Changelog

Names are now escaped correctly when the connector generates constraints. See Schema optimization.

Generated type constraints now use the target property name in preference to the source field name. See Schema optimization.

The driver upgrade unblocks the use of a custom URL resolver. See Options and configuration.

Timestamps, intervals, and byte arrays are now converted differently from previous releases. Set the type.conversion reader option to legacy to keep the pre-5.4.0 behavior. See Data type mapping and Migration 5.3 → 5.4.

When relationship.properties is set, only the listed properties are written; previously, unmapped fields were written too. See Writer options and Migration 5.3 → 5.4.

Spark intervals are now mapped to and from the Neo4j Duration type. See Data type mapping.

Opt back into the legacy conversion logic. See Reader options.

Schema optimization can now await index creation globally. See Schema optimization.

See Spark Structured Streaming.

Adds the skip.null.keys option for write operations.

The connector’s user agent can now be overridden.

Transaction configuration is now also applied to auto-commit transactions.

Adds support for a custom authentication token provider. See Options and configuration.

Restores Java 8 compatibility in the commons library.

Reader queries are now retried automatically to improve reliability. See Reading from Neo4j.

Fixes an overflow in the partition-size computation.

The default build target changed to Scala 2.12. See Installation.

Adds support for configuring a transaction timeout on the Neo4j driver. See Options and configuration.

The Cypher version is prepended to internal queries for explicit version selection, guaranteeing compatibility with Neo4j’s calendar-based versions.

Adds support for the Neo4j 2025.x calendar versions.

Connectivity is now validated through verifyConnectivity.

Partition counting is now correct for custom queries that use a limit.

Restores checkpointing support on structured streams. See Streaming checkpoint.

The connector’s retry logic now aligns with the driver retry mechanisms.

Node properties can now be excluded from relationships.

Schema option validation has been improved.

Fixes resolver handling during driver construction.

Schema type and existence constraints are inferred from the DataFrame definition, with improved management of existing constraints. See Schema optimization.

Adds metrics support on the sink path.

User-agent and query-error management have been improved.

The Spark version can now be configured.

This release requires Apache Spark 3.4 or later in order to support the new push-down Top N optimization. See the compatibility table and Migration 5.1 → 5.2.

Controlled by the pushdown.topN.enabled reader option. See Reader options.

Example notebooks are available from the Quickstart (the legacy /spark/current/playground/ URL redirects there).

The documentation was extracted to a separate repository at neo4j/docs-spark.

This release requires Apache Spark 3.3 or later in order to support push-down v2 filters and limit optimizations. 5.1.0 is the last release compatible with Spark 3.3. See the compatibility table and Migration 5.0 → 5.1.

Controlled by the pushdown.filters.enabled reader option. See Reader options.

Controlled by the pushdown.limit.enabled reader option. See Reader options.

Adds first-class support for Graph Data Science (GDS).

Adds support for bearer authentication. See Options and configuration.

Fixes handling of datetime properties that have null values.

Fixes Spark version validation.

Fixes insertion of the Neo4j Map data type.

Fixes broken documentation links.

Fixes Spark version validation failures on AWS EMR and EMR Serverless.

LocalTime and Time fields are now written as single struct fields instead of individual fields. See Data type mapping.

Fully supports Neo4j 5.0 and Neo4j 4.4, including the Neo4j 5.0 indexes and constraints syntax changes.

Based on the Spark 3.2 DataSource V2 API and controlled by the pushdown.aggregate.enabled reader option (requires Spark > 3.2.2 due to an upstream Spark bug). See Reader options.