Country or region and language

    Choose your country or region and your language.

    Suggested for you

      All countries and regions

        Choose your character

        It shows in the top bar and on your profile. Change it any time.

        Interviews in Apache Airflow

        Work in the tool while you explain your decisions.

        Choose a role and level. Each interview has a new task, follow-up questions and a report that checks your work and your explanation.

        Speak or type. Sign in to save your results across devices.

        Timed assessment

        • 6 practical questions
        • 35 minutes
        • Mid level

        Questions are drawn from a bank of 18. Your report includes a skill breakdown and lessons to practice next.

        Practice interviews

        Junior tasks cover the fundamentals. Mid and senior tasks add debugging, changing requirements and review decisions.

        6 interviews

        • Data engineer
        • Junior
        • 22 min

        A reusable surcharge total

        Rowan Delivery calculates a signed-delivery surcharge. Wire extract into accepted and accepted into total using TaskFlow calls. accepted retains status signed; total sums amount, including negative corrections. Keep the XCom result small in this exercise.

        Start interview
        Skills and rubricTaskFlow dependencies
        • I call total(accepted(extract())) because the TaskFlow values create the upstream dependencies and carry this run’s data. I reconcile the total to signed rows including corrections. For a large export I would pass a durable object reference instead of the full dataset through XCom.
        • Complete the requested work and identify the source fact that makes it valid.
        • I inspect the upstream graph and each run’s XCom because independent task calls can return plausible values while leaving the actual data dependency missing. I include a signed negative correction to expose an incorrect positive-only filter.
        • Analytics engineer
        • Junior
        • 22 min

        One validation per manifest row

        Aspen Parcels provides a list of integer parcel amounts in hub_case.AMOUNTS. Map double_amount across that runtime list with expand. Return twice each input, preserving the mapped input order. Do not construct a fixed number of tasks from the visible list.

        Start interview
        Skills and rubricDynamic task mapping
        • I pass the upstream list to double_amount.expand because the scheduler must create one mapped task instance per runtime input. I check the mapped indexes and their doubled results against the original list.
        • Complete the requested work and identify the source fact that makes it valid.
        • I test a changed list length and a negative amount because a fixed task count or absolute-value shortcut can pass the visible list. I keep map indexes as execution identities, not permanent business keys.
        • Data engineer
        • Mid
        • 30 min

        An export with two transient failures

        Hazel Routes fails its first two fetch calls and succeeds on the third. Give export_rows two retries and return hub_case.fetch(). The task must finish with the actual rows on its third attempt. Keep errors visible instead of returning empty replacement data.

        Start interview
        Skills and rubricTask retries
        • I distinguish a retrying task instance from a new Dag run because clearing or retrying the failed instance should retain its run context. I inspect attempt logs before deciding whether a failure is transient.
        • I make the destination write idempotent under a stable business or partition key because retry configuration does not prevent duplicate side effects. I test an uncertain commit and retain a source-to-destination reconciliation.
        • Use a counterexample and a readback to defend the result under changed inputs.
        • Analytics engineer
        • Mid
        • 30 min

        A join after an optional route

        Juniper Routing chooses fast_lane or standard_lane from hub_case.ROUTE. Keep both branch edges and make finish run after either selected lane succeeds, without accepting an upstream failure. Use none_failed_min_one_success on finish.

        Start interview
        Skills and rubricBranch joins
        • I test skipped and failed states separately because all_success can skip a valid join, while all_done can let a failed selected route publish a completion marker. Neither state should be mistaken for a successful selected branch.
        • I keep finish blocked by the failed route and handle cleanup or alerting in a separately defined path because completion of cleanup does not establish successful business processing. I retain the failed task state and its exception.
        • Use a counterexample and a readback to defend the result under changed inputs.
        • Data engineer
        • Senior
        • 38 min

        A bounded partner endpoint

        Moss Courier allows at most two concurrent partner requests in this local task model. The partner_api pool already has two slots. Assign export_a and export_b to it, each using one slot. Keep both tasks in the Dag. Do not replace pool control with an unrelated Dag-wide limit.

        Start interview
        Skills and rubricPool capacity
        • I add an appropriate shared rate limit or paced request mechanism because concurrency and request rate use different units. I keep the pool for active capacity and test throttling, retry backoff and the vendor’s quota window.
        • I retain the original request, input identities, code or commands, checked output and run evidence because another engineer needs a reproducible result. I name the remaining assumption and the next check that could change it. The browser model cannot establish production performance or external system behavior.
        • Define the replay or recovery boundary and the evidence required before publication.
        • Analytics engineer
        • Senior
        • 38 min

        A historical dispatch date

        Elm Dispatch tests hub_pipeline for November 5 and November 6, 2026. In selected_rows, accept the ds context argument and filter hub_case.ROWS by that date. Return only the requested day; avoid today’s date and a hard-coded visible date.

        Start interview
        Skills and rubricRun-scoped dates
        • I use the specified data interval start and end with an inclusive start and exclusive end because adjacent hourly runs must neither overlap nor omit boundary rows. I verify the timetable and timezone, then test records exactly on both boundaries.
        • I retain the original request, input identities, code or commands, checked output and run evidence because another engineer needs a reproducible result. I name the remaining assumption and the next check that could change it. The browser model cannot establish production performance or external system behavior.
        • Define the replay or recovery boundary and the evidence required before publication.

        Interview Hub access

        The Interview Hub is included in All Access. Choose a plan or sign in to your account to continue.

        View plansSign in or join

        Your Work Gym account

        Save your progress and pick up on another device. Your browser's progress comes with you.

        or

        You can keep practicing without an account.

        Change your plan

        Checking the price

        Help and feedback

        Ask a question, tell us what is not working, or share an idea. We reply by email.

        What is it about?

        We use it only to reply to you.