E-MapReduce:April 15, 2026 release

Category

Update description

References

Data catalog

StarRocks is now supported as a data catalog source. Add StarRocks to your data catalog to access and query its data.

Manage data catalogs

Livy-submitted jobs can now access all DLF Catalogs added to your data catalog, simplifying multi-source configuration.

Data development

Task priority is now configurable. Assign higher priority to critical tasks so they are scheduled first when resources are limited.

Task priority

AI Center

AI Function now supports qwen3.5-plus, qwen3.6-plus, and tongyi-embedding-vision-plus models for inference, vectorization, and analysis of multimodal data including images, videos, and text.

AI Center (Beta)

Session management

DuckDB sessions are now available for lightweight, interactive querying and analysis of small to medium-sized datasets directly on the platform.

Manage DuckDB sessions

Cluster management

Ray clusters are now available as a resource type for multimodal data processing and distributed Python parallel computing.

Use Ray clusters

Skills

The EMR Serverless Spark management Skill lets you manage workspaces and Spark jobs using natural language, lowering the technical barrier for data development.

alibabacloud-emr-spark-manage

Other optimizations

You can now double-click or right-click to copy SparkSQL query results.

Version

Description

esr-5.2.0 (Spark 4.0.1, Scala 2.13)

esr-4.8.0 (Spark 3.5.2, Scala 2.12)

esr-3.7.0 (Spark 3.4.4, Scala 2.12)

• Fusion

  • Improves Hybrid Shuffle stability and performance.

  • Supports Scan Sharing.

  • Optimizes the Window operator.

  • Supports DS V2 Update and Delete.

  • Enables splitting tables by file size during writes.

  • Supports DuckDB sessions.

• Data lake

  • [DLF] Supports row-level permission.

  • [Paimon] Supports Variant.

  • [Delta] Upgrades to 3.3.2 with Delta Sharing support.

  • [Fluss] Supports reads and writes by Spark Batch and Streaming.

• Spark framework & ecosystem

  • Supports the StarRocks catalog.

  • Spark 4 now supports the StarRocks connector.

  • Improves the stability of the Holo Connector.

  • DML statements now return the number of affected rows.

  • distcp can now delete old files at the destination when a job is committed.

  • Spark Streaming can now read from and write to Fluss.

Ray engine err-1.0.0 (Ray 2.47.1, Python 3.12)

• Ray

  • Supports DLF.

  • Supports auto scaling.

  • Supports Kerberos and Ranger authentication and authorization.

  • Adds GPU support.

• Daft

  • Supports Daft on Ray.

Celeborn

• Stability improvements

  • Optimizes the scaling policy.

  • Fixes an issue where threads could exit unexpectedly in extreme cases.