websocket长连接压力测试踩过的坑

Websocket协议压测记录

背景:

    公司的行情系统是采用的websocket协议,有请求和订阅两种方式向服务器申请最新行情信息。请求方式是一次的,订阅方式是建立连接后,服务器定时向客户端推送行情信息。

初步测试方案:

因考虑到websocket是双工通讯,是长连接,并且本次压测的性能指标是系统能建立的最大连接数,并且是建立连接后服务器能持续向客户端推送行情信息。

基于以上原因考虑用python采用多线程建立连接,为了验证能否收到推送的信息,把返回的行情信息保存到文本文件中。Python脚本如下:

import websocket

import time

import threading

import gzip

#import json

#from threadpool import ThreadPool, makeRequests

#from websocket import create_connection

SERVER_URL = "ws://pro-web-new.devtest.exshell-dev.com/r1/main/ws"

#SERVER_URL = "wss://i.cg.net/wi/ws"

#SERVER_URL = "wss://www.exshell.com/r1/main/ws"

def on_message(ws, message):

    print(message)

def on_error(ws, error):

    print(error)

def on_close(ws):

    print("### closed ###")

def on_open(ws):

    def send_trhead():

        send_info = '{"sub": "market.ethusdt.kline.1min","id": "id10"}'

        #send_info = '{"event":"subscribe", "channel":"btc_usdt.deep"}'

        while True:

            #time.sleep(5)

            #ws.send(json.dumps(send_info))

            ws.send(send_info)

            while (1):

                compressData = ws.recv()

                result = gzip.decompress(compressData).decode('utf-8')

                if result[:7] == '{"ping"':

                    ts = result[8:21]

                    pong = '{"pong":' + ts + '}'

                    ws.send(pong)

                    ws.send(send_info)

                else:

                    #print(result)

                    with open('./test_result.txt', 'a') as f:

                        f.write(threading.currentThread().name+' ')

                        f.write(result+' ')

    t = threading.Thread(target=send_trhead)

    t.start()

    print(threading.currentThread().name)

def on_start(a):

    # time.sleep(2)

    # websocket.enableTrace(True)

    # ws = websocket.WebSocketApp(SERVER_URL,

    #                             on_message=on_message,

    #                             on_error=on_error,

    #                             on_close=on_close)

    # ws.on_open = on_open

    # ws.run_forever()

    #print(a[2])

    try:

        ws = websocket.create_connection(SERVER_URL)

        on_open(ws)

    except Exception as e:

        print('error is :',e)

        print('connect ws error,retry...')

        time.sleep(5)

if __name__ == "__main__":

    # pool = ThreadPool(3)

    # test = list()

    # for ir in range(3):

    #     test.append(ir)

    #

    # requests = makeRequests(on_start, test)

    # [pool.putRequest(req) for req in requests]

    # pool.wait()

    # # #on_start(1)

    for ir in range(20):

        on_start(1)

        time.sleep(0.1)

初步测试结果:

    在压测的过程中,发现连接数达到一定程度(单机1400连接),连接就断掉了,监控发现压力机内存基本消耗光了,因建立连接,并接收返回的信息,随着连接数增加,内存消耗大,只能断开连接,释放内存。

调整测试方案:

    和架构、开发讨论后,准备在websocket客户端采用AIO异步通讯方式增大压力,因当时是考虑到长连接未考虑这种方式,查询资料,发现websocket服务端可以采用AIO异步通讯方式,在websocket客户端尝试一下,采用locust + python的方式,也查找了一些资料,发现方案可行。

    Locust是一款可扩展的,分布式的,性能测试的,开源的,用Python编写的性能测试工具。对于测试HTTP协议的接口是比较方便的,但是它也支持测试别的协议的接口,不过需要重写Locust类。脚本如下:

from locust import Locust, events, task, TaskSet

import websocket

import time

import gzip

class WebSocketClient():

     def __init__(self, host, port):

         self.host = host

         self.port = port

class WebSocketLocust(Locust):

     def __init__(self, *args, **kwargs):

         self.client = WebSocketClient("172.31.15.85", 9503)

class UserBehavior(TaskSet):

    ws = websocket.WebSocket()

     #self.ws.connect("ws://10.98.64.103:8807")

     ws.connect("ws://pro-web-new.devtest.exshell-dev.com/r1/main/ws")

     @task(1)

     def buy(self):

         try:

             start_time = time.time()

             #self.ws.send('{"url":"/buy","data":{"id":"123","issue":"20170822","doubled_num":2}}')

            #result = self.ws.recv()

             send_info = '{"sub": "market.ethusdt.kline.1min","id": "id10"}'

