Community Articles

Find and share helpful community-sourced technical articles.
Labels (2)
avatar
Rising Star

The following example provides a guide to connecting to HBase from Spark then perform a few operations using Spark on the data extracted from HBase

Please add the following repo to your package manager

http://repo.hortonworks.com/content/repositories/releases/

and import the following class

com.hortonworks:shc-core:1.1.0-2.1-s_2.11

Keep in mind that new releases may be available

Import the following classes

import org.apache.spark.sql.{SQLContext, _}
import org.apache.spark.sql.execution.datasources.hbase._
import org.apache.spark.{SparkConf, SparkContext}
import spark.sqlContext.implicits._
import org.apache.spark.sql.types._

The first step in connecting to HBase is defining how Spark should interpret each row of data found in HBase. This is done by defining a catalog to use with the connector

def catalog = s"""{
 |"table":{"namespace":"default", "name":"timeseries"},
 |"rowkey":"key",
 |"columns":{
  |"col0":{"cf":"rowkey", "col":"key", "type":"string"},
  |"col1":{"cf":"cf", "col":"t", "type":"string"},
  |"col2":{"cf":"cf", "col":"v", "type":"string"},
  |"col3":{"cf":"cf", "col":"id", "type":"string"}
 |}
|}""".stripMargin

Next we read from Hbase using this catalog

val df = sqlContext
 .read
 .options(Map(HBaseTableCatalog.tableCatalog->catalog.toString))
 .format("org.apache.spark.sql.execution.datasources.hbase")
 .load()

Since the available data formats for the catalog ( I believe these are limited to Avro datatypes ) you will almost always need to cast something into a more meaningful datatype. Also you can see here that I am using "split" to break the composite row key into its two parts

val df2 = df.withColumn("value", df("col2").cast(FloatType))
 .withColumn("tag", split(df("col0"), "-").getItem(0))
 .withColumn("t", unix_timestamp(split(df("col0"), "-").getItem(1), "yyyy/MM/dd HH:mm:ss").cast(TimestampType))

Now that we have a data frame containing the data we want to work with, we can perform SQL simply by registering the data frame as a temporary table.

df2.registerTempTable("Sine1")

Now we can use SQL to explore the data and apply other manipulations

select * from sine1

73397-sine1.png

2,854 Views
Comments
avatar
Contributor

Hi @wsalazar , Thanks for the nice explanation. I was wondering how you do it with composite keys ? We have spent sometime exploring phoenix encoder. Using this, data insertion is good but while reading and doing the range scan it somehow is very slow. Seems only the first part of the composite key is used for filter and rest of the key is not taken into account.