org.elasticsearch.search.aggregations.bucket.AbstractHash Maven / Gradle / Ivy
/*
* Licensed to Elasticsearch under one or more contributor
* license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright
* ownership. Elasticsearch licenses this file to you 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 org.elasticsearch.search.aggregations.bucket;
import com.google.common.base.Preconditions;
import org.elasticsearch.cache.recycler.PageCacheRecycler;
import org.elasticsearch.common.lease.Releasable;
import org.elasticsearch.common.lease.Releasables;
import org.elasticsearch.common.util.BigArrays;
import org.elasticsearch.common.util.LongArray;
/**
* Base implementation for {@link BytesRefHash} and {@link LongHash}.
*/
// IDs are internally stored as id + 1 so that 0 encodes for an empty slot
abstract class AbstractHash implements Releasable {
// Open addressing typically requires having smaller load factors compared to linked lists because
// collisions may result into worse lookup performance.
static final float DEFAULT_MAX_LOAD_FACTOR = 0.6f;
final float maxLoadFactor;
long size, maxSize;
LongArray ids;
long mask;
AbstractHash(long capacity, float maxLoadFactor, PageCacheRecycler recycler) {
Preconditions.checkArgument(capacity >= 0, "capacity must be >= 0");
Preconditions.checkArgument(maxLoadFactor > 0 && maxLoadFactor < 1, "maxLoadFactor must be > 0 and < 1");
this.maxLoadFactor = maxLoadFactor;
long buckets = 1L + (long) (capacity / maxLoadFactor);
buckets = Math.max(1, Long.highestOneBit(buckets - 1) << 1); // next power of two
assert buckets == Long.highestOneBit(buckets);
maxSize = (long) (buckets * maxLoadFactor);
assert maxSize >= capacity;
size = 0;
ids = BigArrays.newLongArray(buckets, recycler, true);
mask = buckets - 1;
}
/**
* Return the number of allocated slots to store this hash table.
*/
public long capacity() {
return ids.size();
}
/**
* Return the number of longs in this hash table.
*/
public long size() {
return size;
}
static long slot(long hash, long mask) {
return hash & mask;
}
static long nextSlot(long curSlot, long mask) {
return (curSlot + 1) & mask; // linear probing
}
/**
* Get the id associated with key at 0 <e; index <e; capacity()
or -1 if this slot is unused.
*/
public long id(long index) {
return ids.get(index) - 1;
}
protected final long id(long index, long id) {
return ids.set(index, id + 1) - 1;
}
/** Resize keys to the given capacity. */
protected void resizeKeys(long capacity) {}
/** Remove key at the given index and */
protected abstract void removeAndAdd(long index, long id);
protected final void grow() {
// The difference of this implementation of grow() compared to standard hash tables is that we are growing in-place, which makes
// the re-mapping of keys to slots a bit more tricky.
assert size == maxSize;
final long prevSize = size;
final long buckets = capacity();
// Resize arrays
final long newBuckets = buckets << 1;
assert newBuckets == Long.highestOneBit(newBuckets) : newBuckets; // power of 2
resizeKeys(newBuckets);
ids = BigArrays.resize(ids, newBuckets);
mask = newBuckets - 1;
// First let's remap in-place: most data will be put in its final position directly
for (long i = 0; i < buckets; ++i) {
final long id = id(i, -1);
if (id != -1) {
removeAndAdd(i, id);
}
}
// The only entries which have not been put in their final position in the previous loop are those that were stored in a slot that
// is < slot(key, mask). This only happens when slot(key, mask) returned a slot that was close to the end of the array and colision
// resolution has put it back in the first slots. This time, collision resolution will have put them at the beginning of the newly
// allocated slots. Let's re-add them to make sure they are in the right slot. This 2nd loop will typically exit very early.
for (long i = buckets; i < newBuckets; ++i) {
final long id = id(i, -1);
if (id != -1) {
removeAndAdd(i, id); // add it back
} else {
break;
}
}
assert size == prevSize;
maxSize = (long) (newBuckets * maxLoadFactor);
assert size < maxSize;
}
@Override
public boolean release() {
Releasables.release(ids);
return true;
}
}