The third installment of a series on running Backstage on Databricks Lakebase demonstrates a FinOps use case: joining live operational Postgres data from Backstage with Databricks billing data in a single SQL query, with zero ETL pipelines. Lakebase isolates compute per workload, allowing analytical queries without impacting the live portal (catalog queries at 55–65 ms, searches at 2–4 ms). Lakehouse Federation connects the Postgres catalog to Unity Catalog, though a workaround is needed since Federation only supports static credentials rather than OAuth JWTs — requiring a separate Postgres role with SCRAM-SHA-256 auth. The result is per-branch, per-developer cost visibility in billing data, including ephemeral CI branches. Lakebase's scale-to-zero behavior means idle branches incur no cost. The post also warns about orphaned branch accumulation and recommends TTL-based lifecycle controls.
