Examples with DataBlockEncoding org.apache.hadoop.hbase.io.encoding.DataBlockEncoding used on opensource projects

Example 1 with DataBlockEncoding

use of org.apache.hadoop.hbase.io.encoding.DataBlockEncoding in project hbase by apache.

the class HFileDataBlockEncoderImpl method createFromFileInfo.

public static HFileDataBlockEncoder createFromFileInfo(FileInfo fileInfo) throws IOException {
    DataBlockEncoding encoding = DataBlockEncoding.NONE;
    byte[] dataBlockEncodingType = fileInfo.get(DATA_BLOCK_ENCODING);
    if (dataBlockEncodingType != null) {
        String dataBlockEncodingStr = Bytes.toString(dataBlockEncodingType);
        try {
            encoding = DataBlockEncoding.valueOf(dataBlockEncodingStr);
        } catch (IllegalArgumentException ex) {
            throw new IOException("Invalid data block encoding type in file info: " + dataBlockEncodingStr, ex);
        }
    }
    if (encoding == DataBlockEncoding.NONE) {
        return NoOpDataBlockEncoder.INSTANCE;
    }
    return new HFileDataBlockEncoderImpl(encoding);
}

Example 2 with DataBlockEncoding

use of org.apache.hadoop.hbase.io.encoding.DataBlockEncoding in project hbase by apache.

the class PerformanceEvaluation method doMultipleClients.

/**
 * Run all clients in this vm each to its own thread.
 * @param cmd Command to run
 * @throws IOException if creating a connection fails
 */
private void doMultipleClients(final Class<? extends Test> cmd) throws IOException {
    final List<Thread> threads = new ArrayList<>(this.N);
    final long[] timings = new long[this.N];
    final int perClientRows = R / N;
    final TableName tableName = this.tableName;
    final DataBlockEncoding encoding = this.blockEncoding;
    final boolean flushCommits = this.flushCommits;
    final Compression.Algorithm compression = this.compression;
    final boolean writeToWal = this.writeToWAL;
    final int preSplitRegions = this.presplitRegions;
    final boolean useTags = this.useTags;
    final int numTags = this.noOfTags;
    final Connection connection = ConnectionFactory.createConnection(getConf());
    for (int i = 0; i < this.N; i++) {
        final int index = i;
        Thread t = new Thread("TestClient-" + i) {

            @Override
            public void run() {
                super.run();
                PerformanceEvaluation pe = new PerformanceEvaluation(getConf());
                pe.tableName = tableName;
                pe.blockEncoding = encoding;
                pe.flushCommits = flushCommits;
                pe.compression = compression;
                pe.writeToWAL = writeToWal;
                pe.presplitRegions = preSplitRegions;
                pe.N = N;
                pe.connection = connection;
                pe.useTags = useTags;
                pe.noOfTags = numTags;
                try {
                    long elapsedTime = pe.runOneClient(cmd, index * perClientRows, perClientRows, R, flushCommits, writeToWAL, useTags, noOfTags, connection, msg -> LOG.info("client-" + getName() + " " + msg));
                    timings[index] = elapsedTime;
                    LOG.info("Finished " + getName() + " in " + elapsedTime + "ms writing " + perClientRows + " rows");
                } catch (IOException e) {
                    throw new RuntimeException(e);
                }
            }
        };
        threads.add(t);
    }
    for (Thread t : threads) {
        t.start();
    }
    for (Thread t : threads) {
        while (t.isAlive()) {
            try {
                t.join();
            } catch (InterruptedException e) {
                LOG.debug("Interrupted, continuing" + e.toString());
            }
        }
    }
    final String test = cmd.getSimpleName();
    LOG.info("[" + test + "] Summary of timings (ms): " + Arrays.toString(timings));
    Arrays.sort(timings);
    long total = 0;
    for (int i = 0; i < this.N; i++) {
        total += timings[i];
    }
    LOG.info("[" + test + "]" + "\tMin: " + timings[0] + "ms" + "\tMax: " + timings[this.N - 1] + "ms" + "\tAvg: " + (total / this.N) + "ms");
}

Example 3 with DataBlockEncoding

use of org.apache.hadoop.hbase.io.encoding.DataBlockEncoding in project hbase by apache.

the class TestHFileOutputFormat2 method testSerializeDeserializeFamilyDataBlockEncodingMap.

