Glossary
Fabric terminology, defined in plain English, each linked to the full reference page.
- Audit log
- Fabric activity is split across several sources: the Purview/M365 unified audit log (user and admin actions, 180-day retention), the Monitoring hub (recent job runs), the Capacity Metrics app (CU consumption), and workspace monitoring via Eventhouse for anything you want to keep longer. Full reference →
- Capacity (F SKU)
- The purchased unit of Fabric compute, sized F2 through F2048. The number is the Capacity Units it provides, and every workload — Spark, Warehouse, pipelines, Direct Lake, Eventstream, KQL — draws from that one shared, smoothed pool. Full reference →
- Capacity Unit (CU)
- Fabric's single billing currency. Every engine's work — a Spark job, a SQL query, a Direct Lake reframe, a pipeline run — converts to CU-seconds drawn from the capacity's shared pool, smoothed over a rolling window rather than billed per-query. Full reference →
- Deployment pipeline
- A Fabric item with up to three stages (dev, test, prod), each bound to a workspace. Deploying copies item definitions from one stage to the next, matching by name and applying deployment rules to rebind stage-specific values. Full reference →
- Direct Lake
- A semantic-model mode that reads Delta tables straight from OneLake into memory on query, with no import step and no refresh schedule — the model reframes to pick up new table versions instead. Its speed ceiling matches Import mode, but only when the source table is V-Order-tuned and compacted. Full reference →
- Git integration
- Binds a workspace to a branch of an Azure DevOps or GitHub repo. Each supported item serializes to a folder containing a `.platform` file plus item-specific definitions — Commit pushes workspace state to the branch, Update pulls it back. Full reference →
- Lakehouse
- A Fabric item combining managed Delta tables (queryable by every engine via a SQL analytics endpoint) with an unmanaged Files area for raw or landing data — the most common landing spot for data in OneLake. Full reference →
- Mirroring
- Continuous, low-latency replication of an external operational database (Snowflake, Azure SQL, Cosmos DB, Postgres) into read-only Delta tables in a Fabric lakehouse — no pipeline to build, and storage is free up to an allowance sized to your capacity. Full reference →
- Native Execution Engine (NEE)
- Replaces parts of the JVM-based Spark physical plan with a vectorized C++ engine (Gluten + Velox) for scans, filters, joins, and aggregations — same SQL and DataFrame API, no separate charge, and a direct cut in CU consumption when it applies. Full reference →
- notebookutils
- The built-in helper module inside a Fabric notebook session (successor to `mssparkutils`, which still works as an alias) for filesystem access, notebook orchestration, secrets, and lakehouse operations. It only exists inside a live Fabric Spark session. Full reference →
- OneLake
- A single, tenant-wide data lake built on ADLS Gen2. Every workspace gets a folder, every lakehouse and warehouse writes Delta-Parquet into it, and every engine — Spark, SQL, Power BI Direct Lake, KQL — reads the same files. Full reference →
- OneLake security
- Row-, column-, and table-level access rules defined once on a lakehouse and enforced at the data layer itself, so Spark, the SQL endpoint, and Direct Lake all honor the same restriction — unlike the older model where Spark could bypass SQL-endpoint or Power BI security. Full reference →
- OPTIMIZE / compaction
- A Spark job that rewrites a Delta table's small files into fewer, larger ones (targeting 128–256 MB), cutting the CU cost of every downstream read. Frequent MERGE or streaming writes create the small-file problem this fixes. Full reference →
- Resource profile
- A named bundle of roughly twenty Spark configuration values — shuffle partitions, AQE thresholds, file-size targets, broadcast limits — tuned for a workload shape (read-heavy, write-heavy, balanced) so you set one option instead of tuning each individually. Full reference →
- Runtime (Fabric Runtime)
- The versioned bundle of Spark, Delta Lake, and Python that a Fabric notebook or job runs on. Runtimes are deprecated on a schedule, and a version bump can change default behavior — plan upgrades rather than letting them force themselves on you. Full reference →
- Shortcut
- A pointer that makes a folder or table in another storage account, cloud, or Dataverse appear inside a lakehouse with zero copy. Every Fabric engine reads through it like a native table; Fabric caches file bytes on read for up to 24 hours. Full reference →
- V-Order
- A write-time Parquet optimization — sorting, row-group reordering, dictionary encoding — tuned for the VertiPaq engine behind Power BI and the SQL analytics endpoint. Costs 10–25% more CU to write; saves 15–50% on Direct Lake and SQL-endpoint reads of the same data. Full reference →
- VACUUM
- Deletes Parquet files no longer referenced by the current Delta table version and older than the retention threshold, reclaiming OneLake storage. It doesn't touch the transaction log — you lose time-travel to removed versions, not the ability to read the current table. Full reference →
- Variable library
- A single workspace item holding named variables, each with a value per value set (typically one per deployment stage). Notebooks and pipelines reference the variable by name, and the active value set switches automatically on deployment. Full reference →
- Warehouse
- A Fabric item providing full T-SQL read/write semantics over Delta tables in OneLake — the same underlying storage a lakehouse uses, accessed through a warehouse-style engine instead of Spark. Full reference →
- Workspace identity
- A managed Entra service principal that Fabric automatically creates and ties to one workspace, letting items in that workspace authenticate to Azure resources — including storage behind a firewall — with no client secret to store or rotate. Full reference →