可测含多进程的app–Python–通过adb命令获取Android App应用的性能数据：CPU、GPU、内存、电池、耗电量，并与Perfdog取值对比结果
1、原理
python脚本通过os.popen()方法运行adb命令，获取性能数据，将性能数据保存在csv文件并算出均值、最大值、最小值。
本脚本可测试一个app包含多个进程的场景，可以获取每个进程的性能数据。
2、环境准备：
2.1 软件环境
具备python环境，Android环境
需要python库：os, csv, time, datetime, sys,time,pandas
2.2 手机状态
Wi-Fi模式连接手机
2、统一手机环境变量：手机满电，将手机降至常温（每个测试场景取接近的初始温度），屏幕亮度和音量调为50%
3、杀掉其他进程
4、手机温度：脚本获取到的是电池温度（Perfdog也是如此）。温枪测温：可将温枪调整为测物体模式测温，每次选取相同测试温点，测温点可选摄像头附近，不要直对摄像头。温枪测的和电池温度相近。
（手机降温：可以使用冰块、小风扇，或者将手机放入冰箱降温）
3、获取性能指标
统一说明：以下adb命令在Mac电脑上使用的是‘grep’，在windows电脑执行时改为‘findstr’
3.1 同时会用到的其他adb命令
# Wi-Fi模式连接手机，手机与电脑连接同一个Wi-Fi，查看手机IP地址，执行命令：adb connect 192.168.88.152# 如果连接失败，可以执行如下命令重启端口5555，再重新连接adb tcpip 5555# 获取app应用名adb shell dumpsys window | grep mCurrentFocus# 获取进程IDadb shell ps | grep im.zego.zegoland# 取出 文件指定行数 区间内容os.popen("sed -n '{},{}p' {} > {}".format(sed_count['start_count'], sed_count['end_count'], original_path,result_path)）3.2 性能指标相关汇总表
| 性能指标 | 在.csv文件中的名称 | 可测目标 | 与Perfdog对比 | adb命令 |
|---|---|---|---|---|
| 电池百分比、 电池温度 | battery_precent、 temperature[°C] | 手机 | 相同 | adb shell dumpsys battery |
| 耗电量：总、 屏幕、CPU、摄像头、Wi-Fi、系统服务、传感器、麦克风 | ‘battery_sum[mAh]’, ‘battery_screen[mAh]’, ‘battery_cpu[mAh]’, ‘battery_camera[mAh]’, ‘battery_wifi[mAh]’, ‘battery_system_services[mAh]’, ‘battery_sensors[mAh]’, ‘battery_audio[mAh]’ | 整个app | Perfdog计算的是功耗 | # 获取耗电量，单位：mAh，一定要WiFi模式连接手机 # 要先清空已有的耗电数据 adb shell dumpsys batterystats --enable full-wake-history # 重置设备耗电数据 adb shell dumpsys batterystats --reset # 执行测试场景后，获取耗电量数据 # 获取应用进ID，去掉下划线，如“u0_a901”，取值“u0a901” # adb shell ps | grep im.zego.zegoland 获取被测应用的耗电量 adb shell dumpsys batterystats > batteryinfo.txt #通过筛选关键字（详细代码中会写）获取耗电量所在开始、结尾行数，再通过sed命令截取区间，最后在这个sed区间内根据uid和耗电项关键字筛选耗电量数据 |
| CPU使用率 | AppCPU[%] | 接近，可替代 | 多进程 | # -n指定刷新次数 adb shell top -n 1 > cpuinfo.txt # 再根据PID筛选被测app的cpu数据 |
| GPU使用率 | GUsage[%] | 与PerfDog误差在0.001位 | 手机 | # 获取GPU数据，使用第2个数除以第1个数 # 部分机型用这个命令，如一加8T,OPPO Reno6 adb shell cat /sys/class/kgsl/kgsl-3d0/gpubusy # 部分机型用这个命令，如红米note7 adb shell su -c ‘cat /sys/class/kgsl/kgsl-3d0/gpubusy’ |
| 内存Pss | PerfDog-Memory-Pss[MB] | 与Perfdog取值相同 | 多进程 | # adb shell dumpsys meminfo > meminfo.txt #通过筛选关键字（详细代码中会写）获取Pss所在开始、结尾行数，再通过sed命令截取区间，最后在这个sed区间内根据PID和内存位置获取内存Pss数据 |
| 内存Uss | Memory-Uss[MB] | 物理专属内存，需要手机root，Perfdog获取的是Pss | / | # 需要手机root #部分手机使用如下命令 adb shell ‘procrank’ # 部分手机使用如下命令 adb shell su -c ‘procrank’ |
关于Android 系统内存耗用：VSS/RSS/PSS/USS 的介绍，可参考：
VSS/RSS/PSS/USS 的介绍【声网】
4、源码
4.1 获取AppCPU使用率，GPU使用率、内存Pss、电池百分比、电池温度、耗电量源码
使用时，根据自己测试信息修改"if main"中的appName、deviceName、性能数据文件名PerfDataFile、收集性能时间runTime（单位秒，可修改为分钟、小时），被测进程（可多个）
#! /usr/bin/env python3# _*_ coding:utf-8 _*_"""@Auth : Dora@ Time： 2023.5.27调整为不依赖外部方法的脚本"""import os, csv, time, datetime, sysfrom time import sleepimport pandas as pdclass Logger(object): def __init__(self, filename='default.log', stream=sys.stdout): self.terminal = stream self.log = open(filename, 'a+') def write(self, message): self.terminal.write(message) self.log.write(message) def flush(self): pass# 将控制台输出到日志文件中，日志是追加模式，记得定时清理# 便于检验数据获取是否正确。