1 package org.apache.lucene.index;
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.IOException;
21 import java.util.ArrayList;
22 import java.util.Collection;
23 import java.util.List;
26 /** <p>This class implements a {@link MergePolicy} that tries
27 * to merge segments into levels of exponentially
28 * increasing size, where each level has fewer segments than
29 * the value of the merge factor. Whenever extra segments
30 * (beyond the merge factor upper bound) are encountered,
31 * all segments within the level are merged. You can get or
32 * set the merge factor using {@link #getMergeFactor()} and
33 * {@link #setMergeFactor(int)} respectively.</p>
35 * <p>This class is abstract and requires a subclass to
36 * define the {@link #size} method which specifies how a
37 * segment's size is determined. {@link LogDocMergePolicy}
38 * is one subclass that measures size by document count in
39 * the segment. {@link LogByteSizeMergePolicy} is another
40 * subclass that measures size as the total byte size of the
41 * file(s) for the segment.</p>
44 public abstract class LogMergePolicy extends MergePolicy {
46 /** Defines the allowed range of log(size) for each
47 * level. A level is computed by taking the max segment
48 * log size, minus LEVEL_LOG_SPAN, and finding all
49 * segments falling within that range. */
50 public static final double LEVEL_LOG_SPAN = 0.75;
52 /** Default merge factor, which is how many segments are
54 public static final int DEFAULT_MERGE_FACTOR = 10;
56 /** Default maximum segment size. A segment of this size
57 * or larger will never be merged. @see setMaxMergeDocs */
58 public static final int DEFAULT_MAX_MERGE_DOCS = Integer.MAX_VALUE;
60 /** Default noCFSRatio. If a merge's size is >= 10% of
61 * the index, then we disable compound file for it.
62 * @see #setNoCFSRatio */
63 public static final double DEFAULT_NO_CFS_RATIO = 0.1;
65 protected int mergeFactor = DEFAULT_MERGE_FACTOR;
67 protected long minMergeSize;
68 protected long maxMergeSize;
69 // Although the core MPs set it explicitly, we must default in case someone
70 // out there wrote his own LMP ...
71 protected long maxMergeSizeForForcedMerge = Long.MAX_VALUE;
72 protected int maxMergeDocs = DEFAULT_MAX_MERGE_DOCS;
74 protected double noCFSRatio = DEFAULT_NO_CFS_RATIO;
76 protected boolean calibrateSizeByDeletes = true;
78 protected boolean useCompoundFile = true;
80 public LogMergePolicy() {
84 protected boolean verbose() {
85 IndexWriter w = writer.get();
86 return w != null && w.verbose();
89 /** @see #setNoCFSRatio */
90 public double getNoCFSRatio() {
94 /** If a merged segment will be more than this percentage
95 * of the total size of the index, leave the segment as
96 * non-compound file even if compound file is enabled.
97 * Set to 1.0 to always use CFS regardless of merge
99 public void setNoCFSRatio(double noCFSRatio) {
100 if (noCFSRatio < 0.0 || noCFSRatio > 1.0) {
101 throw new IllegalArgumentException("noCFSRatio must be 0.0 to 1.0 inclusive; got " + noCFSRatio);
103 this.noCFSRatio = noCFSRatio;
106 protected void message(String message) {
108 writer.get().message("LMP: " + message);
111 /** <p>Returns the number of segments that are merged at
112 * once and also controls the total number of segments
113 * allowed to accumulate in the index.</p> */
114 public int getMergeFactor() {
118 /** Determines how often segment indices are merged by
119 * addDocument(). With smaller values, less RAM is used
120 * while indexing, and searches are
121 * faster, but indexing speed is slower. With larger
122 * values, more RAM is used during indexing, and while
123 * searches is slower, indexing is
124 * faster. Thus larger values (> 10) are best for batch
125 * index creation, and smaller values (< 10) for indices
126 * that are interactively maintained. */
127 public void setMergeFactor(int mergeFactor) {
129 throw new IllegalArgumentException("mergeFactor cannot be less than 2");
130 this.mergeFactor = mergeFactor;
135 public boolean useCompoundFile(SegmentInfos infos, SegmentInfo mergedInfo) throws IOException {
