com.bigdata.htree

Class NodeSerializer

java.lang.Object

com.bigdata.htree.NodeSerializer

public class NodeSerializer
extends Object

An instance of this class is used to serialize and de-serialize the INodeDatas and ILeafDatas of an AbstractBTree. Leaf and non-leaf records have different serialization formats, but their leading bytes use the same format so that you can tell by inspection whether a buffer contains a leaf or a non-leaf node. The header of the record uses a fixed length format so that some fields can be tested without full de-serialization, especially whether the record contains a leaf vs a node. This fixed record also makes it possible to update some fields in the header once the entire record has been serialized, including the checksum, the #of bytes in the serialized record, and the prior/next addresses for leaves.

The methods defined by this class all work with ByteBuffers. On read, the buffer must be positioned to the start of the data to be read. After a read, the buffer will be positioned to the first byte after the data read. If there is insufficient data available in the buffer then an BufferUnderflowException will be thrown. On write, the data will be written on an internal buffer whose size is automatically extended. The write buffer is reused for each write and quickly achieves a maximum size for any given HTree.

Note: while the NodeSerializer is NOT thread-safe for writers, it is thread-safe for readers. This design mirrors the concurrency capabilities of the AbstractBTree.

Version:

$Id: NodeSerializer.java 3305 2010-07-27 15:08:28Z thompsonbry $

Author:

Bryan Thompson

See Also:

AbstractBTree, IndexMetadata, IAbstractNodeData, IAbstractNodeDataCoder

Field Summary

Fields

Modifier and Type	Field and Description
static int	DEFAULT_BUFFER_CAPACITY_PER_ENTRY The default initial capacity multiplier for the (de-)serialization buffer.
protected INodeFactory	nodeFactory An object that knows how to construct DirectoryPages and BucketPages from IDirectoryData and ILeafData objects.

Constructor Summary

Constructors

Constructor and Description
NodeSerializer(IAddressManager addressManager, INodeFactory nodeFactory, int addressBits, int initialBufferCapacity, IndexMetadata indexMetadata, boolean readOnly, IRecordCompressorFactory<?> recordCompressorFactory) Designated constructor.

Method Summary

Methods

Modifier and Type	Method and Description
void	close() Releases any buffers.
IAbstractNodeData	decode(ByteBuffer buf) Decode an INodeData or ILeafData record, wrapping the underlying data record (thread-safe).
AbstractFixedByteArrayBuffer	encode(IAbstractNodeData node) Deprecated. This method is no longer used since I refactored the IndexSegmentBuilder to optionally stuff the generated nodes and leaves into the cache. It still works but it might go away in the future.
<T extends IAbstractNodeData> T	encodeLive(T node) Encode a node or leaf and return a coded instance of the persistent data for that node or leaf backed by an exact fit byte[] (NOT thread-safe).
void	updateLeaf(ByteBuffer b, long priorAddr, long nextAddr) Update the serialization of a leaf to set the prior and next leaf references and change its serialization type from #TYPE_LEAF to #TYPE_LINKED_LEAF.
com.bigdata.htree.AbstractPage	wrap(AbstractHTree btree, long addr, IAbstractNodeData data) Wrap an INodeData or ILeafData instance as a DirectoryPage or a BucketPage.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

nodeFactory

protected final INodeFactory nodeFactory

An object that knows how to construct DirectoryPages and BucketPages from IDirectoryData and ILeafData objects.

DEFAULT_BUFFER_CAPACITY_PER_ENTRY

public static final transient int DEFAULT_BUFFER_CAPACITY_PER_ENTRY

The default initial capacity multiplier for the (de-)serialization buffer. The total initial buffer capacity is this value times the effective branching factor, which is computed from the addressBits constructor parameter.

See Also:

Constant Field Values

Constructor Detail

NodeSerializer

public NodeSerializer(IAddressManager addressManager,
              INodeFactory nodeFactory,
              int addressBits,
              int initialBufferCapacity,
              IndexMetadata indexMetadata,
              boolean readOnly,
              IRecordCompressorFactory<?> recordCompressorFactory)

Designated constructor.