/**
 * Test for {@link HFileOutputFormat2#createFamilyDataBlockEncodingMap(Configuration)}.
 * Tests that the family data block encoding map is correctly serialized into
 * and deserialized from configuration
 *
 * @throws IOException
 */
@Ignore("Goes zombie too frequently; needs work. See HBASE-14563")
@Test
public void testSerializeDeserializeFamilyDataBlockEncodingMap() throws IOException {
    for (int numCfs = 0; numCfs <= 3; numCfs++) {
        Configuration conf = new Configuration(this.util.getConfiguration());
        Map<String, DataBlockEncoding> familyToDataBlockEncoding = getMockColumnFamiliesForDataBlockEncoding(numCfs);
        Table table = Mockito.mock(Table.class);
        setupMockColumnFamiliesForDataBlockEncoding(table, familyToDataBlockEncoding);
        TableDescriptor tableDescriptor = table.getDescriptor();
        conf.set(HFileOutputFormat2.DATABLOCK_ENCODING_FAMILIES_CONF_KEY, HFileOutputFormat2.serializeColumnFamilyAttribute(HFileOutputFormat2.dataBlockEncodingDetails, Arrays.asList(tableDescriptor)));
        // read back family specific data block encoding settings from the
        // configuration
        Map<byte[], DataBlockEncoding> retrievedFamilyToDataBlockEncodingMap = HFileOutputFormat2.createFamilyDataBlockEncodingMap(conf);
        // used mock values
        for (Entry<String, DataBlockEncoding> entry : familyToDataBlockEncoding.entrySet()) {
            assertEquals("DataBlockEncoding configuration incorrect for column family:" + entry.getKey(), entry.getValue(), retrievedFamilyToDataBlockEncodingMap.get(Bytes.toBytes(entry.getKey())));
        }
    }
}

Example 4 with DataBlockEncoding

use of org.apache.hadoop.hbase.io.encoding.DataBlockEncoding in project hbase by apache.

the class HFileOutputFormat2 method createRecordWriter.

static <V extends Cell> RecordWriter<ImmutableBytesWritable, V> createRecordWriter(final TaskAttemptContext context, final OutputCommitter committer) throws IOException {
    // Get the path of the temporary output file
    final Path outputDir = ((FileOutputCommitter) committer).getWorkPath();
    final Configuration conf = context.getConfiguration();
    final boolean writeMultipleTables = conf.getBoolean(MULTI_TABLE_HFILEOUTPUTFORMAT_CONF_KEY, false);
    final String writeTableNames = conf.get(OUTPUT_TABLE_NAME_CONF_KEY);
    if (writeTableNames == null || writeTableNames.isEmpty()) {
        throw new IllegalArgumentException("" + OUTPUT_TABLE_NAME_CONF_KEY + " cannot be empty");
    }
    final FileSystem fs = outputDir.getFileSystem(conf);
    // These configs. are from hbase-*.xml
    final long maxsize = conf.getLong(HConstants.HREGION_MAX_FILESIZE, HConstants.DEFAULT_MAX_FILE_SIZE);
    // Invented config.  Add to hbase-*.xml if other than default compression.
    final String defaultCompressionStr = conf.get("hfile.compression", Compression.Algorithm.NONE.getName());
    final Algorithm defaultCompression = HFileWriterImpl.compressionByName(defaultCompressionStr);
    String compressionStr = conf.get(COMPRESSION_OVERRIDE_CONF_KEY);
    final Algorithm overriddenCompression = compressionStr != null ? Compression.getCompressionAlgorithmByName(compressionStr) : null;
    final boolean compactionExclude = conf.getBoolean("hbase.mapreduce.hfileoutputformat.compaction.exclude", false);
    final Set<String> allTableNames = Arrays.stream(writeTableNames.split(Bytes.toString(tableSeparator))).collect(Collectors.toSet());
    // create a map from column family to the compression algorithm
    final Map<byte[], Algorithm> compressionMap = createFamilyCompressionMap(conf);
    final Map<byte[], BloomType> bloomTypeMap = createFamilyBloomTypeMap(conf);
    final Map<byte[], String> bloomParamMap = createFamilyBloomParamMap(conf);
    final Map<byte[], Integer> blockSizeMap = createFamilyBlockSizeMap(conf);
    String dataBlockEncodingStr = conf.get(DATABLOCK_ENCODING_OVERRIDE_CONF_KEY);
    final Map<byte[], DataBlockEncoding> datablockEncodingMap = createFamilyDataBlockEncodingMap(conf);
    final DataBlockEncoding overriddenEncoding = dataBlockEncodingStr != null ? DataBlockEncoding.valueOf(dataBlockEncodingStr) : null;
    return new RecordWriter<ImmutableBytesWritable, V>() {

