View Javadoc

1   /**
2    *
3    * Licensed to the Apache Software Foundation (ASF) under one
4    * or more contributor license agreements.  See the NOTICE file
5    * distributed with this work for additional information
6    * regarding copyright ownership.  The ASF licenses this file
7    * to you under the Apache License, Version 2.0 (the
8    * "License"); you may not use this file except in compliance
9    * with the License.  You may obtain a copy of the License at
10   *
11   *     http://www.apache.org/licenses/LICENSE-2.0
12   *
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.
18   */
19  
20  package org.apache.hadoop.hbase.regionserver;
21  
22  import java.io.IOException;
23  import java.util.NavigableSet;
24  
25  import org.apache.hadoop.classification.InterfaceAudience;
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  import com.google.common.base.Preconditions;
37  
38  /**
39   * A query matcher that is specifically designed for the scan case.
40   */
41  @InterfaceAudience.Private
42  public class ScanQueryMatcher {
43    // Optimization so we can skip lots of compares when we decide to skip
44    // to the next row.
45    private boolean stickyNextRow;
46    private final byte[] stopRow;
47  
48    private final TimeRange tr;
49  
50    private final Filter filter;
51  
52    /** Keeps track of deletes */
53    private final DeleteTracker deletes;
54  
55    /*
56     * The following three booleans define how we deal with deletes.
57     * There are three different aspects:
58     * 1. Whether to keep delete markers. This is used in compactions.
59     *    Minor compactions always keep delete markers.
60     * 2. Whether to keep deleted rows. This is also used in compactions,
61     *    if the store is set to keep deleted rows. This implies keeping
62     *    the delete markers as well.
63     *    In this case deleted rows are subject to the normal max version
64     *    and TTL/min version rules just like "normal" rows.
65     * 3. Whether a scan can do time travel queries even before deleted
66     *    marker to reach deleted rows.
67     */
68    /** whether to retain delete markers */
69    private boolean retainDeletesInOutput;
70  
71    /** whether to return deleted rows */
72    private final boolean keepDeletedCells;
73    /** whether time range queries can see rows "behind" a delete */
74    private final boolean seePastDeleteMarkers;
75  
76  
77    /** Keeps track of columns and versions */
78    private final ColumnTracker columns;
79  
80    /** Key to seek to in memstore and StoreFiles */
81    private final KeyValue startKey;
82  
83    /** Row comparator for the region this query is for */
84    private final KeyValue.KeyComparator rowComparator;
85  
86    /* row is not private for tests */
87    /** Row the query is on */
88    byte [] row;
89    int rowOffset;
90    short rowLength;
91    
92    /**
93     * Oldest put in any of the involved store files
94     * Used to decide whether it is ok to delete
95     * family delete marker of this store keeps
96     * deleted KVs.
97     */
98    private final long earliestPutTs;
99  
100   /** readPoint over which the KVs are unconditionally included */
101   protected long maxReadPointToTrackVersions;
102 
103   private byte[] dropDeletesFromRow = null, dropDeletesToRow = null;
104 
105   /**
106    * This variable shows whether there is an null column in the query. There
107    * always exists a null column in the wildcard column query.
108    * There maybe exists a null column in the explicit column query based on the
109    * first column.
110    * */
111   private boolean hasNullColumn = true;
112 
113   // By default, when hbase.hstore.time.to.purge.deletes is 0ms, a delete
114   // marker is always removed during a major compaction. If set to non-zero
115   // value then major compaction will try to keep a delete marker around for
116   // the given number of milliseconds. We want to keep the delete markers
117   // around a bit longer because old puts might appear out-of-order. For
118   // example, during log replication between two clusters.
119   //
120   // If the delete marker has lived longer than its column-family's TTL then
121   // the delete marker will be removed even if time.to.purge.deletes has not
122   // passed. This is because all the Puts that this delete marker can influence
123   // would have also expired. (Removing of delete markers on col family TTL will
124   // not happen if min-versions is set to non-zero)
125   //
126   // But, if time.to.purge.deletes has not expired then a delete
127   // marker will not be removed just because there are no Puts that it is
128   // currently influencing. This is because Puts, that this delete can
129   // influence.  may appear out of order.
130   private final long timeToPurgeDeletes;
131   
132   private final boolean isUserScan;
133 
134   /**
135    * Construct a QueryMatcher for a scan
136    * @param scan
137    * @param scanInfo The store's immutable scan info
138    * @param columns
139    * @param scanType Type of the scan
140    * @param earliestPutTs Earliest put seen in any of the store files.
