PROSAGA码农传奇-hbase-如何使用没有HBase API的Spark直接编辑HBase HFile

<div class =“post-text”itemprop =“text”>
  
    回答自己，以防其他人需要这个。
  
  
    可以从HBase快照加载HFile。请遵循以下步骤：
（在HBase shell中）
 1.禁用'namespace：table'
 2.快照'命名空间：表''your_snapshot'
  
  
    这将创建一个可访问的快照，您可以访问/ [HBase_path] /.snapshot/[your_snapshot]
  
  
    将快照加载为RDD [ImmutableBytesWritable，Result]
  
   <pre>
 <code>
 def loadFromSnapshot(sc: SparkContext): RDD[ImmutableBytesWritable, Result] = {

val restorePath =
  new Path(s"hdfs://$storageDirectory/$restoreDirectory/$snapshotName")
val restorePathString = restorePath.toString

// create hbase conf starting from spark's hadoop conf
val hConf = HBaseConfiguration.create()
val hadoopConf = sc.hadoopConfiguration
HBaseConfiguration.merge(hConf, hadoopConf)

// point HBase root dir to snapshot dir
hConf.set("hbase.rootdir",
  s"hdfs://$storageDirectory/$snapshotDirectory/$snapshotName/")

// point Hadoop to the bucket as default fs
hConf.set("fs.default.name", s"hdfs://$storageDirectory/")

// configure serializations
hConf.setStrings("io.serializations",
  hadoopConf.get("io.serializations"),
  classOf[MutationSerialization].getName,
  classOf[ResultSerialization].getName,
  classOf[KeyValueSerialization].getName)

// disable caches
hConf.setFloat(HConstants.HFILE_BLOCK_CACHE_SIZE_KEY, HConstants.HFILE_BLOCK_CACHE_SIZE_DEFAULT)
hConf.setFloat(HConstants.BUCKET_CACHE_SIZE_KEY, 0f)
hConf.unset(HConstants.BUCKET_CACHE_IOENGINE_KEY)

// configure TableSnapshotInputFormat
hConf.set("hbase.TableSnapshotInputFormat.snapshot.name", settingsAccessor.settings.snapshotName)
hConf.set("hbase.TableSnapshotInputFormat.restore.dir", restorePathString)

val scan = new Scan()     // Fake scan which is applied by spark on HFile. Bypass RPC
val scanString = {
  val proto = ProtobufUtil.toScan(scan)
  Base64.encodeBytes(proto.toByteArray)
}
hConf.set(TableInputFormat.SCAN, scanString)

val job = Job.getInstance(hConf)

TableSnapshotInputFormat.setInput(job, settingsAccessor.settings.snapshotName, restorePath)

// create RDD
sc.newAPIHadoopRDD(job.getConfiguration,
  classOf[TableSnapshotInputFormat],
  classOf[ImmutableBytesWritable],
  classOf[Result])
}

</code>
 </pre>
  
    这将从快照目录加载HFile并对它们应用“假”全扫描，这可以避免慢速远程过程调用，但允许具有相同的扫描输出。
  
  
    完成后，您可以重新启用表格
  
  <UL>
    <LI>
      启用'nasmespace：table'
您也可以选择删除快照（实际上不会删除任何数据）
    </LI>
    <LI>
      delete_snapshot'the your_snapshot'
    </LI>
  </UL>
</DIV>