Configure standalone spark for azure storage access - azure

Question

I have a need to be able to run spark on my local machine to access azure wasb and adl urls, but I can't get it to work. I have a stripped down example here:
maven pom.xml (Brand-new pom, only the dependencies have been set):
<dependencies>
<dependency>
<groupId>org.apache.spark</groupId>
<artifactId>spark-core_2.11</artifactId>
<version>2.3.0</version>
</dependency>
<dependency>
<groupId>org.apache.spark</groupId>
<artifactId>spark-sql_2.11</artifactId>
<version>2.3.0</version>
</dependency>
<dependency>
<groupId>org.apache.hadoop</groupId>
<artifactId>hadoop-common</artifactId>
<version>2.8.0</version>
</dependency>
<dependency>
<groupId>org.apache.hadoop</groupId>
<artifactId>hadoop-azure-datalake</artifactId>
<version>3.1.0</version>
</dependency>
<dependency>
<groupId>com.microsoft.azure</groupId>
<artifactId>azure-storage</artifactId>
<version>6.0.0</version>
</dependency>
<dependency>
<groupId>com.microsoft.azure</groupId>
<artifactId>azure-data-lake-store-sdk</artifactId>
<version>2.2.3</version>
</dependency>
<dependency>
<groupId>org.apache.hadoop</groupId>
<artifactId>hadoop-azure</artifactId>
<version>3.1.0</version>
</dependency>
<dependency>
<groupId>com.microsoft.azure</groupId>
<artifactId>azure-storage</artifactId>
<version>7.0.0</version>
</dependency>
Java code (Doesn't need to be java - could be scala):
import org.apache.spark.SparkConf;
import org.apache.spark.SparkContext;
import org.apache.spark.sql.SparkSession;
public class App {
public static void main(String[] args) {
SparkConf config = new SparkConf();
config.setMaster("local");
config.setAppName("app");
SparkSession spark = new SparkSession(new SparkContext(config));
spark.read().parquet("wasb://container#host/path");
spark.read().parquet("adl://host/path");
}
}
No matter what I try I get:
Exception in thread "main" java.io.IOException: No FileSystem for scheme: wasb
Same for adl. Every doc I can find on this either just says to add the azure-storage dependency, which I have done, or says to use HDInsight.
Any thoughts?

Answer 1

I figured this out and decided to post a working project since that is always what I look for. It is hosted here:
azure-spark-local-sample
The crux of it though is as #Shankar Koirala suggested:
For WASB, set the property to allow the url scheme to be recognized:
config.set("spark.hadoop.fs.wasb.impl", "org.apache.hadoop.fs.azure.NativeAzureFileSystem");
Then set the property which authorizes access to the account. You will need one of these for each account you need to access. These are generated through the Azure Portal under the Access Keys section of the Storage Account blade.
config.set("fs.azure.account.key.[storage-account-name].blob.core.windows.net", "[access-key]");
Now for adl, assign the fs scheme as with WASB:
config.set("spark.hadoop.fs.adl.impl", "org.apache.hadoop.fs.adl.AdlFileSystem");
// I don't know why this would be needed, but I saw it
// on an otherwise very helpful page . . .
config.set("spark.fs.AbstractFileSystem.adl.impl", "org.apache.hadoop.fs.adl.Adl");
. . . and finally, set the client access keys in these properties, again for each different account you need to access:
config.set("fs.adl.oauth2.access.token.provider.type", "ClientCredential");
/* Client ID is generally the application ID from the azure portal app registrations*/
config.set("fs.adl.oauth2.client.id", "[client-id]");
/*The client secret is the key generated through the portal*/
config.set("fs.adl.oauth2.credential", "[client-secret]");
/*This is the OAUTH 2.0 TOKEN ENDPOINT under the ENDPOINTS section of the app registrations under Azure Active Directory*/
config.set("fs.adl.oauth2.refresh.url", "[oauth-2.0-token-endpoint]");
I hope this is helpful, and I wish I could give credit to Shankar for the answer, but I also wanted to get the exact details out there.

Answer 2

I am not sure about the adl haven't tested but for the wasb you need to define the file system to be used in the underlying Hadoop configurations.
Since you are using spark 2.3 you can use spark session to create a entry point as
val spark = SparkSession.builder().appName("read from azure storage").master("local[*]").getOrCreate()
Now define the file system
spark.sparkContext.hadoopConfiguration.set("fs.azure", "org.apache.hadoop.fs.azure.NativeAzureFileSystem")
spark.sparkContext.hadoopConfiguration.set("fs.azure.account.key.yourAccount.blob.core.windows.net", "yourKey ")
Now read the parquet file as
val baseDir = "wasb[s]://BlobStorageContainer#yourUser.blob.core.windows.net/"
val dfParquet = spark.read.parquet(baseDir + "pathToParquetFile")
Hope this helps!