src.it.unimi.dsi.big.mg4j.tool.PartitionLexically Maven / Gradle / Ivy
Show all versions of mg4j-big Show documentation
package it.unimi.dsi.big.mg4j.tool;
/*
* MG4J: Managing Gigabytes for Java (big)
*
* Copyright (C) 2006-2011 Sebastiano Vigna
*
* This library is free software; you can redistribute it and/or modify it
* under the terms of the GNU Lesser General Public License as published by the Free
* Software Foundation; either version 3 of the License, or (at your option)
* any later version.
*
* This library 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 Lesser General Public License
* for more details.
*
* You should have received a copy of the GNU Lesser General Public License
* along with this program; if not, see A global index is partitioned lexically by providing a {@link LexicalPartitioningStrategy}
* that specifies a destination local index for each term, and a local term number. The global index
* is read directly at the bit level, and the posting lists are divided among the
* local indices using the provided strategy. For instance,
* an {@link ContiguousLexicalStrategy} divides an index into
* contiguous blocks (of terms) specified by the given strategy.
*
*
By choice, document pointers are not remapped. Thus, it may happen that one of the local indices
* contains no posting with a certain document. However, computing the subset of documents contained
* in each local index to remap them in a contiguous interval is not a good idea, as usually the subset
* of documents appearing in the postings of each local index is large.
*
*
To speed up the search of the right local index of a not-so-frequent term (in
* particular with a {@linkplain it.unimi.dsi.big.mg4j.index.cluster.ChainedLexicalClusteringStrategy chained strategy}),
* after partitioning an index you can create {@linkplain BloomFilter Bloom filters} that will be used to try to avoid
* inquiring indices that do not contain a term. The filters will be automatically loaded
* by {@link it.unimi.dsi.big.mg4j.index.cluster.IndexCluster#getInstance(CharSequence, boolean, boolean)}.
*
*
Note that the size file is the same for each local index and is not copied. Please use
* standard operating system features such as symbolic links to provide size files to
* local indices.
*
*
If you plan to {@linkplain LexicalCluster cluster} the partitioned indices and you need document sizes
* (e.g., for {@linkplain BM25Scorer BM25 scoring}), you can use the index property
* {@link it.unimi.dsi.big.mg4j.index.Index.UriKeys#SIZES} to load the original size file.
*
* If you plan on partitioning an index requiring
* document sizes, you should consider a custom index loading scheme
* that shares the {@linkplain it.unimi.dsi.big.mg4j.index.BitStreamIndex#sizes size list}
* among all local indices.
*
* Important: this class just partitions the index. No auxiliary files (most notably, {@linkplain StringMap term maps}
* or {@linkplain PrefixMap prefix maps}) will be generated. Please refer to a {@link StringMap} implementation (e.g.,
* {@link ShiftAddXorSignedStringMap} or {@link ImmutableExternalPrefixMap}).
*
*
Write-once output and distributed index partitioning
*
* The partitioning process writes each index file sequentially exactly once, so index partitioning
* can output its results to pipes, which in
* turn can spill their content, for instance, through the network. In other words, albeit this
* class theoretically creates a number of local indices on disk, those indices can be
* substituted with suitable pipes creating remote local indices without affecting the partitioning process.
* For instance, the following bash code creates three sets of pipes:
*
* for i in 0 1 2; do
* for e in frequencies globcounts index offsets properties sizes terms; do
* mkfifo pipe-$i.$e
* done
* done
*
*
* Each pipe must be emptied elsewhere, for instance (assuming
* you want local indices index0, index1 and index2 on example.com):
*
* for i in 0 1 2; do
* for e in frequencies globcounts index offsets properties sizes terms; do
* (cat pipe-$i.$e | ssh -x example.com "cat >index-$i.$e" &)
* done
* done
*
* If we now start a partitioning process generating three local indices named pipe-0,
* pipe-1 and pipe-2
* all pipes will be written to by the process, and the data will create remotely
* indices index-0, index-1 and index-2.
