VS Code sync & extension guide

Edit Fabric notebooks, environments, and pipelines from VS Code — with Git, real linting, and no browser tab.

Why leave the web UI

Real editor: multi-cursor, extensions, ruff/black, GitHub Copilot.
Local Git: branch, diff, rebase without the workspace source-control panel.
Run and debug notebooks against a remote Fabric Spark session or locally.

Options

The Fabric Data Engineering VS Code extension (Microsoft, from the Marketplace) is the primary path.

Install the extension + the Jupyter and Python extensions.
Ctrl/Cmd-Shift-P → Fabric: Sign In.
Fabric: Set Workspace → pick your dev workspace.
Notebooks appear under the Fabric explorer. Open one — it edits the .py representation locally.
Pick a kernel: Fabric Runtime (remote Spark) for real capacity, or a local Python env for pure-Python cells.

Recommended local layout

Ingest.Notebook.py

Transform.Notebook.py

pyproject.toml

README.md

// .vscode/settings.json
{
  "notebook.formatOnSave.enabled": true,
  "[python]": { "editor.defaultFormatter": "charliermarsh.ruff" },
  "python.testing.pytestEnabled": true
}

Local unit tests for notebook logic

Extract pure functions out of notebooks into a src/ package and test them with plain pytest + a local pyspark:

# src/transforms.py
from pyspark.sql import DataFrame, functions as F

def add_audit_cols(df: DataFrame, run_date: str) -> DataFrame:
    return df.withColumn("run_date", F.lit(run_date)) \
             .withColumn("_ingested_at", F.current_timestamp())

# tests/test_transforms.py
def test_add_audit_cols(spark):
    out = add_audit_cols(spark.createDataFrame([(1,)], ["id"]), "2026-01-01")
    assert "run_date" in out.columns
    assert out.first()["run_date"] == "2026-01-01"

The notebook then just imports and calls — thin glue, thick tested library.

Keep the notebook's parameter cell as the only stage-specific surface. Everything below it should be importable, testable Python.

Gotchas

mssparkutils / notebookutils only exist in a Fabric session. Guard imports so local test collection doesn't blow up.
Remote kernel start is slow (pool spin-up). Keep a session warm while iterating.
The extension writes the .platform metadata file — commit it; don't edit it.