001/* 002 * Licensed to the Apache Software Foundation (ASF) under one 003 * or more contributor license agreements. See the NOTICE file 004 * distributed with this work for additional information 005 * regarding copyright ownership. The ASF licenses this file 006 * to you under the Apache License, Version 2.0 (the 007 * "License"); you may not use this file except in compliance 008 * with the License. You may obtain a copy of the License at 009 * 010 * http://www.apache.org/licenses/LICENSE-2.0 011 * 012 * Unless required by applicable law or agreed to in writing, software 013 * distributed under the License is distributed on an "AS IS" BASIS, 014 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 015 * See the License for the specific language governing permissions and 016 * limitations under the License. 017 */ 018package org.apache.hadoop.hbase.filter; 019 020import java.util.ArrayList; 021import java.util.Arrays; 022import java.util.Comparator; 023import java.util.List; 024import java.util.Objects; 025import java.util.PriorityQueue; 026import org.apache.hadoop.hbase.Cell; 027import org.apache.hadoop.hbase.CellComparator; 028import org.apache.hadoop.hbase.PrivateCellUtil; 029import org.apache.hadoop.hbase.exceptions.DeserializationException; 030import org.apache.hadoop.hbase.unsafe.HBasePlatformDependent; 031import org.apache.hadoop.hbase.util.Bytes; 032import org.apache.hadoop.hbase.util.Pair; 033import org.apache.yetus.audience.InterfaceAudience; 034 035import org.apache.hbase.thirdparty.com.google.protobuf.InvalidProtocolBufferException; 036import org.apache.hbase.thirdparty.com.google.protobuf.UnsafeByteOperations; 037 038import org.apache.hadoop.hbase.shaded.protobuf.generated.FilterProtos; 039import org.apache.hadoop.hbase.shaded.protobuf.generated.HBaseProtos.BytesBytesPair; 040 041/** 042 * This is optimized version of a standard FuzzyRowFilter Filters data based on fuzzy row key. 043 * Performs fast-forwards during scanning. It takes pairs (row key, fuzzy info) to match row keys. 044 * Where fuzzy info is a byte array with 0 or 1 as its values: 045 * <ul> 046 * <li>0 - means that this byte in provided row key is fixed, i.e. row key's byte at same position 047 * must match</li> 048 * <li>1 - means that this byte in provided row key is NOT fixed, i.e. row key's byte at this 049 * position can be different from the one in provided row key</li> 050 * </ul> 051 * Example: Let's assume row key format is userId_actionId_year_month. Length of userId is fixed and 052 * is 4, length of actionId is 2 and year and month are 4 and 2 bytes long respectively. Let's 053 * assume that we need to fetch all users that performed certain action (encoded as "99") in Jan of 054 * any year. Then the pair (row key, fuzzy info) would be the following: row key = "????_99_????_01" 055 * (one can use any value instead of "?") fuzzy info = 056 * "\x01\x01\x01\x01\x00\x00\x00\x00\x01\x01\x01\x01\x00\x00\x00" I.e. fuzzy info tells the matching 057 * mask is "????_99_????_01", where at ? can be any value. 058 */ 059@InterfaceAudience.Public 060public class FuzzyRowFilter extends FilterBase { 061 private static final boolean UNSAFE_UNALIGNED = HBasePlatformDependent.unaligned(); 062 private List<Pair<byte[], byte[]>> fuzzyKeysData; 063 private boolean done = false; 064 065 /** 066 * The index of a last successfully found matching fuzzy string (in fuzzyKeysData). We will start 067 * matching next KV with this one. If they do not match then we will return back to the one-by-one 068 * iteration over fuzzyKeysData. 