Spark Hbase插入了太多打开的连接

时间:2018-12-06 19:01:45

标签: scala apache-spark hbase apache-zookeeper

我正在尝试在解析一些文本之后对Hbase进行插入,并且代码可以正常工作,但是我认为可以对其进行组织以提高性能。在下面的即时消息中以循环方式打开连接,我想知道我如何能够打开单个连接并将其用于所有插入。我想我需要连接到某个函数才能实现此功能。

def extractInfo(fp:String) = {
val p:Parser = new AutoDetectParser()
val fs = FileSystem.get(new java.net.URI("XXXXXXXXXX"), new Configuration())
val inputPath:Path = new Path(fp)
val is:InputStream = fs.open(inputPath)
val handler:BodyContentHandler = new BodyContentHandler(-1)
val metadata:Metadata = new Metadata()
try{
p.parse(is, handler, metadata, new ParseContext())
is.close()
val hand = handler.toString()
val gson = new Gson
val jsonTree = gson.toJsonTree(metadata)
val metaNode = jsonTree.getAsJsonObject().getAsJsonObject("metadata")
val jsonString = gson.toJson(metaNode)
if (hand.trim().isEmpty()){
println("no Text extracted", inputPath)
} else {
println("Success")
}
val fname = "ABC"
val configuration: Configuration = HBaseConfiguration.create()
configuration.set("hbase.zookeeper.quorum", "XXXX")
configuration.set("hbase.zookeeper.property.clientPort", "XXXX")
configuration.set("zookeeper.znode.parent", "/hbase-XXX") 
configuration.set("hbase.client.keyvalue.maxsize", "0")
val principal = System.getProperty("kerberosPrincipal", "XXXXX")
val keytabLocation = System.getProperty("kerberosKeytab", "XXXXXXXXX")
UserGroupInformation.setConfiguration(configuration)
UserGroupInformation.loginUserFromKeytab(principal, keytabLocation)
val connection = ConnectionFactory.createConnection(HBaseConfiguration.create(configuration))
val admin = connection.getAdmin
val hTable:HTable = new HTable(configuration, "XXXXXXXXX")
val g = new Put(Bytes.toBytes(fname))
g.add(Bytes.toBytes("txt"),Bytes.toBytes("text"),Bytes.toBytes(hand))
hTable.put(g)
val m = new Put(Bytes.toBytes(fname))
m.add(Bytes.toBytes("data"),Bytes.toBytes("info"),Bytes.toBytes(jsonString))
hTable.put(m)
hTable.close()
fs.close()
}
catch {
case e : Throwable => {
println(e.printStackTrace)
}
}
} 


object App {
def main(args : Array[String]) {
val fnames = "/X/X/XXXXX.XXX"
fnames.foreach{x => extractInfo(x) }
}
}

1 个答案:

答案 0 :(得分:1)

首先,如果要从执行程序而不是驱动程序更新Hbase,则为每个执行程序创建连接,因此连接将在同一执行程序中重用。这样,您可以节省连接创建时间,但是每个线程都可以创建表对象,因为hbase表对象不是线程安全的(请参阅Hbase客户端官方文档)。

当然,只要完成,就关闭表和连接。

plot(g,edge.color=E(g)$color)

============如上面的代码示例=========

// this object will make connection reside on executor side not Driver
// and serve as a singleton per executor JVM, which makes connection shared between executor threads, connection is thread-safe!
object HbaseHandler {
  var connection: Option[Connection] = None

  def put(put: Put): Unit = {
    if(connection.isEmpty) {
      val conn = ConnectionFactory.createConnection(HBaseConfiguration.create(configuration))

      connection = Some(conn)
    }

    connection.get.<do some stuff>
    val table = ...
    try {
      table.put(put)
    } finally {  
      table.close()
    }
  }
} 

...

rdd.foreach (
  row => {
    val put: Put = <generate put object>
    HbaseHandler.put(put)
  }
)
相关问题