Summary
The SDK adds `AwsAccessKey` field support for Amazon Bedrock provider configuration and `EntraServicePrincipal` field support for Azure OpenAI and Microsoft Foundry provider configurations. These changes enable additional authentication options when connecting external model providers to Databricks model serving.
Summary generated by brickster.ai. For the full changelog and any code/binary attachments, follow the GitHub link above.
More from databricks/databricks-sdk-go
This release introduces a new AI Gateway workspace-level service for managing AI workloads. Bundle Deployments API is enhanced with an UpdateOperation method and additional fields for tracking deployment versions, operations, and dashboard metadata.
The v0.165.0 SDK adds three new API capabilities: PatchTelemetryConfig for serving endpoints, FanoutOptions for pipeline schemas, and Parameters for SQL alerts. These additions enable more granular configuration of telemetry, pipeline specifications, and alert behavior.
The catalog API adds new special destination enum values for specifying schema, table, volume, function, and registered model owners. The Bundle Deployments API introduces breaking changes: CreateDeployment now requires a different argument order and no longer accepts the DeploymentId field in CreateDeploymentRequest.
The SDK adds DependencyMode configuration options for Spark clusters across creation and management operations, plus new fields for AI runtime tasks, schema registry configuration, and PostgreSQL synced table specifications. The release includes a breaking change removing the Lifetime field from TimeWindow in the ML service.
The SDK now includes effective entitlements for workspace assignments, showing the actual permissions users have after assignment. Job clusters can now be configured with serverless compute IDs for serverless-based job execution.
The SDK adds job performance targeting, enhanced ML streaming with schema support, and expanded pipeline and networking capabilities including Reddit connector options. Breaking changes to IAM workspace assignment methods and ML time window configurations require updating existing code.