        // Map of families to writers and how much has been output on the writer.
        private final Map<byte[], WriterLength> writers = new TreeMap<>(Bytes.BYTES_COMPARATOR);

        private final Map<byte[], byte[]> previousRows = new TreeMap<>(Bytes.BYTES_COMPARATOR);

        private final long now = EnvironmentEdgeManager.currentTime();

        private byte[] tableNameBytes = writeMultipleTables ? null : Bytes.toBytes(writeTableNames);

        @Override
        public void write(ImmutableBytesWritable row, V cell) throws IOException {
            Cell kv = cell;
            // null input == user explicitly wants to flush
            if (row == null && kv == null) {
                rollWriters(null);
                return;
            }
            byte[] rowKey = CellUtil.cloneRow(kv);
            int length = (PrivateCellUtil.estimatedSerializedSizeOf(kv)) - Bytes.SIZEOF_INT;
            byte[] family = CellUtil.cloneFamily(kv);
            if (writeMultipleTables) {
                tableNameBytes = MultiTableHFileOutputFormat.getTableName(row.get());
                tableNameBytes = TableName.valueOf(tableNameBytes).getNameWithNamespaceInclAsString().getBytes(Charset.defaultCharset());
                if (!allTableNames.contains(Bytes.toString(tableNameBytes))) {
                    throw new IllegalArgumentException("TableName " + Bytes.toString(tableNameBytes) + " not expected");
                }
            }
            byte[] tableAndFamily = getTableNameSuffixedWithFamily(tableNameBytes, family);
            WriterLength wl = this.writers.get(tableAndFamily);
            // If this is a new column family, verify that the directory exists
            if (wl == null) {
                Path writerPath = null;
                if (writeMultipleTables) {
                    Path tableRelPath = getTableRelativePath(tableNameBytes);
                    writerPath = new Path(outputDir, new Path(tableRelPath, Bytes.toString(family)));
                } else {
                    writerPath = new Path(outputDir, Bytes.toString(family));
                }
                fs.mkdirs(writerPath);
                configureStoragePolicy(conf, fs, tableAndFamily, writerPath);
            }
            // This can only happen once a row is finished though
            if (wl != null && wl.written + length >= maxsize && Bytes.compareTo(this.previousRows.get(family), rowKey) != 0) {
                rollWriters(wl);
            }
            // create a new WAL writer, if necessary
            if (wl == null || wl.writer == null) {
                InetSocketAddress[] favoredNodes = null;
                if (conf.getBoolean(LOCALITY_SENSITIVE_CONF_KEY, DEFAULT_LOCALITY_SENSITIVE)) {
                    HRegionLocation loc = null;
                    String tableName = Bytes.toString(tableNameBytes);
                    if (tableName != null) {
                        try (Connection connection = ConnectionFactory.createConnection(createRemoteClusterConf(conf));
                            RegionLocator locator = connection.getRegionLocator(TableName.valueOf(tableName))) {
                            loc = locator.getRegionLocation(rowKey);
                        } catch (Throwable e) {
                            LOG.warn("Something wrong locating rowkey {} in {}", Bytes.toString(rowKey), tableName, e);
                            loc = null;
                        }
                    }
                    if (null == loc) {
                        LOG.trace("Failed get of location, use default writer {}", Bytes.toString(rowKey));
                    } else {
                        LOG.debug("First rowkey: [{}]", Bytes.toString(rowKey));
                        InetSocketAddress initialIsa = new InetSocketAddress(loc.getHostname(), loc.getPort());
                        if (initialIsa.isUnresolved()) {
                            LOG.trace("Failed resolve address {}, use default writer", loc.getHostnamePort());
                        } else {
                            LOG.debug("Use favored nodes writer: {}", initialIsa.getHostString());
                            favoredNodes = new InetSocketAddress[] { initialIsa };
                        }
                    }
                }
                wl = getNewWriter(tableNameBytes, family, conf, favoredNodes);
            }
            // we now have the proper WAL writer. full steam ahead
            PrivateCellUtil.updateLatestStamp(cell, this.now);
            wl.writer.append(kv);
            wl.written += length;
            // Copy the row so we know when a row transition.
            this.previousRows.put(family, rowKey);
        }

