Update It turns out this has something to do with the way the Databricks Spark CSV reader is creating the DataFrame. In the example below that does not work, I read the people and address CSV using Databricks CSV reader, then write the resulting DataFrame to HDFS in Parquet format.
I changed the code to create the DataFrame with: (similar for the people.csv)
JavaRDD<Address> address = context.textFile("/Users/sfelsheim/data/address.csv").map(
new Function<String, Address>() {
public Address call(String line) throws Exception {
String[] parts = line.split(",");
Address addr = new Address();
addr.setAddrId(parts[0]);
addr.setCity(parts[1]);
addr.setState(parts[2]);
addr.setZip(parts[3]);
return addr;
}
});
and then write the resulting DataFrame to HDFS in Parquet format, and the join works as expected
I am reading the exact same CSV in both cases.
Running into an issue trying to perform a simple join of two DataFrames created from two different parquet files on HDFS.
[main] INFO org.apache.spark.SparkContext - Running Spark version 1.4.1
Using HDFS from Hadoop 2.7.0
Here is a sample to illustrate.
public void testStrangeness(String[] args) {
SparkConf conf = new SparkConf().setMaster("local[*]").setAppName("joinIssue");
JavaSparkContext context = new JavaSparkContext(conf);
SQLContext sqlContext = new SQLContext(context);
DataFrame people = sqlContext.parquetFile("hdfs://localhost:9000//datalake/sample/people.parquet");
DataFrame address = sqlContext.parquetFile("hdfs://localhost:9000//datalake/sample/address.parquet");
people.printSchema();
address.printSchema();
// yeah, works
DataFrame cartJoin = address.join(people);
cartJoin.printSchema();
// boo, fails
DataFrame joined = address.join(people,
address.col("addrid").equalTo(people.col("addressid")));
joined.printSchema();
}
Contents of people
first,last,addressid
your,mom,1
fred,flintstone,2
Contents of address
addrid,city,state,zip
1,sometown,wi,4444
2,bedrock,il,1111
people.printSchema();
results in...
root
|-- first: string (nullable = true)
|-- last: string (nullable = true)
|-- addressid: integer (nullable = true)
address.printSchema();
results in...
root
|-- addrid: integer (nullable = true)
|-- city: string (nullable = true)
|-- state: string (nullable = true)
|-- zip: integer (nullable = true)
DataFrame cartJoin = address.join(people);
cartJoin.printSchema();
Cartesian join works fine, printSchema() results in...
root
|-- addrid: integer (nullable = true)
|-- city: string (nullable = true)
|-- state: string (nullable = true)
|-- zip: integer (nullable = true)
|-- first: string (nullable = true)
|-- last: string (nullable = true)
|-- addressid: integer (nullable = true)
This join...
DataFrame joined = address.join(people,
address.col("addrid").equalTo(people.col("addressid")));
Results in the following exception.
Exception in thread "main" org.apache.spark.sql.AnalysisException: **Cannot resolve column name "addrid" among (addrid, city, state, zip);**
at org.apache.spark.sql.DataFrame$$anonfun$resolve$1.apply(DataFrame.scala:159)
at org.apache.spark.sql.DataFrame$$anonfun$resolve$1.apply(DataFrame.scala:159)
at scala.Option.getOrElse(Option.scala:121)
at org.apache.spark.sql.DataFrame.resolve(DataFrame.scala:158)
at org.apache.spark.sql.DataFrame.col(DataFrame.scala:558)
at dw.dataflow.DataflowParser.testStrangeness(DataflowParser.java:36)
at dw.dataflow.DataflowParser.main(DataflowParser.java:119)
I tried changing it so people and address have a common key attribute (addressid) and used..
address.join(people, "addressid");
But got the same result.
Any ideas??
Thanks