@Test
public void testTableWithCFNameStartWithUnderScore() throws Exception {
Path dir = util.getDataTestDirOnTestFS("cfNameStartWithUnderScore");
FileSystem fs = util.getTestFileSystem();
dir = dir.makeQualified(fs.getUri(), fs.getWorkingDirectory());
String family = "_cf";
Path familyDir = new Path(dir, family);
byte[] from = Bytes.toBytes("begin");
byte[] to = Bytes.toBytes("end");
Configuration conf = util.getConfiguration();
String tableName = "mytable_cfNameStartWithUnderScore";
Table table = util.createTable(TableName.valueOf(tableName), family);
HFileTestUtil.createHFile(
conf, fs, new Path(familyDir, "hfile"), Bytes.toBytes(family), QUALIFIER, from, to, 1000);
LoadIncrementalHFiles loader = new LoadIncrementalHFiles(conf);
String[] args = {dir.toString(), tableName};
try {
loader.run(args);
assertEquals(1000, util.countRows(table));
} finally {
if (null != table) {
table.close();
}
}
}
private void runIncrementalPELoad(
Configuration conf,
HTableDescriptor tableDescriptor,
RegionLocator regionLocator,
Path outDir)
throws IOException, UnsupportedEncodingException, InterruptedException,
ClassNotFoundException {
Job job = new Job(conf, "testLocalMRIncrementalLoad");
job.setWorkingDirectory(util.getDataTestDirOnTestFS("runIncrementalPELoad"));
job.getConfiguration()
.setStrings(
"io.serializations",
conf.get("io.serializations"),
MutationSerialization.class.getName(),
ResultSerialization.class.getName(),
KeyValueSerialization.class.getName());
setupRandomGeneratorMapper(job);
HFileOutputFormat2.configureIncrementalLoad(job, tableDescriptor, regionLocator);
FileOutputFormat.setOutputPath(job, outDir);
assertFalse(util.getTestFileSystem().exists(outDir));
assertEquals(regionLocator.getAllRegionLocations().size(), job.getNumReduceTasks());
assertTrue(job.waitForCompletion(true));
}
@Test
public void testLoadTooMayHFiles() throws Exception {
Path dir = util.getDataTestDirOnTestFS("testLoadTooMayHFiles");
FileSystem fs = util.getTestFileSystem();
dir = dir.makeQualified(fs);
Path familyDir = new Path(dir, Bytes.toString(FAMILY));
byte[] from = Bytes.toBytes("begin");
byte[] to = Bytes.toBytes("end");
for (int i = 0; i <= MAX_FILES_PER_REGION_PER_FAMILY; i++) {
HFileTestUtil.createHFile(
util.getConfiguration(),
fs,
new Path(familyDir, "hfile_" + i),
FAMILY,
QUALIFIER,
from,
to,
1000);
}
LoadIncrementalHFiles loader = new LoadIncrementalHFiles(util.getConfiguration());
String[] args = {dir.toString(), "mytable_testLoadTooMayHFiles"};
try {
loader.run(args);
fail("Bulk loading too many files should fail");
} catch (IOException ie) {
assertTrue(
ie.getMessage()
.contains("Trying to load more than " + MAX_FILES_PER_REGION_PER_FAMILY + " hfiles"));
}
}
@Test
public void testSplitStoreFile() throws IOException {
Path dir = util.getDataTestDirOnTestFS("testSplitHFile");
FileSystem fs = util.getTestFileSystem();
Path testIn = new Path(dir, "testhfile");
HColumnDescriptor familyDesc = new HColumnDescriptor(FAMILY);
HFileTestUtil.createHFile(
util.getConfiguration(),
fs,
testIn,