也可将这部分删除sys.stdout = Logger('../log/stout_log.log', sys.stdout)sys.stderr = Logger('../log/stout.log_file', sys.stderr)# 可用class PertestInfo: def __init__(self, appName, deviceName, PerfDataFile, runTime, *progress): self.PerfDataFile = PerfDataFile # 分割线，便于定位日志 print("{:*^50s}".format("Split Line")) print("{:+^50s}".format(self.PerfDataFile)) self.appName = appName # 要测试的app self.progress = progress # 要测试的进程，一个应用可能有多个进程 print("self.progress:", self.progress, type(self.progress)) self.deviceName = deviceName # 设备名称 self.current_time = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()) # 运行时间 self.run_time = datetime.timedelta(seconds=runTime) # 收集性能数据时长，时间单位可修改为：hours,minutes,seconds def clear_getsize(self, file_path): # 可以写成一个装饰器，待优化 with open(file_path,'w'): pass start_time = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()) os.popen('adb -s {} shell top -n 1 > {}'.format(deviceName, file_path)) while not os.path.getsize(file_path): # os.path.getsize() 返回文件的字节数，如果为 0，则代表空 # 判断执行top命令返回的数据是否存入文件，若文件为0，继续sleep(1) sleep(1) end_time = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()) print('运行top命令开始时间：', start_time) print('运行top命令数据保存到文件的结束时间：', end_time) def get_PID(self): # 获取pid appPID = {} # result = os.popen('adb -s {} shell ps | grep {}'.format(self.deviceName, appName)) for app in self.progress: # print('app:', app) result = os.popen('adb -s {} shell ps | grep {}'.format(self.deviceName, app)) # print("pid result:", result) for line in result.readlines(): print('line:', line, type(line)) line = '#'.join(line.split()) + '#' # print('line#:', line) appstr = app + '#' if appstr in line: print('line#:', line) pid = line.split('#')[1] # print("pid:", pid) appPID[app] = pid print('appPID:', appPID) sleep(1) return appPID def sed_result(self, original_path, keyword, result_path): # 获取开始、结束行数 start_time = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()) print('获取start_index开始时间：', start_time) sed_index = {'start_index': 0, 'end_index': 0} while sed_index['start_index'] == 0: sleep(1) for index, line in enumerate(open(original_path, 'r')): if keyword in line: sed_index['start_index'] = index print("start_index:", sed_index['start_index']) break end_time = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()) print('获取start_index结束时间：', end_time) with open(original_path, 'r') as f: result = f.readlines()[sed_index['start_index']:] index = sed_index['start_index'] for line in result: index = index + 1 if line in ['\n', '\r\n']: sed_index['end_index'] = index print("end_index:", sed_index['end_index']) break # 从开始、结束行数截取内容 os.popen("sed -n '{},{}p' {} > {}".format(sed_index['start_index'], sed_index['end_index'], original_path, result_path)) # 案例：取出 / etc / passwd的第2行到第5行 # sed - n '2,5p' / etc / passwd # result = os.popen('sed -n {}p {}'.format(500, filename)).read() def get_batteryPercent_temperature(self): # 获取电量半分比和电池温度 result = os.popen("adb -s {} shell dumpsys battery".format(self.deviceName)) # .read() # print(result) battery_percent = {'battery_percent': '', 'temperature': ''} for line in result: # print("line:{}-----", line) if "level" in line: