把 V8 垃圾回收的过程可视化。
// -----------------------------------------------------------------------------
// Heap structures:
//
// A JS heap consists of a young generation, an old generation, and a large
// object space. The young generation is divided into two semispaces. A
// scavenger implements Cheney's copying algorithm. The old generation is
// separated into a map space and an old object space. The map space contains
// all (and only) map objects, the rest of old objects go into the old space.
// The old generation is collected by a mark-sweep-compact collector.
//
// The semispaces of the young generation are contiguous. The old and map
// spaces consists of a list of pages. A page has a page header and an object
// area.
//
// There is a separate large object space for objects larger than
// Page::kMaxHeapObjectSize, so that they do not have to move during
// collection. The large object space is paged. Pages in large object space
// may be larger than the page size.
//
// A store-buffer based write barrier is used to keep track of intergenerational
// references. See heap/store-buffer.h.
//
// During scavenges and mark-sweep collections we sometimes (after a store
// buffer overflow) iterate intergenerational pointers without decoding heap
// object maps so if the page belongs to old pointer space or large object
// space it is essential to guarantee that the page does not contain any
// garbage pointers to new space: every pointer aligned word which satisfies
// the Heap::InNewSpace() predicate must be a pointer to a live heap object in
// new space. Thus objects in old pointer and large object spaces should have a
// special layout (e.g. no bare integer fields). This requirement does not
// apply to map space which is iterated in a special fashion. However we still
// require pointer fields of dead maps to be cleaned.
//
// To enable lazy cleaning of old space pages we can mark chunks of the page
// as being garbage. Garbage sections are marked with a special map. These
// sections are skipped when scanning the page, even if we are otherwise
// scanning without regard for object boundaries. Garbage sections are chained
// together to form a free list after a GC. Garbage sections created outside
// of GCs by object trunctation etc. may not be in the free list chain. Very
// small free spaces are ignored, they need only be cleaned of bogus pointers
// into new space.
//
// Each page may have up to one special garbage section. The start of this
// section is denoted by the top field in the space. The end of the section
// is denoted by the limit field in the space. This special garbage section
// is not marked with a free space map in the data. The point of this section
// is to enable linear allocation without having to constantly update the byte
// array every time the top field is updated and a new object is created. The
// special garbage section is not in the chain of garbage sections.
//
// Since the top and limit fields are in the space, not the page, only one page
// has a special garbage section, and if the top and limit are equal then there
// is no special garbage section.
关键字
-
Young Generation
-
Old Generation
-
Large Object Space
运行一个 JavaScript 应用程序的命令。
node index.js
如果要输出 GC 日志，只需要加上一个配置项：
node --trace_gc index.js
这里的 --trace_gc 是 V8 的一个配置项，所以要放在中间，也就是说如果把命令敲成 node index.js --trace_gc 是不可以的。
如果想知道除了 --trace_gc 还有哪些配置项可以用，可以用命令 node --v8-options | grep gc 列出所有和 GC 相关的选项。
--expose_gc (expose gc extension)
--gc_global (always perform global GCs)
--gc_interval (garbage collect after <n> allocations)
--trace_gc (print one trace line following each garbage collection)
--trace_gc_nvp (print one detailed trace line in name=value format after each garbage collection)
--trace_gc_ignore_scavenger (do not print trace line after scavenger collection)
--print_cumulative_gc_stat (print cumulative GC statistics in name=value format on exit)
--trace_gc_verbose (print more details following each garbage collection)
--flush_code (flush code that we expect not to use again before full gc)
--track_gc_object_stats (track object counts and memory usage)
--cleanup_code_caches_at_gc (Flush inline caches prior to mark compact collection and flush code caches in maps during mark compact cycle.)
--log_gc (Log heap samples on garbage collection for the hp2ps tool.)
--gc_fake_mmap (Specify the name of the file for fake gc mmap used in ll_prof)
需要重点关注的选项：
--trace_gc--trace_gc_nvp--trace_gc_verbose
这些选项之间有叠加和覆盖的关系：
1. 如果启动应用的时候开启了 --trace-gc 选项，则每次GC后输出一行简明扼要的信息。
示例1: node --trace_gc index.js
输出：
[10189] 682 ms: Scavenge 2.3 (36.0) -> 1.9 (37.0) MB, 1 ms [Runtime::PerformGC].
