Release Databricks OSS JDBC driver version 3.3.1

Added

• Added DatabaseMetaData.getProcedures() and DatabaseMetaData.getProcedureColumns() to discover stored procedures and their parameters. Queries information_schema.routines and information_schema.parameters using parameterized SQL for both SEA and Thrift transports.
• Added connection property OAuthWebServerTimeout to configure the OAuth browser authentication timeout for U2M (user-to-machine) flows, and also updated hardcoded 1-hour timeout to default 120 seconds timeout.
• Added connection property UseQueryForMetadata to use SQL SHOW commands instead of Thrift RPCs for metadata operations (getCatalogs, getSchemas, getTables, getColumns, getFunctions). This fixes incorrect wildcard matching where _ was treated as a single-character wildcard in Thrift metadata pattern filters.
• Added connection property TreatMetadataCatalogNameAsPattern to control whether catalog names are treated as patterns in Thrift metadata RPCs. When disabled (default), unescaped _ in catalog names is escaped to prevent single-character wildcard matching. This aligns with JDBC spec which treats catalogName as identifier and not pattern.

Updated

• Bumped com.fasterxml.jackson.core:jackson-core from 2.18.3 to 2.18.6.
• Fat jar now routes SDK and Apache HTTP client logs through Java Util Logging (JUL), removing the need for external logging libraries.
• Added Apache Arrow on-heap memory management for processing Arrow query results. Previously, Arrow result processing was unusable on JDK 16+ without passing the --add-opens=java.base/java.nio=ALL-UNNAMED JVM argument, due to stricter encapsulation of internal APIs. With this change, there is no JVM argument required - the driver automatically falls back to an on-heap memory path that uses standard JVM heap allocation instead of direct memory access.
• Log timestamps now explicitly display timezone.
• [Breaking Change] PreparedStatement.setTimestamp(int, Timestamp, Calendar) now properly applies Calendar timezone conversion using LocalDateTime pattern (inline with getTimestamp). Previously Calendar parameter was ineffective.
• DatabaseMetaData.getColumns() with null catalog parameter now retrieves columns from all available catalogs when using SQL Execution API.

Fixed

• Fixed statement timeout when the server returns TIMEDOUT_STATE directly in the ExecuteStatement response (e.g. query queued under load), the driver now throws SQLTimeoutException instead of DatabricksHttpException.
• Fixed Thrift polling infinite loop when server restarts invalidate operation handles, and added configurable timeout (MetadataOperationTimeout, default 300s) with sleep between polls for metadata operations.
• Fixed DatabricksParameterMetaData.countParameters and DatabricksStatement.trimCommentsAndWhitespaces with a SqlCommentParser utility class.
• Fixed rollback() to throw SQLException when called in auto-commit mode (no active transaction), aligning with JDBC spec. Previously it silently sent a ROLLBACK command to the server.
• Fixed fetchAutoCommitStateFromServer() to accept both "1"/"0" and "true"/"false" responses from SET AUTOCOMMIT query, since different server implementations return different formats.
• Fixed socket leak in SDK HTTP client that prevented CRaC checkpointing. The SDK's connection pool was not shut down on connection.close(), leaving TCP sockets open.
• Fixed IdleConnectionEvictor thread leak in long-running services. The feature-flags context shared per host was ref-counted incorrectly and held a stale connection UUID after the owning connection closed; on the next 15-minute refresh it silently recreated an HTTP client (and its evictor thread) that was never cleaned up. Connection UUIDs are now tracked idempotently and the stored connection context is updated when the owning connection closes.
• Fixed Date fields within complex types (ARRAY, STRUCT, MAP) being returned as epoch day integers instead of proper date values.
• Fixed DatabaseMetaData.getColumns() returning the column type name in COLUMN_DEF for columns with no default value. COLUMN_DEF now correctly returns null per the JDBC specification.
• Coalesce concurrent expired cloud fetch link refreshes into a single batch FetchResults RPC to prevent thread pool exhaustion under high concurrency.

v3.2.1

[v3.2.1] - 2026-02-16

Added

• Added streaming prefetch mode for Thrift inline results (columnar and Arrow) with background batch prefetching and configurable sliding window for improved throughput.
• Added EnableInlineStreaming connection parameter to enable/disable streaming mode (default: enabled).
• Added ThriftMaxBatchesInMemory connection parameter to control the sliding window size for streaming (default: 3).
• Added support for disabling CloudFetch via EnableQueryResultDownload=0 to use inline Arrow results instead.
• Added EnableMetricViewMetadata connection parameter to enable/disable Metric View table type (default: disabled).
• Added NonRowcountQueryPrefixes connection parameter to specify comma-separated query prefixes that should return result sets instead of row counts.