battery_percent['battery_percent'] = int(line.split(":")[1]) # print("battery_percent:{},type(battery_percent):{}".format(battery_percent, type(battery_percent))) if "temperature" in line: battery_percent['temperature'] = int(line.split(":")[1]) / 10 # print("battery_percent:{},type(battery_percent):{}".format(battery_percent, type(battery_percent))) print("battery_percent:", battery_percent) return battery_percent def reset_battery(self): # 执行业务场景前，清空耗电数据，并重置设备耗电数据 os.popen("adb -s {} shell dumpsys batterystats --enable full-wake-history".format(self.deviceName)) # 清空已有的耗电数据 sleep(2) os.popen("adb -s {} shell dumpsys batterystats --reset".format(self.deviceName)) # 重置设备耗电数据 def get_battery(self): # 获取耗电量，单位：mAh # 一定要WiFi模式连接手机 battery = {'battery_sum': '', 'battery_screen': '', 'battery_cpu': '', 'battery_camera': '', 'battery_wifi': '', 'battery_system_services': '', 'battery_sensors': '', 'battery_audio': ''} # 获取进程ID:uid cmd1 = "adb -s {} shell ps | grep {}".format(self.deviceName, self.appName) uid_data = os.popen(cmd1).read() uid_original = uid_data.split()[0] uid = uid_original.replace('_', '') print('uid:', uid) original_path = '../log/original_data.txt' result_path = '../log/sed_result.txt' # 先清空之前的内容 with open(original_path,'w'): pass with open(result_path,'w'): pass start_time = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()) str1 = ' Estimated power use (mAh):' cmd2 = 'adb -s {} shell dumpsys batterystats > {}'.format(self.deviceName, original_path) os.popen(cmd2) while not os.path.getsize(original_path): # os.path.getsize() 返回文件的字节数，如果为 0，则代表空 # 耗电量数据存储需要时间，判断执行top命令返回的数据是否存入文件 sleep(1) end_time = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()) print('运行获取耗电量开始时间：', start_time) print('运行耗电量数据保存到文件的结束时间：', end_time) self.sed_result(original_path=original_path, keyword=str1, result_path=result_path) while not os.path.getsize(result_path): sleep(1) with open(result_path, 'r') as f: for line in f.readlines(): if uid in line: print('battery_line:', line) line = '#'.join(line.split()) battery['battery_sum'] = line.split('#')[2] if 'screen=' in line: battery['battery_screen'] = line.split('screen=')[1].split('#')[0] if 'cpu=' in line: battery['battery_cpu'] = line.split('cpu=')[1].split('#')[0] if 'camera=' in line: battery['battery_camera'] = line.split('camera=')[1].split('#')[0] if 'wifi=' in line: battery['battery_wifi'] = line.split('wifi=')[1].split('#')[0] if 'system_services=' in line: battery['battery_system_services'] = line.split('system_services=')[1].split('#')[0] if 'sensors=' in line: battery['battery_sensors'] = line.split('sensors=')[1].split('#')[0] if 'audio=' in line: battery['battery_audio'] = line.split('audio=')[1].split('#')[0] print('battery:\n', battery) return battery def get_cpuinfo(self): # 获取PID appPID = self.get_PID() # print("appPID:", appPID) appCPU = {} file_path = '../log/cpuinfo.txt' # 先清空之前的内容 with open(file_path,'w'): pass start_time = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()) os.popen('adb -s {} shell top -n 1 > {}'.format(deviceName, file_path)) while not os.path.getsize(file_path): # os.path.getsize() 