Hbase BulkLoad方式导入百亿级数据
2018-12-27 本文已影响20人
大猪大猪
Hbase banner众所周知Hbase的BulkLoad是最快导入数据的方式,在导入历史数据的时候,我们一般会选择使用BulkLoad方式,我们可以借助Spark的计算能力将数据快速地导入。
使用方法
- 导入依赖包
compile group: 'org.apache.spark', name: 'spark-sql_2.11', version: '2.3.1.3.0.0.0-1634'
compile group: 'org.apache.spark', name: 'spark-core_2.11', version: '2.0.0.3.0.0.0-1634'
compile group: 'org.apache.hbase', name: 'hbase-it', version: '2.0.0.3.0.0.0-1634'
- 创建好表与Family
- 编写核心代码
BulkLoad.scala
object BulkLoad {
def main(args: Array[String]): Unit = {
val sparkConf = new SparkConf()
// .setMaster("local[12]")
.setAppName("PhoenixBulkLoad")
val spark = SparkSession
.builder
.config(sparkConf)
.getOrCreate()
val sc = spark.sparkContext
val datas = List(
("abc", ("ext", "type", "login")),
("ccc", ("ext", "type", "logout"))
)
val dataRdd = sc.parallelize(datas)
val output = dataRdd.map {
x => {
val rowKey = Bytes.toBytes(x._1)
val immutableRowKey = new ImmutableBytesWritable(rowKey)
val colFam = x._2._1
val colName = x._2._2
val colValue = x._2._3
val kv = new KeyValue(
rowKey,
Bytes.toBytes(colFam),
Bytes.toBytes(colName),
Bytes.toBytes(colValue.toString)
)
(immutableRowKey, kv)
}
}
val hConf = HBaseConfiguration.create()
hConf.addResource("hbase-site.xml")
val hTableName = "test_log"
hConf.set("hbase.mapreduce.hfileoutputformat.table.name", hTableName)
val job = Job.getInstance(hConf, "HbaseBulkLoad")
job.setMapOutputKeyClass(classOf[ImmutableBytesWritable])
job.setMapOutputValueClass(classOf[KeyValue])
TableMapReduceUtil.initCredentials(job)
val tableName = TableName.valueOf(hTableName)
val conn = ConnectionFactory.createConnection(hConf)
val table = conn.getTable(tableName)
val tableDescr = table.getDescriptor
val regionLocator = conn.getRegionLocator(tableName)
HFileOutputFormat2.configureIncrementalLoad(job, tableDescr, regionLocator)
val hFileOutput = "/tmp/h_file"
output.saveAsNewAPIHadoopFile(hFileOutput,
classOf[ImmutableBytesWritable],
classOf[KeyValue],
classOf[HFileOutputFormat2],
hConf
)
val bulkLoader = new LoadIncrementalHFiles(hConf)
bulkLoader.doBulkLoad(new Path(hFileOutput), conn.getAdmin, table, regionLocator)
}
}
- 提交Spark任务
spark-submit --master yarn --conf spark.yarn.tokens.hbase.enabled=true --class com.dounine.hbase.BulkLoad --executor-memory 2G --num-executors 2G --driver-memory 2G --executor-cores 2 build/libs/hbase-data-insert-1.0.0-SNAPSHOT-all.jar