Java + Spring Boot + Microservices

Audience: Sanjay Naik — 3.5 yrs Java/Spring Boot @ Jio Platforms Ltd (ex-Karkinos Health). This is the round where your prod depth shows. Most non-FAANG tier-1s (Atlassian, Adobe, Microsoft, healthtech: Verily / Tempus / Oscar) weight this round heavily. Treat it as a “do not lose” round.

1. Round structure (2026)

Slot	Duration	What it looks like
Language + framework Q&A	~30 min	Rapid-fire: JVM internals, concurrency, Spring DI, AOP, `@Transactional`, JPA gotchas. Interviewer probes 2-3 levels deep — “why,” then “what breaks.”
Small service / debug	~25-30 min	Design a slim REST service (rate limiter, URL shortener, notification fan-out, idempotent payment endpoint) or debug a broken Spring snippet (circular dep, lost transaction, N+1, race condition).
Wrap / questions	~5 min	Your turn — ask about on-call, deploy cadence, observability stack.

Where this round matters how much:

Company tier	Weight of Java/Spring round
Amazon, Walmart, Meta	Secondary signal — DSA + LP/system-design dominate. Java round is correctness + idioms, not depth.
Microsoft, Atlassian, Adobe	Primary signal alongside DSA. Expect framework depth.
Healthtech (Verily, Tempus, Oscar, Babylon)	Primary — production Java/Spring experience is what they’re buying. Your CDC/Kafka work is gold here.
Indian tier-1 (Razorpay, Swiggy, Flipkart, Zerodha, PhonePe)	Primary — often the longest round.

Default playbook per question: (1) one-line definition, (2) “how it works under the hood,” (3) trade-off / when not to use it, (4) drop a 1-line example from Jio/Karkinos. That fourth bullet is what separates 3.5-yr SDE-II candidates from 6-yr seniors.

2. Top 30 CORE JAVA questions

#	Question	Difficulty	Topic	Asked at
1	Walk me through the JVM memory model — heap, stack, metaspace, code cache.	Med	JVM	Microsoft, Atlassian, Razorpay
2	Difference between G1, ZGC, and Shenandoah on JDK 21 — when do you pick which?	Hard	GC	Adobe, Stripe, Oscar
3	What is the String pool? `new String("x") == "x".intern()` — true or false?	Easy	Strings	Almost everywhere
4	What makes a class truly immutable? List the 5 rules.	Med	OOP	Atlassian, Tempus
5	The `equals` / `hashCode` contract — what breaks if you override one and not the other?	Easy	Collections	Universal
6	`HashMap` internals — load factor, treeify threshold (8), resize cost.	Med	Collections	Universal
7	`ConcurrentHashMap` vs `Hashtable` vs `Collections.synchronizedMap` — internals + perf.	Med	Concurrency	Microsoft, Flipkart
