com.bigdata.rwstore.RWStore Maven / Gradle / Ivy
/**
Copyright (C) SYSTAP, LLC DBA Blazegraph 2006-2016. All rights reserved.
Contact:
SYSTAP, LLC DBA Blazegraph
2501 Calvert ST NW #106
Washington, DC 20008
[email protected]
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; version 2 of the License.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
package com.bigdata.rwstore;
import java.io.ByteArrayInputStream;
import java.io.DataInputStream;
import java.io.File;
import java.io.IOException;
import java.io.InputStream;
import java.io.OutputStream;
import java.io.RandomAccessFile;
import java.lang.ref.WeakReference;
import java.nio.ByteBuffer;
import java.nio.channels.AsynchronousFileChannel;
import java.nio.channels.Channel;
import java.nio.channels.ClosedByInterruptException;
import java.nio.channels.FileChannel;
import java.nio.file.Path;
import java.nio.file.Paths;
import java.nio.file.StandardOpenOption;
import java.security.DigestException;
import java.security.MessageDigest;
import java.util.ArrayList;
import java.util.Collections;
import java.util.HashMap;
import java.util.Iterator;
import java.util.LinkedHashSet;
import java.util.Map;
import java.util.Map.Entry;
import java.util.Set;
import java.util.UUID;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ExecutionException;
import java.util.concurrent.Future;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicBoolean;
import java.util.concurrent.atomic.AtomicReference;
import java.util.concurrent.locks.Lock;
import java.util.concurrent.locks.ReentrantReadWriteLock;
import java.util.concurrent.locks.ReentrantReadWriteLock.ReadLock;
import java.util.concurrent.locks.ReentrantReadWriteLock.WriteLock;
import org.apache.log4j.Logger;
import com.bigdata.btree.BTree.Counter;
import com.bigdata.btree.IIndex;
import com.bigdata.btree.ITuple;
import com.bigdata.btree.ITupleIterator;
import com.bigdata.btree.IndexMetadata;
import com.bigdata.cache.ConcurrentWeakValueCache;
import com.bigdata.counters.CounterSet;
import com.bigdata.counters.Instrument;
import com.bigdata.counters.striped.StripedCounters;
import com.bigdata.ha.HAGlue;
import com.bigdata.ha.HAPipelineGlue;
import com.bigdata.ha.QuorumPipeline;
import com.bigdata.ha.QuorumService;
import com.bigdata.ha.msg.HAWriteMessage;
import com.bigdata.ha.msg.IHALogRequest;
import com.bigdata.ha.msg.IHARebuildRequest;
import com.bigdata.ha.msg.IHAWriteMessage;
import com.bigdata.io.ChecksumUtility;
import com.bigdata.io.DirectBufferPool;
import com.bigdata.io.FileChannelUtility;
import com.bigdata.io.FileChannelUtility.AsyncTransfer;
import com.bigdata.io.IBufferAccess;
import com.bigdata.io.IReopenChannel;
import com.bigdata.io.MergeStreamWithSnapshotData;
import com.bigdata.io.compression.CompressorRegistry;
import com.bigdata.io.compression.IRecordCompressor;
import com.bigdata.io.writecache.BufferedWrite;
import com.bigdata.io.writecache.IBackingReader;
import com.bigdata.io.writecache.IBufferedWriter;
import com.bigdata.io.writecache.WriteCache;
import com.bigdata.io.writecache.WriteCacheService;
import com.bigdata.journal.AbstractBufferStrategy;
import com.bigdata.journal.AbstractJournal;
import com.bigdata.journal.AbstractJournal.ISnapshotData;
import com.bigdata.journal.CommitRecordIndex;
import com.bigdata.journal.CommitRecordSerializer;
import com.bigdata.journal.FileMetadata;
import com.bigdata.journal.ForceEnum;
import com.bigdata.journal.ICommitRecord;
import com.bigdata.journal.ICommitter;
import com.bigdata.journal.IHABufferStrategy;
import com.bigdata.journal.IRootBlockView;
import com.bigdata.journal.RootBlockView;
import com.bigdata.journal.StoreState;
import com.bigdata.journal.StoreTypeEnum;
import com.bigdata.quorum.Quorum;
import com.bigdata.quorum.QuorumException;
import com.bigdata.rawstore.IAllocationContext;
import com.bigdata.rawstore.IPSOutputStream;
import com.bigdata.rawstore.IRawStore;
import com.bigdata.rwstore.StorageStats.Bucket;
import com.bigdata.service.AbstractTransactionService;
import com.bigdata.util.BytesUtil;
import com.bigdata.util.ChecksumError;
/**
* Storage class
*
* Provides an interface to allocating storage within a disk file.
*
* Essentially provides a DiskMalloc interface.
*
* In addition to the DiskMalloc/ReAlloc mechanism, a single root address can be
* associated. This can be used when opening an existing storage file to
* retrieve some management object - such as an object manager!
*
* The allocator also support atomic update via a simple transaction mechanism.
*
* Updates are normally committed immediately, but by using startTransaction and
* commitTransaction, the previous state of the store is retained until the
* moment of commitment.
*
* It would also be possible to add some journaling/version mechanism, where
* snapshots of the allocation maps are retained for sometime. For a store which
* was only added to this would not be an unreasonable overhead and would
* support the rolling back of the database weekly or monthly if required.
*
* The input/output mechanism uses ByteArray Input and Output Streams.
*
* One difference between the disk realloc and in memory realloc is that the
* disk realloc will always return a new address and mark the old address as
* ready to be freed.
*
* The method of storing the allocation headers has been changed from always
* allocating at the end of the file (and moving them on file extend) to
* allocation of fixed areas. The meta-allocation data, containing the bitmap
* that controls these allocations, is itself stored in the heap, and is now
* structured to include both the bit data and the list of meta-storage
* addresses.
*
* Sizing: 256 allocators would reference approximately 2M objects/allocations.
* At 1K per allocator this would require 250K of store. The meta-allocation
* data would therefore need a start address plus 32 bytes (or 8 ints) to
* represent the meta-allocation bits. An array of such data referencing
* sequentially allocated storage areas completes the meta-allocation
* requirements.
*
* A meta-allocation address can therefore be represented as a single bit offset
* from which the block, providing start address, and bit offset can be directly
* determined.
*
* The m_metaBits int array used to be fully used as allocation bits, but now
* stores both the start address plus the 8 ints used to manage that data block.
*
* Allocation is reduced to sets of allocator objects which have a start address
* and a bitmap of allocated storage maps.
*
* Searching thousands of allocation blocks to find storage is not efficient,
* but by utilizing roving pointers and sorting blocks with free space available
* this can be made most efficient.
*
* In order to provide optimum use of bitmaps, this implementation will NOT use
* the BitSet class.
*
* Using the meta-allocation bits, it is straightforward to load ALL the
* allocation headers. A total of (say) 100 allocation headers might provide up
* to 4000 allocations each -> 400 000 objects, while 1000 headers -> 4m objects
* and 2000 -> 8m objects.
*
* The allocators are split into a set of FixedAllocators and then
* BlobAllocation. The FixedAllocators will allocate from 128 to 32K objects,
* with a minimum block allocation of 64K, and a minimum bit number per block of
* 32.
*
* Where possible lists and roving pointers will be used to minimize searching
* of the potentially large structures.
*
* Since the memory is allocated on (at least) a 128 byte boundary, there is
* some leeway on storing the address. Added to the address is the shift
* required to make to the "standard" 128 byte block, e.g. blocksize = 128 <<
* (addr % 8)
*
* NB Useful method on RandomAccessFile.setLength(newLength)
*
* When session data is preserved two things must happen - the allocators must
* not reallocate data that has been freed in this session, or more clearly can
* only free data that has been allocated in this session. That should be it.
*
* The ALLOC_SIZES table is the fibonacci sequence. We multiply by 64 bytes to
* get actual allocation block sizes. We then allocate bits based on 8K
* allocation rounding and 32 bits at a time allocation. Note that 4181 * 64 =
* 267,584 and 256K is 262,144
*
* All data is checksummed, both allocated/saved data and the allocation blocks.
*
* BLOB allocation is not handled using chained data buffers but with a blob
* header record. This is indicated with a BlobAllocator that provides indexed
* offsets to the header record (the address encodes the BlobAllocator and the
* offset to the address). The header record stores the number of component
* allocations and the address of each.
*
* This approach makes for much more efficient freeing/re-allocation of Blob
* storage, in particular avoiding the need to read in the component blocks to
* determine chained blocks for freeing. This is particularly important for
* larger stores where a disk cache could be flushed through simply freeing BLOB
* allocations.
*
* Deferred Free List
*
* The previous implementation has been amended to associate a single set of
* deferredFree blocks with each CommitRecord. The CommitRecordIndex will then
* provide access to the CommitRecords to support the deferred freeing of
* allocations based on age/earliestTxReleaseTime.
*
* The last release time processed is held with the MetaAllocation data
*
* @author Martyn Cutcher
*
* FIXME Release checklist:
*
* Add metabits header record checksum field and verify on read back.
*
* Done. Checksum fixed allocators (needs to be tested on read back).
*
* Done. Add version field to the fixed allocator.
*
* Done. Checksum delete blocks / blob records.
*
* PSOutputStream - remove caching logic. It is unused and makes this
* class much more complex. A separate per-RWStore caching class for
* recycling PSOutputStreams can be added later.
*
* Modify FixedAllocator to use arrayCopy() rather than clone and
* declare more fields to be final. See notes on {@link AllocBlock}.
*
* Done. Implement logic to "abort" a shadow allocation context.
*
* Unit test to verify that we do not recycle allocations from the last
* commit point even when the retention time is zero such that it is
* always possible to re-open the store from the alternative root block
* even after you have allocated things against the current root block
* (but not yet committed).
*
* Read-only mode.
*
* Unit tests looking for persistent memory leaks (e.g., all allocated
* space can be reclaimed).
*/
public class RWStore implements IStore, IBufferedWriter, IBackingReader {
private static final transient Logger log = Logger.getLogger(RWStore.class);
/**
* @see http://sourceforge.net/apps/trac/bigdata/ticket/443 (Logger for
* RWStore transaction service and recycler)
*/
private static final Logger txLog = Logger.getLogger("com.bigdata.txLog");
/**
* Options understood by the {@link RWStore}.
*/
public interface Options {
/**
* Option defines the Allocation block sizes for the RWStore. The values
* defined are multiplied by 64 to provide the actual allocations. The
* list of allocations should be ',' delimited and in increasing order.
* This array is written into the store so changing the values does not
* break older stores. For example,
*
*
* "1,2,4,8,116,32,64"
*
*
* defines allocations from 64 to 4K in size. It is a good to define
* block sizes on 4K boundaries as soon as possible to optimize IO. This
* is particularly relevant for SSDs. A 1K boundary is expressed as
* 16 in the allocation sizes, so a 4K boundary is
* expressed as 64 and an 8k boundary as 128.
*
* The default allocations are {@value #DEFAULT_ALLOCATION_SIZES}.
*
* @see #DEFAULT_ALLOCATION_SIZES
*/
String ALLOCATION_SIZES = RWStore.class.getName() + ".allocationSizes";
/**
* Note: The default allocation sizes SHOULD NOT provide for allocation
* slots larger than an 8k page. This can lead to large allocation slots
* when a B+Tree index is sparsely populated (less efficient prefix
* compression) followed by a gradual reduction in the average page size
* with the net effect that large allocators become unused and turn into
* wasted and unrecoverable space on the backing file. Keeping to an 8k
* maximum allocation slot size means that we have to do a few more IOs
* if the page exceeds the 8k boundary, but we never wind up with those
* large and (mostly) unused allocators. The B+Tree branching factors
* should be tuned to target perhaps 80% of an 8k page in order to have
* only a small number of pages that spill over into blobs.
*
* TODO: We should consider a more adaptable BLOB approach where we
* specify the maximum "slop" in an allocation as the means to determine
* a blob boundary. So, for example, a 5.5K allocation, with maximum slop of
* 1K, would be allocated as a blob of 4K + 2K and not an 8K slot.
*
* @see #ALLOCATION_SIZES
*/
String DEFAULT_ALLOCATION_SIZES = "1, 2, 3, 5, 8, 12, 16, 32, 48, 64, 128";
// String DEFAULT_ALLOCATION_SIZES = "1, 2, 3, 5, 8, 12, 16, 32, 48, 64, 128, 192, 320, 512, 832, 1344, 2176, 3520";
// String DEFAULT_ALLOCATION_SIZES = "1, 2, 3, 5, 8, 12, 16, 32, 48, 64, 128, 192, 320, 512";
// private static final int[] DEFAULT_ALLOC_SIZES = { 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, 233, 377, 610, 987, 1597, 2584, 4181 };
// private static final int[] ALLOC_SIZES = { 1, 2, 4, 8, 16, 32, 64, 128 };
/**
* Option defines the initial size of the meta bits region and effects
* how rapidly this region will grow (default
* {@value #DEFAULT_META_BITS_SIZE}).
*
* Note: A value of 9 may be used to stress the logic which
* is responsible for the growth in the meta bits region.
*
* This has now been deprecated since it adds complexity with no significant benefit
*/
@Deprecated String META_BITS_SIZE = RWStore.class.getName() + ".metaBitsSize";
@Deprecated String DEFAULT_META_BITS_SIZE = "9";
/**
* Defines whether the metabits should be allocated an explicit demispace (default)
* or if not, then to use a standard Allocation (which limits the metabits size to
* the maximum FixedAllocator slot size).
*
* The value should be either "true" or "false"
*/
String META_BITS_DEMI_SPACE = RWStore.class.getName() + ".metabitsDemispace";
String DEFAULT_META_BITS_DEMI_SPACE = "false";
/**
* Defines the number of bits that must be free in a FixedAllocator for
* it to be added to the free list. This is used to ensure a level
* of locality when making large numbers of allocations within a single
* commit.
*
* The value should be >= 1 and <= 5000
*/
String FREE_BITS_THRESHOLD = RWStore.class.getName() + ".freeBitsThreshold";
String DEFAULT_FREE_BITS_THRESHOLD = "300";
/**
* Defines the size of a slot that defines it as a small slot.
*
* Any slot equal to or less than this is considered a small slot and
* its availability for allocation is restricted to ensure a high
* chance that contiguous allocations can be made.
*
* This is arranged by only returning small slot allocators to the free list
* if they have greater than 50% available slots, and then only allocating
* slots from sparse regions with >= 50% free/committed bits.
*
* Small slot processing can be disabled by setting the smallSlotType to zero.
*/
String SMALL_SLOT_TYPE = RWStore.class.getName() + ".smallSlotType";
/**
* Enable the small slot optimization by default.
*
* @see BLZG-1596 (Enable small slot optimization by default)
*/
String DEFAULT_SMALL_SLOT_TYPE = "1024"; // standard default
// String DEFAULT_SMALL_SLOT_TYPE = "0"; // initial default to no special processing
/**
* The #of free bits required to be free in a "small slot" allocator before
* it is automatically returned to the free list. Once the small slot waste
* threshold comes into play, the small slot allocator for a given slot size
* having the maximum free bits will be automatically returned to the free
* list if the percentage of waste in that slot size exceeds a threshold.
*
* @see BLZG-1278 (Implement maximum waste policy for small slot allocators)
*/
String SMALL_SLOT_THRESHOLD = RWStore.class.getName() + ".smallSlotThreshold";
String DEFAULT_SMALL_SLOT_THRESHOLD = "4096"; // 50% of available bits
/**
* We have introduced extra parameters to adjust allocator usage if we notice that
* a significant amount of storage is wasted.
*
* First we check how many allocators of a given slot size have been created. If
* above {@value #SMALL_SLOT_WASTE_CHECK_ALLOCATORS} then we look a little closer.
*
* We retrieve the allocation statistics and determine if the waste threshold is
* exceeded, as determined by {@link SMALL_SLOT_HIGH_WASTE}.
*
* If so, then we attempt to find an available allocator with more free bits as
* determined by {@link SMALL_SLOT_THRESHOLD_HIGH_WASTE}.
*
* @see BLZG-1278 (Implement maximum waste policy for small slot allocators)
*/
String SMALL_SLOT_WASTE_CHECK_ALLOCATORS = RWStore.class.getName() + ".smallSlotWasteCheckAllocators";
String DEFAULT_SMALL_SLOT_WASTE_CHECK_ALLOCATORS = "100"; // Check waste when more than 100 allocators
/**
* Once there are at least {@link #SMALL_SLOT_WASTE_CHECK_ALLOCATORS}
* for a given slot size, then the {@link #SMALL_SLOT_HIGH_WASTE}
* specifies the maximum percentage of waste that will be allowed for
* that slot size. This prevents the amount of waste for small slot
* allocators from growing significantly as the size of the backing
* store increases.
*
* The dynamic policy for small slots can be thought of as follows.
*
*
A normal allocator will be dropped onto the free list once it has
* {@link #FREE_BITS_THRESHOLD} bits free (default 300 bits out of 8192
* = 3.6%).
* For a new store, a small slot allocator will be dropped onto the
* free list once it has {@link #SMALL_SLOT_THRESHOLD} bits free
* (default 4096 bits out of 8192 = 50%).
* Once the #of small slots allocators for a given sized allocator
* exceeds the {@link #DEFAULT_SMALL_SLOT_WASTE_CHECK_ALLOCATORS}, a
* small slot allocator will be dropped onto the free list once it is
* {@link #SMALL_SLOT_HIGH_WASTE} percent free (this amounts to 1638
* bits out of 8192).
*
* Thus, the small slot allocators initially are created freely because
* they need to be highly sparse before they can be on the free list.
* Once we have "enough" small slot allocators, we create them less
* freely - this is achieved by changing the sparsity threshold to a
* value that still requires the small slot allocator to be
* significantly more sparse than a general purpose allocator.
*
* @see BLZG-1278 (Implement maximum waste policy for small slot
* allocators)
*/
String SMALL_SLOT_HIGH_WASTE = RWStore.class.getName() + ".smallSlotHighWaste";
String DEFAULT_SMALL_SLOT_HIGH_WASTE = "20.0f"; // 1638 bits: 20% waste, less than 80% usage
/**
* When true, scattered writes which are strictly ascending
* will be coalesced within a buffer and written out as a single IO
* (default {@value #DEFAULT_DOUBLE_BUFFER_WRITES}). This improves write
* performance for SATA, SAS, and even SSD.
*/
String DOUBLE_BUFFER_WRITES = RWStore.class.getName() + ".doubleBuffer";
String DEFAULT_DOUBLE_BUFFER_WRITES = "true";
// /**
// * When true fills recycled storage with a recognizable
// * byte pattern.
// */
// String OVERWRITE_DELETE = RWStore.class.getName() + ".overwriteDelete";
//
// String DEFAULT_OVERWRITE_DELETE = "false";
//
// /**
// * When true the RWStore will protect any address from
// * recycling, and generate an exception if the address is subsequently
// * accessed
// */
// String MAINTAIN_BLACKLIST = RWStore.class.getName() + ".maintainBlacklist";
//
// String DEFAULT_MAINTAIN_BLACKLIST = "false";
}
/*
* Error messages.
*/
private static final String ERR_WRITE_CACHE_CREATE = "Unable to create write cache service";
/**
* The fixed size of any allocator on the disk in bytes. The #of allocations
* managed by an allocator is this value times 8 because each slot uses one
* bit in the allocator. When an allocator is allocated, the space on the
* persistent heap is reserved for all slots managed by that allocator.
* However, the {@link FixedAllocator} only incrementally allocates the
* {@link AllocBlock}s.
*/
static private final int ALLOC_BLOCK_SIZE = 1024;
// // from 32 bits, need 13 to hold max offset of 8 * 1024, leaving 19 for number of blocks: 256K
// static final int BLOCK_INDEX_BITS = 19;
/**
* The #of low bits in a latched address that encode the offset of the bit
* in a {@link FixedAllocator}. The {@link FixedAllocator} will map the bit
* onto an allocation slot.
*
* The high bits of the latched address is index of the
* {@link FixedAllocator}. The index of the {@link FixedAllocator} is the
* order in which it was created. This is used to index into
* {@link #m_allocs}, which are the {@link FixedAllocator}s.
*/
static final int OFFSET_BITS = 13;
static final int OFFSET_BITS_MASK = 0x1FFF; // was 0xFFFF
static final int ALLOCATION_SCALEUP = 16; // multiplier to convert allocations based on minimum allocation of 64k
static private final int META_ALLOCATION = 8; // 8 * 32K is size of meta Allocation
// If required, then allocate 1M direct buffers
private static final int cDirectBufferCapacity = 1024 * 1024;
private int cMaxDirectBuffers = 20; // 20M of direct buffers
static final int cDirectAllocationOffset = 64 * 1024;
// ///////////////////////////////////////////////////////////////////////////////////////
// RWStore Data
// ///////////////////////////////////////////////////////////////////////////////////////
private final File m_fd;
// private RandomAccessFile m_raf;
// protected FileMetadata m_metadata;
// protected int m_transactionCount;
// private boolean m_committing;
// /**
// * When true the allocations will not actually be recycled
// * until after a store restart. When false, the allocations are
// * recycled once they satisfy the history retention requirement.
// */
// private boolean m_preserveSession = false;
// private boolean m_readOnly;
/**
* The UUID of the backing store.
*
* @see #initfromRootBlock(IRootBlockView)
* @see IRawStore#getUUID()
*/
private UUID m_storeUUID;
/**
* lists of total alloc blocks.
*
* @todo examine concurrency and lock usage for {@link #m_alloc} and the
* rest of these lists.
*/
private final ArrayList m_allocs;
/**
* A fixed length array of lists of free {@link FixedAllocator}s with one
* entry in the array for each configured allocator size. An allocator is
* put onto this free list when it is initially created. When the store is
* opened, it will be added to this list if {@link Allocator#hasFree()}
* returns true. It will be removed when it has no free space remaining. It
* will be added back to the free list when its free slots exceeds a
* configured threshold.
*/
private ArrayList m_freeFixed[];
// /** lists of free blob allocators. */
// private final ArrayList m_freeBlobs;
/** lists of blocks requiring commitment. */
// private final ArrayList m_commitList;
FixedAllocator m_commitHead;
FixedAllocator m_commitTail;
// private WriteBlock m_writes;
private final Quorum m_quorum;
/**
* The #of buffers that will be used by the {@link WriteCacheService}.
*
* @see com.bigdata.journal.Options#WRITE_CACHE_BUFFER_COUNT
*/
private final int m_writeCacheBufferCount;
/**
* @see com.bigdata.journal.Options#WRITE_CACHE_MIN_CLEAN_LIST_SIZE
*/
private final int m_minCleanListSize;
/**
* The #of read buffers that will be used by the {@link WriteCacheService}.
*
* @see com.bigdata.journal.Options#READ_CACHE_BUFFER_COUNT
*/
private final int m_readCacheBufferCount;
/**
* @see com.bigdata.journal.Options#WRITE_CACHE_COMPACTION_THRESHOLD
*/
private final int m_compactionThreshold;
/**
* @see com.bigdata.journal.Options#HOT_CACHE_THRESHOLD
*/
private final int m_hotCacheThreshold;
/**
* @see com.bigdata.journal.Options#HOT_CACHE_SIZE
*/
private final int m_hotCacheSize;
/**
* The key for the {@link CompressorRegistry} which identifies the
* {@link IRecordCompressor} to be applied (optional).
*
* @see com.bigdata.journal.Options#HALOG_COMPRESSOR
*/
private final String m_compressorKey;
/**
* Note: This is not final because we replace the {@link WriteCacheService}
* during {@link #reset(long)} in order to propagate the then current quorum
* token to the {@link WriteCacheService}.
*/
private RWWriteCacheService m_writeCacheService;
/**
* Return the then current {@link WriteCacheService} object.
*
* @see IHABufferStrategy#getWriteCacheService()
*/
public RWWriteCacheService getWriteCacheService() {
m_allocationReadLock.lock();
try {
return m_writeCacheService;
} finally {
m_allocationReadLock.unlock();
}
}
/**
* The actual allocation sizes as read from the store.
*
* @see #DEFAULT_ALLOCATION_SIZES
*/
private int[] m_allocSizes;
/**
* The maximum allocation size (bytes).
*/
final int m_maxFixedAlloc;
/**
* The minimum allocation size (bytes).
*/
final int m_minFixedAlloc;
/**
* We allow blob headers so the maximum blob size is Integer.MAX_VALUE.
*/
final int m_maxBlobAllocSize = Integer.MAX_VALUE;
/**
* This lock is used to exclude readers/writers performing IOs against the
* backing file when the extent of the backing file is about to be changed.
* Readers and writers take the {@link ReadLock}. The {@link WriteLock} is
* taken when the file extent must be changed. This is a workaround for an
* old (an unresolved as of February 2010) Sun bug.
*
* Note: Any public method that ONLY takes the extensionLock MUST NOT make
* calls that could take the {@link #m_allocationLock}. This would cause a
* lock ordering problem. If both locks must be taken, then the
* {@link #m_allocationLock} MUST be taken first.
*
* @see http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6371642
* @see #m_allocationLock
*/
final private ReentrantReadWriteLock m_extensionLock = new ReentrantReadWriteLock();
/**
* An explicit allocation lock supports exclusive access for allocator
* mutation and shared access for readers.
*
* Note: You must hold the {@link #m_allocationReadLock} to read the
* allocators.
*
* Note: You must hold the {@link #m_allocationWriteLock} while allocating
* or clearing allocations.
*
* Note: It is only when an allocation triggers a file extension that the
* {@link WriteLock} of the {@link #m_extensionLock} needs to be taken.
*
* TODO: There is scope to take advantage of the different allocator sizes
* and provide allocation locks on the fixed allocators. We will still need
* a store-wide allocation lock when creating new allocation areas, but
* significant contention may be avoided.
*/
final private ReentrantReadWriteLock m_allocationLock = new ReentrantReadWriteLock();
/**
* Lock used for exclusive access to the allocators.
*
* Note: Historically, this lock was only required for mutation and readers
* did not content for a lock.
*/
final private WriteLock m_allocationWriteLock = m_allocationLock.writeLock();
/**
* Lock used for shared access to allocators.
*
* Note: Historically the allocators were unprotected for shared acccess
* (readers) and protected by a single lock for mutation (writes). Shared
* access by readers was safe since (a) old allocators were never replaced;
* and (b) readers had access only to committed data.
*
* This situation was changed when the {@link #postHACommit(IRootBlockView)}
* method was introduced since it could replace allocators in a manner that
* was not safe for shared access by readers. Methods that were historically
* using unprotected shared access now require protected shared access using
* this lock.
*
* @see #postHACommit(IRootBlockView)
* @see #getData(long, int)
* @see #getData(long, byte[])
* @see #getData(long, byte[], int, int)
*/
final private ReadLock m_allocationReadLock = m_allocationLock.readLock();
/**
* The deferredFreeList is simply an array of releaseTime,freeListAddrs
* stored at commit.
*
* Note that when the deferredFreeList is saved, ONLY thefreeListAddrs
* are stored, NOT the releaseTime. This is because on any open of
* the store, all deferredFrees can be released immediately. This
* mechanism may be changed in the future to enable explicit history
* retention, but if so a different header structure would be used since
* it would not be appropriate to retain a simple header linked to
* thousands if not millions of commit points.
*/
// *
// * If the current txn list exceeds the MAX_DEFERRED_FREE then it is
// * incrementally saved and a new list begun. The master list itself
// * serves as a BLOB header when there is more than a single entry with
// * the same txReleaseTime.
// private static final int MAX_DEFERRED_FREE = 4094; // fits in 16k block
private final long m_minReleaseAge;
/**
* The #of open transactions (read-only or read-write).