返回文件的字节数，如果为 0，则代表空 # 判断执行top命令返回的数据是否存入文件 sleep(1) end_time = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()) print('运行top命令开始时间：', start_time) print('运行top命令数据保存到文件的结束时间：', end_time) with open(file_path, 'r') as f: for line in f.readlines(): for app in self.progress: if app in appPID.keys(): if appPID[app] in line:# print('app:', app)# print('appPID line:', line)# appCPU['time'] = int(time.time()) # 当前时间）cpu = round(float(line.split()[-4]) / 8, 2)appCPU[app] = cpu print("appCPU:", appCPU) return appCPU def get_GUsage(self): # 获取GPU使用率，与PerfDog误差在0.001以内 result1 = os.popen( 'adb -s {} shell "cat /sys/class/kgsl/kgsl-3d0/gpubusy"'.format(deviceName)).read() # 获取GPU占用率,一加8T # print("result:", result1) result2 = os.popen('adb -s {} shell su -c "cat /sys/class/kgsl/kgsl-3d0/gpubusy"'.format( deviceName)).read() # 获取GPU占用率，红米note7 # print("result:", result2) GUsage = 0 if result1: data1 = result1.split() GUsage = round(int(data1[0]) / int(data1[1]) * 100, 3) # round(i,j)给i取j位小数，X100，转化为百分比 print('GUsage1:', GUsage) elif result2: data2 = result2.split() GUsage = round(int(data2[0]) / int(data2[1]) * 100, 3) # round(i,j)给i取j位小数，X100，转化为百分比 print('GUsage2:', GUsage) else: print("GUsage获取失败，可能获取GPU命令不适用于被测机型") return GUsage def get_memory_Pss(self): # 获取pid appPID = self.get_PID() # 获取内存数据 original_path = '../log/original_data.txt' result_path = '../log/sed_result.txt' # 先清空之前的内容 with open(original_path,'w'): pass with open(result_path,'w'): pass start_time = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()) str1 = 'Total PSS by process:' cmd2 = 'adb -s {} shell dumpsys meminfo > {}'.format(self.deviceName, original_path) print('cmd2:', cmd2) os.popen(cmd2) while not os.path.getsize(original_path): # os.path.getsize() 返回文件的字节数，如果为 0，则代表空 # 判断执行top命令返回的数据是否存入文件 sleep(1) end_time = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()) print('运行获取内存开始时间：', start_time) print('运行内存数据保存到文件的结束时间：', end_time) self.sed_result(original_path=original_path, keyword=str1, result_path=result_path) while not os.path.getsize(result_path): sleep(1) # appPSS = {'im.zego.zegoland:zegoland_avatar': 0, 'im.zego.zegoland:zegoland_unity': 0,'im.zego.zegoland:zegoland': 0} appPss = {} with open(result_path, 'r') as f: for line in f.readlines(): for app in self.progress: if app in appPID.keys(): if appPID[app] in line:# print('app:', app)print('app_mem_line:', line, app, appPID[app])Pss = round((int(line.strip().split('K: ')[0].replace(',', ''))) / 1024, 2)# print("Pss:", Pss)appPss[app] = Pss print("appPss", appPss) return appPss def analyse_dumpsys_Csv(self, fileName): # 对性能数据文件进行计算，获取均值、最大值、最小值 analyse_data = [] info = pd.read_csv(fileName, encoding='gbk') rowsName = info.columns # 列名 print('列名：', rowsName) lines = info.shape[0] # 行数 rows = info.shape[1] # 列数 print("行:{},type{}，列：{}".format(lines, type(lines), rows)) # result = {} avg_data = ['', '', '', 'avg'] # 存放均值 max_data = ['', '', '', 'max'] # 存放最大值 min_data = ['', '', '', 'min'] # 存放最小值 for row_name in rowsName: if row_name not in ['id', 'hour', 'minute', 'second']: avg_value = round(info[row_name].mean(), 2) max_value = info[row_name].max() min_value = info[row_name].min() avg_data.append(avg_value) max_data.append(max_value) min_data.append(min_value) # 将均值、最大值、最小值存入analyse_data以便输入csv文件 analyse_data.append(avg_data) analyse_data.append(max_data) analyse_data.append(min_data) print('analyse_data:', analyse_data) # 在数据表中插入平均值、最大值、最小值 with open(fileName, 'a+', newline='') as file: # a+ 追加方式写+读 writer = csv.writer(file) writer.writerows(analyse_data) sleep(1) file.close() def perf_run(self): # 执行获取数据函数 GPU_battery_data = [ ['id', 'hour', 'minute', 'second', 'GUsage[%]', 'battery_percent[%]', 'temperature', 'battery_sum[mAh]', 'battery_screen[mAh]', 'battery_cpu[mAh]', 'battery_camera[mAh]', 'battery_wifi[mAh]', 'battery_system_services[mAh]', 'battery_sensors[mAh]', 'battery_audio[mAh]']] print('GPU_battery_data:', GPU_battery_data) print("now:", datetime.datetime.now()) print('run_time:', self.run_time) self.reset_battery() # 清空耗电数据，并重置耗电数据 sleep(1) id = 1 # id：可以运行的次数 end_time = datetime.datetime.now() + self.run_time mem_Pss_data = [] cpu_data = [] while end_time > datetime.datetime.now(): # 在指定时间内运行 battery_percent = self.get_batteryPercent_temperature() # 获取电量百分数和电池温度 print('battery_percent:', battery_percent) # sleep(1) GUsage = self.get_GUsage() # 获取GPU占用率 # sleep(1) GPU_battery_data.append(( str(id), str(datetime.datetime.now().hour), str(datetime.datetime.now().minute), str(datetime.datetime.now().second), str(GUsage), battery_percent['battery_percent'], battery_percent['temperature'], '', '', '', '', '')) cpu_info = self.get_cpuinfo() # 获取cpu cpu_info['id'] = id cpu_info['hour'] = datetime.datetime.now().hour cpu_info['minute'] = datetime.datetime.now().minute cpu_info['second'] = datetime.datetime.now().second # cpu_info['time'] = str(datetime.datetime.now().hour)+':'+str(datetime.datetime.now().minute)+':'+str(datetime.datetime.now().second) cpu_data.append(cpu_info) # sleep(1) mem_Pss = self.get_memory_Pss() # 获取内存 mem_Pss['id'] = id mem_Pss['hour'] = datetime.datetime.now().hour mem_Pss['minute'] = datetime.datetime.now().minute mem_Pss['second'] = datetime.datetime.now().second mem_Pss_data.append(mem_Pss) # print("多组Pss数据：", mem_Pss_data) id = id + 1 sleep(1) # 最后一组数据展示耗电量 mAh battery_percent = self.get_batteryPercent_temperature() # 获取电量和电池温度 # sleep(1) cpu_info = self.get_cpuinfo() # 获取cpu print('cpu_info:', cpu_info) cpu_info['id'] = id cpu_info['hour'] = datetime.datetime.now().hour cpu_info['minute'] = datetime.datetime.now().minute cpu_info['second'] = datetime.datetime.now().second cpu_data.append(cpu_info) # sleep(1) GUsage = self.get_GUsage() # 获取GPU占用率 # sleep(1) mem_Pss = self.get_memory_Pss() # 获取内存 id = id + 1 mem_Pss['id'] = id mem_Pss['hour'] = datetime.datetime.now().hour mem_Pss['minute'] = datetime.datetime.now().minute mem_Pss['second'] = datetime.datetime.now().second # print("一组Pss数据：", mem_Pss) mem_Pss_data.append(mem_Pss) # print("多组Pss数据：", mem_Pss_data) # sleep(1) battery = self.get_battery() # 耗电量数据，单位：mAh