python分析apache访问日志 python分析apache访问日志脚本分享
人气:0想了解python分析apache访问日志脚本讲解的相关内容吗,在本文为您仔细讲解python分析apache访问日志的相关知识和一些Code实例,欢迎阅读和指正,我们先划重点:python,分析apache访问日志,下面大家一起来学习吧。
#!/usr/bin/env python # coding=utf-8 #------------------------------------------------------ # Name: Apache 日志分析脚本 # Purpose: 此脚本只用来分析Apache的访问日志 # Version: 2.0 # Author: LEO # Created: 2013-4-26 # Modified: 2013-5-4 # Copyright: (c) LEO 2013 #------------------------------------------------------ import sys import time #该类是用来打印格式 class displayFormat(object): def format_size(self,size): '''格式化流量单位''' KB = 1024 MB = 1048576 GB = 1073741824 TB = 1099511627776 if size >= TB : size = str(size / TB) + 'T' elif size < KB : size = str(size) + 'B' elif size >= GB and size < TB: size = str(size / GB) + 'G' elif size >= MB and size < GB : size = str(size / MB) + 'M' else : size = str(size / KB) + 'K' return size formatstring = '%-15s %-10s %-12s %8s %10s %10s %10s %10s %10s %10s %10s' def transverse_line(self) : '''输出横线''' print self.formatstring % ('-'*15,'-'*10,'-'*12,'-'*12,'-'*10,'-'*10,'-'*10,'-'*10,'-'*10,'-'*10,'-'*10) def head(self): '''输出头部信息''' print self.formatstring % ('IP','Traffic','Times','Times%','200','404','500','403','302','304','503') def error_print(self) : '''输出错误信息''' print print 'Usage : ' + sys.argv[0] + ' ApacheLogFilePath [Number]' print sys.exit(1) def execut_time(self): '''输出脚本执行的时间''' print print "Script Execution Time: %.3f second" % time.clock() print #该类是用来生成主机信息的字典 class hostInfo(object): host_info = ['200','404','500','302','304','503','403','times','size'] def __init__(self,host): self.host = host = {}.fromkeys(self.host_info,0) def increment(self,status_times_size,is_size): '''该方法是用来给host_info中的各个值加1''' if status_times_size == 'times': self.host['times'] += 1 elif is_size: self.host['size'] = self.host['size'] + status_times_size else: self.host[status_times_size] += 1 def get_value(self,value): '''该方法是取到各个主机信息中对应的值''' return self.host[value] #该类是用来分析文件 class fileAnalysis(object): def __init__(self): '''初始化一个空字典''' self.report_dict = {} self.total_request_times,self.total_traffic,self.total_200, self.total_404,self.total_500,self.total_403,self.total_302, self.total_304,self.total_503 = 0,0,0,0,0,0,0,0,0 def split_eachline_todict(self,line): '''分割文件中的每一行,并返回一个字典''' split_line = line.split() split_dict = {'remote_host':split_line[0],'status':split_line[-2],'bytes_sent':split_line[-1],} return split_dict def generate_log_report(self,logfile): '''读取文件,分析split_eachline_todict方法生成的字典''' for line in logfile: try: line_dict = self.split_eachline_todict(line) host = line_dict['remote_host'] status = line_dict['status'] except ValueError : continue except IndexError : continue if host not in self.report_dict : host_info_obj = hostInfo(host) self.report_dict[host] = host_info_obj else : host_info_obj = self.report_dict[host] host_info_obj.increment('times',False) if status in host_info_obj.host_info : host_info_obj.increment(status,False) try: bytes_sent = int(line_dict['bytes_sent']) except ValueError: bytes_sent = 0 host_info_obj.increment(bytes_sent,True) return self.report_dict def return_sorted_list(self,true_dict): '''计算各个状态次数、流量总量,请求的总次数,并且计算各个状态的总量 并生成一个正真的字典,方便排序''' for host_key in true_dict : host_value = true_dict[host_key] times = host_value.get_value('times') self.total_request_times = self.total_request_times + times size = host_value.get_value('size') self.total_traffic = self.total_traffic + size o200 = host_value.get_value('200') o404 = host_value.get_value('404') o500 = host_value.get_value('500') o403 = host_value.get_value('403') o302 = host_value.get_value('302') o304 = host_value.get_value('304') o503 = host_value.get_value('503') true_dict[host_key] = {'200':o200,'404':o404,'500':o500,'403':o403,'302':o302,'304':o304, '503':o503,'times':times,'size':size} self.total_200 = self.total_200 + o200 self.total_404 = self.total_404 + o404 self.total_500 = self.total_500 + o500 self.total_302 = self.total_302 + o302 self.total_304 = self.total_304 + o304 self.total_503 = self.total_503 + o503 sorted_list = sorted(true_dict.items(),key=lambda t:(t[1]['times'],t[1]['size']),reverse=True) return sorted_list class Main(object): def main(self) : '''主调函数''' display_format = displayFormat() arg_length = len(sys.argv) if arg_length == 1 : display_format.error_print() elif arg_length == 2 or arg_length == 3: infile_name = sys.argv[1] try : infile = open(infile_name,'r') if arg_length == 3 : lines = int(sys.argv[2]) else : lines = 0 except IOError,e : print print e display_format.error_print() except ValueError : print print "Please Enter A Volid Number !!" display_format.error_print() else : display_format.error_print() fileAnalysis_obj = fileAnalysis() not_true_dict = fileAnalysis_obj.generate_log_report(infile) log_report = fileAnalysis_obj.return_sorted_list(not_true_dict) total_ip = len(log_report) if lines : log_report = log_report[0:lines] infile.close() print total_traffic = display_format.format_size(fileAnalysis_obj.total_traffic) total_request_times = fileAnalysis_obj.total_request_times print 'Total IP: %s Total Traffic: %s Total Request Times: %d' % (total_ip,total_traffic,total_request_times) print display_format.head() display_format.transverse_line() for host in log_report : times = host[1]['times'] times_percent = (float(times) / float(fileAnalysis_obj.total_request_times)) * 100 print display_format.formatstring % (host[0], display_format.format_size(host[1]['size']), times,str(times_percent)[0:5], host[1]['200'],host[1]['404'], host[1]['500'],host[1]['403'], host[1]['302'],host[1]['304'],host[1]['503']) if (not lines) or total_ip == lines : display_format.transverse_line() print display_format.formatstring % (total_ip,total_traffic, total_request_times,'100%', fileAnalysis_obj.total_200, fileAnalysis_obj.total_404, fileAnalysis_obj.total_500, fileAnalysis_obj.total_403, fileAnalysis_obj.total_302, fileAnalysis_obj.total_304, fileAnalysis_obj.total_503) display_format.execut_time() if __name__ == '__main__': main_obj = Main() main_obj.main()
加载全部内容