htsjdk.samtools.util.SortingCollection Maven / Gradle / Ivy
/*
* The MIT License
*
* Copyright (c) 2018 The Broad Institute
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*/
package htsjdk.samtools.util;
import htsjdk.samtools.Defaults;
import java.io.File;
import java.io.IOException;
import java.io.InputStream;
import java.io.OutputStream;
import java.lang.reflect.Array;
import java.nio.file.Files;
import java.nio.file.Path;
import java.nio.file.Paths;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collection;
import java.util.Comparator;
import java.util.Iterator;
import java.util.List;
import java.util.NoSuchElementException;
import java.util.TreeSet;
/**
* Collection to which many records can be added. After all records are added, the collection can be
* iterated, and the records will be returned in order defined by the comparator. Records may be spilled
* to a temporary directory if there are more records added than will fit in memory. As a result of this,
* the objects returned may not be identical to the objects added to the collection, but they should be
* equal as determined by the codec used to write them to disk and read them back.
*
* When iterating over the collection, the number of file handles required is numRecordsInCollection/maxRecordsInRam.
* If this becomes a limiting factor, a file handle cache could be added.
*
* If Snappy DLL is available and snappy.disable system property is not set to true, then Snappy is used
* to compress temporary files.
*/
public class SortingCollection implements Iterable {
private static final Log log = Log.getInstance(SortingCollection.class);
/**
* Client must implement this class, which defines the way in which records are written to and
* read from file.
*/
public interface Codec extends Cloneable {
/**
* Where to write encoded output
*/
void setOutputStream(OutputStream os);
/**
* Where to read encoded input from
*/
void setInputStream(InputStream is);
/**
* Write object to output stream
*
* @param val what to write
*/
void encode(T val);
/**
* Read the next record from the input stream and convert into a java object.
*
* @return null if no more records. Should throw exception if EOF is encountered in the middle of
* a record.
*/
T decode();
/**
* Must return a cloned copy of the codec that can be used independently of
* the original instance. This is required so that multiple codecs can exist simultaneously
* that each is reading a separate file.
*/
Codec clone();
}
/**
* Directories where files of sorted records go.
*/
private final Path[] tmpDirs;
/**
* Used to write records to file, and used as a prototype to create codecs for reading.
*/
private final SortingCollection.Codec codec;
/**
* For sorting, both when spilling records to file, and merge sorting.
*/
private final Comparator comparator;
private final int maxRecordsInRam;
private int numRecordsInRam = 0;
private T[] ramRecords;
private boolean iterationStarted = false;
private boolean doneAdding = false;
/**
* Set to true when all temp files have been cleaned up
*/
private boolean cleanedUp = false;
/**
* List of files in tmpDir containing sorted records
*/
private final List files = new ArrayList<>();
private boolean destructiveIteration = true;
private final TempStreamFactory tempStreamFactory = new TempStreamFactory();
private final boolean printRecordSizeSampling;
/**
* Prepare to accumulate records to be sorted
*
* @param componentType Class of the record to be sorted. Necessary because of Java generic lameness.
