diff --git a/hbase-server/src/main/java/org/apache/hadoop/hbase/mapreduce/HFileOutputFormat2.java b/hbase-server/src/main/java/org/apache/hadoop/hbase/mapreduce/HFileOutputFormat2.java index 91e5b22..e680cf6 100644 --- a/hbase-server/src/main/java/org/apache/hadoop/hbase/mapreduce/HFileOutputFormat2.java +++ b/hbase-server/src/main/java/org/apache/hadoop/hbase/mapreduce/HFileOutputFormat2.java @@ -19,6 +19,7 @@ package org.apache.hadoop.hbase.mapreduce; import java.io.IOException; import java.io.UnsupportedEncodingException; +import java.net.InetSocketAddress; import java.net.URLDecoder; import java.net.URLEncoder; import java.util.ArrayList; @@ -40,11 +41,13 @@ import org.apache.hadoop.hbase.Cell; import org.apache.hadoop.hbase.CellUtil; import org.apache.hadoop.hbase.HColumnDescriptor; import org.apache.hadoop.hbase.HConstants; +import org.apache.hadoop.hbase.HRegionLocation; import org.apache.hadoop.hbase.HTableDescriptor; import org.apache.hadoop.hbase.KeyValue; import org.apache.hadoop.hbase.KeyValueUtil; import org.apache.hadoop.hbase.client.HTable; import org.apache.hadoop.hbase.client.Put; +import org.apache.hadoop.hbase.fs.HFileSystem; import org.apache.hadoop.hbase.io.ImmutableBytesWritable; import org.apache.hadoop.hbase.io.compress.Compression; import org.apache.hadoop.hbase.io.compress.Compression.Algorithm; @@ -105,6 +108,15 @@ public class HFileOutputFormat2 public static final String DATABLOCK_ENCODING_OVERRIDE_CONF_KEY = "hbase.mapreduce.hfileoutputformat.datablock.encoding"; + /** + * Keep locality while generating HFiles for bulkload. See HBASE-12596 + */ + private static final String LOCALITY_SENSITIVE_CONF_KEY = + "hbase.bulkload.locality.sensitive.enabled"; + private static final boolean DEFAULT_LOCALITY_SENSITIVE = true; + private static final String OUTPUT_TABLE_NAME_CONF_KEY = + "hbase.mapreduce.hfileoutputformat.table.name"; + @Override public RecordWriter getRecordWriter( final TaskAttemptContext context) throws IOException, InterruptedException { @@ -188,7 +200,51 @@ public class HFileOutputFormat2 // create a new HLog writer, if necessary if (wl == null || wl.writer == null) { - wl = getNewWriter(family, conf); + if (conf.getBoolean(LOCALITY_SENSITIVE_CONF_KEY, DEFAULT_LOCALITY_SENSITIVE)) { + String tableName = conf.get(OUTPUT_TABLE_NAME_CONF_KEY); + HRegionLocation loc = null; + HTable htable = null; + try { + htable = new HTable(conf, tableName); + loc = htable.getRegionLocation(rowKey); + } catch (Throwable e) { + LOG.warn("there's something wrong when locating rowkey: " + + Bytes.toString(rowKey), e); + loc = null; + } finally { + if(null != htable) { + htable.close(); + } + } + + if (null == loc) { + if (LOG.isTraceEnabled()) { + LOG.trace("failed to get region location, so use default writer: " + + Bytes.toString(rowKey)); + } + wl = getNewWriter(family, conf, null); + } else { + if (LOG.isDebugEnabled()) { + LOG.debug("first rowkey: [" + Bytes.toString(rowKey) + "]"); + } + InetSocketAddress initialIsa = + new InetSocketAddress(loc.getHostname(), loc.getPort()); + if (initialIsa.isUnresolved()) { + if (LOG.isTraceEnabled()) { + LOG.trace("failed to resolve bind address: " + loc.getHostname() + ":" + + loc.getPort() + ", so use default writer"); + } + wl = getNewWriter(family, conf, null); + } else { + if(LOG.isDebugEnabled()) { + LOG.debug("use favored nodes writer: " + initialIsa.getHostString()); + } + wl = getNewWriter(family, conf, new InetSocketAddress[] { initialIsa }); + } + } + } else { + wl = getNewWriter(family, conf, null); + } } // we now have the proper HLog writer. full steam ahead @@ -218,8 +274,8 @@ public class HFileOutputFormat2 * @return A WriterLength, containing a new StoreFile.Writer. * @throws IOException */ - private WriterLength getNewWriter(byte[] family, Configuration conf) - throws IOException { + private WriterLength getNewWriter(byte[] family, Configuration conf, + InetSocketAddress[] favoredNodes) throws IOException { WriterLength wl = new WriterLength(); Path familydir = new Path(outputdir, Bytes.toString(family)); Algorithm compression = compressionMap.get(family); @@ -241,10 +297,18 @@ public class HFileOutputFormat2 contextBuilder.withDataBlockEncoding(encoding); HFileContext hFileContext = contextBuilder.build(); - wl.writer = new StoreFile.WriterBuilder(conf, new CacheConfig(tempConf), fs) - .withOutputDir(familydir).withBloomType(bloomType) - .withComparator(KeyValue.COMPARATOR) - .withFileContext(hFileContext).build(); + if (null == favoredNodes) { + wl.writer = + new StoreFile.WriterBuilder(conf, new