Browse cheat sheets · Microsoft Fabric

Technology: Microsoft FabricCategory: Administration & ConfigurationDifficulty: IntermediateUpdated

Microsoft Fabric Workspace Settings Cheat Sheet

The Microsoft Fabric workspace settings that change how a data engineering workspace behaves: Spark pools, environments and jobs, domain assignment, OneLake shortcut caching and diagnostics, and data workflow (Apache Airflow job) runtime pools, with what to verify for each.

On this page
  1. Problem
  2. Quick answer
  3. Code / commands
  4. 1. Spark workspace settings
  5. 2. Domain workspace settings
  6. 3. OneLake workspace settings
  7. 4. Data workflow workspace settings (Apache Airflow job)
  8. 5. Settings review in one pass
  9. What the result means
  10. What to look for
  11. Production notes
  12. Common mistakes
  13. 1. Confusing workspace settings with tenant or capacity settings
  14. 2. Using domains as a security boundary
  15. 3. Assuming the workspace default always applies
  16. 4. Enabling immutability first and asking about retention later
  17. 5. Searching for “data workflow” and finding nothing
  18. Next step
  19. Related deep dives

Problem

A Fabric workspace has dozens of settings. Four groups change how a data engineering solution runs, costs or is governed:

  • Spark settings: which pool, runtime and environment every notebook starts with, and how long jobs and sessions may live.
  • Domain assignment: which business domain owns the workspace, and which delegated governance settings apply to its items.
  • OneLake settings: shortcut caching for external storage, and diagnostic logs of who read what.
  • Data workflow settings: the runtime pool that Apache Airflow jobs use.

The rest of the settings pane (name, description, license mode, Git, network) matters too, but it is covered elsewhere. This sheet answers one question per group: what is it, when do I care, and how do I prove it is set the way I think?

Last verified against Microsoft Learn: November 2026. Fabric settings move fast; check the current page before you change production.

Quick answer

Group Where in workspace settings Who can change it Why you care
Spark Data Engineering/Science → Spark Compute (Pool, Environment, Jobs, High concurrency) Workspace Admin Start time, runtime version, libraries, concurrency, runaway jobs
Domain The workspace’s domain assignment Workspace Admin who is also a domain contributor (or a Fabric/domain admin from the admin side) Discovery in the OneLake catalog, delegated certification and default label settings
OneLake OneLake tab Workspace Admin (diagnostics also needs Contributor on the target lakehouse) Egress cost of external shortcuts; access audit logs
Data workflow Data Factory → Apache Airflow Runtime Settings Workspace Admin Airflow start latency, always-on cost, DAG concurrency

Three rules that prevent most mistakes:

Workspace settings change behaviour for every item in the workspace.
Domain assignment is never an access control.
"Data workflow" is the earlier name of the Apache Airflow job.

Code / commands

1. Spark workspace settings

Tab Setting What it controls
Pool Default pool: Starter pool or a custom pool Starter pools are prehydrated and start in seconds; a custom pool lets you size nodes, autoscale and dynamically allocate executors, but sessions can take about 3 minutes to start
Pool Customize compute configuration for items Whether environments can override driver and executor cores and memory. Off means every environment uses the default pool’s configuration
Environment Set default environment The environment (runtime, libraries, Spark properties) every notebook and Spark job definition gets unless it attaches its own. Turned off, you choose only a runtime version
Jobs Reserve maximum cores for active Spark jobs Switches off optimistic job admission (the default) and reserves each job’s maximum cores up front
Jobs Spark session timeout Expiry of interactive notebook sessions. Default: 20 minutes
Jobs Set maximum job lifetime Cancels user-submitted jobs (notebooks, Spark job definitions, Livy sessions) that run past the limit. System jobs such as table maintenance and materialized lake view refreshes are not affected
High concurrency For notebooks / for pipeline-run notebooks Whether several notebooks can share one Spark session
Automatic log ML autologging Whether MLflow captures parameters and metrics automatically

Custom pools also depend on a capacity-level switch, Customized workspace pools, which is on by default. If you can’t create a pool, check the capacity before the workspace.

What matters for DP-700 Spark settings
Understand Pool, environment and runtime are three different things: compute shape, packaged configuration, and Spark version
Configure Create a small custom pool, create an environment on a specific runtime, make it the workspace default, set a session timeout
Verify Start a new notebook session and check the runtime and executor settings it actually got (below); compare session start time before and after
Common mistake Switching the default pool to a custom pool to “make Spark faster” and getting slower starts because the starter pool’s prehydrated sessions are no longer used

Verify what a session actually received, from a notebook:

# Spark version of the runtime the session started with
print(spark.version)

# Executor shape the session got (compare with the pool/environment you configured)
for key in ["spark.executor.cores", "spark.executor.memory", "spark.dynamicAllocation.enabled"]:
    print(key, "=", spark.conf.get(key, "not set"))

A notebook that attaches its own environment ignores the workspace default. When the numbers don’t match the workspace settings, check the notebook’s environment selector first.