8	`volatile` vs `synchronized` vs `ReentrantLock` — what guarantees does each give?	Hard	Concurrency	Atlassian, Adobe
9	What is CAS? Walk through `AtomicInteger.incrementAndGet()`.	Hard	Concurrency	Microsoft, Razorpay
10	`ThreadLocal` — when do you use it, and what’s the leak risk in app servers?	Med	Concurrency	Adobe, Verily
11	`ExecutorService` types — fixed vs cached vs scheduled vs single. When does cached blow up?	Med	Concurrency	Walmart, PhonePe
12	`ForkJoinPool` — work-stealing, common pool, when to avoid it.	Hard	Concurrency	Microsoft, Stripe
13	`CompletableFuture` chaining — `thenApply` vs `thenCompose` vs `thenCombine`.	Med	Async	Atlassian, Swiggy
14	Streams — lazy evaluation, terminal vs intermediate, `parallelStream` gotchas.	Med	Streams	Universal
15	`Optional` vs `null` — when to return `Optional` and when not to.	Easy	Idioms	Universal
16	Sealed classes, records, pattern matching for `switch` (JDK 21) — give an example.	Med	JDK 21	Microsoft, Oracle, Adobe
17	Virtual threads (Project Loom, JDK 21) — what problem do they solve over platform threads?	Hard	JDK 21	Atlassian, Stripe, Oscar
18	Reflection cost — why is it slow, and how does method-handle / `LambdaMetafactory` help?	Hard	JVM	Microsoft, Adobe
19	ClassLoader hierarchy — bootstrap → platform → app. Parent-delegation model.	Med	JVM	Microsoft
20	Generics + type erasure — why can’t you do `new T()` or `instanceof T`?	Med	Language	Atlassian
21	Checked vs unchecked exceptions — when do you wrap a checked into a runtime?	Easy	Idioms	Universal
22	`try-with-resources` — what does `AutoCloseable` add vs `Closeable`? Suppressed exceptions?	Med	Idioms	Adobe
23	Java serialization — why do we prefer JSON / protobuf in 2026?	Easy	Idioms	Microsoft, Verily
24	Custom annotation — write `@Retryable` and how would you process it?	Med	Meta	Razorpay
25	SOLID applied in Java — give one real violation you’ve seen and how you fixed it.	Med	Design	Atlassian
26	`==` vs `.equals()` on `Integer` boxed values — `127` vs `128` puzzle.	Easy	Trap	Universal
27	What is a “memory leak” in Java if there’s GC? Give 3 ways to leak.	Med	JVM	Microsoft
28	`finalize()` vs `Cleaner` vs `PhantomReference` — which is preferred in 2026?	Hard	JVM	Adobe
29	Compare `synchronized` block vs `synchronized` method — which monitor is locked?	Easy	Concurrency	Walmart
30	Java module system (JPMS) — `module-info.java`. Why did it never catch on?	Med	JDK 9+	Microsoft, Oracle

