/***********************************************************************************************************************
* Copyright (C) 2010-2013 by the Stratosphere project (http://stratosphere.eu)
*
* Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
* an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
* specific language governing permissions and limitations under the License.
**********************************************************************************************************************/
package eu.stratosphere.pact.runtime.hash;
import java.io.EOFException;
import java.io.IOException;
import java.util.ArrayList;
import java.util.List;
import java.util.concurrent.atomic.AtomicBoolean;
import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;
import eu.stratosphere.api.common.typeutils.TypeComparator;
import eu.stratosphere.api.common.typeutils.TypePairComparator;
import eu.stratosphere.api.common.typeutils.TypeSerializer;
import eu.stratosphere.core.memory.MemorySegment;
import eu.stratosphere.nephele.services.memorymanager.ListMemorySegmentSource;
import eu.stratosphere.pact.runtime.util.MathUtils;
import eu.stratosphere.util.MutableObjectIterator;
/**
* An implementation of an in-memory Hash Table for variable-length records.
* <p>
* The design of this class follows on many parts the design presented in
* "Hash joins and hash teams in Microsoft SQL Server", by Goetz Graefe et al..
*<p>
*
*
* <hr>
*
* The layout of the buckets inside a memory segment is as follows:
*
* <pre>
* +----------------------------- Bucket x ----------------------------
* |Partition (1 byte) | Status (1 byte) | element count (2 bytes) |
* | next-bucket-in-chain-pointer (8 bytes) | reserved (4 bytes) |
* |
* |hashCode 1 (4 bytes) | hashCode 2 (4 bytes) | hashCode 3 (4 bytes) |
* | ... hashCode n-1 (4 bytes) | hashCode n (4 bytes)
* |
* |pointer 1 (8 bytes) | pointer 2 (8 bytes) | pointer 3 (8 bytes) |
* | ... pointer n-1 (8 bytes) | pointer n (8 bytes)
* |
* +---------------------------- Bucket x + 1--------------------------
* |Partition (1 byte) | Status (1 byte) | element count (2 bytes) |
* | next-bucket-in-chain-pointer (8 bytes) | reserved (4 bytes) |
* |
* |hashCode 1 (4 bytes) | hashCode 2 (4 bytes) | hashCode 3 (4 bytes) |
* | ... hashCode n-1 (4 bytes) | hashCode n (4 bytes)
* |
* |pointer 1 (8 bytes) | pointer 2 (8 bytes) | pointer 3 (8 bytes) |
* | ... pointer n-1 (8 bytes) | pointer n (8 bytes)
* +-------------------------------------------------------------------
* | ...
* |
* </pre>
* @param <T>
*
* @param T record type stored in hash table
*
*/
public class CompactingHashTable<T> extends AbstractMutableHashTable<T>{
private static final Log LOG = LogFactory.getLog(CompactingHashTable.class);
// ------------------------------------------------------------------------
// Internal Constants
// ------------------------------------------------------------------------
private static final int MIN_NUM_MEMORY_SEGMENTS = 33;
/**
* The maximum number of partitions
*/
private static final int MAX_NUM_PARTITIONS = 32;
/**
* The default record width that is used when no width is given. The record width is
* used to determine the ratio of the number of memory segments intended for partition
* buffers and the number of memory segments in the hash-table structure.
*/
private static final int DEFAULT_RECORD_LEN = 24; //FIXME maybe find a better default
/**
* The length of the hash code stored in the bucket.
*/
private static final int HASH_CODE_LEN = 4;
/**
* The length of a pointer from a hash bucket to the record in the buffers.
*/
private static final int POINTER_LEN = 8;
/**
* The number of bytes that the entry in the hash structure occupies, in bytes.
* It corresponds to a 4 byte hash value and an 8 byte pointer.
*/
private static final int RECORD_TABLE_BYTES = HASH_CODE_LEN + POINTER_LEN;
/**
* The total storage overhead per record, in bytes. This corresponds to the space in the
* actual hash table buckets, consisting of a 4 byte hash value and an 8 byte
* pointer, plus the overhead for the stored length field.
*/
private static final int RECORD_OVERHEAD_BYTES = RECORD_TABLE_BYTES;
// -------------------------- Bucket Size and Structure -------------------------------------
private static final int NUM_INTRA_BUCKET_BITS = 7;
private static final int HASH_BUCKET_SIZE = 0x1 << NUM_INTRA_BUCKET_BITS;
private static final int BUCKET_HEADER_LENGTH = 16;
private static final int NUM_ENTRIES_PER_BUCKET = (HASH_BUCKET_SIZE - BUCKET_HEADER_LENGTH) / RECORD_OVERHEAD_BYTES;
private static final int BUCKET_POINTER_START_OFFSET = BUCKET_HEADER_LENGTH + (NUM_ENTRIES_PER_BUCKET * HASH_CODE_LEN);
// ------------------------------ Bucket Header Fields ------------------------------
/**
* Offset of the field in the bucket header indicating the bucket's partition.
*/
private static final int HEADER_PARTITION_OFFSET = 0;
/**
* Offset of the field in the bucket header indicating the bucket's status (spilled or in-memory).
*/
private static final int HEADER_COUNT_OFFSET = 4;
/**
* Offset of the field in the bucket header that holds the forward pointer to its
* first overflow bucket.
*/
private static final int HEADER_FORWARD_OFFSET = 8;
/**
* Constant for the forward pointer, indicating that the pointer is not set.
*/
private static final long BUCKET_FORWARD_POINTER_NOT_SET = ~0x0L;
// ------------------------------------------------------------------------
// Members
// ------------------------------------------------------------------------
/**
* The free memory segments currently available to the hash join.
*/
private final ArrayList<MemorySegment> availableMemory;
/**
* The size of the segments used by the hash join buckets. All segments must be of equal size to ease offset computations.