FAMILY,
QUALIFIER,
Bytes.toBytes("aaa"),
Bytes.toBytes("zzz"),
1000);
Path bottomOut = new Path(dir, "bottom.out");
Path topOut = new Path(dir, "top.out");
LoadIncrementalHFiles.splitStoreFile(
util.getConfiguration(), testIn, familyDesc, Bytes.toBytes("ggg"), bottomOut, topOut);
int rowCount = verifyHFile(bottomOut);
rowCount += verifyHFile(topOut);
assertEquals(1000, rowCount);
}
@Test
public void bulkLoadHFileTest() throws Exception {
String testName = TestRegionObserverInterface.class.getName() + ".bulkLoadHFileTest";
TableName tableName = TableName.valueOf(TEST_TABLE.getNameAsString() + ".bulkLoadHFileTest");
Configuration conf = util.getConfiguration();
HTable table = util.createTable(tableName, new byte[][] {A, B, C});
try {
verifyMethodResult(
SimpleRegionObserver.class,
new String[] {"hadPreBulkLoadHFile", "hadPostBulkLoadHFile"},
tableName,
new Boolean[] {false, false});
FileSystem fs = util.getTestFileSystem();
final Path dir = util.getDataTestDirOnTestFS(testName).makeQualified(fs);
Path familyDir = new Path(dir, Bytes.toString(A));
createHFile(util.getConfiguration(), fs, new Path(familyDir, Bytes.toString(A)), A, A);
// Bulk load
new LoadIncrementalHFiles(conf).doBulkLoad(dir, new HTable(conf, tableName));
verifyMethodResult(
SimpleRegionObserver.class,
new String[] {"hadPreBulkLoadHFile", "hadPostBulkLoadHFile"},
tableName,
new Boolean[] {true, true});
} finally {
util.deleteTable(tableName);
table.close();
}
}
public void doAnAction() throws Exception {
long iteration = numBulkLoads.getAndIncrement();
Path dir = UTIL.getDataTestDirOnTestFS(String.format("bulkLoad_%08d", iteration));
// create HFiles for different column families
FileSystem fs = UTIL.getTestFileSystem();
byte[] val = Bytes.toBytes(String.format("%010d", iteration));
final List<Pair<byte[], String>> famPaths = new ArrayList<Pair<byte[], String>>(NUM_CFS);
for (int i = 0; i < NUM_CFS; i++) {
Path hfile = new Path(dir, family(i));
byte[] fam = Bytes.toBytes(family(i));
createHFile(fs, hfile, fam, QUAL, val, 1000);
famPaths.add(new Pair<>(fam, hfile.toString()));
}
// bulk load HFiles
final ClusterConnection conn = (ClusterConnection) UTIL.getAdmin().getConnection();
RegionServerCallable<Void> callable =
new RegionServerCallable<Void>(conn, tableName, Bytes.toBytes("aaa")) {
@Override
public Void call(int callTimeout) throws Exception {
LOG.debug(
"Going to connect to server "
+ getLocation()
+ " for row "
+ Bytes.toStringBinary(getRow()));
byte[] regionName = getLocation().getRegionInfo().getRegionName();
BulkLoadHFileRequest request =
RequestConverter.buildBulkLoadHFileRequest(famPaths, regionName, true);
getStub().bulkLoadHFile(null, request);
return null;
}
};
RpcRetryingCallerFactory factory = new RpcRetryingCallerFactory(conf);
RpcRetryingCaller<Void> caller = factory.<Void>newCaller();
caller.callWithRetries(callable, Integer.MAX_VALUE);
// Periodically do compaction to reduce the number of open file handles.
if (numBulkLoads.get() % 5 == 0) {
// 5 * 50 = 250 open file handles!