             # send_info = '{"event":"subscribe", "channel":"btc_usdt.deep"}'

             while True:

                 # time.sleep(5)

                # ws.send(json.dumps(send_info))

                 ws.send(send_info)

                 while (1):

                    compressData = ws.recv()

                    result = gzip.decompress(compressData).decode('utf-8')

                     if result[:7] == '{"ping"':

                        ts = result[8:21]

                         pong = '{"pong":' + ts + '}'

                         ws.send(pong)

                        ws.send(send_info)

                     else:

                         # print(result)

                        with open('./test_result.txt', 'a') as f:

                             #f.write(threading.currentThread().name + ' ')

                             f.write(result + ' ')

         except Exception as e:

             print("error is:",e)

class ApiUser(WebSocketLocust):

    task_set = UserBehavior

     min_wait = 100

     max_wait = 200

用命令执行脚本:

Locust -f websocket_client_locust.py –-no-web -c 1 -r 1 -t 60s

单个用户执行成功,并能生成文件。但多个用户执行的时候就报错,报错信息如下:This socket is already used by another greenlet: <bound method Waiter.switch of <gevent.hub.Waiter object at 0x7f01f0594900>>

错误原因说的是socke正在被使用,但是我的代码中是新的socket,简单分析了一下,应该不会出现问题,但是我的socek的使用部分是一个全局的client,然后程序运行的时候出现了上述错误.仔细推测我找出了原因:

geven是个协程库,那么多个协程共用一个socek的时候就会出现上述错误了,于是我把socket改成了局部的,问题解决.

修改前:

 

修改后:

 

修改后代码:

from locust import Locust, events, task, TaskSet

import websocket

import time

import gzip

class WebSocketClient():

     def __init__(self, host):

         self.host = host

         #self.port = port

class WebSocketLocust(Locust):

     def __init__(self, *args, **kwargs):

         self.client = WebSocketClient("172.31.15.85")

class UserBehavior(TaskSet):

     # ws = websocket.WebSocket()

    # #self.ws.connect("ws://10.98.64.103:8807")

    # ws.connect("ws://pro-web-new.devtest.exshell-dev.com/r1/main/ws")

     @task(1)

     def buy(self):

         try:

            ws = websocket.WebSocket()

             # self.ws.connect("ws://10.98.64.103:8807")

             ws.connect("ws://pro-web-new.devtest.exshell-dev.com/r1/main/ws")

             start_time = time.time()

             #self.ws.send('{"url":"/buy","data":{"id":"123","issue":"20170822","doubled_num":2}}')

            #result = self.ws.recv()

            send_info = '{"sub": "market.ethusdt.kline.1min","id": "id10"}'

             # send_info = '{"event":"subscribe", "channel":"btc_usdt.deep"}'

             while True:

                 # time.sleep(5)

                # ws.send(json.dumps(send_info))

                 ws.send(send_info)

                 while (1):

                    compressData = ws.recv()

                    result = gzip.decompress(compressData).decode('utf-8')

                     if result[:7] == '{"ping"':

                        ts = result[8:21]

                         pong = '{"pong":' + ts + '}'

                         ws.send(pong)

                        ws.send(send_info)

                     # else:

                    #     # print(result)

                    #     with open('./test_result.txt', 'a') as f:

                    #         #f.write(threading.currentThread().name + ' ')

                    #         f.write(result + ' ')

         except Exception as e:

             print("error is:",e)

class ApiUser(WebSocketLocust):

    task_set = UserBehavior

     min_wait = 100

     max_wait = 200

压测开始,随着用户数上升,压力机端发生如下错误:500和502错误

 

这是协议进行握手时失败,查询后端行情应用服务器,也有大量报错。

查看服务器发现打开最大文件数是1024,调整到65535,用如下命令调整:

第一步,修改/etc/sysctl.conf文件,在文件中添加如下行:

  net.ipv4.ip_local_port_range = 1024 65000

  这表明将系统对本地端口范围限制设置为1024~65000之间。请注意,本地端口范围的最小值必须大于或等于1024;而端口范围的最大值则应小于或等于65535.修改完后保存此文件。

  第二步,执行sysctl命令:

  [speng@as4 ~]$ sysctl -p

  如果系统没有错误提示,就表明新的本地端口范围设置成功。如果按上述端口范围进行设置,则理论上单独一个进程最多可以同时建立60000多个TCP客户端连接。

调整完成后复测,发现还是报这个错误,请开发进行定位分析应用程序。

原文地址:https://www.cnblogs.com/devtest/p/9966465.html