*/
private final int segmentSize;
/**
* The number of hash table buckets in a single memory segment - 1.
* Because memory segments can be comparatively large, we fit multiple buckets into one memory segment.
* This variable is a mask that is 1 in the lower bits that define the number of a bucket
* in a segment.
*/
private final int bucketsPerSegmentMask;
/**
* The number of bits that describe the position of a bucket in a memory segment. Computed as log2(bucketsPerSegment).
*/
private final int bucketsPerSegmentBits;
/**
* An estimate for the average record length.
*/
private final int avgRecordLen;
// ------------------------------------------------------------------------
/**
* The partitions of the hash table.
*/
private final ArrayList<InMemoryPartition<T>> partitions;
/**
* The array of memory segments that contain the buckets which form the actual hash-table
* of hash-codes and pointers to the elements.
*/
private MemorySegment[] buckets;
/**
* temporary storage for partition compaction (always attempts to allocate as many segments as the largest partition)
*/
private InMemoryPartition<T> compactionMemory;
/**
* The number of buckets in the current table. The bucket array is not necessarily fully
* used, when not all buckets that would fit into the last segment are actually used.
*/
private int numBuckets;
private AtomicBoolean closed = new AtomicBoolean();
private boolean running = true;
private int pageSizeInBits;
// ------------------------------------------------------------------------
// Construction and Teardown
// ------------------------------------------------------------------------
public CompactingHashTable(TypeSerializer<T> buildSideSerializer, TypeComparator<T> buildSideComparator, List<MemorySegment> memorySegments)
{
this(buildSideSerializer, buildSideComparator, memorySegments, DEFAULT_RECORD_LEN);
}
public CompactingHashTable(TypeSerializer<T> buildSideSerializer, TypeComparator<T> buildSideComparator, List<MemorySegment> memorySegments, int avgRecordLen)
{
super(buildSideSerializer, buildSideComparator);
// some sanity checks first
if (memorySegments == null) {
throw new NullPointerException();
}
if (memorySegments.size() < MIN_NUM_MEMORY_SEGMENTS) {
throw new IllegalArgumentException("Too few memory segments provided. Hash Join needs at least " +
MIN_NUM_MEMORY_SEGMENTS + " memory segments.");
}
this.availableMemory = (memorySegments instanceof ArrayList) ?
(ArrayList<MemorySegment>) memorySegments :
new ArrayList<MemorySegment>(memorySegments);
this.avgRecordLen = buildSideSerializer.getLength() > 0 ? buildSideSerializer.getLength() : avgRecordLen;
// check the size of the first buffer and record it. all further buffers must have the same size.
// the size must also be a power of 2
this.segmentSize = memorySegments.get(0).size();
if ( (this.segmentSize & this.segmentSize - 1) != 0) {
throw new IllegalArgumentException("Hash Table requires buffers whose size is a power of 2.");
}
int bucketsPerSegment = this.segmentSize >> NUM_INTRA_BUCKET_BITS;
if (bucketsPerSegment == 0) {
throw new IllegalArgumentException("Hash Table requires buffers of at least " + HASH_BUCKET_SIZE + " bytes.");
}
this.bucketsPerSegmentMask = bucketsPerSegment - 1;
this.bucketsPerSegmentBits = MathUtils.log2strict(bucketsPerSegment);
this.partitions = new ArrayList<InMemoryPartition<T>>();
// because we allow to open and close multiple times, the state is initially closed
this.closed.set(true);
// so far no partition has any MemorySegments
}
// ------------------------------------------------------------------------
// Life-Cycle
// ------------------------------------------------------------------------
/**
* Build the hash table
*
* @throws IOException Thrown, if an I/O problem occurs while spilling a partition.
*/
public void open() {
// sanity checks
if (!this.closed.compareAndSet(true, false)) {
throw new IllegalStateException("Hash Table cannot be opened, because it is currently not closed.");
}
// create the partitions
final int partitionFanOut = getPartitioningFanOutNoEstimates(this.availableMemory.size());
createPartitions(partitionFanOut);
// set up the table structure. the write behind buffers are taken away, as are one buffer per partition
final int numBuckets = getInitialTableSize(this.availableMemory.size(), this.segmentSize,
partitionFanOut, this.avgRecordLen);
initTable(numBuckets, (byte) partitionFanOut);
}
/**
* Closes the hash table. This effectively releases all internal structures and closes all
* open files and removes them. The call to this method is valid both as a cleanup after the
* complete inputs were properly processed, and as an cancellation call, which cleans up
* all resources that are currently held by the hash join.