069 */ 070 private int lastFoundIndex = -1; 071 072 /** 073 * Row tracker (keeps all next rows after SEEK_NEXT_USING_HINT was returned) 074 */ 075 private RowTracker tracker; 076 077 public FuzzyRowFilter(List<Pair<byte[], byte[]>> fuzzyKeysData) { 078 List<Pair<byte[], byte[]>> fuzzyKeyDataCopy = new ArrayList<>(fuzzyKeysData.size()); 079 080 for (Pair<byte[], byte[]> aFuzzyKeysData : fuzzyKeysData) { 081 if (aFuzzyKeysData.getFirst().length != aFuzzyKeysData.getSecond().length) { 082 Pair<String, String> readable = new Pair<>(Bytes.toStringBinary(aFuzzyKeysData.getFirst()), 083 Bytes.toStringBinary(aFuzzyKeysData.getSecond())); 084 throw new IllegalArgumentException("Fuzzy pair lengths do not match: " + readable); 085 } 086 087 Pair<byte[], byte[]> p = new Pair<>(); 088 // create a copy of pair bytes so that they are not modified by the filter. 089 p.setFirst(Arrays.copyOf(aFuzzyKeysData.getFirst(), aFuzzyKeysData.getFirst().length)); 090 p.setSecond(Arrays.copyOf(aFuzzyKeysData.getSecond(), aFuzzyKeysData.getSecond().length)); 091 092 // update mask ( 0 -> -1 (0xff), 1 -> 2) 093 p.setSecond(preprocessMask(p.getSecond())); 094 preprocessSearchKey(p); 095 096 fuzzyKeyDataCopy.add(p); 097 } 098 this.fuzzyKeysData = fuzzyKeyDataCopy; 099 this.tracker = new RowTracker(); 100 } 101 102 private void preprocessSearchKey(Pair<byte[], byte[]> p) { 103 if (!UNSAFE_UNALIGNED) { 104 // do nothing 105 return; 106 } 107 byte[] key = p.getFirst(); 108 byte[] mask = p.getSecond(); 109 for (int i = 0; i < mask.length; i++) { 110 // set non-fixed part of a search key to 0. 111 if (mask[i] == 2) { 112 key[i] = 0; 113 } 114 } 115 } 116 117 /** 118 * We need to preprocess mask array, as since we treat 2's as unfixed positions and -1 (0xff) as 119 * fixed positions n * @return mask array 120 */ 121 private byte[] preprocessMask(byte[] mask) { 122 if (!UNSAFE_UNALIGNED) { 123 // do nothing 124 return mask; 125 } 126 if (isPreprocessedMask(mask)) return mask; 127 for (int i = 0; i < mask.length; i++) { 128 if (mask[i] == 0) { 129 mask[i] = -1; // 0 -> -1 130 } else if (mask[i] == 1) { 131 mask[i] = 2;// 1 -> 2 132 } 133 } 134 return mask; 135 } 136 137 private boolean isPreprocessedMask(byte[] mask) { 138 for (int i = 0; i < mask.length; i++) { 139 if (mask[i] != -1 && mask[i] != 2) { 140 return false; 141 } 142 } 143 return true; 144 } 145 146 @Override 147 public ReturnCode filterCell(final Cell c) { 148 final int startIndex = lastFoundIndex >= 0 ? lastFoundIndex : 0; 149 final int size = fuzzyKeysData.size(); 150 for (int i = startIndex; i < size + startIndex; i++) { 151 final int index = i % size; 152 Pair<byte[], byte[]> fuzzyData = fuzzyKeysData.get(index); 153 // This shift is idempotent - always end up with 0 and -1 as mask values. 