*
* @author Sebastiano Vigna
*
* @since 1.0.1
*/
public class PartitionLexically {
private static final Logger LOGGER = Util.getLogger( PartitionLexically.class );
/** The default buffer size for all involved indices. */
public final static int DEFAULT_BUFFER_SIZE = 1024 * 1024;
/** The number of local indices. */
private final int numIndices;
/** The output basenames. */
private final String outputBasename;
/** The array of local output basenames. */
private final String[] localBasename;
/** The input basename. */
private final String inputBasename;
/** The size of I/O buffers. */
private final int bufferSize;
/** The filename of the strategy used to partition the index. */
private final String strategyFilename;
/** The strategy used to partition the index. */
private final LexicalPartitioningStrategy strategy;
/** The additional local properties of each local index. */
private final Properties[] strategyProperties;
/** The logging interval. */
private final long logInterval;
public PartitionLexically( final String inputBasename,
final String outputBasename,
final LexicalPartitioningStrategy strategy,
final String strategyFilename,
final int bufferSize,
final long logInterval ) {
this.inputBasename = inputBasename;
this.outputBasename = outputBasename;
this.strategy = strategy;
this.strategyFilename = strategyFilename;
this.bufferSize = bufferSize;
this.logInterval = logInterval;
numIndices = strategy.numberOfLocalIndices();
strategyProperties = strategy.properties();
localBasename = new String[ numIndices ];
for( int i = 0; i < numIndices; i++ ) localBasename[ i ] = outputBasename + "-" + i;
}
public void runTermsOnly() throws IOException {
final ProgressLogger pl = new ProgressLogger( LOGGER, logInterval );
final PrintWriter[] localTerms = new PrintWriter[ numIndices ];
final int numTerms[] = new int[ numIndices ];
final FastBufferedReader terms = new FastBufferedReader( new InputStreamReader( new FileInputStream( inputBasename + DiskBasedIndex.TERMS_EXTENSION ), "UTF-8" ) );
for( int i = 0; i < numIndices; i++ ) localTerms[ i ] = new PrintWriter( new OutputStreamWriter( new FastBufferedOutputStream( new FileOutputStream( localBasename[ i ] + DiskBasedIndex.TERMS_EXTENSION ) ), "UTF-8" ) );
// The current term
final MutableString currTerm = new MutableString();
pl.itemsName = "terms";
pl.logInterval = logInterval;
pl.start( "Partitioning index terms..." );
int termNumber = 0, k;
while( terms.readLine( currTerm ) != null ) {
k = strategy.localIndex( termNumber ); // The local index for this term
if ( numTerms[ k ] != strategy.localNumber( termNumber ) ) throw new IllegalStateException();
numTerms[ k ]++;
currTerm.println( localTerms[ k ] );
pl.update();
termNumber++;
}
terms.close();
for( int i = 0; i < numIndices; i++ ) localTerms[ i ].close();
pl.done();
}
public void run() throws ConfigurationException, IOException, ClassNotFoundException {
final ProgressLogger pl = new ProgressLogger( LOGGER, logInterval );
final byte[] buffer = new byte[ bufferSize ];
final OutputBitStream[] localIndexStream = new OutputBitStream[ numIndices ];
final OutputBitStream[] localPositionsStream = new OutputBitStream[ numIndices ];
final OutputBitStream[] localOffsets = new OutputBitStream[ numIndices ];
final OutputBitStream[] localPosNumBits = new OutputBitStream[ numIndices ];
final OutputBitStream[] localFrequencies = new OutputBitStream[ numIndices ];
final OutputBitStream[] localGlobCounts = new OutputBitStream[ numIndices ];
final PrintWriter[] localTerms = new PrintWriter[ numIndices ];
final int numTerms[] = new int[ numIndices ];
final long numberOfOccurrences[] = new long[ numIndices ];
