Python connecting to spark in standalone mode and accessing s3 with S3AFileSystem issues

Question

Setup

• Spark spark-2.4.5-bin-without-hadoop configured with hadoop-2.8.5.
• Spark is running in standalone mode.
• PySpark application runs as a separate container and connects to the master from another container.

What needs to happen?

• PySpark application should be able to connect to AWS S3 bucket by using org.apache.hadoop.fs.s3a.TemporaryAWSCredentialsProvider

Attempted solution #1

• Master, workers, and the PySpark application all have the following jars at /

hadoop-aws-2.8.5.jar
hadoop-common-2.8.5.jar 
aws-java-sdk-1.10.6.jar
jackson-core-2.2.3.jar
aws-java-sdk-s3-1.10.6.jar
jackson-databind-2.2.3.jar
jackson-annotations-2.2.3.jar
joda-time-2.9.4.jar

• PySpark application is configured the following way:

builder = SparkSession.builder

builder = builder.config("spark.executor.userClassPathFirst", "true")
builder = builder.config("spark.driver.userClassPathFirst", "true")

class_path = "/hadoop-aws-2.8.5.jar:/hadoop-common-2.8.5.jar:/aws-java-sdk-1.10.6.jar:/jackson-core-2.2.3.jar:/aws-java-sdk-s3-1.10.6.jar:/jackson-databind-2.2.3.jar:/jackson-annotations-2.2.3.jar:/joda-time-2.9.4.jar"
builder = builder.config("spark.driver.extraClassPath", class_path)
builder = builder.config("spark.executor.extraClassPath", class_path)

builder = builder.config("spark.hadoop.fs.s3a.aws.credentials.provider", "org.apache.hadoop.fs.s3a.TemporaryAWSCredentialsProvider")
builder = builder.config("spark.hadoop.fs.s3a.impl", "org.apache.hadoop.fs.s3a.S3AFileSystem")
builder = builder.config("spark.hadoop.fs.s3a.access.key", os.environ.get("AWS_ACCESS_KEY_ID"))
builder = builder.config("spark.hadoop.fs.s3a.secret.key", os.environ.get("AWS_SECRET_ACCESS_KEY"))
builder = builder.config("spark.hadoop.fs.s3a.session.token", os.environ.get("AWS_SESSION_TOKEN"))

Output

Exception in thread "main" java.lang.NoSuchMethodError: org.apache.hadoop.security.authentication.util.KerberosUtil.hasKerberosTicket(Ljavax/security/auth/Subject;)Z
....

• This was happening because hadoop-auth was missing. I downloaded hadoop-auth and got another error. It lead me to conclude that I would need to provide all the dependent jars.
• Instead of going manually through maven. I tried to import hadoop-aws into IntelliJ IDEA to see if it's going to import other dependencies and it gave me this list: (Don't know if there is a mvn command that would do the same thing)

accessors-smart-1.2.jar
activation-1.1.jar
apacheds-i18n-2.0.0-M15.jar
apacheds-kerberos-codec-2.0.0-M15.jar
api-asn1-api-1.0.0-M20.jar
api-util-1.0.0-M20.jar
asm-3.1.jar
asm-5.0.4.jar
avro-1.7.4.jar
aws-java-sdk-core-1.10.6.jar
aws-java-sdk-kms-1.10.6.jar
aws-java-sdk-s3-1.10.6.jar
commons-beanutils-1.7.0.jar
commons-beanutils-core-1.8.0.jar
commons-cli-1.2.jar
commons-codec-1.4.jar
commons-collections-3.2.2.jar
commons-compress-1.4.1.jar
commons-configuration-1.6.jar
commons-digester-1.8.jar
commons-io-2.4.jar
commons-lang-2.6.jar
commons-logging-1.1.3.jar
commons-math3-3.1.1.jar
commons-net-3.1.jar
curator-client-2.7.1.jar
curator-framework-2.7.1.jar
curator-recipes-2.7.1.jar
gson-2.2.4.jar
guava-11.0.2.jar
hadoop-annotations-2.8.5.jar
hadoop-auth-2.8.5.jar
hadoop-aws-2.8.5.jar
hadoop-common-2.8.5.jar
htrace-core4-4.0.1-incubating.jar
httpclient-4.5.2.jar
httpcore-4.4.4.jar
jackson-annotations-2.2.3.jar
jackson-core-2.2.3.jar
jackson-core-asl-1.9.13.jar
jackson-databind-2.2.3.jar
jackson-jaxrs-1.8.3.jar
jackson-mapper-asl-1.9.13.jar
jackson-xc-1.8.3.jar
java-xmlbuilder-0.4.jar
jaxb-api-2.2.2.jar
jaxb-impl-2.2.3-1.jar
jcip-annotations-1.0-1.jar
jersey-core-1.9.jar
jersey-json-1.9.jar
jersey-server-1.9.jar
jets3t-0.9.0.jar
jettison-1.1.jar
jetty-6.1.26.jar
jetty-sslengine-6.1.26.jar
jetty-util-6.1.26.jar
joda-time-2.9.4.jar
jsch-0.1.54.jar
json-smart-2.3.jar
jsp-api-2.1.jar
jsr305-3.0.0.jar
log4j-1.2.17.jar
netty-3.7.0.Final.jar
nimbus-jose-jwt-4.41.1.jar
paranamer-2.3.jar
protobuf-java-2.5.0.jar
servlet-api-2.5.jar
slf4j-api-1.7.10.jar
slf4j-log4j12-1.7.10.jar
snappy-java-1.0.4.1.jar
stax-api-1.0-2.jar
xmlenc-0.52.jar
xz-1.0.jar
zookeeper-3.4.6.jar

