正则表达式基础知识请参阅《正则表达式基础知识》,本文使用正则表达式来匹配多行日志并从中解析出相应的信息。
假设现在有这样的SQL日志:
SELECT * FROM open_app WHERE 1 and `client_id` = 'a08f5e32909cc9418f' and `is_valid` = '1' order by id desc limit 32700,100;
# Time: 160616 10:05:10
# User@Host: shuqin[qqqq] @ [1.1.1.1] Id: 46765069
# Schema: db_xxx Last_errno: 0 Killed: 0
# Query_time: 0.561383 Lock_time: 0.000048 Rows_sent: 100 Rows_examined: 191166 Rows_affected: 0
# Bytes_sent: 14653
SET timestamp=1466042710;
SELECT * FROM open_app WHERE 1 and `client_id` = 'a08f5e32909cc9418f' and `is_valid` = '1' order by id desc limit 36700,100;
# User@Host: shuqin[ssss] @ [2.2.2.2] Id: 46765069
# Schema: db_yyy Last_errno: 0 Killed: 0
# Query_time: 0.501094 Lock_time: 0.000042 Rows_sent: 100 Rows_examined: 192966 Rows_affected: 0
# Bytes_sent: 14966
SET timestamp=1466042727;
要求从中解析出相应的信息, 有如下知识点:
(1) 默认正则是单行模式, 要匹配多行,需要开启 "多行模式": MULTILINE; 对于点号,默认不匹配换行符,为了匹配换行符,也需要开启 "DOTALL模式";
(2) 为了匹配每个多行日志,必须使用非贪婪模式,即在 .* 后面加 ? , 否则第一个匹配会匹配到末尾;
(3) 分而治之。编写正确的正则表达式匹配指定长字符串是不容易的,采用的策略是分而治之,将整个字符串分解成多个子串,分别匹配字串。这里每个字串都是一行,匹配好一行后,可以进一步在行内更细化的匹配;
(4) 无处不在的空格符要使用 \s* 或 \s+ 来增强健壮性; 固定的普通字符串可以在正则表达式中更好地标识各个字串,更容易地匹配到。
(5) Python 正则有两个常用用法: re.findall , re.match ; 前者的匹配结果是一个列表, 每个列表元素是一个元组, 匹配一个多行日志;元组的每个元素用来提取对应捕获分组的字符串; re.match 的匹配结果是一个 Match 对象, 可以通过 group(n) 来获取每个捕获分组的匹配字符串。下面的程序特意两种都用到了。对于多行匹配,使用了 re.findall ; 对于行内匹配,使用了 re.match ; 初学者常问这两者那两者有什么区别, 其实动手试试就知道了。
(6) 展示结构使用 Map. 解析出结果后,必然要展示或做成报告,使用 Map & List 结合的复合结构通常是非常适宜的选择。 比如这一例,如果要展示所有 SQL 日志详情,可以做成
{"tablename1": [{sqlobj11}, {sqlobj12}], ..., "tablenameN": [{sqlobjN1}, {sqlobjN2}] } ,每个 sqlobj 结构为:
{"sql": "select xxx", "QueryTime": 0.5600, ...}
要展示简要的报告,比如每个表的 SQL 统计, 可以做成
{"tablename1": {"sql11": 98, "sql12": 16}, ..., "tablenameN": {"sqlN1": 75, "sqlN2": 23} }
Python 程序实现:
importre
globalRegex= r'^\s*(.*?)# (User@Host:.*?)# (Schema:.*?)# (Query_time:.*?)# Bytes_sent:(.*?)SET timestamp=(\d+);\s*$'costRegex= r'Query_time:\s*(.*)\s*Lock_time:\s*(.*)\s*Rows_sent:\s*(\d+)\s*Rows_examined:\s*(\d+)\s*Rows_affected:\s*(\d+)\s*'schemaRegex= r'Schema:\s*(.*)\s*Last_errno:(.*)\s*Killed:\s*(.*)\s*'
defreadSlowSqlFile(slowSqlFilename):
f=open(slowSqlFilename)
ftext= ''
for line inf:
ftext+=line
f.close()returnftextdeffindInText(regex, text):return re.findall(regex, text, flags=re.DOTALL+re.MULTILINE)defparseSql(sqlobj, sqlText):try:if sqlText.find('#') != -1:
sqlobj['sql'] = sqlText.split('#')[0].strip()
sqlobj['time'] = sqlText.split('#')[1].strip()else:
sqlobj['sql'] =sqlText.strip()
sqlobj['time'] = ''
except:
sqlobj['sql'] =sqlText.strip()defparseCost(sqlobj, costText):
matched=re.match(costRegex, costText)
sqlobj['Cost'] =costTextifmatched:
sqlobj['QueryTime'] = matched.group(1).strip()
sqlobj['LockTime'] = matched.group(2).strip()
