Securing your Hadoop Cluster: Step 1: User Authentication
About Shamsher Ansari
Associate Consultant at Ellicium Solutions Inc.
I will be publishing a series of articles narrating our experience and how to implement security on the Hadoop cluster. In this first article of the series, let’s see how to resolve the challenge of user authentication in Hadoop.
Importance of Big Data Security
The amount of data being generated in enterprises is growing at a fast pace. Big Data technologies help organizations to capture, analyze, and manage this ever-increasing data. With this great power comes the great responsibility of keeping the business-critical data safe and secure.
Above all, Our team of Big Data experts at Ellicium has done extensive work on securing complex and large Hadoop infrastructure. In one such case, we are helping our client in the finance domain with Big Data Managed services where Hadoop security is one of the major areas we focus on.
Here, we have secured a Hadoop cluster with more than 60+ servers/nodes which host sensitive financial data. This cluster was built using Hortonworks Distribution Hadoop (HDP version 18.104.22.168). The cluster runs a diverse range of services like HDFS, YARN, MR2, Tez, Zookeeper, Hive, Spark2, HBase, Phoenix, Kafka, Ambari Metrics, and Ranger.
Moreover, It was quite a challenging task to secure this Hadoop cluster, as it was important to secure it without altering much of the existing workflow. However, we have come up with a series of steps that will make the task of securing your Hadoop cluster based on the collective experience of Ellicians,
Steps to secure Hadoop Cluster:
a) User authentication
b) User authorization
As stated at the beginning, in this article let’s see how to resolve the challenge of user authentication in Hadoop.
What is Authentication?
Authentication is a process in which a user is verified as an authentic user using the provided credentials. Authentication is important because it enables organizations to keep their networks secure by permitting only authenticated users to access its protected resources/services.
Lack of user authenticity in Hadoop
By default, Hadoop does not have a strong authentication mechanism, due to which a user can pretend to be an authentic user and can gain access into cluster exploiting security and accessing unauthorized data. This is a major risk and prevents organizations from adopting Hadoop.
In our case, we were using services like Phoenix and Hive. These are inherently insecure services, as by default they do not have any mechanism to verify the authenticity of the users. We used Kerberos to secure these services in Hadoop. Though there are multiple ways it can be done, other ways are unsafe thus, it can not be recommended.
How to implement strong authentication in Hadoop?
Kerberos is an authentication protocol that allows the system to verify user authenticity using tokens over secured and unsecured networks. However, by implementing Kerberos, you can make it mandatory for users to prove their identity. Thus preventing authorization access against impersonating users.
Steps in Kerberos exchanges diagram are in detail below:
- Client requests to KDC for TGT (ticket-granting ticket) using its principal name.
- In response, AS (authentication server) issues a TGT (ticket-granting ticket) and a session key, the client can use to encrypt and authenticate communication between the KDC for TGT requests.
- Once TGT has obtained further clients can request a ticket from TGS (Ticket Granting Server).
- TGS provides a ticket to the client and it gets stored on the client’s workstation.
- Whenever a client requests an application program, the cached ticket is in use to verify client authenticity.
- After the authenticity of the client has been verified then the client can continue to use the service.
In a Hadoop cluster, Kerberos can be implemented using MIT Kerberos, Active Directory, or FreeIPA server. We recommend that the cluster should get secure using the Enterprise Active Directory server, considering the brief comparison between Active Directory, MIT Kerberos, and FreeIPA.
Why do we recommend the Enterprise Active Directory server? Check this brief comparison with other services.
once the cluster is Kerberized and integrated with Active Directory or FreeIPA using LDAP (Lightweight Directory Access Protocol), users can log in to Gateway nodes/servers using their domain user Id and cache the tickets using kinit command. However, the job will fail with GSSException if users try to use any Hadoop services without caching the tickets, as below:
The token is a must for the execution of jobs in a Kerberized cluster environment.
Brief description of Kinit command
Moreover, Kinit is to obtain and cache Kerberos ticket-granting tickets.
Syntax : kinit <username>@<domain realm>
Once the user executes this command they will get a prompt for a password and after entering the valid password ticket will be cached.
below is an example, in addition, that shows you that Kinit command that can get in use to cache ticket on a Linux server through CLI:
Similarly, we have klist and kdestroy commands.
Klist command: To view the Kerberos cached ticket.
Kdestroy command: Destroys a Kerberos credentials cache.
For automated applications/jobs, users can generate and use keytab. A keytab is a file containing pairs of Kerberos principals and encrypted keys. This can be generated and stored on the gateway nodes and can be used to cache the token without any manual inputs.
Below is an example of how to generate keytab file on a Linux machine using ktutil command when Kerberos integrate with Active Directory server:
Below is an example on how to pass principal(username with realm) and keytab file to an automated spark application:
In short implementation of Kerberos provides strong security features and protects the system from attacks. In the next article moreover, I will share about User authorization. Also how to implement it to strengthen Hadoop security further.