154 for (int j = 0; j < fuzzyData.getSecond().length; j++) { 155 fuzzyData.getSecond()[j] >>= 2; 156 } 157 SatisfiesCode satisfiesCode = satisfies(isReversed(), c.getRowArray(), c.getRowOffset(), 158 c.getRowLength(), fuzzyData.getFirst(), fuzzyData.getSecond()); 159 if (satisfiesCode == SatisfiesCode.YES) { 160 lastFoundIndex = index; 161 return ReturnCode.INCLUDE; 162 } 163 } 164 // NOT FOUND -> seek next using hint 165 lastFoundIndex = -1; 166 167 return ReturnCode.SEEK_NEXT_USING_HINT; 168 169 } 170 171 @Override 172 public Cell getNextCellHint(Cell currentCell) { 173 boolean result = tracker.updateTracker(currentCell); 174 if (result == false) { 175 done = true; 176 return null; 177 } 178 byte[] nextRowKey = tracker.nextRow(); 179 return PrivateCellUtil.createFirstOnRow(nextRowKey, 0, (short) nextRowKey.length); 180 } 181 182 /** 183 * If we have multiple fuzzy keys, row tracker should improve overall performance. It calculates 184 * all next rows (one per every fuzzy key) and put them (the fuzzy key is bundled) into a priority 185 * queue so that the smallest row key always appears at queue head, which helps to decide the 186 * "Next Cell Hint". As scanning going on, the number of candidate rows in the RowTracker will 187 * remain the size of fuzzy keys until some of the fuzzy keys won't possibly have matches any 188 * more. 189 */ 190 private class RowTracker { 191 private final PriorityQueue<Pair<byte[], Pair<byte[], byte[]>>> nextRows; 192 private boolean initialized = false; 193 194 RowTracker() { 195 nextRows = new PriorityQueue<>(fuzzyKeysData.size(), 196 new Comparator<Pair<byte[], Pair<byte[], byte[]>>>() { 197 @Override 198 public int compare(Pair<byte[], Pair<byte[], byte[]>> o1, 199 Pair<byte[], Pair<byte[], byte[]>> o2) { 200 return isReversed() 201 ? Bytes.compareTo(o2.getFirst(), o1.getFirst()) 202 : Bytes.compareTo(o1.getFirst(), o2.getFirst()); 203 } 204 }); 205 } 206 207 byte[] nextRow() { 208 if (nextRows.isEmpty()) { 209 throw new IllegalStateException("NextRows should not be empty, " 210 + "make sure to call nextRow() after updateTracker() return true"); 211 } else { 212 return nextRows.peek().getFirst(); 213 } 214 } 215 216 boolean updateTracker(Cell currentCell) { 217 if (!initialized) { 218 for (Pair<byte[], byte[]> fuzzyData : fuzzyKeysData) { 219 updateWith(currentCell, fuzzyData); 220 } 221 initialized = true; 222 } else { 223 while (!nextRows.isEmpty() && !lessThan(currentCell, nextRows.peek().getFirst())) { 224 Pair<byte[], Pair<byte[], byte[]>> head = nextRows.poll(); 225 Pair<byte[], byte[]> fuzzyData = head.getSecond(); 226 updateWith(currentCell, fuzzyData); 227 } 228 } 229 return !nextRows.isEmpty(); 230 } 231 232 boolean lessThan(Cell currentCell, byte[] nextRowKey) { 233 int compareResult = 234 CellComparator.getInstance().compareRows(currentCell, nextRowKey, 0, nextRowKey.length); 235 return (!isReversed() && compareResult < 0) || (isReversed() && compareResult > 0); 236 } 237 238 void updateWith(Cell currentCell, Pair<byte[], byte[]> fuzzyData) { 239 byte[] nextRowKeyCandidate = 240 getNextForFuzzyRule(isReversed(), currentCell.getRowArray(), currentCell.getRowOffset(), 241 currentCell.getRowLength(), fuzzyData.getFirst(), fuzzyData.getSecond()); 242 if (nextRowKeyCandidate != null) { 243 nextRows.add(new Pair<>(nextRowKeyCandidate, fuzzyData)); 244 } 245 } 246 247 } 248 249 @Override 250 public boolean filterAllRemaining() { 251 return done; 252 } 253 254 /** 255 * @return The filter serialized using pb 256 */ 257 @Override 258 public byte[] toByteArray() { 259 FilterProtos.FuzzyRowFilter.Builder builder = FilterProtos.FuzzyRowFilter.newBuilder(); 260 for (Pair<byte[], byte[]> fuzzyData : fuzzyKeysData) { 261 BytesBytesPair.Builder bbpBuilder = BytesBytesPair.newBuilder(); 262 bbpBuilder.setFirst(UnsafeByteOperations.unsafeWrap(fuzzyData.getFirst())); 263 bbpBuilder.setSecond(UnsafeByteOperations.unsafeWrap(fuzzyData.getSecond())); 264 builder.addFuzzyKeysData(bbpBuilder); 265 } 266 return builder.build().toByteArray(); 267 } 268 269 /** 270 * @param pbBytes A pb serialized {@link FuzzyRowFilter} instance 271 * @return An instance of {@link FuzzyRowFilter} made from <code>bytes</code> n * @see 272 * #toByteArray 273 */ 274 public static FuzzyRowFilter parseFrom(final byte[] pbBytes) throws DeserializationException { 275 FilterProtos.FuzzyRowFilter proto; 276 try { 277 proto = FilterProtos.FuzzyRowFilter.parseFrom(pbBytes); 278 } catch (InvalidProtocolBufferException e) { 279 throw new DeserializationException(e); 280 } 281 int count = proto.getFuzzyKeysDataCount(); 282 ArrayList<Pair<byte[], byte[]>> fuzzyKeysData = new ArrayList<>(count); 283 for (int i = 0; i < count; ++i) { 284 BytesBytesPair current = proto.getFuzzyKeysData(i); 285 byte[] keyBytes = current.getFirst().toByteArray(); 286 byte[] keyMeta = current.getSecond().toByteArray(); 287 fuzzyKeysData.add(new Pair<>(keyBytes, keyMeta)); 288 } 289 return new FuzzyRowFilter(fuzzyKeysData); 290 } 291 292 @Override 293 public String toString() { 294 final StringBuilder sb = new StringBuilder(); 295 sb.append("FuzzyRowFilter"); 296 sb.append("{fuzzyKeysData="); 297 for (Pair<byte[], byte[]> fuzzyData : fuzzyKeysData) { 298 sb.append('{').append(Bytes.toStringBinary(fuzzyData.getFirst())).append(":"); 299 sb.append(Bytes.toStringBinary(fuzzyData.getSecond())).append('}'); 300 } 301 sb.append("}, "); 302 return sb.toString(); 303 } 304 305 // Utility methods 306 307 static enum SatisfiesCode { 308 /** row satisfies fuzzy rule */ 309 YES, 310 /** row doesn't satisfy fuzzy rule, but there's possible greater row that does */ 311 NEXT_EXISTS, 312 /** row doesn't satisfy fuzzy rule and there's no greater row that does */ 313 NO_NEXT 314 } 315 316 static SatisfiesCode satisfies(byte[] row, byte[] fuzzyKeyBytes, byte[] fuzzyKeyMeta) { 317 return satisfies(false, row, 0, row.length, fuzzyKeyBytes, fuzzyKeyMeta); 318 } 319 320 static SatisfiesCode satisfies(boolean reverse, byte[] row, byte[] fuzzyKeyBytes, 321 byte[] fuzzyKeyMeta) { 322 return satisfies(reverse, row, 0, row.length, fuzzyKeyBytes, fuzzyKeyMeta); 323 } 324 325 static SatisfiesCode satisfies(boolean reverse, byte[] row, int offset, int length, 326 byte[] fuzzyKeyBytes, byte[] fuzzyKeyMeta) { 327 328 if (!UNSAFE_UNALIGNED) { 329 return satisfiesNoUnsafe(reverse, row, offset, length, fuzzyKeyBytes, fuzzyKeyMeta); 330 } 331 332 if (row == null) { 333 // do nothing, let scan to proceed 334 return SatisfiesCode.YES; 335 } 336 length = Math.min(length, fuzzyKeyBytes.length); 337 int numWords = length / Bytes.SIZEOF_LONG; 338 339 int j = numWords << 3; // numWords * SIZEOF_LONG; 340 341 for (int i = 0; i < j; i += Bytes.SIZEOF_LONG) { 342 long fuzzyBytes = Bytes.toLong(fuzzyKeyBytes, i); 343 long fuzzyMeta = Bytes.toLong(fuzzyKeyMeta, i); 344 long rowValue = Bytes.toLong(row, offset + i); 345 if ((rowValue & fuzzyMeta) != (fuzzyBytes)) { 346 // We always return