*/
public void close() {
// make sure that we close only once
if (!this.closed.compareAndSet(false, true)) {
return;
}
LOG.debug("Closing hash table and releasing resources.");
// release the table structure
releaseTable();
// clear the memory in the partitions
clearPartitions();
}
public void abort() {
this.running = false;
LOG.debug("Cancelling hash table operations.");
}
public List<MemorySegment> getFreeMemory() {
if (!this.closed.get()) {
throw new IllegalStateException("Cannot return memory while join is open.");
}
return this.availableMemory;
}
public void buildTable(final MutableObjectIterator<T> input) throws IOException {
T record = this.buildSideSerializer.createInstance();
// go over the complete input and insert every element into the hash table
while (this.running && ((record = input.next(record)) != null)) {
insert(record);
}
}
public final void insert(T record) throws IOException {
final int hashCode = hash(this.buildSideComparator.hash(record));
final int posHashCode = hashCode % this.numBuckets;
// get the bucket for the given hash code
final int bucketArrayPos = posHashCode >>> this.bucketsPerSegmentBits;
final int bucketInSegmentPos = (posHashCode & this.bucketsPerSegmentMask) << NUM_INTRA_BUCKET_BITS;
final MemorySegment bucket = this.buckets[bucketArrayPos];
// get the basic characteristics of the bucket
final int partitionNumber = bucket.get(bucketInSegmentPos + HEADER_PARTITION_OFFSET);
final InMemoryPartition<T> p = this.partitions.get(partitionNumber);
long pointer;
try {
pointer = p.appendRecord(record);
if((pointer >> this.pageSizeInBits) > this.compactionMemory.getBlockCount()) {
this.compactionMemory.allocateSegments((int)(pointer >> this.pageSizeInBits));
}
} catch (EOFException e) {
try {
compactPartition(partitionNumber);
// retry append
pointer = this.partitions.get(partitionNumber).appendRecord(record);
} catch (EOFException ex) {
throw new RuntimeException("Memory ran out. Compaction failed. numPartitions: " + this.partitions.size() +
" minPartition: " + getMinPartition() +
" maxPartition: " + getMaxPartition() +
" bucketSize: " + this.buckets.length +
" Message: " + ex.getMessage());
} catch (IndexOutOfBoundsException ex) {
throw new RuntimeException("Memory ran out. Compaction failed. numPartitions: " + this.partitions.size() +
" minPartition: " + getMinPartition() +
" maxPartition: " + getMaxPartition() +
" bucketSize: " + this.buckets.length +
" Message: " + ex.getMessage());
}
} catch (IndexOutOfBoundsException e1) {
try {
compactPartition(partitionNumber);
// retry append
pointer = this.partitions.get(partitionNumber).appendRecord(record);
} catch (EOFException ex) {
throw new RuntimeException("Memory ran out. Compaction failed. numPartitions: " + this.partitions.size() +
" minPartition: " + getMinPartition() +
" maxPartition: " + getMaxPartition() +
" bucketSize: " + this.buckets.length +
" Message: " + ex.getMessage());
} catch (IndexOutOfBoundsException ex) {
throw new RuntimeException("Memory ran out. Compaction failed. numPartitions: " + this.partitions.size() +
" minPartition: " + getMinPartition() +
" maxPartition: " + getMaxPartition() +
" bucketSize: " + this.buckets.length +
" Message: " + ex.getMessage());
}
}
insertBucketEntryFromStart(p, bucket, bucketInSegmentPos, hashCode, pointer);
}
@Override
public <PT> HashTableProber<PT> getProber(TypeComparator<PT> probeSideComparator, TypePairComparator<PT, T> pairComparator) {
return new HashTableProber<PT>(probeSideComparator, pairComparator);
}
/**
*
* @return Iterator over hash table
* @see EntryIterator
*/
public MutableObjectIterator<T> getEntryIterator() {
return new EntryIterator(this);
}
/**
* Replaces record in hash table if record already present or append record if not.
* May trigger expensive compaction.
*
* @param record record to insert or replace
* @param tempHolder instance of T that will be overwritten
* @throws IOException
*/
public void insertOrReplaceRecord(T record, T tempHolder) throws IOException {
final int searchHashCode = hash(this.buildSideComparator.hash(record));
final int posHashCode = searchHashCode % this.numBuckets;
// get the bucket for the given hash code
MemorySegment originalBucket = this.buckets[posHashCode >> this.bucketsPerSegmentBits];
int originalBucketOffset = (posHashCode & this.bucketsPerSegmentMask) << NUM_INTRA_BUCKET_BITS;
MemorySegment bucket = originalBucket;
int bucketInSegmentOffset = originalBucketOffset;
// get the basic characteristics of the bucket
final int partitionNumber = bucket.get(bucketInSegmentOffset + HEADER_PARTITION_OFFSET);
final InMemoryPartition<T> partition = this.partitions.get(partitionNumber);
final MemorySegment[] overflowSegments = partition.overflowSegments;
this.buildSideComparator.setReference(record);
int countInSegment = bucket.getInt(bucketInSegmentOffset + HEADER_COUNT_OFFSET);
int numInSegment = 0;
int posInSegment = bucketInSegmentOffset + BUCKET_HEADER_LENGTH;
long currentForwardPointer = BUCKET_FORWARD_POINTER_NOT_SET;
// loop over all segments that are involved in the bucket (original bucket plus overflow buckets)
while (true) {
while (numInSegment < countInSegment) {
final int thisCode = bucket.getInt(posInSegment);
posInSegment += HASH_CODE_LEN;
// check if the hash code matches
if (thisCode == searchHashCode) {
// get the pointer to the pair