141    * @param oldestUnexpiredTS the oldest timestamp we are interested in,
142    *  based on TTL
143    */
144   public ScanQueryMatcher(Scan scan, ScanInfo scanInfo,
145       NavigableSet<byte[]> columns, ScanType scanType,
146       long readPointToUse, long earliestPutTs, long oldestUnexpiredTS) {
147     this.tr = scan.getTimeRange();
148     this.rowComparator = scanInfo.getComparator().getRawComparator();
149     this.deletes =  new ScanDeleteTracker();
150     this.stopRow = scan.getStopRow();
151     this.startKey = KeyValue.createFirstDeleteFamilyOnRow(scan.getStartRow(),
152         scanInfo.getFamily());
153     this.filter = scan.getFilter();
154     this.earliestPutTs = earliestPutTs;
155     this.maxReadPointToTrackVersions = readPointToUse;
156     this.timeToPurgeDeletes = scanInfo.getTimeToPurgeDeletes();
157 
158     /* how to deal with deletes */
159     this.isUserScan = scanType == ScanType.USER_SCAN;
160     // keep deleted cells: if compaction or raw scan
161     this.keepDeletedCells = (scanInfo.getKeepDeletedCells() && !isUserScan) || scan.isRaw();
162     // retain deletes: if minor compaction or raw scan
163     this.retainDeletesInOutput = scanType == ScanType.COMPACT_RETAIN_DELETES || scan.isRaw();
164     // seePastDeleteMarker: user initiated scans
165     this.seePastDeleteMarkers = scanInfo.getKeepDeletedCells() && isUserScan;
166 
167     int maxVersions = Math.min(scan.getMaxVersions(), scanInfo.getMaxVersions());
168     // Single branch to deal with two types of reads (columns vs all in family)
169     if (columns == null || columns.size() == 0) {
170       // there is always a null column in the wildcard column query.
171       hasNullColumn = true;
172 
173       // use a specialized scan for wildcard column tracker.
174       this.columns = new ScanWildcardColumnTracker(
175           scanInfo.getMinVersions(), maxVersions, oldestUnexpiredTS);
176     } else {
177       // whether there is null column in the explicit column query
178       hasNullColumn = (columns.first().length == 0);
179 
180       // We can share the ExplicitColumnTracker, diff is we reset
181       // between rows, not between storefiles.
182       this.columns = new ExplicitColumnTracker(columns,
183           scanInfo.getMinVersions(), maxVersions, oldestUnexpiredTS);
184     }
185   }
186 
187   /**
188    * Construct a QueryMatcher for a scan that drop deletes from a limited range of rows.
189    * @param scan
190    * @param scanInfo The store's immutable scan info
191    * @param columns
192    * @param earliestPutTs Earliest put seen in any of the store files.
193    * @param oldestUnexpiredTS the oldest timestamp we are interested in,
194    *  based on TTL
195    * @param dropDeletesFromRow The inclusive left bound of the range; can be EMPTY_START_ROW.
196    * @param dropDeletesToRow The exclusive right bound of the range; can be EMPTY_END_ROW.
197    */
198   public ScanQueryMatcher(Scan scan, ScanInfo scanInfo, NavigableSet<byte[]> columns,
199       long readPointToUse, long earliestPutTs, long oldestUnexpiredTS,
200       byte[] dropDeletesFromRow, byte[] dropDeletesToRow) {
201     this(scan, scanInfo, columns, ScanType.COMPACT_RETAIN_DELETES, readPointToUse, earliestPutTs,
202         oldestUnexpiredTS);
203     Preconditions.checkArgument((dropDeletesFromRow != null) && (dropDeletesToRow != null));
204     this.dropDeletesFromRow = dropDeletesFromRow;
205     this.dropDeletesToRow = dropDeletesToRow;
206   }
207 
208   /*
209    * Constructor for tests
210    */
211   ScanQueryMatcher(Scan scan, ScanInfo scanInfo,
212       NavigableSet<byte[]> columns, long oldestUnexpiredTS) {
213     this(scan, scanInfo, columns, ScanType.USER_SCAN,
214           Long.MAX_VALUE, /* max Readpoint to track versions */
215         HConstants.LATEST_TIMESTAMP, oldestUnexpiredTS);
216   }
217 
218   /**
219    *
220    * @return  whether there is an null column in the query
221    */
222   public boolean hasNullColumnInQuery() {
223     return hasNullColumn;
224   }
225 
226   /**
227    * Determines if the caller should do one of several things:
228    * - seek/skip to the next row (MatchCode.SEEK_NEXT_ROW)
229    * - seek/skip to the next column (MatchCode.SEEK_NEXT_COL)
230    * - include the current KeyValue (MatchCode.INCLUDE)
231    * - ignore the current KeyValue (MatchCode.SKIP)
232    * - got to the next row (MatchCode.DONE)
233    *
234    * @param kv KeyValue to check
235    * @return The match code instance.
