话题是测试下多线程和gevent在socket服务端的小包表现能力，测试的方法不太严谨，也没有用event loop + pool池的概念。不管是gevent和threading有pool的情况下，确实很省资源，但是固定的pool线程池容易在突发事件中被堵塞住。 另外提一句，劲量少用multiprocessing，因为他的进程开销有些大，当然如果单纯用multiprocessing做进程池里面worker进程，那还是个好选择，毕竟他是可以跑多核心的。
Hello , 另外请大家多关注下，我的个人博客 blog.xiaorui.cc
不管怎么说，还是有点属于自娱自乐的形态，有问题之处，请大家喷之 ！
话说，我们当时在搞一个回溯任务中心，说白了就是开发任务接口，通过mapreduce计算平均值，关于业务的逻辑我就不多写了，写出来，也只是浪费大家的思考。干脆点，每个连接都特意堵塞了0.5秒钟。
在大量的tcp短连接测试下，threading的开销越来越大，所以造成了在并发数加大的情况下，出现threading崩溃的情况 ！ gevent是 libevent和协程的融合，一个线程里面都可以跑超多的协程！ 利用libevent做io堵塞的调度 ，gevent体系下，同一时间只有一个任务在运行 ！
先来测试下多线程： 我们就不加线程池了
|
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
|
#!/usr/bin/env python
# -*- coding: utf-8 -*-
#xiaorui.cc
import sys
import socket
import time
import threading
#xiaorui.cc
def threads(port):
s = socket.socket()
s.bind((‘0.0.0.0‘, port))
s.listen(500)
while True:
cli, addr = s.accept()
t = threading.Thread(target=handle_request, args=(cli, time.sleep))
t.daemon = True
t.start()
def handle_request(s, sleep):
try:
s.recv(1024)
sleep(0.5)
s.send(‘‘‘http/1.0 200 OK
Hello World! ‘‘‘)
s.shutdown(socket.SHUT_WR)
print ‘.‘,
except Exception, ex:
print ex
finally:
sys.stdout.flush()
s.close()
if __name__ == ‘__main__‘:
threads(4444)
|
用threading跑socket，每个连接堵塞的时间是0.5秒
|
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
|
time ab -n 10000 -c 500 http://127.0.0.1:4444/
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/
Benchmarking 127.0.0.1 (be patient)
Completed 1000 requests
Completed 2000 requests
Completed 3000 requests
Completed 4000 requests
Completed 5000 requests
Completed 6000 requests
Completed 7000 requests
Completed 8000 requests
Completed 9000 requests
Completed 10000 requests
Finished 10000 requests
Server Software:
Server Hostname: 127.0.0.1
Server Port: 4444
Document Path: /
Document Length: 0 bytes
Concurrency Level: 500
Time taken for tests: 11.123 seconds
Complete requests: 10000
Failed requests: 0
Write errors: 0
Total transferred: 470000 bytes
HTML transferred: 0 bytes
Requests per second: 899.01 [#/sec] (mean)
Time per request: 556.166 [ms] (mean)
Time per request: 1.112 [ms] (mean, across all concurrent requests)
Transfer rate: 41.26 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 33 177.0 0 1000
Processing: 500 508 33.9 501 1132
Waiting: 500 508 33.9 501 1132
Total: 500 541 201.8 501 2132
Percentage of the requests served within a certain time (ms)
50% 501
66% 501
75% 502
80% 505
90% 522
95% 532
98% 1534
99% 1722
100% 2132 (longest request)
real 0m11.145s
user 0m0.210s
sys 0m0.961s
|
加到800的时候~ 直接跳出connection reset by peer，这个就是tcp不能正常的建立通信了。
|
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
|
#xiaorui.cc
import sys
import socket
import time
import gevent
from gevent import socket
def server(port):
s = socket.socket()
s.bind((‘0.0.0.0‘, port))
s.listen(500)
while True:
cli, addr = s.accept()
gevent.spawn(handle_request, cli, gevent.sleep)
def handle_request(s, sleep):
try:
data=s.recv(1024)
sleep(0.5)
s.send(‘‘‘http/1.0 200 OK
Hello World! this is xiaorui.cc !!!‘‘‘)
print data
request_string = "GET %s HTTP/1.1\r\nHost: %s\r\n\r\nServer: xiaorui.cc\n" %(‘index.html‘, ‘127.0.0.1‘)
s.send(request_string)
s.shutdown(socket.SHUT_WR)
print ‘.‘,‘be killed’
except Exception, ex:
print ex
finally:
s.close()
if __name__ == ‘__main__‘:
server(7777)
|
那我们在开始用gevent来跑socket server服务：
首先下面是并发数值是500的时候！
|
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
|
time ab -n 10000 -c 500 http://127.0.0.1:7777/
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/
Benchmarking 127.0.0.1 (be patient)
Completed 1000 requests
Completed 2000 requests
Completed 3000 requests
Completed 4000 requests
Completed 5000 requests
Completed 6000 requests
Completed 7000 requests
