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  package org.apache.hadoop.hbase.io.hfile;
20  
21  import static org.apache.hadoop.hbase.io.compress.Compression.Algorithm.GZ;
22  import static org.apache.hadoop.hbase.io.compress.Compression.Algorithm.NONE;
23  import static org.junit.Assert.*;
24  
25  import java.io.ByteArrayOutputStream;
26  import java.io.DataOutputStream;
27  import java.io.IOException;
28  import java.io.OutputStream;
29  import java.nio.ByteBuffer;
30  import java.util.ArrayList;
31  import java.util.Collection;
32  import java.util.Collections;
33  import java.util.HashMap;
34  import java.util.List;
35  import java.util.Map;
36  import java.util.Random;
37  import java.util.concurrent.Callable;
38  import java.util.concurrent.ExecutionException;
39  import java.util.concurrent.Executor;
40  import java.util.concurrent.ExecutorCompletionService;
41  import java.util.concurrent.Executors;
42  import java.util.concurrent.Future;
43  
44  import org.apache.commons.logging.Log;
45  import org.apache.commons.logging.LogFactory;
46  import org.apache.hadoop.fs.FSDataInputStream;
47  import org.apache.hadoop.fs.FSDataOutputStream;
48  import org.apache.hadoop.fs.FileSystem;
49  import org.apache.hadoop.fs.Path;
50  import org.apache.hadoop.hbase.HBaseTestingUtility;
51  import org.apache.hadoop.hbase.HConstants;
52  import org.apache.hadoop.hbase.KeyValue;
53  import org.apache.hadoop.hbase.testclassification.MediumTests;
54  import org.apache.hadoop.hbase.Tag;
55  import org.apache.hadoop.hbase.fs.HFileSystem;
56  import org.apache.hadoop.hbase.io.compress.Compression;
57  import org.apache.hadoop.hbase.io.compress.Compression.Algorithm;
58  import org.apache.hadoop.hbase.io.encoding.DataBlockEncoding;
59  import org.apache.hadoop.hbase.util.Bytes;
60  import org.apache.hadoop.hbase.util.ChecksumType;
61  import org.apache.hadoop.hbase.util.ClassSize;
62  import org.apache.hadoop.io.WritableUtils;
63  import org.apache.hadoop.io.compress.Compressor;
64  import org.junit.Before;
65  import org.junit.Test;
66  import org.junit.experimental.categories.Category;
67  import org.junit.runner.RunWith;
68  import org.junit.runners.Parameterized;
69  import org.junit.runners.Parameterized.Parameters;
70  import org.mockito.Mockito;
71  
72  @Category(MediumTests.class)
73  @RunWith(Parameterized.class)
74  public class TestHFileBlock {
75    // change this value to activate more logs
76    private static final boolean detailedLogging = false;
77    private static final boolean[] BOOLEAN_VALUES = new boolean[] { false, true };
78  
79    private static final Log LOG = LogFactory.getLog(TestHFileBlock.class);
80  
81    static final Compression.Algorithm[] COMPRESSION_ALGORITHMS = { NONE, GZ };
82  
83    private static final int NUM_TEST_BLOCKS = 1000;
84    private static final int NUM_READER_THREADS = 26;
85  
86    // Used to generate KeyValues
87    private static int NUM_KEYVALUES = 50;
88    private static int FIELD_LENGTH = 10;
89    private static float CHANCE_TO_REPEAT = 0.6f;
90  
91    private static final HBaseTestingUtility TEST_UTIL = new HBaseTestingUtility();
92    private FileSystem fs;
93  
94    private final boolean includesMemstoreTS;
95    private final boolean includesTag;
96    public TestHFileBlock(boolean includesMemstoreTS, boolean includesTag) {
97      this.includesMemstoreTS = includesMemstoreTS;
98      this.includesTag = includesTag;
99    }
100 
101   @Parameters
102   public static Collection<Object[]> parameters() {
103     return HBaseTestingUtility.MEMSTORETS_TAGS_PARAMETRIZED;
104   }
105 
106   @Before
107   public void setUp() throws IOException {
108     fs = HFileSystem.get(TEST_UTIL.getConfiguration());
109   }
110 
111   static void writeTestBlockContents(DataOutputStream dos) throws IOException {
112     // This compresses really well.