callable =
new RegionServerCallable<Void>(conn, tableName, Bytes.toBytes("aaa")) {
@Override
public Void call(int callTimeout) throws Exception {
LOG.debug(
"compacting " + getLocation() + " for row " + Bytes.toStringBinary(getRow()));
AdminProtos.AdminService.BlockingInterface server =
conn.getAdmin(getLocation().getServerName());
CompactRegionRequest request =
RequestConverter.buildCompactRegionRequest(
getLocation().getRegionInfo().getRegionName(), true, null);
server.compactRegion(null, request);
numCompactions.incrementAndGet();
return null;
}
};
caller.callWithRetries(callable, Integer.MAX_VALUE);
}
}
private void runTest(
String testName,
HTableDescriptor htd,
BloomType bloomType,
boolean preCreateTable,
byte[][] tableSplitKeys,
byte[][][] hfileRanges)
throws Exception {
Path dir = util.getDataTestDirOnTestFS(testName);
FileSystem fs = util.getTestFileSystem();
dir = dir.makeQualified(fs);
Path familyDir = new Path(dir, Bytes.toString(FAMILY));
int hfileIdx = 0;
for (byte[][] range : hfileRanges) {
byte[] from = range[0];
byte[] to = range[1];
HFileTestUtil.createHFile(
util.getConfiguration(),
fs,
new Path(familyDir, "hfile_" + hfileIdx++),
FAMILY,
QUALIFIER,
from,
to,
1000);
}
int expectedRows = hfileIdx * 1000;
if (preCreateTable) {
util.getHBaseAdmin().createTable(htd, tableSplitKeys);
}
final TableName tableName = htd.getTableName();
LoadIncrementalHFiles loader = new LoadIncrementalHFiles(util.getConfiguration());
String[] args = {dir.toString(), tableName.toString()};
loader.run(args);
Table table = new HTable(util.getConfiguration(), tableName);
try {
assertEquals(expectedRows, util.countRows(table));
} finally {
table.close();
}
// verify staging folder has been cleaned up
Path stagingBasePath = SecureBulkLoadUtil.getBaseStagingDir(util.getConfiguration());
if (fs.exists(stagingBasePath)) {
FileStatus[] files = fs.listStatus(stagingBasePath);
for (FileStatus file : files) {
assertTrue(
"Folder=" + file.getPath() + " is not cleaned up.",
file.getPath().getName() != "DONOTERASE");
}
}
util.deleteTable(tableName);
}
@BeforeClass
public static void setUpBeforeClass() throws Exception {
CONF = TEST_UTIL.getConfiguration();
TEST_UTIL.startMiniDFSCluster(1);
CLUSTER = TEST_UTIL.getDFSCluster();
FS = CLUSTER.getFileSystem();
DIR = TEST_UTIL.getDataTestDirOnTestFS("TestDurability");
FSUtils.setRootDir(CONF, DIR);
}
/** Run small MR job. */
@Ignore("Goes zombie too frequently; needs work. See HBASE-14563")
@Test
public void testWritingPEData() throws Exception {
Configuration conf = util.getConfiguration();
Path testDir = util.getDataTestDirOnTestFS("testWritingPEData");
FileSystem fs = testDir.getFileSystem(conf);
// Set down this value or we OOME in eclipse.
conf.setInt("mapreduce.task.io.sort.mb", 20);
// Write a few files.
conf.setLong(HConstants.HREGION_MAX_FILESIZE, 64 * 1024);
Job job = new Job(conf, "testWritingPEData");
setupRandomGeneratorMapper(job);
// This partitioner doesn't work well for number keys but using it anyways
// just to demonstrate how to configure it.
byte[] startKey = new byte[RandomKVGeneratingMapper.KEYLEN_DEFAULT];
byte[] endKey = new byte[RandomKVGeneratingMapper.KEYLEN_DEFAULT];
Arrays.fill(startKey, (byte) 0);
Arrays.fill(endKey, (byte) 0xff);
job.setPartitionerClass(SimpleTotalOrderPartitioner.class);
// Set start and end rows for partitioner.
SimpleTotalOrderPartitioner.setStartKey(job.getConfiguration(), startKey);
SimpleTotalOrderPartitioner.setEndKey(job.getConfiguration(), endKey);
job.setReducerClass(KeyValueSortReducer.class);
job.setOutputFormatClass(HFileOutputFormat2.class);
job.setNumReduceTasks(4);
job.getConfiguration()
.setStrings(
"io.serializations",
conf.get("io.serializations"),
MutationSerialization.class.getName(),
ResultSerialization.class.getName(),
KeyValueSerialization.class.getName());
FileOutputFormat.setOutputPath(job, testDir);
assertTrue(job.waitForCompletion(false));
FileStatus[] files = fs.listStatus(testDir);
assertTrue(files.length > 0);
}
/**
* Write a random data file and a non-file in a dir with a valid family name but not part of the
* table families. we should we able to bulkload without getting the unmatched family exception.
