Key Big Data Interview Questions and What Interviewers Really Expect

The article is a brief piece based on a real interview, showing that seemingly simple questions can be difficult to answer and that a candidate's response directly influences interview evaluation.

1. What is the QPS of the online real‑time job? A naive answer might be a round number such as 10,000 or 100,000. Interviewers actually want to gauge the candidate's understanding of business scale, data volume, ingestion methods (e.g., CDC), core design, state‑machine complexity, and how QPS would multiply with larger workloads, as well as issues like hot keys and dimension‑table hit rates.

When the data scale is massive, high‑QPS real‑time computation brings challenges such as hot‑key problems and hot‑spot tables; discussing these topics demonstrates deeper insight and exceeds interview expectations.

2. What causes taskmanager full GC and how can it be optimized? A superficial answer would be “increase memory.” Interviewers aim to discover whether the candidate has encountered such issues in production, understands the framework’s memory model, can identify frequent full GC, locate hotspot code, and apply tools like monitoring dashboards and flame‑graphs. Typical root causes include large objects in UDFs, unreasonable heap allocation, or excessive caching of massive data.

3. What are the usage scenarios of HBase? A simplistic reply might be “high‑concurrency read/write for big data.” The interviewer expects a justification for choosing HBase, highlighting its unique advantages for the specific business context, and a discussion of benchmark criteria such as query type (point vs. aggregation), QPS, latency, supported data size, stability, SLA, disaster‑recovery, and future migration options.

These common interview questions test not only knowledge of specific technologies but also the candidate’s practical experience, analytical thinking, and ability to articulate design decisions; neglecting the deeper aspects often leads to poor interview outcomes.

For further learning, the article points to a large‑scale big‑data interview community and a collection of related resources.