113     for (int i = 0; i < 1000; ++i)
114       dos.writeInt(i / 100);
115   }
116 
117   static int writeTestKeyValues(HFileBlock.Writer hbw, int seed, boolean includesMemstoreTS,
118       boolean useTag) throws IOException {
119     List<KeyValue> keyValues = new ArrayList<KeyValue>();
120     Random randomizer = new Random(42l + seed); // just any fixed number
121 
122     // generate keyValues
123     for (int i = 0; i < NUM_KEYVALUES; ++i) {
124       byte[] row;
125       long timestamp;
126       byte[] family;
127       byte[] qualifier;
128       byte[] value;
129 
130       // generate it or repeat, it should compress well
131       if (0 < i && randomizer.nextFloat() < CHANCE_TO_REPEAT) {
132         row = keyValues.get(randomizer.nextInt(keyValues.size())).getRow();
133       } else {
134         row = new byte[FIELD_LENGTH];
135         randomizer.nextBytes(row);
136       }
137       if (0 == i) {
138         family = new byte[FIELD_LENGTH];
139         randomizer.nextBytes(family);
140       } else {
141         family = keyValues.get(0).getFamily();
142       }
143       if (0 < i && randomizer.nextFloat() < CHANCE_TO_REPEAT) {
144         qualifier = keyValues.get(
145             randomizer.nextInt(keyValues.size())).getQualifier();
146       } else {
147         qualifier = new byte[FIELD_LENGTH];
148         randomizer.nextBytes(qualifier);
149       }
150       if (0 < i && randomizer.nextFloat() < CHANCE_TO_REPEAT) {
151         value = keyValues.get(randomizer.nextInt(keyValues.size())).getValue();
152       } else {
153         value = new byte[FIELD_LENGTH];
154         randomizer.nextBytes(value);
155       }
156       if (0 < i && randomizer.nextFloat() < CHANCE_TO_REPEAT) {
157         timestamp = keyValues.get(
158             randomizer.nextInt(keyValues.size())).getTimestamp();
159       } else {
160         timestamp = randomizer.nextLong();
161       }
162       if (!useTag) {
163         keyValues.add(new KeyValue(row, family, qualifier, timestamp, value));
164       } else {
165         keyValues.add(new KeyValue(row, family, qualifier, timestamp, value, new Tag[] { new Tag(
166             (byte) 1, Bytes.toBytes("myTagVal")) }));
167       }
168     }
169 
170     // sort it and write to stream
171     int totalSize = 0;
172     Collections.sort(keyValues, KeyValue.COMPARATOR);
173 
174     for (KeyValue kv : keyValues) {
175       totalSize += kv.getLength();
176       if (includesMemstoreTS) {
177         long memstoreTS = randomizer.nextLong();
178         kv.setSequenceId(memstoreTS);
179         totalSize += WritableUtils.getVIntSize(memstoreTS);
180       }
181       hbw.write(kv);
182     }
183     return totalSize;
184   }
185 
186   public byte[] createTestV1Block(Compression.Algorithm algo)
187       throws IOException {
188     Compressor compressor = algo.getCompressor();
189     ByteArrayOutputStream baos = new ByteArrayOutputStream();
190     OutputStream os = algo.createCompressionStream(baos, compressor, 0);
191     DataOutputStream dos = new DataOutputStream(os);
192     BlockType.META.write(dos); // Let's make this a meta block.
193     writeTestBlockContents(dos);
194     dos.flush();
195     algo.returnCompressor(compressor);
196     return baos.toByteArray();
197   }
198 
199   static HFileBlock.Writer createTestV2Block(Compression.Algorithm algo,
200       boolean includesMemstoreTS, boolean includesTag) throws IOException {
201     final BlockType blockType = BlockType.DATA;
202     HFileContext meta = new HFileContextBuilder()
203                         .withCompression(algo)
204                         .withIncludesMvcc(includesMemstoreTS)
205                         .withIncludesTags(includesTag)
206                         .withBytesPerCheckSum(HFile.DEFAULT_BYTES_PER_CHECKSUM)
207                         .build();
208     HFileBlock.Writer hbw = new HFileBlock.Writer(null, meta);
209     DataOutputStream dos = hbw.startWriting(blockType);
210     writeTestBlockContents(dos);
211     dos.flush();
212     hbw.ensureBlockReady();
213     assertEquals(1000 * 4, hbw.getUncompressedSizeWithoutHeader());
214     hbw.release();
215     return hbw;
216   }
217 
218   public String createTestBlockStr(Compression.Algorithm algo,
219       int correctLength, boolean useTag) throws IOException {
220     HFileBlock.Writer hbw = createTestV2Block(algo, includesMemstoreTS, useTag);
221     byte[] testV2Block = hbw.getHeaderAndDataForTest();
222     int osOffset = HConstants.HFILEBLOCK_HEADER_SIZE + 9;
223     if (testV2Block.length == correctLength) {
224       // Force-set the "OS" field of the gzip header to 3 (Unix) to avoid
225       // variations across operating systems.
226       // See http://www.gzip.org/zlib/rfc-gzip.html for gzip format.
227       // We only make this change when the compressed block length matches.
228       // Otherwise, there are obviously other inconsistencies.