final int pointerOffset = bucketInSegmentOffset + BUCKET_POINTER_START_OFFSET + (numInSegment * POINTER_LEN);
final long pointer = bucket.getLong(pointerOffset);
numInSegment++;
// deserialize the key to check whether it is really equal, or whether we had only a hash collision
try {
partition.readRecordAt(pointer, tempHolder);
if (this.buildSideComparator.equalToReference(tempHolder)) {
long newPointer = partition.appendRecord(record);
bucket.putLong(pointerOffset, newPointer);
partition.setCompaction(false);
if((newPointer >> this.pageSizeInBits) > this.compactionMemory.getBlockCount()) {
this.compactionMemory.allocateSegments((int)(newPointer >> this.pageSizeInBits));
}
return;
}
} catch (EOFException e) {
// system is out of memory so we attempt to reclaim memory with a copy compact run
long newPointer;
try {
compactPartition(partition.getPartitionNumber());
// retry append
newPointer = this.partitions.get(partitionNumber).appendRecord(record);
} catch (EOFException ex) {
throw new RuntimeException("Memory ran out. Compaction failed. numPartitions: " + this.partitions.size() +
" minPartition: " + getMinPartition() +
" maxPartition: " + getMaxPartition() +
" bucketSize: " + this.buckets.length +
" Message: " + ex.getMessage());
} catch (IndexOutOfBoundsException ex) {
throw new RuntimeException("Memory ran out. Compaction failed. numPartitions: " + this.partitions.size() +
" minPartition: " + getMinPartition() +
" maxPartition: " + getMaxPartition() +
" bucketSize: " + this.buckets.length +
" Message: " + ex.getMessage());
}
bucket.putLong(pointerOffset, newPointer);
return;
} catch (IndexOutOfBoundsException e) {
// system is out of memory so we attempt to reclaim memory with a copy compact run
long newPointer;
try {
compactPartition(partition.getPartitionNumber());
// retry append
newPointer = this.partitions.get(partitionNumber).appendRecord(record);
} catch (EOFException ex) {
throw new RuntimeException("Memory ran out. Compaction failed. numPartitions: " + this.partitions.size() +
" minPartition: " + getMinPartition() +
" maxPartition: " + getMaxPartition() +
" bucketSize: " + this.buckets.length +
" Message: " + ex.getMessage());
} catch (IndexOutOfBoundsException ex) {
throw new RuntimeException("Memory ran out. Compaction failed. numPartitions: " + this.partitions.size() +
" minPartition: " + getMinPartition() +
" maxPartition: " + getMaxPartition() +
" bucketSize: " + this.buckets.length +
" Message: " + ex.getMessage());
}
bucket.putLong(pointerOffset, newPointer);
return;
} catch (IOException e) {
throw new RuntimeException("Error deserializing record from the hashtable: " + e.getMessage(), e);
}
}
else {
numInSegment++;
}
}
// this segment is done. check if there is another chained bucket
long newForwardPointer = bucket.getLong(bucketInSegmentOffset + HEADER_FORWARD_OFFSET);
if (newForwardPointer == BUCKET_FORWARD_POINTER_NOT_SET) {
// nothing found. append and insert
long pointer = partition.appendRecord(record);
insertBucketEntryFromSearch(partition, originalBucket, bucket, originalBucketOffset, bucketInSegmentOffset, countInSegment, currentForwardPointer, searchHashCode, pointer);
if((pointer >> this.pageSizeInBits) > this.compactionMemory.getBlockCount()) {
this.compactionMemory.allocateSegments((int)(pointer >> this.pageSizeInBits));
}
return;
}
final int overflowSegNum = (int) (newForwardPointer >>> 32);
bucket = overflowSegments[overflowSegNum];
bucketInSegmentOffset = (int) (newForwardPointer & 0xffffffff);
countInSegment = bucket.getInt(bucketInSegmentOffset + HEADER_COUNT_OFFSET);
posInSegment = bucketInSegmentOffset + BUCKET_HEADER_LENGTH;
numInSegment = 0;
currentForwardPointer = newForwardPointer;
}
}
private final void insertBucketEntryFromStart(InMemoryPartition<T> p, MemorySegment bucket,
int bucketInSegmentPos, int hashCode, long pointer)
throws IOException
{
// find the position to put the hash code and pointer
final int count = bucket.getInt(bucketInSegmentPos + HEADER_COUNT_OFFSET);
if (count < NUM_ENTRIES_PER_BUCKET) {
// we are good in our current bucket, put the values
bucket.putInt(bucketInSegmentPos + BUCKET_HEADER_LENGTH + (count * HASH_CODE_LEN), hashCode); // hash code
bucket.putLong(bucketInSegmentPos + BUCKET_POINTER_START_OFFSET + (count * POINTER_LEN), pointer); // pointer
bucket.putInt(bucketInSegmentPos + HEADER_COUNT_OFFSET, count + 1); // update count
}
else {
// we need to go to the overflow buckets
final long originalForwardPointer = bucket.getLong(bucketInSegmentPos + HEADER_FORWARD_OFFSET);
final long forwardForNewBucket;
if (originalForwardPointer != BUCKET_FORWARD_POINTER_NOT_SET) {
// forward pointer set
final int overflowSegNum = (int) (originalForwardPointer >>> 32);
final int segOffset = (int) (originalForwardPointer & 0xffffffff);
final MemorySegment seg = p.overflowSegments[overflowSegNum];
final int obCount = seg.getInt(segOffset + HEADER_COUNT_OFFSET);
// check if there is space in this overflow bucket
if (obCount < NUM_ENTRIES_PER_BUCKET) {
// space in this bucket and we are done
seg.putInt(segOffset + BUCKET_HEADER_LENGTH + (obCount * HASH_CODE_LEN), hashCode); // hash code
seg.putLong(segOffset + BUCKET_POINTER_START_OFFSET + (obCount * POINTER_LEN), pointer); // pointer
seg.putInt(segOffset + HEADER_COUNT_OFFSET, obCount + 1); // update count
return;
}
else {
// no space here, we need a new bucket. this current overflow bucket will be the