sqlobj['RowsSent'] = int(matched.group(3))
sqlobj['RowsExamined'] = int(matched.group(4))
sqlobj['RowsAffected'] = int(matched.group(5))defparseSchema(sqlobj, schemaText):
matched=re.match(schemaRegex, schemaText)
sqlobj['Schema'] =schemaTextifmatched:
sqlobj['Schema'] = matched.group(1).strip()
sqlobj['LastErrno'] = int(matched.group(2))
sqlobj['Killed'] = int(matched.group(3))defparseSQLObj(matched):
sqlobj={}try:if matched and len(matched) >0:
parseSql(sqlobj, matched[0].strip())
sqlobj['UserHost'] = matched[1].strip()
sqlobj['ByteSent'] = int(matched[4])
sqlobj['timestamp'] = int(matched[5])
parseCost(sqlobj, matched[3].strip())
parseSchema(sqlobj, matched[2].strip())returnsqlobjexcept:returnsqlobjif __name__ == '__main__':
files= ['slow_sqls.txt']
alltext= ''
for f infiles:
text=readSlowSqlFile(f)
alltext+=text
allmatched=findInText(globalRegex, alltext)
tablenames= ['open_app']if not allmatched or len(allmatched) ==0:print 'No matched. exit.'exit(1)
sqlobjMap={}for matched inallmatched:
sqlobj=parseSQLObj(matched)if len(sqlobj) ==0:continue
for tablename intablenames:if sqlobj['sql'].find(tablename) != -1:if notsqlobjMap.get(tablename):
sqlobjMap[tablename]=[]
sqlobjMap[tablename].append(sqlobj)breakresultMap={}for (tablename, sqlobjlist) insqlobjMap.iteritems():
sqlstat={}for sqlobj insqlobjlist:if sqlobj['sql'] not insqlstat:
sqlstat[sqlobj['sql']] =0
sqlstat[sqlobj['sql']] += 1resultMap[tablename]=sqlstat
f_res= open('/tmp/res.txt', 'w')
f_res.write('-------------------------------------: \n')
f_res.write('Bref results: \n')for (tablename, sqlstat) inresultMap.iteritems():
f_res.write('tablename:' + tablename + '\n')
sortedsqlstat= sorted(sqlstat.iteritems(), key=lambda d:d[1], reverse =True)for sortedsql insortedsqlstat:
f_res.write('sql = %s\ncounts: %d\n\n' % (sortedsql[0], sortedsql[1]))
f_res.write('-------------------------------------: \n\n')
f_res.write('-------------------------------------: \n')
f_res.write('Detail results: \n')for (tablename, sqlobjlist) insqlobjMap.iteritems():
f_res.write('tablename:' + tablename + '\n')
f_res.write('sqlinfo: \n')for sqlobj insqlobjlist:
f_res.write('sql:' + sqlobj['sql'] + 'QueryTime:' + str(sqlobj.get('QueryTime')) + 'LockTime:' + str(sqlobj.get('LockTime')) + '\n')
f_res.write(str(sqlobj)+ '\n\n')
f_res.write('-------------------------------------: \n')
f_res.close()
可配置
事实上,可以做成可配置的。只要给定行间及行内关键字集合,可以分割多行及行内字段,就可以分别提取相应的内容。
这里有个基本函数 matchOneLine: 根据一个依序分割一行内容的关键字列表,匹配一行内容,得到每个关键字对应的内容。这个函数用于匹配行内内容。
配置方式: 采用列表的列表。列表中的每个元素列表是可以分割和匹配单行内容的关键字列表。 每个关键字都用于分割单行的某个区域的内容。 为了提升解析性能,这里对关键字列表进行了预编译正则表达式,以便在解析字符串的时候不做重复工作。
见如下代码:
#!/usr/bin/python#_*_encoding:utf-8_*_
importre#config line keywords to seperate lines.
