Changelog

This mode was never fully supported in. Either it failed or was silently equivalent to "Append", depending on the situation. Update your "ErrorIfExists" fields to "Append".

You can no longer pass write-queries via the script option when Spark is in read mode. This creates consistency across the board. For write operations, you can still use script as before.

The parser for property names is updated to distinguish between previous ambiguity. Property names containing dots are no longer ambiguous with alias separators. A backtick-quoted property path such as `my.property` is no longer split on its dots. This makes it possible to consistently read/write maps between the connector and Neo4j.

When the connection drops while a commit is in flight, the server may or may not have applied the transaction, and earlier versions always replayed the batch, silently duplicating it for non-idempotent writes. The new transaction.commit.unknown.outcome option makes that a choice between RETRY (the previous behavior, now also logging a warning) and FAIL. See Transactions, retries, and partial failures.

A generic OIDC token provider prefixed oidc added among available authentication options. You can freely pass OIDC configuration parameters with dot notation. Example: .option("authentication.oidc.clientSecret", "secret")

Replaces the connector’s own quoting function, fixing labels and property names that need escaping.

Previous version accidentally broke Custom authentication supplier. This is now fixed and functionality restored.

Schema optimizations meant for writes are no longer applied by the streaming reader.

There was a bug with labels containing spaces, causing them to degenerate into full graph scans when used in partitions. Such labels are now quoted correctly, and they no longer degenerate into full scans.

A backtick-quoted property path such as my.property is no longer split on its dots.

Fixed an identified NullPointerException when computing the latest offset for streaming reads.

These are now pushed down correctly instead of raising an error.

Improvements made to query options made the validation redundant. Remove the validation since it could reject valid partitioned reads.

Such as option schema.optimization.type, usage of $stream.offset in partitioned reads etc. Despite deprecation, setting data.conversion to legacy is still supported for now.

With Scala version 2.13, minimum Java 17 and targets Neo4j 5.0, 2025 and/or 2026. Dropped support for Scala 2.12, older versions of Java or Neo4j.

Set Cypher version to 25 with .option("cypher.version", "25")

Using the new options-prefix cypher.tuning you can apply CYPHER preambles for any generated query from the connector. See cypher tuning and cypher runtime.

Custom queries passed with the query option is re-written. Custom queries are now supported by Data Source V2’s predicate push-down optimizations.

It is no longer possible to run multiple queries by separating them with a semicolon inside a single script option.

Use the new script.1, script.2, …​, script.N options to run multiple scripts. They are executed in numbered order.

Changed relationship.save.strategy default value from NATIVE to KEYS.

It is now possible for spark clients to set custom Neo4j Driver transaction metadata using option db.transaction.metadata.