// target of the new overflow bucket
forwardForNewBucket = originalForwardPointer;
}
}
else {
// no overflow bucket yet, so we need a first one
forwardForNewBucket = BUCKET_FORWARD_POINTER_NOT_SET;
}
// we need a new overflow bucket
MemorySegment overflowSeg;
final int overflowBucketNum;
final int overflowBucketOffset;
// first, see if there is space for an overflow bucket remaining in the last overflow segment
if (p.nextOverflowBucket == 0) {
// no space left in last bucket, or no bucket yet, so create an overflow segment
overflowSeg = getNextBuffer();
overflowBucketOffset = 0;
overflowBucketNum = p.numOverflowSegments;
// add the new overflow segment
if (p.overflowSegments.length <= p.numOverflowSegments) {
MemorySegment[] newSegsArray = new MemorySegment[p.overflowSegments.length * 2];
System.arraycopy(p.overflowSegments, 0, newSegsArray, 0, p.overflowSegments.length);
p.overflowSegments = newSegsArray;
}
p.overflowSegments[p.numOverflowSegments] = overflowSeg;
p.numOverflowSegments++;
}
else {
// there is space in the last overflow bucket
overflowBucketNum = p.numOverflowSegments - 1;
overflowSeg = p.overflowSegments[overflowBucketNum];
overflowBucketOffset = p.nextOverflowBucket << NUM_INTRA_BUCKET_BITS;
}
// next overflow bucket is one ahead. if the segment is full, the next will be at the beginning
// of a new segment
p.nextOverflowBucket = (p.nextOverflowBucket == this.bucketsPerSegmentMask ? 0 : p.nextOverflowBucket + 1);
// insert the new overflow bucket in the chain of buckets
// 1) set the old forward pointer
// 2) let the bucket in the main table point to this one
overflowSeg.putLong(overflowBucketOffset + HEADER_FORWARD_OFFSET, forwardForNewBucket);
final long pointerToNewBucket = (((long) overflowBucketNum) << 32) | ((long) overflowBucketOffset);
bucket.putLong(bucketInSegmentPos + HEADER_FORWARD_OFFSET, pointerToNewBucket);
// finally, insert the values into the overflow buckets
overflowSeg.putInt(overflowBucketOffset + BUCKET_HEADER_LENGTH, hashCode); // hash code
overflowSeg.putLong(overflowBucketOffset + BUCKET_POINTER_START_OFFSET, pointer); // pointer
// set the count to one
overflowSeg.putInt(overflowBucketOffset + HEADER_COUNT_OFFSET, 1);
}
}
private final void insertBucketEntryFromSearch(InMemoryPartition<T> partition, MemorySegment originalBucket, MemorySegment currentBucket, int originalBucketOffset, int currentBucketOffset, int countInCurrentBucket, long currentForwardPointer, int hashCode, long pointer) {
if (countInCurrentBucket < NUM_ENTRIES_PER_BUCKET) {
// we are good in our current bucket, put the values
currentBucket.putInt(currentBucketOffset + BUCKET_HEADER_LENGTH + (countInCurrentBucket * HASH_CODE_LEN), hashCode); // hash code
currentBucket.putLong(currentBucketOffset + BUCKET_POINTER_START_OFFSET + (countInCurrentBucket * POINTER_LEN), pointer); // pointer
currentBucket.putInt(currentBucketOffset + HEADER_COUNT_OFFSET, countInCurrentBucket + 1); // update count
}
else {
// we need a new overflow bucket
MemorySegment overflowSeg;
final int overflowBucketNum;
final int overflowBucketOffset;
// first, see if there is space for an overflow bucket remaining in the last overflow segment
if (partition.nextOverflowBucket == 0) {
// no space left in last bucket, or no bucket yet, so create an overflow segment
overflowSeg = getNextBuffer();
overflowBucketOffset = 0;
overflowBucketNum = partition.numOverflowSegments;
// add the new overflow segment
if (partition.overflowSegments.length <= partition.numOverflowSegments) {
MemorySegment[] newSegsArray = new MemorySegment[partition.overflowSegments.length * 2];
System.arraycopy(partition.overflowSegments, 0, newSegsArray, 0, partition.overflowSegments.length);
partition.overflowSegments = newSegsArray;
}
partition.overflowSegments[partition.numOverflowSegments] = overflowSeg;
partition.numOverflowSegments++;
}
else {
// there is space in the last overflow segment
overflowBucketNum = partition.numOverflowSegments - 1;
overflowSeg = partition.overflowSegments[overflowBucketNum];
overflowBucketOffset = partition.nextOverflowBucket << NUM_INTRA_BUCKET_BITS;
}
// next overflow bucket is one ahead. if the segment is full, the next will be at the beginning
// of a new segment
partition.nextOverflowBucket = (partition.nextOverflowBucket == this.bucketsPerSegmentMask ? 0 : partition.nextOverflowBucket + 1);
// insert the new overflow bucket in the chain of buckets
// 1) set the old forward pointer
// 2) let the bucket in the main table point to this one
overflowSeg.putLong(overflowBucketOffset + HEADER_FORWARD_OFFSET, currentForwardPointer);
final long pointerToNewBucket = (((long) overflowBucketNum) << 32) | ((long) overflowBucketOffset);
originalBucket.putLong(originalBucketOffset + HEADER_FORWARD_OFFSET, pointerToNewBucket);
// finally, insert the values into the overflow buckets
overflowSeg.putInt(overflowBucketOffset + BUCKET_HEADER_LENGTH, hashCode); // hash code
overflowSeg.putLong(overflowBucketOffset + BUCKET_POINTER_START_OFFSET, pointer); // pointer
// set the count to one
overflowSeg.putInt(overflowBucketOffset + HEADER_COUNT_OFFSET, 1);
}
}
// --------------------------------------------------------------------------------------------
// Setup and Tear Down of Structures
// --------------------------------------------------------------------------------------------
private void createPartitions(int numPartitions) {
