I'm currently in Google's L4 infrastructure group, with 2 years of experience. In the phone interview, I got to speak with Holden Karau, a renowned expert in Spark. He's really nice, and we talked a lot about Netflix and career-related topics. We only spent about ten minutes on a simple question.

HR scheduled the virtual onsite interview for a week later, split over two days. The first day had three technical rounds: 1. Concurrency. 2. Problem solving (practical version of LeetCode questions). 3. System design.

The concurrency round was relatively easy, focusing on basic operations of mutexes/condition variables and their pros & cons. Good communication was key.

The problem-solving round involved a variant of the LRU cache, which was not too difficult.

The system design round was challenging. I had prepared a lot by studying existing system designs and had read through "Designing Data-Intensive Applications" (DDIA), but the interviewer started with a very simple counter concept and dug deep into details (like various open-source tech stack usages, or specific partition/replication algorithms). I think I failed in this round. The question is about designing a counter (think of view count/metrics). Start from single server and bare metal, then scale up and move to cloud. Conversation driven; at least in my case the interviewer asked a lot of questions on every single assumption and components proposed; really deep drill down.

On the second day, I received an email saying there was no need to continue with the interviews.

Overall, the interviewers were very professional and highly skilled. Working in a large company for a long time makes one reliant on existing platforms. It's time to learn more.