Analytics
Amazon EMR
Amazon EMR provides managed Spark, Hadoop, and related big-data frameworks on EC2, EKS, or serverless capacity. It suits large preprocessing and embedding jobs that need distributed compute or custom open-source libraries.
Key points
- EMR Serverless removes cluster management for supported frameworks
- EMR on EKS shares Kubernetes capacity while isolating jobs
- Step and cluster logs should be stored outside ephemeral compute
When to use it
- Normalize a multi-terabyte document corpus
- Generate embeddings or features with distributed Spark
Exam tips
- Prefer Glue for simpler managed ETL and EMR when framework or cluster control matters
- Size shuffle, storage, and executor memory rather than treating an LLM data job as ordinary inference