*
* This is guarded by the {@link #m_allocationLock}.
*/
private int m_activeTxCount = 0;
private volatile long m_lastDeferredReleaseTime = 0L;
// private final ArrayList m_currentTxnFreeList = new ArrayList();
private final PSOutputStream m_deferredFreeOut;
/**
* Used to transparently re-open the backing channel if it has been closed
* by an interrupt during an IO.
*/
private final ReopenFileChannel m_reopener;
private volatile BufferedWrite m_bufferedWrite;
/**
* Our StoreageStats objects
*/
private StorageStats m_storageStats;
private long m_storageStatsAddr = 0;
/**
* true iff the backing store is open.
*/
private volatile boolean m_open = true;
// /**
// * If m_blacklist is non-null then a request to blacklist as address will
// * add the address to the blacklist.
// *
// * When a blacklisted address is freed and is re-allocated, the re-allocation
// * is intercepted (see alloc()), the address is locked and a new allocation is made.
// *
// * The purpose of the blacklist is to trap erroneus references to an
// * address that is retained (and used) after it should be.
// */
// private ConcurrentHashMap m_blacklist = null;
private ConcurrentHashMap m_lockAddresses = null;
class WriteCacheImpl extends WriteCache.FileChannelScatteredWriteCache {
final private String compressorKey;
public WriteCacheImpl(final IBufferAccess buf,
final boolean useChecksum,
final boolean bufferHasData,
final IReopenChannel opener,
final long fileExtent, final String compressorKey)
throws InterruptedException {
super(buf, useChecksum, m_quorum != null
/*&& m_quorum.isHighlyAvailable()*/, bufferHasData, opener,
fileExtent,
m_bufferedWrite);
this.compressorKey = compressorKey;
}
@Override
public String getCompressorKey() {
return compressorKey;
}
/**
* {@inheritDoc}
*
* Note: The performance counters for writes to the disk are reported by
* the {@link WriteCacheService}. The {@link RWStore} never writes
* directly onto the disk (other than the root blocks).
*/
@Override
protected boolean writeOnChannel(final ByteBuffer data,
final long firstOffsetignored,
final Map recordMap,
final long nanos) throws InterruptedException, IOException {
final Lock readLock = m_extensionLock.readLock();
readLock.lock();
try {
boolean ret = super.writeOnChannel(data, firstOffsetignored,
recordMap, nanos);
return ret;
} finally {
readLock.unlock();
}
}
// Added to enable debug of rare problem
// FIXME: disable by removal once solved
protected void registerWriteStatus(long offset, int length, char action) {
m_writeCacheService.debugAddrs(offset, length, action);
}
@Override
protected void addAddress(int latchedAddr, int size) {
// No longer valid
// RWStore.this.addAddress(latchedAddr, size);
}
@Override
protected void removeAddress(int latchedAddr) {
// No longer valid
// RWStore.this.removeAddress(latchedAddr);
}
};
/**
* The ALLOC_SIZES must be initialized from either the file or the
* properties associated with the fileMetadataView
*
* @param fileMetadataView
* @param readOnly
* @param quorum
* @throws InterruptedException
*
* @todo support read-only open.
*/
public RWStore(final FileMetadata fileMetadata, final Quorum quorum) {
if (fileMetadata == null)
throw new IllegalArgumentException();
this.m_minReleaseAge = Long.valueOf(fileMetadata.getProperty(
AbstractTransactionService.Options.MIN_RELEASE_AGE,
AbstractTransactionService.Options.DEFAULT_MIN_RELEASE_AGE));
if (log.isInfoEnabled())
log.info(AbstractTransactionService.Options.MIN_RELEASE_AGE + "="
+ m_minReleaseAge);
// Remove parameterisation, we want to use fixed Allocator block sizing
// there is no significant advantage to parameterize this since file cache
// locality is handled by size of the allocation - 256K is a reasonable
// number as 32 * 8 * 1K size.
//
// Equally there is no benefit to increasing the size of the Allocators beyond
// 1K.
// cDefaultMetaBitsSize = Integer.valueOf(fileMetadata.getProperty(
// Options.META_BITS_SIZE,
// Options.DEFAULT_META_BITS_SIZE));
// cDefaultMetaBitsSize = 9;
// if (cDefaultMetaBitsSize < 9)
// throw new IllegalArgumentException(Options.META_BITS_SIZE
// + " : Must be GTE 9");
m_metaBitsSize = cDefaultMetaBitsSize;
m_useMetabitsDemispace = Boolean.valueOf(fileMetadata.getProperty(
Options.META_BITS_DEMI_SPACE,
Options.DEFAULT_META_BITS_DEMI_SPACE));
cDefaultFreeBitsThreshold = Integer.valueOf(fileMetadata.getProperty(
Options.FREE_BITS_THRESHOLD,
Options.DEFAULT_FREE_BITS_THRESHOLD));
if (cDefaultFreeBitsThreshold < 1 || cDefaultFreeBitsThreshold > 5000) {
throw new IllegalArgumentException(Options.FREE_BITS_THRESHOLD
+ " : Must be between 1 and 5000");
}
cSmallSlot = Integer.valueOf(fileMetadata.getProperty(
Options.SMALL_SLOT_TYPE,
Options.DEFAULT_SMALL_SLOT_TYPE));
cSmallSlotThreshold = Integer.valueOf(fileMetadata.getProperty(
Options.SMALL_SLOT_THRESHOLD,
Options.DEFAULT_SMALL_SLOT_THRESHOLD));
cSmallSlotWasteCheckAllocators = Integer.valueOf(fileMetadata.getProperty(
Options.SMALL_SLOT_WASTE_CHECK_ALLOCATORS,
Options.DEFAULT_SMALL_SLOT_WASTE_CHECK_ALLOCATORS));
cSmallSlotHighWaste = Float.valueOf(fileMetadata.getProperty(
Options.SMALL_SLOT_HIGH_WASTE,
Options.DEFAULT_SMALL_SLOT_HIGH_WASTE));
// cSmallSlotThresholdHighWaste = Integer.valueOf(fileMetadata.getProperty(
// Options.SMALL_SLOT_THRESHOLD_HIGH_WASTE,
// Options.DEFAULT_SMALL_SLOT_THRESHOLD_HIGH_WASTE));
/*
* The highWasteThreshold is more sensibly calculated from
* the high waste value.
*/
cSmallSlotThresholdHighWaste = (int) (cSmallSlotHighWaste * 8192 / 100);
if (cSmallSlot < 0 || cSmallSlot > 2048) {
throw new IllegalArgumentException(Options.SMALL_SLOT_TYPE
+ " : Must be between 0 and 2048");
}
m_metaBits = new int[m_metaBitsSize];
m_metaTransientBits = new int[m_metaBitsSize];
m_quorum = quorum;
m_fd = fileMetadata.file;
// initialize striped performance counters for this store.
this.storeCounters.set(new StoreCounters(10/* batchSize */));
final IRootBlockView m_rb = fileMetadata.rootBlock;
m_allocs = new ArrayList();
// m_freeBlobs = new ArrayList();
try {
final RandomAccessFile m_raf = fileMetadata.getRandomAccessFile();
m_reopener = new ReopenFileChannel(m_fd, m_raf, "rw");
} catch (IOException e1) {
throw new RuntimeException(e1);
}
if (Boolean.valueOf(fileMetadata.getProperty(
Options.DOUBLE_BUFFER_WRITES,
Options.DEFAULT_DOUBLE_BUFFER_WRITES))) {
try {
m_bufferedWrite = new BufferedWrite(this);
} catch (InterruptedException e1) {
m_bufferedWrite = null;
}
} else {
m_bufferedWrite = null;
}
m_writeCacheBufferCount = fileMetadata.writeCacheBufferCount;
m_readCacheBufferCount = Integer.valueOf(fileMetadata.getProperty(
com.bigdata.journal.Options.READ_CACHE_BUFFER_COUNT,
com.bigdata.journal.Options.DEFAULT_READ_CACHE_BUFFER_COUNT));
if (log.isInfoEnabled())
log.info(com.bigdata.journal.Options.WRITE_CACHE_BUFFER_COUNT
+ "=" + m_writeCacheBufferCount);
this.m_minCleanListSize = Integer.valueOf(fileMetadata.getProperty(
com.bigdata.journal.Options.WRITE_CACHE_MIN_CLEAN_LIST_SIZE,
com.bigdata.journal.Options.DEFAULT_WRITE_CACHE_MIN_CLEAN_LIST_SIZE));
if (log.isInfoEnabled())
log.info(com.bigdata.journal.Options.WRITE_CACHE_MIN_CLEAN_LIST_SIZE + "="
+ m_minCleanListSize);
this.m_compactionThreshold = Double.valueOf(fileMetadata.getProperty(
com.bigdata.journal.Options.WRITE_CACHE_COMPACTION_THRESHOLD,
com.bigdata.journal.Options.DEFAULT_WRITE_CACHE_COMPACTION_THRESHOLD)).intValue();
if (log.isInfoEnabled())
log.info(com.bigdata.journal.Options.WRITE_CACHE_COMPACTION_THRESHOLD + "="
+ m_compactionThreshold);
this.m_hotCacheThreshold = Double.valueOf(fileMetadata.getProperty(
com.bigdata.journal.Options.HOT_CACHE_THRESHOLD,
com.bigdata.journal.Options.DEFAULT_HOT_CACHE_THRESHOLD)).intValue();
if (log.isInfoEnabled())
log.info(com.bigdata.journal.Options.HOT_CACHE_THRESHOLD + "="
+ m_hotCacheThreshold);
this.m_hotCacheSize = Double.valueOf(fileMetadata.getProperty(
com.bigdata.journal.Options.HOT_CACHE_SIZE,
com.bigdata.journal.Options.DEFAULT_HOT_CACHE_SIZE)).intValue();
if (log.isInfoEnabled())
log.info(com.bigdata.journal.Options.HOT_CACHE_SIZE + "="
+ m_hotCacheSize);
this.m_compressorKey = fileMetadata.getProperty(
com.bigdata.journal.Options.HALOG_COMPRESSOR,
com.bigdata.journal.Options.DEFAULT_HALOG_COMPRESSOR);
if (log.isInfoEnabled())
log.info(com.bigdata.journal.Options.HALOG_COMPRESSOR + "="
+ m_compressorKey);
// m_writeCache = newWriteCache();
try {
if (m_rb.getNextOffset() == 0) { // if zero then new file
setAllocations(fileMetadata);
/*
* FIXME Martyn, the code paths here are crazy complicated.
* defaultInit() is also invoked from initFromRootBlock().
* Simplify this. BBT
*/
m_storeUUID = m_rb.getUUID();
defaultInit();
m_maxFixedAlloc = m_allocSizes[m_allocSizes.length-1]*64;
m_minFixedAlloc = m_allocSizes[0]*64;
m_storageStats = new StorageStats(m_allocSizes);
// // Check for overwrite option and set overwrite buffer if
// // required
// if (Boolean.valueOf(fileMetadata.getProperty(
// Options.OVERWRITE_DELETE,
// Options.DEFAULT_OVERWRITE_DELETE))) {
// m_writeCache.setOverwriteBuffer(m_maxFixedAlloc);
// }
} else {
initfromRootBlock(m_rb);
m_maxFixedAlloc = m_allocSizes[m_allocSizes.length-1]*64;
m_minFixedAlloc = m_allocSizes[0]*64;
if (m_storageStatsAddr != 0) {
final long statsAddr = m_storageStatsAddr >> 16;
final int statsLen = ((int) m_storageStatsAddr) & 0xFFFF;
final byte[] stats = new byte[statsLen + 4]; // allow for checksum
getData(statsAddr, stats);
final DataInputStream instr = new DataInputStream(new ByteArrayInputStream(stats));
m_storageStats = new StorageStats(instr);
for (FixedAllocator fa: m_allocs) {
m_storageStats.register(fa);
}
} else {
m_storageStats = new StorageStats(m_allocSizes);
}
if (log.isTraceEnabled()) {
final StringBuilder str = new StringBuilder();
this.showAllocators(str);
log.trace(str);
}
}
// Maximum theoretically addressable file size is determined by the
// maximum allocator slot size multiplied by Integer.MAX_VALUE
// FIXME: do we want to constrain this as a system property?
m_maxFileSize = ((long) Integer.MAX_VALUE) * m_maxFixedAlloc;
// setup write cache AFTER init to ensure filesize is correct!
m_writeCacheService = newWriteCacheService();
final int maxBlockLessChk = m_maxFixedAlloc-4;
assert m_maxFixedAlloc > 0;
m_deferredFreeOut = PSOutputStream.getNew(this, m_maxFixedAlloc, null);
// if (Boolean.valueOf(fileMetadata.getProperty(
// Options.MAINTAIN_BLACKLIST,
// Options.DEFAULT_MAINTAIN_BLACKLIST))) {
// m_blacklist = new ConcurrentHashMap();
// m_lockAddresses = new ConcurrentHashMap();
// }
} catch (IOException e) {
throw new StorageTerminalError("Unable to initialize store", e);
}
}
/**
* Called from WriteCache.resetRecordMapFromBuffer
*
* If a FixedAllocator already exists for the address then just set the
* address as active, otherwise, create a new allocator and try again, which
* should work second time around if we are correctly in sync.
*
* @param latchedAddr
* The latched address.
* @param size
* The size of the application data -or- -size if
* this provides notice of the existence of an allocator for that
* latchedAddr but the address itself should not yet be
* allocated.
*/
void addAddress(final int latchedAddr, final int size) {
// ignore zero address
if (latchedAddr == 0)
return;
m_allocationWriteLock.lock();
try {
FixedAllocator alloc = null;
try {
alloc = getBlock(latchedAddr);
} catch (final PhysicalAddressResolutionException par) {
// Must create new allocator
}
final int size2 = size < 0 ? -size : size;
if (alloc == null) {
final int i = fixedAllocatorIndex(size2);
final int block = 64 * m_allocSizes[i];
final ArrayList list = m_freeFixed[i];
if (log.isTraceEnabled())
log.trace("Creating new Allocator for address: "
+ latchedAddr);
final FixedAllocator allocator = new FixedAllocator(this, block);
allocator.setFreeList(list);
allocator.setIndex(m_allocs.size());
m_allocs.add(allocator);
// Check correctly synchronized creation
assert allocator == getBlock(latchedAddr);
alloc = allocator;
}
assert size2 <= alloc.getSlotSize();
if (size > 0) {
/*
* This is a real allocation.
*/
alloc.setAddressExternal(latchedAddr);
}
} finally {
m_allocationWriteLock.unlock();
}
}
/**
* Called from WriteCache.resetRecordMapFromBuffer
*
* Must clear the bit in the allocator.
*
* @param latchedAddr
*/
void removeAddress(final int latchedAddr) {
// ignore zero address
if (latchedAddr == 0)
return;
m_allocationWriteLock.lock();
try {
// assert m_commitList.size() == 0;
final FixedAllocator alloc = getBlockByAddress(latchedAddr);
assert alloc != null;
final int addrOffset = getOffset(latchedAddr);
if (alloc == null) {
throw new IllegalArgumentException(
"Invalid address provided to immediateFree: "
+ latchedAddr);
}
final long pa = alloc.getPhysicalAddress(addrOffset);
if (log.isTraceEnabled())
log.trace("Freeing allocation at " + latchedAddr
+ ", physical address: " + pa);
alloc.free(latchedAddr, 0, false);
// assert m_commitList.size() == 0;
} finally {
m_allocationWriteLock.unlock();
}
}
/**
* Create and return a new {@link RWWriteCacheService} instance. The caller
* is responsible for closing out the old one and must be holding the
* appropriate locks when it switches in the new instance.
*/
private RWWriteCacheService newWriteCacheService() {
try {
// final boolean highlyAvailable = m_quorum != null
// && m_quorum.isHighlyAvailable();
final boolean prefixWrites = m_quorum != null; // highlyAvailable
return new RWWriteCacheService(m_writeCacheBufferCount,
m_minCleanListSize, m_readCacheBufferCount, prefixWrites, m_compactionThreshold, m_hotCacheSize, m_hotCacheThreshold,
convertAddr(m_fileSize), m_reopener, m_quorum, this) {
@Override
@SuppressWarnings("unchecked")
public WriteCache newWriteCache(final IBufferAccess buf,
final boolean useChecksum,
final boolean bufferHasData,
final IReopenChannel opener,
final long fileExtent)
throws InterruptedException {
return new WriteCacheImpl(buf,
useChecksum, bufferHasData,
(IReopenChannel) opener,
fileExtent, m_compressorKey);
}
};
} catch (InterruptedException e) {
throw new IllegalStateException(ERR_WRITE_CACHE_CREATE, e);
} catch (IOException e) {
throw new IllegalStateException(ERR_WRITE_CACHE_CREATE, e);
}
}
private void setAllocations(final FileMetadata fileMetadata)
throws IOException {
final String buckets = fileMetadata.getProperty(
Options.ALLOCATION_SIZES, Options.DEFAULT_ALLOCATION_SIZES);
final String[] specs = buckets.split("\\s*,\\s*");
m_allocSizes = new int[specs.length];
int prevSize = 0;
for (int i = 0; i < specs.length; i++) {
final int nxtSize = Integer.parseInt(specs[i]);
if (nxtSize <= prevSize)
throw new IllegalArgumentException(
"Invalid AllocSizes property");
m_allocSizes[i] = nxtSize;
prevSize = nxtSize;
}
}
private void defaultInit() throws IOException {
final int numFixed = m_allocSizes.length;
m_freeFixed = new ArrayList[numFixed];
for (int i = 0; i < numFixed; i++) {
m_freeFixed[i] = new ArrayList();
}
m_fileSize = convertFromAddr(m_fd.length());
// make space for meta-allocators
m_metaBits[0] = -1;
m_metaTransientBits[0] = -1;
m_nextAllocation = -(1 + META_ALLOCATION); // keep on a minimum 8K boundary
m_committedNextAllocation = m_nextAllocation;
if (m_fileSize > m_nextAllocation) {
m_fileSize = m_nextAllocation;
}
if (log.isInfoEnabled())
log.info("Set default file extent " + convertAddr(m_fileSize));
m_reopener.raf.setLength(convertAddr(m_fileSize));
}
public boolean isOpen() {
return m_open;
}
private void assertOpen() {
if (!m_open)
throw new IllegalStateException(AbstractBufferStrategy.ERR_NOT_OPEN);
}
synchronized public void close() {
m_open = false;
try {
if (m_bufferedWrite != null) {
m_bufferedWrite.release();
m_bufferedWrite = null;
}
m_writeCacheService.close();
m_reopener.raf.close();
} catch (Throwable t) {
throw new RuntimeException(t);
}
}
/**
* Basic check on key root block validity
*
* @param rbv
*/
private void checkRootBlock(final IRootBlockView rbv) {
final long nxtOffset = rbv.getNextOffset();
final int nxtalloc = -(int) (nxtOffset >> 32);
final int metaBitsAddr = -(int) nxtOffset;
final long metaAddr = rbv.getMetaStartAddr();
final long rawMetaBitsAddr = rbv.getMetaBitsAddr();
if (metaAddr == 0 || rawMetaBitsAddr == 0) {
/*
* possible when rolling back to empty file.
*/
log.warn("No meta allocation data included in root block for RWStore");
}
// CANNOT check physicalAddress if follower
if (m_quorum == null && log.isTraceEnabled()) {
final int commitRecordAddr = (int) (rbv.getCommitRecordAddr() >> 32);
log.trace("CommitRecord " + rbv.getCommitRecordAddr()
+ " at physical address: "
+ physicalAddress(commitRecordAddr));
}
final long commitCounter = rbv.getCommitCounter();
// final int metaStartAddr = (int) -(metaAddr >> 32); // void
// final int fileSize = (int) -(metaAddr & 0xFFFFFFFF);
if (log.isTraceEnabled())
log.trace("m_allocation: " + nxtalloc + ", m_metaBitsAddr: "
+ metaBitsAddr + ", m_commitCounter: " + commitCounter);
}
/**
* Utility to encapsulate RootBlock interpretation.
*/
static private class RootBlockInfo {
// int nextAllocation(final IRootBlockView rb) {
// final long nxtOffset = rb.getNextOffset();
//
// // next allocation to be made (in -32K units).
// final int ret = -(int) (nxtOffset >> 32);
//
// /*
// * Skip the first 32K in the file. The root blocks live here but
// * nothing else.
// */
// return ret == 0 ? -(1 + META_ALLOCATION) : ret;
// }
/**
* Used to transparently re-open the backing channel if it has been closed
* by an interrupt during an IO.
*/
private final ReopenFileChannel m_reopener;
/**
* Meta-Allocations stored as {int address; int[8] bits}, so each block
* holds 8*32=256 allocation slots of 1K totaling 256K.
*
* The returned int array is a flattened list of these int[9] blocks
*/
private final int[] m_metabits;
private final long m_storageStatsAddr;
private final long m_lastDeferredReleaseTime;
RootBlockInfo(final IRootBlockView rb,
final ReopenFileChannel reopener) throws IOException {
this.m_reopener = reopener;
final long rawmbaddr = rb.getMetaBitsAddr();
/*
* The #of int32 values in the metabits region.
*
* We get this by taking bottom 16 bits of the metaBitsAddr. This
* gives the #of int32 values in the metabits regions (up to 64k
* int32 values).
*/
final int metaBitsStore = (int) (rawmbaddr & 0xFFFF);
// The byte offset of the metabits region in the file.
final long pmaddr = rawmbaddr >> 16;
/*
* Read the metabits block, including a header and the int32[]
* that encodes both startAddrs and bit vectors.
*/
final byte[] buf = new byte[metaBitsStore * 4];
FileChannelUtility.readAll(m_reopener, ByteBuffer.wrap(buf), pmaddr);
final DataInputStream strBuf = new DataInputStream(new ByteArrayInputStream(buf));
// Can handle minor store version incompatibility
strBuf.readInt(); // STORE VERSION
m_lastDeferredReleaseTime = strBuf.readLong(); // Last Deferred Release Time
strBuf.readInt(); // cDefaultMetaBitsSize
final int allocBlocks = strBuf.readInt();
m_storageStatsAddr = strBuf.readLong(); // m_storageStatsAddr
// step over those reserved ints
for (int i = 0; i < cReservedMetaBits; i++) {
strBuf.readInt();
}
// step over the allocSizes
for (int i = 0; i < allocBlocks; i++) {
strBuf.readInt();
}
final int metaBitsSize = metaBitsStore - allocBlocks - cMetaHdrFields; // allow for header fields
// Must be multiple of 9
assert metaBitsSize % 9 == 0;
final int[] ret = new int[metaBitsSize];
for (int i = 0; i < metaBitsSize; i++) {
ret[i] = strBuf.readInt();
}
/*
* Meta-Allocations stored as {int address; int[8] bits}, so each block
* holds 8*32=256 allocation slots of 1K totaling 256K.
*/
m_metabits = ret;
}
}
/**
* Should be called where previously initFileSpec was used.
*
* Rather than reading from file, instead reads from the current root block.
*
* We use the rootBlock fields, nextOffset, metaStartAddr, metaBitsAddr.
*
* metaBitsAddr indicates where the meta allocation bits are.
*
* metaStartAddr is the offset in the file where the allocation blocks are
* allocated the long value also indicates the size of the allocation, such
* that the address plus the size is the "filesize".
*
* Note that metaBitsAddr must be an absolute address, with the low order 16
* bits used to indicate the size.
*
* @throws IOException
*/
private void initfromRootBlock(final IRootBlockView rb) throws IOException {
// m_rb = m_fmv.getRootBlock();
assert(rb != null);
m_storeUUID = rb.getUUID();
if (rb.getNextOffset() == 0) {
defaultInit();
} else {
/*
* The RWStore stores in IRootBlock.getNextOffset() two distinct
* int32 words.
*
* The high int32 word is the next allocation that will handed out
* and is represented in units of -32K. This is used for things like
* getting a new metabits region or a new region from which fixed
* allocators will be recruited (through the metabits).
*
* The low int32 word is the latched address of the current metabits
* region. It must be interpreted using the metaBits and the
* FixedAllocators in order to turn it into a byte offset on the
* file.
*/
final long nxtOffset = rb.getNextOffset();
// next allocation to be made (in -32K units).
m_nextAllocation = -(int) (nxtOffset >> 32);
if (m_nextAllocation == 0) {
/*
* Skip the first 32K in the file. The root blocks live here but
* nothing else.
*/
m_nextAllocation = -(1 + META_ALLOCATION);
}
m_committedNextAllocation = m_nextAllocation;
// latched offset of the metabits region.
m_metaBitsAddr = -(int) nxtOffset;
if (log.isInfoEnabled()) {
log.info("MetaBitsAddr: " + m_metaBitsAddr);
}
/*
* Get the fileSize in -32K units from the root block.
*/
{
final long metaAddr = rb.getMetaStartAddr();
// in units of -32K.
m_fileSize = (int) -(metaAddr & 0xFFFFFFFF);
if (log.isInfoEnabled())
log.info("InitFromRootBlock m_fileSize: " + convertAddr(m_fileSize));
}
/*
* This stores the byte offset and length of the metabits region in
* the file. The bottom 16-bits are the length (see below). The top
* 48-bits are the byte offset.
*/
long rawmbaddr = rb.getMetaBitsAddr();
/*
* The #of int32 values in the metabits region.
*
* We get this by taking bottom 16 bits of the metaBitsAddr. This
* gives the #of int32 values in the metabits regions (up to 64k
* int32 values). Each int32 value in the metaBits[] gives us 32
* allocators. So, 16-bits gives us up 64k * 32 = 2M allocators.
* Except, that the total #of allocators is reduced by the presence
* of a startAddr every N positions in the metaBits[].
*
* The theoretical maximum number is also reduced since the number
* of "committed" bits could be half the total number of bits.
*
* The theoretical restriction is also limited by the maximum indexable
* allocator, since only 19 bits is available to the index, which, once
* the sign is removed reduces the maximum number of addressable
* allocators to 256K.
*/
final int metaBitsStore = (int) (rawmbaddr & 0xFFFF);
if (metaBitsStore > 0) {
// The byte offset of the metabits region in the file.
rawmbaddr >>= 16;
/*
* Read the metabits block, including a header and the int32[]
* that encodes both startAddrs and bit vectors.
*/
final byte[] buf = new byte[metaBitsStore * 4];
FileChannelUtility.readAll(m_reopener, ByteBuffer.wrap(buf), rawmbaddr);
final DataInputStream strBuf = new DataInputStream(new ByteArrayInputStream(buf));
// Can handle minor store version incompatibility
final int storeVersion = strBuf.readInt();
switch ((storeVersion & 0xFF00)) {
case (cVersion & 0xFF00):
case (cVersionDemispace & 0xFF00):
break;
default:
throw new IllegalStateException(
"Incompatible RWStore header version: storeVersion="
+ storeVersion + ", cVersion=" + cVersion + ", demispace: " + isUsingDemiSpace());
}
m_lastDeferredReleaseTime = strBuf.readLong();
if (strBuf.readInt() != cDefaultMetaBitsSize) {
throw new IllegalStateException("Store opened with unsupported metabits size");
}
final int allocBlocks = strBuf.readInt();
m_storageStatsAddr = strBuf.readLong();
// and let's read in those reserved ints
for (int i = 0; i < cReservedMetaBits; i++) {
strBuf.readInt();
}
m_allocSizes = new int[allocBlocks];
for (int i = 0; i < allocBlocks; i++) {
m_allocSizes[i] = strBuf.readInt();
}
m_metaBitsSize = metaBitsStore - allocBlocks - cMetaHdrFields; // allow for header fields
m_metaBits = new int[m_metaBitsSize];
if (log.isInfoEnabled()) {
log.info("Raw MetaBitsAddr: " + rawmbaddr);
}
for (int i = 0; i < m_metaBitsSize; i++) {
m_metaBits[i] = strBuf.readInt();
}
// m_metaTransientBits = (int[]) m_metaBits.clone();
syncMetaTransients();
final int numFixed = m_allocSizes.length;
m_freeFixed = new ArrayList[numFixed];
for (int i = 0; i < numFixed; i++) {
m_freeFixed[i] = new ArrayList();
}
checkCoreAllocations();
readAllocationBlocks();
}
if (log.isInfoEnabled())
log.info("restored from RootBlock: " + m_nextAllocation
+ ", " + m_metaBitsAddr);
}
}
/**
* Uses System.arraycopy rather than clone() to duplicate the
* metaBits to the metaTransientBits, which will be faster.