CacheConfig(tempConf), fs) + .withOutputDir(familydir).withBloomType(bloomType) + .withComparator(KeyValue.COMPARATOR).withFileContext(hFileContext).build(); + } else { + wl.writer = + new StoreFile.WriterBuilder(conf, new CacheConfig(tempConf), new HFileSystem(fs)) + .withOutputDir(familydir).withBloomType(bloomType) + .withComparator(KeyValue.COMPARATOR).withFileContext(hFileContext) + .withFavoredNodes(favoredNodes).build(); + } this.writers.put(family, wl); return wl; @@ -384,6 +448,12 @@ public class HFileOutputFormat2 MutationSerialization.class.getName(), ResultSerialization.class.getName(), KeyValueSerialization.class.getName()); + if (conf.getBoolean(LOCALITY_SENSITIVE_CONF_KEY, DEFAULT_LOCALITY_SENSITIVE)) { + // record this table name for creating writer by favored nodes + LOG.info("bulkload locality sensitive enabled"); + conf.set(OUTPUT_TABLE_NAME_CONF_KEY, table.getName().getNameAsString()); + } + // Use table's region boundaries for TOP split points. LOG.info("Looking up current regions for table " + Bytes.toString(table.getTableName())); List startKeys = getRegionStartKeys(table); diff --git a/hbase-server/src/test/java/org/apache/hadoop/hbase/mapreduce/TestHFileOutputFormat.java b/hbase-server/src/test/java/org/apache/hadoop/hbase/mapreduce/TestHFileOutputFormat.java index f3157f2..4f8486c 100644 --- a/hbase-server/src/test/java/org/apache/hadoop/hbase/mapreduce/TestHFileOutputFormat.java +++ b/hbase-server/src/test/java/org/apache/hadoop/hbase/mapreduce/TestHFileOutputFormat.java @@ -335,6 +335,7 @@ public class TestHFileOutputFormat { job.setWorkingDirectory(util.getDataTestDir("testJobConfiguration")); HTable table = Mockito.mock(HTable.class); setupMockStartKeys(table); + setupMockTableName(table); HFileOutputFormat.configureIncrementalLoad(job, table); assertEquals(job.getNumReduceTasks(), 4); } @@ -776,6 +777,11 @@ public class TestHFileOutputFormat { Mockito.doReturn(mockKeys).when(table).getStartKeys(); } + private void setupMockTableName(HTable table) throws IOException { + TableName mockTableName = TableName.valueOf("mock_table"); + Mockito.doReturn(mockTableName).when(table).getName(); + } + /** * Test that {@link HFileOutputFormat} RecordWriter uses compression and * bloom filter settings from the column family descriptor @@ -804,6 +810,9 @@ public class TestHFileOutputFormat { // pollutes the GZip codec pool with an incompatible compressor. conf.set("io.seqfile.compression.type", "NONE"); conf.set("hbase.fs.tmp.dir", dir.toString()); + // turn locality off to eliminate getRegionLocation fail-and-retry time when writing kvs + conf.setBoolean("hbase.bulkload.locality.sensitive.enabled", false); + Job job = new Job(conf, "testLocalMRIncrementalLoad"); job.setWorkingDirectory(util.getDataTestDirOnTestFS("testColumnFamilySettings")); setupRandomGeneratorMapper(job); diff --git a/hbase-server/src/test/java/org/apache/hadoop/hbase/mapreduce/TestHFileOutputFormat2.java b/hbase-server/src/test/java/org/apache/hadoop/hbase/mapreduce/TestHFileOutputFormat2.java index 28ca56f..ff80b8a 100644 --- a/hbase-server/src/test/java/org/apache/hadoop/hbase/mapreduce/TestHFileOutputFormat2.java +++ b/hbase-server/src/test/java/org/apache/hadoop/hbase/mapreduce/TestHFileOutputFormat2.java @@ -336,6 +336,7 @@ public class TestHFileOutputFormat2 { job.setWorkingDirectory(util.getDataTestDir("testJobConfiguration")); HTable table = Mockito.mock(HTable.class); setupMockStartKeys(table); + setupMockTableName(table); HFileOutputFormat2.configureIncrementalLoad(job, table); assertEquals(job.getNumReduceTasks(), 4); } @@ -778,6 +779,11 @@ public class TestHFileOutputFormat2 { Mockito.doReturn(mockKeys).when(table).getStartKeys(); } + private void setupMockTableName(HTable table) throws IOException { + TableName mockTableName = TableName.valueOf("mock_table"); + Mockito.doReturn(mockTableName).when(table).getName(); + } + /** * Test that {@link HFileOutputFormat2} RecordWriter uses compression and * bloom filter settings from the column family descriptor @@ -806,6 +812,9 @@ public class TestHFileOutputFormat2 { // pollutes the GZip codec pool with an incompatible compressor. conf.set("io.seqfile.compression.type", "NONE"); conf.set("hbase.fs.tmp.dir", dir.toString()); + // turn locality off to eliminate getRegionLocation fail-and-retry time when writing kvs + conf.setBoolean("hbase.bulkload.locality.sensitive.enabled", false); + Job job = new Job(conf, "testLocalMRIncrementalLoad"); job.setWorkingDirectory(util.getDataTestDirOnTestFS("testColumnFamilySettings")); setupRandomGeneratorMapper(job);