SYSTEM Cited by 1 source
Databricks FEVM¶
The Field Engineering Vending Machine (FEVM) is Databricks' internal self-service infrastructure provisioning platform. It replaces shared workspaces with isolated, just-in-time, use-case-based environments for the field engineering organisation (5,000+ active users across 7,000+ GTM).
Architecture¶
- Deployment: Databricks App (React frontend + Python backend)
- Provisioning engine: Terraform templates fetched per request and fed to a Git Runner
- State store: Lakebase — tracks every resource (identity, owner, purpose, expiration)
- Multi-cloud: AWS, Azure, GCP
- Agent interface: MCP server as single point of control — chat, external services, and AI agents all use the same API
- Admin UI: global configuration (workspace limits by cloud, audit of active deployments, defaults tuning)
Design Principles¶
- Use-case-based provisioning — engineers describe intent ("demo for financial services prospect"), not raw resources; system maps to configured environment via template catalog
- Transparency — every lifecycle event (provisioning, approaching expiration, deletion) triggers Slack notification
- Governed templates — vetted, hardened, tested for security; no ad-hoc provisioning paths
- Resource-level lifecycle independence — catalogs persist independently of workspaces; reattach automatically in the same region
Operational Envelope¶
| Metric | Value |
|---|---|
| Active users | 5,000+ |
| Active deployments | 2,600+ (3 clouds) |
| Peak burst | ~1,200 requests/day (BuildCon) |
| Default TTL | 90 days (builder environments) |
Seen In¶
- sources/2026-07-23-databricks-self-serve-infrastructure-vending-machine — first wiki disclosure (2026-07-25)