• I suspect these need to be included in PySpark application as well. It would be too long to download each .jar separately and probably is not the way to go.

Attempted solution #2

builder = SparkSession.builder

builder = builder.config("spark.jars.packages", "org.apache.hadoop:hadoop-aws:2.8.5,org.apache.hadoop:hadoop-common:2.8.5,com.amazonaws:aws-java-sdk-s3:1.10.6")

builder = builder.config("spark.hadoop.fs.s3a.aws.credentials.provider", "org.apache.hadoop.fs.s3a.TemporaryAWSCredentialsProvider")
builder = builder.config("spark.hadoop.fs.s3a.impl", "org.apache.hadoop.fs.s3a.S3AFileSystem")
builder = builder.config("spark.hadoop.fs.s3a.access.key", os.environ.get("AWS_ACCESS_KEY_ID"))
builder = builder.config("spark.hadoop.fs.s3a.secret.key", os.environ.get("AWS_SECRET_ACCESS_KEY"))
builder = builder.config("spark.hadoop.fs.s3a.session.token", os.environ.get("AWS_SESSION_TOKEN"))

• Instead of providing jars I use --packages it installs the dependencies as well. However, I end up with the following error when trying to read from S3:

java.lang.IllegalAccessError: tried to access method org.apache.hadoop.metrics2.lib.MutableCounterLong.<init>(Lorg/apache/hadoop/metrics2/MetricsInfo;J)V from class org.apache.hadoop.fs.s3a.S3AInstrumentation
...

My research

• Spark uses s3a: java.lang.NoSuchMethodError
• NoClassDefFoundError: org/apache/hadoop/fs/StreamCapabilities while reading s3 Data with spark
• https://hadoop.apache.org/docs/r3.1.0/hadoop-aws/tools/hadoop-aws/troubleshooting_s3a.html#java.lang.NoSuchMethodError_referencing_an_org.apache.hadoop_class

From that, I conclude that my problem has to do with hadoop-aws dependencies. However, I'm having a problem checking this. The versions of Hadoop running with Spark are the same as the jars that I use when connecting and I'm providing all the dependencies.

• Any tools or commands to run for debugging this issue would be helpful.

Answer 1

Try creating an admin user in your S3 bucket and connect with the admin credentials instead of using a temporary connection.

https://bartek-blog.github.io/python/spark/2019/04/22/how-to-access-s3-from-pyspark.html

Answer 2

I found this code to help you from this site

session = boto3.session.Session(profile_name=’MyUserProfile’)
sts_connection = session.client(‘sts’)
response = sts_connection.assume_role(RoleArn=’ARN_OF_THE_ROLE_TO_ASSUME’, RoleSessionName=’THIS_SESSIONS_NAME’,DurationSeconds=3600)
credentials = response[‘Credentials’]

url = str(‘s3a://bucket/key/data.csv’)

spark.sparkContext._jsc.hadoopConfiguration().set(‘fs.s3a.aws.credentials.provider’, ‘org.apache.hadoop.fs.s3a.TemporaryAWSCredentialsProvider’)
spark.sparkContext._jsc.hadoopConfiguration().set(‘fs.s3a.access.key’, credentials[‘AccessKeyId’])
spark.sparkContext._jsc.hadoopConfiguration().set(‘fs.s3a.secret.key’, credentials[‘SecretAccessKey’])
spark.sparkContext._jsc.hadoopConfiguration().set(‘fs.s3a.session.token’, credentials[‘SessionToken’])
spark.read.csv(url).show(1)