1.在生产环境下,如何处理配置文件 && 表的数据处理
配置文件,或者配置表,一般是放在在线db,比如mysql等关系型数据库,或者后台rd直接丢给你一份文件,数据量比起整个离线数据仓库的大表来说算很小,所以这种情况下,一般的做法是将小表,或者小文件广播出去,那么下面一个例子来看,广播表的使用解决ip地址映射问题
数据地址:链接:https://pan.baidu.com/s/1FmFxSrPIynO3udernLU0yQ提取码:hell
2.日志分析案例1
2.1 数据说明
http.log:
用户访问网站所产生的日志。日志格式为:时间戳、IP地址、访问网址、访问数据、浏览器信息等,样例如下:
ip.dat:ip段数据,记录着一些ip段范围对应的位置,总量大概在11万条,数据量也算很小的,样例如下
文件位置:data/http.log、data/ip.dat
链接:https://pan.baidu.com/s/1FmFxSrPIynO3udernLU0yQ提取码:hell
要求:将 http.log 文件中的 ip 转换为地址。如将 122.228.96.111 转为温州,并统计各城市的总访问量
2.2.实现思路和代码如下
有三个关键点,http.log的关键信息是ip地址,所以根据数据的精简原则,只读取ip即可,另外ip映射比对的时候 ,ip地址映射文件是排序的,所以为了提高查找效率,采用将ip地址转为long类型,然后再用二分法来查找,找到地址后映射为地址。
package com.hoult.work
import org.apache.spark.broadcast.Broadcast
import org.apache.spark.rdd.RDD
import org.apache.spark.sql.SparkSession
/**
* 数据源:1.ip地址的访问日志 2.ip地址映射表
* 需要把映射表广播,地址转换为long类型进行比较
*/
object FindIp {
def main(args: Array[String]): Unit = {
val spark = SparkSession
.builder()
.master("local[*]")
.appName(this.getClass.getCanonicalName)
.getOrCreate()
val sc = spark.sparkContext
import spark.implicits._
val ipLogsRDD = sc.textFile("data/http.log")
.map(_.split("\|")(1))
val ipInfoRDD = sc.textFile("data/ip.dat").map {
case line: String => {
val strSplit: Array[String] = line.split("\|")
Ip(strSplit(0), strSplit(1), strSplit(7))
}
}
val brIPInfo = sc.broadcast(ipInfoRDD.map(x => (ip2Long(x.startIp), ip2Long(x.endIp), x.address))collect())
//关联后的结果rdd
ipLogsRDD
.map(x => {
val index = binarySearch(brIPInfo.value, ip2Long(x))
if (index != -1 )
brIPInfo.value(index)._3
else
"NULL"
}).map(x => (x, 1))
.reduceByKey(_ + _)
.map(x => s"城市:${x._1}, 访问量:${x._2}")
.saveAsTextFile("data/work/output_ips")
}
//ip转成long类型
def ip2Long(ip: String): Long = {
val fragments = ip.split("[.]")
var ipNum = 0L
for (i <- 0 until fragments.length) {
ipNum = fragments(i).toLong | ipNum << 8L
}
ipNum
}
//二分法匹配ip规则
def binarySearch(lines: Array[(Long, Long, String)], ip: Long): Int = {
var low = 0
var high = lines.length - 1
while (low <= high) {
val middle = (low + high) / 2
if ((ip >= lines(middle)._1) && (ip <= lines(middle)._2))
return middle
if (ip < lines(middle)._1)
high = middle - 1
else {
low = middle + 1
}
}
-1
}
}
case class Ip(startIp: String, endIp: String, address: String)
结果截图如下:
3.日志分析案例2
3.1 数据说明
日志格式:IP命中率(Hit/Miss)响应时间请求时间请求方法请求URL请求协议状态码响应大小referer 用户代理
日志文件位置:data/cdn.txt
数据case:
任务:
2.1、计算独立IP数
2.2、统计每个视频独立IP数(视频的标志:在日志文件的某些可以找到 *.mp4,代表一个视频文件)
2.3、统计一天中每个小时的流量
分析:刚开始去找格林时间的jod-time解析,找了一圈不知道该怎么写, 后面发现只需要小时即可,使用正则来提取, 注意在求video的访问ip时候,可以用aggregateByKey
来提高性能
3.2 实现代码
package com.hoult.work
import org.apache.spark.rdd.RDD
import org.apache.spark.sql.SparkSession
/**
* 读取日志表到rdd
* 拿到需要的字段:ip, 访问时间:小时即可, 视频名video_name (url中的xx.mp4),
* 分析:
* 1.计算独立IP数
* 2.统计每个视频独立IP数(视频的标志:在日志文件的某些可以找到 *.mp4,代表一个视频文件)
