← Back to Directory

Databricks interview notes, page 9 of 22

Large Enterprises

Most of these Databricks write-ups come from Software Engineers (192), then Machine Learning Engineers (9) and Data Scientists (8). The most reported rounds are the technical phone screen (136), the system design round (57), the behavioral / leadership round (9) and the onsite coding round (6).

216 interview notes · updated Jul 2026

Databricks·Software Engineer·Technical Phone Screen

May 2026
Databricks SWE interview with a file encryption optimization problem that looks straightforward until you actually think about the cost model. The batch API wrapping subdirectories changes everything.
  • You have a root directory with files and nested subdirectories. Some files are already encrypted, some aren't. You have two APIs: one encrypts a single file with a fixed request overhead plus a per-file cost, and another batch-encrypts all unencrypted files in a given directory and all its subdirectories, also with a fixed overhead but only one per-file cost multiplied by the count. Find the minimum total time to encrypt all unencrypted files.

“The part that tripped me up was figuring out when it's actually cheaper to call the directory API versus just looping individual files.”

View Post

Databricks·Software Engineer·Technical Phone Screen

May 2026
Databricks software engineer interview with a tricky tree problem that looks straightforward until you realize the naive approach blows up exponentially. The key insight they were testing was whether you'd recognize the memoization angle before writing anything.
  • Given a recursively defined 'Fibonacci tree' where F(0) and F(1) are leaves and F(n) has F(n-1) and F(n-2) as children, write a solution that operates in O(n) time relative to the tree's height n, not the total node count.

“I started thinking about traversal and immediately went down the wrong path mentally.”

View Post

Databricks·Software Engineer·Technical Phone Screen

May 2026
Databricks SWE interview that went deep on compression internals. The coding portion was more systems-flavored than I expected, less leetcode and more 'show me you understand how data actually moves around'.
  • Implement both Run-Length Encoding and Bit-Packing schemes, each with encode and decode functions.
  • Design a framing or header format so a decoder can determine which compression scheme was used for each segment of data.
  • When would you prefer RLE over Bit-Packing, and how do you decide between them at runtime?
  • Walk through the edge cases: single-value inputs, values at the boundary of their type, and negative numbers in a signed integer scheme.

“I started with RLE because it felt easier and I needed to build confidence.”

View Post

Databricks·Software Engineer·Onsite - System Design / Architecture

May 2026
Databricks systems design round, one big open-ended question about safe concurrent file writes. The depth they expected was pretty serious, covering everything from fsync semantics to back-pressure strategies.
  • Design a routine that safely writes file content to disk under concurrent access. Cover correctness, atomicity, thread and process safety, error handling, crash recovery, and back-pressure when multiple writers target the same file. Also discuss trade-offs between buffered and direct I/O and what durability guarantees you can realistically offer.

“This question is basically a whole interview by itself.”

View Post

Databricks·Software Engineer·Onsite - System Design / Architecture

May 2026
Databricks system design round for a software engineering role. The whole session was basically one deep question about concurrent log writing, and they really wanted you to go end-to-end on it, not just sketch a queue and call it done.
  • Design a thread-safe log writer that supports many concurrent producer threads appending entries while one or more writer threads flush to durable storage. Cover the producer API, buffering strategy, flush triggers, ordering and timestamping guarantees, graceful shutdown, and crash safety.

“This one ran the full session.”

View Post

Databricks·Software Engineer·Technical Phone Screen

May 2026
Databricks software engineering interview with a networking/algorithms problem that pushed well past the basic solution. The follow-up discussion on data structures was where things got interesting and, honestly, a bit uncomfortable.
  • Given an IP address and a list of CIDR blocks, return the first CIDR in the list that contains the given IP. How would you implement this efficiently, beyond just scanning the list linearly?

“I got the basic prefix-matching logic down fine: mask off the host bits, compare to the network address, done.”

View Post

Databricks·Software Engineer·Onsite - System Design / Architecture

May 2026
Databricks system design round, one question the whole time: design Slack. Not a lot of other context to share but it was a full session on that alone.
  • Design a messaging platform like Slack, covering architecture, real-time communication, and scalability.

“This is one of those questions where you think you know where to start and then realize halfway through that you've been going too broad.”

View Post

Databricks·Software Engineer·Onsite - System Design / Architecture

May 2026
Databricks system design round, and they went deep on a problem that looks like a simple logging exercise until you're 20 minutes in and suddenly talking about group commit semantics and WAL internals. Tough but fair.
  • Design an event logger that accepts writes from multiple producers and guarantees durability before each write call returns.
  • How would you batch incoming events to reduce fsync overhead without making write latency unbounded?
  • When exactly does a writer get acknowledged in a group commit scenario? Walk through the coordination between the thread that triggers the fsync and the other threads waiting in the batch.
  • How do you handle concurrent producers safely? What are the tradeoffs between using a single writer thread with a shared queue versus a lock-free approach?
  • After a crash, how do you recover the log to a consistent state? What do you do about partial writes at the tail?

“The base ask seems manageable: append-only log, fsync before ack.”

View Post

Databricks·Software Engineer·Onsite - Behavioral / Leadership

May 2026
Behavioral round at Databricks for a software engineering role. Two questions, both pretty standard in format but the feedback one caught me more off guard than I expected.
  • Walk me through one of your projects in depth, including your specific role, the key decisions you made, trade-offs you navigated, and the impact it had.
  • Tell me about a time you received critical feedback. What was the feedback, how did you react to it initially, what actions did you take, and how has your behavior changed since then?

“I picked a project I knew cold, which was probably the right call, but I spent too long on context and rushed the trade-offs section at the end.”

View Post

Databricks·Software Engineer·Onsite - Coding / Algorithms

May 2026
Databricks onsite coding round with what seemed like a fairly novel string pattern matching problem. The follow-up added real complexity and I wasn't fully prepared for it.
  • Given two strings, represent one string using index ranges from the other. For example, with a='abcdbcd' and b='sabcd', return the index pairs from a that reconstruct b, like [[1,3],[4,4],[2,4]].
  • Follow-up: if you delete exactly one character from the first string, how would you update the solution to minimize the number of partitions used to represent the second string?

“This felt like a fresh problem, not something you'd find recycled on a prep site.”

View Post