com.bigdata.btree.proc.IIndexProcedure Maven / Gradle / Ivy
Show all versions of bigdata-core Show documentation
/*
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
*/
/*
* Created on Mar 15, 2007
*/
package com.bigdata.btree.proc;
import java.io.Serializable;
import com.bigdata.btree.IIndex;
import com.bigdata.btree.IRangeQuery;
import com.bigdata.btree.ISimpleBTree;
import com.bigdata.journal.IReadOnly;
import com.bigdata.service.DataService;
import com.bigdata.service.IDataService;
import com.bigdata.service.ndx.ClientIndexView;
import com.bigdata.sparse.SparseRowStore;
/**
* An arbitrary procedure run against a single index.
*
* There is a fairly direct correspondence between map / reduce processing and
* the use of index procedures. The "map" phase corresponds to the
* {@link IIndexProcedure} while the "reduce" phase corresponds to the
* {@link IResultHandler}. The main difference is that index procedures operate
* on ordered data. {@link IIndexProcedure}s MAY use
* {@link IParallelizableIndexProcedure} to declare that they can be executed in
* parallel when mapped across multiple index partitions - parallelization is
* handled automatically by {@link ClientIndexView}. Otherwise it is assumed
* that the procedure requires sequential execution as it is applied to the
* relevant index partition(s) - again this is handled automatically.
*
* There are basically three kinds of index procedures, each of which determines
* where an index procedure will be mapped in a different manner:
*
*
* - A single key,
byte[] key
*
* - Examples are when reading a logical row from a {@link SparseRowStore} or
* performing a datum {@link ISimpleBTree#lookup(Object)}. These procedures are
* always directed to a single index partition. Since they are never mapped
* across index partitions, there is no "aggregation" phase. Likewise, there is
* no {@link AbstractKeyArrayIndexProcedureConstructor} since the procedure instance is
* always created directly by the application.
*
* - A key range,
byte[] toKey, byte[] fromKey
*
* - Examples are {@link IRangeQuery#rangeCount(byte[], byte[])} and
* {@link IRangeQuery#rangeIterator(byte[], byte[]). Since the same data is sent
* to each index partition there is no requirement for an object to construct
* instances of these procedures. However, the procedures do need a means to
* combine or aggregate the results from each index partition. See
* {@link IKeyRangeIndexProcedure}.
*
* - A set of keys,
byte[][] keys
*
* - Examples are {@link BatchInsert}, {@link BatchLookup}, etc. For this
* case the procedure is not only mapped across one or more index partitions,
* but the procedure's data must be split such that each index partition is sent
* only the data that is relevant to that index partition. See
* {@link IKeyArrayIndexProcedure}.
*
*
*
* Note: this interface extends {@link Serializable}, however that provides
* only for communicating state to the {@link IDataService}. If an instance of
* this procedure will cross a network interface, then the implementation class
* MUST be available to the {@link IDataService} on which it will execute. This
* can be as simple as bundling the procedure into a JAR that is part of the
* CLASSPATH used to start a {@link DataService} or you can use downloaded code
* with the JINI codebase mechanism (java.rmi.server.codebase).
*
* @author Bryan Thompson
*/
public interface IIndexProcedure extends IReadOnly, Serializable {
/**
* Run the procedure.
*
* Note: Unisolated procedures have "auto-commit" ACID properties for a
* local index only. In order for a distributed procedure to be ACID, the
* procedure MUST be executed within a fully isolated transaction.
*
* @param ndx
* The index.
*
* @return The result, which is entirely defined by the procedure
* implementation and which MAY be null. In general,
* this MUST be {@link Serializable} since it may have to pass
* across a network interface.
*/
public T apply(IIndex ndx);
}