1 /**
2 * Copyright 2010 The Apache Software Foundation
3 *
4 * Licensed to the Apache Software Foundation (ASF) under one
5 * or more contributor license agreements. See the NOTICE file
6 * distributed with this work for additional information
7 * regarding copyright ownership. The ASF licenses this file
8 * to you under the Apache License, Version 2.0 (the
9 * "License"); you may not use this file except in compliance
10 * with the License. You may obtain a copy of the License at
11 *
12 * http://www.apache.org/licenses/LICENSE-2.0
13 *
14 * Unless required by applicable law or agreed to in writing, software
15 * distributed under the License is distributed on an "AS IS" BASIS,
16 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17 * See the License for the specific language governing permissions and
18 * limitations under the License.
19 */
20
21 package org.apache.hadoop.hbase.regionserver;
22
23 import java.io.IOException;
24 import java.util.NavigableSet;
25
26 import org.apache.hadoop.hbase.HConstants;
27 import org.apache.hadoop.hbase.KeyValue;
28 import org.apache.hadoop.hbase.client.Scan;
29 import org.apache.hadoop.hbase.filter.Filter;
30 import org.apache.hadoop.hbase.filter.Filter.ReturnCode;
31 import org.apache.hadoop.hbase.io.TimeRange;
32 import org.apache.hadoop.hbase.regionserver.DeleteTracker.DeleteResult;
33 import org.apache.hadoop.hbase.util.Bytes;
34 import org.apache.hadoop.hbase.util.EnvironmentEdgeManager;
35
36 /**
37 * A query matcher that is specifically designed for the scan case.
38 */
39 public class ScanQueryMatcher {
40 // Optimization so we can skip lots of compares when we decide to skip
41 // to the next row.
42 private boolean stickyNextRow;
43 private final byte[] stopRow;
44
45 private final TimeRange tr;
46
47 private final Filter filter;
48
49 /** Keeps track of deletes */
50 private final DeleteTracker deletes;
51
52 /*
53 * The following three booleans define how we deal with deletes.
54 * There are three different aspects:
55 * 1. Whether to keep delete markers. This is used in compactions.
56 * Minor compactions always keep delete markers.
57 * 2. Whether to keep deleted rows. This is also used in compactions,
58 * if the store is set to keep deleted rows. This implies keeping
59 * the delete markers as well.
60 * In this case deleted rows are subject to the normal max version
61 * and TTL/min version rules just like "normal" rows.
62 * 3. Whether a scan can do time travel queries even before deleted
63 * marker to reach deleted rows.
64 */
65 /** whether to retain delete markers */
66 private final boolean retainDeletesInOutput;
67 /** whether to return deleted rows */
68 private final boolean keepDeletedCells;
69 /** whether time range queries can see rows "behind" a delete */
70 private final boolean seePastDeleteMarkers;
71
72
73 /** Keeps track of columns and versions */
74 private final ColumnTracker columns;
75
76 /** Key to seek to in memstore and StoreFiles */
77 private final KeyValue startKey;
78
79 /** Row comparator for the region this query is for */
80 private final KeyValue.KeyComparator rowComparator;
81
82 /* row is not private for tests */
83 /** Row the query is on */
84 byte [] row;
85 int rowOffset;
86 short rowLength;
87
88 /**
89 * Oldest put in any of the involved store files
90 * Used to decide whether it is ok to delete
91 * family delete marker of this store keeps
92 * deleted KVs.
93 */
94 private final long earliestPutTs;
95
96 /** readPoint over which the KVs are unconditionally included */
97 protected long maxReadPointToTrackVersions;
98
99 /**
100 * This variable shows whether there is an null column in the query. There
101 * always exists a null column in the wildcard column query.
102 * There maybe exists a null column in the explicit column query based on the
103 * first column.