* @param codec For writing records to file and reading them back into RAM
* @param comparator Defines output sort order
* @param maxRecordsInRam how many records to accumulate before spilling to disk
* @param printRecordSizeSampling If true record size will be sampled and output at DEBUG log level
* @param tmpDir Where to write files of records that will not fit in RAM
*/
private SortingCollection(final Class componentType, final SortingCollection.Codec codec,
final Comparator comparator, final int maxRecordsInRam,
final boolean printRecordSizeSampling, final Path... tmpDir) {
if (maxRecordsInRam <= 0) {
throw new IllegalArgumentException("maxRecordsInRam must be > 0");
}
if (tmpDir == null || tmpDir.length == 0) {
throw new IllegalArgumentException("At least one temp directory must be provided.");
}
this.tmpDirs = tmpDir;
this.codec = codec;
this.comparator = comparator;
this.maxRecordsInRam = maxRecordsInRam;
@SuppressWarnings("unchecked")
T[] ramRecords = (T[]) Array.newInstance(componentType, maxRecordsInRam);
this.ramRecords = ramRecords;
this.printRecordSizeSampling = printRecordSizeSampling;
}
public void add(final T rec) {
if (doneAdding) {
throw new IllegalStateException("Cannot add after calling doneAdding()");
}
if (iterationStarted) {
throw new IllegalStateException("Cannot add after calling iterator()");
}
if (numRecordsInRam == maxRecordsInRam) {
long startMem = 0;
if (printRecordSizeSampling) {
// Garbage collect and get free memory
Runtime.getRuntime().gc();
startMem = Runtime.getRuntime().freeMemory();
}
spillToDisk();
if (printRecordSizeSampling) {
//Garbage collect again and get free memory
Runtime.getRuntime().gc();
long endMem = Runtime.getRuntime().freeMemory();
long usedBytes = endMem - startMem;
log.debug(String.format("%d records in ram required approximately %s memory or %s per record. ", maxRecordsInRam,
StringUtil.humanReadableByteCount(usedBytes),
StringUtil.humanReadableByteCount(usedBytes / maxRecordsInRam)));
}
}
ramRecords[numRecordsInRam++] = rec;
}
/**
* This method can be called after caller is done adding to collection, in order to possibly free
* up memory. If iterator() is called immediately after caller is done adding, this is not necessary,
* because iterator() triggers the same freeing.
*/
public void doneAdding() {
if (this.cleanedUp) {
throw new IllegalStateException("Cannot call doneAdding() after cleanup() was called.");
}
if (doneAdding) {
return;
}
doneAdding = true;
if (this.files.isEmpty()) {
return;
}
if (this.numRecordsInRam > 0) {
spillToDisk();
}
// Facilitate GC
this.ramRecords = null;
}
/**
* @return True if this collection is allowed to discard data during iteration in order to reduce memory
* footprint, precluding a second iteration over the collection.
*/
public boolean isDestructiveIteration() {
return destructiveIteration;
}
/**
* Tell this collection that it is allowed to discard data during iteration in order to reduce memory footprint,
* precluding a second iteration. This is true by default.
*/
public void setDestructiveIteration(boolean destructiveIteration) {
this.destructiveIteration = destructiveIteration;
}
/**
* Sort the records in memory, write them to a file, and clear the buffer of records in memory.
*/
public void spillToDisk() {
try {
Arrays.parallelSort(this.ramRecords, 0, this.numRecordsInRam, this.comparator);
final Path f = newTempFile();
try (OutputStream os
= tempStreamFactory.wrapTempOutputStream(Files.newOutputStream(f), Defaults.BUFFER_SIZE)) {
this.codec.setOutputStream(os);
for (int i = 0; i < this.numRecordsInRam; ++i) {
this.codec.encode(ramRecords[i]);
// Facilitate GC
this.ramRecords[i] = null;
}
os.flush();
} catch (RuntimeIOException ex) {
throw new RuntimeIOException("Problem writing temporary file " + f.toUri() +
". Try setting TMP_DIR to a file system with lots of space.", ex);
}
this.numRecordsInRam = 0;
this.files.add(f);
} catch (IOException e) {
throw new RuntimeIOException(e);
}
}
/**
* Creates a new tmp file on one of the available temp filesystems, registers it for deletion
* on JVM exit and then returns it.
*/
private Path newTempFile() throws IOException {
/* The minimum amount of space free on a temp filesystem to write a file there. */
return IOUtil.newTempPath("sortingcollection.", ".tmp", this.tmpDirs, IOUtil.FIVE_GBS);
}
/**
* Prepare to iterate through the records in order. This method may be called more than once,
* but add() may not be called after this method has been called.
*/
@Override
public CloseableIterator iterator() {
if (this.cleanedUp) {
throw new IllegalStateException("Cannot call iterator() after cleanup() was called.");
}
doneAdding();
this.iterationStarted = true;
if (this.files.isEmpty()) {
return new InMemoryIterator();
} else {
return new MergingIterator();
}
}
/**
* Delete any temporary files. After this method is called, iterator() may not be called.