236    * @throws IOException in case there is an internal consistency problem
237    *      caused by a data corruption.
238    */
239   public MatchCode match(KeyValue kv) throws IOException {
240     if (filter != null && filter.filterAllRemaining()) {
241       return MatchCode.DONE_SCAN;
242     }
243 
244     byte [] bytes = kv.getBuffer();
245     int offset = kv.getOffset();
246     int initialOffset = offset;
247 
248     int keyLength = Bytes.toInt(bytes, offset, Bytes.SIZEOF_INT);
249     offset += KeyValue.ROW_OFFSET;
250 
251     short rowLength = Bytes.toShort(bytes, offset, Bytes.SIZEOF_SHORT);
252     offset += Bytes.SIZEOF_SHORT;
253 
254     int ret = this.rowComparator.compareRows(row, this.rowOffset, this.rowLength,
255         bytes, offset, rowLength);
256     if (ret <= -1) {
257       return MatchCode.DONE;
258     } else if (ret >= 1) {
259       // could optimize this, if necessary?
260       // Could also be called SEEK_TO_CURRENT_ROW, but this
261       // should be rare/never happens.
262       return MatchCode.SEEK_NEXT_ROW;
263     }
264 
265     // optimize case.
266     if (this.stickyNextRow)
267         return MatchCode.SEEK_NEXT_ROW;
268 
269     if (this.columns.done()) {
270       stickyNextRow = true;
271       return MatchCode.SEEK_NEXT_ROW;
272     }
273 
274     //Passing rowLength
275     offset += rowLength;
276 
277     //Skipping family
278     byte familyLength = bytes [offset];
279     offset += familyLength + 1;
280 
281     int qualLength = keyLength + KeyValue.ROW_OFFSET -
282       (offset - initialOffset) - KeyValue.TIMESTAMP_TYPE_SIZE;
283 
284     long timestamp = kv.getTimestamp();
285     // check for early out based on timestamp alone
286     if (columns.isDone(timestamp)) {
287         return columns.getNextRowOrNextColumn(bytes, offset, qualLength);
288     }
289 
290     /*
291      * The delete logic is pretty complicated now.
292      * This is corroborated by the following:
293      * 1. The store might be instructed to keep deleted rows around.
294      * 2. A scan can optionally see past a delete marker now.
295      * 3. If deleted rows are kept, we have to find out when we can
296      *    remove the delete markers.
297      * 4. Family delete markers are always first (regardless of their TS)
298      * 5. Delete markers should not be counted as version
299      * 6. Delete markers affect puts of the *same* TS
300      * 7. Delete marker need to be version counted together with puts
301      *    they affect
302      */
303     byte type = kv.getType();
304     if (kv.isDelete()) {
305       if (!keepDeletedCells) {
306         // first ignore delete markers if the scanner can do so, and the
307         // range does not include the marker
308         //
309         // during flushes and compactions also ignore delete markers newer
310         // than the readpoint of any open scanner, this prevents deleted
311         // rows that could still be seen by a scanner from being collected
312         boolean includeDeleteMarker = seePastDeleteMarkers ?
313             tr.withinTimeRange(timestamp) :
314             tr.withinOrAfterTimeRange(timestamp);
315         if (includeDeleteMarker
316             && kv.getMemstoreTS() <= maxReadPointToTrackVersions) {
317           this.deletes.add(bytes, offset, qualLength, timestamp, type);
318         }
319         // Can't early out now, because DelFam come before any other keys
320       }
321       if (retainDeletesInOutput
322           || (!isUserScan && (EnvironmentEdgeManager.currentTimeMillis() - timestamp) <= timeToPurgeDeletes)
323           || kv.getMemstoreTS() > maxReadPointToTrackVersions) {
324         // always include or it is not time yet to check whether it is OK
325         // to purge deltes or not
326         return MatchCode.INCLUDE;
327       } else if (keepDeletedCells) {
328         if (timestamp < earliestPutTs) {
329           // keeping delete rows, but there are no puts older than
330           // this delete in the store files.