this.partitions.clear();
ListMemorySegmentSource memSource = new ListMemorySegmentSource(this.availableMemory);
this.pageSizeInBits = MathUtils.log2strict(this.segmentSize);
for (int i = 0; i < numPartitions; i++) {
this.partitions.add(new InMemoryPartition<T>(this.buildSideSerializer, i, memSource, this.segmentSize, pageSizeInBits));
}
this.compactionMemory = new InMemoryPartition<T>(this.buildSideSerializer, -1, memSource, this.segmentSize, pageSizeInBits);
}
private void clearPartitions() {
for (int i = 0; i < this.partitions.size(); i++) {
InMemoryPartition<T> p = this.partitions.get(i);
p.clearAllMemory(this.availableMemory);
}
this.partitions.clear();
this.compactionMemory.clearAllMemory(availableMemory);
}
private void initTable(int numBuckets, byte numPartitions) {
final int bucketsPerSegment = this.bucketsPerSegmentMask + 1;
final int numSegs = (numBuckets >>> this.bucketsPerSegmentBits) + ( (numBuckets & this.bucketsPerSegmentMask) == 0 ? 0 : 1);
final MemorySegment[] table = new MemorySegment[numSegs];
// go over all segments that are part of the table
for (int i = 0, bucket = 0; i < numSegs && bucket < numBuckets; i++) {
final MemorySegment seg = getNextBuffer();
// go over all buckets in the segment
for (int k = 0; k < bucketsPerSegment && bucket < numBuckets; k++, bucket++) {
final int bucketOffset = k * HASH_BUCKET_SIZE;
// compute the partition that the bucket corresponds to
final byte partition = assignPartition(bucket, numPartitions);
// initialize the header fields
seg.put(bucketOffset + HEADER_PARTITION_OFFSET, partition);
seg.putInt(bucketOffset + HEADER_COUNT_OFFSET, 0);
seg.putLong(bucketOffset + HEADER_FORWARD_OFFSET, BUCKET_FORWARD_POINTER_NOT_SET);
}
table[i] = seg;
}
this.buckets = table;
this.numBuckets = numBuckets;
}
private void releaseTable() {
// set the counters back
this.numBuckets = 0;
if (this.buckets != null) {
for (int i = 0; i < this.buckets.length; i++) {
this.availableMemory.add(this.buckets[i]);
}
this.buckets = null;
}
}
private final MemorySegment getNextBuffer() {
// check if the list directly offers memory
int s = this.availableMemory.size();
if (s > 0) {
return this.availableMemory.remove(s-1);
} else {
throw new RuntimeException("Memory ran out. numPartitions: " + this.partitions.size() +
" minPartition: " + getMinPartition() +
" maxPartition: " + getMaxPartition() +
" bucketSize: " + this.buckets.length);
//throw new RuntimeException("The hash table ran out of memory.");
}
}
// --------------------------------------------------------------------------------------------
// Utility Computational Functions
// --------------------------------------------------------------------------------------------
/**
* Gets the number of partitions to be used for an initial hash-table, when no estimates are
* available.
* <p>
* The current logic makes sure that there are always between 10 and 127 partitions, and close
* to 0.1 of the number of buffers.
*
* @param numBuffers The number of buffers available.
* @return The number of partitions to use.
*/
private static final int getPartitioningFanOutNoEstimates(int numBuffers) {
return Math.max(10, Math.min(numBuffers / 10, MAX_NUM_PARTITIONS));
}
private int getMaxPartition() {
int maxPartition = 0;
for(InMemoryPartition<T> p1 : this.partitions) {
if(p1.getBlockCount() > maxPartition) {
maxPartition = p1.getBlockCount();
}
}
return maxPartition;
}
private int getMinPartition() {
int minPartition = Integer.MAX_VALUE;
for(InMemoryPartition<T> p1 : this.partitions) {
if(p1.getBlockCount() < minPartition) {
minPartition = p1.getBlockCount();
}
}
return minPartition;
}
private static final int getInitialTableSize(int numBuffers, int bufferSize, int numPartitions, int recordLenBytes) {
// ----------------------------------------------------------------------------------------
// the following observations hold:
// 1) If the records are assumed to be very large, then many buffers need to go to the partitions
// and fewer to the table
// 2) If the records are small, then comparatively many have to go to the buckets, and fewer to the
// partitions
// 3) If the bucket-table is chosen too small, we will eventually get many collisions and will grow the
// hash table, incrementally adding buffers.
// 4) If the bucket-table is chosen to be large and we actually need more buffers for the partitions, we
// cannot subtract them afterwards from the table
//
// ==> We start with a comparatively small hash-table. We aim for a 200% utilization of the bucket table
// when all the partition buffers are full. Most likely, that will cause some buckets to be re-hashed
// and grab additional buffers away from the partitions.
// NOTE: This decision may be subject to changes after conclusive experiments!
// ----------------------------------------------------------------------------------------
final long totalSize = ((long) bufferSize) * numBuffers;
final long numRecordsStorable = totalSize / (recordLenBytes + RECORD_OVERHEAD_BYTES);
final long bucketBytes = numRecordsStorable * RECORD_TABLE_BYTES;
final long numBuckets = bucketBytes / (2 * HASH_BUCKET_SIZE) + 1;
return numBuckets > Integer.MAX_VALUE ? Integer.MAX_VALUE : (int) numBuckets;
}
/**
* Assigns a partition to a bucket.
*
* @param bucket bucket index
* @param numPartitions number of partitions
* @return The hash code for the integer.