Completed 8000 requests
Completed 9000 requests
Completed 10000 requests
Finished 10000 requests
Server Software:
Server Hostname: 127.0.0.1
Server Port: 7777
Document Path: /
Document Length: 0 bytes
Concurrency Level: 500
Time taken for tests: 11.312 seconds
Complete requests: 10000
Failed requests: 0
Write errors: 0
Total transferred: 20000 bytes
HTML transferred: 0 bytes
Requests per second: 884.04 [#/sec] (mean)
Time per request: 565.584 [ms] (mean)
Time per request: 1.131 [ms] (mean, across all concurrent requests)
Transfer rate: 1.73 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 44 202.7 0 1001
Processing: 500 513 10.1 511 707
Waiting: 500 513 10.1 511 707
Total: 500 557 204.1 512 1525
Percentage of the requests served within a certain time (ms)
50% 512
66% 515
75% 517
80% 519
90% 531
95% 552
98% 1521
99% 1523
100% 1525 (longest request)
real 0m11.334s
user 0m0.159s
sys 0m0.730s
|
并发是1000的时候：
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
|
time ab -n 10000 -c 1000 http://127.0.0.1:7777/
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/
Benchmarking 127.0.0.1 (be patient)
Completed 1000 requests
Completed 2000 requests
Completed 3000 requests
Completed 4000 requests
Completed 5000 requests
Completed 6000 requests
Completed 7000 requests
Completed 8000 requests
Completed 9000 requests
Completed 10000 requests
Finished 10000 requests
Server Software:
Server Hostname: 127.0.0.1
Server Port: 7777
Document Path: /
Document Length: 0 bytes
Concurrency Level: 1000
Time taken for tests: 7.406 seconds
Complete requests: 10000
Failed requests: 0
Write errors: 0
Total transferred: 20000 bytes
HTML transferred: 0 bytes
Requests per second: 1350.22 [#/sec] (mean)
Time per request: 740.623 [ms] (mean)
Time per request: 0.741 [ms] (mean, across all concurrent requests)
Transfer rate: 2.64 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 175 491.7 0 3000
Processing: 500 520 17.7 515 707
Waiting: 500 520 17.7 515 707
Total: 500 695 492.5 517 3521
Percentage of the requests served within a certain time (ms)
50% 517
66% 523
75% 538
80% 569
90% 1515
95% 1530
98% 1539
99% 3514
100% 3521 (longest request)
real 0m7.428s
user 0m0.208s
sys 0m0.741s
|
当并发到1500的时候：
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
|
time ab -n 10000 -c 1500 http://127.0.0.1:7777/
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/
Benchmarking 127.0.0.1 (be patient)
Completed 1000 requests
Completed 2000 requests
Completed 3000 requests
Completed 4000 requests
Completed 5000 requests
Completed 6000 requests
Completed 7000 requests
Completed 8000 requests
Completed 9000 requests
Completed 10000 requests
Finished 10000 requests
Server Software:
Server Hostname: 127.0.0.1
Server Port: 7777
Document Path: /
Document Length: 0 bytes
Concurrency Level: 1500
Time taken for tests: 5.290 seconds
Complete requests: 10000
Failed requests: 0
Write errors: 0
Total transferred: 20000 bytes
HTML transferred: 0 bytes
Requests per second: 1890.27 [#/sec] (mean)
Time per request: 793.536 [ms] (mean)
Time per request: 0.529 [ms] (mean, across all concurrent requests)
Transfer rate: 3.69 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 214 404.9 1 1003
Processing: 500 522 23.0 514 716
Waiting: 500 522 23.0 514 716
Total: 500 736 406.7 520 1712
Percentage of the requests served within a certain time (ms)
50% 520
66% 558
75% 602
80% 1506
90% 1526
95% 1531
98% 1535
99% 1548
100% 1712 (longest request)
real 0m5.313s
user 0m0.275s
sys 0m0.763s
|
出现了少量的报错：
gevent 可以加个队列，来限制协程的数目，但是数目限制了，虽然稳定了，但是并发数上不去。
|
1
2
|
from gevent.pool import Pool
pool = Pool(N)
|
这里测试有点简单，其实我在线上用的方案是prefork+gevent pool，后来因为自己fork出去的进行，不太好互相的通信，pipe实在太难用，所以后来又改用multiprocessing了。 另外监听的机制也把select改成epoll了。