* 3.统计一天中每个小时的流量
*/
object LogAnaylse {
def main(args: Array[String]): Unit = {
val spark = SparkSession
.builder()
.master("local[*]")
.appName(this.getClass.getCanonicalName)
.getOrCreate()
val sc = spark.sparkContext
val cdnRDD = sc.textFile("data/cdn.txt")
//计算独立ips
// aloneIPs(cdnRDD.repartition(1))
//每个视频独立ip数
// videoIPs(cdnRDD.repartition(1))
//每小时流量
hourPoor(cdnRDD.repartition(1))
}
/**
* 独立ip数
*/
def aloneIPs(cdnRDD: RDD[String]) = {
//匹配ip地址
val IPPattern = "((?:(?:25[0-5]|2[0-4]\d|((1\d{2})|([1-9]?\d)))\.){3}(?:25[0-5]|2[0-4]\d|((1\d{2})|([1-9]?\d))))".r
val ipnums = cdnRDD
.flatMap(x => (IPPattern findFirstIn x))
.map(y => (y,1))
.reduceByKey(_+_)
.sortBy(_._2,false)
ipnums.saveAsTextFile("data/cdn/aloneIPs")
}
/**
* 视频独立ip数
*/
def videoIPs(cdnRDD: RDD[String]) = {
//匹配 http 响应码和请求数据大小
val httpSizePattern = ".*\s(200|206|304)\s([0-9]+)\s.*".r
//[15/Feb/2017:11:17:13 +0800] 匹配 2017:11 按每小时播放量统计
val timePattern = ".*(2017):([0-9]{2}):[0-9]{2}:[0-9]{2}.*".r
import scala.util.matching.Regex
// Entering paste mode (ctrl-D to finish)
def isMatch(pattern: Regex, str: String) = {
str match {
case pattern(_*) => true
case _ => false
}
}
def getTimeAndSize(line: String) = {
var res = ("", 0L)
try {
val httpSizePattern(code, size) = line
val timePattern(year, hour) = line
res = (hour, size.toLong)
} catch {
case ex: Exception => ex.printStackTrace()
}
res
}
val IPPattern = "((?:(?:25[0-5]|2[0-4]\d|((1\d{2})|([1-9]?\d)))\.){3}(?:25[0-5]|2[0-4]\d|((1\d{2})|([1-9]?\d))))".r
val videoPattern = "([0-9]+).mp4".r
val res = cdnRDD
.filter(x => x.matches(".*([0-9]+)\.mp4.*"))
.map(x => (videoPattern findFirstIn x toString,IPPattern findFirstIn x toString))
.aggregateByKey(List[String]())(
(lst, str) => (lst :+ str),
(lst1, lst2) => (lst1 ++ lst2)
)
.mapValues(_.distinct)
.sortBy(_._2.size,false)
res.saveAsTextFile("data/cdn/videoIPs")
}
/**
* 一天中每个小时的流量
*
*/
def hourPoor(cdnRDD: RDD[String]) = {
val httpSizePattern = ".*\s(200|206|304)\s([0-9]+)\s.*".r
val timePattern = ".*(2017):([0-9]{2}):[0-9]{2}:[0-9]{2}.*".r
import scala.util.matching.Regex
def isMatch(pattern: Regex, str: String) = {
str match {
case pattern(_*) => true
case _ => false
}
}
def getTimeAndSize(line: String) = {
var res = ("", 0L)
try {
val httpSizePattern(code, size) = line
val timePattern(year, hour) = line
res = (hour, size.toLong)
} catch {
case ex: Exception => ex.printStackTrace()
}
res
}
cdnRDD
.filter(x=>isMatch(httpSizePattern,x))
.filter(x=>isMatch(timePattern,x))
.map(x=>getTimeAndSize(x))
.groupByKey()
.map(x=>(x._1,x._2.sum))
.sortByKey()
.map(x=>x._1+"时 CDN流量="+x._2/(102424*1024)+"G")
.saveAsTextFile("data/cdn/hourPoor")
}
}
运行结果截图:
4. 广告曝光分析案例
假设点击日志文件(click.log)和曝光日志imp.log, 中每行记录格式如下
//点击日志
INFO 2019-09-01 00:29:53 requestURI:/click?app=1&p=1&adid=18005472&industry=469&adid=31