* HBASE-13037/HBASE-13227
*/
private void testNonHfileFolder(String tableName, boolean preCreateTable) throws Exception {
Path dir = util.getDataTestDirOnTestFS(tableName);
FileSystem fs = util.getTestFileSystem();
dir = dir.makeQualified(fs);
Path familyDir = new Path(dir, Bytes.toString(FAMILY));
HFileTestUtil.createHFile(
util.getConfiguration(),
fs,
new Path(familyDir, "hfile_0"),
FAMILY,
QUALIFIER,
Bytes.toBytes("begin"),
Bytes.toBytes("end"),
500);
createRandomDataFile(fs, new Path(familyDir, "012356789"), 16 * 1024);
final String NON_FAMILY_FOLDER = "_logs";
Path nonFamilyDir = new Path(dir, NON_FAMILY_FOLDER);
fs.mkdirs(nonFamilyDir);
fs.mkdirs(new Path(nonFamilyDir, "non-file"));
createRandomDataFile(fs, new Path(nonFamilyDir, "012356789"), 16 * 1024);
Table table = null;
try {
if (preCreateTable) {
table = util.createTable(TableName.valueOf(tableName), FAMILY);
} else {
table = util.getConnection().getTable(TableName.valueOf(tableName));
}
final String[] args = {dir.toString(), tableName};
new LoadIncrementalHFiles(util.getConfiguration()).run(args);
assertEquals(500, util.countRows(table));
} finally {
if (table != null) {
table.close();
}
fs.delete(dir, true);
}
}
/**
* Run an ImportTsv job and perform basic validation on the results. Returns the ImportTsv <code>
* Tool</code> instance so that other tests can inspect it for further validation as necessary.
* This method is static to insure non-reliance on instance's util/conf facilities.
*
* @param args Any arguments to pass BEFORE inputFile path is appended.
* @param dataAvailable
* @return The Tool instance used to run the test.
*/
private Tool doMROnTableTest(
HBaseTestingUtility util,
String family,
String data,
String[] args,
int valueMultiplier,
boolean dataAvailable)
throws Exception {
String table = args[args.length - 1];
Configuration conf = new Configuration(util.getConfiguration());
// populate input file
FileSystem fs = FileSystem.get(conf);
Path inputPath = fs.makeQualified(new Path(util.getDataTestDirOnTestFS(table), "input.dat"));
FSDataOutputStream op = fs.create(inputPath, true);
op.write(Bytes.toBytes(data));
op.close();
LOG.debug(String.format("Wrote test data to file: %s", inputPath));
if (conf.getBoolean(FORCE_COMBINER_CONF, true)) {
LOG.debug("Forcing combiner.");
conf.setInt("mapreduce.map.combine.minspills", 1);
}
// run the import
List<String> argv = new ArrayList<String>(Arrays.asList(args));
argv.add(inputPath.toString());
Tool tool = new ImportTsv();
LOG.debug("Running ImportTsv with arguments: " + argv);
assertEquals(0, ToolRunner.run(conf, tool, argv.toArray(args)));
validateTable(conf, TableName.valueOf(table), family, valueMultiplier, dataAvailable);
if (conf.getBoolean(DELETE_AFTER_LOAD_CONF, true)) {
LOG.debug("Deleting test subdirectory");
util.cleanupDataTestDirOnTestFS(table);
}
return tool;
}
@Override
public int run(String[] args) throws Exception {
Path rootRegionDir = null;
int numThreads = 1;
long numIterations = 1000000;
int numFamilies = 1;
int syncInterval = 0;
boolean noSync = false;
boolean verify = false;
boolean verbose = false;
boolean cleanup = true;
boolean noclosefs = false;
long roll = Long.MAX_VALUE;
boolean compress = false;
String cipher = null;
int numRegions = 1;
String spanReceivers = getConf().get("hbase.trace.spanreceiver.classes");
boolean trace = spanReceivers != null && !spanReceivers.isEmpty();
double traceFreq = 1.0;
// Process command line args
for (int i = 0; i < args.length; i++) {
String cmd = args[i];
try {
if (cmd.equals("-threads")) {
numThreads = Integer.parseInt(args[++i]);
} else if (cmd.equals("-iterations")) {
numIterations = Long.parseLong(args[++i]);
} else if (cmd.equals("-path")) {
rootRegionDir = new Path(args[++i]);
} else if (cmd.equals("-families")) {
numFamilies = Integer.parseInt(args[++i]);
} else if (cmd.equals("-qualifiers")) {
numQualifiers = Integer.parseInt(args[++i]);
} else if (cmd.equals("-keySize")) {
keySize = Integer.parseInt(args[++i]);
} else if (cmd.equals("-valueSize")) {
valueSize = Integer.parseInt(args[++i]);
} else if (cmd.equals("-syncInterval")) {