138 if (!useCompoundFile) {
140 } else if (noCFSRatio == 1.0) {
144 for (SegmentInfo info : infos)
145 totalSize += size(info);
147 doCFS = size(mergedInfo) <= noCFSRatio * totalSize;
152 /** Sets whether compound file format should be used for
153 * newly flushed and newly merged segments. */
154 public void setUseCompoundFile(boolean useCompoundFile) {
155 this.useCompoundFile = useCompoundFile;
158 /** Returns true if newly flushed and newly merge segments
159 * are written in compound file format. @see
160 * #setUseCompoundFile */
161 public boolean getUseCompoundFile() {
162 return useCompoundFile;
165 /** Sets whether the segment size should be calibrated by
166 * the number of deletes when choosing segments for merge. */
167 public void setCalibrateSizeByDeletes(boolean calibrateSizeByDeletes) {
168 this.calibrateSizeByDeletes = calibrateSizeByDeletes;
171 /** Returns true if the segment size should be calibrated
172 * by the number of deletes when choosing segments for merge. */
173 public boolean getCalibrateSizeByDeletes() {
174 return calibrateSizeByDeletes;
178 public void close() {}
180 abstract protected long size(SegmentInfo info) throws IOException;
182 protected long sizeDocs(SegmentInfo info) throws IOException {
183 if (calibrateSizeByDeletes) {
184 int delCount = writer.get().numDeletedDocs(info);
185 assert delCount <= info.docCount;
186 return (info.docCount - (long)delCount);
188 return info.docCount;
192 protected long sizeBytes(SegmentInfo info) throws IOException {
193 long byteSize = info.sizeInBytes(true);
194 if (calibrateSizeByDeletes) {
195 int delCount = writer.get().numDeletedDocs(info);
196 double delRatio = (info.docCount <= 0 ? 0.0f : ((float)delCount / (float)info.docCount));
197 assert delRatio <= 1.0;
198 return (info.docCount <= 0 ? byteSize : (long)(byteSize * (1.0 - delRatio)));
204 protected boolean isMerged(SegmentInfos infos, int maxNumSegments, Map<SegmentInfo,Boolean> segmentsToMerge) throws IOException {
205 final int numSegments = infos.size();
207 SegmentInfo mergeInfo = null;
208 boolean segmentIsOriginal = false;
209 for(int i=0;i<numSegments && numToMerge <= maxNumSegments;i++) {
210 final SegmentInfo info = infos.info(i);
211 final Boolean isOriginal = segmentsToMerge.get(info);
212 if (isOriginal != null) {
213 segmentIsOriginal = isOriginal;
219 return numToMerge <= maxNumSegments &&
220 (numToMerge != 1 || !segmentIsOriginal || isMerged(mergeInfo));
223 /** Returns true if this single info is already fully merged (has no
224 * pending norms or deletes, is in the same dir as the
225 * writer, and matches the current compound file setting */
226 protected boolean isMerged(SegmentInfo info)
228 IndexWriter w = writer.get();
230 boolean hasDeletions = w.numDeletedDocs(info) > 0;
231 return !hasDeletions &&
232 !info.hasSeparateNorms() &&
233 info.dir == w.getDirectory() &&
234 (info.getUseCompoundFile() == useCompoundFile || noCFSRatio < 1.0);
238 * Returns the merges necessary to merge the index, taking the max merge
239 * size or max merge docs into consideration. This method attempts to respect
240 * the {@code maxNumSegments} parameter, however it might be, due to size
241 * constraints, that more than that number of segments will remain in the
242 * index. Also, this method does not guarantee that exactly {@code
243 * maxNumSegments} will remain, but <= that number.
245 private MergeSpecification findForcedMergesSizeLimit(
246 SegmentInfos infos, int maxNumSegments, int last) throws IOException {
247 MergeSpecification spec = new MergeSpecification();
248 final List<SegmentInfo> segments = infos.asList();
250 int start = last - 1;
252 SegmentInfo info = infos.info(start);
253 if (size(info) > maxMergeSizeForForcedMerge || sizeDocs(info) > maxMergeDocs) {
255 message("findForcedMergesSizeLimit: skip segment=" + info + ": size is > maxMergeSize (" + maxMergeSizeForForcedMerge + ") or sizeDocs is > maxMergeDocs (" + maxMergeDocs + ")");
257 // need to skip that segment + add a merge for the 'right' segments,
258 // unless there is only 1 which is merged.