INFO 2019-09-01 00:30:31 requestURI:/click?app=2&p=1&adid=18005472&industry=469&adid=31
INFO 2019-09-01 00:31:03 requestURI:/click?app=1&p=1&adid=18005472&industry=469&adid=32
INFO 2019-09-01 00:31:51 requestURI:/click?app=1&p=1&adid=18005472&industry=469&adid=33
//曝光日志
INFO 2019-09-01 00:29:53 requestURI:/imp?app=1&p=1&adid=18005472&industry=469&adid=31
INFO 2019-09-01 00:29:53 requestURI:/imp?app=1&p=1&adid=18005472&industry=469&adid=31
INFO 2019-09-01 00:29:53 requestURI:/imp?app=1&p=1&adid=18005472&industry=469&adid=34
用Spark-Core实现统计每个adid的曝光数与点击数,思路较简单,直接上代码
代码:
package com.hoult.work
import org.apache.spark.sql.SparkSession
object AddLog {
def main(args: Array[String]): Unit = {
val spark = SparkSession
.builder()
.master("local[*]")
.appName(this.getClass.getCanonicalName)
.getOrCreate()
val sc = spark.sparkContext
val clickRDD = sc.textFile("data/click.log")
val impRDD = sc.textFile("data/imp.log")
val clickRes = clickRDD.map{line => {
val arr = line.split("\s+")
val adid = arr(3).substring(arr(3).lastIndexOf("=") + 1)
(adid, 1)
}}.reduceByKey(_ + _)
val impRes = impRDD.map { line =>
val arr = line.split("\s+")
val adid = arr(3).substring(arr(3).lastIndexOf("=") + 1)
(adid, 1)
}.reduceByKey(_ + _)
//保存到hdfs
clickRes.fullOuterJoin(impRes)
.map(x => x._1 + "," + x._2._1.getOrElse(0) + "," + x._2._2.getOrElse(0))
.repartition(1)
// .saveAsTextFile("hdfs://linux121:9000/data/")
.saveAsTextFile("data/add_log")
sc.stop()
}
}
分析:共有两次shuffle
, fulljon
可以修改为union + reduceByKey
,将shuffle
减少到一次
5.使用spark-sql完成下面的转换
A表有三个字段:ID、startdate、enddate,有3条数据:
1 2019-03-04 2020-02-03
2 2020-04-05 2020-08-04
3 2019-10-09 2020-06-11
写SQL(需要SQL和DSL)将以上数据变化为:
2019-03-04 2019-10-09
2019-10-09 2020-02-03
2020-02-03 2020-04-05
2020-04-05 2020-06-11
2020-06-11 2020-08-04
2020-08-04 2020-08-04
分析:观察,可以得到,第一列实际上是startdate 和 enddate两列叠加的结果,而第二列是下一个,可以用lead
窗口函数
代码如下:
package com.hoult.work
import org.apache.spark.sql.{DataFrame, SparkSession}
object DataExchange {
def main(args: Array[String]): Unit = {
val spark = SparkSession
.builder()
.appName("DateSort")
.master("local[*]")
.getOrCreate()
spark.sparkContext.setLogLevel("warn")
// 原数据
val tab = List((1, "2019-03-04", "2020-02-03"),(2, "2020-04-05", "2020-08-04"),(3, "2019-10-09", "2020-06-11"))
val df: DataFrame = spark.createDataFrame(tab).toDF("ID", "startdate", "enddate")
val dateset: DataFrame = df.select("startdate").union(df.select("enddate"))
dateset.createOrReplaceTempView("t")
val result: DataFrame = spark.sql(
"""
|select tmp.startdate, nvl(lead(tmp.startdate) over(partition by col order by tmp.startdate), startdate) enddate from
|(select "1" col, startdate from t) tmp
|""".stripMargin)
result.show()
}
}
运行结果:
吴邪,小三爷,混迹于后台,大数据,人工智能领域的小菜鸟。
更多请关注