Parameters:

nodeFactory - An object that knows how to construct INodeDatas and leaves.

addressBits - The #of address bits for target HTree.

initialBufferCapacity - The initial capacity for internal buffer used to serialize nodes and leaves. The buffer will be resized as necessary until it is sufficient for the records being serialized for the HTree. When zero (0), a default is used. A non-zero value is worth specifying only when the actual buffer size is consistently less than the default for some HTree. See DEFAULT_BUFFER_CAPACITY_PER_ENTRY

indexMetadata - The IndexMetadata record for the index.

readOnly - true IFF the caller is asserting that they WILL NOT attempt to serialize any nodes or leaves using this NodeSerializer instance. FIXME IRecordCompressorFactory should either be used here or moved into the IRawStore impl.

Method Detail

close

public void close()

Releases any buffers. They will be automatically reallocated if the NodeSerializer is used again.

decode

public IAbstractNodeData decode(ByteBuffer buf)

Decode an INodeData or ILeafData record, wrapping the underlying data record (thread-safe). The decision to decode as an INodeData or ILeafData instance is made based on inspection of the first byte byte in the supplied buffer, which codes for a node, leaf, or linked-leaf.

Parameters:

buf - The data record.

Returns:

A INodeData or ILeafData instance for that data record. FIXME modify to accept IDataRecord rather than ByteBuffer.

wrap

public com.bigdata.htree.AbstractPage wrap(AbstractHTree btree,
                                  long addr,
                                  IAbstractNodeData data)

Wrap an INodeData or ILeafData instance as a DirectoryPage or a BucketPage. This DOES NOT set the parent of the new DirectoryPage or BucketPage.

Parameters:

btree - The owning B+Tree.

addr - The address of the data record in the backing store.

data - The data record.

Returns:

The node or leaf.

encodeLive

public <T extends IAbstractNodeData> T encodeLive(T node)

Encode a node or leaf and return a coded instance of the persistent data for that node or leaf backed by an exact fit byte[] (NOT thread-safe). The operation writes on an internal buffer which is automatically extended as required.

Parameters:

node - The node or leaf.

Returns:

The buffer containing the coded data record for the node or leaf in an exact fit byte[] owned by the coded node or leaf.

encode

public AbstractFixedByteArrayBuffer encode(IAbstractNodeData node)

Deprecated. This method is no longer used since I refactored the IndexSegmentBuilder to optionally stuff the generated nodes and leaves into the cache. It still works but it might go away in the future.

Encode a node or leaf onto an internal buffer and return that buffer (NOT thread-safe). This is a slight optimization of encodeLive(IAbstractNodeData) which is used when the written node or leaf will not be reused, e.g., by the IndexSegmentBuilder . The operation writes on an internal buffer which is automatically extended as required.

Parameters:

node - The node or leaf.

Returns:

The buffer containing the coded data record for the node or leaf in a shared buffer. The contents of this buffer may be overwritten by the next node or leaf serialized the same instance of this class. The position will be zero and the limit will be the #of bytes in the coded representation.

updateLeaf

public void updateLeaf(ByteBuffer b,
              long priorAddr,
              long nextAddr)

Update the serialization of a leaf to set the prior and next leaf references and change its serialization type from #TYPE_LEAF to #TYPE_LINKED_LEAF.

Note: In order to use this method to write linked leaves on the store you have to either write behind at a pre-determined address on the store or settle for writing only the prior or the next leaf address, but not both. It is up to the caller to perform these tricks. All this method does is to touch up the serialized record.

Note: This method has NO side-effects on the position or limit of the caller's ByteBuffer.

Parameters:

b - The serialization leaf.

priorAddr - The address of the previous leaf in key order, 0L if it is known that there is no previous leaf, and -1L if either: (a) it is not known whether there is a previous leaf; or (b) it is known but the address of that leaf is not known to the caller.

nextAddr - The address of the next leaf in key order, 0L if it is known that there is no next leaf, and -1L if either: (a) it is not known whether there is a next leaf; or (b) it is known but the address of that leaf is not known to the caller.

See Also:

IndexSegmentBuilder, DefaultLeafCoder