2. Domain workspace settings

Assigning a workspace to a domain associates every item in the workspace with that domain. It does three things:

  • Consumers can filter the OneLake catalog by the domain.
  • Settings the tenant admin delegated to the domain apply to the workspace’s items: currently a domain default sensitivity label (where enabled) and certification settings (who may certify, and whether certification is allowed).
  • It records business ownership, which governance reports and lineage reviews can group by.

It does not change who can see or open an item. Access still comes from workspace roles, item permissions and data-level security.

What matters for DP-700 Domain settings
Understand Domains (and subdomains) are created by Fabric admins; domain admins manage them; domain contributors assign workspaces they administer
Configure As a workspace Admin who is a domain contributor, assign the workspace to a domain or subdomain from the workspace’s own settings
Verify Open the OneLake catalog, filter by the domain, and confirm the workspace’s items appear; check whether the domain overrides certification settings
Common mistake Treating a “Finance” domain as a security boundary. A user with no role or permission still can’t open Finance items, and a user with one still can, whatever the domain

Two admin-side behaviours explain “why did my workspace move?”:

Default domain: workspaces whose admins are listed users or groups are assigned automatically,
                if they aren't assigned yet, and so are new workspaces those users create.
Override setting: Fabric and domain admins can reassign already-assigned workspaces only when
                  the tenant setting that allows overriding workspace assignments is enabled.

3. OneLake workspace settings

The workspace OneLake tab currently holds two things.

Setting What it does Key facts
Shortcut cache Caches files read through external shortcuts in the workspace, so repeated reads don’t go back to the remote store Retention 1–28 days, reset on each access; files over 1 GB aren’t cached; supported for Google Cloud Storage, Amazon S3, S3-compatible and on-premises data gateway shortcuts (not ADLS Gen2); Reset cache empties it
Diagnostic events Streams OneLake data-access events as JSON into a lakehouse you choose Lakehouse must be on the same capacity; needs workspace Admin plus Contributor on the lakehouse; events can take up to an hour to appear; moving the workspace to another capacity disables it
Diagnostic immutability Makes the diagnostic JSON files write-once for a retention period Configured on the workspace that holds the diagnostic lakehouse; can’t be shortened or removed once applied

What is not a workspace OneLake setting: whether apps outside Fabric (OneLake file explorer, ADLS APIs, external tools) may reach OneLake. That is the tenant setting Users can access data stored in OneLake with apps external to Fabric. Data access rules for tables and folders are OneLake security roles on the item, not workspace settings.

What matters for DP-700 OneLake settings
Understand Caching is a cost and latency control for cross-cloud shortcuts; diagnostics is an audit trail of data access
Configure Turn on the shortcut cache with a retention period; turn on diagnostic events into a dedicated lakehouse
Verify Read through an S3 or GCS shortcut twice and compare; an hour after enabling diagnostics, read the events (below)
Common mistake Expecting the cache to help an ADLS Gen2 shortcut, or applying an immutability period before the retention requirement is agreed

Read the diagnostic events from a notebook attached to the diagnostic lakehouse:

# Path documented by Microsoft: Files/DiagnosticLogs/OneLake/Workspaces/<WorkspaceId>/y=YYYY/m=MM/d=DD/h=HH/m=00/PT1H.json
events = spark.read.json("Files/DiagnosticLogs/OneLake/Workspaces/*/*/*/*/*/*/*.json")

(events
 .select("accessStartTime", "executingUPN", "originatingApp", "operationName", "resource", "isShortcut", "httpStatusCode")
 .orderBy("accessStartTime", ascending=False)
 .show(20, truncate=False))

executingUPN is empty when the tenant setting that includes end-user identifiers in OneLake diagnostic logs is turned off. Access through Fabric engines is logged as temporary access granted to the workload; the engine’s own logs show the query.

4. Data workflow workspace settings (Apache Airflow job)

The DP-700 skills outline still says data workflow; the Fabric item is now the Apache Airflow job. The workspace setting picks the default Airflow runtime pool for the workspace.

Property Starter pool (default) Custom pool
Size Large compute node Choose node size (Small or Large), autoscale and extra nodes
Start Instant Created in a stopped state
Idle behaviour Shuts down after 20 minutes of inactivity Always on until you pause it
Resume Up to about 5 minutes Up to about 5 minutes
Suggested for Development Production

Each extra node in a custom pool adds capacity for three more workers, so more DAGs can run concurrently. Private networks and virtual networks aren’t currently supported for Apache Airflow jobs.

