HBASE-22612 Address the final overview reviewing comments of HBASE-21879 (#331)

Author: openinx

hbase-common/src/main/java/org/apache/hadoop/hbase/io/ByteBufferListOutputStream.java

@@ -41,7 +41,7 @@
 public class ByteBufferListOutputStream extends ByteBufferOutputStream {
   private static final Logger LOG = LoggerFactory.getLogger(ByteBufferListOutputStream.class);
 
-  private ByteBuffAllocator allocator;
+  private final ByteBuffAllocator allocator;
   // Keep track of the BBs where bytes written to. We will first try to get a BB from the pool. If
   // it is not available will make a new one our own and keep writing to that. We keep track of all
   // the BBs that we got from pool, separately so that on closeAndPutbackBuffers, we can make sure

hbase-common/src/main/java/org/apache/hadoop/hbase/io/hfile/HFileContext.java

@@ -193,8 +193,8 @@ public String getHFileName() {
   }
 
   /**
-   * HeapSize implementation. NOTE : The heapsize should be altered as and when new state variable
-   * are added
+   * HeapSize implementation. NOTE : The heap size should be altered when new state variable are
+   * added.
    * @return heap size of the HFileContext
    */
   @Override

hbase-common/src/main/java/org/apache/hadoop/hbase/nio/RefCnt.java

@@ -26,7 +26,7 @@
 
 /**
  * Maintain an reference count integer inside to track life cycle of {@link ByteBuff}, if the
- * reference count become 0, it'll call {@link Recycler#free()} once.
+ * reference count become 0, it'll call {@link Recycler#free()} exactly once.
  */
 @InterfaceAudience.Private
 public class RefCnt extends AbstractReferenceCounted {
@@ -36,8 +36,8 @@ public class RefCnt extends AbstractReferenceCounted {
   /**
    * Create an {@link RefCnt} with an initial reference count = 1. If the reference count become
    * zero, the recycler will do nothing. Usually, an Heap {@link ByteBuff} will use this kind of
-   * refCnt to track its life cycle, it help to abstract the code path although it's meaningless to
-   * use an refCnt for heap ByteBuff.
+   * refCnt to track its life cycle, it help to abstract the code path although it's not really
+   * needed to track on heap ByteBuff.
    */
   public static RefCnt create() {
     return new RefCnt(ByteBuffAllocator.NONE);

hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/LruBlockCache.java

@@ -153,8 +153,14 @@ public class LruBlockCache implements FirstLevelBlockCache {
   private static final String LRU_MAX_BLOCK_SIZE = "hbase.lru.max.block.size";
   private static final long DEFAULT_MAX_BLOCK_SIZE = 16L * 1024L * 1024L;
 
-  /** Concurrent map (the cache) */
-  private transient final Map<BlockCacheKey, LruCachedBlock> map;
+  /**
+   * Defined the cache map as {@link ConcurrentHashMap} here, because in
+   * {@link LruBlockCache#getBlock}, we need to guarantee the atomicity of map#computeIfPresent
+   * (key, func). Besides, the func method must execute exactly once only when the key is present
+   * and under the lock context, otherwise the reference count will be messed up. Notice that the
+   * {@link java.util.concurrent.ConcurrentSkipListMap} can not guarantee that.
+   */
+  private transient final ConcurrentHashMap<BlockCacheKey, LruCachedBlock> map;
 
   /** Eviction lock (locked when eviction in process) */
   private transient final ReentrantLock evictionLock = new ReentrantLock(true);

hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/bucket/BucketCache.java

@@ -140,7 +140,7 @@ public class BucketCache implements BlockCache, HeapSize {
   transient final RAMCache ramCache;
   // In this map, store the block's meta data like offset, length
   @VisibleForTesting
-  transient ConcurrentMap<BlockCacheKey, BucketEntry> backingMap;
+  transient ConcurrentHashMap<BlockCacheKey, BucketEntry> backingMap;
 
   /**
    * Flag if the cache is enabled or not... We shut it off if there are IO
@@ -1526,7 +1526,16 @@ float getMemoryFactor() {
    * Wrapped the delegate ConcurrentMap with maintaining its block's reference count.
    */
   static class RAMCache {
-    final ConcurrentMap<BlockCacheKey, RAMQueueEntry> delegate = new ConcurrentHashMap<>();
+    /**
+     * Defined the map as {@link ConcurrentHashMap} explicitly here, because in
+     * {@link RAMCache#get(BlockCacheKey)} and
+     * {@link RAMCache#putIfAbsent(BlockCacheKey, RAMQueueEntry)} , we need to guarantee the
+     * atomicity of map#computeIfPresent(key, func) and map#putIfAbsent(key, func). Besides, the
+     * func method can execute exactly once only when the key is present(or absent) and under the
+     * lock context. Otherwise, the reference count of block will be messed up. Notice that the
+     * {@link java.util.concurrent.ConcurrentSkipListMap} can not guarantee that.
+     */
+    final ConcurrentHashMap<BlockCacheKey, RAMQueueEntry> delegate = new ConcurrentHashMap<>();
 
     public boolean containsKey(BlockCacheKey key) {
       return delegate.containsKey(key);