1.1 — Configure Fabric Workspace Settings
DP-700 Domain 1 · Skill 1.1
What You Need to Know
Workspace settings decide how every Spark job, domain assignment, shortcut read and Airflow job in a workspace behaves. You need the workspace Admin role to change the Spark settings. The exam tests which setting solves which problem.
Spark settings (Workspace settings → Data Engineering/Science → Spark settings)
| Setting | What it does | Pick it when |
|---|---|---|
| Starter pool (default) | Prehydrated, medium-size nodes; sessions start in about 5–10 seconds | You want fast startup for everyday notebooks |
| Custom pool | You choose node size, autoscale and dynamic executor allocation; a session takes about 3 minutes to start | Jobs need bigger or smaller nodes. A capacity admin must allow Customized workspace pools |
| Default environment | An Environment item with the runtime version, Spark properties and libraries | Every new notebook should get the same libraries or Spark settings by default |
| High concurrency | Notebooks (and notebooks run by pipelines) share one Spark session | Faster starts and lower cost when the same user runs several notebooks |
| Jobs | Optimistic job admission (default) or Reserve maximum cores; session timeout (20 minutes by default); maximum job lifetime | You need to stop runaway jobs or guarantee cores |
Making a library available to all notebooks takes three steps. Create an Environment, add the library (a public package or an uploaded custom .whl) and publish it. Then set it as the workspace default environment.
High concurrency in detail
- A session is shared only when the notebooks run under the same user, use the same default lakehouse and have the same Spark compute settings. Otherwise Fabric starts a separate session.
- Each notebook gets its own REPL, so variables don't collide, but the notebooks share executors.
- Up to 5 notebooks share a session by default. You can raise this to 50 with
spark.highConcurrency.maxin the environment. - Only the notebook that started the session is billed.
- If notebooks must each run in their own session (for example so that a heavy job can't take executors from another), turn high concurrency off.
Domain settings
- A domain groups workspaces by business area (data mesh). Assigning a workspace assigns all of its items. Domains help people find content but don't grant access.
- Fabric admin: creates and deletes domains, names domain admins, and can do everything below.
- Domain admin: creates subdomains, assigns workspaces and names domain contributors. A domain admin can't delete or rename the domain or add other domain admins.
- Domain contributor: a workspace Admin who is allowed to assign their own workspaces to the domain, from the workspace's settings.
- A default domain for users or security groups automatically assigns workspaces that those people create.
OneLake settings: shortcut cache
Turn on the cache under Workspace settings → OneLake to cut cross-cloud egress costs.
| What | Detail |
|---|---|
| Sources cached | Google Cloud Storage, Amazon S3, S3-compatible and on-premises data gateway shortcuts. ADLS Gen2 is not cached, because there's no cross-cloud egress to save |
| Retention | 1–28 days, and the clock resets each time a file is read |
| Size limit | Files larger than 1 GB aren't cached |
| When the source changes | If the source has a newer version of a file, OneLake reads it from the source and refreshes the cache |
| Reset cache | Clears the cache |
Data workflow settings: Apache Airflow jobs
- Under Workspace settings → Data Factory → Apache Airflow runtime settings, you choose the default pool.
- The starter pool suits development and shuts down after 20 minutes of inactivity.
- A custom pool stays on until you pause it and suits production. You can set the node size, autoscale and extra nodes. You pay for its base nodes while it runs.
Key Concept: Decide whether the requirement is about speed (starter pool, high concurrency), isolation (high concurrency off, or a custom pool), defaults (default environment) or cost (shortcut cache, high concurrency billing).
Exam Traps
- Installing a shared library with
%pip installin every notebook instead of publishing it in a default Environment - Expecting a custom pool to start as fast as the starter pool
- Expecting two users to share one high-concurrency session (sharing stays within one user)
- Thinking that assigning a workspace to a domain gives domain members access to its data
- Turning on the shortcut cache for ADLS Gen2 shortcuts, or for files over 1 GB
Check your understanding
Practice Scenario
An admin changed the default pool of a Fabric workspace from the starter pool to a custom pool with large nodes. Engineers now wait about three minutes for every interactive notebook session to start. Their notebooks process small files and don't need large nodes. What should the admin do to restore session start times of a few seconds?
Log in to save your progress.
Sources
- Workspace administration settings in Microsoft Fabric, Microsoft Learn
- High concurrency mode in Apache Spark for Fabric, Microsoft Learn
- Fabric domains, Microsoft Learn
- OneLake shortcuts: caching, Microsoft Learn
- Apache Airflow Job workspace settings, Microsoft Learn