Snippets that earn extra points

equals / hashCode (record auto-implements both — show you know):

public record Patient(UUID id, String mrn) {
    // equals & hashCode auto-generated from all components — JDK 16+
}

Virtual threads:

try (var executor = Executors.newVirtualThreadPerTaskExecutor()) {
    IntStream.range(0, 10_000).forEach(i ->
        executor.submit(() -> blockingDbCall(i)));
} // 10k virtual threads, ~few MB heap — not 10k OS threads

CompletableFuture — thenCompose flattens:

CompletableFuture<User> u = fetchUser(id);
CompletableFuture<Order> o = u.thenCompose(this::fetchLastOrder);  // not thenApply

CAS-style retry:

AtomicReference<State> ref = new AtomicReference<>(initial);
State cur, next;
do {
    cur = ref.get();
    next = mutate(cur);
} while (!ref.compareAndSet(cur, next));

3. Top 25 SPRING / SPRING BOOT questions

#	Question	Difficulty	Topic	Asked at
1	Explain IoC + DI like I’m new — then go three layers deeper.	Easy	Core	Universal
2	`BeanFactory` vs `ApplicationContext` — when do you ever touch `BeanFactory`?	Med	Core	Microsoft
3	Bean scopes — singleton, prototype, request, session, application, websocket.	Med	Core	Adobe, Razorpay
4	`@Component` vs `@Bean` vs `@Configuration` — when do you pick which?	Easy	Core	Universal
5	`@Autowired` resolution order — by type, then by `@Qualifier`, then by name.	Med	DI	Universal
6	How do you break a circular dependency? List 4 ways.	Med	DI	Atlassian, Microsoft
7	AOP proxy types — JDK dynamic proxy vs CGLIB — when does Spring pick which?	Hard	AOP	Atlassian, Adobe
8	`@Transactional` propagation levels — REQUIRED vs REQUIRES_NEW vs NESTED.	Hard	Tx	Microsoft, Razorpay, Tempus
9	`@Transactional` isolation — READ_COMMITTED vs REPEATABLE_READ.	Hard	Tx	Adobe, Stripe
10	Rollback rules — why doesn’t `@Transactional` roll back on checked exceptions by default?	Med	Tx	Universal
11	Self-invocation gotcha — why does calling a `@Transactional` method from within the same class not start a transaction?	Hard	Tx, Trap	Atlassian, Microsoft, Razorpay
12	Spring Data JPA N+1 — how do you detect it, how do you fix it?	Hard	JPA	Universal
13	Hibernate L1 vs L2 cache — when have you actually enabled L2?	Med	JPA	Adobe
14	Optimistic (`@Version`) vs pessimistic (`PESSIMISTIC_WRITE`) locking — when which?	Med	JPA	Stripe, Razorpay
15	How does Spring Boot auto-configuration actually work? Walk through `@EnableAutoConfiguration` → `AutoConfiguration.imports`.	Hard	Boot	Microsoft, Atlassian
16	Profiles — how do you load `application-dev-int.yml` vs `application-prod.yml`?	Easy	Boot	Universal
17	Actuator — which endpoints are safe to expose in prod?	Easy	Boot	Adobe, Verily
18	Write a custom starter — what files matter? (`spring.factories` legacy vs `AutoConfiguration.imports` for 3.x).	Hard	Boot	Microsoft
19	Spring Security filter chain — order matters. Where does JWT slot in?	Hard	Security	Atlassian, Razorpay
20	JWT vs server-side session — trade-offs (revocation, size, rotation).	Med	Security	Universal
21	OAuth2 resource server — how does Spring validate the token?	Med	Security	Adobe, Oscar
22	WebFlux vs MVC — Mono/Flux, when to actually use WebFlux?	Med	Reactive	Stripe, Atlassian
23	Bean Validation (Jakarta) — `@Valid` vs `@Validated`, group sequences.	Easy	Validation	Universal
24	Error handling — `@ControllerAdvice` + `@ExceptionHandler` + RFC 7807 problem details.	Med	API	Atlassian, Verily
25	Testing — `@SpringBootTest` (full) vs `@WebMvcTest` (slice) vs `@DataJpaTest` + Testcontainers.	Med	Test	Microsoft, Atlassian

Snippets

Self-invocation trap:

@Service
class OrderService {
    public void outer() {
        inner();  // <-- this call bypasses the proxy. @Transactional NOT applied.
    }
    @Transactional
    public void inner() { /* writes */ }
}

Fix: inject self (@Autowired private OrderService self;) and call self.inner(), or use AopContext.currentProxy(), or split into two beans.

Fixing N+1 with fetch join:

@Query("select p from Patient p join fetch p.treatmentPlans where p.org = :org")
List<Patient> findAllWithPlans(@Param("org") UUID org);

Or @EntityGraph(attributePaths = "treatmentPlans") on the repo method.

Auto-config — peek under the hood:

java -jar app.jar --debug
# prints positive + negative auto-configuration matches

@ControllerAdvice with problem details:

@RestControllerAdvice
class ApiExceptionHandler {
    @ExceptionHandler(EntityNotFoundException.class)
    ProblemDetail notFound(EntityNotFoundException e) {
        return ProblemDetail.forStatusAndDetail(HttpStatus.NOT_FOUND, e.getMessage());
    }
}

Testcontainers slice — what real teams ship:

@SpringBootTest
@Testcontainers
class TreatmentPlanIT {
    @Container static PostgreSQLContainer<?> pg = new PostgreSQLContainer<>("postgres:16");
    @DynamicPropertySource
    static void props(DynamicPropertyRegistry r) {
        r.add("spring.datasource.url", pg::getJdbcUrl);
    }
}

