending...............: avg=256.06µs min=24.69µs  med=55.3µs   max=38.98ms  p(90)=466.9µs p(95)=859.19µs
     http_req_tls_handshaking.......: avg=2.72µs   min=0s       med=0s       max=39.14ms  p(90)=0s      p(95)=0s      
     http_req_waiting...............: avg=1.1ms    min=0s       med=837.62µs max=28.51ms  p(90)=2.21ms  p(95)=2.87ms  
     http_reqs......................: 38321  3831.506661/s
     iteration_duration.............: avg=2.06ms   min=165.5µs  med=1.59ms   max=47.18ms  p(90)=4.07ms  p(95)=5.23ms  
     iterations.....................: 38321  3831.506661/s
     vus............................: 8      min=8         max=8  
     vus_max........................: 8      min=8         max=8  


running (10.0s), 0/8 VUs, 38321 complete and 0 interrupted iterations
default ✓ [======================================] 8 VUs  10s
root@kusaka-virtual-machine:~/test# k6 run script.js 

          /\      |‾‾| /‾‾/   /‾‾/   
     /\  /  \     |  |/  /   /  /    
    /  \/    \    |     (   /   ‾‾\  
   /          \   |  |\  \ |  (‾)  | 
  / __________ \  |__| \__\ \_____/ .io

     execution: local
        script: script.js
        output: -

     scenarios: (100.00%) 1 scenario, 8 max VUs, 40s max duration (incl. graceful stop):
              * default: 8 looping VUs for 10s (gracefulStop: 30s)


     data_received..................: 5.2 MB 518 kB/s
     data_sent......................: 3.0 MB 302 kB/s
     http_req_blocked...............: avg=5.5µs    min=1.09µs   med=1.78µs  max=6.61ms  p(90)=2.63µs  p(95)=3.46µs  
     http_req_connecting............: avg=322ns    min=0s       med=0s      max=3.45ms  p(90)=0s      p(95)=0s      
     http_req_duration..............: avg=1.96ms   min=124.23µs med=1.54ms  max=39.52ms p(90)=3.93ms  p(95)=4.95ms  
       { expected_response:true }...: avg=1.96ms   min=124.23µs med=1.54ms  max=39.52ms p(90)=3.93ms  p(95)=4.95ms  
     http_req_failed................: 0.00%  ✓ 0           ✗ 37811
     http_req_receiving.............: avg=120.47µs min=10.45µs  med=24.95µs max=26.94ms p(90)=230.5µs p(95)=387.88µs
     http_req_sending...............: avg=29.93µs  min=4.73µs   med=7.73µs  max=14.72ms p(90)=15.27µs p(95)=56.93µs 
     http_req_tls_handshaking.......: avg=0s       min=0s       med=0s      max=0s      p(90)=0s      p(95)=0s      
     http_req_waiting...............: avg=1.81ms   min=103.29µs med=1.41ms  max=39.48ms p(90)=3.71ms  p(95)=4.68ms  
     http_reqs......................: 37811  3780.428615/s
     iteration_duration.............: avg=2.08ms   min=152.17µs med=1.67ms  max=39.58ms p(90)=4.09ms  p(95)=5.14ms  
     iterations.....................: 37811  3780.428615/s
     vus............................: 5      min=5         max=8  
     vus_max........................: 8      min=8         max=8  


running (10.0s), 0/8 VUs, 37811 complete and 0 interrupted iterations
default ✓ [======================================] 8 VUs  10s

k6 的不足

k6 目前不支持压测非加密的http2 server(H2C)，也就是上面代码中的 9000 端口对应的服务端。如果使用 k6 去测试，那么只能使用 HTTP/1.1 的协议，无法像 curl 那样使用额外的参数 ``–http2-prior-knowledge` 去指定 HTTP/2.0 协议。

相关的讨论可以看这里。https://github.com/grafana/k6/issues/970 ，早在 2019 年就已经在讨论了，不过看起来目前还是没有办法在没有 breaking change 的情况下引入对 H2C 的支持。