final long numberOfPostings[] = new long[ numIndices ];
final boolean isHighPerformance = BitStreamHPIndex.class.isAssignableFrom( Class.forName( new Properties( inputBasename + DiskBasedIndex.PROPERTIES_EXTENSION ).getString( Index.PropertyKeys.INDEXCLASS ) ) );
final InputBitStream globalIndex = new InputBitStream( inputBasename + DiskBasedIndex.INDEX_EXTENSION, bufferSize );
final long globalPositionsLength = new File( inputBasename + DiskBasedIndex.POSITIONS_EXTENSION ).length();
final InputBitStream globalPositions = isHighPerformance ? new InputBitStream( inputBasename + DiskBasedIndex.POSITIONS_EXTENSION, bufferSize ) : null;
final FastBufferedReader terms = new FastBufferedReader( new InputStreamReader( new FileInputStream( inputBasename + DiskBasedIndex.TERMS_EXTENSION ), "UTF-8" ) );
final InputBitStream offsets = new InputBitStream( inputBasename + DiskBasedIndex.OFFSETS_EXTENSION );
final File posNumBitsFile = new File( inputBasename + DiskBasedIndex.POSITIONS_NUMBER_OF_BITS_EXTENSION );
final InputBitStream posNumBits = posNumBitsFile.exists() ? new InputBitStream( inputBasename + DiskBasedIndex.POSITIONS_NUMBER_OF_BITS_EXTENSION ) : null;
final InputBitStream frequencies = new InputBitStream( inputBasename + DiskBasedIndex.FREQUENCIES_EXTENSION );
final InputBitStream globCounts = new InputBitStream( inputBasename + DiskBasedIndex.GLOBCOUNTS_EXTENSION );
offsets.readGamma();
for( int i = 0; i < numIndices; i++ ) {
localIndexStream[ i ] = new OutputBitStream( localBasename[ i ] + DiskBasedIndex.INDEX_EXTENSION, bufferSize );
if ( isHighPerformance ) localPositionsStream[ i ] = new OutputBitStream( localBasename[ i ] + DiskBasedIndex.POSITIONS_EXTENSION, bufferSize );
localFrequencies[ i ] = new OutputBitStream( localBasename[ i ] + DiskBasedIndex.FREQUENCIES_EXTENSION );
localGlobCounts[ i ] = new OutputBitStream( localBasename[ i ] + DiskBasedIndex.GLOBCOUNTS_EXTENSION );
localTerms[ i ] = new PrintWriter( new OutputStreamWriter( new FastBufferedOutputStream( new FileOutputStream( localBasename[ i ] + DiskBasedIndex.TERMS_EXTENSION ) ), "UTF-8" ) );
localOffsets[ i ] = new OutputBitStream( localBasename[ i ] + DiskBasedIndex.OFFSETS_EXTENSION );
if ( posNumBits != null ) localPosNumBits[ i ] = new OutputBitStream( localBasename[ i ] + DiskBasedIndex.POSITIONS_NUMBER_OF_BITS_EXTENSION );
localOffsets[ i ].writeGamma( 0 );
}
// The current term
final MutableString currTerm = new MutableString();
pl.expectedUpdates = ( new File( inputBasename + DiskBasedIndex.INDEX_EXTENSION ).length() + ( isHighPerformance ? new File( inputBasename + DiskBasedIndex.POSITIONS_EXTENSION ).length() : 0 ) ) * 8;
pl.itemsName = "bits";
pl.logInterval = logInterval;
pl.start( "Partitioning index..." );
int termNumber = 0, k, prevK = -1, previousHeaderLength = 0, newHeaderLength = 0;
long length, count, positionsOffset = 0;
int res;
long frequency;
while( terms.readLine( currTerm ) != null ) {
k = strategy.localIndex( termNumber ); // The local index for this term
if ( numTerms[ k ] != strategy.localNumber( termNumber ) ) throw new IllegalStateException();
numTerms[ k ]++;
if ( isHighPerformance ) {
final long temp = globalIndex.readBits();
positionsOffset = globalIndex.readLongDelta();
previousHeaderLength = (int)( globalIndex.readBits() - temp );
if ( prevK != -1 ) {
length = positionsOffset - globalPositions.readBits();
pl.count += length;
while( length > 0 ) {
res = (int)Math.min( bufferSize * 8, length );
globalPositions.read( buffer, res );
localPositionsStream[ prevK ].write( buffer, res );
length -= res;
}
}
newHeaderLength = localIndexStream[ k ].writeLongDelta( localPositionsStream[ k ].writtenBits() );
}
frequency = frequencies.readLongGamma();
