DataAI ETL AWS (Glue / EMR) — Production Installation Guide

Installation and Usage for AWS Marketplace

Delivery model

Build from an AWS-provided EMR Serverless Spark image, preserving its entrypoint and hadoop user. The Dockerfile copies DataAI JARs to /opt/dataai/lib and /usr/lib/spark/jars. Customers select the resulting Marketplace image when creating EMR Serverless Spark compute. No DataAI service runs.

Production-candidate artifacts

Install these matching-version modules together:

• dataai-spark-api-1.0.0.jar
• dataai-spark-quality-1.0.0.jar
• dataai-spark-core-1.0.0.jar
• dataai-spark-functions-1.0.0.jar

The CLI JAR is optional. Spark and Hadoop are provided by the customer runtime and are not bundled by DataAI.

Generic Spark usage

Add all required JARs to the Spark driver and executor classpaths using the target platform's supported library or --jars mechanism. Then call DataAiPipeline and the functions in com.dataai.etl.spark.functions. The complete inventory is in FUNCTION_CATALOG.md.

Customer code supplies Dataset<Row> inputs and decides whether returned DataFrames are written. DataAI performs no automatic persistence.

Production

Deploy immutable version 1.0.0 only after Yanbor release authorization, commercial entitlement, and target-platform validation. Obtain it from the approved marketplace artifact, authenticated Maven repository, or customer artifact repository. Verify checksums and pin the exact version.

Troubleshooting

• UnsupportedClassVersionError: select a Java 17-compatible runtime.
• Class not found: install API, quality, core, and functions at the same version.
• Scala linkage error: use a runtime compatible with Scala 2.12.
• Spark linkage error: validate against the documented Spark 3.5.x baseline.
• Permission failure: use customer-approved identity and storage permissions; DataAI does not bypass platform controls.