print("") GPU_battery_data.append(( str(id), str(datetime.datetime.now().hour), str(datetime.datetime.now().minute), str(datetime.datetime.now().second), str(GUsage), battery_percent['battery_percent'], battery_percent['temperature'], battery['battery_sum'], battery['battery_screen'], battery['battery_cpu'], battery['battery_camera'], battery['battery_wifi'], battery['battery_system_services'], battery['battery_sensors'], battery['battery_audio'])) print("收集性能数据结束时间:", datetime.datetime.now()) sleep(1) print("保存数据") # 保存数据GPU和battery数据 with open('../data/batatery_GPU-%s' % self.PerfDataFile, 'w', newline='') as file: writer = csv.writer(file, quoting=csv.QUOTE_ALL) writer.writerows(GPU_battery_data) sleep(10) file.flush() sleep(2) file.close() print("分析数据") # 分析数据 # self.analyse_perf_Csv('../data/batatery_GPU-%s' % self.PerfDataFile) # 分析GPU和battery数据 self.analyse_dumpsys_Csv('../data/batatery_GPU-%s' % self.PerfDataFile) # 保存内存数据 fieldName = [] for app in self.progress: # print('app:', app) fieldName.append(app) els = ['id', 'hour', 'minute', 'second'] fieldName = els + fieldName print('filedName:', fieldName) with open('../data/mem_Pss-%s' % self.PerfDataFile, 'w', newline='') as csvfile: writer = csv.DictWriter(csvfile, fieldnames=fieldName) writer.writeheader() for row in mem_Pss_data: writer.writerow(row) self.analyse_dumpsys_Csv('../data/mem_Pss-%s' % self.PerfDataFile) # 保存cpu数据 """ cpu_fieldName = [] for app in self.progress: # print('app:', app) cpu_fieldName.append(app) cpu_els = ['id', 'time'] cpu_fieldName = cpu_els + cpu_fieldName """ with open('../data/AppCPU-%s' % self.PerfDataFile, 'w', newline='') as csvfile: writer = csv.DictWriter(csvfile, fieldnames=fieldName) writer.writeheader() for row in cpu_data: writer.writerow(row) self.analyse_dumpsys_Csv('../data/AppCPU-%s' % self.PerfDataFile, )if __name__ == '__main__': # 一定要Wi-Fi模式连接手机 # deviceName = '192.168.89.48:5555' # deviceName = '192.168.1.105:5555' # deviceName = '192.168.88.152:5555' # 一加8T # deviceName = '192.168.88.145:5555' # 华为P50 deviceName = '192.168.88.210:5555' # oppo reno6 # deviceName = '192.168.1.106:5555' # deviceName = '192.168.88.192:5555' # 小米11青春版 # deviceName = '192.168.88.198:5555' # 红米note7 # deviceName = '192.168.88.199:5555' # oppo reno3 pro appName = 'im.zego.zegoland' # ZegoLand # appName = 'com.zego.goavatar' # avatar # 运行时常 runTime = 60 PerfDataFile = "OPPOreno6-语聊房-{}.csv".format( str(time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()))) # 应用包含多个进程时，存放于*progress # ZL phone = PertestInfo(appName, deviceName, PerfDataFile, runTime, 'im.zego.zegoland:zegoland_avatar', 'im.zego.zegoland:zegoland_unity', 'im.zego.zegoland') # phone = PertestInfo(appName, deviceName, PerfDataFile, runTime,'com.zego.goavatar') # appPID = phone.get_PID() # phone.get_battery() # phone.get_GUsage() # phone.get_cpuinfo() # phone.get_memory_Pss() phone.perf_run()5、测试数据示例
测试数据输入到csv文件，并算出平均值、最大值、最小值。示例截图如下：
6、使用场景：
手动将app设置到测试场景，运行性能测试脚本perftestino.py，收集性能数据；
2、使用多线程，在执行monkey测试的同时收集性能数据和log cat日志，同时检测crash，导出crash日志；
3、使用多线程，在执行UI自动化测试场景的同时收集性能数据
4、手工测试功能时，运行性能测试脚本，捕捉可能存在的内存泄露等问题
来源地址：https://blog.csdn.net/u012089395/article/details/126668946