4. Top 20 MICROSERVICES questions

#	Question	Difficulty	Topic	Asked at
1	When not to do microservices — give 3 anti-cases.	Easy	Strategy	Atlassian, Verily
2	Walk me through a real monolith → MS migration you did.	Med	Strategy	Universal
3	Sync (REST/gRPC) vs async (Kafka/queues) — decision criteria.	Med	Comms	Razorpay, Stripe
4	REST vs gRPC vs GraphQL — pick one for service-to-service, one for mobile, one for analytics.	Med	API	Microsoft, Adobe
5	API Gateway pattern — what does it solve? (auth, rate-limit, routing, fan-out). You used Kong/Konga at Jio — lead with that.	Med	Pattern	Atlassian, Razorpay
6	Service discovery — Eureka vs Consul vs k8s-native DNS.	Med	Infra	Microsoft, PhonePe
7	Client-side vs server-side load balancing — Ribbon (deprecated) vs Spring Cloud LoadBalancer.	Med	Infra	Adobe
8	Circuit breaker (Resilience4j) — CLOSED → OPEN → HALF_OPEN. Walk through state transitions + config.	Hard	Resilience	Atlassian, Razorpay, Stripe
9	Bulkhead — semaphore vs thread-pool, what does it isolate?	Med	Resilience	Adobe
10	Retry with exponential backoff + jitter — why jitter? (thundering herd).	Med	Resilience	Microsoft, Stripe
11	Distributed tracing — Sleuth → Micrometer Tracing → Zipkin/Jaeger. How is trace-id propagated?	Hard	Observability	Atlassian, Adobe
12	Centralized logging — ELK / Loki, structured (JSON) logs, correlation-id pattern.	Med	Observability	Verily, Oscar
13	Saga pattern — choreography (event-driven) vs orchestration (central coordinator). Failure & compensation.	Hard	Distributed Tx	Razorpay, Stripe, Atlassian
14	Two-phase commit — why is it avoided in MS? (blocking coordinator, partition intolerance).	Med	Distributed Tx	Microsoft
15	Outbox pattern + CDC (Debezium) — what problem does it solve? Dual-write hazard. You shipped this at Jio — own this question.	Hard	Pattern	Atlassian, Stripe, Tempus
16	Event sourcing — when is it worth it? When is it overkill?	Hard	Pattern	Adobe, Razorpay
17	CQRS — separating read/write models. Eventual consistency window.	Med	Pattern	Microsoft, Stripe
18	Idempotency keys — how do you implement a `POST /payments` that is safe to retry?	Hard	Pattern	Stripe, Razorpay, PhonePe
19	Health checks — liveness vs readiness vs startup probes. Why the distinction matters in k8s.	Med	Ops	Microsoft, Atlassian
20	Blue-green vs canary vs rolling — and how do you DB-migrate without downtime?	Hard	Ops	Adobe, Stripe

Snippets

Resilience4j circuit breaker:

@CircuitBreaker(name = "mdmAdmin", fallbackMethod = "fallbackPatient")
@Retry(name = "mdmAdmin")
@Bulkhead(name = "mdmAdmin", type = Bulkhead.Type.SEMAPHORE)
public Patient fetch(UUID id) { return mdmClient.getPatient(id); }

private Patient fallbackPatient(UUID id, Throwable t) {
    return cache.getOrDefault(id, Patient.empty());
}

resilience4j.circuitbreaker.instances.mdmAdmin:
  slidingWindowSize: 20
  failureRateThreshold: 50
  waitDurationInOpenState: 10s
  permittedNumberOfCallsInHalfOpenState: 5

Idempotent endpoint pattern:

@PostMapping("/payments")
public ResponseEntity<Payment> create(
        @RequestHeader("Idempotency-Key") String key,
        @RequestBody PaymentRequest req) {
    return idempotencyStore.findResponse(key)
        .map(ResponseEntity::ok)
        .orElseGet(() -> {
            Payment p = service.charge(req);
            idempotencyStore.save(key, p, Duration.ofHours(24));
            return ResponseEntity.ok(p);
        });
}

Outbox + Debezium (the headline diagram):

[App tx] ──INSERT──► [orders + outbox] (same Postgres tx)
                           │
                      [Debezium WAL reader]
                           │
                           ▼
                        [Kafka topic]
                           │
                           ▼
                  [consumer services]

The dual-write hazard (write DB + publish Kafka, one of them fails) is eliminated because both writes are in one Postgres transaction, and Debezium replays the WAL.