...
2. 如果加上 --trace_gc_verbose 则输出一些列的键值对。
示例2： node --trace_gc --trace_gc_nvp index.js
输出：
[9893] 636 ms: pause=0 mutator=0 gc=s external=0 mark=0 sweep=0 sweepns=0 evacuate=0 new_new=0 root_new=0 old_new=0 compaction_ptrs=0 intracompaction_ptrs=0 misc_compaction=0 total_size_before=2398448 total_size_after=2017168 holes_size_before=169032 holes_size_after=176640 allocated=2398448 promoted=392648 stepscount=0 stepstook=0
3. 第三个选项和 GC 没有直接的关系，但是可以在每次 GC 结束后输出各个区域的分配情况。
示例3: node --trace_gc --trace_gc_verbose index.js
[9800] 9870 ms: Scavenge 3.0 (37.0) -> 2.2 (37.0) MB, 1 ms [allocation failure].
[9800] Memory allocator, used: 37904 KB, available: 1461232 KB
[9800] New space, used: 87 KB, available: 1960 KB, committed: 4096 KB
[9800] Old pointers, used: 1011 KB, available: 164 KB, committed: 1519 KB
[9800] Old data space, used: 579 KB, available: 7 KB, committed: 1199 KB
[9800] Code space, used: 430 KB, available: 0 KB, committed: 996 KB
[9800] Map space, used: 93 KB, available: 0 KB, committed: 128 KB
[9800] Cell space, used: 15 KB, available: 0 KB, committed: 128 KB
[9800] Large object space, used: 0 KB, available: 1460191 KB, committed: 0 KB
[9800] All spaces, used: 2218 KB, available: 2132 KB, committed: 8067 KB
[9800] Total time spent in GC : 4 ms
这三个选项的对应的关系可以从源码中找到： void GCTracer::Stop() void Heap::PrintShortHeapStatistics()
通过在 Node.JS 启动过程中加上相应的配置项，把 GC 日志输出到命令行，接着通过操作系统的管道传递给另外一个 Node.JS 实现的工具。 这个工具能够解析命令行的输出，并通过 WebSocket 传递给浏览器里的应用，由浏览器负责视觉呈现。
实现的难点在与对 GC 日志格式的理解。
| 属性 | 变量 | 解释 |
|---|---|---|
| pause | duration | |
| mutator | spent_in_mutator | |
| gc | TypeName(true) | |
| external | ||
| mark | ||
| sweep | ||
| sweepns | ||
| sweepos | ||
| sweepcode | ||
| sweepcell | ||
| sweepmap | ||
| evacuate | ||
| new_new | ||
| root_new | ||
| old_new | ||
| compaction_ptrs | ||
| intracompaction_ptrs | ||
| misc_compaction | ||
| weakcollection_process | ||
| weakcollection_clear | ||
| weakcollection_abort | ||
| total_size_before | ||
| total_size_after | ||
| holes_size_before | ||
| holes_size_after | ||
| allocated | ||
| promoted | ||
| nodes_died_in_new | ||
| nodes_copied_in_new | ||
| nodes_promoted | ||
| promotion_rate | ||
| semi_space_copy_rate | ||
| steps_count | ||
| steps_took | ||
| longest_step | ||
| incremental_marking_throughput |
V8 对内存的管理和与内存分配相关的系统调用密切相关：
- mmap
- munmap
mmap(start,length,prot,flags,fd,offset) 将一个文件或其它对象映射进内存。
start 映射区开始地址
length 映射区的长度
prot 内存保护标志
flags 对象的类型
fd 文件描述符
offset 被映射对象内容的起点
munmap(start,length) 删除特定区域的对象映射。如果映射关系解除，访问原来的地址将发生段错误。
start 映射区的起点
length 映射区的长度
实际工作中可以使用 strace 记录系统调用的情况。
sudo strace -p pid -e mmap,munmap -ttt
1.安装命令行工具 onegc
npm install onegc -g
2.用下面的命令启动 Node.JS 应用
node --trace_gc --trace_gc_nvp --trace_gc_verbose server.js | onegc
3.在浏览器里打开浏览器端