Databricks Lakebase introduces autoscaling for serverless Postgres, solving the classic provisioning paradox where developers must choose between over-provisioning or risking failures during traffic spikes. The system uses Compute Units (CUs), where 1 CU = 2 GB of memory, and dynamically adjusts resources by monitoring CPU load, memory usage, and working set size — all without database restarts. Developers set a min/max CU range (max spread of 8 CU, up to 32 CU total). A scale-to-zero feature suspends idle instances after a configurable timeout, cutting costs by 70%+ for dev branches and bursty workloads. The approach is particularly useful for AI agent workloads and database branching in CI/CD pipelines.
Table of contents
What is Lakebase Autoscaling?The Mechanics: How the Scaling Algorithm ThinksConfiguring the Guardrails: Min and Max CUsReal-World Scenarios: When Autoscaling WinsSort: