Databricks Runtime Versioning and Deprecation Policy

Azure Databricks offers two types of runtime for the clusters that you create:

  • Databricks Runtime is the set of core components that run on the clusters managed by Azure Databricks. It includes Apache Spark but also adds a number of components and updates that substantially improve the usability, performance, and security of big data analytics.
  • Databricks Light (also known as Data Engineering Light) is the Azure Databricks packaging of the Apache Spark runtime and excludes many of the components that Databricks Runtime adds to open source Spark, in order to provide a runtime option for jobs that don’t need the advanced benefits provided by Databricks Runtime.

You can choose from among many supported runtime versions when you create a cluster.

../../_images/runtime-version.png

Runtime components

Databricks Runtime consists of the following components:

  • Apache Spark: each runtime version contains a specific Apache Spark version
  • Delta Lake: a next-generation engine built on top of Apache Spark that provides ACID transactions, optimized layouts and indexes, and execution engine improvements for building data pipelines.
  • Databricks Serverless: a layer on top of Apache Spark that provides fine-grained resource sharing to optimize cloud costs
  • Ubuntu and its accompanying system libraries
  • Pre-installed Java, Scala, Python, and R languages
  • Pre-installed Java, Scala, Python, and R libraries
  • GPU libraries for GPU-enabled clusters
  • Databricks services that integrate with other components of the platform, such as notebooks, jobs, and cluster manager

Databricks Light is the Azure Databricks packaging of the open source Apache Spark runtime. It excludes many of the libraries and services listed above. For details, see Overview of Databricks Light.

The Databricks Runtime Release Notes list the library versions included in each runtime version.

Versioning

New versions of Databricks Runtime are released on a regular basis.

  • Major Releases are represented by an increment to the version number that precedes the decimal point (the jump from 3.5 to 4.0, for example). They are released when there are major changes, some of which may not be backwards-compatible.
  • Feature Releases are represented by an increment to the version number that follows the decimal point (the jump from 3.4 to 3.5, for example). Each major release includes multiple feature releases. Feature releases are always backwards compatible with previous releases within their major release.
  • Long Term Support releases are represented by an “-LTS” suffix (for example, 3.5-LTS) and are the “canonical” feature version of the major release, for which we provide two full years of support. We recommend these releases for your production jobs. See the deprecation policy described below for more information.

Databricks Light releases closely follow those of open-source Apache Spark and are named after the Apache Spark release.

Phases of a version

Phase Guarantees
Beta Support SLAs are not applicable. For more information, see runtime-releases.
Full Support Major stability and security fixes are backported. New features will be available in the next version.
Marked for deprecation Version will be deprecated (unsupported) soon. This phase will last no less than 3 months, although the actual duration will vary depending on version type. See Deprecation policy for more information.
Deprecated

Version is unsupported:

  • Not available for selection in the UI
  • Workloads running on these versions receive no Databricks support
  • Databricks will not backport fixes
  • Users can continue to use the API to create clusters
Sunset Typically, Databricks removes a version completely from the API only when its usage drops to 0. Your scheduled workloads are therefore guaranteed to run properly regardless of the deprecation schedule. If we make exceptions to this rule, we will give ample notice.

Important

The policies enumerated above to do not apply to Databricks Light. Policies for Databricks Light will be published here soon.

Deprecation policy

The Databricks deprecation policy depends on version type.

Feature version

We give 4 months notice before deprecating a feature version. After 4 months from the deprecation announcement, we remove the version from the Create Cluster and Edit Cluster pages. We do not backport fixes and we provide no support for those versions. You can continue to create clusters from the API if you have automated jobs running.

Long Term Support (LTS) version

For some major versions, we identify a pinned version for which we offer two years of support from the date of release. After two years, we mark the LTS version for deprecation, deprecating it one year later.

Important

The policies enumerated above to do not apply to Databricks Light. Policies for Databricks Light will be published here soon.

List of releases

Current Databricks Runtime releases

Version Spark Version Release Date Deprecation Announcement Deprecation Date
5.3 Spark 2.4 Apr 03, 2019 Aug 03, 2019 Dec 03, 2019
5.2 Spark 2.4 Jan 24, 2019 May 27, 2019 Sep 30, 2019
3.5-LTS Spark 2.2 Dec 21, 2017 Jan 02, 2019 Jan 02, 2020

Current Databricks Light releases

Version Spark Version Release Date Deprecation Announcement Deprecation Date
2.4 Spark 2.4 Feb 27, 2019

Databricks Runtime versions marked for Deprecation

Version Spark Version Release Date Deprecation Announcement Deprecation Date
5.0 Spark 2.4 Nov 08, 2018 Mar 08, 2019 Jul 08, 2019
3.5-LTS Spark 2.2 Dec 21, 2017 Jan 02, 2019 Jan 02, 2020

Deprecated Databricks Runtime releases

Version Spark Version Release Date Deprecation Announcement Deprecation Date
4.3 Spark 2.3 Aug 09, 2018 Dec 09, 2018 Apr 09, 2019
4.2 Spark 2.3 Jul 05, 2018 Nov 05, 2018 Mar 05, 2019
4.1 Spark 2.3 May 17, 2018 Sep 17, 2018 Jan 17, 2019
4.0 Spark 2.3 Mar 01, 2018 Jul 01, 2018 Nov 01, 2018
3.4 Spark 2.2 Nov 20, 2017 Mar 31, 2018 Jul 31, 2018

REST API version string

The structure of a Databricks runtime version string in the REST API is:

Databricks Runtime:

<M>.<F>.x[-cpu][-gpu][-ml][-hls]-scala<scala-version>

where

  • M - Databricks Runtime major release
  • F - Databricks Runtime feature release
  • cpu - CPU version (with -ml only)
  • gpu - GPU-enabled
  • ml - Machine learning
  • hls - Health and life sciences
  • scala-version - version of Scala used to compile Spark: 2.10 or 2.11

For example, 3.5.x-scala2.10 and 4.1.x-gpu-scala2.11. The List of releases tables map Databricks Runtime versions to the Spark version contained in the Runtime.

Databricks Light:

apache-spark.<M>.<F>.x-scala<scala-version>

where

  • M - Apache Spark major release
  • F - Apache Spark feature release
  • scala-version - version of Scala used to compile Spark: 2.10 or 2.11

For example, apache-spark-2.4.x-scala2.11.