331           return columns.getNextRowOrNextColumn(bytes, offset, qualLength);
332         }
333         // else: fall through and do version counting on the
334         // delete markers
335       } else {
336         return MatchCode.SKIP;
337       }
338       // note the following next else if...
339       // delete marker are not subject to other delete markers
340     } else if (!this.deletes.isEmpty()) {
341       DeleteResult deleteResult = deletes.isDeleted(bytes, offset, qualLength,
342           timestamp);
343       switch (deleteResult) {
344         case FAMILY_DELETED:
345         case COLUMN_DELETED:
346           return columns.getNextRowOrNextColumn(bytes, offset, qualLength);
347         case VERSION_DELETED:
348           return MatchCode.SKIP;
349         case NOT_DELETED:
350           break;
351         default:
352           throw new RuntimeException("UNEXPECTED");
353         }
354     }
355 
356     int timestampComparison = tr.compare(timestamp);
357     if (timestampComparison >= 1) {
358       return MatchCode.SKIP;
359     } else if (timestampComparison <= -1) {
360       return columns.getNextRowOrNextColumn(bytes, offset, qualLength);
361     }
362 
363     /**
364      * Filters should be checked before checking column trackers. If we do
365      * otherwise, as was previously being done, ColumnTracker may increment its
366      * counter for even that KV which may be discarded later on by Filter. This
367      * would lead to incorrect results in certain cases.
368      */
369     ReturnCode filterResponse = ReturnCode.SKIP;
370     if (filter != null) {
371       filterResponse = filter.filterKeyValue(kv);
372       if (filterResponse == ReturnCode.SKIP) {
373         return MatchCode.SKIP;
374       } else if (filterResponse == ReturnCode.NEXT_COL) {
375         return columns.getNextRowOrNextColumn(bytes, offset, qualLength);
376       } else if (filterResponse == ReturnCode.NEXT_ROW) {
377         stickyNextRow = true;
378         return MatchCode.SEEK_NEXT_ROW;
379       } else if (filterResponse == ReturnCode.SEEK_NEXT_USING_HINT) {
380         return MatchCode.SEEK_NEXT_USING_HINT;
381       }
382     }
383 
384     MatchCode colChecker = columns.checkColumn(bytes, offset, qualLength,
385         timestamp, type, kv.getMemstoreTS() > maxReadPointToTrackVersions);
386     /*
387      * According to current implementation, colChecker can only be
388      * SEEK_NEXT_COL, SEEK_NEXT_ROW, SKIP or INCLUDE. Therefore, always return
389      * the MatchCode. If it is SEEK_NEXT_ROW, also set stickyNextRow.
390      */
391     if (colChecker == MatchCode.SEEK_NEXT_ROW) {
392       stickyNextRow = true;
393     } else if (filter != null && colChecker == MatchCode.INCLUDE &&
394                filterResponse == ReturnCode.INCLUDE_AND_NEXT_COL) {
395       return MatchCode.INCLUDE_AND_SEEK_NEXT_COL;
396     }
397     return colChecker;
398 
399   }
400 
401   /** Handle partial-drop-deletes. As we match keys in order, when we have a range from which
402    * we can drop deletes, we can set retainDeletesInOutput to false for the duration of this
403    * range only, and maintain consistency. */
404   private void checkPartialDropDeleteRange(byte [] row, int offset, short length) {
405     // If partial-drop-deletes are used, initially, dropDeletesFromRow and dropDeletesToRow
406     // are both set, and the matcher is set to retain deletes. We assume ordered keys. When
407     // dropDeletesFromRow is leq current kv, we start dropping deletes and reset
408     // dropDeletesFromRow; thus the 2nd "if" starts to apply.
409     if ((dropDeletesFromRow != null)
410         && ((dropDeletesFromRow == HConstants.EMPTY_START_ROW)
411           || (Bytes.compareTo(row, offset, length,
412               dropDeletesFromRow, 0, dropDeletesFromRow.length) >= 0))) {
413       retainDeletesInOutput = false;
414       dropDeletesFromRow = null;
415     }
416     // If dropDeletesFromRow is null and dropDeletesToRow is set, we are inside the partial-
417     // drop-deletes range. When dropDeletesToRow is leq current kv, we stop dropping deletes,
418     // and reset dropDeletesToRow so that we don't do any more compares.
