com.bigdata.rdf.spo

Class XXXCShardSplitHandler

java.lang.Object

com.bigdata.rdf.spo.XXXCShardSplitHandler

All Implemented Interfaces:

ISimpleSplitHandler, Serializable

public class XXXCShardSplitHandler
extends Object
implements ISimpleSplitHandler, Serializable

A split handler for the xxxC quads indices which ensures that all quads for the same "triple" are in the same shard. The use of this split handler allows optimizations for default graph queries in scale-out. Since all quads for a given triple are known to be on the same shard, we can use the ContextAdvancer to skip to the next possible triple without imposing a "DISTINCT" filter.

Version:

$Id$

Author:

Bryan Thompson

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
XXXCShardSplitHandler()

Method Summary

Methods

Modifier and Type	Method and Description
byte[]	getSeparatorKey(IndexSegment seg, int fromIndex, int toIndex, int splitAt) Return a separator key which does not split quads for the same triple prefix.

Methods inherited from class java.lang.Object

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

Constructor Detail

XXXCShardSplitHandler

public XXXCShardSplitHandler()

Method Detail

getSeparatorKey

public byte[] getSeparatorKey(IndexSegment seg,
                     int fromIndex,
                     int toIndex,
                     int splitAt)

Return a separator key which does not split quads for the same triple prefix.

A "quad" key is formed by an IV[4] array. For an index whose last key component is "C" (context), the first three key components identify some permutation of a triple. This will return a separator key such that all "quads" for a given triple are on the same shard.

On entry, splitAt will be the index of some quads key. From that key, this method generates (a) the key corresponding to that triple from IV[3]; and (b) the key corresponding to the next triple from successor(IV[3]).

The ILinearList.indexOf(byte[]) positions of the two keys are considered within the index segment. The key is returned whose index position is closest to the given splitAt index. If neither of these two keys lies within the allowable half-open range (fromIndex:toIndex]) then an error message is logged and null is returned, indicating that the IndexSegment should not be split.

Specified by:

getSeparatorKey in interface ISimpleSplitHandler

Parameters:

seg - The IndexSegment containing the data to be split.

fromIndex - The index of the first tuple which may be considered (inclusive lower bound).

toIndex - The index of the last tuple which may be considered (exclusive upper bound).

splitAt - The index of the recommended separator key.

Returns:

The desired separator key. If no acceptable separator key could be found or constructed then return null.