104 * */
105 private boolean hasNullColumn = true;
106
107 // By default, when hbase.hstore.time.to.purge.deletes is 0ms, a delete
108 // marker is always removed during a major compaction. If set to non-zero
109 // value then major compaction will try to keep a delete marker around for
110 // the given number of milliseconds. We want to keep the delete markers
111 // around a bit longer because old puts might appear out-of-order. For
112 // example, during log replication between two clusters.
113 //
114 // If the delete marker has lived longer than its column-family's TTL then
115 // the delete marker will be removed even if time.to.purge.deletes has not
116 // passed. This is because all the Puts that this delete marker can influence
117 // would have also expired. (Removing of delete markers on col family TTL will
118 // not happen if min-versions is set to non-zero)
119 //
120 // But, if time.to.purge.deletes has not expired then a delete
121 // marker will not be removed just because there are no Puts that it is
122 // currently influencing. This is because Puts, that this delete can
123 // influence. may appear out of order.
124 private final long timeToPurgeDeletes;
125
126 private final boolean isUserScan;
127
128 /**
129 * Construct a QueryMatcher for a scan
130 * @param scan
131 * @param scanInfo The store's immutable scan info
132 * @param columns
133 * @param scanType Type of the scan
134 * @param earliestPutTs Earliest put seen in any of the store files.
135 * @param oldestUnexpiredTS the oldest timestamp we are interested in,
136 * based on TTL
137 */
138 public ScanQueryMatcher(Scan scan, Store.ScanInfo scanInfo,
139 NavigableSet<byte[]> columns, ScanType scanType,
140 long readPointToUse, long earliestPutTs, long oldestUnexpiredTS) {
141 this.tr = scan.getTimeRange();
142 this.rowComparator = scanInfo.getComparator().getRawComparator();
143 this.deletes = new ScanDeleteTracker();
144 this.stopRow = scan.getStopRow();
145 this.startKey = KeyValue.createFirstDeleteFamilyOnRow(scan.getStartRow(),
146 scanInfo.getFamily());
147 this.filter = scan.getFilter();
148 this.earliestPutTs = earliestPutTs;
149 this.maxReadPointToTrackVersions = readPointToUse;
150 this.timeToPurgeDeletes = scanInfo.getTimeToPurgeDeletes();
151
152 /* how to deal with deletes */
153 this.isUserScan = scanType == ScanType.USER_SCAN;
154 // keep deleted cells: if compaction or raw scan
155 this.keepDeletedCells = (scanInfo.getKeepDeletedCells() && !isUserScan) || scan.isRaw();
156 // retain deletes: if minor compaction or raw scan
157 this.retainDeletesInOutput = scanType == ScanType.MINOR_COMPACT || scan.isRaw();
158 // seePastDeleteMarker: user initiated scans
159 this.seePastDeleteMarkers = scanInfo.getKeepDeletedCells() && isUserScan;
160
161 int maxVersions = Math.min(scan.getMaxVersions(), scanInfo.getMaxVersions());
162 // Single branch to deal with two types of reads (columns vs all in family)
163 if (columns == null || columns.size() == 0) {
164 // there is always a null column in the wildcard column query.
165 hasNullColumn = true;
166
167 // use a specialized scan for wildcard column tracker.
168 this.columns = new ScanWildcardColumnTracker(
169 scanInfo.getMinVersions(), maxVersions, oldestUnexpiredTS);
170 } else {
171 // whether there is null column in the explicit column query
172 hasNullColumn = (columns.first().length == 0);
173
174 // We can share the ExplicitColumnTracker, diff is we reset
175 // between rows, not between storefiles.