5. Top 15 KAFKA & DATA-PIPELINE questions

#	Question	Difficulty	Topic	Asked at
1	Kafka vs RabbitMQ vs SQS — log vs queue semantics, replay, ordering.	Med	Brokers	Atlassian, Stripe
2	Partitions and ordering — what guarantees ordering within a partition?	Med	Kafka	Razorpay, PhonePe
3	Consumer groups — rebalance triggers, sticky assignor, cooperative rebalance.	Hard	Kafka	Atlassian, Adobe
4	Exactly-once semantics — idempotent producer + transactional producer + `isolation.level=read_committed`.	Hard	Kafka	Stripe, Microsoft
5	Idempotent producer — what does `enable.idempotence=true` actually do? (PID + seq).	Hard	Kafka	Stripe
6	Transactional producer — `initTransactions`, `beginTransaction`, send, `commitTransaction`.	Hard	Kafka	Atlassian
7	Log compaction vs retention — when to use which? Compaction’s “tombstone” message.	Med	Kafka	Adobe, Razorpay
8	Kafka Streams vs KSQL vs Flink — when to pick which?	Med	Streaming	Stripe
9	Debezium CDC architecture — connector → WAL → topic per table → SMTs. Your sub-3-sec lag story goes here.	Hard	CDC	Atlassian, Tempus, Stripe
10	Schema Registry — Avro / Protobuf, BACKWARD vs FORWARD vs FULL compatibility.	Med	Schema	Adobe, Verily
11	Replay strategy — how do you replay last 24 hours into a new consumer without double-processing?	Hard	Ops	Stripe, Atlassian
12	DLQ pattern — when does a message go to DLQ vs retry topic? Tiered retries.	Med	Pattern	Razorpay
13	Backpressure — how does a slow consumer affect producers?	Med	Kafka	Microsoft
14	`acks=0` vs `acks=1` vs `acks=all` — durability vs latency.	Easy	Kafka	Universal
15	Why is consumer offset committed to a Kafka topic (`__consumer_offsets`) and not Zookeeper?	Easy	Kafka	Adobe

Kafka producer config that earns nods:

enable.idempotence=true
acks=all
max.in.flight.requests.per.connection=5
retries=Integer.MAX_VALUE
compression.type=zstd
linger.ms=10
batch.size=65536

6. Hibernate & PostgreSQL “make-or-break”

#	Question	Why it matters
1	What is the N+1 query problem? Show me the SQL it produces.	Most common JPA bug — interviewers love it.
2	`LazyInitializationException` — root cause and 3 fixes.	Open-session-in-view debate.
3	Fetch join (`join fetch`) vs `@EntityGraph` — when does which work?	EntityGraph is preferred for repository methods.
4	`hibernate.jdbc.batch_size` + `order_inserts` + `order_updates` — why all three?	Real perf knob in prod.
5	L1 (per-session) vs L2 (per-`SessionFactory`) cache — when have you enabled L2 in prod?	Most people haven’t — be honest.
6	`save` vs `persist` vs `merge` vs `saveAndFlush` — semantic differences.	Trap question.
7	Dirty checking — how does Hibernate know an entity changed? (snapshot at load).	Trips juniors.
8	`EXPLAIN ANALYZE` walkthrough — seq scan vs index scan vs bitmap heap scan.	Postgres depth.
9	Index types — B-tree, Hash, GIN (full-text, jsonb), BRIN (time-series).	Healthtech & analytics.
10	Partitioning — RANGE (by date) vs LIST vs HASH. When does the planner do partition pruning?	Tempus, Verily — large datasets.
11	VACUUM vs autovacuum vs VACUUM FULL — bloat, freeze, wraparound.	Stripe, Razorpay love this.
12	MVCC — how does Postgres handle two concurrent updates to the same row?	Atlassian, Microsoft.
13	Isolation levels — `READ COMMITTED` (default) vs `REPEATABLE READ` vs `SERIALIZABLE`. Phantom reads in Postgres `REPEATABLE READ`? Trick: Postgres `REPEATABLE READ` already prevents phantoms via snapshot isolation.	Trick question — most candidates miss it.
14	HikariCP — what’s `maximumPoolSize` set to, and how do you size it? (`connections = ((core_count * 2) + effective_spindle_count)`).	Stripe, PhonePe.
15	Read replicas — how do you route reads in Spring? (`AbstractRoutingDataSource` or `@Transactional(readOnly = true)` + transparent routing).	Adobe, Razorpay.

