How quickly can a serverless application start running?

Summary

• In seconds. Serverless compute on the Databricks Platform starts in seconds because Databricks keeps warm pools of pre-initialized instances ready, so there is no wait for infrastructure to provision.
• SQL warehouses: about 2 to 6 seconds. Serverless SQL warehouses typically start in roughly 2 to 6 seconds, so queries begin almost immediately.
• Jobs and notebooks: seconds. In performance-optimized mode, serverless compute for jobs and notebooks starts up in seconds; interactive notebooks always start in seconds and run fast.
• Standard mode trades startup for cost. Standard mode has a longer startup (about 4 to 6 minutes) in exchange for lower cost, so you choose the mode that fits the workload.
• Environment caching speeds repeat runs. You define an environment once, and future runs load it from cache in seconds.

How quickly can a serverless application start running?

When an application runs on serverless compute on the Databricks Platform, it starts in seconds rather than waiting for infrastructure to be provisioned. Databricks keeps warm pools of pre-initialized instances ready so compute is available the moment a workload is triggered. Exact startup depends on the workload type and the compute mode you choose.

Why Databricks Platform serverless starts fast

• Warm pools of ready instances. Databricks maintains warm pools of instances so compute is ready when teams are, which removes the cold-start delay of provisioning new machines on demand. See What is serverless computing?.
• SQL warehouses start in about 2 to 6 seconds. Serverless SQL warehouses typically achieve a startup time of roughly 2 to 6 seconds, so interactive queries and BI dashboards begin almost immediately.
• Jobs and notebooks start in seconds. In performance-optimized mode, serverless compute for jobs and notebooks starts up in seconds by drawing from warm pools and scaling resources aggressively, which suits interactive and time-sensitive workloads. Interactive notebooks always start in seconds and run fast. See The evolution of data engineering: how serverless compute is transforming notebooks and Lakeflow Jobs.
• Standard mode optimizes for cost. Standard mode has a longer startup latency of about 4 to 6 minutes, depending on compute availability and whether the workload uses Spark, in exchange for lower cost. You pick performance-optimized mode for the fastest startup or standard mode for cost efficiency. See Best practices for serverless compute.
• Environment caching speeds up repeat runs. You define your environment once, and future runs load that environment from cache in seconds, so recurring workloads avoid repeating setup work.

Getting started

• Read Use serverless architectures for platform best practices.
• Turn on serverless SQL warehouses for interactive analytics and BI with near-instant startup.
• Choose the serverless compute mode that matches each workload: performance-optimized for fastest startup, standard for cost efficiency.

FAQs

How fast does serverless compute start?

Serverless compute starts in seconds because Databricks keeps warm pools of pre-initialized instances ready, so there is no wait for infrastructure to provision.

How quickly does a serverless SQL warehouse start?

Serverless SQL warehouses typically start in about 2 to 6 seconds, so queries and dashboards begin almost immediately.

Why is standard-mode startup longer?

Standard mode optimizes for cost rather than the fastest startup, so it has a longer startup latency of about 4 to 6 minutes depending on compute availability and whether the workload uses Spark.

Can repeat runs start even faster?

Yes. You define an environment once, and future runs load that environment from cache in seconds, so recurring workloads skip repeated setup.