*/
private void syncMetaTransients() {
if (m_metaTransientBits == null || m_metaTransientBits.length != m_metaBits.length) {
m_metaTransientBits = (int[]) m_metaBits.clone();
} else {
System.arraycopy(m_metaBits, 0, m_metaTransientBits, 0, m_metaTransientBits.length);
}
}
// /*
// * Called when store is opened to make sure any deferred frees are
// * cleared.
// *
// * Stored persistently is only the list of addresses of blocks to be freed,
// * the knowledge of the txn release time does not need to be held persistently,
// * this is only relevant for transient state while the RWStore is open.
// *
// * The deferredCount is the number of entries - integer address and integer
// * count at each address
// */
// private void clearOutstandingDeferrels(final int deferredAddr, final int deferredCount) {
// if (deferredAddr != 0) {
// assert deferredCount != 0;
// final int sze = deferredCount * 8 + 4; // include space for checksum
//
// if (log.isDebugEnabled())
// log.debug("Clearing Outstanding Deferrals: " + deferredCount);
//
// byte[] buf = new byte[sze];
// getData(deferredAddr, buf);
//
// final byte[] blockBuf = new byte[8 * 1024]; // maximum size required
//
// ByteBuffer in = ByteBuffer.wrap(buf);
// for (int i = 0; i < deferredCount; i++) {
// int blockAddr = in.getInt();
// int addrCount = in.getInt();
//
// // now read in this block and free all addresses referenced
// getData(blockAddr, blockBuf, 0, addrCount*4 + 4);
// ByteBuffer inblock = ByteBuffer.wrap(blockBuf);
// for (int b = 0; b < addrCount; b++) {
// final int defAddr = inblock.getInt();
// Allocator alloc = getBlock(defAddr);
// if (alloc instanceof BlobAllocator) {
// b++;
// assert b < addrCount;
// alloc.free(defAddr, inblock.getInt());
// } else {
// alloc.free(defAddr, 0); // size ignored for FreeAllocators
// }
// }
// // once read then free the block allocation
// free(blockAddr, 0);
// }
//
// // lastly free the deferredAddr
// free(deferredAddr, 0);
// }
//
// }
/*********************************************************************
* make sure resource is closed!
**/
protected void finalize() {
close();
}
@SuppressWarnings("unchecked")
protected void readAllocationBlocks() throws IOException {
assert m_allocs.size() == 0;
if (log.isInfoEnabled())
log.info("readAllocationBlocks, m_metaBits.length: "
+ m_metaBits.length);
/**
* Allocators are sorted in StartAddress order (which MUST be the order
* they were created and therefore will correspond to their index) The
* comparator also checks for equality, which would indicate an error in
* the metaAllocation if two allocation blocks were loaded for the same
* address (must be two version of same Allocator).
*
* Meta-Allocations stored as {int address; int[8] bits}, so each block
* holds 8*32=256 allocation slots of 1K totaling 256K.
*/
for (int b = 0; b < m_metaBits.length; b += cDefaultMetaBitsSize) {
final long blockStart = convertAddr(m_metaBits[b]);
final int startBit = (b * 32) + 32;
final int endBit = startBit + ((cDefaultMetaBitsSize-1)*32);
for (int i = startBit; i < endBit; i++) {
if (tstBit(m_metaBits, i)) {
final long addr = blockStart + ((i-startBit) * ALLOC_BLOCK_SIZE);
final FixedAllocator allocator = readAllocator(addr);
allocator.setDiskAddr(i); // store bit, not physical address!
m_allocs.add(allocator);
if (m_storageStats != null) {
m_storageStats.register(allocator);
}
}
}
}
// add sorted blocks into index array and set index number for address
// encoding
// m_allocs.addAll(blocks);
Collections.sort(m_allocs);
for (int index = 0; index < m_allocs.size(); index++) {
((Allocator) m_allocs.get(index)).setIndex(index);
}
}
private FixedAllocator readAllocator(final long addr) throws IOException {
final byte buf[] = new byte[ALLOC_BLOCK_SIZE];
FileChannelUtility.readAll(m_reopener, ByteBuffer.wrap(buf), addr);
final ByteArrayInputStream baBuf = new ByteArrayInputStream(buf);
final DataInputStream strBuf = new DataInputStream(baBuf);
final int allocSize = strBuf.readInt(); // if Blob < 0
assert allocSize > 0;
final int slotSizeIndex = slotSizeIndex(allocSize);
if (slotSizeIndex == -1) {
throw new IllegalStateException("Unexpected allocation size of: " + allocSize);
}
final FixedAllocator fa = new FixedAllocator(this, allocSize);//, m_writeCache);
fa.read(strBuf);
final int chk = ChecksumUtility.getCHK().checksum(buf,
buf.length - baBuf.available());
int tstChk = strBuf.readInt();
if (tstChk != chk) {
throw new IllegalStateException("FixedAllocator checksum error");
}
if (slotSizeIndex == -1) {
throw new IllegalStateException("Unexpected allocation size of: " + allocSize);
}
final ArrayList freeList;
freeList = m_freeFixed[slotSizeIndex];
fa.setFreeList(freeList);
return fa;
}
/**
* Computes the slot size index given the absolute slot size.
*
* If the slotSizes are [1,2,4] this corresponds to absolute sizes by
* multiplying by 64 of [64, 128, 256], so slotSizeIndex(64) would return 0,
* and any parameter other than 64, 128 or 256 would return -1.
*
* @param allocSize - absolute slot size
* @return
*/
private int slotSizeIndex(final int allocSize) {
if (allocSize % 64 != 0)
return -1;
final int slotSize = allocSize / 64;
int slotSizeIndex = -1;
for (int index = 0; index < m_allocSizes.length; index++) {
if (m_allocSizes[index] == slotSize) {
slotSizeIndex = index;
break;
}
}
return slotSizeIndex;
}
/**
* Required for HA to support post commit message to synchronize allocators
* with new state. By this time the new allocator state will have been flushed
* to the disk, so should be 1) On disk, 2) Probably in OS cache and 3) Possibly
* in the WriteCache.
*
* For efficiency we do not want to default to reading from disk.
*
* If there is an existing allocator, then we can compare the old with the new state
* to determine which addresses have been freed and hence which addresses should be
* removed from the external cache.
*
* @param index of Alloctor to be updated
* @param addr on disk to be read
* @throws InterruptedException
* @throws ChecksumError
* @throws IOException
*/
private void updateFixedAllocator(final int index, final long addr) throws ChecksumError, InterruptedException, IOException {
final ByteBuffer buf = m_writeCacheService.read(addr, ALLOC_BLOCK_SIZE);
final ByteArrayInputStream baBuf = new ByteArrayInputStream(buf.array());
final DataInputStream strBuf = new DataInputStream(baBuf);
final int allocSize = strBuf.readInt(); // if Blob < 0
assert allocSize > 0;
final int slotIndex = slotSizeIndex(allocSize);
if (slotIndex == -1)
throw new IllegalStateException("Invalid allocation size: " + allocSize);
final FixedAllocator allocator = new FixedAllocator(this, allocSize);
final ArrayList freeList = m_freeFixed[slotIndex];
if (index < m_allocs.size()) {
final FixedAllocator old = m_allocs.get(index);
freeList.remove(old);
m_allocs.set(index, allocator);
allocator.setFreeList(freeList);
// Need to iterate over all allocated bits in "old" and see if they
// are clear in "new". If so then clear from externalCache
} else {
assert index == m_allocs.size();
m_allocs.add(allocator);
}
}
/**
* Called from ContextAllocation when no free FixedAllocator is immediately
* available. First the free list will be checked to see if one is
* available, otherwise it will be created. When the calling
* ContextAllocation is released, its allocators will be added to the
* global free lists.
*
* @param block - the index of the Fixed size allocation
* @return the FixedAllocator
*/
private FixedAllocator establishFreeFixedAllocator(final int block) {
final ArrayList list = m_freeFixed[block];
for (int i = 0; i < list.size(); i++) {
FixedAllocator f = list.get(i);
if (!isOnCommitList(f)) {
list.remove(i);
return f;
}
}
// no valid free allocators, so create a new one
final int allocSize = 64 * m_allocSizes[block];
final FixedAllocator allocator = new FixedAllocator(this,
allocSize);//, m_writeCache);
allocator.setIndex(m_allocs.size());
m_allocs.add(allocator);
if (m_storageStats != null) {
m_storageStats.register(allocator, true);
}
return allocator;
}
// // Root interface
// public long getRootAddr() {
// return m_rootAddr;
// }
//
// // Root interface
// public PSInputStream getRoot() {
// try {
// return getData(m_rootAddr);
// } catch (Exception e) {
// throw new StorageTerminalError("Unable to read root data", e);
// }
// }
//
// public void setRootAddr(long rootAddr) {
// m_rootAddr = (int) rootAddr;
// }
// // Limits
// public void setMaxFileSize(final int maxFileSize) {
// m_maxFileSize = maxFileSize;
// }
public long getMaxFileSize() {
return m_maxFileSize;
}
// // Allocators
// public PSInputStream getData(final long addr) {
// return getData((int) addr, addr2Size((int) addr));
// }
//
// // Allocators
// public PSInputStream getData(final int addr, final int size) {
// final Lock readLock = m_extensionLock.readLock();
//
// readLock.lock();
//
// try {
// try {
// m_writeCache.flush(false);
// } catch (InterruptedException e1) {
// throw new RuntimeException(e1);
// }
//
// if (addr == 0) {
// return null;
// }
//
// final PSInputStream instr = PSInputStream.getNew(this, size);
//
// try {
//// m_raf.seek(physicalAddress(addr));
//// m_raf.readFully(instr.getBuffer(), 0, size);
//// m_raf.getChannel().read(ByteBuffer.wrap(instr.getBuffer(), 0, size), physicalAddress(addr));
// FileChannelUtility.readAll(m_reopener, ByteBuffer.wrap(instr.getBuffer(), 0, size),
// physicalAddress(addr));
// } catch (IOException e) {
// throw new StorageTerminalError("Unable to read data", e);
// }
//
// return instr;
// } finally {
// readLock.unlock();
// }
// }
volatile private long m_cacheReads = 0;
volatile private long m_diskReads = 0;
volatile private int m_allocations = 0;
volatile private int m_frees = 0;
volatile private long m_nativeAllocBytes = 0;
/**
* Alternative method signature returning a ByteBuffer rather than receiving a
* byte array.
*
* If a blob then an extra byte array is required in which to build the data,
* but otherwise extra buffering could be avoided be reading directly from
* the WriteCacheService.
*
* @param rwaddr
* @param sze
* @return
*/
public ByteBuffer getData(final long rwaddr, final int sze) {
/*
* Note: Contend with postHACommit().
*/
final Lock lock = m_allocationReadLock;
lock.lock();
try {
// must allow for checksum
if (sze > (m_maxFixedAlloc-4) || m_writeCacheService == null) {
final byte buf[] = new byte[sze + 4]; // 4 bytes for checksum
getData(rwaddr, buf, 0, sze+4);
return ByteBuffer.wrap(buf, 0, sze);
} else {
final long paddr = physicalAddress((int) rwaddr);
if (paddr == 0) {
assertAllocators();
throw new PhysicalAddressResolutionException(rwaddr);
}
assert paddr > 0;
try {
return m_writeCacheService.read(paddr, sze+4);
} catch (Throwable e) {
/*
* Note: ClosedByInterruptException can be thrown out of
* FileChannelUtility.readAll(), typically because the LIMIT on
* a query was satisfied, but we do not want to log that as an
* error.
*/
// log.error(e,e);
throw new RuntimeException("addr=" + rwaddr + " : cause=" + e, e);
}
}
} finally {
lock.unlock();
}
}
/**
* If the buf[] size is greater than the maximum fixed allocation, then the
* direct read will be the blob header record. In this case we should hand
* over the streaming to a PSInputStream.
*
* FIXME: Javadoc update (was: For now we do not use the PSInputStream but instead process
* directly...)
*
* If it is a BlobAllocation, then the BlobAllocation address points to the
* address of the BlobHeader record.
*/
public void getData(final long addr, final byte buf[]) {
getData(addr, buf, 0, buf.length);
}
/*
* Set the option below to true to enable asynchronous reads of blob data.
* The aim is to reduce latency when reading blobs from disk as it will enable the disk controllers
* to re-order IO requests nd where possible process in parallel.
* This should benefit all Blob reads but specifically helps large deferredFree data to reduce commit latency
* as described in BLZG-1663.
*/
static boolean s_readBlobsAsync = true;
public void getData(final long addr, final byte buf[], final int offset,
final int length) {
assertOpen();
if (addr == 0) {
return;
}
final long begin = System.nanoTime();
/*
* Note: Contend with postHACommit().
*/
final Lock lock = m_allocationReadLock;
lock.lock();
try {
assertOpen(); // check again after taking lock
// assertNoRebuild();
// length includes space for the checksum
if (length > m_maxFixedAlloc) {
try {
final int alloc = m_maxFixedAlloc-4;
final int nblocks = (alloc - 1 + (length-4))/alloc;
if (nblocks < 0)
throw new IllegalStateException(
"Allocation error, m_maxFixedAlloc: "
+ m_maxFixedAlloc);
final byte[] hdrbuf = new byte[4 * (nblocks + 1) + 4]; // plus 4 bytes for checksum
if (hdrbuf.length > m_maxFixedAlloc) {
if (log.isInfoEnabled()) {
log.info("LARGE BLOB - header is BLOB");
}
}
getData(addr, hdrbuf); // will work even if header is also a blob
final DataInputStream hdrstr = new DataInputStream(new ByteArrayInputStream(hdrbuf));
final int rhdrs = hdrstr.readInt();
if (rhdrs != nblocks) {
throw new IllegalStateException(
"Incompatible BLOB header record, expected: "
+ nblocks + ", got: " + rhdrs);
}
final int[] blobHdr = new int[nblocks];
for (int i = 0; i < nblocks; i++) {
blobHdr[i] = hdrstr.readInt();
}
// Now we have the header addresses, we can read MAX_FIXED_ALLOCS until final buffer
if (!s_readBlobsAsync) { // synchronous read of blob data
int cursor = 0;
int rdlen = m_maxFixedAlloc;
for (int i = 0; i < nblocks; i++) {
if (i == (nblocks - 1)) {
rdlen = length - cursor;
}
getData(blobHdr[i], buf, cursor, rdlen); // include space for checksum
cursor += rdlen-4; // but only increase cursor by data
}
// } else { // s_readBlobsAsync
// final AsynchronousFileChannel channel = m_reopener.getAsyncChannel();
// final ArrayList> reads = new ArrayList>();
// try {
// int cursor = 0;
// int rdlen = m_maxFixedAlloc;
// int cacheReads = 0;
// for (int i = 0; i < nblocks; i++) {
// if (i == (nblocks - 1)) {
// rdlen = length - cursor;
// }
// final ByteBuffer bb = ByteBuffer.wrap(buf,
// cursor, rdlen-4); // strip off checksum to avoid overlapping buffer reads!
// final long paddr = physicalAddress(blobHdr[i]);
// final ByteBuffer cache = m_writeCacheService._readFromCache(paddr, rdlen);
// if (cache != null) {
// bb.put(cache); // write cached data!
// cacheReads++;
// } else {
// reads.add(channel.read(bb,
// paddr));
// }
// cursor += rdlen - 4; // but only increase cursor by data
// }
// for (Future r : reads) {
// r.get();
// }
// } catch (Exception e) {
// throw new IOException("Error from async IO", e);
// } finally {
// for (Future r : reads) {
// r.cancel(true);
// }
// }
} else { // read non-cached data with FileChannelUtility
final ArrayList transfers = new ArrayList();
int cursor = 0;
int rdlen = m_maxFixedAlloc;
for (int i = 0; i < nblocks; i++) {
if (i == (nblocks - 1)) {
rdlen = length - cursor;
}
final ByteBuffer bb = ByteBuffer.wrap(buf,
cursor, rdlen - 4); // strip off
// checksum to avoid
// overlapping
// buffer reads!
final long paddr = physicalAddress(blobHdr[i]);
final ByteBuffer cache;
try {
cache = m_writeCacheService._readFromCache(paddr, rdlen);
} catch (Exception e) {
throw new IOException("Error from async IO", e);
}
if (cache != null) {
bb.put(cache); // write cached data!
} else {
transfers.add(new AsyncTransfer(paddr, bb));
}
cursor += rdlen - 4; // but only increase cursor
// by data
}
FileChannelUtility.readAllAsync(m_reopener, transfers);
}
return;
} catch (IOException e) {
log.error(e,e);
throw new IllegalStateException("Unable to restore Blob allocation", e);
}
}
{
final StoreCounters storeCounters = (StoreCounters) this.storeCounters
.get().acquire();
try {
final int nbytes = length;
if (nbytes > storeCounters.maxReadSize) {
storeCounters.maxReadSize = nbytes;
}
} finally {
storeCounters.release();
}
}
try {
final int slotSize = getBlock((int) addr).getBlockSize();
if (slotSize < length) {
throw new IllegalStateException("Bad Address: length requested greater than allocated slot: " + slotSize + " < " + length);
}
final long paddr = physicalAddress((int) addr);
if (paddr == 0) {
assertAllocators();
throw new PhysicalAddressResolutionException(addr);
}
assert paddr > 0;
/**
* Check WriteCache first
*
* Note that the buffer passed in should include the checksum
* value, so the cached data is 4 bytes less than the buffer
* size.
*/
final ByteBuffer bbuf;
try {
bbuf = m_writeCacheService != null ? m_writeCacheService.read(paddr, length) : null;
} catch (Throwable t) {
throw new IllegalStateException(
"Error reading from WriteCache addr: " + paddr
+ " length: " + (length - 4)
+ ", writeCacheDebug: "
+ m_writeCacheService.addrDebugInfo(paddr), t);
}
if (bbuf != null) {
if (bbuf.limit() != length-4) {
assertAllocators();
throw new IllegalStateException(
"Incompatible buffer size for addr: " + paddr
+ ", " + bbuf.limit() + " != "
+ (length - 4) + " writeCacheDebug: "
+ m_writeCacheService.addrDebugInfo(paddr));
}
final byte[] in = bbuf.array(); // reads in with checksum - no need to check if in cache
for (int i = 0; i < length-4; i++) {
buf[offset+i] = in[i];
}
m_cacheReads++;
/*
* Hit on the write cache.
*
* Update the store counters.
*/
final StoreCounters c = (StoreCounters) storeCounters
.get().acquire();
try {
final int nbytes = length;
c.nreads++;
c.bytesRead += nbytes;
c.elapsedReadNanos += (System.nanoTime() - begin);
} finally {
c.release();
}
} else {
// Read through to the disk.
// With a non-null WCS, the actual read should be via a callback to readRaw, it should not get here
// unless it is not possible to cache - but maybe even then the WCS should read into a temporary
// buffer
// If checksum is required then the buffer should be sized to include checksum in final 4 bytes
final ByteBuffer bb = ByteBuffer.wrap(buf, offset, length);
// Use ReadRaw - should be the same read all
readRaw(paddr, bb);
final int chk = ChecksumUtility.getCHK().checksum(buf, offset, length-4); // read checksum
final int tstchk = bb.getInt(offset + length-4);
if (chk != tstchk) {
assertAllocators();
if (m_writeCacheService != null) {
final String cacheDebugInfo = m_writeCacheService.addrDebugInfo(paddr);
log.warn("Invalid data checksum for addr: " + paddr
+ ", chk: " + chk + ", tstchk: " + tstchk + ", length: " + length
+ ", first bytes: " + toHexString(buf, 32) + ", successful reads: " + m_diskReads
+ ", at last extend: " + m_readsAtExtend + ", cacheReads: " + m_cacheReads
+ ", writeCacheDebug: " + cacheDebugInfo);
}
throw new IllegalStateException(
"Invalid data checksum from address: " + paddr
+ ", size: " + (length - 4));
}
// do not explicitly cache the read, it will be cached by the WCS!
// if (m_writeCache != null) { // cache the read!
// m_writeCache.cache(paddr, bb);
// }
}
} catch (PhysicalAddressResolutionException e) {
throw new IllegalArgumentException("Unable to read data: "+e, e);
} catch (Throwable e) {
/*
* Note: ClosedByInterruptException can be thrown out of
* FileChannelUtility.readAll(), typically because the LIMIT on
* a query was satisfied, but we do not want to log that as an
* error.
*/
// log.error(e,e);
throw new RuntimeException("addr=" + addr + " : cause=" + e, e);
}
} finally {
lock.unlock();
}
}
// /**
// * Convenience check for thoseA batch invoice public methods that must be restricted if a rebuild is in progress
// */
// private void assertNoRebuild() {
// if (m_rebuildRequest != null)
// throw new IllegalStateException("Invalid when rebuilding");
// }
private void assertAllocators() {
final Lock lock = m_allocationReadLock;
lock.lock();
try {
for (int i = 0; i < m_allocs.size(); i++) {
if (m_allocs.get(i).getIndex() != i) {
throw new IllegalStateException("Allocator at invalid index: " + i + ", index stored as: "
+ m_allocs.get(i).getIndex());
}
}
} finally {
lock.unlock();
}
}
// static private final char[] HEX_CHAR_TABLE = {
// '0', '1','2','3',
// '4','5','6','7',
// '8','9','a','b',
// 'c','d','e','f'
// };
// utility to display byte array of maximum i bytes as hexString
static private String toHexString(final byte[] buf, int n) {
// n = n < buf.length ? n : buf.length;
// final StringBuffer out = new StringBuffer();
// for (int i = 0; i < n; i++) {
// final int v = buf[i] & 0xFF;
// out.append(HEX_CHAR_TABLE[v >>> 4]);
// out.append(HEX_CHAR_TABLE[v &0xF]);
// }
// return out.toString();
return BytesUtil.toHexString(buf, n);
}
public void free(final long laddr, final int sze) {
free(laddr, sze, null/* AlocationContext */);
}
// private long m_unsafeFrees = 0;
/**
* free
*
* If the address is greater than zero than it is interpreted as a physical
* address and the allocators are searched to find the allocations.
* Otherwise the address directly encodes the allocator index and bit
* offset, allowing direct access to clear the allocation.
*
* A blob allocator contains the allocator index and offset, so an allocator
* contains up to 245 blob references.
*
* @param laddr
* @param sze
* @param context
*/
public void free(final long laddr, final int sze, final IAllocationContext context) {
assertOpen();
// assertNoRebuild();
final int addr = (int) laddr;
switch (addr) {
case 0:
case -1:
case -2:
return;
}
m_allocationWriteLock.lock();
try {
checkContext(context);
if (m_lockAddresses != null && m_lockAddresses.containsKey((int)laddr))
throw new IllegalStateException("address locked: " + laddr);
if (sze > m_maxFixedAlloc-4) {
freeBlob(addr, sze, context);
} else {
final FixedAllocator alloc = getBlockByAddress(addr);
/*
* There are a few conditions here. If the context owns the
* allocator and the allocation was made by this context then it
* can be freed immediately. The problem comes when the context
* is null and the allocator is NOT owned, BUT there are active
* AllocationContexts, in this situation, the free must ALWAYS
* be deferred.
*
* If the MIN_RELEASE_AGE is ZERO then we can protect allocations
* and read-only transactions with Session protection, avoiding
* the need to manage deferred frees.
*
* FIXME We need unit tests when MIN_RELEASE_AGE is GT ZERO.
*
* FIXME We need unit test when MIN_RELEASE_AGE is ZERO AND
* there are open read-only transactions.
*/
if (m_minReleaseAge == 0) {
/*
* The session protection is complicated by the mix of
* transaction protection and isolated AllocationContexts.
*
* If this is the first use of an IAllocationContext then
* then isSessionProtected may return false, so check the
* context first.
*/
if (context != null && context.isIsolated()) {
if (alloc.canImmediatelyFree(addr, sze, context)) {
immediateFree(addr, sze, true);
} else {
getContextAllocation(context).deferFree(encodeAddr(addr, sze));
}
} else if (this.isSessionProtected()) {
immediateFree(addr, sze, false);
} else {
immediateFree(addr, sze);
}
} else if (context != null && (context.isIsolated()) && alloc.canImmediatelyFree(addr, sze, context)){
immediateFree(addr, sze);
} else {
// if a free request is made within a context not managed by
// the allocator then it is not safe to free
boolean alwaysDefer = m_activeTxCount > 0;
if (!alwaysDefer)
alwaysDefer = context == null && !m_contexts.isEmpty();
if (alwaysDefer)
if (log.isDebugEnabled())
log.debug("Should defer " + addr + " real: " + physicalAddress(addr));
if (alwaysDefer || !alloc.canImmediatelyFree(addr, sze, context)) {
// If the context is != null, then the deferral must be against that context!
if (context != null && context.isIsolated()) {
getContextAllocation(context).deferFree(encodeAddr(addr, sze));
} else {
deferFree(addr, sze);
}
} else {
immediateFree(addr, sze);
}
}
}
} finally {
m_allocationWriteLock.unlock();
}
}
private void checkContext(final IAllocationContext context) {
if (context != null) {
context.checkActive();
}
}
private long encodeAddr(long alloc, final int nbytes) {
alloc <<= 32;
alloc += nbytes;
return alloc;
}
long getHistoryRetention() {
return m_minReleaseAge;
}
/**
* Session protection can only be used in preference to deferred frees when
* the minReleaseAge is zero. If so then two protection states are checked:
* either a positive activeTxCount incremented by the TransactionManager
* or if there are active AllocationContexts.
*
* The activeTxCount essentially protects read-only transactions while the
* AllocationContexts enable concurrent store allocations, whilst also
* supporting immediate re-cycling of localized allocations (those made
* and released within the same AllocationContext).
*
* Also check to see if there is an uncomplete quorum being established, in
* which case provide default session protection to avoid recycling.
*
* @return whether there is a logical active session
*/
boolean isSessionProtected() {
if (!m_allocationWriteLock.isHeldByCurrentThread()) {
/*
* In order for changes to m_activeTxCount to be visible the caller
* MUST be holding the lock.
*/
throw new IllegalMonitorStateException();
}
// backoff until synchronization is implemented
// // protect recyling with unmet quorum
// if (m_quorum != null && !m_quorum.isQuorumMet()) {
// return true;
// }
return m_minReleaseAge == 0 && (m_activeTxCount > 0 || !m_contexts.isEmpty());
}
/**
* Sessions will only be used to protect transactions and read-only views
* when the m_minReleaseAge is no zero, otherwise the deferredFree
* approach will be used.
*
* When called, will call through to the Allocators to re-sync the
* transient bits with the committed and live.
*
* The writeCache is passed into the allocator to enable any "now free"
* allocations to be cleared from the cache. Until the session is released
* the writeCache must be maintained to support readers of uncommitted and
* unwritten allocations.
