Azure Data Factory interviews test whether you can design and run pipelines: linked services and datasets, the three integration runtimes, triggers, expressions, control-flow activities and their limits, incremental loads, mapping data flows, security and CI/CD. Limits quoted below are from Microsoft Learn, checked November 17, 2026. Below are 30 questions by level, five troubleshooting scenarios and how interviews test the service.
Data Factory limits and defaults worth quoting
| Item | Per Microsoft Learn (November 17, 2026) |
|---|---|
| ForEach parallelism | batchCount defaults to 20; at most 50 iterations run in parallel |
| Nesting | A ForEach can't be nested inside another ForEach or an Until; call an inner pipeline instead |
| Lookup activity | Returns at most the first 5,000 rows and fails if the output exceeds 4 MB |
| Integration runtimes | Azure, self-hosted and Azure-SSIS; a self-hosted IR can have up to four nodes |
| Triggers | Pipelines and triggers are many-to-many, except tumbling window triggers |
| Variables | Pipeline-scoped, so Set Variable is unsafe inside a parallel ForEach |
Fundamentals
- What are the main Data Factory components? Pipelines (groups of activities), activities, datasets (the shape and location of data), linked services (connection information), integration runtimes (the compute), triggers, parameters and variables, and mapping data flows.
- Linked service or dataset? A linked service is the connection, like a connection string; a dataset points to specific data through it, such as a table or a folder of files.
- Which integration runtime do you use when? The Azure IR for cloud-to-cloud copies and data flows; the self-hosted IR for on-premises or private-network sources; the Azure-SSIS IR to run existing SSIS packages.
- Which activity types exist? Data movement (Copy), transformation (Data Flow, Stored Procedure, Databricks and other compute activities) and control flow (ForEach, If Condition, Switch, Until, Wait, Lookup, Get Metadata, Execute Pipeline, Set Variable, Filter, Web and Fail).
- Which triggers are there? Schedule, tumbling window, storage event and custom event triggers, plus manual or on-demand runs.
- Parameters or variables? Parameters are passed in and fixed for a run; variables can change during the run through Set Variable or Append Variable; global parameters apply across the factory.
- How do expressions work? With
@functions and system variables, such as@pipeline().parameters.runDate,@activity('GetWm').output.firstRow.wm,@item()inside ForEach and@formatDateTime(utcNow(), 'yyyy-MM-dd'). - What does the Copy activity do, and how do you tune it? It copies between a source and a sink with column mapping, optional staging and fault tolerance for incompatible rows. Tune it with data integration units, parallel copies, source partition options and sensible file sizes.
- How do you process every file in a folder? Get Metadata returns the child items, a ForEach iterates over them, and a parameterized Copy activity handles each file name.
- What are the Lookup activity's limits? It returns at most the first 5,000 rows and fails above 4 MB of output, so page through larger sets or redesign with a staging table.
- How do you monitor and rerun pipelines? Use the monitor view for pipeline and activity runs, rerun from the failed activity, send diagnostics to Log Analytics and set Azure Monitor alerts on failures.
Intermediate
- How do you build an incremental load? Look up the old watermark, look up the new maximum, copy rows between them with a query such as
WHERE ModifiedAt > '@{activity('OldWm').output.firstRow.wm}', then update the watermark with a Stored Procedure activity. - What makes a tumbling window trigger different? It fires for fixed-size, non-overlapping, contiguous windows, keeps state, can backfill past windows, supports dependencies and retries, and passes
@trigger().outputs.windowStartTimeand windowEndTime to the pipeline. Each one serves a single pipeline. - Schedule or tumbling window trigger? A schedule trigger runs on a wall-clock calendar and doesn't track state; a tumbling window trigger processes each time slice exactly once and can rerun or backfill specific windows.
- What are mapping data flows? Visual transformations that Data Factory runs on Spark compute it manages, defined by the Azure IR's data flow settings. Use debug mode for interactive testing and watch cluster time, because it drives cost.
- How do you handle nested loops? A ForEach can't contain another ForEach or an Until, so put the inner loop in a child pipeline and call it with Execute Pipeline, passing the item as a parameter.
- Why is Set Variable risky inside a parallel ForEach? Variables are scoped to the whole pipeline, not to each iteration, so parallel iterations overwrite each other. Pass values as parameters to a child pipeline or write them to a table.
- How do you make datasets and linked services reusable? Parameterize them, for example a schema and table name on one dataset, and pull secrets from Azure Key Vault through a Key Vault linked service.
- How do you secure Data Factory? Managed identity for authentication to storage and databases, Key Vault for secrets, a managed virtual network with private endpoints, self-hosted IR for private sources and role-based access for authors.
- How do you handle errors? Activity dependencies on Succeeded, Failed, Completed and Skipped give try-catch patterns; activity retries cover transient faults; the Fail activity raises a clear custom error.