259 if (last - start - 1 > 1 || (start != last - 1 && !isMerged(infos.info(start + 1)))) {
260 // there is more than 1 segment to the right of
261 // this one, or a mergeable single segment.
262 spec.add(new OneMerge(segments.subList(start + 1, last)));
265 } else if (last - start == mergeFactor) {
266 // mergeFactor eligible segments were found, add them as a merge.
267 spec.add(new OneMerge(segments.subList(start, last)));
273 // Add any left-over segments, unless there is just 1
274 // already fully merged
275 if (last > 0 && (++start + 1 < last || !isMerged(infos.info(start)))) {
276 spec.add(new OneMerge(segments.subList(start, last)));
279 return spec.merges.size() == 0 ? null : spec;
283 * Returns the merges necessary to forceMerge the index. This method constraints
284 * the returned merges only by the {@code maxNumSegments} parameter, and
285 * guaranteed that exactly that number of segments will remain in the index.
287 private MergeSpecification findForcedMergesMaxNumSegments(SegmentInfos infos, int maxNumSegments, int last) throws IOException {
288 MergeSpecification spec = new MergeSpecification();
289 final List<SegmentInfo> segments = infos.asList();
291 // First, enroll all "full" merges (size
292 // mergeFactor) to potentially be run concurrently:
293 while (last - maxNumSegments + 1 >= mergeFactor) {
294 spec.add(new OneMerge(segments.subList(last - mergeFactor, last)));
298 // Only if there are no full merges pending do we
299 // add a final partial (< mergeFactor segments) merge:
300 if (0 == spec.merges.size()) {
301 if (maxNumSegments == 1) {
303 // Since we must merge down to 1 segment, the
305 if (last > 1 || !isMerged(infos.info(0))) {
306 spec.add(new OneMerge(segments.subList(0, last)));
308 } else if (last > maxNumSegments) {
310 // Take care to pick a partial merge that is
311 // least cost, but does not make the index too
312 // lopsided. If we always just picked the
313 // partial tail then we could produce a highly
314 // lopsided index over time:
316 // We must merge this many segments to leave
317 // maxNumSegments in the index (from when
318 // forceMerge was first kicked off):
319 final int finalMergeSize = last - maxNumSegments + 1;
321 // Consider all possible starting points:
325 for(int i=0;i<last-finalMergeSize+1;i++) {
327 for(int j=0;j<finalMergeSize;j++)
328 sumSize += size(infos.info(j+i));
329 if (i == 0 || (sumSize < 2*size(infos.info(i-1)) && sumSize < bestSize)) {
335 spec.add(new OneMerge(segments.subList(bestStart, bestStart + finalMergeSize)));
338 return spec.merges.size() == 0 ? null : spec;
341 /** Returns the merges necessary to merge the index down
342 * to a specified number of segments.
343 * This respects the {@link #maxMergeSizeForForcedMerge} setting.
344 * By default, and assuming {@code maxNumSegments=1}, only
345 * one segment will be left in the index, where that segment
346 * has no deletions pending nor separate norms, and it is in
347 * compound file format if the current useCompoundFile
348 * setting is true. This method returns multiple merges
349 * (mergeFactor at a time) so the {@link MergeScheduler}
350 * in use may make use of concurrency. */
352 public MergeSpecification findForcedMerges(SegmentInfos infos,
353 int maxNumSegments, Map<SegmentInfo,Boolean> segmentsToMerge) throws IOException {
355 assert maxNumSegments > 0;
357 message("findForcedMerges: maxNumSegs=" + maxNumSegments + " segsToMerge="+ segmentsToMerge);
360 // If the segments are already merged (e.g. there's only 1 segment), or
361 // there are <maxNumSegements:.
362 if (isMerged(infos, maxNumSegments, segmentsToMerge)) {
364 message("already merged; skip");
369 // Find the newest (rightmost) segment that needs to
370 // be merged (other segments may have been flushed
371 // since merging started):
372 int last = infos.size();
374 final SegmentInfo info = infos.info(--last);
375 if (segmentsToMerge.get(info) != null) {
381 if (last == 0) return null;
383 // There is only one segment already, and it is merged
384 if (maxNumSegments == 1 && last == 1 && isMerged(infos.info(0))) {
386 message("already 1 seg; skip");
391 // Check if there are any segments above the threshold
392 boolean anyTooLarge = false;
393 for (int i = 0; i < last; i++) {
394 SegmentInfo info = infos.info(i);
395 if (size(info) > maxMergeSizeForForcedMerge || sizeDocs(info) > maxMergeDocs) {
402 return findForcedMergesSizeLimit(infos, maxNumSegments, last);
404 return findForcedMergesMaxNumSegments(infos, maxNumSegments, last);
409 * Finds merges necessary to force-merge all deletes from the
410 * index. We simply merge adjacent segments that have
411 * deletes, up to mergeFactor at a time.
