Databricks Interview Questions

Given firewall rules as CIDR blocks with allow or deny actions, return the verdict for an IP: the first rule it falls within wins. Reports agree on the follow-up — the query becomes a CIDR block itself, allowed only if every address in it is covered by allow rules — and on the ladder after it: write your own test cases, pseudocode the CIDR-versus-CIDR overlap check, then scale to firewall-log volumes. One candidate's recruiter debrief named the weak spot exactly: building the bit mask for the prefix length. This is the free preview — full statement, follow-ups, and a solution run against its test cases.

You are the middleman: a customer submits an ISBN, a bid price, and a payment method; your system fans out to hundreds of partner bookstores, buys automatically if the lowest price beats the bid, and otherwise returns the best price found. The interviewer hints are consistent across reports — the downstream calls are async and the point is not overloading the sellers, you design the seller API yourself, and when you reach for a lock the interviewer asks whether you could just return a payment failure instead. The recorded failures are the sold-out-between-quote-and-hold race, and spending the hour on clarification without ever reaching a schema.

Implement a logger shared by many threads where a write must not return until the record is on disk, then raise throughput without giving up that guarantee — batching writers behind one fsync while each caller still blocks for its own record. Reports say the interviewer has a specific implementation in mind, so clarify preferences before coding. One debrief was blunt: do not reach for thread-safe containers, show that you know when to lock and unlock; another candidate was graded on whether busy-waiting had actually been eliminated. Know your condition variables.