*/
private void releaseSessions() {
assert(m_activeTxCount == 0 && m_contexts.isEmpty());
if (m_minReleaseAge == 0) {
if (log.isDebugEnabled())
log.debug("RELEASE SESSIONS");
for (FixedAllocator fa : m_allocs) {
fa.releaseSession(m_writeCacheService);
}
}
}
private boolean freeBlob(final int hdr_addr, final int sze, final IAllocationContext context) {
if (sze <= (m_maxFixedAlloc-4))
throw new IllegalArgumentException("Unexpected address size");
if (m_storageStats != null) {
m_storageStats.deleteBlob(sze);
}
final int alloc = m_maxFixedAlloc-4;
final int blcks = (alloc - 1 + sze)/alloc;
// read in header block, then free each reference
final byte[] hdr = new byte[(blcks+1) * 4 + 4]; // add space for checksum
getData(hdr_addr, hdr);
final DataInputStream instr = new DataInputStream(
new ByteArrayInputStream(hdr, 0, hdr.length-4) );
try {
final int allocs = instr.readInt();
int rem = sze;
for (int i = 0; i < allocs; i++) {
final int nxt = instr.readInt();
free(nxt, rem < alloc ? rem : alloc, context);
rem -= alloc;
}
free(hdr_addr, hdr.length, context);
return true;
} catch (IOException ioe) {
throw new RuntimeException(ioe);
}
}
private boolean freeImmediateBlob(final int hdr_addr, final int sze) {
if (sze <= (m_maxFixedAlloc-4))
throw new IllegalArgumentException("Unexpected address size");
if (m_storageStats != null) {
m_storageStats.deleteBlob(sze);
}
final int alloc = m_maxFixedAlloc-4;
final int blcks = (alloc - 1 + sze)/alloc;
// read in header block, then free each reference
final byte[] hdr = new byte[(blcks+1) * 4 + 4]; // add space for checksum
getData(hdr_addr, hdr);
final DataInputStream instr = new DataInputStream(
new ByteArrayInputStream(hdr, 0, hdr.length-4) );
// retain lock for all frees
m_allocationWriteLock.lock();
try {
final int allocs = instr.readInt();
int rem = sze;
for (int i = 0; i < allocs; i++) {
final int nxt = instr.readInt();
immediateFree(nxt, rem <= alloc ? rem : alloc);
rem -= alloc;
}
immediateFree(hdr_addr, hdr.length);
return true;
} catch (IOException ioe) {
throw new RuntimeException(ioe);
} finally {
m_allocationWriteLock.unlock();
}
}
// private long immediateFreeCount = 0;
private void immediateFree(final int addr, final int sze) {
immediateFree(addr, sze, false);
}
private void immediateFree(final int addr, final int sze, final boolean overrideSession) {
switch (addr) {
case 0:
case -1:
case -2:
return;
}
if (sze > (this.m_maxFixedAlloc-4)) {
freeImmediateBlob(addr, sze);
return;
}
m_allocationWriteLock.lock();
try {
final FixedAllocator alloc = getBlockByAddress(addr);
final int addrOffset = getOffset(addr);
if (alloc == null) {
throw new IllegalArgumentException("Invalid address provided to immediateFree: " + addr + ", size: " + sze);
}
final long pa = alloc.getPhysicalAddress(addrOffset);
// In a tight loop, this log level test shows up as a hotspot
// if (log.isTraceEnabled())
// log.trace("Freeing allocation at " + addr + ", physical address: " + pa);
alloc.free(addr, sze, overrideSession);
// must clear after free in case is a blobHdr that requires reading!
// the allocation lock protects against a concurrent re-allocation
// of the address before the cache has been cleared
assert pa != 0;
// only clear any existing write to cache if no active session
if (overrideSession || !this.isSessionProtected()) {
// Only overwrite if NOT committed
if (!alloc.isCommitted(addrOffset)) {
m_writeCacheService.clearWrite(pa,addr);
// m_writeCache.overwrite(pa, sze);
/*
* Pass the size of the allocator, NOT the size of the
* allocation.
*
* @see RWStore immedateFree() not removing Checkpoint
* addresses from the historical index cache.
*/
// removeFromExternalCache(pa, sze);
removeFromExternalCache(pa, alloc.m_size);
}
}
m_frees++;
if (alloc.isAllocated(addrOffset))
throw new IllegalStateException("Reallocation problem with WriteCache");
if (alloc.isUnlocked()) {
addToCommit(alloc);
}
m_recentAlloc = true;
} finally {
m_allocationWriteLock.unlock();
}
}
/**
* We need to remove entries from the historicalIndexCache for checkpoint
* records when the allocations associated with those checkpoint records are
* freed.
*
* @param clr
* The physical address that is being deleted.
* @param slotSize
* The size of the allocator slot for that physical address.
*
* @see
* RWStore immedateFree() not removing Checkpoint addresses from the
* historical index cache.
*/
void removeFromExternalCache(final long clr, final int slotSize) {
assert m_allocationWriteLock.isHeldByCurrentThread();
if (m_externalCache == null)
return;
if (slotSize == 0 || slotSize == m_cachedDatasize) {
/*
* Either known to be the same slot size as a checkpoint record -or-
* the slot size is not known.
*/
m_externalCache.remove(clr);
}
}
/**
* alloc
*
* Alloc always allocates from a FixedAllocation. Blob allocations are
* implemented using largest Fixed blocks as specified in MAX_FIXED_ALLOC.
*
* The previous Stream method chained blocks together, but the new approach
* uses a master block and a list of allocations. Since we now have a
* MAX-FIXED_ALLOC of 256K this means that we would represent a 1MB
* allocation as a 64byte masters and four 256K blocks. For BigData 1MB
* bloom filters we would probably handle all in a single FixedAllocator of
* 256K allocations since we would hold 4096 of these in a single allocator,
* which with (say) 12 1MB bloom filters with 2-phase commit would only
* require 2 * (4 * 12) = 48 bits plus 12 64 byte headers. The maximum BLOB
* would be determined by a 256K header record with 64K * 256K allocations
* or 16GB, which is larger than MAXINT (we use an int to store allocation
* size in the address).
*
* The use of a IAllocationContext adds some complexity to the previous
* simple freelist management. The problem is two-fold.
*
* Firstly it is okay for an Allocator on the free list to return a null
* address, since it may be managing storage for a specific context.
*
* Secondly we must try and ensure that Allocators used by a specific
* context can be found again. For example, if allocator#1 is assigned to
* context#1 and allocator#2 to context#2, when context#1 is detached we
* want context#2 to first find allocator#2. This is further complicated
* by the finer granularity of the AllocBlocks within a FixedAllocator.
*/
// private volatile long m_maxAllocation = 0;
private volatile long m_spareAllocation = 0;
/** Core allocation method. */
public int alloc(final int size, final IAllocationContext context) {
if (size > m_maxFixedAlloc) {
throw new IllegalArgumentException("Allocation size to big: " + size + " > " + m_maxFixedAlloc);
}
m_allocationWriteLock.lock();
try {
checkContext(context);
try {
final FixedAllocator allocator;
final int i = fixedAllocatorIndex(size);
if (context != null && context.isIsolated()) {
allocator = getContextAllocation(context).getFreeFixed(i);
if (allocator.checkBlock0()) {
if (log.isInfoEnabled())
log.info("Adding new shadowed allocator, index: " + allocator.getIndex() + ", diskAddr: " + allocator.getDiskAddr());
addToCommit(allocator);
}
} else {
final int block = 64 * m_allocSizes[i];
m_spareAllocation += (block - size); // Isn't adjusted by frees!
final ArrayList list = m_freeFixed[i];
if (list.size() == 0) {
/*
* No allocator on the free list for that slot size.
*/
final FixedAllocator candidate;
if (size < this.cSmallSlot) {
/*
* Check to see if can locate a good enough
* Allocator
*
* @see BLZG-1278 (Small slot optimization to
* minimize waste).
*/
candidate = findAllocator(block);
} else {
candidate = null;
}
if (candidate != null) {
candidate.addToFreeList();
allocator = candidate;
} else {
/*
* We need a new allocator.
*/
allocator = new FixedAllocator(this, block);
allocator.setFreeList(list);
allocator.setIndex(m_allocs.size());
if (log.isTraceEnabled())
log.trace("New FixedAllocator for " + block);
m_allocs.add(allocator);
if (m_storageStats != null) {
m_storageStats.register(allocator, true);
}
}
if (allocator.checkBlock0()) {
addToCommit(allocator);
}
} else {
// Verify free list only has allocators with free bits
if (log.isDebugEnabled()){
int tsti = 0;
final Iterator allocs = list.iterator();
while (allocs.hasNext()) {
final Allocator tstAlloc = allocs.next();
if (!tstAlloc.hasFree()) {
throw new IllegalStateException("Free list contains full allocator, " + tsti + " of " + list.size());
}
tsti++;
}
}
allocator = list.get(0);
}
}
final int addr = allocator.alloc(this, size, context);
if (addr == 0) {
throw new IllegalStateException("Free Allocator unable to allocate address: " + allocator.getSummaryStats());
}
if (allocator.isUnlocked()) {
addToCommit(allocator);
}
m_recentAlloc = true;
final long pa = physicalAddress(addr);
if (pa == 0L) {
throw new IllegalStateException(
"No physical address found for " + addr);
}
m_allocations++;
m_nativeAllocBytes += size;
return addr;
} catch (Throwable t) {
log.error(t,t);
throw new RuntimeException(t);
}
} finally {
m_allocationWriteLock.unlock();
}
}
/**
* For a small slot size only, look for an existing allocator that has a
* sufficient percentage of free bits and add it to the free list. If this
* test fails then the caller must allocate a new allocator.
*
* @param block
*
* @return
*
* @see BLZG-1278 (Small slot optimization to minimize waste).
*/
private FixedAllocator findAllocator(final int block) {
// only look if small slot
if (block > cSmallSlot) {
return null;
}
// Look up the statistics for that slot size.
final Bucket stats = m_storageStats.findBucket(block);
if (stats == null) {
// Can't do anything. This is not an expected code path.
return null;
}
/*
* Only check waste if number of allocators is greater than some
* configurable amount.
*
* The thought here is that it is not necessary to focus on minimizing
* waste for small stores and that by allowing that waste we permit
* better locality (co-location on a page) for small slots. Once we
* start to limit the small slot waste we essentially just change the
* #of free bits before we are willing to allow a small slot allocator
* onto the free list.
*/
if (stats.m_allocators < cSmallSlotWasteCheckAllocators) {
return null;
}
// only check small slots if total waste is larger than some configurable amount
final float slotWaste = stats.slotsUnused();
if (slotWaste < cSmallSlotHighWaste) {
return null;
}
// Now find candidate allocator with maximum free slots above a minimum threshold
FixedAllocator candidate = null;
int candidateFreeBits = cSmallSlotThresholdHighWaste; // minimum threshold
for (int i = 0; i < m_allocs.size(); i++) {
final FixedAllocator tst = m_allocs.get(i);
if (tst.getBlockSize() == block) { // right size
if (tst.m_freeBits > candidateFreeBits) {
candidate = tst;
candidateFreeBits = candidate.m_freeBits;
}
}
}
if (candidate != null) {
candidate.m_smallSlotHighWaste = true;
if (log.isDebugEnabled()) {
log.debug("Found candidate small slot allocator");
}
}
return candidate;
}
private int fixedAllocatorIndex(final int size) {
int i = 0;
int cmp = m_minFixedAlloc;
while (size > cmp) {
i++;
cmp = 64 * m_allocSizes[i];
}
return i;
}
/****************************************************************************
* The base realloc method that returns a stream for writing to rather than
* handle the reallocation immediately.
**/
public PSOutputStream realloc(final long oldAddr, final int size) {
free(oldAddr, size);
return PSOutputStream.getNew(this, m_maxFixedAlloc, null);
}
/****************************************************************************
* Called by PSOutputStream to make to actual allocation or directly by
* lower level API clients.
*
* If the allocation is for greater than MAX_FIXED_ALLOC, then a
* PSOutputStream is used to manage the chained buffers.
*
* TODO: Instead of using PSOutputStream, manage allocations written to the
* WriteCacheService, building BlobHeader as you go.
**/
public long alloc(final byte buf[], final int size,
final IAllocationContext context) {
m_allocationWriteLock.lock();
try {
checkContext(context);
final long begin = System.nanoTime();
if (size > (m_maxFixedAlloc - 4)) {
if (size > getMaxBlobSize())
throw new IllegalArgumentException(
"Allocation request beyond maximum BLOB of "
+ getMaxBlobSize());
if (log.isTraceEnabled())
log.trace("BLOB ALLOC: " + size);
if (m_storageStats != null) {
m_storageStats.allocateBlob(size);
}
final PSOutputStream psout = PSOutputStream.getNew(this,
m_maxFixedAlloc, context);
try {
int i = 0;
final int blocks = size / 512;
for (int b = 0; b < blocks; b++) {
psout.write(buf, i, 512); // add 512 bytes at a time
i += 512;
}
psout.write(buf, i, size - i);
return psout.save();
} catch (IOException e) {
throw new RuntimeException("Closed Store?", e);
} finally {
try {
psout.close(); // return stream
} catch (IOException ioe) {
// should not happen, since this should only be
// recycling
log.warn("Unexpected error closing PSOutputStream", ioe);
}
}
}
final int newAddr = alloc(size + 4, context); // allow size for
// checksum
if (newAddr == 0)
throw new IllegalStateException("NULL address allocated");
final int chk = ChecksumUtility.getCHK().checksum(buf, size);
final long pa = physicalAddress(newAddr);
try {
m_writeCacheService.write(pa, ByteBuffer.wrap(buf, 0, size),
chk, true/* writeChecksum */, newAddr/* latchedAddr */);
} catch (InterruptedException e) {
throw new RuntimeException("Closed Store?", e);
}
// Update counters.
final StoreCounters c = (StoreCounters) storeCounters.get()
.acquire();
try {
final int nwrite = size + 4;// size plus checksum.
c.nwrites++;
c.bytesWritten += nwrite;
c.elapsedWriteNanos += (System.nanoTime() - begin);
if (nwrite > c.maxWriteSize) {
c.maxWriteSize = nwrite;
}
} finally {
c.release();
}
return newAddr;
} finally {
m_allocationWriteLock.unlock();
}
}
// /****************************************************************************
// * Fixed buffer size reallocation
// **/
// public long realloc(final long oldAddr, final int oldSize, final byte buf[]) {
//
// free(oldAddr, oldSize);
//
// return alloc(buf, buf.length);
// }
// /**
// * Must handle valid possibility that a request to start/commit transaction
// * could be made within a commitCallback request
// */
// synchronized public void startTransaction() {
// if (m_committing) {
// return;
// }
//
// m_transactionCount++;
// }
//
// synchronized public void commitTransaction() {
// if (m_committing) {
// return;
// }
//
// if (log.isDebugEnabled())
// log.debug("Commit Transaction");
//
// if (--m_transactionCount <= 0) {
// commitChanges();
//
// m_transactionCount = 0;
// }
// }
//
// public int getTransactionCount() {
// return m_transactionCount;
// }
//
// // --------------------------------------------------------------------------------------------
// // rollbackTransaction
// //
// // clear write cache
// // read in last committed header
// synchronized public void rollbackTransaction() {
// if (m_transactionCount > 0 || m_readOnly) { // hack for resync
// baseInit();
//
// try {
// m_writeCache.reset(); // dirty writes are discarded
//
// readAllocationBlocks();
// } catch (Exception e) {
// throw new StorageTerminalError("Unable to rollback transaction", e);
// }
// }
// }
// /*
// * Slug
// */
// private int fibslug(int n) {
// if (n < 2)
// return 1;
// else
// return fibslug(n-1) + fibslug(n-2);
// }
/**
* The semantics of reset are to revert unisolated writes to committed
* state.
*
* Unisolated writes must also be removed from the write cache.
*
* The AllocBlocks of the FixedAllocators maintain the state to determine
* the correct reset behavior.
*
* If the store is using DirectFixedAllocators then an IllegalStateException
* is thrown.
*
* If there is an active {@link #m_commitStateRef}, then this indicates a
* failure after the {@link RWStore#commit()} had "succeeded".
*/
public void reset() {
if (log.isInfoEnabled()) {
log.info("RWStore Reset");
}
m_allocationWriteLock.lock();
try {
// DEBUG
// fibslug(40); // slug to improve odds of interruption of reset (if possible)
assertOpen();
// assertNoRebuild();
final CommitState commitState = m_commitStateRef
.getAndSet(null/* newValue */);
if (commitState != null) {
commitState.reset(); // restore state values on RWStore.
}
boolean isolatedWrites = false;
/**
* Clear all allocators, not just dirty allocators, since we also
* need to reset the transient bits associated with session
* protection.
*
* Need to know if there are any isolated modifications, in which case
* we must remember so that we avoid clearing down the store.
*/
for (FixedAllocator fa : m_allocs) {
isolatedWrites |= fa.reset(m_writeCacheService, m_committedNextAllocation);
}
/**
* Now clone the transient metabits for protection if this service becomes leader
*/
syncMetaTransients();
if (!isolatedWrites) {
/**
* Now we should be able to unwind any unused allocators and unused
* alloc blocks. An unused allocator is one with no diskAddr (never
* committed). But it may be more difficult to determine if
* an alloc block has never been used, for that we really need to
* know what the nextAllocationOffset was at the previous commit.
* This could be cached as lastCommittedOffset, in which case we can unwind any
* allocBlocks with addresses >= to that.
*/
int origAllocs = m_allocs.size();
while (m_allocs.size() > 0) {
final int last = m_allocs.size()-1;
final FixedAllocator fa = m_allocs.get(last);
if (fa.getDiskAddr() == 0) {
fa.setIndex(-1);
// must remove from free list!
m_freeFixed[fixedAllocatorIndex(fa.m_size)].remove(fa);
// ..and then from main allocation list
m_allocs.remove(last);
} else {
break;
}
}
m_nextAllocation = m_committedNextAllocation;
if (log.isDebugEnabled())
log.debug("Reset allocators, old: " + origAllocs + ", now: " + m_allocs.size());
// Clear the dirty list.
// FIXME: we should be able to clear the dirty list, but this currently causes
// problems in HA.
// If the allocators are torn down correctly, we should be good to clear the commitList
clearCommitList();
// Flag no allocations since last commit
m_recentAlloc = false;
} else {
// there are isolated writes, so we must not clear the commit list since otherwise
// the Alloction index wil get out of sync as per Ticket #1136
}
if (m_quorum != null) {
/**
* When the RWStore is part of an HA quorum, we need to close
* out and then reopen the WriteCacheService every time the
* quorum token is changed. For convienence, this is handled by
* extending the semantics of abort() on the Journal and reset()
* on the RWStore.
*
* @see
* HA Journal
*/
m_writeCacheService.close();
m_writeCacheService = newWriteCacheService();
} else if (m_writeCacheService != null) {
/*
* Note: We DO NOT need to reset() the WriteCacheService. If a
* record was already flushed to the disk, then it is on the
* disk and clearing the record from the cache will not change
* that. If the record has not yet been flushed to the disk,
* then we already cleared it from the WCS when we reset the
* FixedAllocators (above).
*/
// m_writeCacheService.reset();
// m_writeCacheService.setExtent(convertAddr(m_fileSize));
}
/*
* Discard any writes on the delete blocks. Those deletes MUST NOT
* be applied after a reset() on the RWStore.
*
* @see https://sourceforge.net/apps/trac/bigdata/ticket/602
* (RWStore does not discard deferred deletes on reset)
*/
m_deferredFreeOut.reset();
/*
* Reset any storage stats
*/
if (m_storageStatsAddr != 0) {
m_storageStats.reset();
} else {
m_storageStats = new StorageStats(m_allocSizes);
}
} catch (Exception e) {
throw new IllegalStateException("Unable to reset the store", e);
} finally {
m_allocationWriteLock.unlock();
}
}
// synchronized public boolean isActiveTransaction() {
// return m_transactionCount > 0;
// }
/**
* writeMetaBits must be called after all allocations have been made, the
* last one being the allocation for the metabits themselves (allowing for
* an extension!).
*
* Ticket #936: The meta-bits allocation is currently made from the FixedAllocator
* region. This works well providing the required allocation bits is less than
* the maximum FixedAllocator slot size. While this is neat, there are problems at scale
* for maximum slot sizes less than 64K.
*
* To address the 8K bits in a 1K alloctor, 13 bits are required, this leaves 19 bits
* to index an Allocator, or 18 bits without the sign => 256K maximum index.
*
* To be able to commit changes to all 256K allocators requires 512K metabits => 64K bytes.
* We would like to associate the 64K allocations with the root block, so a single 128K
* allocation would be split into 64K demi-spaces, one for each root block.
*
* While a negative address indicates a standard RW allocation a ositive address can be used
* to indicate an explicitly allocated region. The trick is to ensure that the region is
* allocated on a 128K boundary, then the lower bits can indicate which demi-space is used with
* a simple XOR.
*
* Note that we must ensure that any previous demi-space write is removed from the WCS.
*
* @throws IOException
*/
private void writeMetaBits() throws IOException {
final byte buf[] = genMetabitsData();
/*
* Note: this address is set by commit() prior to calling
* writeMetaBits().
*/
//final long addr = physicalAddress(m_metaBitsAddr);
final long addr = m_metaBitsAddr < 0 ? physicalAddress(m_metaBitsAddr) : ((long) m_metaBitsAddr) << ALLOCATION_SCALEUP;
if (addr == 0) {
throw new IllegalStateException("Invalid metabits address: " + m_metaBitsAddr);
}
assert addr > 0;
try {
if (log.isDebugEnabled())
log.debug("writing metabits at: " + addr);
// Similar to writeMetaBits, we are no longer writing to a FixedAllocator managed region,
// so no latched address is provided
m_writeCacheService.write(addr, ByteBuffer.wrap(buf), 0/*chk*/, false/*useChecksum*/, m_metaBitsAddr < 0 ? m_metaBitsAddr : 0 /*latchedAddr*/);
} catch (InterruptedException e) {
throw new RuntimeException(e);
}
}
private byte[] genMetabitsData() throws IOException {
// the metabits is now prefixed by a long specifying the lastTxReleaseTime
// used to free the deferedFree allocations. This is used to determine
// which commitRecord to access to process the nextbatch of deferred
// frees.
// the cDefaultMetaBitsSize is also written since this can now be
// parameterized.
final int len = 4 * (cMetaHdrFields + m_allocSizes.length + m_metaBits.length);
final byte buf[] = new byte[len];
final FixedOutputStream str = new FixedOutputStream(buf);
try {
str.writeInt(m_metaBitsAddr > 0 ? cVersionDemispace : cVersion);
str.writeLong(m_lastDeferredReleaseTime);
str.writeInt(cDefaultMetaBitsSize);
str.writeInt(m_allocSizes.length);
str.writeLong(m_storageStatsAddr);
// Let's reserve ourselves some space
for (int i = 0; i < cReservedMetaBits; i++) {
str.writeInt(0);
}
/*
* Write out the size of the allocation slots as defined by
* Options.ALLOCATION_SIZES (this is where we store that
* information).
*/
for (int i = 0; i < m_allocSizes.length; i++) {
str.writeInt(m_allocSizes[i]);
}
/*
* Write out XXX
*/
for (int i = 0; i < m_metaBits.length; i++) {
str.writeInt(m_metaBits[i]);
}
str.flush();
} finally {
str.close();
}
return buf;
}
/**
*
* @return
*/
public boolean isDirty() {
return requiresCommit();
}
/**
* Object recording the undo state for the {@link RWStore#commit()} ...
* {@link RWStore#postCommit()} sequence. The {@link CommitState} must
* either {@link CommitState#commit()} or {@link CommitState#reset()}. Those
* {@link CommitState} methods are invoked out of the corresponding
* {@link RWStore} methods.
*
* @see RWStore commit is not
* robust to internal failure.
*/
private class CommitState {
/*
* Critical pre-commit state that must be restored if a commit is
* discarded.
*/
private final int m_lastCommittedNextAllocation;
private final long m_storageStatsAddr;
private final int m_metaBitsAddr;
CommitState() {
// retain copy of critical pre-commit state
if (!m_allocationWriteLock.isHeldByCurrentThread())
throw new IllegalMonitorStateException();
m_lastCommittedNextAllocation = RWStore.this.m_committedNextAllocation;
m_storageStatsAddr = RWStore.this.m_storageStatsAddr;
m_metaBitsAddr = RWStore.this.m_metaBitsAddr;
}
void postCommit() {
// NOP
}
/** Reset pre-commit state to support reset/abort/rollback. */
void reset() {
if (!m_allocationWriteLock.isHeldByCurrentThread())
throw new IllegalMonitorStateException();
RWStore.this.m_storageStatsAddr = m_storageStatsAddr;
RWStore.this.m_committedNextAllocation = m_lastCommittedNextAllocation;
RWStore.this.m_metaBitsAddr = m_metaBitsAddr;
}
}
/**
* @see RWStore commit is not
* robust to internal failure.
*/
private final AtomicReference m_commitStateRef = new AtomicReference();
/**
* Package private method used by the test suite.
*/
void clearCommitStateRef() {
m_commitStateRef.set(null/* newValue */);
}
@Override
public void commit() {
assertOpen();
// assertNoRebuild();
checkCoreAllocations();
// take allocation lock to prevent other threads allocating during commit
m_allocationWriteLock.lock();
try {
/*
* Create a transient object to retain values of previous
* commitState to support abort/reset/rollback if requested after
* this commit() is requested.
*/
if (!m_commitStateRef.compareAndSet(null/* expect */,
new CommitState())) {
throw new IllegalStateException(
"RWStore commitState found, incomplete previous commit must be rolled back/aborted");
}
// final int totalFreed = checkDeferredFrees(true, journal); // free now if possible
//
// if (totalFreed > 0 && log.isInfoEnabled()) {
// log.info("Freed " + totalFreed + " deferralls on commit");
// }
// free old storageStatsAddr
if (m_storageStatsAddr != 0) {
final int len = (int) (m_storageStatsAddr & 0xFFFF);
final int addr = (int) (m_storageStatsAddr >> 16);
immediateFree(addr, len);
}
if (m_storageStats != null) {
final byte[] buf = m_storageStats.getData();
final long addr = alloc(buf, buf.length, null);
m_storageStatsAddr = (addr << 16) + buf.length;
}
/*
* Pre-allocate storage for metaBits from FixedAllocators (ensure
* that we do not need to reallocate the metabits region when we are
* writing out the updated versions of the FixedAllocators).
*/
if (m_metaBitsAddr > 0) {
// already using demi-space, remove from WCS
m_writeCacheService.removeWriteToAddr(convertAddr(-m_metaBitsAddr), 0);
} else {
final int reqmbc = getRequiredMetaBitsStorage();
int nmbaddr = 0;
// if > max alloc or explicitly use the demi-space, then drop through for demi-space
if ((!m_useMetabitsDemispace) && reqmbc < m_maxFixedAlloc) {
nmbaddr = alloc(reqmbc, null);
}
// If existing allocation, then free it
if (m_metaBitsAddr < 0) {
final int oldMetaBitsSize = (m_metaBits.length
+ m_allocSizes.length + 1) * 4;
// Call immediateFree - no need to defer freeof metaBits, this
// has to stop somewhere!
// No more allocations must be made
immediateFree((int) m_metaBitsAddr, oldMetaBitsSize);
}
m_metaBitsAddr = nmbaddr;
}
if (m_metaBitsAddr == 0) {
// Allocate special region to be able to store maximum metabits (128k of 2 64K demi-space
// Must be aligned on 128K boundary and allocations are made in units of 64K.
//
// May need to extend the file for teh demi-space!
while (m_nextAllocation % 2 != 0) {
m_nextAllocation--;
}
m_metaBitsAddr = -m_nextAllocation; // must be positive to differentiate from FixedAllocator address
m_nextAllocation -= 2; // allocate 2 * 64K
// Check for file extension
while (m_nextAllocation <= m_fileSize) {
extendFile();
}
if (log.isInfoEnabled())
log.info("Using Demi-space metabits");
}
if (m_metaBitsAddr > 0) { // Demi-Space
// Now "toggle" m_metaBitsAddr - 64K boundary
m_metaBitsAddr ^= 0x01; // toggle zero or 64K offset
}
if (log.isDebugEnabled()) {
final long mbaddr;
if (m_metaBitsAddr < 0) {
mbaddr = physicalAddress((int) m_metaBitsAddr);
} else {
mbaddr = convertAddr(-m_metaBitsAddr); // maximum 48 bit address range
}
log.debug("Writing metabits at " + mbaddr);
}
// There must be no buffered deferred frees
// assert m_deferredFreeOut.getBytesWritten() == 0;
// save allocation headers
FixedAllocator fa = m_commitHead;
while (fa != null) {
final FixedAllocator allocator = fa;
// the bit in metabits for the old allocator version.
final int old = allocator.getDiskAddr();
// mark old version - reclaimed after commit.
metaFree(old);
// the bit in metabits for the new allocator version.
final int naddr = metaAlloc();
// set that bit on the allocator.
allocator.setDiskAddr(naddr);
if (log.isTraceEnabled())
log.trace("Update allocator " + allocator.getIndex()
+ ", old addr: " + old + ", new addr: " + naddr);
try {
// do not use checksum
m_writeCacheService.write(metaBit2Addr(naddr), ByteBuffer
.wrap(allocator.write()), 0/*chk*/, false/*useChecksum*/,0/*latchedAddr*/);
} catch (InterruptedException e) {
throw new RuntimeException(e);
}
fa = fa.m_nextCommit;
}
// DO NOT clear the commit list until the writes have been flushed
// m_commitList.clear();
writeMetaBits();
try {
m_writeCacheService.flush(true);
lastBlockSequence = m_writeCacheService.resetSequence();
} catch (InterruptedException e) {
log.error(e, e);
throw new RuntimeException(e);
}
// Should not write rootBlock, this is responsibility of client
// to provide control
// writeFileSpec();
syncMetaTransients();
// Must be called from AbstractJournal commitNow after writeRootBlock
// postCommit();
// if (m_commitCallback != null) {
// m_commitCallback.commitComplete();
// }
// The Journal handles the force in doubleSync
// m_reopener.reopenChannel().force(false); // TODO, check if required!
} catch (IOException e) {
throw new StorageTerminalError("Unable to commit transaction", e);
} finally {
m_recentAlloc = false;
m_allocationWriteLock.unlock();
}
checkCoreAllocations();
if (log.isTraceEnabled())
log.trace("commitChanges for: " + m_nextAllocation + ", "
+ m_metaBitsAddr + ", active contexts: "
+ m_contexts.size());
if (log.isDebugEnabled() && m_quorum != null && m_quorum.isHighlyAvailable()) {
log.debug(showAllocatorList());
}
}
/**
* {@inheritDoc}
*/
@Override
public Lock getCommitLock() {
return m_allocationWriteLock;
}
/**
* {@inheritDoc}
*
* Commits the FixedAllocator bits
*/
@Override
public void postCommit() {
if (!m_allocationWriteLock.isHeldByCurrentThread())
throw new IllegalMonitorStateException();
final CommitState commitState = m_commitStateRef.getAndSet(null/* newValue */);
if (commitState == null) {
throw new IllegalStateException(
"No current CommitState found on postCommit");
} else {
commitState.postCommit();
}
{
FixedAllocator fa = m_commitHead;
while (fa != null) {
fa.postCommit();
fa = fa.m_nextCommit;
}
}
if (m_storageStats != null) {
m_storageStats.commit();
}
clearCommitList();
}
@Override
public int checkDeferredFrees(final AbstractJournal journal) {
if (journal == null)
return 0;
/*
* Note: since this is now called directly from the AbstractJournal
* commit method (and is part of a public API) we must take the
* allocation lock.
*
* This may have adverse effects wrt concurrency deadlock issues, but
* none have been noticed so far.
*/
m_allocationWriteLock.lock();
try {
/**
* if session protected then do not free any deferrals!
*/
if (isSessionProtected()) {
return 0;
}
final AbstractTransactionService transactionService = (AbstractTransactionService) journal
.getLocalTransactionManager().getTransactionService();
// the previous commit point.
final long lastCommitTime = journal.getLastCommitTime();
if (lastCommitTime == 0L) {
// Nothing committed.
return 0;
}
/*
* The timestamp for which we may release commit state.
*/
final long latestReleasableTime = transactionService.getReleaseTime();
if (lastCommitTime <= latestReleasableTime) {
throw new AssertionError("lastCommitTime=" + lastCommitTime
+ ", latestReleasableTime=" + latestReleasableTime
+ ", lastDeferredReleaseTime="
+ m_lastDeferredReleaseTime + ", activeTxCount="
+ m_activeTxCount);
}
// Note: This is longer true. Delete blocks are attached to the
// commit point in which the deletes were made.
// /*
// * add one because we want to read the delete blocks for all
// * commit points up to and including the first commit point that
// * we may not release.
// */
// latestReleasableTime++;
// /*
// * add one to give this inclusive upper bound semantics to the
// * range scan.
// */
// latestReleasableTime++;
if (txLog.isInfoEnabled())
txLog.info("RECYCLER: lastCommitTime=" + lastCommitTime
+ ", latestReleasableTime=" + latestReleasableTime
+ ", lastDeferredReleaseTime="
+ m_lastDeferredReleaseTime + ", activeTxCount="
+ m_activeTxCount);
/*
* Free deferrals.
*
* Note: Per ticket#480, we can not begin recycling from the first
* commit point in the commit record index as there are some bigdata
* versions (1.0.4) where we did not prune the commit record index.
* Therefore, this relies on the (lastDeferredReleaseTime+1) for the
* exclusive lower bound. This is avoids triggering an exception
* from an attempt to process deferred free blocks which have
* already been released.
*
* @see https://sourceforge.net/apps/trac/bigdata/ticket/480
*/
if (m_lastDeferredReleaseTime >= latestReleasableTime) {
/**
* Note: Added for HA. I have observed both values equal to
* ZERO. Since we add ONE (1) to the lastDeferredReleaseTime it
* MUST BE LT the latestReleasableTime or we will get a
* "toKey LT fromKey" exception.
*
* @see
* Journal HA
*/
return 0;
}
return freeDeferrals(journal, m_lastDeferredReleaseTime + 1,
latestReleasableTime);
} finally {
m_allocationWriteLock.unlock();
}
}
/**
*
* @return conservative requirement for metabits storage, mindful that the
* request to allocate the metabits may require an increase in the
* number of allocation blocks and therefore an extension to the
* number of metabits.
*/
private int getRequiredMetaBitsStorage() {
int ints = cMetaHdrFields;
ints += m_allocSizes.length + m_metaBits.length;
// add the maximum number of new metaBits storage that may be
// needed to save the current committed objects
final int commitInts = ((32 + commitListSize()) / 32);
final int allocBlocks = (cDefaultMetaBitsSize - 1 + commitInts)/(cDefaultMetaBitsSize-1);
ints += cDefaultMetaBitsSize * allocBlocks;
return ints*4; // return as bytes
}
// Header Data
// volatile private long m_curHdrAddr = 0;
// volatile private int m_rootAddr;
/**
* {@link #m_fileSize} is in units of -32K.
*/
volatile private int m_fileSize;
volatile private int m_nextAllocation;
/**
* The value of nextAllocation at commit is cached and used
* in reset() to unwind new FixedAllocators and/or AllocBlocks
*/
volatile private int m_committedNextAllocation;
final private long m_maxFileSize;
// private int m_headerSize = 2048;
/*
* Meta Allocator
*/
/**
* MetaBits HEADER version must be changed when the header or allocator
* serialization changes
*
* Use BCD-style numbering so
* 0x0200 == 2.00
* 0x0320 == 3.20
*
* The minor byte values should maintain binary compatibility, with
* major bytes
* Versions
* 0x0300 - extended header to include reserved ints
* 0x0400 - removed explicit BlobAllocators
* 0x0500 - using metaBits demi-space
*/
final private int cVersion = 0x0400;
/**
* The {@link #cVersion} value corresponding to the use of the demi-space
* for the metabits.
*
* @see Support larger metabit
* allocations
* @see
* Data migration
*/
final private int cVersionDemispace = 0x0500;
/**
* cReservedMetaBits is the reserved space in the metaBits header
* to alloc for binary compatibility moving forward.
*
* If we need to add int values to the header we can do so and reduce the
* reservation by 1 each time
*/
final static int cReservedMetaBits = 20;
/**
* MetaBits Header
* 0 int version
* 1-2 int[2] long deferredFree
* 3 int defaultMetaBitsSize
* 4 int length of allocation sizes
* 5-6 int[2] storage stats addr
* + 20 reserved
*/
final static private int cMetaHdrFields = 7 + cReservedMetaBits;
/**
* @see Options#META_BITS_SIZE
*/
final private int cDefaultMetaBitsSize = 9;
/**
* @see Options#META_BITS_SIZE
*/
volatile private int m_metaBitsSize;
volatile private boolean m_useMetabitsDemispace = true;
/**
* Package private since is uded by FixedAllocators
*
* @see Options#META_BITS_SIZE
*/
final int cDefaultFreeBitsThreshold;
/**
* The smallSlotThreshold, when activated, is intended to ensure improve the
* opportunity for write elissions (to mechanical disks) whilst also reducing
* the read-backs on current generation (2014-15) SSDs that can impact
* write throughput.
* Given that the objective is to statistically improve write elission,
* the number of required free bits needs to be large - around 50%.
* However, this can result in a large amount of store waste for certain
* patterns of data - for example when small slots are used to store large
* literals that will not be recycled. In this scenario it is possible
* that allocators are not recycled.
* Some further thoughts:
* 1) The more efficient elission of small slots for the allocation of large literals
* is probably the major throughput benefit
* 2) OTOH, at a lower level, small sparse but localised writes (eg 16 64 byte writes to a 4k
* sector) may only incur a single read-back with good firmware.
* To address the concern for high waste, when a statistically large number of allocators have
* been created, and the waste is beyond some threshold, then a lower small slot threshold
* is used. The logic for this is implemented in {@link FixedAllocator#meetsSmallSlotThreshold()}
*/
int cSmallSlot = 1024; // @see from Options#SMALL_SLOT_TYPE
int cSmallSlotThreshold = 4096; // @see from Options#SMALL_SLOT_THRESHOLD
/**
* High Waste Criteria
*/
int cSmallSlotThresholdHighWaste = 2048; // @see from Options#SMALL_SLOT_THRESHOLD_HIGH_WASTE
int cSmallSlotWasteCheckAllocators = 100; // @see from Options#SMALL_SLOT_WASTE_CHECK_ALLOCATORS
float cSmallSlotHighWaste = 0.2f; // @see from Options#SMALL_SLOT_HIGH_WASTE
/**
* Each "metaBit" is a file region
*/
private int m_metaBits[];
private int m_metaTransientBits[];
// volatile private int m_metaStartAddr;
private volatile int m_metaBitsAddr;
// @todo javadoc please.
volatile private boolean m_recentAlloc = false;
/**
* Return the address of a contiguous region on the persistent heap.
*
* @param size
* The size of that region (this is not bytes, but something a
* bit more complicated).
*/
protected int allocBlock(final int size) {
// minimum 1
if (size <= 0) {
throw new Error("allocBlock called with zero size request");
}
final int allocAddr = m_nextAllocation;
m_nextAllocation -= size;
while (convertAddr(m_nextAllocation) >= convertAddr(m_fileSize)) {
extendFile();
}
checkCoreAllocations();
if (log.isTraceEnabled())
log.trace("allocation created at " + convertAddr(allocAddr) + " for " + convertAddr(-size));
return allocAddr;
}
private void checkCoreAllocations() {
final long lfileSize = convertAddr(m_fileSize);
final long lnextAlloc = convertAddr(m_nextAllocation);
if (lnextAlloc >= lfileSize) {
throw new IllegalStateException("Core Allocation Error - file size: "
+ lfileSize + ", nextAlloc: " + lnextAlloc);
}
}
/**
* meta allocation/free
*
* Allocates persistent store for allocation blocks.
*
* grows data from the top to the file, e.g. bit 0 is 1024 from end-of-file.
*
* If metaStart <= nextAllocation, then the file must be extended. All the
* allocation blocks are moved to the new end of file area, and the
* metaStartAddress is incremented by the same delta value.
*
* NB the metaStart calculation uses an address rounded to 8k, so on
* extension the new metaStart may be up to 8K less than the true start
* address.
*
* The updated approach to metaAllocation uses native allocation from
* the heap (by simply incrementing from m_nextAllocation) to provide
* space for the allocation blocks.
*
* This approach means that the file only needs to be extended when
* m_nextAllocation passes the m_fileSize, since we no longer store
* the allocation blocks at the end of the file.
*/
int metaAlloc() {
int bit = fndMetabit();
if (bit < 0) {
// reallocate metaBits and recalculate m_headerSize
// extend m_metaBits by 8 ints of bits plus start address!
final int nsize = m_metaBits.length + cDefaultMetaBitsSize;
// arrays initialized to zero by default
final int[] nbits = new int[nsize];
final int[] ntransients = new int[nsize];
// copy existing values
for (int i = 0; i < m_metaBits.length; i++) {
nbits[i] = m_metaBits[i];
ntransients[i] = m_metaTransientBits[i];
}
m_metaBits = nbits;
m_metaTransientBits = ntransients;
m_metaBits[m_metaBitsSize] = m_nextAllocation;
m_nextAllocation -= META_ALLOCATION; // 256K
m_metaBitsSize = nsize;
// now get new allocation!
bit = fndMetabit();
assert bit >= 0;
}
setBit(m_metaTransientBits, bit);
setBit(m_metaBits, bit);
if (m_nextAllocation <= m_fileSize) {
if (log.isInfoEnabled())
log.info("ExtendFile called from metaAlloc");
extendFile();
}
// cat.info("meta allocation at " + addr);
checkCoreAllocations();
return bit;
}
/**
* Search the metabits for a bit that is free for allocation of space that
* an allocator could write on.
*
* @return The bit -or- -1 if the meta bits region is currently
* ful.
*/
private int fndMetabit() {
final int blocks = m_metaBits.length / cDefaultMetaBitsSize;
for (int b = 0; b < blocks; b++) {
final int ret = fndBit(m_metaTransientBits,
(b * cDefaultMetaBitsSize) + 1, cDefaultMetaBitsSize-1);
if (ret != -1) {
// The assumption is that this bit is also NOT set in m_metaBits
assert !tstBit(m_metaBits, ret);
return ret;
}
}
return -1; // none found
}
void metaFree(final int bit) {
if (!m_allocationWriteLock.isHeldByCurrentThread()) {
/*
* Must hold the allocation lock while allocating or clearing
* allocations.
*/
throw new IllegalMonitorStateException();
}
if (bit <= 0) {
return;
}
if (tstBit(m_metaBits, bit)) {
clrBit(m_metaBits, bit);
} else {
clrBit(m_metaTransientBits, bit);
}
m_writeCacheService.clearWrite(metaBit2Addr(bit),0/*latchedAddr*/);
}
/**
* The metabits are encoded in {@link #cDefaultMetaBitsSize} int runs as
* follows
*
*
* [startAddr1][bits0][bits1]...[bitsN]
* [startAddr2]...
* ...
*
*
* where N is {@link #cDefaultMetaBitsSize} MINUS TWO and
* [bits0]...[bitsN] are interpreted as a bit map.
*
* The bit parameter is processed to determine which run it is part of.
*
* Note that the bit offsets are not contiguous since there are "holes"
* where the meta allocation [startAddr] are stored.
*
* When the metabits region is first created, and each time it is grown, a
* region is reserved at the then current nextOffset on the file that is
* used for {@link FixedAllocator}s associated with the bit vector in the
* next run of the metabits block. Those {@link FixedAllocator}s will be
* recruited and used as needed. Note that {@link FixedAllocator}s are
* always written onto an unused "bit" at each commit, and the old "bit" is
* then freed. Thus dirty {@link FixedAllocator}s move at each commit and
* can move between runs in the metabits.
*/
long metaBit2Addr(final int bit) {
// final int bitsPerBlock = 9 * 32;
/*
* The integer index into the m_metaBits[].
*/
final int intIndex = bit / 32; // divide 32;
/*
* Make sure that the [bit] is a bit that falls into one of the bit
* regions (versus one of the startAddr int32 values).
*/
assert intIndex % cDefaultMetaBitsSize != 0; // used by the start addrs!
/*
* The index into the metabits region corresponding to the int32 value
* before the start of the bit vector in which this bit falls. This
* offset is relative to the start of the m_metaBits[].
*/
final int addrIndex = (intIndex / cDefaultMetaBitsSize)
* cDefaultMetaBitsSize;
/*
* Pull out convert the startAddr for the bit vector addressed by that
* bit. This gives us the int64 byte offset of some region on the
* backing file.
*/
final long addr = convertAddr(m_metaBits[addrIndex]);
/*
* The bit index of this bit in the bit vector for this region in the
* metaBits[].
*/
final int intOffset = bit - ((addrIndex + 1) * 32);
/*
* The byte offset into the backing file of the FixedAllocator for that
* bit. All FixedAllocators are the same size [ALLOC_BLOCK_SIZE]. The
* FixedAllocator knows what size allocations it makes and manages the
* regions on the backing store in which those allocation are made.
*/
final long ret = addr + (ALLOC_BLOCK_SIZE * intOffset);
return ret;
}
/**
* Convert an implicitly scaled int32 offset into the backing file into an
* int64 address into the backing file.
*
* @param addr
* An int32 offset into the backing file formed by
* {@link #convertFromAddr(long)}. The representation is a
* negative integer that has been left shifted by
* {@link #ALLOCATION_SCALEUP} to reduce its bit size.
*
* @return A byte offset in the backing file.
*
* @see #convertFromAddr(long)
* @see #ALLOCATION_SCALEUP
*/
public static long convertAddr(final int addr) {
final long laddr = addr;
if (laddr < 0) {
final long ret = (-laddr) << ALLOCATION_SCALEUP;
return ret;
} else {
return laddr & 0xFFFFFFF0;
}
}
/**
* Convert an int64 address into the backing file into an int32 offset that
* is implicitly scaled by {@link #ALLOCATION_SCALEUP}.
*
* @param addr
* An int64 offset into the backing file.
*
* @return The implicitly scaled int32 offset.
* @see #convertAddr(int)
* @see #ALLOCATION_SCALEUP
*/
public int convertFromAddr(final long addr) {
return (int) -(addr >> ALLOCATION_SCALEUP);
}
private volatile boolean m_extendingFile = false;
/**
* extendFile will extend by 10% and round up to be a multiple of 16k
*
* The allocation blocks must also be moved. Note that it would be a bad
* idea if these were moved with an overlapping copy!
*
* After moving the physical memory the in-memory allocation blocks must
* then be updated with their new position.
*
* Note that since version 3.0 the size of the metaBits is variable. This
* must be taken into consideration when moving data. - Has the location
* changed as a result of the "reallocation". If this is incorrect then the
* wrong commit blocks will be copied, resulting in a corrupt data file.
*
* There are two approaches to this problem. The first is only to copy the
* known committed (written) allocation blocks - but this cannot be implied
* by "zero'd" bits since this can indicate that memory has been cleared.
*
* Synchronization
*
* The writecache may contain allocation block writes that must be flushed
* before the file can be extended. The extend file explicitly moves the
* written allocation blocks to there new location at the new end of the
* file and then updates the rootblocks to ensure they point to the new
* allocation areas.
*
* Extend file is only triggered by either alloc or metaAlloc which are s
* synchronized by the allocation lock. So extend file ends up being
* synchronized by the same lock.
*
* If we knew that the write cache had no writes to the allocation areas,
* we would not need to flush, but calling flush prior to the extend is
* sufficient to guarantee, in conjunction with holding the allocation lock,
* that no new writes to the allocation areas will be made.
*
* Once the flush is complete we take the extension writeLock to prevent
* further reads or writes, extend the file, moving the allocation areas on
* the disk, then force the new rootblocks to disk.
*/
private void extendFile() {
final int adjust = -1200 + (m_fileSize / 10);
extendFile(adjust);
}
private volatile long m_readsAtExtend = 0;
private void extendFile(final int adjust) {
if (m_extendingFile) {
throw new IllegalStateException("File concurrently extended");
}
/**
* Note: Synchronous flush of the WriteCacheService should not be
* required. It has been commented out in support of
*
*
* Coalesce records in write cache
*/
// try {
// /*
// * The call to flush the cache cannot be made while holding the
// * extension writeLock, since the writeOnChannel takes the
// * extension readLock.
// */
// m_writeCache.flush(true);
// } catch (InterruptedException e) {
// throw new RuntimeException("Flush interrupted in extend file");
// }
final Lock lock = this.m_extensionLock.writeLock();
lock.lock();
try {
m_extendingFile = true;
// final long curSize = convertAddr(m_fileSize);
m_fileSize += adjust;
final long toAddr = convertAddr(m_fileSize);
if (getMaxFileSize() < toAddr) {
// whoops!! How to exit more gracefully?
throw new Error("System greater than maximum size");
}
if (log.isInfoEnabled()) log.info("Extending file to: " + toAddr);
m_reopener.reopenChannel();
m_reopener.raf.setLength(toAddr);
storeCounters.get().ntruncate++;
// must ensure writeCache is in sync for HA
m_writeCacheService.setExtent(toAddr);
if (log.isInfoEnabled()) log.info("Extend file done");
} catch (Throwable t) {
throw new RuntimeException("Force Reopen", t);
} finally {
m_extendingFile = false;
m_readsAtExtend = this.m_diskReads;
lock.unlock();
}
}
static void setBit(final int[] bits, final int bitnum) {
final int index = bitnum / 32;
final int bit = bitnum % 32;
bits[(int) index] |= 1 << bit;
}
static boolean tstBit(final int[] bits, final int bitnum) {
final int index = bitnum / 32;
final int bit = bitnum % 32;
if (index >= bits.length)
throw new IllegalArgumentException("Accessing bit index: " + index
+ " of array length: " + bits.length);
return (bits[(int) index] & 1 << bit) != 0;
}
static void clrBit(final int[] bits, final int bitnum) {
final int index = bitnum / 32;
final int bit = bitnum % 32;
int val = bits[index];
val &= ~(1 << bit);
bits[index] = val;
}
static int fndBit(final int[] bits, final int size) {
return fndBit(bits, 0, size);
}
static int fndBit(final int[] bits, final int offset, final int size) {
final int eob = size + offset;
for (int i = offset; i < eob; i++) {
final int b = fndBit(bits[i]);
if (b != -1) {
return (i * 32) + b;
}
}
return -1;
}
static int fndBit(final int bits) {
if (bits != 0xFFFFFFFF) {
for (int k = 0; k < 32; k++) {
if ((bits & (1 << k)) == 0) {
return k;
}
}
}
return -1;
}
public static class AllocationStats {
public AllocationStats(final int i) {
m_blockSize = i;
}
long m_blockSize;
long m_reservedSlots;
long m_filledSlots;
}
/**
* Utility debug outputing the allocator array, showing index, start
* address and alloc type/size
*
* Collected statistics are against each Allocation Block size:
* total number of slots | store size
* number of filled slots | store used
*
* - AllocatorSize
- The #of bytes in the allocated slots issued by this allocator.
* - AllocatorCount
- The #of fixed allocators for that slot size.
* - SlotsInUse
- The difference between the two previous columns (net slots in use for this slot size).
* - SlotsReserved
- The #of slots in this slot size which have had storage reserved for them.
* - SlotsAllocated
- Cumulative allocation of slots to date in this slot size (regardless of the transaction outcome).
* - SlotsRecycled
- Cumulative recycled slots to date in this slot size (regardless of the transaction outcome).
* - SlotsChurn
- How frequently slots of this size are re-allocated (SlotsInUse/SlotsAllocated).
* - %SlotsUnused
- The percentage of slots of this size which are not in use (1-(SlotsInUse/SlotsReserved)).
* - BytesReserved
- The space reserved on the backing file for those allocation slots
* - BytesAppData
- The #of bytes in the allocated slots which are used by application data (including the record checksum).
* - %SlotWaste
- How well the application data fits in the slots (BytesAppData/(SlotsInUse*AllocatorSize)).
* - %AppData
- How much of your data is stored by each allocator (BytesAppData/Sum(BytesAppData)).
* - %StoreFile
- How much of the backing file is reserved for each allocator (BytesReserved/Sum(BytesReserved)).
* - %StoreWaste
- How much of the total waste on the store is waste for this allocator size ((BytesReserved-BytesAppData)/(Sum(BytesReserved)-Sum(BytesAppData))).
*
* @see StorageStats#showStats(StringBuilder)
*/
public void showAllocators(final StringBuilder str) {
m_storageStats.showStats(str);
str.append("\nChecking regions.....");
// Now check all allocators to confirm that each file region maps to only one allocator
final Lock lock = m_allocationLock.readLock();
lock.lock();
try {
final HashMap map = new HashMap();
for (FixedAllocator fa : m_allocs) {
fa.addToRegionMap(map);
}
str.append("okay\n");
} catch (IllegalStateException is) {
str.append(is.getMessage() + "\n");
} finally {
lock.unlock();
}
}
/**
* Given a physical address (byte offset on the store), return true if that
* address could be managed by an allocated block.
*
* @param a
* the storage address to be tested.
*/
public boolean verify(final long laddr) {
final int addr = (int) laddr;
if (addr == 0) {
return false;
}
return getBlockByAddress(addr) != null;
}
/*****************************************************************************
* Address transformation: latched2Physical
*/
/**
* Return the byte offset in the file.
*
* @param addr
* The latched address.
*
* @return The byte offset in the file.
*/
final private long physicalAddress(final int addr, final boolean nocheck) {
/*
* Guard against concurrent mutation.
*
* Note: Taking the lock here is necessary since physicalAddress/1 is
* public.
*/
final Lock lock = m_allocationReadLock;
lock.lock();
try {
if (addr >= 0) {
return addr & 0xFFFFFFE0;
} else {
// Find the allocator.
final FixedAllocator allocator = getBlock(addr);
// Get the bit index into the allocator.
final int offset = getOffset(addr);
// Translate the bit index into a byte offset on the file.
final long laddr = allocator
.getPhysicalAddress(offset, nocheck);
return laddr;
}
} finally {
lock.unlock();
}
}
/**
* Return the byte offset in the file.
*
* @param addr
* A latched address.
*
* @return The byte offset.
*/
final public long physicalAddress(final int addr) {
return physicalAddress(addr, false/* nocheck */);
}
/********************************************************************************
* handle dual address format, if addr is positive then it is the physical
* address, so the Allocators must be searched.
**/
FixedAllocator getBlockByAddress(final int addr) {
if (addr < 0) {
return getBlock(addr);
}
final Iterator allocs = m_allocs.iterator();
FixedAllocator alloc = null;
while (allocs.hasNext()) {
alloc = allocs.next();
if (alloc.addressInRange(addr)) {
break;
}
alloc = null;
}
return alloc;
}
/**
* Get the {@link FixedAllocator} for a latched address.
*
* @param addr
* The latched address.
*
* @return The {@link FixedAllocator} for that latched address.
