调用逻辑:
hmaster.handleCreateTable->HRegion.createHRegion-> HRegion. initialize->initializeRegionInternals->instantiateHStore
->Store.Store->new CacheConfig(conf, family)-> CacheConfig.instantiateBlockCache->new LruBlockCache
传入参数
/** * Configurable constructor. Use this constructor if not using defaults. * @param maxSize maximum size of this cache, in bytes * @param blockSize expected average size of blocks, in bytes * @param evictionThread whether to run evictions in a bg thread or not * @param mapInitialSize initial size of backing ConcurrentHashMap * @param mapLoadFactor initial load factor of backing ConcurrentHashMap * @param mapConcurrencyLevel initial concurrency factor for backing CHM * @param minFactor percentage of total size that eviction will evict until * @param acceptableFactor percentage of total size that triggers eviction * @param singleFactor percentage of total size for single-access blocks * @param multiFactor percentage of total size for multiple-access blocks * @param memoryFactor percentage of total size for in-memory blocks */ public LruBlockCache(long maxSize, long blockSize, boolean evictionThread, int mapInitialSize, float mapLoadFactor, int mapConcurrencyLevel, float minFactor, float acceptableFactor, float singleFactor, float multiFactor, float memoryFactor)
new LruBlockCache时除了设置默认的参数外,还会创建evictionThread并wait和一个定时打印的线程StatisticsThread
当执行HFileReaderV2的readBlock时,会先看判断是否开户了Cache ,如果开启,则使用cache中block
// Check cache for block. If found return. if (cacheConf.isBlockCacheEnabled()) { // Try and get the block from the block cache. If the useLock variable is true then this // is the second time through the loop and it should not be counted as a block cache miss. HFileBlock cachedBlock = (HFileBlock) cacheConf.getBlockCache().getBlock(cacheKey, cacheBlock, useLock); if (cachedBlock != null) { BlockCategory blockCategory = cachedBlock.getBlockType().getCategory(); getSchemaMetrics().updateOnCacheHit(blockCategory, isCompaction); if (cachedBlock.getBlockType() == BlockType.DATA) { HFile.dataBlockReadCnt.incrementAndGet(); } validateBlockType(cachedBlock, expectedBlockType); // Validate encoding type for encoded blocks. We include encoding // type in the cache key, and we expect it to match on a cache hit. if (cachedBlock.getBlockType() == BlockType.ENCODED_DATA && cachedBlock.getDataBlockEncoding() != dataBlockEncoder.getEncodingInCache()) { throw new IOException("Cached block under key " + cacheKey + " " + "has wrong encoding: " + cachedBlock.getDataBlockEncoding() + " (expected: " + dataBlockEncoder.getEncodingInCache() + ")"); } return cachedBlock; } // Carry on, please load. }在getBlock方法中,会更新一些统计数据,重要的时更新
BlockPriority.SINGLE为BlockPriority.MULTI public Cacheable getBlock(BlockCacheKey cacheKey, boolean caching, boolean repeat) { CachedBlock cb = map.get(cacheKey); if(cb == null) { if (!repeat) stats.miss(caching); return null; } stats.hit(caching); cb.access(count.incrementAndGet()); return cb.getBuffer(); }---------------------
若是第一次读,则将block加入Cache.