What matters for DP-700 Data workflow settings
Understand Airflow is code-first orchestration (Python DAGs) inside Fabric; pipelines are the low-code orchestrator. The workspace setting decides the runtime, not the DAGs
Configure Workspace settings → Data Factory → Apache Airflow Runtime Settings → Default Data Workflow Setting → New Pool
Verify Create an Apache Airflow job, check which pool it uses, run a simple DAG, and note start and resume times
Common mistake Leaving a custom pool running for a development workspace: it is always on until paused, and it consumes capacity while idle

5. Settings review in one pass

Use this as a review list for any workspace that runs data engineering workloads:

[ ] Default Spark pool, and why it is not the starter pool (if it isn't)
[ ] Default environment and its runtime version; notebooks that override it
[ ] Session timeout and maximum job lifetime
[ ] High concurrency on or off, deliberately
[ ] Domain assignment and the delegated settings it brings
[ ] Shortcut cache: on only where external S3/GCS/gateway shortcuts are read repeatedly
[ ] OneLake diagnostics: destination lakehouse, who can read it, immutability decision
[ ] Airflow runtime pool: starter for dev, custom (paused when idle) for prod

What the result means

  • Notebook start time jumped after a settings change: the default pool or environment changed. Custom pools and environments with extra libraries start slower than the starter pool.
  • A notebook runs a different Spark version than the workspace default: it attaches its own environment, which wins over the workspace default.
  • Jobs are cancelled at the same duration every time: a maximum job lifetime is set. Table maintenance and materialized lake view refreshes continue regardless.
  • The workspace appears under a domain nobody chose: a default domain matched one of its admins.
  • Shortcut reads are no faster with caching on: the shortcut type isn’t cached (ADLS Gen2, for example), the files exceed 1 GB, or the remote file changed so OneLake served the newer version.
  • No diagnostic events: wait up to an hour after enabling; then check the lakehouse is on the same capacity and the workspace hasn’t moved capacity.

What to look for

  • A custom default pool with no written reason.
  • Many notebooks attaching their own environments, so the workspace default no longer describes what runs.
  • A Spark session timeout raised to hours “to avoid restarts”: idle sessions hold capacity.
  • No maximum job lifetime in shared or production workspaces.
  • Workspaces with no domain, or domain names used in access-request processes as if they granted access.
  • Diagnostic logs stored in a lakehouse that the people being audited can also write to.
  • An always-on custom Airflow pool in a development workspace.

Production notes

  • Write settings down. Most workspace settings are not in Git and do not move with deployment pipelines. Record them per environment (Dev, Test, Prod) so a rebuilt or new workspace can be configured the same way.
  • Keep the environment as code where you can. Environments are supported items in Git integration and deployment pipelines; the workspace default that points at them is a setting you must set in each workspace.
  • Isolate audit data. Microsoft recommends a dedicated workspace for OneLake diagnostic logs, with few admins, especially when immutability is on.
  • Plan log retention. Immutable diagnostic files can’t be deleted until their period expires, so storage grows; schedule a cleanup that skips files still inside the window.
  • Domains belong to the business. Assign workspaces to domains with the domain owners, then use the domain’s delegated certification settings for endorsement.

Common mistakes

1. Confusing workspace settings with tenant or capacity settings

Custom pools need the capacity-level Customized workspace pools switch. External app access to OneLake is a tenant setting. Domain creation is a Fabric admin task. Workspace Admins configure what the higher level allows.

2. Using domains as a security boundary

Domain assignment never grants or removes access. Secure data with workspace roles, item permissions, OneLake security and SQL security.

3. Assuming the workspace default always applies

Notebooks and Spark job definitions can attach their own environment, and “Customize compute configuration for items” lets environments change compute. Verify from inside the session.

4. Enabling immutability first and asking about retention later

The immutability period can’t be changed once it is applied.

5. Searching for “data workflow” and finding nothing

The current product name is the Apache Airflow job. The workspace section is Apache Airflow Runtime Settings, and its dropdown still says Default Data Workflow Setting.

Next step

Configure one test workspace end to end:

  1. Create an environment on the runtime you use, make it the workspace default, and measure a notebook’s start time.
  2. Set a session timeout and a maximum job lifetime, and record both.
  3. Assign the workspace to a domain and find its items through the OneLake catalog domain filter.
  4. Enable OneLake diagnostics into a separate lakehouse and read the first events.
  5. Create an Apache Airflow job on the starter pool and note its start and resume behaviour.

Then secure the same workspace and promote it to Test: the security and lifecycle labs below use exactly this kind of workspace.

← All cheat sheets