229       testV2Block[osOffset] = 3;
230     }
231     return Bytes.toStringBinary(testV2Block);
232   }
233 
234   @Test
235   public void testNoCompression() throws IOException {
236     CacheConfig cacheConf = Mockito.mock(CacheConfig.class);
237     Mockito.when(cacheConf.isBlockCacheEnabled()).thenReturn(false);
238 
239     HFileBlock block =
240       createTestV2Block(NONE, includesMemstoreTS, false).getBlockForCaching(cacheConf);
241     assertEquals(4000, block.getUncompressedSizeWithoutHeader());
242     assertEquals(4004, block.getOnDiskSizeWithoutHeader());
243     assertTrue(block.isUnpacked());
244   }
245 
246   @Test
247   public void testGzipCompression() throws IOException {
248     final String correctTestBlockStr =
249         "DATABLK*\\x00\\x00\\x00>\\x00\\x00\\x0F\\xA0\\xFF\\xFF\\xFF\\xFF"
250             + "\\xFF\\xFF\\xFF\\xFF"
251             + "\\x01\\x00\\x00@\\x00\\x00\\x00\\x00["
252             // gzip-compressed block: http://www.gzip.org/zlib/rfc-gzip.html
253             + "\\x1F\\x8B"  // gzip magic signature
254             + "\\x08"  // Compression method: 8 = "deflate"
255             + "\\x00"  // Flags
256             + "\\x00\\x00\\x00\\x00"  // mtime
257             + "\\x00"  // XFL (extra flags)
258             // OS (0 = FAT filesystems, 3 = Unix). However, this field
259             // sometimes gets set to 0 on Linux and Mac, so we reset it to 3.
260             // This appears to be a difference caused by the availability
261             // (and use) of the native GZ codec.
262             + "\\x03"
263             + "\\xED\\xC3\\xC1\\x11\\x00 \\x08\\xC00DD\\xDD\\x7Fa"
264             + "\\xD6\\xE8\\xA3\\xB9K\\x84`\\x96Q\\xD3\\xA8\\xDB\\xA8e\\xD4c"
265             + "\\xD46\\xEA5\\xEA3\\xEA7\\xE7\\x00LI\\x5Cs\\xA0\\x0F\\x00\\x00"
266             + "\\x00\\x00\\x00\\x00"; //  4 byte checksum (ignored)
267     final int correctGzipBlockLength = 95;
268     final String testBlockStr = createTestBlockStr(GZ, correctGzipBlockLength, false);
269     // We ignore the block checksum because createTestBlockStr can change the
270     // gzip header after the block is produced
271     assertEquals(correctTestBlockStr.substring(0, correctGzipBlockLength - 4),
272       testBlockStr.substring(0, correctGzipBlockLength - 4));
273   }
274 
275   @Test
276   public void testReaderV2() throws IOException {
277     testReaderV2Internals();
278   }
279 
280   protected void testReaderV2Internals() throws IOException {
281     if(includesTag) {
282       TEST_UTIL.getConfiguration().setInt("hfile.format.version", 3);
283     }
284     for (Compression.Algorithm algo : COMPRESSION_ALGORITHMS) {
285       for (boolean pread : new boolean[] { false, true }) {
286           LOG.info("testReaderV2: Compression algorithm: " + algo +
287                    ", pread=" + pread);
288         Path path = new Path(TEST_UTIL.getDataTestDir(), "blocks_v2_"
289             + algo);
290         FSDataOutputStream os = fs.create(path);
291         HFileContext meta = new HFileContextBuilder()
292                            .withCompression(algo)
293                            .withIncludesMvcc(includesMemstoreTS)
294                            .withIncludesTags(includesTag)
295                            .withBytesPerCheckSum(HFile.DEFAULT_BYTES_PER_CHECKSUM)
296                            .build();
297         HFileBlock.Writer hbw = new HFileBlock.Writer(null,
298            meta);
299         long totalSize = 0;
300         for (int blockId = 0; blockId < 2; ++blockId) {
301           DataOutputStream dos = hbw.startWriting(BlockType.DATA);
302           for (int i = 0; i < 1234; ++i)
303             dos.writeInt(i);
304           hbw.writeHeaderAndData(os);
305           totalSize += hbw.getOnDiskSizeWithHeader();
306         }
307         os.close();
308 
309         FSDataInputStream is = fs.open(path);
310         meta = new HFileContextBuilder()
311         .withHBaseCheckSum(true)
312         .withIncludesMvcc(includesMemstoreTS)
313         .withIncludesTags(includesTag)
314         .withCompression(algo).build();
315         HFileBlock.FSReader hbr = new HFileBlock.FSReaderImpl(is, totalSize, meta);
316         HFileBlock b = hbr.readBlockData(0, -1, -1, pread);
317         is.close();
318         assertEquals(0, HFile.getChecksumFailuresCount());
319 
320         b.sanityCheck();
321         assertEquals(4936, b.getUncompressedSizeWithoutHeader());
322         assertEquals(algo == GZ ? 2173 : 4936,
323                      b.getOnDiskSizeWithoutHeader() - b.totalChecksumBytes());
324         HFileBlock expected = b;
325 
326         if (algo == GZ) {
327           is = fs.open(path);
328           hbr = new HFileBlock.FSReaderImpl(is, totalSize, meta);
329           b = hbr.readBlockData(0, 2173 + HConstants.HFILEBLOCK_HEADER_SIZE +
330                                 b.totalChecksumBytes(), -1, pread);
331           assertEquals(expected, b);
332           int wrongCompressedSize = 2172;
333           try {
334             b = hbr.readBlockData(0, wrongCompressedSize
335                 + HConstants.HFILEBLOCK_HEADER_SIZE, -1, pread);
336             fail("Exception expected");
337           } catch (IOException ex) {
338             String expectedPrefix = "On-disk size without header provided is "
339                 + wrongCompressedSize + ", but block header contains "
340                 + b.getOnDiskSizeWithoutHeader() + ".";
341             assertTrue("Invalid exception message: '" + ex.getMessage()
342                 + "'.\nMessage is expected to start with: '" + expectedPrefix
343                 + "'", ex.getMessage().startsWith(expectedPrefix));
344           }
345           is.close();
346         }
347       }
348     }
349   }
350 
351   /**
352    * Test encoding/decoding data blocks.
