1 package org.apache.lucene.search;
4 * Licensed to the Apache Software Foundation (ASF) under one or more
5 * contributor license agreements. See the NOTICE file distributed with
6 * this work for additional information regarding copyright ownership.
7 * The ASF licenses this file to You under the Apache License, Version 2.0
8 * (the "License"); you may not use this file except in compliance with
9 * the License. You may obtain a copy of the License at
11 * http://www.apache.org/licenses/LICENSE-2.0
13 * Unless required by applicable law or agreed to in writing, software
14 * distributed under the License is distributed on an "AS IS" BASIS,
15 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
16 * See the License for the specific language governing permissions and
17 * limitations under the License.
20 import java.io.Serializable;
21 import java.io.IOException;
23 import java.util.WeakHashMap;
25 import org.apache.lucene.index.IndexReader;
26 import org.apache.lucene.util.FixedBitSet;
29 * Wraps another filter's result and caches it. The purpose is to allow
30 * filters to simply filter, and then wrap with this class
33 * <p><b>NOTE</b>: if you wrap this filter as a query (eg,
34 * using ConstantScoreQuery), you'll likely want to enforce
35 * deletions (using either {@link DeletesMode#RECACHE} or
36 * {@link DeletesMode#DYNAMIC}).
38 public class CachingWrapperFilter extends Filter {
42 * Expert: Specifies how new deletions against a reopened
43 * reader should be handled.
45 * <p>The default is IGNORE, which means the cache entry
46 * will be re-used for a given segment, even when that
47 * segment has been reopened due to changes in deletions.
48 * This is a big performance gain, especially with
49 * near-real-timer readers, since you don't hit a cache
50 * miss on every reopened reader for prior segments.</p>
52 * <p>However, in some cases this can cause invalid query
53 * results, allowing deleted documents to be returned.
54 * This only happens if the main query does not rule out
55 * deleted documents on its own, such as a toplevel
56 * ConstantScoreQuery. To fix this, use RECACHE to
57 * re-create the cached filter (at a higher per-reopen
58 * cost, but at faster subsequent search performance), or
59 * use DYNAMIC to dynamically intersect deleted docs (fast
60 * reopen time but some hit to search performance).</p>
62 public static enum DeletesMode {IGNORE, RECACHE, DYNAMIC};
64 protected final FilterCache<DocIdSet> cache;
66 static abstract class FilterCache<T> implements Serializable {
69 * A transient Filter cache (package private because of test)
71 // NOTE: not final so that we can dynamically re-init
73 transient Map<Object,T> cache;
75 private final DeletesMode deletesMode;
77 public FilterCache(DeletesMode deletesMode) {
78 this.deletesMode = deletesMode;
81 public synchronized T get(IndexReader reader, Object coreKey, Object delCoreKey) throws IOException {
85 cache = new WeakHashMap<Object,T>();
88 if (deletesMode == DeletesMode.IGNORE) {
90 value = cache.get(coreKey);
91 } else if (deletesMode == DeletesMode.RECACHE) {
92 // key on deletes, if any, else core
93 value = cache.get(delCoreKey);
95 assert deletesMode == DeletesMode.DYNAMIC;
97 // first try for exact match
98 value = cache.get(delCoreKey);
101 // now for core match, but dynamically AND NOT
103 value = cache.get(coreKey);
104 if (value != null && reader.hasDeletions()) {
105 value = mergeDeletes(reader, value);
113 protected abstract T mergeDeletes(IndexReader reader, T value);
115 public synchronized void put(Object coreKey, Object delCoreKey, T value) {
116 if (deletesMode == DeletesMode.IGNORE) {
117 cache.put(coreKey, value);
118 } else if (deletesMode == DeletesMode.RECACHE) {
119 cache.put(delCoreKey, value);
121 cache.put(coreKey, value);
122 cache.put(delCoreKey, value);
128 * New deletes are ignored by default, which gives higher
129 * cache hit rate on reopened readers. Most of the time
130 * this is safe, because the filter will be AND'd with a
131 * Query that fully enforces deletions. If instead you
132 * need this filter to always enforce deletions, pass
133 * either {@link DeletesMode#RECACHE} or {@link
134 * DeletesMode#DYNAMIC}.
135 * @param filter Filter to cache results of
137 public CachingWrapperFilter(Filter filter) {
138 this(filter, DeletesMode.IGNORE);
142 * Expert: by default, the cached filter will be shared
143 * across reopened segments that only had changes to their
146 * @param filter Filter to cache results of
147 * @param deletesMode See {@link DeletesMode}
149 public CachingWrapperFilter(Filter filter, DeletesMode deletesMode) {
150 this.filter = filter;
151 cache = new FilterCache<DocIdSet>(deletesMode) {
153 public DocIdSet mergeDeletes(final IndexReader r, final DocIdSet docIdSet) {
154 return new FilteredDocIdSet(docIdSet) {
156 protected boolean match(int docID) {
157 return !r.isDeleted(docID);
164 /** Provide the DocIdSet to be cached, using the DocIdSet provided
165 * by the wrapped Filter.
166 * <p>This implementation returns the given {@link DocIdSet}, if {@link DocIdSet#isCacheable}
167 * returns <code>true</code>, else it copies the {@link DocIdSetIterator} into
168 * an {@link FixedBitSet}.
170 protected DocIdSet docIdSetToCache(DocIdSet docIdSet, IndexReader reader) throws IOException {
171 if (docIdSet == null) {
172 // this is better than returning null, as the nonnull result can be cached
173 return DocIdSet.EMPTY_DOCIDSET;
174 } else if (docIdSet.isCacheable()) {
177 final DocIdSetIterator it = docIdSet.iterator();
178 // null is allowed to be returned by iterator(),
179 // in this case we wrap with the empty set,
180 // which is cacheable.
182 return DocIdSet.EMPTY_DOCIDSET;
184 final FixedBitSet bits = new FixedBitSet(reader.maxDoc());
192 int hitCount, missCount;
195 public DocIdSet getDocIdSet(IndexReader reader) throws IOException {
197 final Object coreKey = reader.getCoreCacheKey();
198 final Object delCoreKey = reader.hasDeletions() ? reader.getDeletesCacheKey() : coreKey;
200 DocIdSet docIdSet = cache.get(reader, coreKey, delCoreKey);
201 if (docIdSet != null) {
209 docIdSet = docIdSetToCache(filter.getDocIdSet(reader), reader);
211 if (docIdSet != null) {
212 cache.put(coreKey, delCoreKey, docIdSet);
219 public String toString() {
220 return "CachingWrapperFilter("+filter+")";
224 public boolean equals(Object o) {
225 if (!(o instanceof CachingWrapperFilter)) return false;
226 return this.filter.equals(((CachingWrapperFilter)o).filter);
230 public int hashCode() {
231 return filter.hashCode() ^ 0x1117BF25;