176 this.columns = new ExplicitColumnTracker(columns,
177 scanInfo.getMinVersions(), maxVersions, oldestUnexpiredTS);
178 }
179 }
180
181 /*
182 * Constructor for tests
183 */
184 ScanQueryMatcher(Scan scan, Store.ScanInfo scanInfo,
185 NavigableSet<byte[]> columns, long oldestUnexpiredTS) {
186 this(scan, scanInfo, columns, ScanType.USER_SCAN,
187 Long.MAX_VALUE, /* max Readpoint to track versions */
188 HConstants.LATEST_TIMESTAMP, oldestUnexpiredTS);
189 }
190
191 /**
192 *
193 * @return whether there is an null column in the query
194 */
195 public boolean hasNullColumnInQuery() {
196 return hasNullColumn;
197 }
198
199 /**
200 * Determines if the caller should do one of several things:
201 * - seek/skip to the next row (MatchCode.SEEK_NEXT_ROW)
202 * - seek/skip to the next column (MatchCode.SEEK_NEXT_COL)
203 * - include the current KeyValue (MatchCode.INCLUDE)
204 * - ignore the current KeyValue (MatchCode.SKIP)
205 * - got to the next row (MatchCode.DONE)
206 *
207 * @param kv KeyValue to check
208 * @return The match code instance.
209 * @throws IOException in case there is an internal consistency problem
210 * caused by a data corruption.
211 */
212 public MatchCode match(KeyValue kv) throws IOException {
213 if (filter != null && filter.filterAllRemaining()) {
214 return MatchCode.DONE_SCAN;
215 }
216
217 byte [] bytes = kv.getBuffer();
218 int offset = kv.getOffset();
219 int initialOffset = offset;
220
221 int keyLength = Bytes.toInt(bytes, offset, Bytes.SIZEOF_INT);
222 offset += KeyValue.ROW_OFFSET;
223
224 short rowLength = Bytes.toShort(bytes, offset, Bytes.SIZEOF_SHORT);
225 offset += Bytes.SIZEOF_SHORT;
226
227 int ret = this.rowComparator.compareRows(row, this.rowOffset, this.rowLength,
228 bytes, offset, rowLength);
229 if (ret <= -1) {
230 return MatchCode.DONE;
231 } else if (ret >= 1) {
232 // could optimize this, if necessary?
233 // Could also be called SEEK_TO_CURRENT_ROW, but this
234 // should be rare/never happens.
235 return MatchCode.SEEK_NEXT_ROW;
236 }
237
238 // optimize case.
239 if (this.stickyNextRow)
240 return MatchCode.SEEK_NEXT_ROW;
241
242 if (this.columns.done()) {
243 stickyNextRow = true;
244 return MatchCode.SEEK_NEXT_ROW;
245 }
246
247 //Passing rowLength
248 offset += rowLength;
249
250 //Skipping family
251 byte familyLength = bytes [offset];
252 offset += familyLength + 1;
253
254 int qualLength = keyLength + KeyValue.ROW_OFFSET -
255 (offset - initialOffset) - KeyValue.TIMESTAMP_TYPE_SIZE;
256
257 long timestamp = kv.getTimestamp();
258 // check for early out based on timestamp alone
259 if (columns.isDone(timestamp)) {
260 return columns.getNextRowOrNextColumn(bytes, offset, qualLength);
261 }
262
263 /*
264 * The delete logic is pretty complicated now.
265 * This is corroborated by the following:
266 * 1. The store might be instructed to keep deleted rows around.
267 * 2. A scan can optionally see past a delete marker now.
268 * 3. If deleted rows are kept, we have to find out when we can
269 * remove the delete markers.
270 * 4. Family delete markers are always first (regardless of their TS)
271 * 5. Delete markers should not be counted as version
272 * 6. Delete markers affect puts of the *same* TS
273 * 7. Delete marker need to be version counted together with puts
274 * they affect
275 */
276 byte type = kv.getType();
277 if (kv.isDelete()) {
278 if (!keepDeletedCells) {
279 // first ignore delete markers if the scanner can do so, and the
280 // range does not include the marker
281 //
282 // during flushes and compactions also ignore delete markers newer
283 // than the readpoint of any open scanner, this prevents deleted
284 // rows that could still be seen by a scanner from being collected
285 boolean includeDeleteMarker = seePastDeleteMarkers ?
