Class DfsBlockCache


  • public final class DfsBlockCache
    extends java.lang.Object
    Caches slices of a BlockBasedFile in memory for faster read access.

    The DfsBlockCache serves as a Java based "buffer cache", loading segments of a BlockBasedFile into the JVM heap prior to use. As JGit often wants to do reads of only tiny slices of a file, the DfsBlockCache tries to smooth out these tiny reads into larger block-sized IO operations.

    Whenever a cache miss occurs, loading is invoked by exactly one thread for the given (DfsStreamKey,position) key tuple. This is ensured by an array of locks, with the tuple hashed to a lock instance.

    Its too expensive during object access to be accurate with a least recently used (LRU) algorithm. Strictly ordering every read is a lot of overhead that typically doesn't yield a corresponding benefit to the application. This cache implements a clock replacement algorithm, giving each block at least one chance to have been accessed during a sweep of the cache to save itself from eviction. The number of swipe chances is configurable per pack extension.

    Entities created by the cache are held under hard references, preventing the Java VM from clearing anything. Blocks are discarded by the replacement algorithm when adding a new block would cause the cache to exceed its configured maximum size.

    The key tuple is passed through to methods as a pair of parameters rather than as a single Object, thus reducing the transient memory allocations of callers. It is more efficient to avoid the allocation, as we can't be 100% sure that a JIT would be able to stack-allocate a key tuple.

    The internal hash table does not expand at runtime, instead it is fixed in size at cache creation time. The internal lock table used to gate load invocations is also fixed in size.

    • Field Detail

      • tableSize

        private final int tableSize
        Number of entries in table.
      • table

        private final java.util.concurrent.atomic.AtomicReferenceArray<DfsBlockCache.HashEntry> table
        Hash bucket directory; entries are chained below.
      • loadLocks

        private final java.util.concurrent.locks.ReentrantLock[] loadLocks
        Locks to prevent concurrent loads for same (PackFile,position) block. The number of locks is DfsBlockCacheConfig.getConcurrencyLevel() to cap the overall concurrent block loads.
      • refLocks

        private final java.util.concurrent.locks.ReentrantLock[][] refLocks
        A separate pool of locks per pack extension to prevent concurrent loads for same index or bitmap from PackFile.
      • maxBytes

        private final long maxBytes
        Maximum number of bytes the cache should hold.
      • maxStreamThroughCache

        private final long maxStreamThroughCache
        Pack files smaller than this size can be copied through the cache.
      • blockSize

        private final int blockSize
        Suggested block size to read from pack files in.

        If a pack file does not have a native block size, this size will be used.

        If a pack file has a native size, a whole multiple of the native size will be used until it matches this size.

        The value for blockSize must be a power of 2.

      • blockSizeShift

        private final int blockSizeShift
        As blockSize is a power of 2, bits to shift for a / blockSize.
      • statHit

        private final java.util.concurrent.atomic.AtomicReference<java.util.concurrent.atomic.AtomicLong[]> statHit
        Number of times a block was found in the cache, per pack file extension.
      • statMiss

        private final java.util.concurrent.atomic.AtomicReference<java.util.concurrent.atomic.AtomicLong[]> statMiss
        Number of times a block was not found, and had to be loaded, per pack file extension.
      • statEvict

        private final java.util.concurrent.atomic.AtomicReference<java.util.concurrent.atomic.AtomicLong[]> statEvict
        Number of blocks evicted due to cache being full, per pack file extension.
      • liveBytes

        private final java.util.concurrent.atomic.AtomicReference<java.util.concurrent.atomic.AtomicLong[]> liveBytes
        Number of bytes currently loaded in the cache, per pack file extension.
      • clockLock

        private final java.util.concurrent.locks.ReentrantLock clockLock
        Protects the clock and its related data.
      • refLockWaitTime

        private final java.util.function.Consumer<java.lang.Long> refLockWaitTime
        A consumer of object reference lock wait time milliseconds. May be used to build a metric.
      • cacheHotLimits

        private final int[] cacheHotLimits
        Limits of cache hot count per pack file extension.
      • indexEvictionMap

        private final java.util.Map<DfsBlockCache.EvictKey,​java.lang.Long> indexEvictionMap
        Stores timestamps of the last eviction of indexes.
    • Method Detail

      • reconfigure

        public static void reconfigure​(DfsBlockCacheConfig cfg)
        Modify the configuration of the window cache.

        The new configuration is applied immediately, and the existing cache is cleared.