*/
private static final byte assignPartition(int bucket, byte numPartitions) {
return (byte) (bucket % numPartitions);
}
/**
* Compacts (garbage collects) partition with copy-compact strategy using compaction partition
*
* @param partition partition number
* @throws IOException
*/
private void compactPartition(int partitionNumber) throws IOException {
// stop if no garbage exists
if(this.partitions.get(partitionNumber).isCompacted()) {
return;
}
// release all segments owned by compaction partition
this.compactionMemory.clearAllMemory(availableMemory);
this.compactionMemory.allocateSegments(1);
T tempHolder = this.buildSideSerializer.createInstance();
InMemoryPartition<T> partition = this.partitions.remove(partitionNumber);
final int numPartitions = this.partitions.size() + 1; // dropped one earlier
long pointer = 0L;
int pointerOffset = 0;
int bucketOffset = 0;
final int bucketsPerSegment = this.bucketsPerSegmentMask + 1;
for (int i = 0, bucket = partitionNumber; i < this.buckets.length && bucket < this.numBuckets; i++) {
MemorySegment segment = this.buckets[i];
// go over all buckets in the segment belonging to the partition
for (int k = bucket % bucketsPerSegment; k < bucketsPerSegment && bucket < this.numBuckets; k += numPartitions, bucket += numPartitions) {
bucketOffset = k * HASH_BUCKET_SIZE;
if((int)segment.get(bucketOffset + HEADER_PARTITION_OFFSET) != partitionNumber) {
throw new IOException("Accessed wrong bucket! ");
}
int count = segment.getInt(bucketOffset + HEADER_COUNT_OFFSET);
for (int j = 0; j < NUM_ENTRIES_PER_BUCKET && j < count; j++) {
pointerOffset = bucketOffset + BUCKET_POINTER_START_OFFSET + (j * POINTER_LEN);
pointer = segment.getLong(pointerOffset);
partition.readRecordAt(pointer, tempHolder);
pointer = this.compactionMemory.appendRecord(tempHolder);
segment.putLong(pointerOffset, pointer);
}
long overflowPointer = segment.getLong(bucketOffset + HEADER_FORWARD_OFFSET);
if(overflowPointer != BUCKET_FORWARD_POINTER_NOT_SET) {
// scan overflow buckets
int current = NUM_ENTRIES_PER_BUCKET;
bucketOffset = (int) (overflowPointer & 0xffffffff);
pointerOffset = ((int) (overflowPointer & 0xffffffff)) + BUCKET_POINTER_START_OFFSET;
int overflowSegNum = (int) (overflowPointer >>> 32);
count += partition.overflowSegments[overflowSegNum].getInt(bucketOffset + HEADER_COUNT_OFFSET);
while(current < count) {
pointer = partition.overflowSegments[overflowSegNum].getLong(pointerOffset);
partition.readRecordAt(pointer, tempHolder);
pointer = this.compactionMemory.appendRecord(tempHolder);
partition.overflowSegments[overflowSegNum].putLong(pointerOffset, pointer);
current++;
if(current % NUM_ENTRIES_PER_BUCKET == 0) {
count += partition.overflowSegments[overflowSegNum].getInt(bucketOffset + HEADER_COUNT_OFFSET);
overflowPointer = partition.overflowSegments[overflowSegNum].getLong(bucketOffset + HEADER_FORWARD_OFFSET);
if(overflowPointer == BUCKET_FORWARD_POINTER_NOT_SET) {
break;
}
overflowSegNum = (int) (overflowPointer >>> 32);
bucketOffset = (int) (overflowPointer & 0xffffffff);
pointerOffset = ((int) (overflowPointer & 0xffffffff)) + BUCKET_POINTER_START_OFFSET;
} else {
pointerOffset += POINTER_LEN;
}
}
}
}
}
// swap partition with compaction partition
this.compactionMemory.setPartitionNumber(partitionNumber);
this.partitions.add(partitionNumber, compactionMemory);
this.compactionMemory = partition;
this.partitions.get(partitionNumber).overflowSegments = this.compactionMemory.overflowSegments;
this.partitions.get(partitionNumber).numOverflowSegments = this.compactionMemory.numOverflowSegments;
this.partitions.get(partitionNumber).nextOverflowBucket = this.compactionMemory.nextOverflowBucket;
this.partitions.get(partitionNumber).setCompaction(true);
this.compactionMemory.resetRecordCounter();
this.compactionMemory.setPartitionNumber(-1);
// try to allocate maximum segment count
int maxSegmentNumber = 0;
for (InMemoryPartition<T> e : this.partitions) {
if(e.getBlockCount() > maxSegmentNumber) {
maxSegmentNumber = e.getBlockCount();
}
}
this.compactionMemory.allocateSegments(maxSegmentNumber);
if(this.compactionMemory.getBlockCount() > maxSegmentNumber) {
this.compactionMemory.releaseSegments(maxSegmentNumber, availableMemory);
}
}
/**
* Compacts partition but may not reclaim all garbage
*
* @param partition partition number
* @throws IOException
*/
@SuppressWarnings("unused")
private void fastCompactPartition(int partitionNumber) throws IOException {
// stop if no garbage exists
if(this.partitions.get(partitionNumber).isCompacted()) {
return;
}
//TODO IMPLEMENT ME
return;
}
/**
* This function hashes an integer value. It is adapted from Bob Jenkins' website
* <a href="http://www.burtleburtle.net/bob/hash/integer.html">http://www.burtleburtle.net/bob/hash/integer.html</a>.
* The hash function has the <i>full avalanche</i> property, meaning that every bit of the value to be hashed
* affects every bit of the hash value.
*
* @param code The integer to be hashed.
* @return The hash code for the integer.