Updated

• Enhanced error logging for token exchange failures.
• Geospatial column type names now include SRID information (e.g., GEOMETRY(4326) instead of GEOMETRY).
• Implemented lazy loading for inline Arrow results, fetching arrow batches on demand instead of all at once. This improves memory usage and initial response time for large result sets when using the Thrift protocol with Arrow format.
• Enhanced enableMultipleCatalogSupport behavior: When this parameter is disabled (enableMultipleCatalogSupport=0), metadata operations (such as getSchemas(), getTables(), getColumns(), etc.) now return results only when the catalog parameter is either null or matches the current catalog. For any other catalog name, an empty result set is returned. This ensures metadata queries are restricted to the current catalog context. When enabled (enableMultipleCatalogSupport=1), metadata operations continue to work across all accessible catalogs.

Fixed

• Fixed getTypeInfo() and getClientInfoProperties() to return fresh ResultSet instances on each call instead of shared static instances. This resolves issues where calling these methods multiple times would fail due to exhausted cursor state (Issue #1178).
• Fixed complex data type metadata support when retrieving 0 rows in Arrow format
• Normalized TIMESTAMP_NTZ to TIMESTAMP in Thrift path for consistency with SEA behavior
• Fixed complex types not being returned as objects in SEA Inline mode when EnableComplexDatatypeSupport=true.
• Fixed StringIndexOutOfBoundsException when parsing complex data types in Thrift CloudFetch mode. The issue occurred when metadata contained incomplete type information (e.g., "ARRAY" instead of "ARRAY"). Now retrieves complete type information from Arrow metadata.
• Fixed timeout exception handling to throw SQLTimeoutException instead of DatabricksHttpException when queries timeout during result fetching phase. This completes the timeout exception fix to handle both query execution polling and result fetching phases.
• Fixed getResultSet() to return null in case of DML statements to honour JDBC spec.

v3.1.1

Added

• Added token caching for all authentication providers to reduce token endpoint calls.
• We will be rolling out the use of Databricks SQL Execution API by default for queries submitted on DBSQL. To continue using Databricks Thrift Server backend for execution, set UseThriftClient to 1.

Updated

• Changed default value of IgnoreTransactions from 0 to 1 to disable multi-statement transactions by default. Preview participants can opt-in by setting IgnoreTransactions=0. Also updated supportsTransactions() to respect this flag.

Fixed

• [PECOBLR-1131] Fix incorrect refetching of expired CloudFetch links when using Thrift protocol.
• Fixed logging to respect params when the driver is shaded.
• Fixed isWildcard to return true only when the value is *

Release 3.0.7

Updated

• Log timestamps now explicitly display timezone.
• [Breaking Change] PreparedStatement.setTimestamp(int, Timestamp, Calendar) now properly applies Calendar timezone conversion using LocalDateTime pattern (inline with getTimestamp). Previously Calendar parameter was ineffective.
• DatabaseMetaData.getColumns() with null catalog parameter now retrieves columns from all catalogs when using SQL Execution API, aligning the behaviour with thrift.
• DatabaseMetaData.getFunctions() with null catalog parameter now retrieves columns from the current catalog when using SQL Execution API, aligning the behaviour with thrift.

Fixed

• Fix timeout exception handling to throw SQLTimeoutException instead of DatabricksSQLException when queries timeout.
• Removes dangerous global timezone modification that caused race conditions.
• Fixed Statement.getLargeUpdateCount() to return -1 instead of throwing Exception when there were no more results or result is not an update count.
• CVE-2025-66566. Updated lz4-java dependency to 1.10.1.
• Fix INVALID_IDENTIFIER error when using catalog/schema/table names for SQL Exec API with hyphens or special characters in metadata operations (getSchemas(), getTables(), getColumns(), etc.) and connection methods (setCatalog(), setSchema()). Per Databricks identifier rules, special characters are now properly enclosed in backticks.
• Fix Auth_Scope handling inconsistency in Azure U2M OAuth.

Release Databricks OSS JDBC driver version 3.0.6

Added

• Added the EnableTokenFederation url param to enable or disable Token federation feature. By default it is set to 1
• Added the ApiRetriableHttpCodes, ApiRetryTimeout url params to enable retries for specific HTTP codes irrespective of Retry-After header. By default the HTTP codes list is empty.

Updated

• Added validation for positive integer configuration properties (RowsFetchedPerBlock, BatchInsertSize, etc.) to prevent hangs and errors when set to zero or negative values.
• Updated Circuit breaker to be triggered by 429 errors too.
• Refactored chunk download to keep a sliding window of chunk links. The window advances as the main thread consumes chunks. These changes can be enabled using the connection property EnableStreamingChunkProvider=1. The changes are expected to make chunk download faster and robust.
• Added separate circuit breaker to handle 429 from SQL Exec API connection creation calls, and fall back to Thrift.

Fixed

• Fix driver crash when using INTERVAL types.
• Fix connection failure in restricted environments when LogLevel.OFF is used.
• Fix U2M by including SDK OAuth HTML callback resources.
• Fix microsecond precision loss in PreparedStatement.setTimestamp(int,Timestamp, Calendar) and address thread-safety issues with global timezone modification.
• Fix metadata methods (getColumns, getFunctions, getPrimaryKeys, getImportedKeys) to return empty ResultSets instead of throwing exceptions when catalog parameter is NULL, for SQL Exec API.