NEXT_EXISTS 347 return SatisfiesCode.NEXT_EXISTS; 348 } 349 } 350 351 int off = j; 352 353 if (length - off >= Bytes.SIZEOF_INT) { 354 int fuzzyBytes = Bytes.toInt(fuzzyKeyBytes, off); 355 int fuzzyMeta = Bytes.toInt(fuzzyKeyMeta, off); 356 int rowValue = Bytes.toInt(row, offset + off); 357 if ((rowValue & fuzzyMeta) != (fuzzyBytes)) { 358 // We always return NEXT_EXISTS 359 return SatisfiesCode.NEXT_EXISTS; 360 } 361 off += Bytes.SIZEOF_INT; 362 } 363 364 if (length - off >= Bytes.SIZEOF_SHORT) { 365 short fuzzyBytes = Bytes.toShort(fuzzyKeyBytes, off); 366 short fuzzyMeta = Bytes.toShort(fuzzyKeyMeta, off); 367 short rowValue = Bytes.toShort(row, offset + off); 368 if ((rowValue & fuzzyMeta) != (fuzzyBytes)) { 369 // We always return NEXT_EXISTS 370 // even if it does not (in this case getNextForFuzzyRule 371 // will return null) 372 return SatisfiesCode.NEXT_EXISTS; 373 } 374 off += Bytes.SIZEOF_SHORT; 375 } 376 377 if (length - off >= Bytes.SIZEOF_BYTE) { 378 int fuzzyBytes = fuzzyKeyBytes[off] & 0xff; 379 int fuzzyMeta = fuzzyKeyMeta[off] & 0xff; 380 int rowValue = row[offset + off] & 0xff; 381 if ((rowValue & fuzzyMeta) != (fuzzyBytes)) { 382 // We always return NEXT_EXISTS 383 return SatisfiesCode.NEXT_EXISTS; 384 } 385 } 386 return SatisfiesCode.YES; 387 } 388 389 static SatisfiesCode satisfiesNoUnsafe(boolean reverse, byte[] row, int offset, int length, 390 byte[] fuzzyKeyBytes, byte[] fuzzyKeyMeta) { 391 if (row == null) { 392 // do nothing, let scan to proceed 393 return SatisfiesCode.YES; 394 } 395 396 Order order = Order.orderFor(reverse); 397 boolean nextRowKeyCandidateExists = false; 398 399 for (int i = 0; i < fuzzyKeyMeta.length && i < length; i++) { 400 // First, checking if this position is fixed and not equals the given one 401 boolean byteAtPositionFixed = fuzzyKeyMeta[i] == 0; 402 boolean fixedByteIncorrect = byteAtPositionFixed && fuzzyKeyBytes[i] != row[i + offset]; 403 if (fixedByteIncorrect) { 404 // in this case there's another row that satisfies fuzzy rule and bigger than this row 405 if (nextRowKeyCandidateExists) { 406 return SatisfiesCode.NEXT_EXISTS; 407 } 408 409 // If this row byte is less than fixed then there's a byte array bigger than 410 // this row and which satisfies the fuzzy rule. Otherwise there's no such byte array: 411 // this row is simply bigger than any byte array that satisfies the fuzzy rule 412 boolean rowByteLessThanFixed = (row[i + offset] & 0xFF) < (fuzzyKeyBytes[i] & 0xFF); 413 if (rowByteLessThanFixed && !reverse) { 414 return SatisfiesCode.NEXT_EXISTS; 415 } else if (!rowByteLessThanFixed && reverse) { 416 return SatisfiesCode.NEXT_EXISTS; 417 } else { 418 return SatisfiesCode.NO_NEXT; 419 } 420 } 421 422 // Second, checking if this position is not fixed and byte value is not the biggest. In this 423 // case there's a byte array bigger than this row and which satisfies the fuzzy rule. To get 424 // bigger byte array that satisfies the rule we need to just increase this byte 425 // (see the code of getNextForFuzzyRule below) by one. 426 // Note: if non-fixed byte is already at biggest value, this doesn't allow us to say there's 427 // bigger one that satisfies the rule as it can't be increased. 