*/
public void cleanup() {
this.iterationStarted = true;
this.cleanedUp = true;
IOUtil.deletePaths(this.files);
}
/**
* Syntactic sugar around the ctor, to save some typing of type parameters
*
* @param componentType Class of the record to be sorted. Necessary because of Java generic lameness.
* @param codec For writing records to file and reading them back into RAM
* @param comparator Defines output sort order
* @param maxRecordsInRAM how many records to accumulate in memory before spilling to disk
* @param tmpDir Where to write files of records that will not fit in RAM
* @deprecated since 2017-09. Use {@link #newInstance(Class, Codec, Comparator, int, Path...)} instead
*/
@Deprecated
public static SortingCollection newInstance(final Class componentType,
final SortingCollection.Codec codec,
final Comparator comparator,
final int maxRecordsInRAM,
final File... tmpDir) {
return new SortingCollection<>(componentType, codec, comparator, maxRecordsInRAM, false, Arrays.stream(tmpDir).map(File::toPath).toArray(Path[]::new));
}
/**
* Syntactic sugar around the ctor, to save some typing of type parameters
*
* @param componentType Class of the record to be sorted. Necessary because of Java generic lameness.
* @param codec For writing records to file and reading them back into RAM
* @param comparator Defines output sort order
* @param maxRecordsInRAM how many records to accumulate in memory before spilling to disk
* @param tmpDirs Where to write files of records that will not fit in RAM
* @deprecated since 2017-09. Use {@link #newInstanceFromPaths(Class, Codec, Comparator, int, Collection)} instead
*/
@Deprecated
public static SortingCollection newInstance(final Class componentType,
final SortingCollection.Codec codec,
final Comparator comparator,
final int maxRecordsInRAM,
final Collection tmpDirs) {
return new SortingCollection<>(componentType,
codec,
comparator,
maxRecordsInRAM,
false,
tmpDirs.stream().map(File::toPath).toArray(Path[]::new));
}
/**
* Syntactic sugar around the ctor, to save some typing of type parameters. Writes files to java.io.tmpdir
*
* @param componentType Class of the record to be sorted. Necessary because of Java generic lameness.
* @param codec For writing records to file and reading them back into RAM
* @param comparator Defines output sort order
* @param maxRecordsInRAM how many records to accumulate in memory before spilling to disk
* @param printRecordSizeSampling If true record size will be sampled and output at DEBUG log level
*/
public static SortingCollection newInstance(final Class componentType,
final SortingCollection.Codec codec,
final Comparator comparator,
final int maxRecordsInRAM,
final boolean printRecordSizeSampling) {
final Path tmpDir = Paths.get(System.getProperty("java.io.tmpdir"));
return new SortingCollection<>(componentType, codec, comparator, maxRecordsInRAM, printRecordSizeSampling, tmpDir);
}
/**
* Syntactic sugar around the ctor, to save some typing of type parameters
*
* @param componentType Class of the record to be sorted. Necessary because of Java generic lameness.
* @param codec For writing records to file and reading them back into RAM
* @param comparator Defines output sort order
* @param maxRecordsInRAM how many records to accumulate in memory before spilling to disk
* @param printRecordSizeSampling If true record size will be sampled and output at DEBUG log level
* @param tmpDir Where to write files of records that will not fit in RAM
*/
public static SortingCollection newInstance(final Class componentType,
final SortingCollection.Codec codec,
final Comparator comparator,
final int maxRecordsInRAM,
final boolean printRecordSizeSampling,
final Path... tmpDir) {
return new SortingCollection<>(componentType, codec, comparator, maxRecordsInRAM, printRecordSizeSampling, tmpDir);
}
/**
* Syntactic sugar around the ctor, to save some typing of type parameters. Writes files to java.io.tmpdir
*
* @param componentType Class of the record to be sorted. Necessary because of Java generic lameness.
