一瞥之见

0%

Python(django & aiohttp) vs Go(Echo) web server压测性能对比

发表于 更新于 分类于

对于一个真实工程稍做简化后分别使用Python(django & aiohttp) & Go(Echo)实现了一版进行压测

背景说明

  • web后端server场景：全局initial时访问了一次数据库user表、形成本地内存的name-map-passwd的dict/map。对于进来的每个请求，检测了必须的参数是否齐全、而后从name-map-passwd检测是否有权限。最后json序列化后lpush到redis(均没有限制redis pool的max size)。
  • python3.7, aiohttp v3.5.4，nginx, 5worker uvloop gunicorn based on .sock file
  • python3.7, django v3.0.2，nginx，uwsgi，5worker based on .sock file
  • go1.13.6，Echo v4.0.0，nginx，5worker based on http port
  • 压测机器为8核16G内存，被压测机器4核16G内存
  • 压测工具使用同事开发的ysab
    • 重要参数说明； -n：Number of multiple requests to make at a time； -r：round；

结果展示

Echo(Go)

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
$ ysab -n 900 -r 100 -u 'http://host:port/sendMsgAll/?userkey=xxx&type=xxx&content=dfhajklshfjkld&tolist=156526xxxxx&username=pd'

Summary:
  Complete requests:		90000
  Failed requests:		0
  Time taken (s):		10.66654527
  Total data size (Byte):	10530000
  Data size/request (Byte):	117
  Max use time (ms):		1276
  Min use time (ms):		6
  Average use time (ms):	103.338
  Requests/sec:			8437.596027753041
  SuccessRequests/sec:		8437.596027753041

Percentage of waiting time (ms):
    10.00%:		34
    25.00%:		54
    50.00%:		86
    75.00%:		131
    90.00%:		180
    95.00%:		221
    99.00%:		335
    99.90%:		1098
    99.99%:		1274


Time detail (ms)
  item		min		mean		max
  dns		0		0		0
  conn		0		0.154		40
  wait		6		103.029		1276
  resp		0		0.035		22

Response Time histogram (code: requests):
  200:		90000

Aiohttp(Python)

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
$ ysab -n 900 -r 100 -u 'http://host:port/sendMsgAll/?userkey=xxx&type=xxx&content=dfhajklshfjkld&tolist=156526xxxxx&username=pd'

Summary:
  Complete requests:		90000
  Failed requests:		1304
  Time taken (s):		35.491095644
  Total data size (Byte):	11401288
  Data size/request (Byte):	126
  Max use time (ms):		1368
  Min use time (ms):		1
  Average use time (ms):	353.329
  Requests/sec:			2535.8473264043932
  SuccessRequests/sec:		2499.105716252934

Percentage of waiting time (ms):
    10.00%:		172
    25.00%:		243
    50.00%:		324
    75.00%:		437
    90.00%:		578
    95.00%:		679
    99.00%:		885
    99.90%:		1117
    99.99%:		1368


Time detail (ms)
  item		min		mean		max
  dns		0		0		0
  conn		0		0.182		58
  wait		1		352.967		1368
  resp		0		0.002		12

Response Time histogram (code: requests):
  200:		88696
  502:		1304
Django(Python)
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
$ ysab -n 900 -r 100 -u 'http://host:port/sendMsgAll/?userkey=xxx&type=xxx&content=dfhajklshfjkld&tolist=156526xxxxx&username=pd'

Summary:
	Complete requests:		90000
  Failed requests:		86630
  Time taken (s):		3.070857802
  Total data size (Byte):	15411610
  Data size/request (Byte):	171
  Max use time (ms):		327
  Min use time (ms):		0
  Average use time (ms):	17.152
  Requests/sec:			29307.771900536867
  SuccessRequests/sec:		1097.4132367201025

Percentage of waiting time (ms):
    10.00%:		1
    25.00%:		2
    50.00%:		14
    75.00%:		20
    90.00%:		31
    95.00%:		45
    99.00%:		126
    99.90%:		279
    99.99%:		322


Time detail (ms)
  item		min		mean		max
  dns		0		0		0
  conn		0		0.191		88
  wait		0		7.667		160
  resp		0		0.295		197

Response Time histogram (code: requests):
  200:		3370
  502:		86630

可见，同步django根本不能够承受到-n为900的情况，经过测试，其只能达到-n参数为100的程度，数据如下：

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
Summary:
  Complete requests:		10000
  Failed requests:		0
  Time taken (s):		6.128261992
  Total data size (Byte):	1260000
  Data size/request (Byte):	126
  Max use time (ms):		91
  Min use time (ms):		5
  Average use time (ms):	60.374
  Requests/sec:			1631.7840218081851
  SuccessRequests/sec:		1631.7840218081851

Percentage of waiting time (ms):
    10.00%:		57
    25.00%:		58
    50.00%:		60
    75.00%:		62
    90.00%:		64
    95.00%:		66
    99.00%:		81
    99.90%:		89
    99.99%:		90


Time detail (ms)
  item		min		mean		max
  dns		0		0		0
  conn		0		0.009		6
  wait		3		60.277		91
  resp		0		0.003		10

Response Time histogram (code: requests):
  200:		10000

总结

  • 在稍微需要考虑并发的情况下，不要使用同步模式下的Django。
  • aiohttp，由于已出现1.4%的fail，故认为其已达极限，结合我做的别的压测，其能承受qps应该在2.2k左右。
  • Go，在qps达8.4k的情况下，未出现失败请求，且总体请求时间仅为aiohttp的28.6%，可见其真实性能应该高于此。