        private Path getTableRelativePath(byte[] tableNameBytes) {
            String tableName = Bytes.toString(tableNameBytes);
            String[] tableNameParts = tableName.split(":");
            Path tableRelPath = new Path(tableNameParts[0]);
            if (tableNameParts.length > 1) {
                tableRelPath = new Path(tableRelPath, tableNameParts[1]);
            }
            return tableRelPath;
        }

        private void rollWriters(WriterLength writerLength) throws IOException {
            if (writerLength != null) {
                closeWriter(writerLength);
            } else {
                for (WriterLength wl : this.writers.values()) {
                    closeWriter(wl);
                }
            }
        }

        private void closeWriter(WriterLength wl) throws IOException {
            if (wl.writer != null) {
                LOG.info("Writer=" + wl.writer.getPath() + ((wl.written == 0) ? "" : ", wrote=" + wl.written));
                close(wl.writer);
                wl.writer = null;
            }
            wl.written = 0;
        }

        private Configuration createRemoteClusterConf(Configuration conf) {
            final Configuration newConf = new Configuration(conf);
            final String quorum = conf.get(REMOTE_CLUSTER_ZOOKEEPER_QUORUM_CONF_KEY);
            final String clientPort = conf.get(REMOTE_CLUSTER_ZOOKEEPER_CLIENT_PORT_CONF_KEY);
            final String parent = conf.get(REMOTE_CLUSTER_ZOOKEEPER_ZNODE_PARENT_CONF_KEY);
            if (quorum != null && clientPort != null && parent != null) {
                newConf.set(HConstants.ZOOKEEPER_QUORUM, quorum);
                newConf.setInt(HConstants.ZOOKEEPER_CLIENT_PORT, Integer.parseInt(clientPort));
                newConf.set(HConstants.ZOOKEEPER_ZNODE_PARENT, parent);
            }
            for (Entry<String, String> entry : conf) {
                String key = entry.getKey();
                if (REMOTE_CLUSTER_ZOOKEEPER_QUORUM_CONF_KEY.equals(key) || REMOTE_CLUSTER_ZOOKEEPER_CLIENT_PORT_CONF_KEY.equals(key) || REMOTE_CLUSTER_ZOOKEEPER_ZNODE_PARENT_CONF_KEY.equals(key)) {
                    // Handled them above
                    continue;
                }
                if (entry.getKey().startsWith(REMOTE_CLUSTER_CONF_PREFIX)) {
                    String originalKey = entry.getKey().substring(REMOTE_CLUSTER_CONF_PREFIX.length());
                    if (!originalKey.isEmpty()) {
                        newConf.set(originalKey, entry.getValue());
                    }
                }
            }
            return newConf;
        }