* @param codec For writing records to file and reading them back into RAM
* @param comparator Defines output sort order
* @param maxRecordsInRAM how many records to accumulate in memory before spilling to disk
*/
public static SortingCollection newInstance(final Class componentType,
final SortingCollection.Codec codec,
final Comparator comparator,
final int maxRecordsInRAM) {
final Path tmpDir = Paths.get(System.getProperty("java.io.tmpdir"));
return new SortingCollection<>(componentType, codec, comparator, maxRecordsInRAM, false, tmpDir);
}
/**
* Syntactic sugar around the ctor, to save some typing of type parameters
*
* @param componentType Class of the record to be sorted. Necessary because of Java generic lameness.
* @param codec For writing records to file and reading them back into RAM
* @param comparator Defines output sort order
* @param maxRecordsInRAM how many records to accumulate in memory before spilling to disk
* @param tmpDir Where to write files of records that will not fit in RAM
*/
public static SortingCollection newInstance(final Class componentType,
final SortingCollection.Codec codec,
final Comparator comparator,
final int maxRecordsInRAM,
final Path... tmpDir) {
return new SortingCollection<>(componentType, codec, comparator, maxRecordsInRAM, false, tmpDir);
}
/**
* Syntactic sugar around the ctor, to save some typing of type parameters
*
* @param componentType Class of the record to be sorted. Necessary because of Java generic lameness.
* @param codec For writing records to file and reading them back into RAM
* @param comparator Defines output sort order
* @param maxRecordsInRAM how many records to accumulate in memory before spilling to disk
* @param tmpDirs Where to write files of records that will not fit in RAM
*/
public static SortingCollection newInstanceFromPaths(final Class componentType,
final SortingCollection.Codec codec,
final Comparator comparator,
final int maxRecordsInRAM,
final Collection tmpDirs) {
return new SortingCollection<>(componentType,
codec,
comparator,
maxRecordsInRAM,
false,
tmpDirs.toArray(new Path[tmpDirs.size()]));
}
/**
* For iteration when number of records added is less than the threshold for spilling to disk.
*/
class InMemoryIterator implements CloseableIterator {
private int iterationIndex = 0;
InMemoryIterator() {
Arrays.parallelSort(SortingCollection.this.ramRecords,
0,
SortingCollection.this.numRecordsInRam,
SortingCollection.this.comparator);
}
@Override
public void close() {
// nothing to do
}
@Override
public boolean hasNext() {
return this.iterationIndex < SortingCollection.this.numRecordsInRam;
}
@Override
public T next() {
if (!hasNext()) {
throw new NoSuchElementException();
}
T ret = SortingCollection.this.ramRecords[iterationIndex];
if (destructiveIteration) SortingCollection.this.ramRecords[iterationIndex] = null;
++iterationIndex;
return ret;
}
@Override
public void remove() {
throw new UnsupportedOperationException();
}
}
/**
* For iteration when spilling to disk has occurred.
* Each file is has records in sort order within the file.
* This iterator automatically closes when it iterates to the end, but if not iterating
* to the end it is a good idea to call close().
*
* Algorithm: MergingIterator maintains a PriorityQueue of PeekFileRecordIterators.
* Each PeekFileRecordIterator iterates through a file in which the records are sorted.
* The comparator for PeekFileRecordIterator used by the PriorityQueue peeks at the next record from
* the file, so the first element in the PriorityQueue is the file that has the next record to be emitted.
