Lesson 06 of 7 · Git and deployment
Every change through the same gate.
You already keep the kit in Git. In Fabric, the whole workspace joins in, and I get a proper gate.
The kit is already in Git
Today the kit is a folder in a Git repo: the model as , the report as JSON. In Fabric, you connect the workspace to a Git branch with .
What format does Fabric write the model and report in?
In Git or not?
Git holds what should exist. It never holds the data.
The semantic model's TMDL
The rows in gold_supply
The notebook that builds silver
The lakehouse definition: its name and shortcuts
Last night's pipeline run history
The Dataflow Gen2 queries
One change, from branch to production
The kit's workflow scales up. The stops being something the agent runs by hand and becomes a . Tap the steps in order.
Start with the first step.
Three stages
A moves items from one workspace to the next. Each stage has its own lakehouse, so each model must read its own stage's data.
Tap a stage
Dev, test, prod
Same items, three workspaces, three sets of data.
After promoting to production, the prod report shows test data. What went wrong?
Drift
Someone fixes a typo in a measure directly in the production workspace. What's the problem?
Three things to keep
- 1 uses the same text format as PBIP, so the kit's Git habits carry straight over.
- 2The validator becomes a on every pull request: exit code 1 blocks the merge.
- 3A promotes dev to test to prod. Each stage reads its own data, and nobody edits prod directly.