Spark读取Hbase的数据

val conf = HBaseConfiguration.create()
conf.addResource(new Path("/opt/cloudera/parcels/CDH-5.4.4-1.cdh5.4.4.p0.4/lib/hbase/conf/hbase-site.xml"))
conf.addResource(new Path("/opt/cloudera/parcels/CDH-5.4.4-1.cdh5.4.4.p0.4/lib/hadoop/etc/hadoop/core-site.xml"))
conf.set(TableInputFormat.INPUT_TABLE, "FLOW") //添加过滤条件,年龄大于 18 岁
//val scan = new Scan()
//conf.set(TableInputFormat.SCAN, convertScanToString(scan))
/*
scan.setFilter(new SingleColumnValueFilter("basic".getBytes, "age".getBytes,
CompareOp.GREATER_OR_EQUAL, Bytes.toBytes(18)))
*/ val usersRDD = sc.newAPIHadoopRDD(conf, classOf[TableInputFormat],
classOf[org.apache.hadoop.hbase.io.ImmutableBytesWritable],
classOf[org.apache.hadoop.hbase.client.Result]) val data1 = usersRDD.count() val sf = new SimpleDateFormat("yyyy-MM-dd HH:mm:ss.SSSSS") println("data length:" + data1) var map = HashMap[String, HashMap[String, collection.mutable.ArrayBuffer[Double]]]() usersRDD.collect().map {
case (_, result) =>
val key = Bytes.toInt(result.getRow)
println("Key:" + key)
val ip = Bytes.toString(result.getValue("F".getBytes, "SADDR".getBytes))
val port = Bytes.toString(result.getValue("F".getBytes, "SPORT".getBytes))
val startTimeLong = Bytes.toString(result.getValue("F".getBytes, "STIME".getBytes))
val endTimeLong = Bytes.toString(result.getValue("F".getBytes, "LTIME".getBytes))
val protocol = Bytes.toString(result.getValue("F".getBytes, "PROTO".getBytes))
val sumTime = Bytes.toString(result.getValue("F".getBytes, "DUR".getBytes))
val sum = Bytes.toString(result.getValue("F".getBytes, "DBYTES".getBytes)).toDouble println("ip:" + ip + ",port:" + port + ",startTime:" + startTimeLong + ",endTime:" + endTimeLong + ",protocol:" + protocol + ",sum:" + sum) //ip+port+udp,14:02 14:07 List
//ip+port+tcp,15:02 15:07 List
val startTimeDate = sf.parse(startTimeLong)
val endTimeLongDate = sf.parse(endTimeLong)
val startHours = startTimeDate.getHours
val startMinutes = startTimeDate.getMinutes val endHours = endTimeLongDate.getHours
val endMinutes = endTimeLongDate.getMinutes val key1 = ip + "_" + port + "_" + protocol
println("key1:" + key1) val key2 = startHours + ":" + startMinutes + "_" + endHours + ":" + endMinutes println("key2:" + key2) val tmpMap = map.get(key1) if (!tmpMap.isEmpty) {
println("--------------------map is not null:" + tmpMap.size + "--------------------")
val sumArray = tmpMap.get.get(key2)
if (!sumArray.isEmpty) {
sumArray.get += sum
}
} else {
println("--------------------map is null--------------------")
//如果当前Key不存在的话,是一个全新的Ip
val sumArray = collection.mutable.ArrayBuffer[Double]()
sumArray += sum val secondMap = HashMap[String, collection.mutable.ArrayBuffer[Double]]()
secondMap += (key2 -> sumArray)
map += (key1 -> secondMap)
}
map
println("map size-----------------:" + map.size)
} println("map size:" + map.size) map.map(e => {
println("--------------------Statistics start --------------------")
val resultKey1 = e._1
val resultVal1 = e._2
println("resultKey1:" + resultKey1)
resultVal1.foreach(f => {
val resultKey2 = f._1
val resultVal2 = f._2
println("resultKey2:" + resultKey2)
println("-----------------resultVal2:" + resultVal2.length) resultVal2.map(f=>{
println("------------------------f:"+f)
}) val dataArray = resultVal2.map(f => Vectors.dense(f)) val summary: MultivariateStatisticalSummary = Statistics.colStats(sc.parallelize(dataArray)) //
println("--------------------mean:" + summary.mean + " --------------------")
println("--------------------variance:" + summary.variance + " --------------------") println("--------------------mean apply 0:" + summary.mean.toArray.apply(0) + " --------------------")
println("--------------------variance apply 0:" + summary.variance.apply(0) + " --------------------") val upbase = summary.mean.toArray.apply(0) + 1.960 * Math.sqrt(summary.variance.apply(0))
val downbase = summary.mean.toArray.apply(0) - 1.960 * Math.sqrt(summary.variance.apply(0))
println("------------------- " + upbase + " ---------- " + downbase)
val df = new DecimalFormat(".##")
val upbaseString = df.format(upbase)
val downbaseString = df.format(downbase)
//resultMap.put(key, value)
val result3 = HashMap[Double, Double]()
//result3 +=(upbase -> downbase)
println("ip port:" + resultKey1 + ",time:" + resultKey2 + ",upbase:" + upbase + ",downbase:" + downbase)
})
}) println("--------------------baseLine end --------------------")
sc.stop()
上一篇:用jq获取元素内文本,但不包括其子元素内的文本值的方法


下一篇:使用百度网盘+Git,把版本控制托管到云端