*/
private FixedAllocator getBlock(final int addr) {
// index of the FixedAllocator for that latched address.
final int index = (-addr) >>> OFFSET_BITS;
if (index >= m_allocs.size()) {
throw new PhysicalAddressResolutionException(addr);
}
// Return the FixedAllocator for that index.
return m_allocs.get(index);
}
/**
* Return the bit index into a {@link FixedAllocator}.
*
* Note: This is directly encoded by the latched address. You do not need to
* know which {@link FixedAllocator} is being addressed in order to figure
* this out.
*
* @param addr
* A latched address.
*
* @return The bit index into the {@link FixedAllocator}.
*/
private int getOffset(final int addr) {
return (-addr) & OFFSET_BITS_MASK; // OFFSET_BITS
}
/**
* The {@link RWStore} always generates negative address values.
*
* @return whether the address given is a native IStore address
*/
public boolean isNativeAddress(final long addr) {
return addr <= 0;
}
public File getStoreFile() {
return m_fd;
}
public boolean requiresCommit() {
return m_recentAlloc;
}
/**
* Since we need to store the absolute address and the size can be
* a maximum of 64K, the absolute address is limited to 48 bits, setting
* the maximum address as 140T, which is sufficient.
*
* @return long representation of metaBitsAddr PLUS the size
*/
public long getMetaBitsAddr() {
long ret = 0;
if (m_metaBitsAddr < 0) {
ret = physicalAddress((int) m_metaBitsAddr);
} else {
// long ret = physicalAddress((int) m_metaBitsAddr);
ret = convertAddr(-m_metaBitsAddr); // maximum 48 bit address range
}
ret <<= 16;
// include space for version, allocSizes and deferred free info AND
// cDefaultMetaBitsSize
final int metaBitsSize = cMetaHdrFields + m_metaBits.length
+ m_allocSizes.length;
ret += metaBitsSize;
if (log.isTraceEnabled())
log.trace("Returning metabitsAddr: " + ret + ", for "
+ m_metaBitsAddr + " - " + m_metaBits.length + ", "
+ metaBitsSize);
return ret;
}
/**
*
* @return the address of the metaBits
*/
public long getMetaBitsStoreAddress() {
if (m_metaBitsAddr < 0) {
return physicalAddress((int) m_metaBitsAddr);
} else {
return convertAddr(-m_metaBitsAddr); // maximum 48 bit address range
}
}
/**
* @return long representation of metaStartAddr PLUS the size where addr +
* size is fileSize (not necessarily physical size)
*/
public long getMetaStartAddr() {
return -m_fileSize;
}
/**
*
* @return the nextAllocation from the file Heap to be provided to an
* Allocation Block
*/
public long getNextOffset() {
long ret = -m_nextAllocation;
if (m_metaBitsAddr > 0) {
// FIX for sign use in m_metaBitsAddr when packing into long
ret++;
}
ret <<= 32;
ret += -m_metaBitsAddr;
if (log.isTraceEnabled())
log.trace("Returning nextOffset: " + ret + ", for " + m_metaBitsAddr);
return ret;
}
public void flushWrites(final boolean metadata) throws IOException {
assertOpen();
try {
m_writeCacheService.flush(metadata);
// sync the disk.
m_reopener.reopenChannel().force(metadata);
final StoreCounters c = (StoreCounters) storeCounters.get()
.acquire();
try {
c.nforce++;
} finally {
c.release();
}
} catch (InterruptedException e) {
throw new ClosedByInterruptException();
}
}
/** The # of allocation requests made. */
public long getTotalAllocations() {
return m_allocations;
}
/**
* The # of free requests made
*/
public long getTotalFrees() {
return m_frees;
}
/**
* The # of bytes requested - as opposed to the size of the slots allocated.
*/
public long getTotalAllocationsSize() {
return m_nativeAllocBytes;
}
/**
* A Blob Allocator maintains a list of Blob headers. The allocator stores
* up to 255 blob headers plus a checksum. When a request is made to read the
* blob data, the blob allocator retrieves the blob header and reads the
* data from that into the passed byte array.
*/
// public int registerBlob(final int addr) {
// m_allocationLock.lock();
// try {
// BlobAllocator ba = null;
// if (m_freeBlobs.size() > 0) {
// ba = (BlobAllocator) m_freeBlobs.get(0);
// }
// if (ba == null) {
// final Allocator lalloc = (Allocator) m_allocs.get(m_allocs.size() - 1);
// // previous block start address
// final int psa = lalloc.getRawStartAddr();
// assert (psa - 1) > m_nextAllocation;
// ba = new BlobAllocator(this, psa - 1);
// ba.setFreeList(m_freeBlobs); // will add itself to the free list
// ba.setIndex(m_allocs.size());
// m_allocs.add(ba);
// }
//
// if (!m_commitList.contains(ba)) {
// m_commitList.add(ba);
// }
//
// return ba.register(addr);
// } finally {
// m_allocationLock.unlock();
// }
// }
void addToCommit(final FixedAllocator allocator) {
if (allocator.m_prevCommit == null && m_commitHead != allocator) { // not on list
allocator.m_prevCommit = m_commitTail;
if (allocator.m_prevCommit != null) {
allocator.m_prevCommit.m_nextCommit = allocator;
m_commitTail = allocator;
} else {
m_commitHead = m_commitTail = allocator;
}
}
}
final boolean isOnCommitList(final FixedAllocator allocator) {
return allocator.m_prevCommit != null || allocator == m_commitHead;
}
final void clearCommitList() {
FixedAllocator cur = m_commitHead;
while (cur != null) {
final FixedAllocator t = cur;
cur = t.m_nextCommit;
t.m_prevCommit = t.m_nextCommit = null;
}
m_commitHead = m_commitTail = null;
}
final int commitListSize() {
int count = 0;
FixedAllocator cur = m_commitHead;
while (cur != null) {
count++;
cur = cur.m_nextCommit;
}
return count;
}
// void removeFromCommit(final Allocator allocator) {
// m_commitList.remove(allocator);
// }
public Allocator getAllocator(final int i) {
return (Allocator) m_allocs.get(i);
}
/**
* Simple implementation for a {@link RandomAccessFile} to handle the direct
* backing store.
*/
private class ReopenFileChannel implements
IReopenChannel, FileChannelUtility.IAsyncOpener {
final private File file;
private final String mode;
private volatile RandomAccessFile raf;
private final Path path;
private volatile AsynchronousFileChannel asyncChannel;
public ReopenFileChannel(final File file, final RandomAccessFile raf,
final String mode) throws IOException {
this.file = file;
this.mode = mode;
this.raf = raf;
this.path = Paths.get(file.getAbsolutePath());
reopenChannel();
}
public AsynchronousFileChannel getAsyncChannel() {
if (asyncChannel != null) {
if (asyncChannel.isOpen())
return asyncChannel;
}
try {
asyncChannel = AsynchronousFileChannel.open(path, StandardOpenOption.READ);
} catch (IOException e) {
throw new RuntimeException(e);
}
return asyncChannel;
}
public String toString() {
return file.toString();
}
public FileChannel reopenChannel() throws IOException {
/*
* Note: This is basically a double-checked locking pattern. It is
* used to avoid synchronizing when the backing channel is already
* open.
*/
{
final RandomAccessFile tmp = raf;
if (tmp != null) {
final FileChannel channel = tmp.getChannel();
if (channel.isOpen()) {
// The channel is still open.
return channel;
}
}
}
synchronized(this) {
if (raf != null) {
final FileChannel channel = raf.getChannel();
if (channel.isOpen()) {
/*
* The channel is still open. If you are allowing
* concurrent reads on the channel, then this could
* indicate that two readers each found the channel
* closed and that one was able to re-open the channel
* before the other such that the channel was open again
* by the time the 2nd reader got here.
*/
return channel;
}
}
// open the file.
this.raf = new RandomAccessFile(file, mode);
// Update counters.
final StoreCounters c = (StoreCounters) storeCounters
.get().acquire();
try {
c.nreopen++;
} finally {
c.release();
}
return raf.getChannel();
}
}
}
/**
* If the current file extent is different from the required extent then the
* call is made to {@link #extendFile(int)}.
*
* @param extent
* The new file extent.
*/
public void establishExtent(final long extent) {
assertOpen();
final long currentExtent = convertAddr(m_fileSize);
if (extent > currentExtent) {
extendFile(convertFromAddr(extent - currentExtent));
} else if (extent < currentExtent) {
//See https://github.com/SYSTAP/db-enterprise/issues/12
//TODO: Determine if there is a more graceful way to handle this.
// throw new IllegalArgumentException(
log.warn("Cannot shrink RWStore extent: currentExtent="
+ currentExtent + ", fileSize=" + m_fileSize
+ ", newValue=" + extent);
}
}
/**
* @return number of FixedAllocators
*/
public int getFixedAllocatorCount() {
final Lock lock = m_allocationReadLock;
lock.lock();
try {
int fixed = 0;
final Iterator allocs = m_allocs.iterator();
while (allocs.hasNext()) {
if (allocs.next() instanceof FixedAllocator) {
fixed++;
}
}
return fixed;
} finally {
lock.unlock();
}
}
/**
* @return the number of heap allocations made to the FixedAllocators.
*/
public int getAllocatedBlocks() {
final Lock lock = m_allocationReadLock;
lock.lock();
try {
int allocated = 0;
final Iterator allocs = m_allocs.iterator();
while (allocs.hasNext()) {
final Allocator alloc = allocs.next();
if (alloc instanceof FixedAllocator) {
allocated += ((FixedAllocator) alloc).getAllocatedBlocks();
}
}
return allocated;
} finally {
lock.unlock();
}
}
/**
* @return the amount of heap storage assigned to the FixedAllocators.
*/
public long getFileStorage() {
final Lock lock = m_allocationReadLock;
lock.lock();
try {
long allocated = 0;
final Iterator allocs = m_allocs.iterator();
while (allocs.hasNext()) {
final FixedAllocator alloc = allocs.next();
allocated += ((FixedAllocator) alloc).getFileStorage();
}
return allocated;
} finally {
lock.unlock();
}
}
/**
* Computes the amount of utilised storage
*
* @return the amount of storage to alloted slots in the allocation blocks
*/
public long getAllocatedSlots() {
final Lock lock = m_allocationReadLock;
lock.lock();
try {
long allocated = 0;
final Iterator allocs = m_allocs.iterator();
while (allocs.hasNext()) {
final Allocator alloc = allocs.next();
if (alloc instanceof FixedAllocator) {
allocated += ((FixedAllocator) alloc).getAllocatedSlots();
}
}
return allocated;
} finally {
lock.unlock();
}
}
/**
* Adds the address for later freeing to the deferred free list.
*
* If the allocation is for a BLOB then the sze is also stored
*
* The deferred list is checked on AllocBlock and prior to commit.
*
* DeferredFrees are written to the deferred PSOutputStream
*/
public void deferFree(final int rwaddr, final int sze) {
m_allocationWriteLock.lock();
try {
if (sze > (this.m_maxFixedAlloc-4)) {
m_deferredFreeOut.writeInt(-rwaddr);
m_deferredFreeOut.writeInt(sze);
/*
* rather than write out blob address, instead flatten the blob addresses and
* write all to remove the latency on commit caused by reading potentially many blob headers.
*
* This idea was propposed to support BLZG-641/BLZG-1663 to redcue commit latency.
*
* However, it appears that deferFree is not called with the raw blob size and is already
* reduced to the blob part addrs.
*/
log.debug("Unexpected code path deferring free of direct blob address");
// final int alloc = m_maxFixedAlloc-4;
// final int nblocks = (alloc - 1 + (sze-4))/alloc;
// if (nblocks < 0)
// throw new IllegalStateException(
// "Allocation error, m_maxFixedAlloc: "
// + m_maxFixedAlloc);
//
// final byte[] hdrbuf = new byte[4 * (nblocks + 1) + 4]; // plus 4 bytes for checksum
// if (hdrbuf.length > m_maxFixedAlloc) {
// if (log.isInfoEnabled()) {
// log.info("LARGE BLOB - header is BLOB");
// }
// }
//
// getData(rwaddr, hdrbuf); // will work even if header is also a blob
//
// // deferFree header
// deferFree(rwaddr, hdrbuf.length);
//
// // Now read all blob part addresses
// final DataInputStream hdrstr = new DataInputStream(new ByteArrayInputStream(hdrbuf));
// final int rhdrs = hdrstr.readInt();
// if (rhdrs != nblocks) {
// throw new IllegalStateException(
// "Incompatible BLOB header record, expected: "
// + nblocks + ", got: " + rhdrs);
// }
//
// int remaining = sze;
// int partSize = alloc;
// for (int i = 0; i < nblocks; i++) {
// final int blobpartAddr = hdrstr.readInt();
// // deferFree(blobpartAddr, partSize);
// m_deferredFreeOut.writeInt(blobpartAddr);
//
// remaining -= partSize;
//
// if (remaining < partSize) {
// partSize = remaining;
// }
// }
} else {
m_deferredFreeOut.writeInt(rwaddr);
}
} catch (IOException e) {
throw new RuntimeException("Could not free: rwaddr=" + rwaddr
+ ", size=" + sze, e);
} finally {
m_allocationWriteLock.unlock();
}
}
// private void checkFreeable(final JournalTransactionService transactionService) {
// if (transactionService == null) {
// return;
// }
//
// try {
// final Long freeTime = transactionService.tryCallWithLock(new Callable() {
//
// public Long call() throws Exception {
// final long now = transactionService.nextTimestamp();
// final long earliest = transactionService.getEarliestTxStartTime();
// final long aged = now - transactionService.getMinReleaseAge();
//
// if (transactionService.getActiveCount() == 0) {
// return aged;
// } else {
// return aged < earliest ? aged : earliest;
// }
// }
//
// }, 5L, TimeUnit.MILLISECONDS);
// } catch (RuntimeException e) {
// // fine, will try again later
// } catch (Exception e) {
// throw new RuntimeException(e);
// }
// }
public long saveDeferrals() {
m_allocationWriteLock.lock();
try {
if (m_deferredFreeOut.getBytesWritten() == 0) {
return 0;
}
m_deferredFreeOut.writeInt(0); // terminate!
final int outlen = m_deferredFreeOut.getBytesWritten();
long addr = m_deferredFreeOut.save();
addr <<= 32;
addr += outlen;
// Ensure added to blob allocation stats: BLZG-1646
if (outlen > this.m_maxFixedAlloc && m_storageStats != null) {
m_storageStats.allocateBlob(outlen);
}
m_deferredFreeOut.reset();
return addr;
} catch (IOException e) {
throw new RuntimeException("Cannot write to deferred free", e);
} finally {
m_allocationWriteLock.unlock();
}
}
/**
* Provided with the address of a block of addresses to be freed
* @param blockAddr
* @return the total number of addresses freed
*/
private int freeDeferrals(final long blockAddr, final long lastReleaseTime) {
final int addr = (int) (blockAddr >> 32);
final int sze = (int) blockAddr & 0xFFFFFFFF; // Resolution for BLZG-1236 (recycler error)
if (log.isTraceEnabled())
log.trace("freeDeferrals at " + physicalAddress(addr) + ", size: " + sze + " releaseTime: " + lastReleaseTime);
final byte[] buf = new byte[sze+4]; // allow for checksum
getData(addr, buf);
final DataInputStream strBuf = new DataInputStream(new ByteArrayInputStream(buf));
m_allocationWriteLock.lock();
int totalFreed = 0;
try {
int nxtAddr = strBuf.readInt();
int cnt = 0;
while (nxtAddr != 0) { // while (false && addrs-- > 0) {
if (nxtAddr > 0) { // Blob
final int bloblen = strBuf.readInt();
assert bloblen > 0; // a Blob address MUST have a size
immediateFree(-nxtAddr, bloblen);
} else {
// The lack of size messes with the stats
immediateFree(nxtAddr, 1); // size ignored for FixedAllocators
}
totalFreed++;
nxtAddr = strBuf.readInt();
}
// now free delete block
immediateFree(addr, sze);
m_lastDeferredReleaseTime = lastReleaseTime;
if (log.isTraceEnabled())
log.trace("Updated m_lastDeferredReleaseTime="
+ m_lastDeferredReleaseTime);
} catch (IOException e) {
throw new RuntimeException("Problem freeing deferrals", e);
} finally {
m_allocationWriteLock.unlock();
}
return totalFreed;
}
/**
* Provided with an iterator of CommitRecords, process each and free any
* deferred deletes associated with each.
*
* @param journal
* @param fromTime
* The inclusive lower bound.
* @param toTime
* The exclusive upper bound.
*/
private int freeDeferrals(final AbstractJournal journal,
final long fromTime,
final long toTime) {
final ITupleIterator commitRecords;
/*
* Commit can be called prior to Journal initialisation, in which
* case the commitRecordIndex will not be set.
*/
final IIndex commitRecordIndex = journal.getReadOnlyCommitRecordIndex();
if (commitRecordIndex == null) { // TODO Why is this here?
return 0;
}
final IndexMetadata metadata = commitRecordIndex
.getIndexMetadata();
final byte[] fromKey = metadata.getTupleSerializer()
.serializeKey(fromTime);
final byte[] toKey = metadata.getTupleSerializer()
.serializeKey(toTime);
commitRecords = commitRecordIndex
.rangeIterator(fromKey, toKey);
int totalFreed = 0;
int commitPointsRecycled = 0;
while (commitRecords.hasNext()) {
final ITuple tuple = commitRecords.next();
final CommitRecordIndex.Entry entry = tuple.getObject();
try {
final ICommitRecord record = CommitRecordSerializer.INSTANCE
.deserialize(journal.read(entry.addr));
final long blockAddr = record
.getRootAddr(AbstractJournal.DELETEBLOCK);
if (blockAddr != 0) {
totalFreed += freeDeferrals(blockAddr,
record.getTimestamp());
}
// Note: This is releasing the ICommitRecord itself. I've moved the responsibilty
// for that into AbstractJournal#removeCommitRecordEntries() (invoked below).
//
// immediateFree((int) (entry.addr >> 32), (int) entry.addr);
commitPointsRecycled++;
} catch (RuntimeException re) {
throw new RuntimeException("Problem with entry at "
+ entry.addr, re);
}
}
/*
*
*
* @see https://sourceforge.net/apps/trac/bigdata/ticket/440
*/
// Now remove the commit record entries from the commit record index.
final int commitPointsRemoved = journal.removeCommitRecordEntries(
fromKey, toKey);
if (txLog.isInfoEnabled())
txLog.info("RECYCLED: fromTime=" + fromTime + ", toTime=" + toTime
+ ", totalFreed=" + totalFreed
+ ", commitPointsRecycled=" + commitPointsRecycled
+ ", commitPointsRemoved=" + commitPointsRemoved
);
if (commitPointsRecycled != commitPointsRemoved)
throw new AssertionError("commitPointsRecycled="
+ commitPointsRecycled + " != commitPointsRemoved="
+ commitPointsRemoved);
return totalFreed;
}
/**
* {@inheritDoc}
*
* The {@link ContextAllocation} object manages a freeList of associated
* allocators and an overall list of allocators. When the context is
* detached, all allocators must be released and any that has available
* capacity will be assigned to the global free lists. See
* {@link AllocBlock #releaseSession}
*
* @param context
* The context to be released from all {@link FixedAllocator}s.
*/
public void detachContext(final IAllocationContext context) {
assertOpen();
m_allocationWriteLock.lock();
try {
context.release();
if (context.isIsolated()) {
final ContextAllocation alloc = m_contexts.remove(context);
if (alloc != null) {
alloc.release();
} else {
throw new IllegalStateException("Multiple call to detachContext");
}
if (m_contexts.isEmpty() && this.m_activeTxCount == 0) {
releaseSessions();
}
}
} finally {
m_allocationWriteLock.unlock();
}
}
/**
* The ContextAllocation object manages a freeList of associated allocators
* and an overall list of allocators. When the context is aborted then
* allocations made by that context should be released.
* See {@link AllocBlock #abortShadow}
*
* @param context
* The context to be released from all FixedAllocators.
*/
public void abortContext(final IAllocationContext context) {
assertOpen();
m_allocationWriteLock.lock();
try {
context.release();
if (context.isIsolated()) {
final ContextAllocation alloc = m_contexts.remove(context);
if (alloc != null) {
alloc.abort();
}
}
} finally {
m_allocationWriteLock.unlock();
}
}
/**
* The ContextAllocation class manages a set of Allocators.
*
* A ContextAllocation can have a parent ContextAllocation such that when
* it is released it will transfer its Allocators to its parent.
*
*
*
* @author Martyn Cutcher
*
*/
static class ContextAllocation {
private final RWStore m_store;
private final ArrayList m_freeFixed[];
private final ArrayList m_allFixed;
private final ArrayList m_deferredFrees;
// lists of free blob allocators
// private final ArrayList m_freeBlobs;
private final ContextAllocation m_parent;
private final IAllocationContext m_context;
@SuppressWarnings("unchecked")
ContextAllocation(final RWStore store,
final int fixedBlocks,
final ContextAllocation parent,
final IAllocationContext acontext) {
m_store = store;
m_parent = parent;
m_context = acontext;
m_freeFixed = new ArrayList[fixedBlocks];
for (int i = 0; i < m_freeFixed.length; i++) {
m_freeFixed[i] = new ArrayList();
}
m_allFixed = new ArrayList();
m_deferredFrees = new ArrayList();
// m_freeBlobs = new ArrayList();
}
/**
* For frees made against a shadowed FixedAlocator that is NOT owned
* by the context, the physical free must be deferred until the
* context is deshadowed or aborted.
*
* @param encodeAddr
*/
public void deferFree(final long encodeAddr) {
m_deferredFrees.add(encodeAddr);
}
/**
* Must return the shadowed allocators to the parent/global
* environment, resetting the freeList association.
*/
void release() {
final ArrayList freeFixed[] = m_parent != null ? m_parent.m_freeFixed
: m_store.m_freeFixed;
final IAllocationContext pcontext = m_parent == null ? null
: m_parent.m_context;
for (FixedAllocator f : m_allFixed) {
f.setAllocationContext(pcontext);
// will add to free list if required
f.setFreeList(freeFixed[m_store.fixedAllocatorIndex(f.m_size)]);
}
// for (int i = 0; i < m_freeFixed.length; i++) {
// freeFixed[i].addAll(m_freeFixed[i]);
// }
// freeBlobs.addAll(m_freeBlobs);
// now free all deferred frees made within this context for other
// allocators
if (log.isDebugEnabled())
log.debug("Releasing " + m_deferredFrees.size() + " deferred frees");
final boolean defer = m_store.m_minReleaseAge > 0 || m_store.m_activeTxCount > 0 || m_store.m_contexts.size() > 0;
for (Long l : m_deferredFrees) {
final int addr = (int) (l >> 32);
final int sze = l.intValue();
if (defer) {
m_store.deferFree(addr, sze);
} else {
m_store.immediateFree(addr, sze);
}
}
m_deferredFrees.clear();
}
void abort() {
final ArrayList freeFixed[] = m_parent != null ? m_parent.m_freeFixed
: m_store.m_freeFixed;
final IAllocationContext pcontext = m_parent == null ? null
: m_parent.m_context;
for (FixedAllocator f : m_allFixed) {
f.abortAllocationContext(pcontext, m_store.m_writeCacheService);
f.setFreeList(freeFixed[m_store.fixedAllocatorIndex(f.m_size)]);
}
if (log.isDebugEnabled())
log.debug("Aborting " + m_deferredFrees.size() + " deferred frees");
m_deferredFrees.clear();
}
FixedAllocator getFreeFixed(final int i) {
final ArrayList free = m_freeFixed[i];
if (free.size() == 0) {
final FixedAllocator falloc = establishFixedAllocator(i);
if (falloc.m_pendingContextCommit) {
throw new IllegalStateException("Allocator on free list while pendingContextCommit");
}
falloc.setAllocationContext(m_context);
// The normal check for adding to the free list is whether to return to the free list,
// but in this case, we are moving to another free list, so we should not need to
// check for the smallAllocation threshold.
falloc.setFreeList(free, true/*force*/);
if (free.size() == 0 ) {
throw new IllegalStateException("Free list should not be empty, pendingContextCommit: " + falloc.m_pendingContextCommit);
}
m_allFixed.add(falloc);
}
return free.get(0); // take first in list
}
/**
*
* @param i - the block-index for the allocator required
* @return
*/
FixedAllocator establishFixedAllocator(final int i) {
if (m_parent == null) {
return m_store.establishFreeFixedAllocator(i);
} else {
return m_parent.establishFixedAllocator(i);
}
}
}
/**
* A map of the {@link IAllocationContext}s.
*
* Note: This map must be thread-safe since it is referenced from various
* methods outside of the governing {@link #m_allocationLock}.
*/
private final Map m_contexts =
new ConcurrentHashMap();
private ContextAllocation getContextAllocation(
final IAllocationContext context) {
/*
* The allocation lock MUST be held to make changes in the membership of
* m_contexts atomic with respect to free().
*/
assert m_allocationWriteLock.isHeldByCurrentThread();
ContextAllocation ret = m_contexts.get(context);
if (ret == null) {
// This is no longer a valid state
throw new IllegalStateException("No associated ContextAllocation");
// ret = new ContextAllocation(this, m_freeFixed.length, null, context);
//
// if (m_contexts.put(context, ret) != null) {
//
// throw new AssertionError();
//
// }
//
// if (log.isTraceEnabled())
// log.trace("Establish ContextAllocation: " + ret
// + ", total: " + m_contexts.size()
// + ", requests: " + ++m_contextRequests
// + ", removals: " + m_contextRemovals
// + ", allocators: " + m_allocs.size() );
//
//
// if (log.isInfoEnabled())
// log.info("Context: ncontexts=" + m_contexts.size()
// + ", context=" + context);
}
return ret;
}
public int getSlotSize(final int data_len) {
int i = 0;
int ret = m_minFixedAlloc;
while (data_len > ret) {
i++;
// If we write directly to the writeCache then the data_len
// may be larger than largest slot
if (i == m_allocSizes.length)
return data_len;
ret = 64 * m_allocSizes[i];
}
return ret;
}
/**
* The maximum allocation size (bytes).
*/
public int getMaxAllocSize() {
return m_maxFixedAlloc;
}
/**
* This can be called as part of the HA downstream replication.
*
* FIXME: If part of downstream replication then the current metabits
* held by the RWStore will not be in sync with that stored on disk.
*
* This will only be a problem if the RWStore needs to take over as
* leader and be able to allocate and write to the store.
*
* Note that the metabits are not needed in order to determine the
* physical address mapping of an rw-native address.
*
* @param rootBlock
* @param forceOnCommit
*/
public void writeRootBlock(final IRootBlockView rootBlock,
final ForceEnum forceOnCommit) {
if (rootBlock == null)
throw new IllegalArgumentException();
checkRootBlock(rootBlock);
assertOpen();
if (log.isTraceEnabled()) {
log.trace("Writing new rootblock with commitCounter: "
+ rootBlock.getCommitCounter() + ", commitRecordAddr: "
+ rootBlock.getCommitRecordAddr()
+ ", commitRecordIndexAddr: "
+ rootBlock.getCommitRecordIndexAddr());
}
try {
final ByteBuffer data = rootBlock.asReadOnlyBuffer();
final long pos = rootBlock.isRootBlock0()
? FileMetadata.OFFSET_ROOT_BLOCK0
: FileMetadata.OFFSET_ROOT_BLOCK1;
/*
* Note: This uses the [opener] to automatically retry the operation
* in case concurrent readers are interrupting, causing an
* asynchronous close of the backing channel.
*/
// Note: extensionLock required for file IO.
final Lock lock = m_extensionLock.readLock();
lock.lock();
try {
// Update the root block.
FileChannelUtility.writeAll(m_reopener, data, pos);
/*
* Generally, you want to force the file data to the disk here.
