xiaofucode.com – a programmer learning guide driven by the pursuit of profit
Spring’s bean factory uses a three‑level cache—singletonObjects, earlySingletonObjects, and singletonFactories—to handle circular dependencies, but only the third level is needed when AOP proxies are involved, allowing lazy proxy creation without breaking the standard bean lifecycle.
The article explains why Java's ThreadPoolExecutor places tasks into a queue before creating non‑core threads, discusses the cost of thread creation and CPU contention, illustrates the design with a bank‑counter analogy, and shows how Tomcat customizes the queue to prioritize thread creation for low‑latency web workloads.
Even though a gateway authenticates tokens and forwards user info, microservices must perform a second authorization check to enforce business rules, protect against internal threats, and uphold a zero‑trust security model, as explained with concrete examples and code.
The article explains why InheritableThreadLocal fails to propagate TraceId correctly in thread‑pool scenarios under high load, illustrates the problem with concrete request examples, and presents manual propagation and Alibaba's TransmittableThreadLocal as solutions.
The article examines how Java 21’s cheap virtual threads change the role of traditional thread pools, explaining why pooling virtual threads is a bad idea, how thread pools still act as natural throttlers, and which scenarios—especially CPU‑bound work and synchronized blocks—remain unsuitable for virtual threads.
The article explains why, despite the apparent speed of direct internal service calls, using an API gateway provides essential benefits such as unified governance, version‑aware routing, protocol translation, security isolation, and high‑availability traffic buffering, making it a worthwhile trade‑off in most microservice environments.
The article explains why fixed‑size chunking harms RAG recall, demonstrates three semantic‑chunking strategies—including recursive punctuation splitting, overlapping windows, and parent‑child document mapping—and provides complete Java/LangChain4j code that integrates tokenizers, Redis, and Qdrant to boost retrieval performance.
The article explains how using plain prompts to force large language models to output JSON for intent recognition and named‑entity extraction leads to unpredictable extra text, key mismatches, and hallucinated fields, and it presents three robust alternatives—placeholders, function calling, and constrained decoding—to achieve reliable structured outputs.
This article analyzes the emerging Java AI ecosystem, comparing Spring AI, LangChain4j, Spring AI Alibaba, AgentScope‑Java, and Microsoft Semantic Kernel, and provides guidance on selecting the right framework based on features such as RAG, agent support, observability, security sandbox, and cloud integration.
The article explains that Nacos‑based service deregistration is not instantly reflected in client load‑balancer caches due to periodic polling and unreliable UDP pushes, leading to a delay where gateways still route requests to a terminated instance, and then presents step‑by‑step lossless shutdown solutions for small teams and large‑scale deployments.