syncInterval = Integer.parseInt(args[++i]);
} else if (cmd.equals("-nosync")) {
noSync = true;
} else if (cmd.equals("-verify")) {
verify = true;
} else if (cmd.equals("-verbose")) {
verbose = true;
} else if (cmd.equals("-nocleanup")) {
cleanup = false;
} else if (cmd.equals("-noclosefs")) {
noclosefs = true;
} else if (cmd.equals("-roll")) {
roll = Long.parseLong(args[++i]);
} else if (cmd.equals("-compress")) {
compress = true;
} else if (cmd.equals("-encryption")) {
cipher = args[++i];
} else if (cmd.equals("-regions")) {
numRegions = Integer.parseInt(args[++i]);
} else if (cmd.equals("-traceFreq")) {
traceFreq = Double.parseDouble(args[++i]);
} else if (cmd.equals("-h")) {
printUsageAndExit();
} else if (cmd.equals("--help")) {
printUsageAndExit();
} else {
System.err.println("UNEXPECTED: " + cmd);
printUsageAndExit();
}
} catch (Exception e) {
printUsageAndExit();
}
}
if (compress) {
Configuration conf = getConf();
conf.setBoolean(HConstants.ENABLE_WAL_COMPRESSION, true);
}
if (cipher != null) {
// Set up WAL for encryption
Configuration conf = getConf();
conf.set(HConstants.CRYPTO_KEYPROVIDER_CONF_KEY, KeyProviderForTesting.class.getName());
conf.set(HConstants.CRYPTO_MASTERKEY_NAME_CONF_KEY, "hbase");
conf.setClass(
"hbase.regionserver.hlog.reader.impl", SecureProtobufLogReader.class, WAL.Reader.class);
conf.setClass(
"hbase.regionserver.hlog.writer.impl", SecureProtobufLogWriter.class, Writer.class);
conf.setBoolean(HConstants.ENABLE_WAL_ENCRYPTION, true);
conf.set(HConstants.CRYPTO_WAL_ALGORITHM_CONF_KEY, cipher);
}
if (numThreads < numRegions) {
LOG.warn("Number of threads is less than the number of regions; some regions will sit idle.");
}
// Internal config. goes off number of threads; if more threads than handlers, stuff breaks.
// In regionserver, number of handlers == number of threads.
getConf().setInt(HConstants.REGION_SERVER_HANDLER_COUNT, numThreads);
// Run WAL Performance Evaluation
// First set the fs from configs. In case we are on hadoop1
FSUtils.setFsDefault(getConf(), FSUtils.getRootDir(getConf()));
FileSystem fs = FileSystem.get(getConf());
LOG.info("FileSystem: " + fs);
SpanReceiverHost receiverHost = trace ? SpanReceiverHost.getInstance(getConf()) : null;
TraceScope scope = Trace.startSpan("WALPerfEval", trace ? Sampler.ALWAYS : Sampler.NEVER);
try {
if (rootRegionDir == null) {
rootRegionDir = TEST_UTIL.getDataTestDirOnTestFS("WALPerformanceEvaluation");
}
rootRegionDir = rootRegionDir.makeQualified(fs);
cleanRegionRootDir(fs, rootRegionDir);
FSUtils.setRootDir(getConf(), rootRegionDir);
final WALFactory wals = new WALFactory(getConf(), null, "wals");
final HRegion[] regions = new HRegion[numRegions];
final Runnable[] benchmarks = new Runnable[numRegions];
final MockRegionServerServices mockServices = new MockRegionServerServices(getConf());
final LogRoller roller = new LogRoller(mockServices, mockServices);
Threads.setDaemonThreadRunning(roller.getThread(), "WALPerfEval.logRoller");
try {
for (int i = 0; i < numRegions; i++) {
// Initialize Table Descriptor
// a table per desired region means we can avoid carving up the key space
final HTableDescriptor htd = createHTableDescriptor(i, numFamilies);
regions[i] = openRegion(fs, rootRegionDir, htd, wals, roll, roller);
benchmarks[i] =
Trace.wrap(
new WALPutBenchmark(
regions[i], htd, numIterations, noSync, syncInterval, traceFreq));
}
ConsoleReporter.enable(this.metrics, 30, TimeUnit.SECONDS);
long putTime = runBenchmark(benchmarks, numThreads);
logBenchmarkResult(
"Summary: threads="
+ numThreads
+ ", iterations="
+ numIterations
+ ", syncInterval="
+ syncInterval,
numIterations * numThreads,
putTime);
for (int i = 0; i < numRegions; i++) {
if (regions[i] != null) {
closeRegion(regions[i]);
regions[i] = null;
}
}
if (verify) {
LOG.info("verifying written log entries.");
Path dir =
new Path(
FSUtils.getRootDir(getConf()), DefaultWALProvider.getWALDirectoryName("wals"));
long editCount = 0;
FileStatus[] fsss = fs.listStatus(dir);
if (fsss.length == 0) throw new IllegalStateException("No WAL found");
for (FileStatus fss : fsss) {
Path p = fss.getPath();