Release 3.0.5

Added

• Added support for high-performance batched writes with parameter interpolation:
  • supportManyParameters=1: Enables parameter interpolation to bypass 256-parameter limit (default: 0)
  • EnableBatchedInserts=1: Enables multi-row INSERT batching (default: 0)
  • BatchInsertSize=<SIZE>: Maximum rows per batch (default: 1000)
  • Note: Large batches are chunked for execution. If a chunk fails, previous chunks remain committed (no transaction rollback). Consider using staging tables for critical workflows.
• Added Feature-flag integration for SQL Exec API rollout
• Call statements will return result sets in response
• Add a gating flag for enabling GeoSpatial support: EnableGeoSpatialSupport. By default, it will be disabled

Updated

• Minimized OAuth requests by reducing calls in feature flags and telemetry.
• Geospatial getWKB() now returns OGC-compliant WKB values.

Fixed

• Fix: SQLInterpolator failing to escape temporal fields and special characters.
• Fixed: Errors in table creation when using BIGINT, SMALLINT, TINYINT, or VOID types.
• Fixed: PreparedStatement.getMetaData() now correctly reports TINYINT columns as Types.TINYINT (java.lang.Byte) instead of Types.SMALLINT (java.lang.Integer).
• Fixed: TINYINT to String conversion to return numeric representation (e.g., "65") instead of character representation (e.g., "A").
• Fixed: Complex types (Structs, arrays, maps) now show detailed type information in metadata calls in Thrift mode
• Fixed: incorrect chunk download/processing status codes.
• Shade SLF4J to avoid conflicts with user applications.

v3.0.4

Added

• Added support for telemetry log levels, which can be controlled via the connection parameter TelemetryLogLevel. This allows users to configure the verbosity of telemetry logging from OFF to TRACE.
• Added full support for JDBC transaction control methods in Databricks. Transaction support in Databricks is currently available as a Private Preview. The IgnoreTransactions connection parameter can be set to 1 to disable or no-op transaction control methods.
• Added a new config attribute DisableOauthRefreshToken to control whether refresh tokens are requested in OAuth exchanges. By default, the driver does not include the offline_access scope. If offline_access is explicitly provided by the user, it is preserved and not removed.

Updated

• Updated sdk version from 0.65.0 to 0.69.0

Fixed

• Fixed SQL syntax error when LIKE queries contain empty ESCAPE clauses.
• Fix: driver failing to authenticate on token update in U2M flow.
• Fix: driver failing to parse complex data types with nullable attributes.
• Fixed: Resolved SDK token-caching regression causing token refresh on every call. SDK is now configured once to avoid excessive token endpoint hits and rate limiting.
• Fixed: TimestampConverter.toString() returning ISO8601 format with timezone conversion instead of SQL standard format.
• Fixed: Driver not loading complete JSON result in the case of SEA Inline without Arrow

Releasing 3.0.3 to Maven

Fixing regression in increased traffic for Token endpoint

v1.0.9-oss

Added

• Added support for providing custom HTTP options: HttpMaxConnectionsPerRoute and HttpConnectionRequestTimeout.
• Add V2 of chunk download using async http client with corresponding implementations of AbstractRemoteChunkProvider and
  AbstractArrowResultChunk
• Telemetry is being enabled by default, with a configuration to opt-out.

Updated

Fixed

• Fixed Statement.getUpdateCount to return -1 for non-DML queries.
• Fixed Statement.setMaxRows(0) to be interepeted as no limit.
• Fixed retry behaviour to not throw an exception when there is no retry-after header for 503 and 429 status codes.
• Fixed encoded UserAgent parsing in BI tools.
• Fixed setting empty schema as the default schema in the spark session.

v1.0.8-oss

Added

• Added DCO (Developer Certificate of Origin) check workflow for pull requests to ensure all commits are properly signed-off
• Added support for SSL client certificate authentication via parameter: SSLTrustStoreProvider
• Provide an option to push telemetry logs (using the flag ForceEnableTelemetry=1). For more details see documentation
• Added putFiles methods in DBFSVolumeClient for async multi-file upload.
• Added validation on UID param to ensure it is either not set or set to 'token'.
• Added CloudFetch download speed logging at INFO level
• Added vendor error codes to SQLExceptions raised for incorrect UID, host or token.

Updated

• Column name support for JDBC ResultSet operations is now case-insensitive
• Updated arrow to 17.0.0 to resolve CVE-2024-52338
• Updated commons-lang3 to 3.18.0 to resolve CVE-2025-48924
• Enhanced SSL certificate path validation error messages to provide actionable troubleshooting steps.

Fixed

• Fixed Bouncy Castle registration conflicts by using local provider instance instead of global security registration.
• Fixed Azure U2M authentication issue.
• Fixed unchecked exception thrown in delete session
• Fixed ParameterMetaData.getParameterCount() to return total parameter count from SQL parsing instead of bound parameter count, aligning with JDBC standards