*/
private static final int hash(int code) {
code = (code + 0x7ed55d16) + (code << 12);
code = (code ^ 0xc761c23c) ^ (code >>> 19);
code = (code + 0x165667b1) + (code << 5);
code = (code + 0xd3a2646c) ^ (code << 9);
code = (code + 0xfd7046c5) + (code << 3);
code = (code ^ 0xb55a4f09) ^ (code >>> 16);
return code >= 0 ? code : -(code + 1);
}
/**
* Iterator that traverses the whole hash table once
*
* If entries are inserted during iteration they may be overlooked by the iterator
*/
public class EntryIterator implements MutableObjectIterator<T> {
private CompactingHashTable<T> table;
private ArrayList<T> cache; // holds full bucket including its overflow buckets
private int currentBucketIndex = 0;
private int currentSegmentIndex = 0;
private int currentBucketOffset = 0;
private int bucketsPerSegment;
private boolean done;
private EntryIterator(CompactingHashTable<T> compactingHashTable) {
this.table = compactingHashTable;
this.cache = new ArrayList<T>(64);
this.done = false;
this.bucketsPerSegment = table.bucketsPerSegmentMask + 1;
}
@Override
public T next(T reuse) throws IOException {
if(done) {
return null;
} else if(!cache.isEmpty()) {
reuse = cache.remove(cache.size()-1);
return reuse;
} else {
while(!done && cache.isEmpty()) {
done = !fillCache();
}
if(!done) {
reuse = cache.remove(cache.size()-1);
return reuse;
} else {
return null;
}
}
}
private boolean fillCache() throws IOException {
if(currentBucketIndex >= table.numBuckets) {
return false;
}
MemorySegment bucket = table.buckets[currentSegmentIndex];
// get the basic characteristics of the bucket
final int partitionNumber = bucket.get(currentBucketOffset + HEADER_PARTITION_OFFSET);
final InMemoryPartition<T> partition = table.partitions.get(partitionNumber);
final MemorySegment[] overflowSegments = partition.overflowSegments;
int countInSegment = bucket.getInt(currentBucketOffset + HEADER_COUNT_OFFSET);
int numInSegment = 0;
int posInSegment = currentBucketOffset + BUCKET_POINTER_START_OFFSET;
int bucketOffset = currentBucketOffset;
// loop over all segments that are involved in the bucket (original bucket plus overflow buckets)
while (true) {
while (numInSegment < countInSegment) {
long pointer = bucket.getLong(posInSegment);
posInSegment += POINTER_LEN;
numInSegment++;
T target = table.buildSideSerializer.createInstance();
try {
partition.readRecordAt(pointer, target);
cache.add(target);
} catch (IOException e) {
throw new RuntimeException("Error deserializing record from the hashtable: " + e.getMessage(), e);
}
}
// this segment is done. check if there is another chained bucket
final long forwardPointer = bucket.getLong(bucketOffset + HEADER_FORWARD_OFFSET);
if (forwardPointer == BUCKET_FORWARD_POINTER_NOT_SET) {
break;
}
final int overflowSegNum = (int) (forwardPointer >>> 32);
bucket = overflowSegments[overflowSegNum];
bucketOffset = (int)(forwardPointer & 0xffffffff);
countInSegment = bucket.getInt(bucketOffset + HEADER_COUNT_OFFSET);
posInSegment = bucketOffset + BUCKET_POINTER_START_OFFSET;
numInSegment = 0;
}
currentBucketIndex++;
if(currentBucketIndex % bucketsPerSegment == 0) {
currentSegmentIndex++;
currentBucketOffset = 0;
} else {
currentBucketOffset += HASH_BUCKET_SIZE;
}
return true;
}
}
public final class HashTableProber<PT> extends AbstractHashTableProber<PT, T>{
private InMemoryPartition<T> partition;
private MemorySegment bucket;
private int pointerOffsetInBucket;
private HashTableProber(TypeComparator<PT> probeTypeComparator, TypePairComparator<PT, T> pairComparator)
{
super(probeTypeComparator, pairComparator);
}
public boolean getMatchFor(PT probeSideRecord, T targetForMatch) {
final int searchHashCode = hash(this.probeTypeComparator.hash(probeSideRecord));
final int posHashCode = searchHashCode % numBuckets;
// get the bucket for the given hash code
MemorySegment bucket = buckets[posHashCode >> bucketsPerSegmentBits];
int bucketInSegmentOffset = (posHashCode & bucketsPerSegmentMask) << NUM_INTRA_BUCKET_BITS;
// get the basic characteristics of the bucket
final int partitionNumber = bucket.get(bucketInSegmentOffset + HEADER_PARTITION_OFFSET);
final InMemoryPartition<T> partition = partitions.get(partitionNumber);
final MemorySegment[] overflowSegments = partition.overflowSegments;
this.pairComparator.setReference(probeSideRecord);
int countInSegment = bucket.getInt(bucketInSegmentOffset + HEADER_COUNT_OFFSET);
int numInSegment = 0;
int posInSegment = bucketInSegmentOffset + BUCKET_HEADER_LENGTH;
// loop over all segments that are involved in the bucket (original bucket plus overflow buckets)
while (true) {
while (numInSegment < countInSegment) {
final int thisCode = bucket.getInt(posInSegment);
posInSegment += HASH_CODE_LEN;
// check if the hash code matches
if (thisCode == searchHashCode) {
// get the pointer to the pair
final int pointerOffset = bucketInSegmentOffset + BUCKET_POINTER_START_OFFSET + (numInSegment * POINTER_LEN);
final long pointer = bucket.getLong(pointerOffset);
numInSegment++;
// deserialize the key to check whether it is really equal, or whether we had only a hash collision
try {
partition.readRecordAt(pointer, targetForMatch);
if (this.pairComparator.equalToReference(targetForMatch)) {
this.partition = partition;
this.bucket = bucket;
this.pointerOffsetInBucket = pointerOffset;
return true;
}
}
catch (IOException e) {
throw new RuntimeException("Error deserializing record from the hashtable: " + e.getMessage(), e);
}
}
else {
numInSegment++;
}
}
// this segment is done. check if there is another chained bucket
final long forwardPointer = bucket.getLong(bucketInSegmentOffset + HEADER_FORWARD_OFFSET);
if (forwardPointer == BUCKET_FORWARD_POINTER_NOT_SET) {
return false;
}
final int overflowSegNum = (int) (forwardPointer >>> 32);
bucket = overflowSegments[overflowSegNum];
bucketInSegmentOffset = (int) (forwardPointer & 0xffffffff);
countInSegment = bucket.getInt(bucketInSegmentOffset + HEADER_COUNT_OFFSET);
posInSegment = bucketInSegmentOffset + BUCKET_HEADER_LENGTH;
numInSegment = 0;
}
}
public void updateMatch(T record) throws IOException {
long newPointer = this.partition.appendRecord(record);
this.bucket.putLong(this.pointerOffsetInBucket, newPointer);
this.partition.setCompaction(false); //FIXME Do we really create garbage here?
}
}
}