414 public MergeSpecification findForcedDeletesMerges(SegmentInfos segmentInfos)
415 throws CorruptIndexException, IOException {
416 final List<SegmentInfo> segments = segmentInfos.asList();
417 final int numSegments = segments.size();
420 message("findForcedDeleteMerges: " + numSegments + " segments");
422 MergeSpecification spec = new MergeSpecification();
423 int firstSegmentWithDeletions = -1;
424 IndexWriter w = writer.get();
426 for(int i=0;i<numSegments;i++) {
427 final SegmentInfo info = segmentInfos.info(i);
428 int delCount = w.numDeletedDocs(info);
431 message(" segment " + info.name + " has deletions");
432 if (firstSegmentWithDeletions == -1)
433 firstSegmentWithDeletions = i;
434 else if (i - firstSegmentWithDeletions == mergeFactor) {
435 // We've seen mergeFactor segments in a row with
436 // deletions, so force a merge now:
438 message(" add merge " + firstSegmentWithDeletions + " to " + (i-1) + " inclusive");
439 spec.add(new OneMerge(segments.subList(firstSegmentWithDeletions, i)));
440 firstSegmentWithDeletions = i;
442 } else if (firstSegmentWithDeletions != -1) {
443 // End of a sequence of segments with deletions, so,
444 // merge those past segments even if it's fewer than
445 // mergeFactor segments
447 message(" add merge " + firstSegmentWithDeletions + " to " + (i-1) + " inclusive");
448 spec.add(new OneMerge(segments.subList(firstSegmentWithDeletions, i)));
449 firstSegmentWithDeletions = -1;
453 if (firstSegmentWithDeletions != -1) {
455 message(" add merge " + firstSegmentWithDeletions + " to " + (numSegments-1) + " inclusive");
456 spec.add(new OneMerge(segments.subList(firstSegmentWithDeletions, numSegments)));
462 private static class SegmentInfoAndLevel implements Comparable<SegmentInfoAndLevel> {
467 public SegmentInfoAndLevel(SegmentInfo info, float level, int index) {
473 // Sorts largest to smallest
474 public int compareTo(SegmentInfoAndLevel other) {
475 if (level < other.level)
477 else if (level > other.level)
484 /** Checks if any merges are now necessary and returns a
485 * {@link MergePolicy.MergeSpecification} if so. A merge
486 * is necessary when there are more than {@link
487 * #setMergeFactor} segments at a given level. When
488 * multiple levels have too many segments, this method
489 * will return multiple merges, allowing the {@link
490 * MergeScheduler} to use concurrency. */
492 public MergeSpecification findMerges(SegmentInfos infos) throws IOException {
494 final int numSegments = infos.size();
496 message("findMerges: " + numSegments + " segments");
498 // Compute levels, which is just log (base mergeFactor)
499 // of the size of each segment
500 final List<SegmentInfoAndLevel> levels = new ArrayList<SegmentInfoAndLevel>();
501 final float norm = (float) Math.log(mergeFactor);
503 final Collection<SegmentInfo> mergingSegments = writer.get().getMergingSegments();
505 for(int i=0;i<numSegments;i++) {
506 final SegmentInfo info = infos.info(i);
507 long size = size(info);
509 // Floor tiny segments
514 final SegmentInfoAndLevel infoLevel = new SegmentInfoAndLevel(info, (float) Math.log(size)/norm, i);
515 levels.add(infoLevel);
518 final long segBytes = sizeBytes(info);
519 String extra = mergingSegments.contains(info) ? " [merging]" : "";
520 if (size >= maxMergeSize) {
521 extra += " [skip: too large]";
523 message("seg=" + writer.get().segString(info) + " level=" + infoLevel.level + " size=" + String.format("%.3f MB", segBytes/1024/1024.) + extra);
527 final float levelFloor;
528 if (minMergeSize <= 0)
529 levelFloor = (float) 0.0;
531 levelFloor = (float) (Math.log(minMergeSize)/norm);
533 // Now, we quantize the log values into levels. The