* The file metadata MIGHT not matter since we always force it
* to the disk when we change the file size (unless the file
* system updates other aspects of file metadata during normal
* writes).
*/
// sync the disk.
m_reopener.reopenChannel().force(forceOnCommit == ForceEnum.ForceMetadata);
// Update counters.
final StoreCounters c = (StoreCounters) storeCounters.get()
.acquire();
try {
c.nwriteRootBlock++;
} finally {
c.release();
}
// ensure cached commitNextAllocation
if (m_committedNextAllocation != m_nextAllocation ) {
if (log.isTraceEnabled())
log.trace("Updating committedNextAllocation from writeRootBlock");
m_committedNextAllocation = m_nextAllocation;
}
} finally {
lock.unlock();
}
} catch (IOException ex) {
throw new RuntimeException(ex);
}
if (log.isDebugEnabled())
log.debug("wrote root block: "+rootBlock);
}
public ByteBuffer readRootBlock(final boolean rootBlock0) {
assertOpen();
// assertNoRebuild();
final ByteBuffer tmp = ByteBuffer
.allocate(RootBlockView.SIZEOF_ROOT_BLOCK);
// Guard IO against concurrent file extension.
final Lock lock = m_extensionLock.readLock();
lock.lock();
try {
FileChannelUtility.readAll(m_reopener, tmp,
rootBlock0 ? FileMetadata.OFFSET_ROOT_BLOCK0
: FileMetadata.OFFSET_ROOT_BLOCK1);
tmp.position(0); // resets the position.
} catch (IOException ex) {
throw new RuntimeException(ex);
} finally {
lock.unlock();
}
return tmp;
}
/**
* Striped performance counters for {@link IRawStore} access, including
* operations that read or write through to the underlying media.
*
* Note: The performance counters for writes to the disk are reported by the
* {@link WriteCacheService}. The {@link RWStore} never writes directly onto
* the disk (other than the root blocks).
*
* @author Bryan
* Thompson
* @param
*
* @todo report elapsed time and average latency for force, reopen, and
* writeRootBlock.
*
* FIXME CAT may be much faster than striped locks (2-3x faster).
*/
static public class StoreCounters> extends
StripedCounters {
/**
* #of read requests.
*/
public volatile long nreads;
/**
* #of read requests that read through to the backing file.
*/
public volatile long ndiskRead;
/**
* #of bytes read.
*/
public volatile long bytesRead;
/**
* #of bytes that have been read from the disk.
*/
public volatile long bytesReadFromDisk;
/**
* Total elapsed time for reads.
*/
public volatile long elapsedReadNanos;
/**
* Total elapsed time for reading on the disk.
*/
public volatile long elapsedDiskReadNanos;
/**
* The #of checksum errors while reading on the local disk.
*/
public volatile long checksumErrorCount;
/**
* #of write requests.
*/
public volatile long nwrites;
// This is reported by the WriteCacheService.
// /**
// * #of write requests that write through to the backing file.
// */
// public volatile long ndiskWrite;
/**
* The size of the largest record read.
*/
public volatile long maxReadSize;
/**
* The size of the largest record written.
*/
public volatile long maxWriteSize;
/**
* #of bytes written.
*/
public volatile long bytesWritten;
// This is reported by the WriteCacheService.
// /**
// * #of bytes that have been written on the disk.
// */
// public volatile long bytesWrittenOnDisk;
/**
* Total elapsed time for writes.
*/
public volatile long elapsedWriteNanos;
// This is reported by the WriteCacheService.
// /**
// * Total elapsed time for writing on the disk.
// */
// public volatile long elapsedDiskWriteNanos;
/**
* #of times the data were forced to the disk.
*/
public volatile long nforce;
/**
* #of times the length of the file was changed (typically, extended).
*/
public volatile long ntruncate;
/**
* #of times the file has been reopened after it was closed by an
* interrupt.
*/
public volatile long nreopen;
/**
* #of times one of the root blocks has been written.
*/
public volatile long nwriteRootBlock;
/**
* buffer counters
*/
public volatile long bufferDataBytes;
public volatile long bufferDataWrites;
public volatile long bufferFileWrites;
/**
* {@inheritDoc}
*/
public StoreCounters() {
super();
}
/**
* {@inheritDoc}
*/
public StoreCounters(final int batchSize) {
super(batchSize);
}
/**
* {@inheritDoc}
*/
public StoreCounters(final int nstripes, final int batchSize) {
super(nstripes, batchSize);
}
@Override
public void add(final T o) {
super.add(o);
nreads += o.nreads;
ndiskRead += o.ndiskRead;
bytesRead += o.bytesRead;
bytesReadFromDisk += o.bytesReadFromDisk;
maxReadSize = Math.max(maxReadSize, o.maxReadSize);
elapsedReadNanos += o.elapsedReadNanos;
elapsedDiskReadNanos += o.elapsedDiskReadNanos;
checksumErrorCount += o.checksumErrorCount;
nwrites += o.nwrites;
// ndiskWrite += o.ndiskWrite;
maxWriteSize = Math.max(maxWriteSize, o.maxWriteSize);
bytesWritten += o.bytesWritten;
// bytesWrittenOnDisk += o.bytesWrittenOnDisk;
elapsedWriteNanos += o.elapsedWriteNanos;
// elapsedDiskWriteNanos += o.elapsedDiskWriteNanos;
nforce += o.nforce;
ntruncate += o.ntruncate;
nreopen += o.nreopen;
nwriteRootBlock += o.nwriteRootBlock;
}
@Override
public T subtract(final T o) {
// make a copy of the current counters.
final T t = super.subtract(o);
// subtract out the given counters.
t.nreads -= o.nreads;
t.ndiskRead -= o.ndiskRead;
t.bytesRead -= o.bytesRead;
t.bytesReadFromDisk -= o.bytesReadFromDisk;
t.maxReadSize -= o.maxReadSize; // @todo report max? min?
t.elapsedReadNanos -= o.elapsedReadNanos;
t.elapsedDiskReadNanos -= o.elapsedDiskReadNanos;
t.checksumErrorCount -= o.checksumErrorCount;
t.nwrites -= o.nwrites;
// t.ndiskWrite -= o.ndiskWrite;
t.maxWriteSize -= o.maxWriteSize; // @todo report max? min?
t.bytesWritten -= o.bytesWritten;
// t.bytesWrittenOnDisk -= o.bytesWrittenOnDisk;
t.elapsedWriteNanos -= o.elapsedWriteNanos;
// t.elapsedDiskWriteNanos -= o.elapsedDiskWriteNanos;
t.nforce -= o.nforce;
t.ntruncate -= o.ntruncate;
t.nreopen -= o.nreopen;
t.nwriteRootBlock -= o.nwriteRootBlock;
return t;
}
@Override
public void clear() {
// subtract out the given counters.
nreads = 0;
ndiskRead = 0;
bytesRead = 0;
bytesReadFromDisk = 0;
maxReadSize = 0;
elapsedReadNanos = 0;
elapsedDiskReadNanos = 0;
checksumErrorCount = 0;
nwrites = 0;
// ndiskWrite = 0;
maxWriteSize = 0;
bytesWritten = 0;
// bytesWrittenOnDisk = 0;
elapsedWriteNanos = 0;
// elapsedDiskWriteNanos = 0;
nforce = 0;
ntruncate = 0;
nreopen = 0;
nwriteRootBlock = 0;
}
@Override
public CounterSet getCounters() {
final CounterSet root = super.getCounters();
// IRawStore API
{
/*
* reads
*/
root.addCounter("nreads", new Instrument() {
public void sample() {
setValue(nreads);
}
});
root.addCounter("bytesRead", new Instrument() {
public void sample() {
setValue(bytesRead);
}
});
root.addCounter("readSecs", new Instrument() {
public void sample() {
final double elapsedReadSecs = (elapsedReadNanos / 1000000000.);
setValue(elapsedReadSecs);
}
});
root.addCounter("bytesReadPerSec", new Instrument() {
public void sample() {
final double readSecs = (elapsedReadNanos / 1000000000.);
final double bytesReadPerSec = (readSecs == 0L ? 0d
: (bytesRead / readSecs));
setValue(bytesReadPerSec);
}
});
root.addCounter("maxReadSize", new Instrument() {
public void sample() {
setValue(maxReadSize);
}
});
root.addCounter("checksumErrorCount", new Instrument() {
public void sample() {
setValue(checksumErrorCount);
}
});
/*
* writes
*/
root.addCounter("nwrites", new Instrument() {
public void sample() {
setValue(nwrites);
}
});
root.addCounter("bytesWritten", new Instrument() {
public void sample() {
setValue(bytesWritten);
}
});
root.addCounter("writeSecs", new Instrument() {
public void sample() {
final double writeSecs = (elapsedWriteNanos / 1000000000.);
setValue(writeSecs);
}
});
root.addCounter("bytesWrittenPerSec", new Instrument() {
public void sample() {
final double writeSecs = (elapsedWriteNanos / 1000000000.);
final double bytesWrittenPerSec = (writeSecs == 0L ? 0d
: (bytesWritten / writeSecs));
setValue(bytesWrittenPerSec);
}
});
root.addCounter("maxWriteSize", new Instrument() {
public void sample() {
setValue(maxWriteSize);
}
});
} // IRawStore
// BufferedWriter
final CounterSet bc = root.makePath("buffer");
bc.addCounter("ndataWrites", new Instrument() {
public void sample() {
setValue(bufferDataWrites);
}
});
bc.addCounter("nfileWrites", new Instrument() {
public void sample() {
setValue(bufferFileWrites);
}
});
// disk statistics
{
final CounterSet disk = root.makePath("disk");
/*
* read
*/
disk.addCounter("nreads", new Instrument() {
public void sample() {
setValue(ndiskRead);
}
});
disk.addCounter("bytesRead", new Instrument() {
public void sample() {
setValue(bytesReadFromDisk);
}
});
disk.addCounter("bytesPerRead", new Instrument() {
public void sample() {
final double bytesPerDiskRead = (ndiskRead == 0 ? 0d
: (bytesReadFromDisk / (double) ndiskRead));
setValue(bytesPerDiskRead);
}
});
disk.addCounter("readSecs", new Instrument() {
public void sample() {
final double diskReadSecs = (elapsedDiskReadNanos / 1000000000.);
setValue(diskReadSecs);
}
});
disk.addCounter("bytesReadPerSec", new Instrument() {
public void sample() {
final double diskReadSecs = (elapsedDiskReadNanos / 1000000000.);
final double bytesReadPerSec = (diskReadSecs == 0L ? 0d
: bytesReadFromDisk / diskReadSecs);
setValue(bytesReadPerSec);
}
});
disk.addCounter("secsPerRead", new Instrument() {
public void sample() {
final double diskReadSecs = (elapsedDiskReadNanos / 1000000000.);
final double readLatency = (diskReadSecs == 0 ? 0d
: diskReadSecs / ndiskRead);
setValue(readLatency);
}
});
/*
* write
*/
// disk.addCounter("nwrites", new Instrument() {
// public void sample() {
// setValue(ndiskWrite);
// }
// });
//
// disk.addCounter("bytesWritten", new Instrument() {
// public void sample() {
// setValue(bytesWrittenOnDisk);
// }
// });
//
// disk.addCounter("bytesPerWrite", new Instrument() {
// public void sample() {
// final double bytesPerDiskWrite = (ndiskWrite == 0 ? 0d
// : (bytesWrittenOnDisk / (double) ndiskWrite));
// setValue(bytesPerDiskWrite);
// }
// });
//
// disk.addCounter("writeSecs", new Instrument() {
// public void sample() {
// final double diskWriteSecs = (elapsedDiskWriteNanos / 1000000000.);
// setValue(diskWriteSecs);
// }
// });
//
// disk.addCounter("bytesWrittenPerSec", new Instrument() {
// public void sample() {
// final double diskWriteSecs = (elapsedDiskWriteNanos / 1000000000.);
// final double bytesWrittenPerSec = (diskWriteSecs == 0L ? 0d
// : bytesWrittenOnDisk / diskWriteSecs);
// setValue(bytesWrittenPerSec);
// }
// });
//
// disk.addCounter("secsPerWrite", new Instrument() {
// public void sample() {
// final double diskWriteSecs = (elapsedDiskWriteNanos / 1000000000.);
// final double writeLatency = (diskWriteSecs == 0 ? 0d
// : diskWriteSecs / ndiskWrite);
// setValue(writeLatency);
// }
// });
/*
* other
*/
disk.addCounter("nforce", new Instrument() {
public void sample() {
setValue(nforce);
}
});
disk.addCounter("nextend", new Instrument() {
public void sample() {
setValue(ntruncate);
}
});
disk.addCounter("nreopen", new Instrument() {
public void sample() {
setValue(nreopen);
}
});
disk.addCounter("rootBlockWrites", new Instrument() {
public void sample() {
setValue(nwriteRootBlock);
}
});
} // disk
return root;
} // getCounters()
} // class StoreCounters
/**
* Striped performance counters for this class.
*/
@SuppressWarnings("unchecked")
private final AtomicReference storeCounters = new AtomicReference();
/**
* Returns the striped performance counters for the store.
*/
public StoreCounters getStoreCounters() {
return storeCounters.get();
}
/**
* Replaces the {@link StoreCounters} object.
*
* @param storeCounters
* The new {@link Counter}s.
*
* @throws IllegalArgumentException
* if the argument is null.
*/
public void setStoreCounters(final StoreCounters storeCounters) {
if (storeCounters == null)
throw new IllegalArgumentException();
this.storeCounters.set(storeCounters);
}
/**
* Return interesting information about the write cache and file operations.
*
* @todo allocations data? user extent allocated? user extent used? etc.
*/
public CounterSet getCounters() {
final CounterSet root = new CounterSet();
// root.addCounter("nextOffset", new Instrument() {
// public void sample() {
// setValue(nextOffset.get());
// }
// });
root.addCounter("extent", new Instrument() {
public void sample() {
setValue(getStoreFile().length());
}
});
// attach the most recently updated values from the striped counters.
root.attach(storeCounters.get().getCounters());
if (m_writeCacheService != null) {
final CounterSet tmp = root.makePath("writeCache");
tmp.attach(m_writeCacheService.getCounters());
}
return root;
}
public void writeRawBuffer(final IHAWriteMessage msg, final IBufferAccess b)
throws IOException, InterruptedException {
// expand buffer before writing on the store.
final ByteBuffer xb = msg.expand(b.buffer());
if (log.isTraceEnabled()) {
log.trace("expanded buffer, position: " + xb.position()
+ ", limit: " + xb.limit());
}
final IBufferAccess ba = new IBufferAccess() {
@Override
public ByteBuffer buffer() {
return xb;
}
@Override
public void release() throws InterruptedException {
}
@Override
public void release(long timeout, TimeUnit unit)
throws InterruptedException {
}
};
/*
* Wrap up the data from the message as a WriteCache object. This will
* build up a RecordMap containing the allocations to be made, and
* including a ZERO (0) data length if any offset winds up being deleted
* (released).
*
* Note: We do not need to pass in the compressorKey here. It is ignored
* by WriteCache.flush(). We have expanded the payload above. Now we are
* just flushing the write cache onto the disk.
*/
final WriteCache writeCache = m_writeCacheService.newWriteCache(ba,
true/* useChecksums */, true/* bufferHasData */, m_reopener,
msg.getFileExtent());
// Ensure that replicated buffers are not compacted.
writeCache.closeForWrites();
/*
* Setup buffer for writing. We receive the buffer with pos=0,
* limit=#ofbyteswritten. However, flush() expects pos=limit, will
* clear pos to zero and then write bytes up to the limit. So,
* we set the position to the limit before calling flush.
*/
final ByteBuffer bb = ba.buffer();
final int limit = bb.limit();
bb.position(limit);
/*
* Flush the scattered writes in the write cache to the backing store.
*
* Note: WriteCacheImpl.writeOnChannel() will take the extensionLock for
* the IO against the channel.
*/
// final Lock lock = m_allocationReadLock; // TODO Is the allocation lock required here? I doubt it.
// lock.lock();
// try {
// Flush writes.
writeCache.flush(false/* force */);
// } finally {
// lock.unlock();
// }
// install reads into readCache (if any)
m_writeCacheService.installReads(writeCache);
}
public Future sendHALogBuffer(final IHALogRequest req,
final IHAWriteMessage msg, final IBufferAccess buf)
throws IOException, InterruptedException {
final ByteBuffer b = buf.buffer();
assert b.remaining() > 0 : "Empty buffer: " + b;
@SuppressWarnings("unchecked")
final QuorumPipeline quorumMember = (QuorumPipeline) m_quorum
.getMember();
final Future remoteWriteFuture = quorumMember.replicate(req, msg, b);
return remoteWriteFuture;
}
/**
* @see IHABufferStrategy#sendRawBuffer(IHARebuildRequest, long,
* long, long, long, int, ByteBuffer)
*/
public Future sendRawBuffer(final IHARebuildRequest req,
// final long commitCounter, final long commitTime,
final long sequence, final long quorumToken, final long fileExtent,
final long offset, final int nbytes, final ByteBuffer b)
throws IOException, InterruptedException {
// read direct from store
final ByteBuffer clientBuffer = b;
clientBuffer.position(0);
clientBuffer.limit(nbytes);
readRaw(/*nbytes,*/ offset, clientBuffer);
assert clientBuffer.remaining() > 0 : "Empty buffer: " + clientBuffer;
@SuppressWarnings("unchecked")
final QuorumPipeline quorumMember = (QuorumPipeline) m_quorum
.getMember();
final int chk = ChecksumUtility.threadChk.get().checksum(b);
final IHAWriteMessage msg = new HAWriteMessage(m_storeUUID,
-1L/* commitCounter */, -1L/* commitTime */, sequence, nbytes,
chk, StoreTypeEnum.RW, quorumToken, fileExtent, offset/* firstOffset */);
final Future remoteWriteFuture = quorumMember.replicate(req, msg,
clientBuffer);
return remoteWriteFuture;
}
public void writeOnStream(final OutputStream os, final ISnapshotData snapshotData,
final Quorum> quorum, final long token)
throws IOException, QuorumException, InterruptedException {
// final FileInputStream filein = new FileInputStream(this.m_fd);
final FileChannelUtility.ReopenerInputStream filein = new FileChannelUtility.ReopenerInputStream(m_reopener);
try {
MergeStreamWithSnapshotData.process(filein, snapshotData, os);
} finally {
filein.close();
}
if (quorum!=null&&!quorum.getClient().isJoinedMember(token)) {
// See #1172
throw new QuorumException();
}
}
public void writeOnStream2(final OutputStream os, final Set> snapshotData,
final Quorum> quorum, final long token)
throws IOException, QuorumException {
IBufferAccess buf = null;
try {
try {
// Acquire a buffer.
buf = DirectBufferPool.INSTANCE.acquire();
} catch (InterruptedException ex) {
// Wrap and re-throw.
throw new IOException(ex);
}
// The backing ByteBuffer.
final ByteBuffer b = buf.buffer();
// The capacity of that buffer (typically 1MB).
final int bufferCapacity = b.capacity();
// A big enough byte[].
final byte[] a = new byte[bufferCapacity];
// The size of the root blocks (which we skip).
final int headerSize = FileMetadata.headerSize0;
/*
* The size of the file at the moment we begin. We will not
* replicate data on new extensions of the file. Those data will
* be captured by HALog files that are replayed by the service
* that is doing the rebuild.
*/
// final long fileExtent = getExtent();
final long fileExtent = getStoreFile().length();
// The #of bytes to be transmitted.
final long totalBytes = fileExtent - headerSize;
// The #of bytes remaining.
long remaining = totalBytes;
// The offset from which data is retrieved.
long offset = headerSize;
long sequence = 0L;
if (log.isInfoEnabled())
log.info("Writing on stream: nbytes=" + totalBytes);
while (remaining > 0) {
int nbytes = (int) Math.min((long) bufferCapacity,
remaining);
if (sequence == 0L && nbytes == bufferCapacity
&& remaining > bufferCapacity) {
/*
* Adjust the first block so the remainder will be
* aligned on the bufferCapacity boundaries (IO
* efficiency).
*/
nbytes -= headerSize;
}
if (log.isDebugEnabled())
log.debug("Writing block: sequence=" + sequence
+ ", offset=" + offset + ", nbytes=" + nbytes);
// read direct from store
final ByteBuffer clientBuffer = b;
clientBuffer.position(0);
clientBuffer.limit(nbytes);
readRaw(/*nbytes,*/ offset, clientBuffer);
assert clientBuffer.remaining() > 0 : "Empty buffer: " + clientBuffer;
if (BytesUtil
.toArray(clientBuffer, false/* forceCopy */, a/* dst */) != a) {
// Should have copied into our array.
throw new AssertionError();
}
// write onto the stream.
os.write(a, 0/* off */, nbytes/* len */);
remaining -= nbytes;
offset += nbytes;
sequence++;
if (!quorum.getClient().isJoinedMember(token))
throw new QuorumException();
}
if (log.isInfoEnabled())
log.info("Wrote on stream: #blocks=" + sequence + ", #bytes="
+ (fileExtent - headerSize));
} finally {
if (buf != null) {
try {
// Release the direct buffer.
buf.release();
} catch (InterruptedException e) {
log.warn(e);
}
}
}
}
/**
* Read on the backing file. {@link ByteBuffer#remaining()} bytes will be
* read into the caller's buffer, starting at the specified offset in the
* backing file.
*
* @param offset
* The offset of the first byte (relative to the start of the
* data region).
* @param dst
* Where to put the data. Bytes will be written at position until
* limit.
*
* @return The caller's buffer, prepared for reading.
*/
public ByteBuffer readRaw(final long offset, final ByteBuffer dst) {
// Guard against concurrent file extension.
final Lock lock = m_extensionLock.readLock();
lock.lock();
try {
final int position = dst.position();
try {
final long beginDisk = System.nanoTime();
// the offset into the disk file.
// final long pos = FileMetadata.headerSize0 + offset;
final long pos = offset;
final int length = dst.limit();
// read on the disk.
final int ndiskRead = FileChannelUtility.readAll(m_reopener,
dst, pos);
m_diskReads += ndiskRead;
final long now = System.nanoTime();
// update performance counters.
final StoreCounters c = (StoreCounters) storeCounters
.get().acquire();
try {
c.ndiskRead += ndiskRead;
final int nbytes = length;
c.nreads++;
c.bytesRead += nbytes;
c.bytesReadFromDisk += nbytes;
c.elapsedReadNanos += now - beginDisk;
c.elapsedDiskReadNanos += now - beginDisk;
} finally {
c.release();
}
} catch (IOException ex) {
throw new RuntimeException(ex);
}
// reset for reading
dst.position(position);
return dst;
} finally {
lock.unlock();
}
}
public int getMaxBlobSize() {
return m_maxBlobAllocSize-4; // allow for checksum
}
public StorageStats getStorageStats() {
return m_storageStats;
}
private final class RawTx implements IRawTx {
private final AtomicBoolean m_open = new AtomicBoolean(true);
RawTx() {
activateTx();
}
@Override
public void close() {
if (m_open.compareAndSet(true/*expect*/, false/*update*/)) {
deactivateTx();
}
}
}
@Override
public IRawTx newTx() {
return new RawTx();
}
private void activateTx() {
m_allocationWriteLock.lock();
try {
m_activeTxCount++;
if(log.isInfoEnabled())
log.info("#activeTx="+m_activeTxCount);
} finally {
m_allocationWriteLock.unlock();
}
}
private void deactivateTx() {
m_allocationWriteLock.lock();
try {
if (log.isInfoEnabled())
log.info("Deactivating TX " + m_activeTxCount);
if (m_activeTxCount == 0) {
throw new IllegalStateException("Tx count must be positive!");
}
m_activeTxCount--;
if(log.isInfoEnabled())
log.info("#activeTx="+m_activeTxCount);
if (m_activeTxCount == 0 && m_contexts.isEmpty()) {
releaseSessions();
}
} finally {
m_allocationWriteLock.unlock();
}
}
/**
* Debug ONLY method added to permit unit tests to be written that the
* native transaction counter is correctly decremented to zero. The returned
* value is ONLY valid while holding the {@link #m_allocationLock}.
* Therefore this method MAY NOT be used reliably outside of code that can
* guarantee that there are no concurrent committers on the {@link RWStore}.
*
* @see Journal file growth
* reported with 1.3.3
*/
public int getActiveTxCount() {
m_allocationWriteLock.lock();
try {
return m_activeTxCount;
} finally {
m_allocationWriteLock.unlock();
}
}
/**
* Returns the slot size associated with this address
*/
@Override
public int getAssociatedSlotSize(final int addr) {
return getBlock(addr).getBlockSize();
}
/**
* lockAddress adds the address passed to a lock list. This is for
* debug only and is not intended to be used generally for the live system.
*
* @param addr - address to be locked
*/
public void lockAddress(final int addr) {
if (m_lockAddresses.putIfAbsent(addr, System.currentTimeMillis()) != null) {
throw new IllegalStateException("address already locked, logical: " + addr + ", physical: " + physicalAddress(addr, true));
}
}
public void showWriteCacheDebug(final long paddr) {
log.warn("WriteCacheDebug: " + paddr + " - " + m_writeCacheService.addrDebugInfo(paddr));
}
public CounterSet getWriteCacheCounters() {
return m_writeCacheService.getCounters();
}
// /**
// * If historical data is maintained then this will return the earliest time for which
// * data can be safely retrieved.
// *
// * @return time of last release
// */
@Override
public long getLastReleaseTime() {
return m_lastDeferredReleaseTime;
}
private ConcurrentWeakValueCache m_externalCache = null;
private int m_cachedDatasize = 0;
@Override
public void registerExternalCache(
final ConcurrentWeakValueCache externalCache,
final int dataSize) {
m_allocationWriteLock.lock();
try {
m_externalCache = externalCache;
m_cachedDatasize = getSlotSize(dataSize);
} finally {
m_allocationWriteLock.unlock();
}
}
/**
* Return true iff the allocation having that address is
* flagged as committed. The caller must be holding the allocation lock in
* order for the result to remain valid outside of the method call.
*
* @param addr
* The address.
*
* @return true iff the address is currently committed.
*/
public boolean isCommitted(final int rwaddr) {
// FIXME ALLOCATION LOCK : Why not use the ReadLock here?
final Lock lock = m_allocationWriteLock;
lock.lock();
try {
final FixedAllocator alloc = getBlockByAddress(rwaddr);
final int offset = getOffset(rwaddr);
return alloc.isCommitted(offset);
} finally {
lock.unlock();
}
}
public boolean inWriteCache(final int rwaddr) {
return m_writeCacheService.isPresent(physicalAddress(rwaddr, true));
}
@Override
public InputStream getInputStream(long addr) {
return new PSInputStream(this, addr);
}
@Override
public IPSOutputStream getOutputStream() {
return getOutputStream(null);
}
public IPSOutputStream getOutputStream(final IAllocationContext context) {
checkContext(context);
return PSOutputStream.getNew(this, m_maxFixedAlloc, context);
}
/**
* Low level routine used when we replace the root blocks of an empty
* journal in HA with those from the leader.
*
* Note: This method is only invoked in contexts where there should not be
* concurrent access to the journal. This we should not need to worry about
* concurrent readers during {@link #resetFromHARootBlock(IRootBlockView)}.
*
* @see #postHACommit(IRootBlockView)
*/
public void resetFromHARootBlock(final IRootBlockView rootBlock) {
/*
* Acquire exclusive access to the allocators.
*
* Note: Since the allocation lock must be held before you may take the
* extensionLock, and we have exclusive access to the allocation lock,
* we SHOULD NOT need to take the extension lock as well.
