Databricks platform self-check
01 Move access to Entra ID groups defined in Terraform
Give access to groups, not users, with reader and contributor groups per schema, and define them in code. Then review them every quarter.
02 Give every workload a cost owner
Apply cluster policies and tags per team, set budgets with alerts and review the system billing tables monthly with the owners.
03 Deploy Asset Bundles only through pipelines
Use Asset Bundles for jobs and pipelines, deploy to development, test, acceptance and production from a pipeline, and validate and scan every pull request.
04 Close public paths and remove long-lived tokens
Use VNet injection and private endpoints, turn public access off, use managed identities and Key Vault-backed secret scopes, and run the Security Analysis Tool to find the rest.
05 Get an independent second opinion
A strong platform still drifts. A short review of cost, security, governance and delivery finds unused permissions, idle compute and exceptions that never expired.
Data platform maturity check
06 Build one governed foundation with a first pipeline
Bring the data for one business question into one platform with raw and curated layers, scheduled pipelines and alerts. Start small and put it in production.
07 Move access to groups and name owners
Give access through Entra ID groups per catalogue and schema, name an owner group for every dataset and review access every quarter. Define the groups in code.
08 Deliver everything as code through reviewed pull requests
Stop changing resources by hand. Put infrastructure and pipelines in Git, add a test and acceptance stage and validate every pull request before it reaches production.
09 Agree one version of the key numbers
Pick the five numbers management asks for most, define each once, name an owner and publish them from governed data. A quick scan can rank which to fix first.
10 Get an independent review
A platform that scores well still has blind spots. A short external review of cost, security, governance and delivery finds them before an incident does.
A checklist is a starting point. It lists what we would ask first, not everything a platform or pilot needs.