D8LooP
Start building

Interviewing

Senior Data Engineer Interview Preparation

Four things an interviewer reads for when deciding the level, not the offer.

4 signals37 steps~10h total
Start this plan

It opens on your dashboard and ticks itself off as you go. Every step below is a chapter or an exercise that already ships.

01

When you finish, you can

You argue cost, failure and change without being asked to.

The level call You argue trade-offs, not features

02

The plan, 4 signals

Ordered by what each one buys you, not by topic. Every step says why it is here.

  1. Signal 01 · ~3h

    You can put a number on it

    You can price a design and say which knob moves the bill.

    Skip this if you already quote cost per run before proposing anything.

    1. SectionPaying for performance7 minEvery scaling answer is a spending answer. Say the number.
    2. SectionRent a clock, or buy bytes7 minTwo billing models, two different optimisations. Know which you are on.
    3. SectionSlow is not expensive6 minThe distinction juniors never make, and seniors are asked to make.
    4. SectionSizing, dynamic allocation & cluster managers7 minCluster shape is a cost decision wearing a performance costume.
    5. InterviewSize a cluster for 1 TB6 minSubtract the overhead out loud before you divide. That is the signal.
    6. InterviewCapacity is not just throughput6 minTotal capacity can be fine while one partition is on fire.
    7. ExerciseIt works, and it costs too much45 minin the studioIt works and costs too much. The most senior kind of review.
    8. ExerciseOne tenant ate the cluster50 minin the studioOne customer, everyone else degraded. Isolation costs money too.
    9. ExerciseThree ways to compute customer spend: rank them by shuffle size before anyone guesses20 minin the studioOn Spark, the first number to price is shuffle bytes.
    10. QuestionsData Pipeline · Cost & scale15 minEstimate before you build. Bytes per day, then everything else.

  2. Signal 02 · ~3h

    You name the failure before it happens

    You can list what breaks and what each break costs downstream.

    Skip this if you already state blast radius when proposing a change.

    1. SectionThe failure modes you must expect7 minNaming them unprompted is most of what "senior" means here.
    2. SectionRecovery: retries & checkpoints6 minA non-idempotent retry is a second incident, not a fix.
    3. SectionLineage: cause and blast radius6 minWho else is wrong right now? The first question a lead asks.
    4. SectionIdempotency, recovery & monitoring7 minThe same three properties, argued at cluster scale.
    5. InterviewClassify the failure first6 minSeven classes. Reading an error for its class bounds the search.
    6. InterviewFail the record or fail the job?6 minA declared threshold, not a judgement call at three in the morning.
    7. ExerciseThe source that lies60 minin the studioThe upstream is confidently wrong. Nothing errors. Now what?
    8. ExerciseMoving without a gap60 minin the studioMigrate a live pipeline with nobody noticing. The staff-level move.
    9. QuestionsData Pipeline · Failure & recovery15 minTwenty reruns and replays, each with a real outcome attached.

  3. Signal 03 · ~2h

    You can change it without an outage

    You can ship a breaking change to a table other teams depend on.

    Skip this if expand-migrate-contract is already how you ship schema changes.

    1. SectionShipping a breaking change anyway7 minExpand, migrate, contract. The answer to "without downtime".
    2. SectionThe break that passes every check7 minThe column still parses and the number changed meaning. Nothing catches that.
    3. SectionThe producer-consumer handshake6 minA contract is who you woke up, not a YAML file.
    4. SectionTwo shapes, one model6 minRun both versions until the last consumer moves. Then delete one.
    5. InterviewBatch to streaming, safely6 minParallel run, compare, cut over. Never a big bang.
    6. InterviewThe source deleted a row6 minDeleting the fact makes last month unreproducible. Reverse instead.
    7. ExerciseIs this change safe?45 minin the studioDecide before you ship, not after somebody notices.
    8. ExerciseThe table everyone depends on45 minin the studioForty consumers you did not know about. Change it anyway.
    9. QuestionsData Modeling · Evolution & migration15 minChanging a model without breaking a report is the whole skill.

  4. Signal 04 · ~2h

    You know when not to build it

    You can argue against a design, including one you proposed.

    Nothing to skip — the strongest design answer is often "do not".

    1. SectionThe best pipeline is often no pipeline7 minThe most senior sentence available in a design round. Mean it.
    2. SectionWhat changing your mind actually costs6 minPrice the reversal before you pick. That is what a decision is.
    3. InterviewAirflow, Glue or Step Functions6 minArguing against your own tool is the whole point of the question.
    4. InterviewOne stream or several?6 minFailure isolation is a property of the split, not of the code.
    5. ExerciseWhen you cannot do it all60 minin the studioNot everything ships. Choosing out loud is the exercise.
    6. SectionNarrating a design6 minHaving judgement and showing it are two different exercises.
    7. SectionNarrating a modeling answer6 minThe same skill on the modelling side, where it is scored harder.
    8. QuestionsData Pipeline · Trade-offs & tool choice15 minTwenty decisions with no snippet. Say what would change your mind.
    9. QuestionsData Modeling · Approach & trade-offs15 minKimball or Vault, star or wide, mesh or central. Argue one.

03

Start it now

Reading is free and needs no account. The plan itself tracks on your dashboard.

Start Senior / staff depthTracked, 4 signals, ~10hSee the other plans16 more, by what you are aiming at