428 if (fuzzyKeyMeta[i] == 1 && !order.isMax(fuzzyKeyBytes[i])) { 429 nextRowKeyCandidateExists = true; 430 } 431 } 432 return SatisfiesCode.YES; 433 } 434 435 static byte[] getNextForFuzzyRule(byte[] row, byte[] fuzzyKeyBytes, byte[] fuzzyKeyMeta) { 436 return getNextForFuzzyRule(false, row, 0, row.length, fuzzyKeyBytes, fuzzyKeyMeta); 437 } 438 439 static byte[] getNextForFuzzyRule(boolean reverse, byte[] row, byte[] fuzzyKeyBytes, 440 byte[] fuzzyKeyMeta) { 441 return getNextForFuzzyRule(reverse, row, 0, row.length, fuzzyKeyBytes, fuzzyKeyMeta); 442 } 443 444 /** Abstracts directional comparisons based on scan direction. */ 445 private enum Order { 446 ASC { 447 @Override 448 public boolean lt(int lhs, int rhs) { 449 return lhs < rhs; 450 } 451 452 @Override 453 public boolean gt(int lhs, int rhs) { 454 return lhs > rhs; 455 } 456 457 @Override 458 public byte inc(byte val) { 459 // TODO: what about over/underflow? 460 return (byte) (val + 1); 461 } 462 463 @Override 464 public boolean isMax(byte val) { 465 return val == (byte) 0xff; 466 } 467 468 @Override 469 public byte min() { 470 return 0; 471 } 472 }, 473 DESC { 474 @Override 475 public boolean lt(int lhs, int rhs) { 476 return lhs > rhs; 477 } 478 479 @Override 480 public boolean gt(int lhs, int rhs) { 481 return lhs < rhs; 482 } 483 484 @Override 485 public byte inc(byte val) { 486 // TODO: what about over/underflow? 487 return (byte) (val - 1); 488 } 489 490 @Override 491 public boolean isMax(byte val) { 492 return val == 0; 493 } 494 495 @Override 496 public byte min() { 497 return (byte) 0xFF; 498 } 499 }; 500 501 public static Order orderFor(boolean reverse) { 502 return reverse ? DESC : ASC; 503 } 504 505 /** Returns true when {@code lhs < rhs}. */ 506 public abstract boolean lt(int lhs, int rhs); 507 508 /** Returns true when {@code lhs > rhs}. */ 509 public abstract boolean gt(int lhs, int rhs); 510 511 /** Returns {@code val} incremented by 1. */ 512 public abstract byte inc(byte val); 513 514 /** Return true when {@code val} is the maximum value */ 515 public abstract boolean isMax(byte val); 516 517 /** Return the minimum value according to this ordering scheme. */ 518 public abstract byte min(); 519 } 520 521 /** 522 * @return greater byte array than given (row) which satisfies the fuzzy rule if it exists, null 523 * otherwise 524 */ 525 static byte[] getNextForFuzzyRule(boolean reverse, byte[] row, int offset, int length, 526 byte[] fuzzyKeyBytes, byte[] fuzzyKeyMeta) { 527 // To find out the next "smallest" byte array that satisfies fuzzy rule and "greater" than 528 // the given one we do the following: 529 // 1. setting values on all "fixed" positions to the values from fuzzyKeyBytes 530 // 2. if during the first step given row did not increase, then we increase the value at 531 // the first "non-fixed" position (where it is not maximum already) 532 533 // It is easier to perform this by using fuzzyKeyBytes copy and setting "non-fixed" position 534 // values than otherwise. 