Databricks SDKs introduced IAM v2 methods and enforced pagination across identity APIs.

August brought significant identity and access management changes across both the Python and Go SDKs. Both libraries received expanded IAM v2 methods to manage users, groups, service principals, and workspace assignments at account and workspace levels [11, 12]. These additions accompanied a series of breaking changes, as list methods for IAM entities and workspace assignments began requiring explicit pagination handling [6, 7, 9, 10]. Earlier releases also removed legacy internal identifiers from core IAM models [20].

Pipeline connectivity and resource management broadened across multiple releases. Developers gained integration support for NetSuite, LinkedIn Ads, Marketo, RabbitMQ, and external API sources, alongside Postgres snapshot operations and halfvec column types [1, 9, 15, 17]. Compute configurations added H100 GPU and GpuXlarge8 options, while billing APIs received granular principal overrides and usage-blocking alert actions [1, 15, 17, 18]. Under the hood, the Go SDK raised its baseline to Go 1.25, resolved concurrent OAuth token cache conflicts, and adjusted bundle deployment models [6, 7, 8].

In developer tooling, Databricks accelerated the shift away from Terraform state in asset bundles, transitioning default bundle workflows to a direct deployment engine [2, 16]. In the broader ecosystem, the community introduced Minilake, an emulator built to run Databricks SDK and Terraform code locally against real SQL and job runtimes [4].