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.Comparator;
24  import java.util.List;
25  import java.util.PriorityQueue;
26  
27  import org.apache.hadoop.classification.InterfaceAudience;
28  import org.apache.hadoop.hbase.Cell;
29  import org.apache.hadoop.hbase.KeyValue;
30  import org.apache.hadoop.hbase.KeyValue.KVComparator;
31  
32  /**
33   * Implements a heap merge across any number of KeyValueScanners.
34   * <p>
35   * Implements KeyValueScanner itself.
36   * <p>
37   * This class is used at the Region level to merge across Stores
38   * and at the Store level to merge across the memstore and StoreFiles.
39   * <p>
40   * In the Region case, we also need InternalScanner.next(List), so this class
41   * also implements InternalScanner.  WARNING: As is, if you try to use this
42   * as an InternalScanner at the Store level, you will get runtime exceptions.
43   */
44  @InterfaceAudience.Private
45  public class KeyValueHeap extends NonReversedNonLazyKeyValueScanner
46      implements KeyValueScanner, InternalScanner {
47    protected PriorityQueue<KeyValueScanner> heap = null;
48  
49    /**
50     * The current sub-scanner, i.e. the one that contains the next key/value
51     * to return to the client. This scanner is NOT included in {@link #heap}
52     * (but we frequently add it back to the heap and pull the new winner out).
53     * We maintain an invariant that the current sub-scanner has already done
54     * a real seek, and that current.peek() is always a real key/value (or null)
55     * except for the fake last-key-on-row-column supplied by the multi-column
56     * Bloom filter optimization, which is OK to propagate to StoreScanner. In
57     * order to ensure that, always use {@link #pollRealKV()} to update current.
58     */
59    protected KeyValueScanner current = null;
60  
61    protected KVScannerComparator comparator;
62    
63    /**
64     * Constructor.  This KeyValueHeap will handle closing of passed in
65     * KeyValueScanners.
66     * @param scanners
67     * @param comparator
68     */
69    public KeyValueHeap(List<? extends KeyValueScanner> scanners,
70        KVComparator comparator) throws IOException {
71      this(scanners, new KVScannerComparator(comparator));
72    }
73  
74    /**
75     * Constructor.
76     * @param scanners
77     * @param comparator
78     * @throws IOException
79     */
80    KeyValueHeap(List<? extends KeyValueScanner> scanners,
81        KVScannerComparator comparator) throws IOException {
82      this.comparator = comparator;
83      if (!scanners.isEmpty()) {
84        this.heap = new PriorityQueue<KeyValueScanner>(scanners.size(),
85            this.comparator);
86        for (KeyValueScanner scanner : scanners) {
87          if (scanner.peek() != null) {
88            this.heap.add(scanner);
89          } else {
90            scanner.close();
91          }
92        }
93        this.current = pollRealKV();
94      }
95    }
96  
97    public KeyValue peek() {
98      if (this.current == null) {
99        return null;
100     }
101     return this.current.peek();
102   }
103 
104   public KeyValue next()  throws IOException {
105     if(this.current == null) {
106       return null;
107     }
108     KeyValue kvReturn = this.current.next();
109     KeyValue kvNext = this.current.peek();
110     if (kvNext == null) {
111       this.current.close();
112       this.current = pollRealKV();
113     } else {
114       KeyValueScanner topScanner = this.heap.peek();
115       if (topScanner == null ||
116           this.comparator.compare(kvNext, topScanner.peek()) >= 0) {
117         this.heap.add(this.current);
118         this.current = pollRealKV();
119       }
120     }
121     return kvReturn;
122   }
123 
124   /**
125    * Gets the next row of keys from the top-most scanner.
126    * <p>
127    * This method takes care of updating the heap.
128    * <p>
129    * This can ONLY be called when you are using Scanners that implement
130    * InternalScanner as well as KeyValueScanner (a {@link StoreScanner}).
131    * @param result
132    * @param limit
133    * @return true if there are more keys, false if all scanners are done
134    */
135   public boolean next(List<Cell> result, int limit) throws IOException {
136     if (this.current == null) {
137       return false;
138     }
139     InternalScanner currentAsInternal = (InternalScanner)this.current;
140     boolean mayContainMoreRows = currentAsInternal.next(result, limit);
141     KeyValue pee = this.current.peek();
142     /*
143      * By definition, any InternalScanner must return false only when it has no
144      * further rows to be fetched. So, we can close a scanner if it returns
145      * false. All existing implementations seem to be fine with this. It is much
146      * more efficient to close scanners which are not needed than keep them in
147      * the heap. This is also required for certain optimizations.