        /*
       * Create a new StoreFile.Writer.
       * @return A WriterLength, containing a new StoreFile.Writer.
       */
        @edu.umd.cs.findbugs.annotations.SuppressWarnings(value = "BX_UNBOXING_IMMEDIATELY_REBOXED", justification = "Not important")
        private WriterLength getNewWriter(byte[] tableName, byte[] family, Configuration conf, InetSocketAddress[] favoredNodes) throws IOException {
            byte[] tableAndFamily = getTableNameSuffixedWithFamily(tableName, family);
            Path familydir = new Path(outputDir, Bytes.toString(family));
            if (writeMultipleTables) {
                familydir = new Path(outputDir, new Path(getTableRelativePath(tableName), Bytes.toString(family)));
            }
            WriterLength wl = new WriterLength();
            Algorithm compression = overriddenCompression;
            compression = compression == null ? compressionMap.get(tableAndFamily) : compression;
            compression = compression == null ? defaultCompression : compression;
            BloomType bloomType = bloomTypeMap.get(tableAndFamily);
            bloomType = bloomType == null ? BloomType.NONE : bloomType;
            String bloomParam = bloomParamMap.get(tableAndFamily);
            if (bloomType == BloomType.ROWPREFIX_FIXED_LENGTH) {
                conf.set(BloomFilterUtil.PREFIX_LENGTH_KEY, bloomParam);
            }
            Integer blockSize = blockSizeMap.get(tableAndFamily);
            blockSize = blockSize == null ? HConstants.DEFAULT_BLOCKSIZE : blockSize;
            DataBlockEncoding encoding = overriddenEncoding;
            encoding = encoding == null ? datablockEncodingMap.get(tableAndFamily) : encoding;
            encoding = encoding == null ? DataBlockEncoding.NONE : encoding;
            HFileContextBuilder contextBuilder = new HFileContextBuilder().withCompression(compression).withDataBlockEncoding(encoding).withChecksumType(StoreUtils.getChecksumType(conf)).withBytesPerCheckSum(StoreUtils.getBytesPerChecksum(conf)).withBlockSize(blockSize).withColumnFamily(family).withTableName(tableName);
            if (HFile.getFormatVersion(conf) >= HFile.MIN_FORMAT_VERSION_WITH_TAGS) {
                contextBuilder.withIncludesTags(true);
            }
            HFileContext hFileContext = contextBuilder.build();
            if (null == favoredNodes) {
                wl.writer = new StoreFileWriter.Builder(conf, CacheConfig.DISABLED, fs).withOutputDir(familydir).withBloomType(bloomType).withFileContext(hFileContext).build();
            } else {
                wl.writer = new StoreFileWriter.Builder(conf, CacheConfig.DISABLED, new HFileSystem(fs)).withOutputDir(familydir).withBloomType(bloomType).withFileContext(hFileContext).withFavoredNodes(favoredNodes).build();
            }
            this.writers.put(tableAndFamily, wl);
            return wl;
        }

        private void close(final StoreFileWriter w) throws IOException {
            if (w != null) {
                w.appendFileInfo(BULKLOAD_TIME_KEY, Bytes.toBytes(EnvironmentEdgeManager.currentTime()));
                w.appendFileInfo(BULKLOAD_TASK_KEY, Bytes.toBytes(context.getTaskAttemptID().toString()));
                w.appendFileInfo(MAJOR_COMPACTION_KEY, Bytes.toBytes(true));
                w.appendFileInfo(EXCLUDE_FROM_MINOR_COMPACTION_KEY, Bytes.toBytes(compactionExclude));
                w.appendTrackedTimestampsToMetadata();
                w.close();
            }
        }

        @Override
        public void close(TaskAttemptContext c) throws IOException, InterruptedException {
            for (WriterLength wl : this.writers.values()) {
                close(wl.writer);
            }
        }
    };
}

Example 5 with DataBlockEncoding

use of org.apache.hadoop.hbase.io.encoding.DataBlockEncoding in project hbase by apache.

the class TestCacheOnWrite method readStoreFile.

