Databricks Interview Questions — Page 19

Databricks·Machine Learning Engineer·Onsite - System Design / Architecture

Jun 2026

Databricks ML engineer round that was basically a system design session disguised as a coding question. The core problem sounds deceptively simple but the follow-ups kept piling on until I was rethinking my entire approach from scratch.

Design an in-memory key-value store that also tracks query-per-second load over a 5-minute sliding window, with APIs for put, get, and get_qps.
How would you optimize memory usage so you're not storing a record for every single operation?
How would you extend the tracking window from 5 minutes to 24 hours without blowing up memory?
If multiple operations share the exact same timestamp, how do you count them correctly and what are the concurrency implications?
How would you support per-key QPS tracking and identify the top-K hottest keys over the sliding window?

“I started with a naive list of timestamped ops and immediately the interviewer asked what happens after a few hours of traffic.”

View Post

Databricks·Software Engineer·Technical Phone Screen

Jun 2026

Databricks coding round, one problem the whole time. It was a string compression thing with DP and honestly more involved than it sounds at first glance.

Implement a string compressor using run-length encoding, but use dynamic programming over the string's prefixes to find the optimal segmentation that minimizes the total encoded length. Account for encoding overhead (e.g. a count of 1 might be written as just the character, not 'c1'). Return either the minimum encoded length or the actual compressed string.

“I knew RLE from the classic interview version and jumped straight into a greedy pass.”

View Post

Databricks·Software Engineer·Technical Phone Screen

Jun 2026

Databricks coding round, one question the whole time. It was a networking/bit-manipulation problem that looks deceptively clean on paper but has enough edge cases to make you sweat if you haven't seen it before.

Given a starting IP address and a positive integer n, return the smallest list of CIDR blocks that exactly covers n consecutive IP addresses beginning at that IP. Each block must be aligned to its size, which must be a power of 2.

“The core loop isn't too bad once you see it: figure out the largest aligned power-of-2 block you can emit without overshooting n, write out the CIDR, subtract from n, move the pointer forward, repeat.”

View Post

Databricks·Software Engineer·Onsite - Behavioral / Leadership

Jun 2026

Databricks internship debrief round, basically a behavioral interview stitched together with a pretty thorough walkthrough of your most recent internship. They want specifics, not a highlight reel.

Walk me through your most recent internship: your role, how the team was structured, and what you personally owned.
What specific technical decisions did you make during the internship, and what trade-offs were involved?
Why was the architecture designed the way it was, and what would you change about it now?
Tell me about a problem you ran into during the internship and how you resolved it.

“I started with a decent overview but got too high-level too fast.”

View Post

Databricks·Data Scientist·Technical Phone Screen

Jun 2026

Databricks data scientist interview with a SQL/Python take-home style problem. Pretty computational, no behavioral stuff from what I could tell. The question itself is well-designed but there are a few edge cases that'll bite you if you're not careful.

Given a transaction table with customer IDs, dates, and spend amounts, compute a weekly time series showing how many distinct customers have reached at least $1000 in year-to-date cumulative spend as of each week. Handle multi-year data by resetting the YTD counter on January 1st each year. Output year, week start date, and the customer count, ordered chronologically.

“The core join/window logic isn't too bad once you see it, but I fumbled the multi-year reset part initially.”

View Post

Databricks·Software Engineer·Technical Phone Screen

Jun 2026

Databricks technical phone screen, one meaty graph problem that took up most of the time. The multi-modal twist is what makes it interesting, otherwise it's just shortest path.

Given a 2D grid with a start, a destination, open cells, and blocked cells, find the minimum-cost path where you can switch between multiple transportation modes (each with different step costs and movement constraints), and switching modes at a cell also carries a cost.

“The key insight I kept circling around was that you can't just track position, you have to track (position, current mode) as your state.”

View Post

Databricks·Software Engineer·Onsite - System Design / Architecture

Jun 2026

Databricks system design round focused entirely on designing a file system from scratch. No shortcuts allowed, they explicitly said you can't just say 'use S3 as the blob store' and call it a day. It went deep fast.

Design a file system from scratch, covering the on-disk layout including inodes, directory entries, and free-space management. You cannot use an existing blob or object store as a building block.
How would you support fast directory listing with pagination when a directory contains a very large number of entries?
How do you implement recursive directory deletion that is atomic, handles very large subtrees efficiently, and stays safe under concurrent access?
How do you ensure crash consistency in your file system design? Walk through journaling, copy-on-write, and fsync semantics.
How would you scale this file system design to a distributed setting, covering the metadata service, data servers, replication, and garbage collection?

“The constraint about not using an object store tripped me up early.”

View Post

Databricks·Software Engineer·Technical Phone Screen

Jun 2026

Databricks SWE interview that centered on designing a lazy array data structure from scratch. The real emphasis wasn't the implementation itself but writing genuinely thorough tests for it, which I did not fully appreciate until I was already in the weeds.

Design and implement a lazy array data structure where values at each index are computed on demand from a generator or deferred source, then cached for future accesses. Define the API collaboratively with the interviewer.
Write thorough test cases for your lazy array. Cover empty arrays, repeated access to verify memoization, out-of-bounds behavior, infinite or lazy sources using a take(n) pattern, chained operations like map followed by filter, and confirm the underlying generator is never called more times than necessary.

“I spent too long bikeshedding the API before writing a single line.”

View Post

Databricks·Software Engineer·Onsite - Behavioral / Leadership

Jun 2026

Behavioral round for a software engineering role at Databricks. Three questions, all pretty standard retrospective stuff, but the wording was specific enough that vague answers wouldn't cut it.

Describe the hardest challenge you've faced, either technical or interpersonal, and walk through how you dealt with it.
Tell me about your biggest professional mistake, what the impact was, and what you took away from it.
Describe a time you received feedback you disagreed with. How did you push back, and what ended up happening?

“I went technical because I panicked a bit and figured it was safer ground.”

View Post

Databricks·Software Engineer·Technical Phone Screen

Jun 2026

Databricks software engineer coding round. The problem itself was manageable but the interview dynamic felt off, like the interviewer was more interested in steering me toward their preferred solution than actually evaluating my thinking. Came out of it pretty frustrated.

Implement a snapshot set iterator where the return order of elements does not matter.

“I came in with a design using a shared log list plus a dict and explained the tradeoff.”

View Post