Question 1

How would you design a thread-safe, high-performance in-process cache in Java?

Accepted Answer

Use ConcurrentHashMap with computeIfAbsent for atomic get-or-load; or use Caffeine library for LRU/W-TinyLFU eviction, size limits, TTL expiry, and async loading. Caffeine is the state-of-the-art Java cache: W-TinyLFU policy, higher hit rate than LRU, async refresh, stats. ConcurrentHashMap is sufficient for simple cases. In-process cache is 100–1000× faster than Redis for hot data — use both layers (L1 local + L2 Redis).

Question 2

What are the trade-offs between microservices and a monolith for a Java system?

Accepted Answer

Microservices give independent deployment, scaling per service, and technology diversity — but add operational complexity, network latency, distributed transactions, and observability challenges; monolith is simpler but harder to scale independently. Start with a monolith, extract services when you feel specific pain (a team stepping on each other, one component needing 10× the resources). Premature decomposition creates distributed monoliths — the worst of both worlds. Martin Fowler: 'MonolithFirst'.

Question 3

How would you handle distributed transactions across Java microservices?

Accepted Answer

Use the Saga pattern: each service performs a local transaction and publishes an event; compensating transactions roll back on failure — avoiding the performance and availability problems of 2PC. 2PC requires a coordinator and locks resources across services — slow, single point of failure. Saga (choreography or orchestration) uses local transactions + compensating actions. E.g., OrderSaga: create order → reserve inventory → charge payment; if payment fails, release inventory.

Question 4

What is reactive programming in Java (e.g., Project Reactor / Spring WebFlux)?

Accepted Answer

A non-blocking, event-driven programming model using backpressure-aware streams — ideal for high-throughput I/O-bound systems where threads should never block waiting for I/O. WebFlux uses Reactor: Mono for 0-1 results, Flux for 0-N. A small thread pool handles thousands of concurrent requests because threads never block. Backpressure prevents fast producers from overwhelming slow consumers. Trade-off: harder to debug, different mental model.

Question 5

How would you diagnose and fix a memory leak in a production Java application?

Accepted Answer

Take a heap dump (jcmd <pid> GC.heap_dump or jmap), analyze with Eclipse MAT or JProfiler, correlate with GC logs showing Old Gen growth, and use APM tools (Datadog, Dynatrace) for continuous monitoring. Process: 1) Confirm leak via GC logs (growing Old Gen after Full GC). 2) Take heap dump at different times. 3) Compare in MAT — find growing object types. Common causes: static collections holding references, unclosed resources, listener registration without deregistration, ThreadLocal misuse.

Java — Tech Lead Interview Questions

More Tech Lead Topics

Java at Other Levels