if (!fs.exists(p)) throw new IllegalStateException(p.toString());
editCount += verify(wals, p, verbose);
}
long expected = numIterations * numThreads;
if (editCount != expected) {
throw new IllegalStateException("Counted=" + editCount + ", expected=" + expected);
}
}
} finally {
mockServices.stop("test clean up.");
for (int i = 0; i < numRegions; i++) {
if (regions[i] != null) {
closeRegion(regions[i]);
}
}
if (null != roller) {
LOG.info("shutting down log roller.");
Threads.shutdown(roller.getThread());
}
wals.shutdown();
// Remove the root dir for this test region
if (cleanup) cleanRegionRootDir(fs, rootRegionDir);
}
} finally {
// We may be called inside a test that wants to keep on using the fs.
if (!noclosefs) fs.close();
scope.close();
if (receiverHost != null) receiverHost.closeReceivers();
}
return (0);
}
/**
* Test that {@link HFileOutputFormat2} RecordWriter uses compression and bloom filter settings
* from the column family descriptor
*/
@Ignore("Goes zombie too frequently; needs work. See HBASE-14563")
@Test
public void testColumnFamilySettings() throws Exception {
Configuration conf = new Configuration(this.util.getConfiguration());
RecordWriter<ImmutableBytesWritable, Cell> writer = null;
TaskAttemptContext context = null;
Path dir = util.getDataTestDir("testColumnFamilySettings");
// Setup table descriptor
Table table = Mockito.mock(Table.class);
RegionLocator regionLocator = Mockito.mock(RegionLocator.class);
HTableDescriptor htd = new HTableDescriptor(TABLE_NAME);
Mockito.doReturn(htd).when(table).getTableDescriptor();
for (HColumnDescriptor hcd : HBaseTestingUtility.generateColumnDescriptors()) {
htd.addFamily(hcd);
}
// set up the table to return some mock keys
setupMockStartKeys(regionLocator);
try {
// partial map red setup to get an operational writer for testing
// We turn off the sequence file compression, because DefaultCodec
// 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(HFileOutputFormat2.LOCALITY_SENSITIVE_CONF_KEY, false);
Job job = new Job(conf, "testLocalMRIncrementalLoad");
job.setWorkingDirectory(util.getDataTestDirOnTestFS("testColumnFamilySettings"));
setupRandomGeneratorMapper(job);
HFileOutputFormat2.configureIncrementalLoad(job, table.getTableDescriptor(), regionLocator);
FileOutputFormat.setOutputPath(job, dir);
context = createTestTaskAttemptContext(job);
HFileOutputFormat2 hof = new HFileOutputFormat2();
writer = hof.getRecordWriter(context);
// write out random rows
writeRandomKeyValues(writer, context, htd.getFamiliesKeys(), ROWSPERSPLIT);
writer.close(context);
// Make sure that a directory was created for every CF
FileSystem fs = dir.getFileSystem(conf);
// commit so that the filesystem has one directory per column family
hof.getOutputCommitter(context).commitTask(context);
hof.getOutputCommitter(context).commitJob(context);
FileStatus[] families = FSUtils.listStatus(fs, dir, new FSUtils.FamilyDirFilter(fs));
assertEquals(htd.getFamilies().size(), families.length);
for (FileStatus f : families) {
String familyStr = f.getPath().getName();
HColumnDescriptor hcd = htd.getFamily(Bytes.toBytes(familyStr));
// verify that the compression on this file matches the configured
// compression
Path dataFilePath = fs.listStatus(f.getPath())[0].getPath();
Reader reader = HFile.createReader(fs, dataFilePath, new CacheConfig(conf), conf);
Map<byte[], byte[]> fileInfo = reader.loadFileInfo();
byte[] bloomFilter = fileInfo.get(StoreFile.BLOOM_FILTER_TYPE_KEY);
if (bloomFilter == null) bloomFilter = Bytes.toBytes("NONE");
assertEquals(
"Incorrect bloom filter used for column family "
+ familyStr
+ "(reader: "
+ reader
+ ")",
hcd.getBloomFilterType(),
BloomType.valueOf(Bytes.toString(bloomFilter)));
assertEquals(
"Incorrect compression used for column family "
+ familyStr
+ "(reader: "
+ reader
+ ")",
hcd.getCompressionType(),
reader.getFileContext().getCompression());
}
} finally {
dir.getFileSystem(conf).delete(dir, true);
}
}
/**
* This test is to test the scenario happened in HBASE-6901. All files are bulk loaded and
* excluded from minor compaction. Without the fix of HBASE-6901, an
* ArrayIndexOutOfBoundsException will be thrown.
