Manifest与Current文件是LevelDB保存在稳定存储中的文件版本信息,在LevelDB被打开后,其会先通过Current文件找到当前的Manifest文件,读取并反序列化其中数据,并在内存中维护文件版本信息,以便后续操作。
先来说说Version,Version表示了当前leveldb的版本信息,
版本信息内容包括:1当前每一层的SSTable文件元信息。2记录被Seek太多次需要Compact的文件元信息,以及文件所在的level。3记录所有level层中compaction score 最大的那一层及其level,用于比较判断是否需要对此level层进行compact。
leveldb/Version.java at master · dain/leveldb · GitHub
public class Version implements SeekingIterable<InternalKey, Slice> { private final AtomicInteger retained = new AtomicInteger(1); private final VersionSet versionSet; private final Level0 level0; private final List<Level> levels; // move these mutable fields somewhere else private int compactionLevel; private double compactionScore; private FileMetaData fileToCompact; private int fileToCompactLevel;
AtomicInteger是提供原子操作的Integer类,通过线程安全的方式操作加减。
leveldb内部只有一个VersionSet进行管理所有的Version. 当不断有新版本生成的时候,那么就需要不断地 Append 到 versionset 里面。 当旧的 Version 不再服务读请求之后,这个 Version 就会从Versionset中移除。
leveldb的第一层level0 并新建List<Level> levels记录文件所在level信息。
FileMetaData使用来表示sstable的元数据。fileTocompact记录下一个等待compact的file。当fileMataData文件查找到一定次数时就要执行合并操作。
version改变时结合compactionLevel和compactionScore的信息确定下次compaction的level和文件。
fileToCompactLevel记录文件压缩到哪个level。
对 Version 的修改主要发生于 compaction 之后。compaction 分为 minor compaction 和 major compaction,其中 minor compaction 是落 memtable 到 L0,只会新增文件,而 major compaction 会跨 level 做合并,既新增文件也删除文件。每当这时,便会生成一个新的 VersionEdit 产生新的 Version。
public Version(VersionSet versionSet) { this.versionSet = versionSet; checkArgument(NUM_LEVELS > 1, "levels must be at least 2"); this.level0 = new Level0(new ArrayList<FileMetaData>(), getTableCache(), getInternalKeyComparator()); Builder<Level> builder = ImmutableList.builder(); for (int i = 1; i < NUM_LEVELS; i++) { List<FileMetaData> files = new ArrayList<>(); builder.add(new Level(i, files, getTableCache(), getInternalKeyComparator())); } this.levels = builder.build(); }
先判断versionset里的level层数>1 若为Level0则新建对象level0,由FileMetaData TableCache 和 InternalKeyComparartor组成。
ImmutableList.builder方法新建builder对象,更新产生新的version,将新的level对象add到builder中。
public void assertNoOverlappingFiles() { for (int level = 1; level < NUM_LEVELS; level++) { assertNoOverlappingFiles(level); } }
判断SStables有没有重复的Key范围
public void assertNoOverlappingFiles(int level) { if (level > 0) { Collection<FileMetaData> files = getFiles().asMap().get(level); if (files != null) { long previousFileNumber = 0; InternalKey previousEnd = null; for (FileMetaData fileMetaData : files) { if (previousEnd != null) { checkArgument(getInternalKeyComparator().compare( previousEnd, fileMetaData.getSmallest() ) < 0, "Overlapping files %s and %s in level %s", previousFileNumber, fileMetaData.getNumber(), level); } previousFileNumber = fileMetaData.getNumber(); previousEnd = fileMetaData.getLargest(); } } } }
根据PreviousFileNumber(InternalKey类) ,fileMetaData 和所在的 level,检查是否有重叠的Files
private TableCache getTableCache() { return versionSet.getTableCache(); }
获取TableCache
public final InternalKeyComparator getInternalKeyComparator() { return versionSet.getInternalKeyComparator(); }
获取InternalKey比较器
public synchronized int getCompactionLevel() { return compactionLevel; }
获取compaction的level
public synchronized void setCompactionLevel(int compactionLevel) { this.compactionLevel = compactionLevel; }
设置compaction的level
public synchronized double getCompactionScore() { return compactionScore; }
获取compactionScore,决定是否触发compaction
public synchronized void setCompactionScore(double compactionScore) { this.compactionScore = compactionScore; }
设置CampactionScore