localFrequencies[ k ].writeLongGamma( frequency );
numberOfPostings[ k ] += frequency;
if ( posNumBits != null ) localPosNumBits[ k ].writeGamma( posNumBits.readGamma() );
count = globCounts.readLongGamma();
numberOfOccurrences[ k ] += count;
localGlobCounts[ k ].writeLongGamma( count );
currTerm.println( localTerms[ k ] );
length = offsets.readLongGamma() - previousHeaderLength;
localOffsets[ k ].writeLongGamma( length + newHeaderLength );
pl.count += length + previousHeaderLength - 1;
while( length > 0 ) {
res = (int)Math.min( bufferSize * 8, length );
globalIndex.read( buffer, res );
localIndexStream[ k ].write( buffer, res );
length -= res;
}
pl.update();
prevK = k;
termNumber++;
}
// We pour the last piece of positions
if ( isHighPerformance ) {
if ( prevK != -1 ) {
length = globalPositionsLength * 8 - globalPositions.readBits();
System.err.println( globalPositionsLength * 8 - globalPositions.readBits() );
while( length > 0 ) {
res = (int)Math.min( bufferSize * 8, length );
globalPositions.read( buffer, res );
localPositionsStream[ prevK ].write( buffer, res );
length -= res;
}
}
}
pl.done();
terms.close();
offsets.close();
frequencies.close();
globCounts.close();
globalIndex.close();
if ( posNumBits != null ) posNumBits.close();
if ( isHighPerformance ) globalPositions.close();
// We copy the relevant properties from the original
Properties properties = new Properties( inputBasename + DiskBasedIndex.PROPERTIES_EXTENSION );
Properties globalProperties = new Properties();
if ( strategyFilename != null ) globalProperties.setProperty( IndexCluster.PropertyKeys.STRATEGY, strategyFilename );
globalProperties.setProperty( DocumentalCluster.PropertyKeys.BLOOM, false );
globalProperties.setProperty( Index.PropertyKeys.INDEXCLASS, LexicalCluster.class.getName() );
for( int i = 0; i < numIndices; i++ ) globalProperties.addProperty( IndexCluster.PropertyKeys.LOCALINDEX, localBasename[ i ] );
globalProperties.setProperty( Index.PropertyKeys.FIELD, properties.getProperty( Index.PropertyKeys.FIELD ) );
globalProperties.setProperty( Index.PropertyKeys.POSTINGS, properties.getProperty( Index.PropertyKeys.POSTINGS ) );
globalProperties.setProperty( Index.PropertyKeys.OCCURRENCES, properties.getProperty( Index.PropertyKeys.OCCURRENCES ) );
globalProperties.setProperty( Index.PropertyKeys.DOCUMENTS, properties.getProperty( Index.PropertyKeys.DOCUMENTS ) );
globalProperties.setProperty( Index.PropertyKeys.TERMS, properties.getProperty( Index.PropertyKeys.TERMS ) );
globalProperties.setProperty( Index.PropertyKeys.TERMPROCESSOR, properties.getProperty( Index.PropertyKeys.TERMPROCESSOR ) );
globalProperties.setProperty( Index.PropertyKeys.MAXCOUNT, properties.getProperty( Index.PropertyKeys.MAXCOUNT ) );
globalProperties.setProperty( Index.PropertyKeys.MAXDOCSIZE, properties.getProperty( Index.PropertyKeys.MAXDOCSIZE ) );
globalProperties.save( outputBasename + DiskBasedIndex.PROPERTIES_EXTENSION );
LOGGER.debug( "Properties for clustered index " + outputBasename + ": " + new ConfigurationMap( globalProperties ) );
for( int i = 0; i < numIndices; i++ ) {
localIndexStream[ i ].close();
if ( isHighPerformance ) localPositionsStream[ i ].close();
localOffsets[ i ].close();
if ( posNumBits != null ) localPosNumBits[ i ].close();
localFrequencies[ i ].close();
localGlobCounts[ i ].close();
localTerms[ i ].close();
final InputStream input = new FileInputStream( inputBasename + DiskBasedIndex.SIZES_EXTENSION );
final OutputStream output = new FileOutputStream( localBasename[ i ] + DiskBasedIndex.SIZES_EXTENSION );
IOUtils.copy( input, output );
input.close();
output.close();
Properties localProperties = new Properties();