353    * @throws IOException a bug or a problem with temporary files.
354    */
355   @Test
356   public void testDataBlockEncoding() throws IOException {
357     testInternals();
358   }
359 
360   private void testInternals() throws IOException {
361     final int numBlocks = 5;
362     if(includesTag) {
363       TEST_UTIL.getConfiguration().setInt("hfile.format.version", 3);
364     }
365     for (Compression.Algorithm algo : COMPRESSION_ALGORITHMS) {
366       for (boolean pread : new boolean[] { false, true }) {
367         for (DataBlockEncoding encoding : DataBlockEncoding.values()) {
368           Path path = new Path(TEST_UTIL.getDataTestDir(), "blocks_v2_"
369               + algo + "_" + encoding.toString());
370           FSDataOutputStream os = fs.create(path);
371           HFileDataBlockEncoder dataBlockEncoder = (encoding != DataBlockEncoding.NONE) ?
372               new HFileDataBlockEncoderImpl(encoding) : NoOpDataBlockEncoder.INSTANCE;
373           HFileContext meta = new HFileContextBuilder()
374                               .withCompression(algo)
375                               .withIncludesMvcc(includesMemstoreTS)
376                               .withIncludesTags(includesTag)
377                               .withBytesPerCheckSum(HFile.DEFAULT_BYTES_PER_CHECKSUM)
378                               .build();
379           HFileBlock.Writer hbw = new HFileBlock.Writer(dataBlockEncoder, meta);
380           long totalSize = 0;
381           final List<Integer> encodedSizes = new ArrayList<Integer>();
382           final List<ByteBuffer> encodedBlocks = new ArrayList<ByteBuffer>();
383           for (int blockId = 0; blockId < numBlocks; ++blockId) {
384             hbw.startWriting(BlockType.DATA);
385             writeTestKeyValues(hbw, blockId, includesMemstoreTS, includesTag);
386             hbw.writeHeaderAndData(os);
387             int headerLen = HConstants.HFILEBLOCK_HEADER_SIZE;
388             byte[] encodedResultWithHeader = hbw.getUncompressedBufferWithHeader().array();
389             final int encodedSize = encodedResultWithHeader.length - headerLen;
390             if (encoding != DataBlockEncoding.NONE) {
391               // We need to account for the two-byte encoding algorithm ID that
392               // comes after the 24-byte block header but before encoded KVs.
393               headerLen += DataBlockEncoding.ID_SIZE;
394             }
395             byte[] encodedDataSection =
396                 new byte[encodedResultWithHeader.length - headerLen];
397             System.arraycopy(encodedResultWithHeader, headerLen,
398                 encodedDataSection, 0, encodedDataSection.length);
399             final ByteBuffer encodedBuf =
400                 ByteBuffer.wrap(encodedDataSection);
401             encodedSizes.add(encodedSize);
402             encodedBlocks.add(encodedBuf);
403             totalSize += hbw.getOnDiskSizeWithHeader();
404           }
405           os.close();
406 
407           FSDataInputStream is = fs.open(path);
408           meta = new HFileContextBuilder()
409                 .withHBaseCheckSum(true)
410                 .withCompression(algo)
411                 .withIncludesMvcc(includesMemstoreTS)
412                 .withIncludesTags(includesTag)
413                 .build();
414           HFileBlock.FSReaderImpl hbr = new HFileBlock.FSReaderImpl(is, totalSize, meta);
415           hbr.setDataBlockEncoder(dataBlockEncoder);
416           hbr.setIncludesMemstoreTS(includesMemstoreTS);
417           HFileBlock blockFromHFile, blockUnpacked;
418           int pos = 0;
419           for (int blockId = 0; blockId < numBlocks; ++blockId) {
420             blockFromHFile = hbr.readBlockData(pos, -1, -1, pread);
421             assertEquals(0, HFile.getChecksumFailuresCount());
422             blockFromHFile.sanityCheck();
423             pos += blockFromHFile.getOnDiskSizeWithHeader();
424             assertEquals((int) encodedSizes.get(blockId),
425               blockFromHFile.getUncompressedSizeWithoutHeader());
426             assertEquals(meta.isCompressedOrEncrypted(), !blockFromHFile.isUnpacked());
427             long packedHeapsize = blockFromHFile.heapSize();
428             blockUnpacked = blockFromHFile.unpack(meta, hbr);
429             assertTrue(blockUnpacked.isUnpacked());
430             if (meta.isCompressedOrEncrypted()) {
431               LOG.info("packedHeapsize=" + packedHeapsize + ", unpackedHeadsize=" + blockUnpacked
432                 .heapSize());
433               assertFalse(packedHeapsize == blockUnpacked.heapSize());
434               assertTrue("Packed heapSize should be < unpacked heapSize",
435                 packedHeapsize < blockUnpacked.heapSize());
436             }
437             ByteBuffer actualBuffer = blockUnpacked.getBufferWithoutHeader();
438             if (encoding != DataBlockEncoding.NONE) {
439               // We expect a two-byte big-endian encoding id.