419     if ((dropDeletesFromRow == null)
420         && (dropDeletesToRow != null) && (dropDeletesToRow != HConstants.EMPTY_END_ROW)
421         && (Bytes.compareTo(row, offset, length,
422             dropDeletesToRow, 0, dropDeletesToRow.length) >= 0)) {
423       retainDeletesInOutput = true;
424       dropDeletesToRow = null;
425     }
426   }
427 
428   public boolean moreRowsMayExistAfter(KeyValue kv) {
429     if (!Bytes.equals(stopRow , HConstants.EMPTY_END_ROW) &&
430         rowComparator.compareRows(kv.getBuffer(),kv.getRowOffset(),
431             kv.getRowLength(), stopRow, 0, stopRow.length) >= 0) {
432       // KV >= STOPROW
433       // then NO there is nothing left.
434       return false;
435     } else {
436       return true;
437     }
438   }
439 
440   /**
441    * Set current row
442    * @param row
443    */
444   public void setRow(byte [] row, int offset, short length) {
445     checkPartialDropDeleteRange(row, offset, length);
446     this.row = row;
447     this.rowOffset = offset;
448     this.rowLength = length;
449     reset();
450   }
451 
452   public void reset() {
453     this.deletes.reset();
454     this.columns.reset();
455 
456     stickyNextRow = false;
457   }
458 
459   /**
460    *
461    * @return the start key
462    */
463   public KeyValue getStartKey() {
464     return this.startKey;
465   }
466 
467   /**
468    *
469    * @return the Filter
470    */
471   Filter getFilter() {
472     return this.filter;
473   }
474 
475   public KeyValue getNextKeyHint(KeyValue kv) throws IOException {
476     if (filter == null) {
477       return null;
478     } else {
479       return filter.getNextKeyHint(kv);
480     }
481   }
482 
483   public KeyValue getKeyForNextColumn(KeyValue kv) {
484     ColumnCount nextColumn = columns.getColumnHint();
485     if (nextColumn == null) {
486       return KeyValue.createLastOnRow(
487           kv.getBuffer(), kv.getRowOffset(), kv.getRowLength(),
488           kv.getBuffer(), kv.getFamilyOffset(), kv.getFamilyLength(),
489           kv.getBuffer(), kv.getQualifierOffset(), kv.getQualifierLength());
490     } else {
491       return KeyValue.createFirstOnRow(
492           kv.getBuffer(), kv.getRowOffset(), kv.getRowLength(),
493           kv.getBuffer(), kv.getFamilyOffset(), kv.getFamilyLength(),
494           nextColumn.getBuffer(), nextColumn.getOffset(), nextColumn.getLength());
495     }
496   }
497 
498   public KeyValue getKeyForNextRow(KeyValue kv) {
499     return KeyValue.createLastOnRow(
500         kv.getBuffer(), kv.getRowOffset(), kv.getRowLength(),
501         null, 0, 0,
502         null, 0, 0);
503   }
504 
505   /**
506    * {@link #match} return codes.  These instruct the scanner moving through
507    * memstores and StoreFiles what to do with the current KeyValue.
508    * <p>
509    * Additionally, this contains "early-out" language to tell the scanner to
510    * move on to the next File (memstore or Storefile), or to return immediately.
511    */
512   public static enum MatchCode {
513     /**
514      * Include KeyValue in the returned result
515      */
516     INCLUDE,
517 
518     /**
519      * Do not include KeyValue in the returned result
520      */
521     SKIP,
522 
523     /**
524      * Do not include, jump to next StoreFile or memstore (in time order)
525      */
526     NEXT,
527 
528     /**
529      * Do not include, return current result
530      */
531     DONE,
532 
533     /**
534      * These codes are used by the ScanQueryMatcher
535      */
536 
537     /**
538      * Done with the row, seek there.
539      */
540     SEEK_NEXT_ROW,
541     /**
542      * Done with column, seek to next.
543      */
544     SEEK_NEXT_COL,
545 
546     /**
547      * Done with scan, thanks to the row filter.
548      */
549     DONE_SCAN,
550 
551     /*
552      * Seek to next key which is given as hint.
553      */
554     SEEK_NEXT_USING_HINT,
555 
556     /**
557      * Include KeyValue and done with column, seek to next.
558      */
559     INCLUDE_AND_SEEK_NEXT_COL,
560 
561     /**
562      * Include KeyValue and done with row, seek to next.
563      */
564     INCLUDE_AND_SEEK_NEXT_ROW,
565   }
566 }