535 byte[] result = 536 Arrays.copyOf(fuzzyKeyBytes, length > fuzzyKeyBytes.length ? length : fuzzyKeyBytes.length); 537 if (reverse && length > fuzzyKeyBytes.length) { 538 // we need trailing 0xff's instead of trailing 0x00's 539 for (int i = fuzzyKeyBytes.length; i < result.length; i++) { 540 result[i] = (byte) 0xFF; 541 } 542 } 543 int toInc = -1; 544 final Order order = Order.orderFor(reverse); 545 546 boolean increased = false; 547 for (int i = 0; i < result.length; i++) { 548 if (i >= fuzzyKeyMeta.length || fuzzyKeyMeta[i] == 0 /* non-fixed */) { 549 result[i] = row[offset + i]; 550 if (!order.isMax(row[offset + i])) { 551 // this is "non-fixed" position and is not at max value, hence we can increase it 552 toInc = i; 553 } 554 } else if (i < fuzzyKeyMeta.length && fuzzyKeyMeta[i] == -1 /* fixed */) { 555 if (order.lt((row[i + offset] & 0xFF), (fuzzyKeyBytes[i] & 0xFF))) { 556 // if setting value for any fixed position increased the original array, 557 // we are OK 558 increased = true; 559 break; 560 } 561 562 if (order.gt((row[i + offset] & 0xFF), (fuzzyKeyBytes[i] & 0xFF))) { 563 // if setting value for any fixed position makes array "smaller", then just stop: 564 // in case we found some non-fixed position to increase we will do it, otherwise 565 // there's no "next" row key that satisfies fuzzy rule and "greater" than given row 566 break; 567 } 568 } 569 } 570 571 if (!increased) { 572 if (toInc < 0) { 573 return null; 574 } 575 result[toInc] = order.inc(result[toInc]); 576 577 // Setting all "non-fixed" positions to zeroes to the right of the one we increased so 578 // that found "next" row key is the smallest possible 579 for (int i = toInc + 1; i < result.length; i++) { 580 if (i >= fuzzyKeyMeta.length || fuzzyKeyMeta[i] == 0 /* non-fixed */) { 581 result[i] = order.min(); 582 } 583 } 584 } 585 586 return reverse ? result : trimTrailingZeroes(result, fuzzyKeyMeta, toInc); 587 } 588 589 /** 590 * For forward scanner, next cell hint should not contain any trailing zeroes unless they are part 591 * of fuzzyKeyMeta hint = '\x01\x01\x01\x00\x00' will skip valid row '\x01\x01\x01' nn * @param 592 * toInc - position of incremented byte 593 * @return trimmed version of result 594 */ 595 596 private static byte[] trimTrailingZeroes(byte[] result, byte[] fuzzyKeyMeta, int toInc) { 597 int off = fuzzyKeyMeta.length >= result.length ? result.length - 1 : fuzzyKeyMeta.length - 1; 598 for (; off >= 0; off--) { 599 if (fuzzyKeyMeta[off] != 0) break; 600 } 601 if (off < toInc) off = toInc; 602 byte[] retValue = new byte[off + 1]; 603 System.arraycopy(result, 0, retValue, 0, retValue.length); 604 return retValue; 605 } 606 607 /** 608 * @return true if and only if the fields of the filter that are serialized are equal to the 609 * corresponding fields in other. Used for testing. 610 */ 611 @Override 612 boolean areSerializedFieldsEqual(Filter o) { 613 if (o == this) return true; 614 if (!(o instanceof FuzzyRowFilter)) return false; 615 616 FuzzyRowFilter other = (FuzzyRowFilter) o; 617 if (this.fuzzyKeysData.size() != other.fuzzyKeysData.size()) return false; 618 for (int i = 0; i < fuzzyKeysData.size(); ++i) { 619 Pair<byte[], byte[]> thisData = this.fuzzyKeysData.get(i); 620 Pair<byte[], byte[]> otherData = other.fuzzyKeysData.get(i); 621 if ( 622 !(Bytes.equals(thisData.getFirst(), otherData.getFirst()) 623 && Bytes.equals(thisData.getSecond(), otherData.getSecond())) 624 ) { 625 return false; 626 } 627 } 628 return true; 629 } 630 631 @Override 632 public boolean equals(Object obj) { 633 return obj instanceof Filter && areSerializedFieldsEqual((Filter) obj); 634 } 635 636 @Override 637 public int hashCode() { 638 return Objects.hash(this.fuzzyKeysData); 639 } 640}