localProperties.addAll( globalProperties );
localProperties.setProperty( Index.PropertyKeys.TERMS, numTerms[ i ] );
localProperties.setProperty( Index.PropertyKeys.OCCURRENCES, numberOfOccurrences[ i ] );
localProperties.setProperty( Index.PropertyKeys.POSTINGS, numberOfPostings[ i ] );
localProperties.setProperty( Index.PropertyKeys.POSTINGS, numberOfPostings[ i ] );
localProperties.setProperty( Index.PropertyKeys.INDEXCLASS, properties.getProperty( Index.PropertyKeys.INDEXCLASS ) );
localProperties.addProperties( Index.PropertyKeys.CODING, properties.getStringArray( Index.PropertyKeys.CODING ) );
localProperties.setProperty( BitStreamIndex.PropertyKeys.SKIPQUANTUM, properties.getProperty( BitStreamIndex.PropertyKeys.SKIPQUANTUM ) );
localProperties.setProperty( BitStreamIndex.PropertyKeys.SKIPHEIGHT, properties.getProperty( BitStreamIndex.PropertyKeys.SKIPHEIGHT ) );
if ( strategyProperties[ i ] != null ) localProperties.addAll( strategyProperties[ i ] );
localProperties.save( localBasename[ i ] + DiskBasedIndex.PROPERTIES_EXTENSION );
LOGGER.debug( "Post-partitioning properties for index " + localBasename[ i ] + ": " + new ConfigurationMap( localProperties ) );
}
}
public static void main( final String[] arg ) throws JSAPException, ConfigurationException, IOException, ClassNotFoundException, SecurityException, InstantiationException, IllegalAccessException {
SimpleJSAP jsap = new SimpleJSAP( PartitionLexically.class.getName(), "Partitions an index lexically.",
new Parameter[] {
new FlaggedOption( "bufferSize", JSAP.INTSIZE_PARSER, Util.formatBinarySize( DEFAULT_BUFFER_SIZE ), JSAP.NOT_REQUIRED, 'b', "buffer-size", "The size of an I/O buffer." ),
new FlaggedOption( "logInterval", JSAP.LONG_PARSER, Long.toString( ProgressLogger.DEFAULT_LOG_INTERVAL ), JSAP.NOT_REQUIRED, 'l', "log-interval", "The minimum time interval between activity logs in milliseconds." ),
new FlaggedOption( "strategy", JSAP.STRING_PARSER, JSAP.NO_DEFAULT, JSAP.NOT_REQUIRED, 's', "strategy", "A serialised lexical partitioning strategy." ),
new FlaggedOption( "uniformStrategy", JSAP.INTEGER_PARSER, JSAP.NO_DEFAULT, JSAP.NOT_REQUIRED, 'u', "uniform", "Requires a uniform partitioning in the given number of parts." ),
new Switch( "termsOnly", 't', "terms-only", "Just partition the term list." ),
new UnflaggedOption( "inputBasename", JSAP.STRING_PARSER, JSAP.REQUIRED, "The basename of the global index." ),
new UnflaggedOption( "outputBasename", JSAP.STRING_PARSER, JSAP.REQUIRED, "The basename of the local indices." )
});
JSAPResult jsapResult = jsap.parse( arg );
if ( jsap.messagePrinted() ) return;
String inputBasename = jsapResult.getString( "inputBasename" );
String outputBasename = jsapResult.getString( "outputBasename" );
String strategyFilename = jsapResult.getString( "strategy" );
LexicalPartitioningStrategy strategy = null;
if ( jsapResult.userSpecified( "uniformStrategy" ) ) {
strategy = LexicalStrategies.uniform( jsapResult.getInt( "uniformStrategy" ), DiskBasedIndex.getInstance( inputBasename, false, false, true ) );
BinIO.storeObject( strategy, strategyFilename = outputBasename + IndexCluster.STRATEGY_DEFAULT_EXTENSION );
}
else if ( strategyFilename != null ) strategy = (LexicalPartitioningStrategy)BinIO.loadObject( strategyFilename );
else throw new IllegalArgumentException( "You must specify a splitting strategy" );
final PartitionLexically partitionLexically = new PartitionLexically( inputBasename,
outputBasename,
strategy,
strategyFilename,
jsapResult.getInt( "bufferSize" ),
jsapResult.getLong( "logInterval" ) );
if ( jsapResult.getBoolean( "termsOnly" ) ) partitionLexically.runTermsOnly();
else partitionLexically.run();
}
}