Spot-the-N+1:

List<Patient> patients = patientRepo.findAll();
for (Patient p : patients) {
    p.getTreatmentPlans().size();   // fires 1 query per patient — N+1
}

Fix with @EntityGraph(attributePaths = "treatmentPlans") on findAll.

7. “Tricky” / trap questions

These are the gotcha ones — interviewers ask them because the right answer requires knowing Spring’s proxy internals. Burn these in.

#	Question	Answer in one line
1	Why does autowiring a prototype bean into a singleton break?	Singleton is constructed once → it holds the same prototype instance forever. Fix: `ObjectProvider<T>`, `Provider<T>` (JSR-330), or `@Scope(proxyMode = TARGET_CLASS)`.
2	What happens when `equals` is overridden but `hashCode` is not?	`HashMap`/`HashSet` lose the object — `equals`-equal objects land in different buckets.
3	Why does `@Async` on a `@Transactional` method “lose” the transaction?	`@Async` runs on a new thread → new thread has no `ThreadLocal` `TransactionSynchronizationManager` context → no transaction propagates.
4	Why does `@Transactional` on a private method do nothing?	Spring uses proxies (JDK dynamic or CGLIB). Proxies can only intercept `public` (CGLIB also intercepts `protected`/package). Private methods bypass the proxy.
5	Why does `@Transactional` not work on a method called from inside the same bean (self-invocation)?	Same reason — call is on `this`, not on the proxy.
6	`HashMap` in concurrent code — what’s the worst that happens?	Pre-JDK 8: infinite loop on resize. JDK 8+: still data corruption / lost writes. Use `ConcurrentHashMap`.
7	`Integer` cache — why is `Integer.valueOf(127) == Integer.valueOf(127)` true but `Integer.valueOf(128) == Integer.valueOf(128)` false?	Integer cache: -128 to 127 are cached as the same instance.
8	`String s = "a" + "b"` vs `String s = a + b` (where a, b are vars) — what’s different at bytecode?	First is a compile-time constant in the String pool. Second is a `StringBuilder.append` at runtime.
9	Why is overriding `finalize()` an anti-pattern in 2026?	Unpredictable timing, can resurrect objects, slow GC. Use `Cleaner` (JDK 9+) or `try-with-resources`.
10	Why can’t you `throw new T()` in a generic method?	Type erasure — `T` is `Object` at runtime, JVM can’t verify it’s a `Throwable`.
11	Why does `Arrays.asList(1, 2, 3).add(4)` throw `UnsupportedOperationException`?	Returns a fixed-size wrapper over the array, not an `ArrayList`.
12	What happens if a `@Transactional` method calls a `REQUIRES_NEW` method on the same bean?	Same self-invocation trap — `REQUIRES_NEW` is ignored, you’re still in the outer tx.
13	Will a circuit breaker `OPEN` ever close on its own?	Yes — after `waitDurationInOpenState`, it goes `HALF_OPEN`, lets N probe calls through. If they succeed, `CLOSED`.
14	`@PostConstruct` vs `InitializingBean.afterPropertiesSet()` vs `@Bean(initMethod=...)` — order?	`@PostConstruct` → `afterPropertiesSet` → `initMethod`.
15	Two `@Configuration` classes both define `@Bean DataSource` — what happens?	`NoUniqueBeanDefinitionException` unless one is `@Primary` or you use `@Qualifier`.

8. Using Sanjay’s experience as concrete examples

Drop these as 1-2 sentence “I’ve shipped this” lines. They turn a generic answer into a senior answer.