ksconf = [['S'], ['# User@Host:','Id:'] , ['# Schema:', 'Last_errno:', 'Killed:'], ['# Query_time:','Lock_time:', 'Rows_sent:', 'Rows_examined:', 'Rows_affected:'], ['# Bytes_sent:'], ['SET timestamp=']]
files= ['slow_sqls.txt']#ksconf = [['id:'], ['name:'], ['able:']]#files = ['stu.txt']
globalConf= {'ksconf': ksconf, 'files': files}
defproduceRegex(keywordlistInOneLine):'''build the regex to match keywords in the list of keywordlistInOneLine'''oneLineRegex= "^\s*"oneLineRegex+= "(.*?)".join(keywordlistInOneLine)
oneLineRegex+= "(.*?)\s*$"
returnoneLineRegexdefreadFile(filename):
f=open(filename)
ftext= ''
for line inf:
ftext+=line
f.close()returnftextdefreadAllFiles(files):return ''.join(map(readFile, files))deffindInText(regex, text, linesConf):'''return a list of maps, each map is a match to multilines,
in a map, key is the line keyword
and value is the content corresponding to the key'''matched=regex.findall(text)ifempty(matched):return[]
allMatched=[]
linePatternMap=buildLinePatternMap(linesConf)for onematch inmatched:
oneMatchedMap=buildOneMatchMap(linesConf, onematch, linePatternMap)
allMatched.append(oneMatchedMap)returnallMatcheddefbuildOneMatchMap(linesConf, onematch, linePatternMap):
sepLines= map(lambdaks:ks[0], linesConf)
lenOflinesInOneMatch=len(sepLines)
lineMatchedMap={}for i inrange(lenOflinesInOneMatch):
lineContent= sepLines[i] +onematch[i].strip()
linekey=getLineKey(linesConf[i])
lineMatchedMap.update(matchOneLine(linesConf[i], lineContent, linePatternMap))returnlineMatchedMapdefmatchOneLine(keywordlistOneLine, lineContent, patternMap):'''match lineContent with a list of keywords , and return a map
in which key is the keyword and value is the content matched the key.
eg.
keywordlistOneLine = ["host:", "ip:"] , lineContent = "host: qinhost ip: 1.1.1.1"
return {"host:": "qinhost", "ip": "1.1.1.1"}'''ksmatchedResult={}if len(keywordlistOneLine) == 0 or lineContent.strip() == "":return{}
linekey=getLineKey(keywordlistOneLine)ifempty(patternMap):
linePattern=getLinePattern(keywordlistOneLine)else:
linePattern=patternMap.get(linekey)
lineMatched=linePattern.findall(lineContent)ifempty(lineMatched):return{}
kslen=len(keywordlistOneLine)if kslen == 1:
ksmatchedResult[cleankey(keywordlistOneLine[0])]=lineMatched[0].strip()else:for i inrange(kslen):
ksmatchedResult[cleankey(keywordlistOneLine[i])]=lineMatched[0][i].strip()returnksmatchedResultdefempty(obj):return obj is None or len(obj) ==0defcleankey(dirtykey):'''clean unused characters in key'''
return re.sub(r"[# :]", "", dirtykey)defprintMatched(allMatched, linesConf):
allks=[]for kslist inlinesConf:
allks.extend(kslist)for matched inallMatched:for k inallks:print cleankey(k) , "=>", matched.get(cleankey(k))print '\n'
defbuildLinePatternMap(linesConf):
linePatternMap={}for keywordlistOneLine inlinesConf:
linekey=getLineKey(keywordlistOneLine)
linePatternMap[linekey]=getLinePattern(keywordlistOneLine)returnlinePatternMapdefgetLineKey(keywordlistForOneLine):return "_".join(keywordlistForOneLine)defgetLinePattern(keywordlistForOneLine):returnre.compile(produceRegex(keywordlistForOneLine))deftestMatchOneLine():assert len(matchOneLine([], "haha", {})) ==0assert len(matchOneLine(["host"], "", {})) ==0assert len(matchOneLine("", "haha", {})) ==0assert len(matchOneLine(["host", "ip"], "host:qqq addr: 1.1.1.1", {})) ==0
lineMatchMap1= matchOneLine(["id:"], "id: 123456", {"id:": re.compile(produceRegex(["id:"]))})assert lineMatchMap1.get("id") == "123456"lineMatchMap2= matchOneLine(["host:", "ip:"], "host: qinhost ip: 1.1.1.1", {"host:_ip:": re.compile(produceRegex(["host:", "ip:"]))})assert lineMatchMap2.get("host") == "qinhost"
assert lineMatchMap2.get("ip") == "1.1.1.1"
print 'testMatchOneLine passed.'
if __name__ == '__main__':
testMatchOneLine()
files= globalConf['files']
linesConf= globalConf['ksconf']
sepLines= map(lambdaks:ks[0], linesConf)
text=readAllFiles(files)
wholeRegex=produceRegex(sepLines)print 'wholeRegex:', wholeRegex
compiledPattern= re.compile(wholeRegex, flags=re.DOTALL+re.MULTILINE)
allMatched=findInText(compiledPattern, text, linesConf)
printMatched(allMatched, linesConf)
如果想以下多行解析文本文件,只需要修改下 ksconf = [['id:'], ['name:'], ['able:']]。
id:1name:shu
able:swim,study
id:2name:qin
able:sleep,run