@Override public MergingIterator iterator() { Builder<InternalIterator> builder = ImmutableList.builder(); builder.add(level0.iterator()); builder.addAll(getLevelIterators()); return new MergingIterator(builder.build(), getInternalKeyComparator()); }
如果每个iterator中的key有序,但是所有iterator中的所有key全局无序,此时,需要一种能够“归并”多路有序iterator的结构。这一结构即MergingIterator
传入level0的iterator和所有Level的iterator,以及用来比较InternalKey的Comparator。InternalKey迭代器迭代器组合了MemTable Iterator、Immutable MemTable Iterator、每个Level-0 SSTable的Iterator,和level>1的所有SSTable的Concatenating Iterator。
List<InternalTableIterator> getLevel0Files() { Builder<InternalTableIterator> builder = ImmutableList.builder(); for (FileMetaData file : level0.getFiles()) { builder.add(getTableCache().newIterator(file)); } return builder.build(); }
新建InternalTableIterator的对象获取TableCache中的level0的file。迭代方法使用时都反复需要获取其中iterator是否为valid或获取其value,不需要每次都访问到最下层的iterator,只需要访问缓存状态即可。
List<LevelIterator> getLevelIterators() { Builder<LevelIterator> builder = ImmutableList.builder(); for (Level level : levels) { if (!level.getFiles().isEmpty()) { builder.add(level.iterator()); } } return builder.build(); }
遍历每一层Level,获取level的Iterator
public LookupResult get(LookupKey key) { // We can search level-by-level since entries never hop across // levels. Therefore we are guaranteed that if we find data // in an smaller level, later levels are irrelevant. ReadStats readStats = new ReadStats(); LookupResult lookupResult = level0.get(key, readStats); if (lookupResult == null) { for (Level level : levels) { lookupResult = level.get(key, readStats); if (lookupResult != null) { break; } } } updateStats(readStats.getSeekFileLevel(), readStats.getSeekFile()); return lookupResult; }
获取level0的LookupKey
int pickLevelForMemTableOutput(Slice smallestUserKey, Slice largestUserKey) { int level = 0; if (!overlapInLevel(0, smallestUserKey, largestUserKey)) { // Push to next level if there is no overlap in next level, // and the #bytes overlapping in the level after that are limited. InternalKey start = new InternalKey(smallestUserKey, MAX_SEQUENCE_NUMBER, ValueType.VALUE); InternalKey limit = new InternalKey(largestUserKey, 0, ValueType.VALUE); while (level < MAX_MEM_COMPACT_LEVEL) { if (overlapInLevel(level + 1, smallestUserKey, largestUserKey)) { break; } long sum = Compaction.totalFileSize(versionSet.getOverlappingInputs(level + 2, start, limit)); if (sum > MAX_GRAND_PARENT_OVERLAP_BYTES) { break; } level++; } } return level; }
pickLevelForMemTableOutput:从Memtable中dump到level0
leveldb中的db文件本身没有层次概念,所有的db文件都一样,如何确定这个文件是在哪一层由pickLevelForMemTableOutput方法实现。
public boolean overlapInLevel(int level, Slice smallestUserKey, Slice largestUserKey) { checkPositionIndex(level, levels.size(), "Invalid level"); requireNonNull(smallestUserKey, "smallestUserKey is null"); requireNonNull(largestUserKey, "largestUserKey is null"); if (level == 0) { return level0.someFileOverlapsRange(smallestUserKey, largestUserKey); } return levels.get(level - 1).someFileOverlapsRange(smallestUserKey, largestUserKey); }public boolean overlapInLevel(int level, Slice smallestUserKey, Slice largestUserKey) { checkPositionIndex(level, levels.size(), "Invalid level"); requireNonNull(smallestUserKey, "smallestUserKey is null"); requireNonNull(largestUserKey, "largestUserKey is null"); if (level == 0) { return level0.someFileOverlapsRange(smallestUserKey, largestUserKey); } return levels.get(level - 1).someFileOverlapsRange(smallestUserKey, largestUserKey); }
参考:深入浅出LevelDB —— 08 Iterator – 叉鸽 MrCroxx 的博客
原创文章,作者:3628473679,如若转载,请注明出处:https://blog.ytso.com/275864.html