Configuring an Amazon EMR cluster for Kerberos-authenticated HDFS users and SSH connections

Amazon EMR creates Kerberos-authenticated user clients for the applications that run on the cluster—for example, the hadoop user, spark user, and others. You can also add users who are authenticated to cluster processes using Kerberos. Authenticated users can then connect to the cluster with their Kerberos credentials and work with applications. For a user to authenticate to the cluster, the following configurations are required:

Adding Linux users and Kerberos principals to the primary node

If you do not use Active Directory, you must create Linux accounts on the cluster primary node and add principals for these Linux users to the KDC. This includes a principal in the KDC for the primary node. In addition to the user principals, the KDC running on the primary node needs a principal for the local host.

When your architecture includes Active Directory integration, Linux users and principals on the local KDC, if applicable, are created automatically. You can skip this step. For more information, see Cross-realm trust and External KDC—cluster KDC on a different cluster with Active Directory cross-realm trust.

Submitting a step to the cluster after you create it or when you create the cluster is the easiest way to add users and KDC principals. Alternatively, you can connect to the primary node using an EC2 key pair as the default hadoop user to run the commands. For more information, see Connect to the Amazon EMR cluster primary node using SSH.

The following example submits a bash script configureCluster.sh to a cluster that already exists, referencing its cluster ID. The script is saved to Amazon S3.

aws emr add-steps --cluster-id <j-2AL4XXXXXX5T9> \
--steps Type=CUSTOM_JAR,Name=CustomJAR,ActionOnFailure=CONTINUE,\
Jar=s3://region.elasticmapreduce/libs/script-runner/script-runner.jar,\
Args=["s3://amzn-s3-demo-bucket/configureCluster.sh"]

The following example demonstrates the contents of the configureCluster.sh script. The script also handles creating HDFS user directories and enabling GSSAPI for SSH, which are covered in the following sections.

#!/bin/bash
#Add a principal to the KDC for the primary node, using the primary node's returned host name
sudo kadmin.local -q "ktadd -k /etc/krb5.keytab host/`hostname -f`"
#Declare an associative array of user names and passwords to add
declare -A arr
arr=([lijuan]=pwd1 [marymajor]=pwd2 [richardroe]=pwd3)
for i in ${!arr[@]}; do
    #Assign plain language variables for clarity
     name=${i} 
     password=${arr[${i}]}

     # Create a principal for each user in the primary node and require a new password on first logon
     sudo kadmin.local -q "addprinc -pw $password +needchange $name"

     #Add hdfs directory for each user
     hdfs dfs -mkdir /user/$name

     #Change owner of each user's hdfs directory to that user
     hdfs dfs -chown $name:$name /user/$name
done

# Enable GSSAPI authentication for SSH and restart SSH service
sudo sed -i 's/^.*GSSAPIAuthentication.*$/GSSAPIAuthentication yes/' /etc/ssh/sshd_config
sudo sed -i 's/^.*GSSAPICleanupCredentials.*$/GSSAPICleanupCredentials yes/' /etc/ssh/sshd_config
sudo systemctl restart sshd

Adding user HDFS directories

To allow your users to log in to the cluster to run Hadoop jobs, you must add HDFS user directories for their Linux accounts, and grant each user ownership of their directory.

Submitting a step to the cluster after you create it or when you create the cluster is the easiest way to create HDFS directories. Alternatively, you could connect to the primary node using an EC2 key pair as the default hadoop user to run the commands. For more information, see Connect to the Amazon EMR cluster primary node using SSH.

The following example submits a bash script AddHDFSUsers.sh to a cluster that already exists, referencing its cluster ID. The script is saved to Amazon S3.

aws emr add-steps --cluster-id <j-2AL4XXXXXX5T9> \
--steps Type=CUSTOM_JAR,Name=CustomJAR,ActionOnFailure=CONTINUE,\
Jar=s3://region.elasticmapreduce/libs/script-runner/script-runner.jar,Args=["s3://amzn-s3-demo-bucket/AddHDFSUsers.sh"]

The following example demonstrates the contents of the AddHDFSUsers.sh script.

#!/bin/bash
# AddHDFSUsers.sh script

# Initialize an array of user names from AD, or Linux users created manually on the cluster
ADUSERS=("lijuan" "marymajor" "richardroe" "myusername")

# For each user listed, create an HDFS user directory
# and change ownership to the user

for username in ${ADUSERS[@]}; do
     hdfs dfs -mkdir /user/$username
     hdfs dfs -chown $username:$username /user/$username
done
     

Enabling GSSAPI for SSH

For Kerberos-authenticated users to connect to the primary node using SSH, the SSH service must have GSSAPI authentication enabled. To enable GSSAPI, run the following commands from the primary node command line or use a step to run it as a script. After reconfiguring SSH, you must restart the service.

sudo sed -i 's/^.*GSSAPIAuthentication.*$/GSSAPIAuthentication yes/' /etc/ssh/sshd_config
sudo sed -i 's/^.*GSSAPICleanupCredentials.*$/GSSAPICleanupCredentials yes/' /etc/ssh/sshd_config
sudo systemctl restart sshd