440               assertEquals(
441                 "Unexpected first byte with " + buildMessageDetails(algo, encoding, pread),
442                 Long.toHexString(0), Long.toHexString(actualBuffer.get(0)));
443               assertEquals(
444                 "Unexpected second byte with " + buildMessageDetails(algo, encoding, pread),
445                 Long.toHexString(encoding.getId()), Long.toHexString(actualBuffer.get(1)));
446               actualBuffer.position(2);
447               actualBuffer = actualBuffer.slice();
448             }
449 
450             ByteBuffer expectedBuffer = encodedBlocks.get(blockId);
451             expectedBuffer.rewind();
452 
453             // test if content matches, produce nice message
454             assertBuffersEqual(expectedBuffer, actualBuffer, algo, encoding, pread);
455 
456             // test serialized blocks
457             for (boolean reuseBuffer : new boolean[] { false, true }) {
458               ByteBuffer serialized = ByteBuffer.allocate(blockFromHFile.getSerializedLength());
459               blockFromHFile.serialize(serialized);
460               HFileBlock deserialized =
461                 (HFileBlock) blockFromHFile.getDeserializer().deserialize(serialized, reuseBuffer);
462               assertEquals(
463                 "Serialization did not preserve block state. reuseBuffer=" + reuseBuffer,
464                 blockFromHFile, deserialized);
465               // intentional reference comparison
466               if (blockFromHFile != blockUnpacked) {
467                 assertEquals("Deserializaed block cannot be unpacked correctly.",
468                   blockUnpacked, deserialized.unpack(meta, hbr));
469               }
470             }
471           }
472           is.close();
473         }
474       }
475     }
476   }
477 
478   static String buildMessageDetails(Algorithm compression, DataBlockEncoding encoding,
479       boolean pread) {
480     return String.format("compression %s, encoding %s, pread %s", compression, encoding, pread);
481   }
482 
483   static void assertBuffersEqual(ByteBuffer expectedBuffer,
484       ByteBuffer actualBuffer, Compression.Algorithm compression,
485       DataBlockEncoding encoding, boolean pread) {
486     if (!actualBuffer.equals(expectedBuffer)) {
487       int prefix = 0;
488       int minLimit = Math.min(expectedBuffer.limit(), actualBuffer.limit());
489       while (prefix < minLimit &&
490           expectedBuffer.get(prefix) == actualBuffer.get(prefix)) {
491         prefix++;
492       }
493 
494       fail(String.format(
495           "Content mismatch for %s, commonPrefix %d, expected %s, got %s",
496           buildMessageDetails(compression, encoding, pread), prefix,
497           nextBytesToStr(expectedBuffer, prefix),
498           nextBytesToStr(actualBuffer, prefix)));
499     }
500   }
501 
502   /**
503    * Convert a few next bytes in the given buffer at the given position to
504    * string. Used for error messages.
505    */
506   private static String nextBytesToStr(ByteBuffer buf, int pos) {
507     int maxBytes = buf.limit() - pos;
508     int numBytes = Math.min(16, maxBytes);
509     return Bytes.toStringBinary(buf.array(), buf.arrayOffset() + pos,
510         numBytes) + (numBytes < maxBytes ? "..." : "");
511   }
512 
513   @Test
514   public void testPreviousOffset() throws IOException {
515     testPreviousOffsetInternals();
516   }
517 
518   protected void testPreviousOffsetInternals() throws IOException {
519     // TODO: parameterize these nested loops.