* In order to get the next record, the first PeekFileRecordIterator in the PriorityQueue is popped,
* the record is obtained from that iterator, and then if that iterator is not empty, it is pushed back into
* the PriorityQueue. Because it now has a different record as its next element, it may go into another
* location in the PriorityQueue
*/
class MergingIterator implements CloseableIterator {
private final TreeSet queue;
MergingIterator() {
this.queue = new TreeSet<>(new PeekFileRecordIteratorComparator());
int n = 0;
log.debug(String.format("Creating merging iterator from %d files", files.size()));
int suggestedBufferSize = checkMemoryAndAdjustBuffer(files.size());
for (final Path f : files) {
final FileRecordIterator it = new FileRecordIterator(f, suggestedBufferSize);
if (it.hasNext()) {
this.queue.add(new PeekFileRecordIterator(it, n++));
} else {
it.close();
}
}
}
// Since we need to open and buffer all temp files in the sorting collection at once it is important
// to have enough memory left to do this. This method checks to make sure that, given the number of files and
// the size of the buffer, we can reasonably open all files. If we can't it will return a buffer size that
// is appropriate given the number of temp files and the amount of memory left on the heap. If there isn't
// enough memory for buffering it will return zero and all reading will be unbuffered.
private int checkMemoryAndAdjustBuffer(int numFiles) {
int bufferSize = Defaults.BUFFER_SIZE;
// garbage collect so that our calculation is accurate.
Runtime.getRuntime().gc();
// There is ~20k in overhead per file.
final long freeMemory = Runtime.getRuntime().freeMemory() - (numFiles * 20 * 1024);
// use the floor value from the divide
final int memoryPerFile = (int) (freeMemory / numFiles);
if (memoryPerFile < 0) {
log.warn("There is not enough memory per file for buffering. Reading will be unbuffered.");
bufferSize = 0;
} else if (bufferSize > memoryPerFile) {
log.warn(String.format("Default io buffer size of %s is larger than available memory per file of %s.",
StringUtil.humanReadableByteCount(bufferSize),
StringUtil.humanReadableByteCount(memoryPerFile)));
bufferSize = memoryPerFile;
}
return bufferSize;
}
@Override
public boolean hasNext() {
return !this.queue.isEmpty();
}
@Override
public T next() {
if (!hasNext()) {
throw new NoSuchElementException();
}
final PeekFileRecordIterator fileIterator = queue.pollFirst();
final T ret = fileIterator.next();
if (fileIterator.hasNext()) {
this.queue.add(fileIterator);
} else {
((CloseableIterator) fileIterator.getUnderlyingIterator()).close();
}
return ret;
}
@Override
public void remove() {
throw new UnsupportedOperationException();
}
@Override
public void close() {
while (!this.queue.isEmpty()) {
final PeekFileRecordIterator it = this.queue.pollFirst();
((CloseableIterator) it.getUnderlyingIterator()).close();
}
}
}
/**
* Read a file of records in format defined by the codec
*/
class FileRecordIterator implements CloseableIterator {
private final Path file;
private final InputStream is;
private final Codec codec;
private T currentRecord = null;
FileRecordIterator(final Path file, final int bufferSize) {
this.file = file;
try {
this.is = Files.newInputStream(file);
this.codec = SortingCollection.this.codec.clone();
this.codec.setInputStream(tempStreamFactory.wrapTempInputStream(this.is, bufferSize));
advance();
} catch (IOException e) {
throw new RuntimeIOException(e);
}
}
@Override
public boolean hasNext() {
return this.currentRecord != null;
}
@Override
public T next() {
if (!hasNext()) {
throw new NoSuchElementException();
}
final T ret = this.currentRecord;
advance();
return ret;
}
@Override
public void remove() {
throw new UnsupportedOperationException();
}
private void advance() {
this.currentRecord = this.codec.decode();
}
@Override
public void close() {
CloserUtil.close(this.is);
}
}
/**
* Just a typedef
*/
class PeekFileRecordIterator extends PeekIterator {
final int n; // A serial number used for tie-breaking in the sort
PeekFileRecordIterator(final Iterator underlyingIterator, final int n) {
super(underlyingIterator);
this.n = n;
}
}
class PeekFileRecordIteratorComparator implements Comparator {
@Override
public int compare(final PeekFileRecordIterator lhs, final PeekFileRecordIterator rhs) {
final int result = comparator.compare(lhs.peek(), rhs.peek());
if (result == 0) return lhs.n - rhs.n;
else return result;
}
}
}