v0.157.0
Summary
This release adds the Mason service, AI Gateway skill management methods, and new Unity Catalog securable types for skills and agent services. It also introduces maintenance notifications for jobs, support for TikTok Ads and Smartsheet ingestion sources, and budget policy controls for ML feature operations.
Summary generated by brickster.ai. For the full changelog and any code/binary attachments, follow the GitHub link above.
More from databricks/databricks-sdk-java
The SDK adds support for configuring environment variables across Jobs and tasks, alongside new aiDecide and listCommands client methods. It also makes commandPath in deployment specifications and apiSecretRef in schema registry configurations optional.
The SDK adds MCP service user mapped credential management APIs to the AI Gateway service, enabling better credential handling for AI workloads. Pipeline transformers now support AVRO and PROTOBUF formats and include a groupName field for RunAs configuration.
This release adds new API fields across multiple Databricks services, including support for Unity Catalog image paths in AI runtime tasks, feature view sources in ML data sources, budget policies and tags in ML publishing, and a new GPU_8X_B300 compute type. These additions enable Java SDK users to access recently added Databricks platform capabilities for jobs, ML workflows, pipelines, and workspace settings management.
The Java SDK now supports purging feature entities via a new method and adds deny-based policy controls in the catalog service. ML feature ingestion and materialized features now support budget policies and tags for improved cost management and organization.
Breaking changes remove rate limiting fields and enums while adding a new domains service and sandbox command execution method. The ML feature engineering APIs gain new fields for handling data sources, time windows, and scheduling.
This release adds SDK support for AI Functions and Sandboxes, as well as new backfill management methods for Feature Engineering. Breaking changes remove the owner field across multiple catalog model services, along with trafficSplitting from model routing and disabled from inference table configurations.