databricks/databricks-jdbc v3.4.4

v3.4.4

Repository: databricks/databricks-jdbc

Tag: v3.4.4

Published: 2026-10-22T18:05:36Z

Prerelease: no

Release notes:

Added

• Added typed getters for all six SAFE feature flag types using the existing cache.
• Added inline Arrow result support when UseBoundedSeaApi=1 and EnableQueryResultDownload=0, including multi-chunk streaming without relying on the deprecated total_row_count manifest field.

Updated

• Bumped Apache HttpClient 5 (httpclient5) from 5.6.3 to 5.6.4.
• Updated bundled Jackson from 2.18.9 to 2.18.11, lz4-java from 1.11.1 to 1.11.4, and Bouncy Castle from 1.84 to 1.85 to address security findings.
• SQL Execution API requests now skip HTTP response compression by default to avoid gzipping inline LZ4-compressed Arrow results. Set EnableSeaResponseCompression=1 to restore compression on bandwidth-constrained connections.
• Allow the server-side SQL Execution API feature flag to select SEA when CloudFetch is disabled.

Fixed

• Decode Reyden's native Arrow struct results for GEOMETRY and GEOGRAPHY, including

nested values in arrays, map values, and structs in both native-object and EWKT string modes.

• Fixed UserAgentEntry attribution in Query History for SQL Execution API connections.
• Honor Auth_Scope for OAuth client-secret M2M; unify scope parsing with JWT-assertion M2M and U2M. (#1706)
• Fixed request latency and lock contention that grew with every connection opened in a long-running JVM. The driver now registers each User-Agent entry once instead of on every connection. At most 64 distinct UserAgentEntry values are added to the shared User-Agent per JVM; SQL Execution API requests still send their own entry.
• Fixed queries failing with Multiple entries with same key when result column names differ only in case (e.g. policyid and policyId). Lookups by column label now return the first matching column, as the JDBC spec requires. (#1727)