        Parameters:
        cfg - the new window cache configuration.
        Throws:
        java.lang.IllegalArgumentException - the cache configuration contains one or more invalid settings, usually too low of a limit.
      • getInstance

        public static DfsBlockCache getInstance()
        Get the currently active DfsBlockCache.
        Returns:
        the currently active DfsBlockCache.
      • shouldCopyThroughCache

        boolean shouldCopyThroughCache​(long length)
      • getCurrentSize

        public long[] getCurrentSize()
        Get total number of bytes in the cache, per pack file extension.
        Returns:
        total number of bytes in the cache, per pack file extension.
      • getFillPercentage

        public long getFillPercentage()
        Get 0..100, defining how full the cache is.
        Returns:
        0..100, defining how full the cache is.
      • getHitCount

        public long[] getHitCount()
        Get number of requests for items in the cache, per pack file extension.
        Returns:
        number of requests for items in the cache, per pack file extension.
      • getMissCount

        public long[] getMissCount()
        Get number of requests for items not in the cache, per pack file extension.
        Returns:
        number of requests for items not in the cache, per pack file extension.
      • getTotalRequestCount

        public long[] getTotalRequestCount()
        Get total number of requests (hit + miss), per pack file extension.
        Returns:
        total number of requests (hit + miss), per pack file extension.
      • getHitRatio

        public long[] getHitRatio()
        Get hit ratios
        Returns:
        hit ratios
      • getEvictions

        public long[] getEvictions()
        Get number of evictions performed due to cache being full, per pack file extension.
        Returns:
        number of evictions performed due to cache being full, per pack file extension.
      • hasBlock0

        public boolean hasBlock0​(DfsStreamKey key)
        Quickly check if the cache contains block 0 of the given stream.

        This can be useful for sophisticated pre-read algorithms to quickly determine if a file is likely already in cache, especially small reftables which may be smaller than a typical DFS block size.

        Parameters:
        key - the file to check.
        Returns:
        true if block 0 (the first block) is in the cache.
      • hash

        private int hash​(int packHash,
                         long off)
      • getBlockSize

        int getBlockSize()
      • getOrLoad

        DfsBlock getOrLoad​(BlockBasedFile file,
                           long position,
                           DfsReader ctx,
                           DfsBlockCache.ReadableChannelSupplier fileChannel)
                    throws java.io.IOException
        Look up a cached object, creating and loading it if it doesn't exist.
        Parameters:
        file - the pack that "contains" the cached object.
        position - offset within pack of the object.
        ctx - current thread's reader.
        fileChannel - supplier for channel to read pack.
        Returns:
        the object reference.
        Throws:
        java.io.IOException - the reference was not in the cache and could not be loaded.
      • reserveSpace

        private void reserveSpace​(long reserve,
                                  DfsStreamKey key)
      • creditSpace

        private void creditSpace​(long credit,
                                 DfsStreamKey key)
      • getOrLoadRef

        <T> DfsBlockCache.Ref<T> getOrLoadRef​(DfsStreamKey key,
                                              long position,
                                              DfsBlockCache.RefLoader<T> loader)
                                       throws java.io.IOException
        Look up a cached object, creating and loading it if it doesn't exist.
        Parameters:
        key - the stream key of the pack.
        position - the position in the key. The default should be 0.
        loader - the function to load the reference.
        Returns:
        the object reference.
        Throws:
        java.io.IOException - the reference was not in the cache and could not be loaded.
      • contains

        boolean contains​(DfsStreamKey key,
                         long position)
      • slot

        private int slot​(DfsStreamKey key,
                         long position)
      • lockFor

        private java.util.concurrent.locks.ReentrantLock lockFor​(DfsStreamKey key,
                                                                 long position)
      • lockForRef

        private java.util.concurrent.locks.ReentrantLock lockForRef​(DfsStreamKey key)
      • newCounters

        private static java.util.concurrent.atomic.AtomicLong[] newCounters()
      • getStat

        private static java.util.concurrent.atomic.AtomicLong getStat​(java.util.concurrent.atomic.AtomicReference<java.util.concurrent.atomic.AtomicLong[]> stats,
                                                                      DfsStreamKey key)
      • getStatVals

        private static long[] getStatVals​(java.util.concurrent.atomic.AtomicReference<java.util.concurrent.atomic.AtomicLong[]> stat)
      • reportIndexRequested

        private void reportIndexRequested​(DfsBlockCache.Ref<?> ref,
                                          boolean cacheHit,
                                          long start)
      • reportIndexEvicted

        private void reportIndexEvicted​(DfsBlockCache.Ref<?> dead)
      • isIndexOrBitmapExtPos

        private static boolean isIndexOrBitmapExtPos​(int packExtPos)