286 tr.withinTimeRange(timestamp) :
287 tr.withinOrAfterTimeRange(timestamp);
288 if (includeDeleteMarker
289 && kv.getMemstoreTS() <= maxReadPointToTrackVersions) {
290 this.deletes.add(bytes, offset, qualLength, timestamp, type);
291 }
292 // Can't early out now, because DelFam come before any other keys
293 }
294 if (retainDeletesInOutput
295 || (!isUserScan && (EnvironmentEdgeManager.currentTimeMillis() - timestamp) <= timeToPurgeDeletes)
296 || kv.getMemstoreTS() > maxReadPointToTrackVersions) {
297 // always include or it is not time yet to check whether it is OK
298 // to purge deltes or not
299 return MatchCode.INCLUDE;
300 } else if (keepDeletedCells) {
301 if (timestamp < earliestPutTs) {
302 // keeping delete rows, but there are no puts older than
303 // this delete in the store files.
304 return columns.getNextRowOrNextColumn(bytes, offset, qualLength);
305 }
306 // else: fall through and do version counting on the
307 // delete markers
308 } else {
309 return MatchCode.SKIP;
310 }
311 // note the following next else if...
312 // delete marker are not subject to other delete markers
313 } else if (!this.deletes.isEmpty()) {
314 DeleteResult deleteResult = deletes.isDeleted(bytes, offset, qualLength,
315 timestamp);
316 switch (deleteResult) {
317 case FAMILY_DELETED:
318 case COLUMN_DELETED:
319 return columns.getNextRowOrNextColumn(bytes, offset, qualLength);
320 case VERSION_DELETED:
321 return MatchCode.SKIP;
322 case NOT_DELETED:
323 break;
324 default:
325 throw new RuntimeException("UNEXPECTED");
326 }
327 }
328
329 int timestampComparison = tr.compare(timestamp);
330 if (timestampComparison >= 1) {
331 return MatchCode.SKIP;
332 } else if (timestampComparison <= -1) {
333 return columns.getNextRowOrNextColumn(bytes, offset, qualLength);
334 }
335
336 /**
337 * Filters should be checked before checking column trackers. If we do
338 * otherwise, as was previously being done, ColumnTracker may increment its
339 * counter for even that KV which may be discarded later on by Filter. This
340 * would lead to incorrect results in certain cases.
341 */
342 ReturnCode filterResponse = ReturnCode.SKIP;
343 if (filter != null) {
344 filterResponse = filter.filterKeyValue(kv);
345 if (filterResponse == ReturnCode.SKIP) {
346 return MatchCode.SKIP;
347 } else if (filterResponse == ReturnCode.NEXT_COL) {
348 return columns.getNextRowOrNextColumn(bytes, offset, qualLength);
349 } else if (filterResponse == ReturnCode.NEXT_ROW) {
350 stickyNextRow = true;
351 return MatchCode.SEEK_NEXT_ROW;
352 } else if (filterResponse == ReturnCode.SEEK_NEXT_USING_HINT) {
353 return MatchCode.SEEK_NEXT_USING_HINT;
354 }
355 }
356
357 MatchCode colChecker = columns.checkColumn(bytes, offset, qualLength,
358 timestamp, type, kv.getMemstoreTS() > maxReadPointToTrackVersions);
359 /*
360 * According to current implementation, colChecker can only be
361 * SEEK_NEXT_COL, SEEK_NEXT_ROW, SKIP or INCLUDE. Therefore, always return
362 * the MatchCode. If it is SEEK_NEXT_ROW, also set stickyNextRow.
363 */
364 if (colChecker == MatchCode.SEEK_NEXT_ROW) {
365 stickyNextRow = true;
366 } else if (filter != null && colChecker == MatchCode.INCLUDE &&
367 filterResponse == ReturnCode.INCLUDE_AND_NEXT_COL) {
368 return MatchCode.INCLUDE_AND_SEEK_NEXT_COL;
369 }
370 return colChecker;
371
372 }
373
374 public boolean moreRowsMayExistAfter(KeyValue kv) {
375 if (!Bytes.equals(stopRow , HConstants.EMPTY_END_ROW) &&
376 rowComparator.compareRows(kv.getBuffer(),kv.getRowOffset(),
377 kv.getRowLength(), stopRow, 0, stopRow.length) >= 0) {
378 // KV >= STOPROW
379 // then NO there is nothing left.