- How do you copy large volumes quickly? Partitioned source reads, enough data integration units, parallel copies, staged copy into Synapse or similar sinks, compressed files of reasonable size and sinks that support bulk load.
Experienced and senior
- How does CI/CD work for Data Factory? Only the development factory is linked to Git. Merged changes are published as ARM templates and deployed to test and production by a release pipeline with environment parameters. Microsoft recommends pre- and post-deployment scripts to stop and restart triggers, and the same integration runtime names in every stage.
- What is automated publish? Instead of clicking Publish, a build pipeline uses Microsoft's @microsoft/azure-data-factory-utilities npm package to validate resources and generate the ARM templates.
- What is a metadata-driven pipeline? A control table lists sources, targets, load types and watermarks; a Lookup reads it and a ForEach runs one parameterized copy per row, so adding a table is a data change, not a code change.
- How do mapping data flows handle schema drift? Enable Allow schema drift on sources and sinks, use rule-based mappings, and reference columns by name with functions such as
byName(). - How do you control Data Factory cost? Keep data flow clusters short-lived, size data integration units sensibly, avoid needless activity runs in tight loops, and review cost per pipeline in Azure Cost Management.
- How do you build event-driven ingestion? Storage event triggers fire when blobs are created or deleted; make the pipeline idempotent, filter on path and file pattern, and handle late or duplicate files.
- How do you make a self-hosted IR highly available? Register up to four nodes to the same self-hosted integration runtime, which removes the single point of failure and spreads concurrent jobs.
- Data Factory, Synapse pipelines or Fabric? Synapse pipelines share the Data Factory engine but support only Azure and self-hosted runtimes; Microsoft Fabric has its own Data Factory experience. Ask which one the team runs before you answer in depth.
- How do you test pipelines? Debug runs against test data, parameterized test environments, validation activities and row-count checks after copies, and automated runs triggered through the REST API or SDK in CI.
Five troubleshooting scenarios
- A copy from an on-premises SQL Server can't connect. Check that the self-hosted IR is online, the machine can reach the server through its firewall, and the credentials and driver are right.
- A ForEach over 200 tables takes hours. It is sequential, or batchCount is low. Run it in parallel with a batchCount up to 50, within what the source can bear.
- A Lookup returns only part of the control table. It returned the first 5,000 rows. Page the query or move the logic into a staging table.
- A new tumbling window trigger starts running dozens of old windows. Its start time is in the past, so it is backfilling. Set the intended start time or cap concurrency.
- Deployment to production fails. Triggers were still running, or integration runtime names differ between factories. Stop triggers in the pre-deployment script and align runtime names.
How interviews test Data Factory
Employers don't publish question banks, but Azure data engineer candidates commonly report scenario design (an incremental or metadata-driven load), expression questions, CI/CD, and security, alongside SQL and Spark rounds; see our Spark interview questions and Azure interview questions. Teams migrating packages also ask about SSIS.
Where CoPilot Interview fits
CoPilot Interview is an AI interview assistant whose free AI mock interview builds Azure Data Factory and data engineering questions from your résumé and the job description, reads them aloud and scores each spoken answer out of 10. The browser demo at /try has 145 practice questions in 10 tracks, scored out of 10, with no signup. The free plan gives 15 minutes a day of live help in sessions of up to 15 minutes, with AI mock interviews included; Standard costs $24.99 a month and Pro $39.99. For data roles, see our data engineer interview questions. Live coding tasks, take-homes and online assessments usually forbid outside help, so prepare for those; in any live interview, follow your employer's rules.
Disclosure: CoPilot Interview is our product, so weigh what we say about it accordingly.
Frequently asked questions
What is an integration runtime in Azure Data Factory?
The compute that runs activities: the Azure IR for cloud work, the self-hosted IR for private networks and on-premises data, and the Azure-SSIS IR for SSIS packages.
How do you do an incremental load in Data Factory?
Store a watermark in a control table, copy only rows changed since it, then update it; tumbling window triggers suit time-sliced loads.
What is the difference between Data Factory and Databricks?
Data Factory orchestrates and moves data with low-code activities; Databricks is a Spark platform for heavy transformation and data science. Many pipelines use Data Factory to call Databricks.
How many parallel iterations can a ForEach run?
Up to 50, with a default batchCount of 20, per Microsoft Learn on November 17, 2026.
How do I prepare for an Azure Data Factory interview?
Build a metadata-driven incremental pipeline with Key Vault secrets and a tumbling window trigger, deploy it to a second factory through ARM templates, and practice explaining each choice.
Practice this interview before the real one
The AI mock interview builds questions from your resume and the job description, reads them aloud and scores each spoken answer out of 10. Included on the free plan.
Start a mock interview