520     for (Compression.Algorithm algo : COMPRESSION_ALGORITHMS) {
521       for (boolean pread : BOOLEAN_VALUES) {
522         for (boolean cacheOnWrite : BOOLEAN_VALUES) {
523           Random rand = defaultRandom();
524           LOG.info("testPreviousOffset:Compression algorithm: " + algo +
525                    ", pread=" + pread +
526                    ", cacheOnWrite=" + cacheOnWrite);
527           Path path = new Path(TEST_UTIL.getDataTestDir(), "prev_offset");
528           List<Long> expectedOffsets = new ArrayList<Long>();
529           List<Long> expectedPrevOffsets = new ArrayList<Long>();
530           List<BlockType> expectedTypes = new ArrayList<BlockType>();
531           List<ByteBuffer> expectedContents = cacheOnWrite
532               ? new ArrayList<ByteBuffer>() : null;
533           long totalSize = writeBlocks(rand, algo, path, expectedOffsets,
534               expectedPrevOffsets, expectedTypes, expectedContents);
535 
536           FSDataInputStream is = fs.open(path);
537           HFileContext meta = new HFileContextBuilder()
538                               .withHBaseCheckSum(true)
539                               .withIncludesMvcc(includesMemstoreTS)
540                               .withIncludesTags(includesTag)
541                               .withCompression(algo).build();
542           HFileBlock.FSReader hbr = new HFileBlock.FSReaderImpl(is, totalSize, meta);
543           long curOffset = 0;
544           for (int i = 0; i < NUM_TEST_BLOCKS; ++i) {
545             if (!pread) {
546               assertEquals(is.getPos(), curOffset + (i == 0 ? 0 :
547                   HConstants.HFILEBLOCK_HEADER_SIZE));
548             }
549 
550             assertEquals(expectedOffsets.get(i).longValue(), curOffset);
551             if (detailedLogging) {
552               LOG.info("Reading block #" + i + " at offset " + curOffset);
553             }
554             HFileBlock b = hbr.readBlockData(curOffset, -1, -1, pread);
555             if (detailedLogging) {
556               LOG.info("Block #" + i + ": " + b);
557             }
558             assertEquals("Invalid block #" + i + "'s type:",
559                 expectedTypes.get(i), b.getBlockType());
560             assertEquals("Invalid previous block offset for block " + i
561                 + " of " + "type " + b.getBlockType() + ":",
562                 (long) expectedPrevOffsets.get(i), b.getPrevBlockOffset());
563             b.sanityCheck();
564             assertEquals(curOffset, b.getOffset());
565 
566             // Now re-load this block knowing the on-disk size. This tests a
567             // different branch in the loader.
568             HFileBlock b2 = hbr.readBlockData(curOffset,
569                 b.getOnDiskSizeWithHeader(), -1, pread);
570             b2.sanityCheck();
571 
572             assertEquals(b.getBlockType(), b2.getBlockType());
573             assertEquals(b.getOnDiskSizeWithoutHeader(),
574                 b2.getOnDiskSizeWithoutHeader());
575             assertEquals(b.getOnDiskSizeWithHeader(),
576                 b2.getOnDiskSizeWithHeader());
577             assertEquals(b.getUncompressedSizeWithoutHeader(),
578                 b2.getUncompressedSizeWithoutHeader());
579             assertEquals(b.getPrevBlockOffset(), b2.getPrevBlockOffset());
580             assertEquals(curOffset, b2.getOffset());
581             assertEquals(b.getBytesPerChecksum(), b2.getBytesPerChecksum());
582             assertEquals(b.getOnDiskDataSizeWithHeader(),
583                          b2.getOnDiskDataSizeWithHeader());
584             assertEquals(0, HFile.getChecksumFailuresCount());
585 
586             curOffset += b.getOnDiskSizeWithHeader();
587 
588             if (cacheOnWrite) {
589               // NOTE: cache-on-write testing doesn't actually involve a BlockCache. It simply
590               // verifies that the unpacked value read back off disk matches the unpacked value
591               // generated before writing to disk.
592               b = b.unpack(meta, hbr);
593               // b's buffer has header + data + checksum while
594               // expectedContents have header + data only
595               ByteBuffer bufRead = b.getBufferWithHeader();
596               ByteBuffer bufExpected = expectedContents.get(i);
597               boolean bytesAreCorrect = Bytes.compareTo(bufRead.array(),
598                   bufRead.arrayOffset(),
599                   bufRead.limit() - b.totalChecksumBytes(),
600                   bufExpected.array(), bufExpected.arrayOffset(),
601                   bufExpected.limit()) == 0;
602               String wrongBytesMsg = "";
603 
604               if (!bytesAreCorrect) {
605                 // Optimization: only construct an error message in case we
606                 // will need it.
