ETL CompareETL Compare

Review · Figures checked September 2026

Apache Auron (formerly Blaze) review: a native engine for Spark

ETL Compare staff · Figures checked September 2026 · Sourced from vendor docs, project pages and public benchmarks

In brief

Apache Auron, the project formerly known as Blaze, maps Spark physical plans onto the DataFusion native engine and adds its own compacted shuffle format and multi-level memory management. It scores 3.42 out of 5 on our lead ranking: free and well aimed at ETL bottlenecks, but self-managed and light on published platform and version detail.

3.42 / 5

ETL fit score (editorial assessment, 0-5)

ETL fit score (editorial assessment, 0-5)

Rank 6 of 9

Open-source option with its own shuffle and memory layer

Auron at a glance
ApproachOpen-source top-level Apache Software Foundation project that maps Spark physical plans onto DataFusion native execution, with its own shuffle format
Runs onSelf-managed Spark on JDK 8, 11, 17 or 21
InstancesStandard CPU instances
LicenseApache License 2.0
PriceNo license fee
Code changesNone; Spark settings and custom shuffle manager
Figures checkedSeptember 2026

Sources: auron.apache.org, apache/auron on GitHub, kwai/blaze on GitHub · Fetched 27 Sep 2026

How does it score on each criterion?

Score breakdown by criterion, with reasons
CriterionWeightScoreWhy
Adoption effort20%3.4No code changes, but you enable it through Spark settings and a custom shuffle manager and install it yourself.
Stage coverage20%3.8Native vectorized execution on DataFusion, compacted shuffle formats and multi-level memory management are all documented.
Tuning and operating burden20%2.6Tied lowestSelf-managed with community support through the Apache mailing list.
Platform and instance portability10%3.8Runs on standard CPU instances and supports JDK 8, 11, 17 and 21; the project says it is adapted to Spark mainline versions without listing them on the page we reviewed.
Cost model transparency5%5.0Tied highestApache License 2.0, no license fee.
Published evidence15%3.4The project states about 2x faster than Spark 3.5 on TPC-DS with about 50% cluster resources saved; project-published.
Maturity and community10%3.2A top-level project of the Apache Software Foundation (auron.apache.org, read 26 October 2026), with a history under the Blaze name.

What is Apache Auron?

Auron describes itself as an accelerator for big data engines that uses native vectorized execution. It maps optimized Spark physical plans into Apache DataFusion's execution engine, written in Rust on Apache Arrow's columnar format. Auron states it is a top-level project of the Apache Software Foundation (read 26 October 2026), and it is Apache 2.0 licensed. The GitHub repository at kwai/blaze now serves the Apache Auron repository, which is how the Blaze name carried over.

Sources: auron.apache.org, github.com/apache/auron, github.com/kwai/blaze · Fetched 27 Sep 2026

What does setup involve?

You enable Auron through Spark settings, including spark.auron.enable set to true, and a custom shuffle manager. It supports JDK 8, 11, 17 and 21. The project says it is adapted to Apache Spark mainline versions but does not list them on its homepage, so confirm your version in the repository before planning a trial.

What does the project claim?

Project states

“~2x faster” than Apache Spark 3.5 on TPC-DS while saving “~50% cluster resources”.

Source: auron.apache.org · Fetched 27 Sep 2026

Project figure. Not measured by ETL Compare.

Where is it strong?

  • Designed around ETL bottlenecks: compacted shuffle formats and multi-level memory management are documented features, not add-ons.
  • No license fee (cost model 5.0, tied highest).
  • Broad JDK support, including JDK 8 for older clusters.

What are the watch-outs?

  • Young top-level project (maturity 3.2): Auron states it is a top-level Apache Software Foundation project, but its track record under the Auron name is short.
  • Self-managed with mailing-list support; tuning and operating burden is tied lowest in the set (2.6).
  • Supported Spark versions and managed-platform guides are not listed on the homepage.

Who should shortlist it?

Engineering-led teams that self-manage Spark, want an open-source engine that treats shuffle and memory as first-class problems, and can accept a young project with community support.

Which job stages does it address?

Spark ETL job stages and which stages each accelerator documents addressingA waterfall of five Spark ETL job stages (read, transform, shuffle, spill, write) with illustrative proportions, and below it a grid showing, for each accelerator, whether its own documentation says it addresses that stage.Anatomy of a Spark ETL jobRead: scan and decode filesTransform: filter, join, aggregateShuffle: write and fetch between stagesSpill: memory pressure pushes data to diskWrite: encode and commit outputIllustrative proportions, not measured data. Your own split comes from the Spark UI: see Profile a slow Spark job.
Stage coverage by accelerator, from vendor documentation
AcceleratorReadTransformShuffleSpillWrite
AuronPartialPartialDocumentedDocumentedDocumentedDocumentedPartialPartialNot statedNot stated
  • DocumentedDocumented: the vendor's own documentation says it addresses this stage
  • PartialPartial: indirect or experimental, or covered only by an end-to-end claim
  • Not statedNot stated: not found on the pages we reviewed

This shows what each vendor says, not what we measured. Sources are listed on each review.

Native vectorized execution, compacted shuffle formats, multi-level memory management, extensible object store and file format support.

Source: apache/auron on GitHub · Fetched 27 Sep 2026

Frequently asked questions

Is Auron the same project as Blaze?

The kwai/blaze GitHub repository now resolves to the Apache Auron repository, so Auron is the continuation of Blaze, which Auron states is now a top-level project of the Apache Software Foundation.

Does Auron need code changes?

No. It is enabled with Spark configuration and a shuffle manager setting.

Compare with

Before you buy

Find your slowest stage with Profile a slow Spark job, then work out the money with Spark cost per job, explained.

Sources