AWS Batch
- What it is
- Fully managed batch computing: you submit jobs, Batch provisions the compute to run them.
- Dynamically launches EC2 or AWS Fargate capacity sized to the job queue, then shuts it down.
- Key traits
- Jobs run to completion, then exit. No long-running server.
- Handles queueing, dependencies between jobs, retries, and priority.
- Pairs well with Spot Instances — batch work is usually interruption-tolerant, so cost drops sharply.
- Not to be confused with
- AWS Lambda — event-driven, short-lived (15-minute ceiling). Batch is for long-running compute jobs (hours).
- AWS Step Functions — orchestrates a workflow of steps; Batch executes the heavy compute inside a step.
- Amazon EMR — managed Hadoop/Spark specifically; Batch runs any containerized job.