July 29, 2026 - v3.4.0
Download the Spark Connector JAR.
[v3.4.0-v6.2.4]⚓︎
Added⚓︎
- Introduced a new optional configuration setting:
service_account_token, a JWT used to authenticate to the Hydrolix cluster, as an alternative tousername/password.
- Added additional Spark UI metrics:
PartitionReader fetch time(min, max, median, p90),SparkScanBuilder.build time, partition reader start/last timestamps, andTurbine query timestatistics - Added optional configuration setting
hdx_partitions_per_task, which specifies the number of Hydrolix partitions read by each Spark partition - Added pushdown support (performance optimization) for ORDER BY + LIMIT (top-N) queries
- Added pushdown support (performance optimization) for the OFFSET clause
- Added shading rules for Apache httpcore and httpclient libraries to avoid classpath conflicts on Microsoft Fabric
- Added logging of the embedded Turbine version at startup
- Added an experimental columnar (transposition-based) partition reader for use with the
force_columnarquery mode; the transposition batch size may be set with thecolumnar_batch_sizeconfig - Added the optional config keys
cluster_url,jdbc_protocol, andjdbc_portas an alternative toapi_urlandjdbc_url - Added support for page-number-paginated Hydrolix API endpoints, recursively fetching all pages when more than one page of data is available
Changed⚓︎
- Bundled Turbine server updated to v6.2.4
- Changed the embedded Turbine's listen port from 8088 to 8123
- Reworked summary table interaction:
- All summary aliases are now accessed with the syntax
SELECT hdxAgg('my_summary_alias') FROM hydrolix.my_project.my_summary_table - No user interaction is required to pre-register summary aliases before querying them
- Using summary tables requires the SQL extension
spark.sql.extensions=io.hydrolix.connectors.spark.SummaryUdfExtension
- All summary aliases are now accessed with the syntax
Fixed⚓︎
- Added workaround for reads failing for tables whose partition catalog spans more than one page (>1000 partitions by default)
- Fixed presigned reads failing with a
config not loadederror when the bundled Turbine server is v6.2.0 or later - Fixed the API access-token cache not renewing, which could cause long-running sessions to fail with authentication errors until restarted
- Fixed
TURBINE_VERSIONnot accepting released Turbine versions when building from source, and the JAR's internal version metadata reporting a stalev1.0.0