534 // first level is any segment whose log size is within
535 // LEVEL_LOG_SPAN of the max size, or, who has such as
536 // segment "to the right". Then, we find the max of all
537 // other segments and use that to define the next level
540 MergeSpecification spec = null;
542 final int numMergeableSegments = levels.size();
545 while(start < numMergeableSegments) {
547 // Find max level of all segments not already
549 float maxLevel = levels.get(start).level;
550 for(int i=1+start;i<numMergeableSegments;i++) {
551 final float level = levels.get(i).level;
552 if (level > maxLevel)
556 // Now search backwards for the rightmost segment that
557 // falls into this level:
559 if (maxLevel <= levelFloor)
560 // All remaining segments fall into the min level
563 levelBottom = (float) (maxLevel - LEVEL_LOG_SPAN);
565 // Force a boundary at the level floor
566 if (levelBottom < levelFloor && maxLevel >= levelFloor)
567 levelBottom = levelFloor;
570 int upto = numMergeableSegments-1;
571 while(upto >= start) {
572 if (levels.get(upto).level >= levelBottom) {
578 message(" level " + levelBottom + " to " + maxLevel + ": " + (1+upto-start) + " segments");
580 // Finally, record all merges that are viable at this level:
581 int end = start + mergeFactor;
582 while(end <= 1+upto) {
583 boolean anyTooLarge = false;
584 boolean anyMerging = false;
585 for(int i=start;i<end;i++) {
586 final SegmentInfo info = levels.get(i).info;
587 anyTooLarge |= (size(info) >= maxMergeSize || sizeDocs(info) >= maxMergeDocs);
588 if (mergingSegments.contains(info)) {
596 } else if (!anyTooLarge) {
598 spec = new MergeSpecification();
599 final List<SegmentInfo> mergeInfos = new ArrayList<SegmentInfo>();
600 for(int i=start;i<end;i++) {
601 mergeInfos.add(levels.get(i).info);
602 assert infos.contains(levels.get(i).info);
605 message(" add merge=" + writer.get().segString(mergeInfos) + " start=" + start + " end=" + end);
607 spec.add(new OneMerge(mergeInfos));
608 } else if (verbose()) {
609 message(" " + start + " to " + end + ": contains segment over maxMergeSize or maxMergeDocs; skipping");
613 end = start + mergeFactor;
622 /** <p>Determines the largest segment (measured by
623 * document count) that may be merged with other segments.
624 * Small values (e.g., less than 10,000) are best for
625 * interactive indexing, as this limits the length of
626 * pauses while indexing to a few seconds. Larger values
627 * are best for batched indexing and speedier
630 * <p>The default value is {@link Integer#MAX_VALUE}.</p>
632 * <p>The default merge policy ({@link
633 * LogByteSizeMergePolicy}) also allows you to set this
634 * limit by net size (in MB) of the segment, using {@link
635 * LogByteSizeMergePolicy#setMaxMergeMB}.</p>
637 public void setMaxMergeDocs(int maxMergeDocs) {
638 this.maxMergeDocs = maxMergeDocs;
641 /** Returns the largest segment (measured by document
642 * count) that may be merged with other segments.
643 * @see #setMaxMergeDocs */
644 public int getMaxMergeDocs() {
649 public String toString() {
650 StringBuilder sb = new StringBuilder("[" + getClass().getSimpleName() + ": ");
651 sb.append("minMergeSize=").append(minMergeSize).append(", ");
652 sb.append("mergeFactor=").append(mergeFactor).append(", ");
653 sb.append("maxMergeSize=").append(maxMergeSize).append(", ");
654 sb.append("maxMergeSizeForForcedMerge=").append(maxMergeSizeForForcedMerge).append(", ");
655 sb.append("calibrateSizeByDeletes=").append(calibrateSizeByDeletes).append(", ");
656 sb.append("maxMergeDocs=").append(maxMergeDocs).append(", ");
657 sb.append("useCompoundFile=").append(useCompoundFile).append(", ");
658 sb.append("noCFSRatio=").append(noCFSRatio);
660 return sb.toString();