Topic	Sanjay’s drop-in line
CDC / Outbox pattern	“At Jio, we replaced a 30-min nightly batch with a Debezium → Kafka → Postgres pipeline. Sub-3-second end-to-end lag from source DB write to downstream consumer. The outbox was in the same Postgres tx as the business write, so we eliminated dual-write hazards entirely.”
API Gateway	“We fronted ~12 microservices with Kong + Konga at Jio. JWT validation, per-consumer rate limits, and request transformation lived in Kong — services stayed thin.”
Circuit breaker	“We wrapped every cross-service call in Resilience4j. Failure threshold 50% over a 20-call sliding window, 10s open state. When MDM Admin started returning 5xx during a deploy, the Treatment Plan service degraded gracefully instead of cascading.”
N+1 / EntityGraph	“Our Treatment Plan list endpoint went from ~800 ms (N+1) to ~80 ms after switching to `@EntityGraph(attributePaths = …)` on the repo method.”
Kafka exactly-once	“For the Treatment Plan event stream we used `enable.idempotence=true` + transactional producer + `read_committed` consumer. Replays were safe because the consumer was already idempotent on `event_id`.”
Saga / orchestration	“Patient onboarding spanned 4 services (MDM, Treatment Plan, Notification, Billing). We did orchestration via a state machine in the MDM Admin service — easier to debug than choreography for a non-trivial flow.”
Distributed tracing	“We instrumented every service with Micrometer Tracing → Jaeger. The trace-id was propagated via the gateway and stamped into structured JSON logs, so an SRE could pivot from a Kibana log line to the full request trace in one click.”
Spring Profiles	“Our config was Spring Cloud Config + profile-per-env (`dev-int`, `qa`, `prod`). Treatment Plan ran on `8901/treatment-plan`; MDM Admin on `8085/mdm-admin`. Liquibase migrations were profile-gated.”
Hibernate L1 / dirty checking	“We hit a `dirty-checking` regression — an entity was being modified inside a read-only tx by mistake, causing autoflush. Switching to a DTO projection killed the surprise updates.”
PostgreSQL MVCC + isolation	“We saw a `serialization_failure` under load on a `SERIALIZABLE` tx. Dropped to `READ COMMITTED` with explicit `SELECT … FOR UPDATE` on the contended row — throughput went up 3x.”
Idempotency keys	“Every external-facing write at Karkinos required an `Idempotency-Key` header. We stored `(key → response)` in Redis with a 24h TTL. Retries returned the cached response, no DB hit.”
HikariCP	“We sized HikariCP to `(cores * 2) + 1` for the CPU-bound services and tracked `hikari.connections.pending` in Prometheus. Caught a leak via the metric before it hit prod.”
Spring Security + JWT	“JWT validated at the gateway, with a downstream `JwtAuthenticationConverter` in each MS so `@PreAuthorize` worked against extracted scopes.”
Liquibase	“Every schema change was a Liquibase changeset, applied at boot. Rollback scripts were mandatory in PR review.”
Docker / k8s	“Both MDM Admin (`8085`) and Treatment Plan (`8901`) ran as separate compose services in dev-int, then as separate k8s Deployments with readiness probes hitting `/actuator/health/readiness`.”
Virtual threads	“Haven’t shipped Loom to prod yet, but I prototyped it for an I/O-bound enrichment service — `Executors.newVirtualThreadPerTaskExecutor()` let us drop from a 200-thread pool to virtual threads and the heap stayed flat.”

Tip: Don’t volunteer every story. Let the interviewer ask “have you done this?” — then drop the 1-liner. Brevity reads as senior.

9. Sources

Appendix — 10-minute pre-interview warm-up

Run this checklist 30 minutes before the round:

Open IntelliJ / a Spring Boot 3.x scratch project. Don’t go in cold.
Re-read the trap section (section 7) — those are the cheapest points to lose.
Re-read your own CLAUDE.md for the Treatment Plan + MDM Admin port + JVM args. The “what does your prod config look like” question is free if you can rattle it off.
Have a 60-second version of the Debezium CDC story rehearsed. Specifics: source DB, target, lag, what it replaced.
Have a 60-second version of one failure story — a prod incident, what you debugged, what you fixed. Senior-ness is calibrated on failure stories more than success ones.

Java + Spring Boot + Microservices — Top Interview Questions (2026)

1. Round structure (2026)

2. Top 30 CORE JAVA questions

Snippets that earn extra points

3. Top 25 SPRING / SPRING BOOT questions

Snippets

4. Top 20 MICROSERVICES questions

Snippets

5. Top 15 KAFKA & DATA-PIPELINE questions

6. Hibernate & PostgreSQL “make-or-break”

7. “Tricky” / trap questions

8. Using Sanjay’s experience as concrete examples

9. Sources

Appendix — 10-minute pre-interview warm-up