windows - Spark 2.0: Relative path in absolute URI (spark-warehouse)

Question

I'm trying to migrate from Spark 1.6.1 to Spark 2.0.0 and I am getting a weird error when trying to read a csv file into SparkSQL. Previously, when I would read a file from local disk in pyspark I would do:

Spark 1.6

df = sqlContext.read 
        .format('com.databricks.spark.csv') 
        .option('header', 'true') 
        .load('file:///C:/path/to/my/file.csv', schema=mySchema)

In the latest release I think it should look like this:

Spark 2.0

spark = SparkSession.builder 
           .master('local[*]') 
           .appName('My App') 
           .getOrCreate()

df = spark.read 
        .format('csv') 
        .option('header', 'true') 
        .load('file:///C:/path/to/my/file.csv', schema=mySchema)

But I am getting this error no matter how many different ways I try to adjust the path:

IllegalArgumentException: 'java.net.URISyntaxException: Relative path in 
absolute URI: file:/C:/path//to/my/file/spark-warehouse'

Not sure if this is just an issue with Windows or there is something I am missing. I was excited that the spark-csv package is now a part of Spark right out of the box, but I can't seem to get it to read any of my local files anymore. Any ideas?

Answer 1

I was able to do some digging around in the latest Spark documentation, and I notice they have a new configuration setting that I hadn't noticed before:

spark.sql.warehouse.dir

So I went ahead and added this setting when I set up my SparkSession:

spark = SparkSession.builder 
           .master('local[*]') 
           .appName('My App') 
           .config('spark.sql.warehouse.dir', 'file:///C:/path/to/my/') 
           .getOrCreate()

That seems to set the working directory, and then I can just feed my filename directly into the csv reader:

df = spark.read 
        .format('csv') 
        .option('header', 'true') 
        .load('file.csv', schema=mySchema) 

Once I set the spark warehouse, Spark was able to locate all of my files and my app finishes successfully now. The amazing thing is that it runs about 20 times faster than it did in Spark 1.6. So they really have done some very impressive work optimizing their SQL engine. Spark it up!