*/
final Lock outerLock = m_allocationWriteLock;
outerLock.lock();
try {
// Exclude IOs.
final Lock innerLock = m_extensionLock.writeLock();
innerLock.lock();
try {
// should not be any dirty allocators
// assert m_commitList.size() == 0;
// Remove all current allocators
m_allocs.clear();
assert m_nextAllocation != 0;
m_nextAllocation = 0;
initfromRootBlock(rootBlock);
// KICK external cache into touch - FIXME: handle with improved Allocator synchronization
m_externalCache.clear();
assert m_nextAllocation != 0;
} finally {
innerLock.unlock();
}
} catch (IOException e) {
throw new RuntimeException(e);
} finally {
outerLock.unlock();
}
}
/**
* Called from {@link AbstractJournal} commit2Phase to ensure that a
* downstream HA quorum member ensures it is able to read committed data
* that has been streamed directly to the backing store.
*
* The data stream will have included metabits and modified
* {@link FixedAllocator}s so these must be reset using the metabitsAddr
* data in the root block.
*
* Note: Reads on the {@link RWStore} MUST block during this method since
* some allocators may be replaced as part of the post-commit protocol.
*
* Ticket #778 was for a problem when a follower takes over as leader and
* was not correctly synchronised. This was traced, eventually, to a problem
* in calculating the diskAddr metabit for the modified Allocator. The problem
* was demonstrated by a temporary method to reserve metaAllocations by extending and
* setting the m_transient bits. But that has to be done within the commit() method
* before it attempts to save all the dirty allocators. If we need to contrive a similar
* scenario in the future a better approach would be a special debug property on the
* RWStore that indicates a "TRANSIENT_RESERVE" or something similar.
*
* @param rbv
* The new {@link IRootBlockView}.
*/
@SuppressWarnings("unchecked")
public void postHACommit(final IRootBlockView rbv) {
/*
* Acquire exclusive access to the allocators.
*
* Note: Since the allocation lock must be held before you may take the
* extensionLock, and we have exclusive access to the allocation lock,
* we SHOULD NOT need to take the extension lock as well.
*/
final Lock outerLock = m_allocationWriteLock;
outerLock.lock();
try {
final Lock innerLock = m_extensionLock.writeLock();
innerLock.lock();
try {
// Current FixedAllocators for sanity
if (log.isTraceEnabled())
{
log.trace("POSTHACOMMIT START");
for (int index = 0; index < m_allocs.size(); index++) {
final FixedAllocator xfa = m_allocs.get(index);
log.trace("Allocator " + index + ", size: " + xfa.m_size + ", startAddress: " + xfa.getStartAddr() + ", allocated: " + (xfa.getAllocatedSlots()/xfa.m_size));
}
}
// Update m_metaBits addr and m_nextAllocation to ensure able to allocate as well as read!
{
final long nxtOffset = rbv.getNextOffset();
// next allocation to be made (in -32K units).
m_nextAllocation = -(int) (nxtOffset >> 32);
if (m_nextAllocation == 0) {
throw new IllegalStateException("Invalid state for non-empty store");
}
m_committedNextAllocation = m_nextAllocation;
// latched offset of the metabits region.
m_metaBitsAddr = -(int) nxtOffset;
}
final ArrayList nallocs = new ArrayList();
// current metabits
final int[] oldmetabits = m_metaBits;
// new metabits
final RootBlockInfo rbi = new RootBlockInfo(rbv, m_reopener);
m_metaBits = rbi.m_metabits;
// and grab the last deferred release and storageStats!
m_lastDeferredReleaseTime = rbi.m_lastDeferredReleaseTime;
m_storageStatsAddr = rbi.m_storageStatsAddr;
if(log.isTraceEnabled())
log.trace("Metabits length: " + m_metaBits.length);
// Valid metabits should be multiples of default sizes
if (oldmetabits.length % cDefaultMetaBitsSize != 0)
throw new AssertionError();
if (m_metaBits.length % cDefaultMetaBitsSize != 0)
throw new AssertionError("New metabits: " + m_metaBits.length + ", old: " + oldmetabits.length);
// Is it always valid to assume that:
// metabits.length >= oldmetabits.length
if (m_metaBits.length < oldmetabits.length)
throw new AssertionError();
// need to compute modded metabits, those newly written slots by ANDing
// new bits with compliment of current
final int[] moddedBits = m_metaBits.clone();
for (int b = 0; b < oldmetabits.length; b+=cDefaultMetaBitsSize) {
// int[0] is startAddr, int[1:cDefaultMetaBitsSize] bits
for (int i = 1; i < cDefaultMetaBitsSize; i++) {
moddedBits[b+i] &= ~oldmetabits[b+i];
}
}
if (log.isTraceEnabled()) {
StringBuffer sb = new StringBuffer();
Iterator>> entries = m_externalCache.entryIterator();
while (entries.hasNext()) {
sb.append(entries.next().getKey() + "|");
}
log.trace("External Cache Start Size: " + m_externalCache.size() + ", entries: " + sb.toString());
}
// Stage One: Count moddedBits
// Stage Two: Compute Address of modded bits
// Stage Three: Read Allocator from modded address
// Stage Four: Update Live Allocators
int modCount = 0;
int totalFreed = 0;
for (int i = 0; i < moddedBits.length; i+=cDefaultMetaBitsSize) {
final long startAddr = convertAddr(m_metaBits[i]);
for (int j = 1; j < cDefaultMetaBitsSize; j++) {
final int chkbits = moddedBits[i+j];
for (int b = 0; b < 32; b++) {
if ((chkbits & (1 << b)) != 0) {
modCount++;
// Calculate address
final int bit = b + (32 * (j-1));
final long paddr = startAddr + (bit * ALLOC_BLOCK_SIZE);
if (log.isTraceEnabled())
log.trace("Allocator at: " + paddr);
// metaBit
// final int metaBit = (i * cDefaultMetaBitsSize * 32) + (j * 32) + b;
final int metaBit = ((i + j) * 32) + b;
// Now try to read it in
final FixedAllocator nalloc = readAllocator(paddr);
if (log.isTraceEnabled())
log.trace("Allocator read of size: " + nalloc.m_size + ", metaBit: " + metaBit);
nalloc.setDiskAddr(metaBit);
// Now can we find an existing one to replace, otherwise we need to add to the new list
boolean found = false;
if (log.isTraceEnabled())
log.trace("Checking allocator at " + nalloc.getStartAddr());
for (int index = 0; !found && index < m_allocs.size(); index++) {
final FixedAllocator xfa = m_allocs.get(index);
if (xfa.getStartAddr() == nalloc.getStartAddr()) {
if (log.isTraceEnabled())
log.trace("Found updated allocator at " + index
+ ", size: " + xfa.m_size + " vs " + nalloc.m_size + ", allocated slots: " + (xfa.getAllocatedSlots()/xfa.m_size) + " vs " + (nalloc.getAllocatedSlots()/xfa.m_size));
// Compare allocators to see if same
found = true;
// Replace old with new
m_allocs.set(index, nalloc);
nalloc.setIndex(index);
// remove old from free list (if set)
xfa.removeFromFreeList();
// now clear any cached writes now freed
totalFreed +=nalloc.removeFreedWrites(xfa, m_externalCache);
}
}
if (!found) {
nallocs.add(nalloc);
}
}
}
}
}
if (log.isInfoEnabled())
log.info("Released: " + totalFreed + " addresses from " + modCount + " modified Allocators");
if (log.isTraceEnabled()) {
log.trace("OLD BITS: " + BytesUtil.toHexString(oldmetabits));
log.trace("NEW BITS: " + BytesUtil.toHexString(m_metaBits));
log.trace("MODDED BITS: " + BytesUtil.toHexString(moddedBits));
log.trace("MODDED COUNT: " + modCount + " from " + m_allocs.size() + " Allocators");
}
// Now add in any new allocators, first sorting and setting their index number
if (nallocs.size() > 0) {
Collections.sort(nallocs);
final int sindex = m_allocs.size();
for (int index = 0; index < nallocs.size(); index++) {
((Allocator) nallocs.get(index)).setIndex(sindex + index);
if (log.isTraceEnabled())
log.trace("New Allocator, index: " + (sindex + index));
}
if (log.isTraceEnabled())
log.trace("Adding new allocators: " + sindex);
m_allocs.addAll(nallocs);
}
{
final long nxtOffset = rbv.getNextOffset();
// next allocation to be made (in -32K units).
m_nextAllocation = -(int) (nxtOffset >> 32);
if (m_nextAllocation == 0) {
/*
* Skip the first 32K in the file. The root blocks live here but
* nothing else.
*/
m_nextAllocation = -(1 + META_ALLOCATION);
}
m_committedNextAllocation = m_nextAllocation;
}
if (log.isTraceEnabled()) {
log.trace("POSTHACOMMIT END");
for (int index = 0; index < m_allocs.size(); index++) {
final FixedAllocator xfa = m_allocs.get(index);
log.trace("Allocator " + index + ", startAddress: " + xfa.getStartAddr() + ", allocated: " + xfa.getAllocatedSlots());
}
}
if (log.isTraceEnabled())
log.trace("External Cache Pre Clear Size: " + m_externalCache.size());
// If FixedAllocator.removeFreedWrites does its job then we do not
// need to clear the external cache
// m_externalCache.clear();
assert m_nextAllocation != 0;
} finally {
innerLock.unlock();
}
} catch (IOException e) {
throw new RuntimeException(e);
} finally {
outerLock.unlock();
}
// FIXME: Remove once allocators are synced
// log.error("Complete implementation of postHACommit()");
//
// resetFromHARootBlock(rbv);
//
// log.warn("POSTHACOMMIT AFTER RESET");
// for (int index = 0; index < m_allocs.size(); index++) {
// final FixedAllocator xfa = m_allocs.get(index);
// log.warn("Allocator " + index + ", startAddress: " + xfa.getStartAddr() + ", allocated: " + xfa.getAllocatedSlots());
// }
}
/**
* Simple class to collect up DeleteBlockStats and returned by
* checkDeleteBlocks, called from DumpJournal.
*/
public static class DeleteBlockStats {
private int m_commitRecords = 0;;
private int m_addresses = 0;
private int m_blobs = 0;
private int m_badAddresses = 0;
private final HashMap m_freed = new HashMap();
/**
* The latched address of each address that appears more than once
* across the delete blocks.
*/
private final Set m_duplicates = new LinkedHashSet();
// /**
// * The hexstring version of the data associated with the addresses that
// * are present more than once in the delete blocks.
// */
// private final ArrayList m_dupData = new ArrayList();
/**
* The #of commit records that would be processed.
*/
public int getCommitRecords() {
return m_commitRecords;
}
/**
* Return the #of addresses in the delete blocks acrosss the commit
* records.
*/
public int getAddresses() {
return m_addresses;
}
/**
* Return the #of addresses that are not committed data across the
* commit records.
*/
public int getBadAddresses() {
return m_badAddresses;
}
/**
* Return the latched addresses that appear more than once in the delete
* blocks across the commit records.
*/
public Set getDuplicateAddresses() {
return m_duplicates;
}
public String toString(final RWStore store) {
final StringBuilder sb = new StringBuilder();
sb.append("CommitRecords: " + m_commitRecords + ", Addresses: " + m_addresses
+ ", Blobs: " + m_blobs + ", bad: " + + m_badAddresses);
if (!m_duplicates.isEmpty()) {
for (int latchedAddr : m_duplicates) {
// final int latchedAddr = m_duplicates.get(i);
sb.append("\nDuplicate: latchedAddr=" + latchedAddr + "\n");
/*
* Note: Now dumped by DumpJournal.
*/
// final byte[] data;
// try {
// data = store.readFromLatchedAddress(latchedAddr);
// } catch (IOException ex) {
// final String msg = "Could not read data: addr="
// + latchedAddr;
// log.error(msg, ex);
// sb.append(msg);
// continue;
// }
//
// final String hexStr = BytesUtil.toHexString(data,
// data.length);
//
// BytesUtil.printHexString(sb, hexStr);
}
}
return sb.toString();
}
}
/**
* Utility to check the deleteBlocks associated with each active CommitRecord
*/
public DeleteBlockStats checkDeleteBlocks(final AbstractJournal journal) {
final DeleteBlockStats stats = new DeleteBlockStats();
/*
* Commit can be called prior to Journal initialisation, in which case
* the commitRecordIndex will not be set.
*/
final IIndex commitRecordIndex = journal.getReadOnlyCommitRecordIndex();
if (commitRecordIndex == null) {
return stats;
}
@SuppressWarnings("unchecked")
final ITupleIterator commitRecords = commitRecordIndex
.rangeIterator();
while (commitRecords.hasNext()) {
final ITuple tuple = commitRecords.next();
final CommitRecordIndex.Entry entry = tuple.getObject();
try {
final ICommitRecord record = CommitRecordSerializer.INSTANCE
.deserialize(journal.read(entry.addr));
final long blockAddr = record
.getRootAddr(AbstractJournal.DELETEBLOCK);
if (blockAddr != 0) {
checkDeferrals(blockAddr, record.getTimestamp(), stats);
}
stats.m_commitRecords++;
} catch (RuntimeException re) {
throw new RuntimeException("Problem with entry at "
+ entry.addr, re);
}
}
return stats;
}
/**
* Utility method to verify the deferred delete blocks.
*
* @param blockAddr
* The address of a deferred delete block.
* @param commitTime
* The commitTime associated with the {@link ICommitRecord}.
* @param stats
* Where to collect statistics.
*/
private void checkDeferrals(final long blockAddr,
final long commitTime, final DeleteBlockStats stats) {
/**
* Debug flag. When true, writes all frees onto stderr so they can be
* read into a worksheet for analysis.
*/
final boolean writeAll = false;
final int addr = (int) (blockAddr >> 32);
final int sze = (int) blockAddr & 0xFFFFFFFF; // Resolution for BLZG-1236 (recycler error)
if (log.isTraceEnabled())
log.trace("freeDeferrals at " + physicalAddress(addr) + ", size: "
+ sze + " releaseTime: " + commitTime);
final byte[] buf = new byte[sze + 4]; // allow for checksum
getData(addr, buf);
final DataInputStream strBuf = new DataInputStream(
new ByteArrayInputStream(buf));
m_allocationWriteLock.lock();
// int totalFreed = 0;
try {
int nxtAddr = strBuf.readInt();
// int cnt = 0;
while (nxtAddr != 0) { // while (false && addrs-- > 0) {
stats.m_addresses++;
if (nxtAddr > 0) { // Blob
stats.m_blobs++;
final int bloblen = strBuf.readInt();
assert bloblen > 0; // a Blob address MUST have a size
nxtAddr = -nxtAddr;
}
if (!isCommitted(nxtAddr)) {
stats.m_badAddresses++;
}
if (stats.m_freed.containsKey(nxtAddr)) {
stats.m_duplicates.add(nxtAddr);
if (writeAll) {
log.warn("" + commitTime + " " + nxtAddr
+ " FREE DUP");
}
} else {
stats.m_freed.put(nxtAddr, nxtAddr);
if (writeAll) {
log.warn("" + commitTime + " " + nxtAddr
+ " FREE");
}
}
nxtAddr = strBuf.readInt();
}
// now check delete block
assert isCommitted(addr);
} catch (IOException e) {
throw new RuntimeException("Problem checking deferrals: " + e, e);
} finally {
m_allocationWriteLock.unlock();
}
}
/**
* A low level utility method that reads directly from the backing
* {@link FileChannel}.
*
* Note: The latched address does not encode the actual length of the data.
* Therefore, all data in the slot addressed by the latched address will be
* returned.
*
* @param nxtAddr
* The latched address.
*
* @return The byte[] in the addressed slot.
*
* @throws IOException
*/
public final byte[] readFromLatchedAddress(final int nxtAddr)
throws IOException {
final Lock outerLock = m_allocationReadLock;
try {
final FixedAllocator alloc = getBlockByAddress(nxtAddr);
final byte[] data = new byte[alloc.m_size];
final ByteBuffer bb = ByteBuffer.wrap(data);
final int offset = getOffset(nxtAddr);
final long paddr = alloc.getPhysicalAddress(offset);
// Guard IO against concurrent file extension.
final Lock innerLock = m_extensionLock.readLock();
try {
FileChannelUtility.readAll(m_reopener, bb, paddr);
} finally {
innerLock.unlock();
}
return data;
} finally {
outerLock.unlock();
}
}
/**
* @see IHABufferStrategy#getBlockSequence()
*/
public long getBlockSequence() {
return lastBlockSequence;
}
private long lastBlockSequence = 0;
/**
* @see IHABufferStrategy#getCurrentBlockSequence()
*/
public long getCurrentBlockSequence() {
final WriteCacheService tmp = m_writeCacheService;
if (tmp == null) {
/*
* Either this is not an HA strategy mode -or- we are in abort() and
* the value temporarily [null]. If there is an abort(), then the
* counter will be reset to 0L.
*/
return 0L;
}
return tmp.getSequence();
}
// private HARebuildRequest m_rebuildRequest = null;
// /**
// * Only blacklist the addr if not already available, in other words
// * a blacklisted address only makes sense if it for previously
// * committed data and not instantly recyclable.
// */
// public void blacklistAddress(int addr, final String info) {
// if (m_blacklist == null) {
// // add delay/synchronization to emulate blacklist delay?
// return;
// }
//
// if (physicalAddress(addr) == 0)
// throw new IllegalStateException("Blacklist should only be called for a valid address");
//
// if (info == null) {
// throw new IllegalStateException("Blacklist must have info String");
// }
//
// if ( m_blacklist.putIfAbsent(addr, info) != null)
// throw new IllegalStateException("Address already blacklisted: "
// + addr + ", info: " + info + ", prev: " + m_blacklist.get(addr));
//
// ;
// }
/**
* @see IHABufferStrategy#computeDigest(Object, MessageDigest)
*/
public void computeDigest(final Object snapshot, final MessageDigest digest)
throws DigestException, IOException {
if(true) {
computeDigestOld(snapshot, digest);
} else {
computeDigestAlt(snapshot, digest);
}
}
private void computeDigestOld(final Object snapshot, final MessageDigest digest)
throws DigestException, IOException {
if (snapshot != null)
throw new UnsupportedOperationException();
IBufferAccess buf = null;
try {
try {
// Acquire a buffer.
buf = DirectBufferPool.INSTANCE.acquire();
} catch (InterruptedException ex) {
// Wrap and re-throw.
throw new IOException(ex);
}
// The backing ByteBuffer.
final ByteBuffer b = buf.buffer();
// // A byte[] with the same capacity as that ByteBuffer.
// final byte[] a = new byte[b.capacity()];
// The capacity of that buffer (typically 1MB).
final int bufferCapacity = b.capacity();
// The size of the file at the moment we begin.
final long fileExtent = getStoreFile().length();
// The #of bytes whose digest will be computed.
final long totalBytes = fileExtent;
// The #of bytes remaining.
long remaining = totalBytes;
// The offset of the current block.
long offset = 0L;
// The block sequence.
long sequence = 0L;
if (log.isInfoEnabled())
log.info("Computing digest: nbytes=" + totalBytes);
while (remaining > 0) {
final int nbytes = (int) Math.min((long) bufferCapacity,
remaining);
if (log.isTraceEnabled())
log.trace("Computing digest: sequence=" + sequence
+ ", offset=" + offset + ", nbytes=" + nbytes);
// Setup for read.
b.position(0);
b.limit(nbytes);
// read block
readRaw(/*nbytes,*/ offset, b);
// // Copy data into our byte[].
// final byte[] c = BytesUtil.toArray(b, false/* forceCopy */, a);
// update digest
//digest.update(c, 0/* off */, nbytes/* len */);
digest.update(b);
remaining -= nbytes;
offset += nbytes;
sequence++;
}
if (log.isInfoEnabled())
log.info("Computed digest: #blocks=" + sequence + ", #bytes="
+ totalBytes);
// Done.
return;
} finally {
if (buf != null) {
try {
// Release the direct buffer.
buf.release();
} catch (InterruptedException e) {
log.warn(e);
}
}
}
}
/**
* This alternative implementation checks only the live allocations
*
* @param snapshot
* @param digest
* @throws DigestException
* @throws IOException
*/
private void computeDigestAlt(final Object snapshot, final MessageDigest digest)
throws DigestException, IOException {
if (snapshot != null)
throw new UnsupportedOperationException();
m_allocationWriteLock.lock();
try {
// FIXME add digest for RootBlocks!
for (FixedAllocator fa : m_allocs) {
fa.computeDigest(snapshot, digest);
}
} finally {
m_allocationWriteLock.unlock();
}
{
final byte[] data = digest.digest();
final StringBuffer sb = new StringBuffer();
for (byte b : data) {
if (sb.length() > 0)
sb.append(",");
sb.append(b);
}
log.warn("STORE DIGEST: " + sb.toString());
log.warn("Free Deferrals: " + this.m_deferredFreeOut.getBytesWritten());
}
}
/**
* Used as part of the rebuild protocol
* @throws IOException
*/
public void writeRaw(final long offset, final ByteBuffer transfer) throws IOException {
if (log.isDebugEnabled())
log.debug("writeRaw: " + offset);
// Guard IO against concurrent file extension.
final Lock lock = m_extensionLock.readLock();
lock.lock();
try {
FileChannelUtility.writeAll(m_reopener, transfer, offset);
} finally {
lock.unlock();
}
}
private String showAllocatorList() {
final StringBuilder sb = new StringBuilder();
for (int index = 0; index < m_allocs.size(); index++) {
final FixedAllocator xfa = m_allocs.get(index);
sb.append("Allocator " + index + ", size: " + xfa.m_size + ", startAddress: " + xfa.getStartAddr() + ", allocated: " + xfa.getAllocatedSlots() + "\n");
}
return sb.toString();
}
// /**
// *
// * @return whether WCS is flushed
// *
// * @see IBufferStrategy#isFlushed()
// */
// public boolean isFlushed() {
// return this.m_writeCacheService.isFlushed();
// }
public static class RWStoreState implements StoreState {
/**
* Generated ID
*/
private static final long serialVersionUID = 4315400143557397323L;
/*
* Transient state necessary for consistent ha leader transition
*/
private final int m_fileSize;
private final int m_nextAllocation;
private final int m_committedNextAllocation;
private final long m_minReleaseAge;
private final long m_lastDeferredReleaseTime;
private final long m_storageStatsAddr;
private final int m_allocsSize;
private final int m_metaBitsAddr;
private final int m_metaBitsSize;
private RWStoreState(final RWStore store) {
m_fileSize = store.m_fileSize;
m_nextAllocation = store.m_nextAllocation;
m_committedNextAllocation = store.m_committedNextAllocation;
m_minReleaseAge = store.m_minReleaseAge;
m_lastDeferredReleaseTime = store.m_lastDeferredReleaseTime;
m_storageStatsAddr = store.m_storageStatsAddr;
m_allocsSize = store.m_allocs.size();
m_metaBitsAddr = store.m_metaBitsAddr;
m_metaBitsSize = store.m_metaBits.length;
}
@Override
public boolean equals(final Object obj) {
if (obj == null || !(obj instanceof RWStoreState))
return false;
final RWStoreState other = (RWStoreState) obj;
return m_fileSize == other.m_fileSize
&& m_nextAllocation == other.m_nextAllocation
&& m_committedNextAllocation == other.m_committedNextAllocation
&& m_minReleaseAge == other.m_minReleaseAge
&& m_lastDeferredReleaseTime == other.m_lastDeferredReleaseTime
&& m_storageStatsAddr == other.m_storageStatsAddr
&& m_allocsSize == other.m_allocsSize
&& m_metaBitsAddr == other.m_metaBitsAddr
&& m_metaBitsSize == other.m_metaBitsSize;
}
@Override
public String toString() {
final StringBuilder sb = new StringBuilder();
sb.append("RWStoreState\n");
sb.append("fileSize: " + m_fileSize + "\n");
sb.append("nextAllocation: " + m_nextAllocation + "\n");
sb.append("committedNextAllocation: " + m_committedNextAllocation + "\n");
sb.append("minReleaseAge: " + m_minReleaseAge + "\n");
sb.append("lastDeferredReleaseTime: " + m_lastDeferredReleaseTime + "\n");
sb.append("storageStatsAddr: " + m_storageStatsAddr + "\n");
sb.append("allocsSize: " + m_allocsSize + "\n");
sb.append("metaBitsAddr: " + m_metaBitsAddr + "\n");
sb.append("metaBitsSize: " + m_metaBitsSize + "\n");
return sb.toString();
}
}
/**
* Can be used to determine if an address is within an allocated slot.
*
* @param addr
* @return whether addr is within slot allocated area
*/
public boolean verifyAllocatedAddress(final long addr) {
for (int index = 0; index < m_allocs.size(); index++) {
final FixedAllocator xfa = m_allocs.get(index);
if (xfa.verifyAllocatedAddress(addr))
return true;
}
return false;
}
public StoreState getStoreState() {
final RWStoreState ret = new RWStoreState(this);
return ret;
}
/**
* Forces a reset of the metabits allocation on the next commit.
*
* Note that a side-effect of this is that there will be a memory leak
* of either a FixedAllocation slot or an existing demi-space.
*
* @param useDemispace
* @return whether the storage has been modified.
*/
public boolean ensureMetabitsDemispace(final boolean useDemispace) {
final boolean isDemispace = m_metaBitsAddr > 0;
if (isDemispace != useDemispace || m_useMetabitsDemispace != useDemispace) {
m_useMetabitsDemispace = useDemispace;
m_metaBitsAddr = 0;
m_recentAlloc = true; // force commit
return true;
} else {
return false;
}
}
public boolean isUsingDemiSpace() {
return m_metaBitsAddr > 0;
}
/**
* Add the address/byte[] to the snapshot representing the metabits allocaiton data
*
* @throws IOException
*/
public void snapshotMetabits(final ISnapshotData tm) throws IOException {
final long mba;
if (m_metaBitsAddr < 0) {
mba = physicalAddress((int) m_metaBitsAddr);
} else {
// long ret = physicalAddress((int) m_metaBitsAddr);
mba = convertAddr(-m_metaBitsAddr); // maximum 48 bit address range
}
tm.put(mba, genMetabitsData());
}
/**
* Add the address/allocator associated with each FixedAllocator to the snapshot map
*/
public void snapshotAllocators(final ISnapshotData tm) {
for(FixedAllocator alloc : m_allocs) {
alloc.snapshot(tm);
}
}
class AllocationContext implements IAllocationContext {
boolean m_active = true;
final boolean m_isolated;
public AllocationContext(boolean isolated) {
m_isolated = isolated;
}
final public void checkActive() {
if (!m_active) {
throw new IllegalStateException();
}
}
final public void release() {
checkActive();
m_active = false;
}
@Override
public boolean isIsolated() {
return m_isolated;
}
}
public IAllocationContext newAllocationContext(final boolean isolated) {
m_allocationWriteLock.lock();
try {
final IAllocationContext ret = new AllocationContext(isolated);
if (isolated) {
final ContextAllocation ca = new ContextAllocation(this,
m_freeFixed.length, null, ret);
if (m_contexts.put(ret, ca) != null) {
throw new AssertionError();
}
}
return ret;
} finally {
m_allocationWriteLock.unlock();
}
}
// public void prepareForRebuild(final HARebuildRequest req) {
// assert m_rebuildRequest == null;
//
// m_rebuildRequest = req;
// }
//
// public void completeRebuild(final HARebuildRequest req, final IRootBlockView rbv) {
// assert m_rebuildRequest != null;
//
// assert m_rebuildRequest.equals(req);
//
// // TODO: reinit from file
// this.resetFromHARootBlock(rbv);
//
// m_rebuildRequest = null;
// }
}