How does serverless handle runtime upgrades?

Summary

• Upgrades are automatic and managed. Databricks periodically upgrades the serverless runtime for all users; you don't select, patch, or maintain runtime versions.
• Versionless architecture. A stable, versioned client API (based on Spark Connect) decouples your code from the server, so Databricks upgrades the engine underneath without requiring code changes.
• No downtime, with automatic rollback. Over 18 months Databricks automatically upgraded over 2 billion Spark workloads across 25 Databricks Runtime releases at a 99.998% success rate; rare issues auto-roll-back to the last stable version.
• Security patching built in. Serverless compute has a maximum lifetime of 7 days before restarting onto a freshly patched machine image; new images ship roughly every 2–4 weeks.
• Predictable lifecycle. Each environment version pins a Python version and package set and is supported for three years; new workloads use the latest version by default, and you can pin a supported older version.

How does serverless handle runtime upgrades?

With Databricks serverless compute, runtime upgrades are fully managed — there is no infrastructure to maintain and no runtime version to pick, patch, or upgrade yourself. Databricks periodically rolls out updates to the serverless runtime so every workload runs on a current, secure engine.

Why Databricks serverless makes runtime upgrades effortless

• Automatic, managed upgrades. Databricks periodically releases updates and automatically upgrades the serverless runtime to deliver performance improvements, security fixes, and new capabilities; all users receive the same updates, rolled out over a short window.
• Versionless architecture. A stable, versioned client API based on Spark Connect decouples the client from the Spark server, so Databricks can upgrade the server independently — with no code changes required in your workloads.
• Proven at scale, with rollback. Over the past 18 months Databricks automatically upgraded over 2 billion Spark workloads across 25 Databricks Runtime releases with a 99.998% success rate; if an issue is detected, workloads are automatically rolled back to the previous stable version.
• Security patching without intervention. Serverless compute has a maximum lifetime of 7 days before it automatically restarts onto the latest machine image with current security patches; new images are released roughly every 2–4 weeks.
• Immutable infrastructure. Systems are replaced rather than patched in place, which improves reliability and security.
• Predictable environment lifecycle. Each environment version specifies a Python version and package set and is supported for three years; new workloads use the latest version by default, and you can select a supported older version if needed.

Getting started

1. Choose serverless compute for notebooks, jobs, or SQL so runtime management is handled for you.
2. Optionally pin a specific supported environment version for reproducibility.
3. Let Databricks roll out runtime and security updates automatically.
4. See the serverless documentation and the serverless release notes for environment-version details, and the versionless Apache Spark blog for how upgrades work.

FAQs

Do I need to upgrade the serverless runtime myself?

No. Databricks automatically upgrades the serverless runtime for all users; there is no runtime version to install, patch, or maintain.

Will an upgrade break my workloads?

A versionless, stable client API keeps applications compatible while the server is upgraded, and rare failures are automatically rolled back to the last stable version.

How does serverless stay patched for security?

Serverless compute has a maximum lifetime of 7 days before restarting onto a freshly patched machine image, and new images are released roughly every 2–4 weeks.