148      */
149     if (pee == null || !mayContainMoreRows) {
150       this.current.close();
151     } else {
152       this.heap.add(this.current);
153     }
154     this.current = pollRealKV();
155     return (this.current != null);
156   }
157 
158   /**
159    * Gets the next row of keys from the top-most scanner.
160    * <p>
161    * This method takes care of updating the heap.
162    * <p>
163    * This can ONLY be called when you are using Scanners that implement
164    * InternalScanner as well as KeyValueScanner (a {@link StoreScanner}).
165    * @param result
166    * @return true if there are more keys, false if all scanners are done
167    */
168   public boolean next(List<Cell> result) throws IOException {
169     return next(result, -1);
170   }
171 
172   protected static class KVScannerComparator implements Comparator<KeyValueScanner> {
173     protected KVComparator kvComparator;
174     /**
175      * Constructor
176      * @param kvComparator
177      */
178     public KVScannerComparator(KVComparator kvComparator) {
179       this.kvComparator = kvComparator;
180     }
181     public int compare(KeyValueScanner left, KeyValueScanner right) {
182       int comparison = compare(left.peek(), right.peek());
183       if (comparison != 0) {
184         return comparison;
185       } else {
186         // Since both the keys are exactly the same, we break the tie in favor
187         // of the key which came latest.
188         long leftSequenceID = left.getSequenceID();
189         long rightSequenceID = right.getSequenceID();
190         if (leftSequenceID > rightSequenceID) {
191           return -1;
192         } else if (leftSequenceID < rightSequenceID) {
193           return 1;
194         } else {
195           return 0;
196         }
197       }
198     }
199     /**
200      * Compares two KeyValue
201      * @param left
202      * @param right
203      * @return less than 0 if left is smaller, 0 if equal etc..
204      */
205     public int compare(KeyValue left, KeyValue right) {
206       return this.kvComparator.compare(left, right);
207     }
208     /**
209      * @return KVComparator
210      */
211     public KVComparator getComparator() {
212       return this.kvComparator;
213     }
214   }
215 
216   public void close() {
217     if (this.current != null) {
218       this.current.close();
219     }
220     if (this.heap != null) {
221       KeyValueScanner scanner;
222       while ((scanner = this.heap.poll()) != null) {
223         scanner.close();
224       }
225     }
226   }
227 
228   /**
229    * Seeks all scanners at or below the specified seek key.  If we earlied-out
230    * of a row, we may end up skipping values that were never reached yet.
231    * Rather than iterating down, we want to give the opportunity to re-seek.
232    * <p>
233    * As individual scanners may run past their ends, those scanners are
234    * automatically closed and removed from the heap.
235    * <p>
236    * This function (and {@link #reseek(KeyValue)}) does not do multi-column
237    * Bloom filter and lazy-seek optimizations. To enable those, call
238    * {@link #requestSeek(KeyValue, boolean, boolean)}.
239    * @param seekKey KeyValue to seek at or after
240    * @return true if KeyValues exist at or after specified key, false if not
241    * @throws IOException
242    */
243   @Override
244   public boolean seek(KeyValue seekKey) throws IOException {
245     return generalizedSeek(false,    // This is not a lazy seek
246                            seekKey,
247                            false,    // forward (false: this is not a reseek)
248                            false);   // Not using Bloom filters
249   }
250 
251   /**
252    * This function is identical to the {@link #seek(KeyValue)} function except
253    * that scanner.seek(seekKey) is changed to scanner.reseek(seekKey).
254    */
255   @Override
256   public boolean reseek(KeyValue seekKey) throws IOException {
257     return generalizedSeek(false,    // This is not a lazy seek
258                            seekKey,
259                            true,     // forward (true because this is reseek)
260                            false);   // Not using Bloom filters
261   }
262 
263   /**
264    * {@inheritDoc}
265    */
266   @Override
267   public boolean requestSeek(KeyValue key, boolean forward,
268       boolean useBloom) throws IOException {
269     return generalizedSeek(true, key, forward, useBloom);
270   }
271 
272   /**
273    * @param isLazy whether we are trying to seek to exactly the given row/col.
274    *          Enables Bloom filter and most-recent-file-first optimizations for
275    *          multi-column get/scan queries.