607                 wrongBytesMsg = "Expected bytes in block #" + i + " (algo="
608                     + algo + ", pread=" + pread
609                     + ", cacheOnWrite=" + cacheOnWrite + "):\n";
610                 wrongBytesMsg += Bytes.toStringBinary(bufExpected.array(),
611                   bufExpected.arrayOffset(), Math.min(32 + 10, bufExpected.limit()))
612                     + ", actual:\n"
613                     + Bytes.toStringBinary(bufRead.array(),
614                   bufRead.arrayOffset(), Math.min(32 + 10, bufRead.limit()));
615                 if (detailedLogging) {
616                   LOG.warn("expected header" +
617                            HFileBlock.toStringHeader(bufExpected) +
618                            "\nfound    header" +
619                            HFileBlock.toStringHeader(bufRead));
620                   LOG.warn("bufread offset " + bufRead.arrayOffset() +
621                            " limit " + bufRead.limit() +
622                            " expected offset " + bufExpected.arrayOffset() +
623                            " limit " + bufExpected.limit());
624                   LOG.warn(wrongBytesMsg);
625                 }
626               }
627               assertTrue(wrongBytesMsg, bytesAreCorrect);
628             }
629           }
630 
631           assertEquals(curOffset, fs.getFileStatus(path).getLen());
632           is.close();
633         }
634       }
635     }
636   }
637 
638   private Random defaultRandom() {
639     return new Random(189237);
640   }
641 
642   private class BlockReaderThread implements Callable<Boolean> {
643     private final String clientId;
644     private final HFileBlock.FSReader hbr;
645     private final List<Long> offsets;
646     private final List<BlockType> types;
647     private final long fileSize;
648 
649     public BlockReaderThread(String clientId,
650         HFileBlock.FSReader hbr, List<Long> offsets, List<BlockType> types,
651         long fileSize) {
652       this.clientId = clientId;
653       this.offsets = offsets;
654       this.hbr = hbr;
655       this.types = types;
656       this.fileSize = fileSize;
657     }
658 
659     @Override
660     public Boolean call() throws Exception {
661       Random rand = new Random(clientId.hashCode());
662       long endTime = System.currentTimeMillis() + 10000;
663       int numBlocksRead = 0;
664       int numPositionalRead = 0;
665       int numWithOnDiskSize = 0;
666       while (System.currentTimeMillis() < endTime) {
667         int blockId = rand.nextInt(NUM_TEST_BLOCKS);
668         long offset = offsets.get(blockId);
669         boolean pread = rand.nextBoolean();
670         boolean withOnDiskSize = rand.nextBoolean();
671         long expectedSize =
672           (blockId == NUM_TEST_BLOCKS - 1 ? fileSize
673               : offsets.get(blockId + 1)) - offset;
674 
675         HFileBlock b;
676         try {
677           long onDiskSizeArg = withOnDiskSize ? expectedSize : -1;
678           b = hbr.readBlockData(offset, onDiskSizeArg, -1, pread);
679         } catch (IOException ex) {
680           LOG.error("Error in client " + clientId + " trying to read block at "
681               + offset + ", pread=" + pread + ", withOnDiskSize=" +
682               withOnDiskSize, ex);
683           return false;
684         }
685 
686         assertEquals(types.get(blockId), b.getBlockType());
687         assertEquals(expectedSize, b.getOnDiskSizeWithHeader());
688         assertEquals(offset, b.getOffset());
689 
690         ++numBlocksRead;
691         if (pread)
692           ++numPositionalRead;
693         if (withOnDiskSize)
694           ++numWithOnDiskSize;
695       }
696       LOG.info("Client " + clientId + " successfully read " + numBlocksRead +
697         " blocks (with pread: " + numPositionalRead + ", with onDiskSize " +
698         "specified: " + numWithOnDiskSize + ")");
699 
700       return true;
701     }
702 
703   }
704 
705   @Test
706   public void testConcurrentReading() throws Exception {
707     testConcurrentReadingInternals();
708   }
709 
710   protected void testConcurrentReadingInternals() throws IOException,
711       InterruptedException, ExecutionException {
712     for (Compression.Algorithm compressAlgo : COMPRESSION_ALGORITHMS) {
713       Path path =
714           new Path(TEST_UTIL.getDataTestDir(), "concurrent_reading");
715       Random rand = defaultRandom();
716       List<Long> offsets = new ArrayList<Long>();
717       List<BlockType> types = new ArrayList<BlockType>();
718       writeBlocks(rand, compressAlgo, path, offsets, null, types, null);
719       FSDataInputStream is = fs.open(path);
720       long fileSize = fs.getFileStatus(path).getLen();
721       HFileContext meta = new HFileContextBuilder()
722                           .withHBaseCheckSum(true)
723                           .withIncludesMvcc(includesMemstoreTS)
724                           .withIncludesTags(includesTag)
725                           .withCompression(compressAlgo)
726                           .build();
727       HFileBlock.FSReader hbr = new HFileBlock.FSReaderImpl(is, fileSize, meta);
728 
729       Executor exec = Executors.newFixedThreadPool(NUM_READER_THREADS);
730       ExecutorCompletionService<Boolean> ecs =
