JVM 内存和垃圾回收如何工作？

Question

JVM 内存和垃圾回收如何工作？

Accepted Answer

JVM **自动管理内存** — 分配对象并通过 **垃圾回收 (GC)** 释放无法到达的对象 — 因此您不需要手动释放内存。理解内存区域和 GC 行为是调优性能和诊断内存问题的关键。 ## JVM 内存区域 ```text HEAP — where all OBJECTS live (shared across threads), GC operates here ├── Young Generation — new objects (most die young) │ ├── Eden — new allocations go here │ └── Survivor — objects that survived a young GC └── Old (Tenured) Gen — long-lived objects (promoted from young) STACK — per-thread; method frames, local variables, primitive locals, references Metaspace — class metadata (replaced PermGen in Java 8+) ``` 对象存在于 **heap** 上；本地变量和引用存在于每线程的 **stacks** 上。GC 关注堆。 ## 分代假说 ```text Observation: MOST objects die young (short-lived). → GC splits the heap into generations and collects the young gen FREQUENTLY (cheap, fast) and the old gen RARELY (expensive). ``` ## GC 如何工作：标记和清扫 ```text 1. MARK — starting from GC roots (stack refs, static fields), mark all REACHABLE objects 2. SWEEP — reclaim memory of unmarked (unreachable) objects 3. COMPACT— (often) move survivors together to reduce fragmentation ``` 一个对象是垃圾当它不再从任何 GC root **可达**。您不释放它 — GC 查找并回收它。 ## Minor vs Major GC ```text Minor GC — collects the Young Generation; frequent, fast, short pause Major/Full GC — collects the Old Generation (and more); rarer, longer pause → long Full GC pauses ("stop-the-world") are a common performance concern ``` ## GC 算法（根据工作负载选择） ```text G1 GC (default, Java 9+) — balanced, predictable pause times, good general choice Parallel GC — max throughput (batch jobs), longer pauses acceptable ZGC / Shenandoah — very low pause times for large heaps (latency-sensitive apps) ``` ## 内存泄漏仍然会发生（尽管有 GC） ```java // ❌ a leak: a static collection that only grows keeps objects REACHABLE forever static List

JVM 内存区域

分代假说

GC 如何工作：标记和清扫

Minor vs Major GC

GC 算法（根据工作负载选择）

内存泄漏仍然会发生（尽管有 GC）

为什么这很重要