276    * @param seekKey key to seek to
277    * @param forward whether to seek forward (also known as reseek)
278    * @param useBloom whether to optimize seeks using Bloom filters
279    */
280   private boolean generalizedSeek(boolean isLazy, KeyValue seekKey,
281       boolean forward, boolean useBloom) throws IOException {
282     if (!isLazy && useBloom) {
283       throw new IllegalArgumentException("Multi-column Bloom filter " +
284           "optimization requires a lazy seek");
285     }
286 
287     if (current == null) {
288       return false;
289     }
290     heap.add(current);
291     current = null;
292 
293     KeyValueScanner scanner;
294     while ((scanner = heap.poll()) != null) {
295       KeyValue topKey = scanner.peek();
296       if (comparator.getComparator().compare(seekKey, topKey) <= 0) {
297         // Top KeyValue is at-or-after Seek KeyValue. We only know that all
298         // scanners are at or after seekKey (because fake keys of
299         // scanners where a lazy-seek operation has been done are not greater
300         // than their real next keys) but we still need to enforce our
301         // invariant that the top scanner has done a real seek. This way
302         // StoreScanner and RegionScanner do not have to worry about fake keys.
303         heap.add(scanner);
304         current = pollRealKV();
305         return current != null;
306       }
307 
308       boolean seekResult;
309       if (isLazy && heap.size() > 0) {
310         // If there is only one scanner left, we don't do lazy seek.
311         seekResult = scanner.requestSeek(seekKey, forward, useBloom);
312       } else {
313         seekResult = NonLazyKeyValueScanner.doRealSeek(
314             scanner, seekKey, forward);
315       }
316 
317       if (!seekResult) {
318         scanner.close();
319       } else {
320         heap.add(scanner);
321       }
322     }
323 
324     // Heap is returning empty, scanner is done
325     return false;
326   }
327 
328   /**
329    * Fetches the top sub-scanner from the priority queue, ensuring that a real
330    * seek has been done on it. Works by fetching the top sub-scanner, and if it
331    * has not done a real seek, making it do so (which will modify its top KV),
332    * putting it back, and repeating this until success. Relies on the fact that
333    * on a lazy seek we set the current key of a StoreFileScanner to a KV that
334    * is not greater than the real next KV to be read from that file, so the
335    * scanner that bubbles up to the top of the heap will have global next KV in
336    * this scanner heap if (1) it has done a real seek and (2) its KV is the top
337    * among all top KVs (some of which are fake) in the scanner heap.
338    */
339   protected KeyValueScanner pollRealKV() throws IOException {
340     KeyValueScanner kvScanner = heap.poll();
341     if (kvScanner == null) {
342       return null;
343     }
344 
345     while (kvScanner != null && !kvScanner.realSeekDone()) {
346       if (kvScanner.peek() != null) {
347         kvScanner.enforceSeek();
348         KeyValue curKV = kvScanner.peek();
349         if (curKV != null) {
350           KeyValueScanner nextEarliestScanner = heap.peek();
351           if (nextEarliestScanner == null) {
352             // The heap is empty. Return the only possible scanner.
353             return kvScanner;
354           }
355 
356           // Compare the current scanner to the next scanner. We try to avoid
357           // putting the current one back into the heap if possible.
358           KeyValue nextKV = nextEarliestScanner.peek();
359           if (nextKV == null || comparator.compare(curKV, nextKV) < 0) {
360             // We already have the scanner with the earliest KV, so return it.
361             return kvScanner;
362           }
363 
364           // Otherwise, put the scanner back into the heap and let it compete
365           // against all other scanners (both those that have done a "real
366           // seek" and a "lazy seek").
367           heap.add(kvScanner);
368         } else {
369           // Close the scanner because we did a real seek and found out there
370           // are no more KVs.
371           kvScanner.close();
372         }
373       } else {
374         // Close the scanner because it has already run out of KVs even before
375         // we had to do a real seek on it.
376         kvScanner.close();
377       }
378       kvScanner = heap.poll();
379     }
380 
381     return kvScanner;
382   }
383 
384   /**
385    * @return the current Heap
386    */
387   public PriorityQueue<KeyValueScanner> getHeap() {
388     return this.heap;
389   }
390 
391   @Override
392   public long getSequenceID() {
393     return 0;
394   }
395 
396   KeyValueScanner getCurrentForTesting() {
397     return current;
398   }
399 }