731           new ExecutorCompletionService<Boolean>(exec);
732 
733       for (int i = 0; i < NUM_READER_THREADS; ++i) {
734         ecs.submit(new BlockReaderThread("reader_" + (char) ('A' + i), hbr,
735             offsets, types, fileSize));
736       }
737 
738       for (int i = 0; i < NUM_READER_THREADS; ++i) {
739         Future<Boolean> result = ecs.take();
740         assertTrue(result.get());
741         if (detailedLogging) {
742           LOG.info(String.valueOf(i + 1)
743             + " reader threads finished successfully (algo=" + compressAlgo
744             + ")");
745         }
746       }
747 
748       is.close();
749     }
750   }
751 
752   private long writeBlocks(Random rand, Compression.Algorithm compressAlgo,
753       Path path, List<Long> expectedOffsets, List<Long> expectedPrevOffsets,
754       List<BlockType> expectedTypes, List<ByteBuffer> expectedContents
755   ) throws IOException {
756     boolean cacheOnWrite = expectedContents != null;
757     FSDataOutputStream os = fs.create(path);
758     HFileContext meta = new HFileContextBuilder()
759                         .withHBaseCheckSum(true)
760                         .withIncludesMvcc(includesMemstoreTS)
761                         .withIncludesTags(includesTag)
762                         .withCompression(compressAlgo)
763                         .withBytesPerCheckSum(HFile.DEFAULT_BYTES_PER_CHECKSUM)
764                         .build();
765     HFileBlock.Writer hbw = new HFileBlock.Writer(null, meta);
766     Map<BlockType, Long> prevOffsetByType = new HashMap<BlockType, Long>();
767     long totalSize = 0;
768     for (int i = 0; i < NUM_TEST_BLOCKS; ++i) {
769       long pos = os.getPos();
770       int blockTypeOrdinal = rand.nextInt(BlockType.values().length);
771       if (blockTypeOrdinal == BlockType.ENCODED_DATA.ordinal()) {
772         blockTypeOrdinal = BlockType.DATA.ordinal();
773       }
774       BlockType bt = BlockType.values()[blockTypeOrdinal];
775       DataOutputStream dos = hbw.startWriting(bt);
776       int size = rand.nextInt(500);
777       for (int j = 0; j < size; ++j) {
778         // This might compress well.
779         dos.writeShort(i + 1);
780         dos.writeInt(j + 1);
781       }
782 
783       if (expectedOffsets != null)
784         expectedOffsets.add(os.getPos());
785 
786       if (expectedPrevOffsets != null) {
787         Long prevOffset = prevOffsetByType.get(bt);
788         expectedPrevOffsets.add(prevOffset != null ? prevOffset : -1);
789         prevOffsetByType.put(bt, os.getPos());
790       }
791 
792       expectedTypes.add(bt);
793 
794       hbw.writeHeaderAndData(os);
795       totalSize += hbw.getOnDiskSizeWithHeader();
796 
797       if (cacheOnWrite)
798         expectedContents.add(hbw.getUncompressedBufferWithHeader());
799 
800       if (detailedLogging) {
801         LOG.info("Written block #" + i + " of type " + bt
802             + ", uncompressed size " + hbw.getUncompressedSizeWithoutHeader()
803             + ", packed size " + hbw.getOnDiskSizeWithoutHeader()
804             + " at offset " + pos);
805       }
806     }
807     os.close();
808     LOG.info("Created a temporary file at " + path + ", "
809         + fs.getFileStatus(path).getLen() + " byte, compression=" +
810         compressAlgo);
811     return totalSize;
812   }
813 
814   @Test
815   public void testBlockHeapSize() {
816     testBlockHeapSizeInternals();
817   }
818 
819   protected void testBlockHeapSizeInternals() {
820     if (ClassSize.is32BitJVM()) {
821       assertTrue(HFileBlock.BYTE_BUFFER_HEAP_SIZE == 64);
822     } else {
823       assertTrue(HFileBlock.BYTE_BUFFER_HEAP_SIZE == 80);
824     }
825 
826     for (int size : new int[] { 100, 256, 12345 }) {
827       byte[] byteArr = new byte[HConstants.HFILEBLOCK_HEADER_SIZE + size];
828       ByteBuffer buf = ByteBuffer.wrap(byteArr, 0, size);
829       HFileContext meta = new HFileContextBuilder()
830                           .withIncludesMvcc(includesMemstoreTS)
831                           .withIncludesTags(includesTag)
832                           .withHBaseCheckSum(false)
833                           .withCompression(Algorithm.NONE)
834                           .withBytesPerCheckSum(HFile.DEFAULT_BYTES_PER_CHECKSUM)
835                           .withChecksumType(ChecksumType.NULL).build();
836       HFileBlock block = new HFileBlock(BlockType.DATA, size, size, -1, buf,
837           HFileBlock.FILL_HEADER, -1, 
838           0, meta);
839       long byteBufferExpectedSize =
840           ClassSize.align(ClassSize.estimateBase(buf.getClass(), true)
841               + HConstants.HFILEBLOCK_HEADER_SIZE + size);
842       long hfileMetaSize =  ClassSize.align(ClassSize.estimateBase(HFileContext.class, true));
843       long hfileBlockExpectedSize =
844           ClassSize.align(ClassSize.estimateBase(HFileBlock.class, true));
845       long expected = hfileBlockExpectedSize + byteBufferExpectedSize + hfileMetaSize;
846       assertEquals("Block data size: " + size + ", byte buffer expected " +
847           "size: " + byteBufferExpectedSize + ", HFileBlock class expected " +
848           "size: " + hfileBlockExpectedSize + ";", expected,
849           block.heapSize());
850     }
851   }
852 }