3

Related to save to JDBC, trying to import a text file and save to a Hive JDBC file for import by reporting tools.

We are running spark-1.5.1-bin-hadoop2.6 (master + 1 slave), the JDBC thrift server and the beeline client. They all seem to interconnect and communicate. From what I can understand, Hive is included in this release in the datanucleus jars. I have configured directories to hold the Hive files, but have no conf/hive-config.xml.

Simple input CSV file:

Administrator,FiveHundredAddresses1,92121
Ann,FiveHundredAddresses2,92109
Bobby,FiveHundredAddresses3,92101
Charles,FiveHundredAddresses4,92111

The users table has been pre-created in the beeline client using

 CREATE TABLE users(first_name STRING, last_name STRING, zip_code STRING);
 show tables;    // it's there

For the scala REPL session on the master:

 val connectionUrl = "jdbc:hive2://x.y.z.t:10000/users?user=blah&password="
 val userCsvFile = sc.textFile("/home/blah/Downloads/Users4.csv")
 case class User(first_name:String, last_name:String, work_zip:String)
 val users = userCsvFile.map(_.split(",")).map(l => User(l(0), l(1), l(2)))
 val usersDf = sqlContext.createDataFrame(users)
 usersDf.count()  // 4
 usersDf.schema  // res92: org.apache.spark.sql.types.StructType = StructType(StructField(first_name,StringType,true), StructField(last_name,StringType,true), StructField(work_zip,StringType,true))
 usersDf.insertIntoJDBC(connectionUrl,"users",true)

OR

 usersDf.createJDBCTable(connectionUrl, "users", true)  // w/o beeline creation

OR

val properties = new java.util.Properties
properties.setProperty("user", "blah")
properties.setProperty("password", "blah")
val connectionUrl = "jdbc:hive2://172.16.3.10:10000"
contactsDf.write.jdbc(connectionUrl,"contacts", properties)

throws 

warning: there were 1 deprecation warning(s); re-run with -deprecation for details
java.sql.SQLException: org.apache.spark.sql.AnalysisException: cannot recognize input near 'TEXT' ',' 'last_name' in column type; line 1 pos  
at org.apache.hive.jdbc.HiveStatement.execute(HiveStatement.java:296)
at org.apache.hive.jdbc.HiveStatement.executeUpdate(HiveStatement.java:406)
at org.apache.hive.jdbc.HivePreparedStatement.executeUpdate(HivePreparedStatement.java:119)
at org.apache.spark.sql.DataFrameWriter.jdbc(DataFrameWriter.scala:275)
at org.apache.spark.sql.DataFrame.insertIntoJDBC(DataFrame.scala:1629)

Any ideas where I'm going wrong? Can this version actually write JDBC files from a DataFrame?

Thanks for any help!

Jon

JP-SD
  • 41
  • 4

1 Answers1

1

After a lot of searching (now it works) you can do this in the REPL:

import org.apache.spark.sql.SaveMode
contactsDf.saveAsTable("contacts", SaveMode.Overwrite)

I also configured $SPARK_INSTALL_LOC/conf/hive-site.xml as follows:

<property>
  <name>javax.jdo.option.ConnectionURL</name>
  <value>jdbc:derby:;databaseName=metastore_db;create=true</value>
  <description>JDBC connect string for a JDBC metastore</description>
</property>

<property>
  <name>javax.jdo.option.ConnectionDriverName</name>
  <value>org.apache.derby.jdbc.EmbeddedDriver</value>
  <description>Driver class name for a JDBC metastore</description>
</property>

<property>
  <name>hive.metastore.warehouse.dir</name>
  <value>/user/hive-warehouse</value>
  <description>Where to store metastore data</description>
</property>

</configuration>

Another key is that with Derby as the Hive backing database you cannot (at least how I have it configured) have both the ThriftJdbc Server and a REPL running simultaneously because of Derby's threading limitations. However, maybe if it's reconfigured with Postgres or MySQL or the like simultaneous access might be possible.

JP-SD
  • 41
  • 4