A real estate company has a mission-critical application using Apache HBase in Amazon EMR. Amazon EMR is configured with a single master node. The company has over 5 TB of data stored on an Hadoop…

Question

Options

• AUse Spot Instances for core and task nodes and a Reserved Instance for the EMR master node.
• BStore the data on an EMR File System (EMRFS) instead of HDFS.
• CStore the data on an EMR File System (EMRFS) instead of HDFS and enable EMRFS consistent
• DStore the data on an EMR File System (EMRFS) instead of HDFS and enable EMRFS consistent

• A
  12% (7)
• B
  82% (49)
• C
  3% (2)
• D
  3% (2)

Explanation

The core problem is that HDFS stores data on the local disks of EC2 instances. If the single EMR master node fails, the cluster (and access to HDFS data) is lost. Migrating to EMRFS (EMR File System) moves the data to Amazon S3, which provides 11-nines durability and 99.99% availability by design - making the data highly available independently of the EMR cluster's health. If the cluster fails, the data persists in S3 and a new cluster can be launched against it. This is also cost-effective since S3 storage is cheaper than EBS volumes used by HDFS. Options C and D mention enabling EMRFS consistent view (which added DynamoDB cost and was used for eventual-consistency workarounds), but S3 is now strongly consistent by default, making that option unnecessary and less cost-effective. Option A uses Spot Instances, which reduces cost but does not inherently make the data highly available.

No community discussion yet for this question.