Spring Boot Interview Questions — Medium

• Instantiation → populate properties → Aware interfaces (BeanNameAware...) → BeanPostProcessor.before → @PostConstruct/InitializingBean.afterPropertiesSet/initMethod → ready → @PreDestroy on shutdown.
• BeanPostProcessor power: proxies (transactions, async, caching) wrap here — explains annotation self-invocation failures. Ordering knowledge separates debugging heroes from sufferers.

---

• JDK dynamic proxies (interfaces) or CGLIB subclassing weave advice around beans: @Transactional, @Cacheable, @Async, custom aspects.
• Limits: self-invocation bypasses proxy (internal call misses transaction); final methods/classes defeat CGLIB; private methods never advised. Fixes: split beans, self-injection via ObjectProvider, or AspectJ weaving for exotic cases.

---

Key ones:

• REQUIRED (default): join or create.
• REQUIRES_NEW: suspend outer, fresh tx — audit log must survive outer rollback.
• NESTED: savepoint inside outer — partial rollback possible.
• NOT_SUPPORTED/MANDATORY/NEVER: suspension/assertions for special flows. Scenario drill: batch processing where per-row failure shouldn't sink whole batch → REQUIRES_NEW per row + collect errors.

---

• DEFAULT→DB; READ_UNCOMMITTED (dirty reads), READ_COMMITTED (default PG), REPEATABLE_READ (MySQL default), SERIALIZABLE.
• Anomalies ladder: dirty → non-repeatable → phantom; serialization prevents all incl. write-skew (PG SSI). Practical guidance: raise level only for the critical section; pair with retries on serialization failures.

---

• One EntityManager per transaction via shared proxy; entities loaded become managed — changes auto-flushed at commit (dirty checking).
• Read-only hint (@Transactional(readOnly=true)) enables flush-mode manual/Hibernate optimizations + replica routing hints.
• Detached entity pitfalls: modifying after context close silently lost; merging copies state into new managed instance.

---