380 return false;
381 } else {
382 return true;
383 }
384 }
385
386 /**
387 * Set current row
388 * @param row
389 */
390 public void setRow(byte [] row, int offset, short length) {
391 this.row = row;
392 this.rowOffset = offset;
393 this.rowLength = length;
394 reset();
395 }
396
397 public void reset() {
398 this.deletes.reset();
399 this.columns.reset();
400
401 stickyNextRow = false;
402 }
403
404 /**
405 *
406 * @return the start key
407 */
408 public KeyValue getStartKey() {
409 return this.startKey;
410 }
411
412 /**
413 *
414 * @return the Filter
415 */
416 Filter getFilter() {
417 return this.filter;
418 }
419
420 public KeyValue getNextKeyHint(KeyValue kv) {
421 if (filter == null) {
422 return null;
423 } else {
424 return filter.getNextKeyHint(kv);
425 }
426 }
427
428 public KeyValue getKeyForNextColumn(KeyValue kv) {
429 ColumnCount nextColumn = columns.getColumnHint();
430 if (nextColumn == null) {
431 return KeyValue.createLastOnRow(
432 kv.getBuffer(), kv.getRowOffset(), kv.getRowLength(),
433 kv.getBuffer(), kv.getFamilyOffset(), kv.getFamilyLength(),
434 kv.getBuffer(), kv.getQualifierOffset(), kv.getQualifierLength());
435 } else {
436 return KeyValue.createFirstOnRow(
437 kv.getBuffer(), kv.getRowOffset(), kv.getRowLength(),
438 kv.getBuffer(), kv.getFamilyOffset(), kv.getFamilyLength(),
439 nextColumn.getBuffer(), nextColumn.getOffset(), nextColumn.getLength());
440 }
441 }
442
443 public KeyValue getKeyForNextRow(KeyValue kv) {
444 return KeyValue.createLastOnRow(
445 kv.getBuffer(), kv.getRowOffset(), kv.getRowLength(),
446 null, 0, 0,
447 null, 0, 0);
448 }
449
450 /**
451 * {@link #match} return codes. These instruct the scanner moving through
452 * memstores and StoreFiles what to do with the current KeyValue.
453 * <p>
454 * Additionally, this contains "early-out" language to tell the scanner to
455 * move on to the next File (memstore or Storefile), or to return immediately.
456 */
457 public static enum MatchCode {
458 /**
459 * Include KeyValue in the returned result
460 */
461 INCLUDE,
462
463 /**
464 * Do not include KeyValue in the returned result
465 */
466 SKIP,
467
468 /**
469 * Do not include, jump to next StoreFile or memstore (in time order)
470 */
471 NEXT,
472
473 /**
474 * Do not include, return current result
475 */
476 DONE,
477
478 /**
479 * These codes are used by the ScanQueryMatcher
480 */
481
482 /**
483 * Done with the row, seek there.
484 */
485 SEEK_NEXT_ROW,
486 /**
487 * Done with column, seek to next.
488 */
489 SEEK_NEXT_COL,
490
491 /**
492 * Done with scan, thanks to the row filter.
493 */
494 DONE_SCAN,
495
496 /*
497 * Seek to next key which is given as hint.
498 */
499 SEEK_NEXT_USING_HINT,
500
501 /**
502 * Include KeyValue and done with column, seek to next.
503 */
504 INCLUDE_AND_SEEK_NEXT_COL,
505
506 /**
507 * Include KeyValue and done with row, seek to next.
508 */
509 INCLUDE_AND_SEEK_NEXT_ROW,
510 }
511 }