private void readStoreFile(boolean useTags) throws IOException {
    HFile.Reader reader = HFile.createReader(fs, storeFilePath, cacheConf, true, conf);
    LOG.info("HFile information: " + reader);
    HFileContext meta = new HFileContextBuilder().withCompression(compress).withBytesPerCheckSum(CKBYTES).withChecksumType(ChecksumType.NULL).withBlockSize(DATA_BLOCK_SIZE).withDataBlockEncoding(NoOpDataBlockEncoder.INSTANCE.getDataBlockEncoding()).withIncludesTags(useTags).build();
    final boolean cacheBlocks = false;
    final boolean pread = false;
    HFileScanner scanner = reader.getScanner(conf, cacheBlocks, pread);
    assertTrue(testDescription, scanner.seekTo());
    long offset = 0;
    EnumMap<BlockType, Integer> blockCountByType = new EnumMap<>(BlockType.class);
    DataBlockEncoding encodingInCache = NoOpDataBlockEncoder.INSTANCE.getDataBlockEncoding();
    List<Long> cachedBlocksOffset = new ArrayList<>();
    Map<Long, Pair<HFileBlock, HFileBlock>> cachedBlocks = new HashMap<>();
    while (offset < reader.getTrailer().getLoadOnOpenDataOffset()) {
        // Flags: don't cache the block, use pread, this is not a compaction.
        // Also, pass null for expected block type to avoid checking it.
        HFileBlock block = reader.readBlock(offset, -1, false, true, false, true, null, encodingInCache);
        BlockCacheKey blockCacheKey = new BlockCacheKey(reader.getName(), offset);
        HFileBlock fromCache = (HFileBlock) blockCache.getBlock(blockCacheKey, true, false, true);
        boolean isCached = fromCache != null;
        cachedBlocksOffset.add(offset);
        cachedBlocks.put(offset, fromCache == null ? null : Pair.newPair(block, fromCache));
        boolean shouldBeCached = cowType.shouldBeCached(block.getBlockType());
        assertTrue("shouldBeCached: " + shouldBeCached + "\n" + "isCached: " + isCached + "\n" + "Test description: " + testDescription + "\n" + "block: " + block + "\n" + "encodingInCache: " + encodingInCache + "\n" + "blockCacheKey: " + blockCacheKey, shouldBeCached == isCached);
        if (isCached) {
            if (cacheConf.shouldCacheCompressed(fromCache.getBlockType().getCategory())) {
                if (compress != Compression.Algorithm.NONE) {
                    assertFalse(fromCache.isUnpacked());
                }
                fromCache = fromCache.unpack(meta, reader.getUncachedBlockReader());
            } else {
                assertTrue(fromCache.isUnpacked());
            }
            // block we cached at write-time and block read from file should be identical
            assertEquals(block.getChecksumType(), fromCache.getChecksumType());
            assertEquals(block.getBlockType(), fromCache.getBlockType());
            assertNotEquals(BlockType.ENCODED_DATA, block.getBlockType());
            assertEquals(block.getOnDiskSizeWithHeader(), fromCache.getOnDiskSizeWithHeader());
            assertEquals(block.getOnDiskSizeWithoutHeader(), fromCache.getOnDiskSizeWithoutHeader());
            assertEquals(block.getUncompressedSizeWithoutHeader(), fromCache.getUncompressedSizeWithoutHeader());
        }
        offset += block.getOnDiskSizeWithHeader();
        BlockType bt = block.getBlockType();
        Integer count = blockCountByType.get(bt);
        blockCountByType.put(bt, (count == null ? 0 : count) + 1);
    }
    LOG.info("Block count by type: " + blockCountByType);
    String countByType = blockCountByType.toString();
    if (useTags) {
        assertEquals("{" + BlockType.DATA + "=2663, LEAF_INDEX=297, BLOOM_CHUNK=9, INTERMEDIATE_INDEX=32}", countByType);
    } else {
        assertEquals("{" + BlockType.DATA + "=2498, LEAF_INDEX=278, BLOOM_CHUNK=9, INTERMEDIATE_INDEX=31}", countByType);
    }
    // iterate all the keyvalue from hfile
    while (scanner.next()) {
        scanner.getCell();
    }
    Iterator<Long> iterator = cachedBlocksOffset.iterator();
    while (iterator.hasNext()) {
        Long entry = iterator.next();
        BlockCacheKey blockCacheKey = new BlockCacheKey(reader.getName(), entry);
        Pair<HFileBlock, HFileBlock> blockPair = cachedBlocks.get(entry);
        if (blockPair != null) {
            // Call return twice because for the isCache cased the counter would have got incremented
            // twice. Notice that here we need to returnBlock with different blocks. see comments in
            // BucketCache#returnBlock.
            blockPair.getSecond().release();
            if (cacheCompressedData) {
                if (this.compress == Compression.Algorithm.NONE || cowType == CacheOnWriteType.INDEX_BLOCKS || cowType == CacheOnWriteType.BLOOM_BLOCKS) {
                    blockPair.getFirst().release();
                }
            } else {
                blockPair.getFirst().release();
            }
        }
    }
    scanner.shipped();
    reader.close();
}