// Cache the block if necessary if (cacheBlock && cacheConf.shouldCacheBlockOnRead( hfileBlock.getBlockType().getCategory())) { cacheConf.getBlockCache().cacheBlock(cacheKey, hfileBlock, cacheConf.isInMemory()); }
2. LRU evict
写入cache时就是将block加入到 一个 ConcurrentHashMap中,并更新Metrics,之后判断if(newSize > acceptableSize() && !evictionInProgress), acceptableSize是初始化时给的值(long)Math.floor(this.maxSize * this.acceptableFactor),acceptableFactor是一个百分比,是可以配置的:"hbase.lru.blockcache.acceptable.factor"(0.85f), 这里的意思就是判断总Size是不是大于这个值,如果大于并且没有正在执行的eviction线程, 那么就执行evict。
/** * Cache the block with the specified name and buffer. * <p> * It is assumed this will NEVER be called on an already cached block. If * that is done, an exception will be thrown. * @param cacheKey block's cache key * @param buf block buffer * @param inMemory if block is in-memory */ public void cacheBlock(BlockCacheKey cacheKey, Cacheable buf, boolean inMemory) { CachedBlock cb = map.get(cacheKey); if(cb != null) { throw new RuntimeException("Cached an already cached block"); } cb = new CachedBlock(cacheKey, buf, count.incrementAndGet(), inMemory); long newSize = updateSizeMetrics(cb, false); map.put(cacheKey, cb); elements.incrementAndGet(); if(newSize > acceptableSize() && !evictionInProgress) { runEviction(); } }
在evict方法中,
1. 计算总size和需要free的size, minsize = (long)Math.floor(this.maxSize * this.minFactor);其中minFactor是可配置的"hbase.lru.blockcache.min.factor"(0.75f);
long currentSize = this.size.get(); long bytesToFree = currentSize - minSize();
2. 初始化三种BlockBucket:bucketSingle,bucketMulti,bucketMemory并遍历map,按照三种类型分别add进各自的queue(MinMaxPriorityQueue.expectedSize(initialSize).create();)中, 并按照访问的次数逆序。
三种类型的区别是:
SINGLE对应第一次读的
MULTI对应多次读
MEMORY是设定column family中的IN_MEMORY为true的
// Instantiate priority buckets BlockBucket bucketSingle = new BlockBucket(bytesToFree, blockSize, singleSize()); BlockBucket bucketMulti = new BlockBucket(bytesToFree, blockSize, multiSize()); BlockBucket bucketMemory = new BlockBucket(bytesToFree, blockSize, memorySize());
其中三种BlockBuckt Size大小分配比例默认是:
static final float DEFAULT_SINGLE_FACTOR = 0.25f;
static final float DEFAULT_MULTI_FACTOR = 0.50f;
static final float DEFAULT_MEMORY_FACTOR = 0.25f;
private long singleSize() { return (long)Math.floor(this.maxSize * this.singleFactor * this.minFactor); } private long multiSize() { return (long)Math.floor(this.maxSize * this.multiFactor * this.minFactor); } private long memorySize() { return (long)Math.floor(this.maxSize * this.memoryFactor * this.minFactor); }
并将三种BlockBuckt 加入到优先队列中,按照totalSize - bucketSize排序,,再计算需要free大小,执行free:
PriorityQueue<BlockBucket> bucketQueue = new PriorityQueue<BlockBucket>(3); bucketQueue.add(bucketSingle); bucketQueue.add(bucketMulti); bucketQueue.add(bucketMemory); int remainingBuckets = 3; long bytesFreed = 0; BlockBucket bucket; while((bucket = bucketQueue.poll()) != null) { long overflow = bucket.overflow(); if(overflow > 0) { long bucketBytesToFree = Math.min(overflow, (bytesToFree - bytesFreed) / remainingBuckets); bytesFreed += bucket.free(bucketBytesToFree); } remainingBuckets--; }
free方法中一个一个取出queue中block,由于是按照访问次数逆序,所以从后面取出就是先取出访问次数少的,将其在map中一个一个remove, 并更新Mertrics.
public long free(long toFree) { CachedBlock cb; long freedBytes = 0; while ((cb = queue.pollLast()) != null) { freedBytes += evictBlock(cb); if (freedBytes >= toFree) { return freedBytes; } } return freedBytes; } protected long evictBlock(CachedBlock block) { map.remove(block.getCacheKey()); updateSizeMetrics(block, true); elements.decrementAndGet(); stats.evicted(); return block.heapSize(); }
3. HBase LruBlockCache的特点是针对不同的访问次数使用不同的策略,避免频繁的更新的Cache(便如Scan),这样更加有利于提高读的性能。