Freshworks SSE Interview (Virtual)

Round 1: Data Structures & Algorithms

This round was taken by a lead software engineer and focused mainly on DSA. I was asked about my current work and to provide an example for meta-information of a report, as my team deals with reporting.

• Given a set of intervals representing meeting start and end times, I had to determine if a person could attend all meetings. I was then asked to refactor my code, specifically to decrease the number of lines by moving my custom comparator into lambda expressions.
• I was asked to find the maximum subarray sum.

Round 2: System Design

A week later, I had the System Design round with another lead software engineer. This round included some questions based on my current team and how I spend a day at my company.

• I was asked to explain how Authentication and Authorization work.
• The interviewer asked about the difference between PATCH and PUT HTTP methods.
• We discussed sharding: what it is, its usefulness, the 'celebrity problem,' and how to handle changes in the number of servers.
• I had to differentiate between partitioning and sharding.
• The discussion moved to indexing: how it helps in query optimization and its internal workings.
• I was asked if I review PRs and how I provide review comments to colleagues.
• The interviewer presented a scenario of facing a 504 gateway timeout and asked about potential causes, troubleshooting steps, and solutions.
• I was asked to design a high-level architecture for a WhatsApp-like chat system. I wasn't thoroughly prepared for system design, but I talked about websockets and message queues for notifications. He then asked me to look deeper into how websockets work.

Round 3: Managerial Round

Another week later, I had the managerial round with a staff engineer. We discussed query optimizations, why Redshift was chosen over PostgreSQL in my current team's architecture, and a shard migration tool I had worked on.

• The core question was to design a web scraper given a start URL and a level 'n' until which URLs need to be scraped, returning a list of all scraped URLs. I initially proposed a simple HLD where requests are processed synchronously by hitting URLs, downloading data, and processing recursively. However, the interviewer was not satisfied, pointing out its limitations for very large depths (e.g., billions). I then suggested using a Message Queue with client requests as producers and worker threads as subscribers to process results asynchronously. For responses, I thought of using a storage like Amazon S3, providing pre-signed URLs to the client's email. When asked about status updates without email, I suggested a UI logs page to show 'Queued / IN_PROGRESS / SUCCESS / FAILED' statuses. He then presented a follow-up: what if a worker thread processing 1 million requests crashes after processing 0.5 million? I explained that we could run NFRs to determine a threshold level for crashes, use that threshold to set breakpoints or checkpoints to store partial results, and maintain meta-information about the request. If a failure occurs, we would resume processing from the last successful checkpoint.