*/
@Ignore("Flakey: See HBASE-9051")
@Test
public void testExcludeAllFromMinorCompaction() throws Exception {
Configuration conf = util.getConfiguration();
conf.setInt("hbase.hstore.compaction.min", 2);
generateRandomStartKeys(5);
util.startMiniCluster();
try (Connection conn = ConnectionFactory.createConnection();
Admin admin = conn.getAdmin();
Table table = util.createTable(TABLE_NAME, FAMILIES);
RegionLocator locator = conn.getRegionLocator(TABLE_NAME)) {
final FileSystem fs = util.getDFSCluster().getFileSystem();
assertEquals("Should start with empty table", 0, util.countRows(table));
// deep inspection: get the StoreFile dir
final Path storePath =
new Path(
FSUtils.getTableDir(FSUtils.getRootDir(conf), TABLE_NAME),
new Path(
admin.getTableRegions(TABLE_NAME).get(0).getEncodedName(),
Bytes.toString(FAMILIES[0])));
assertEquals(0, fs.listStatus(storePath).length);
// Generate two bulk load files
conf.setBoolean("hbase.mapreduce.hfileoutputformat.compaction.exclude", true);
for (int i = 0; i < 2; i++) {
Path testDir = util.getDataTestDirOnTestFS("testExcludeAllFromMinorCompaction_" + i);
runIncrementalPELoad(
conf, table.getTableDescriptor(), conn.getRegionLocator(TABLE_NAME), testDir);
// Perform the actual load
new LoadIncrementalHFiles(conf).doBulkLoad(testDir, admin, table, locator);
}
// Ensure data shows up
int expectedRows = 2 * NMapInputFormat.getNumMapTasks(conf) * ROWSPERSPLIT;
assertEquals(
"LoadIncrementalHFiles should put expected data in table",
expectedRows,
util.countRows(table));
// should have a second StoreFile now
assertEquals(2, fs.listStatus(storePath).length);
// minor compactions shouldn't get rid of the file
admin.compact(TABLE_NAME);
try {
quickPoll(
new Callable<Boolean>() {
@Override
public Boolean call() throws Exception {
return fs.listStatus(storePath).length == 1;
}
},
5000);
throw new IOException("SF# = " + fs.listStatus(storePath).length);
} catch (AssertionError ae) {
// this is expected behavior
}
// a major compaction should work though
admin.majorCompact(TABLE_NAME);
quickPoll(
new Callable<Boolean>() {
@Override
public Boolean call() throws Exception {
return fs.listStatus(storePath).length == 1;
}
},
5000);
} finally {
util.shutdownMiniCluster();
}
}
@Override
public int run(String[] args) throws Exception {
Path rootRegionDir = null;
int numThreads = 1;
long numIterations = 10000;
int numFamilies = 1;
boolean noSync = false;
boolean verify = false;
boolean verbose = false;
boolean cleanup = true;
boolean noclosefs = false;
long roll = Long.MAX_VALUE;
// Process command line args
for (int i = 0; i < args.length; i++) {
String cmd = args[i];
try {
if (cmd.equals("-threads")) {
numThreads = Integer.parseInt(args[++i]);
} else if (cmd.equals("-iterations")) {
numIterations = Long.parseLong(args[++i]);
} else if (cmd.equals("-path")) {
rootRegionDir = new Path(args[++i]);
} else if (cmd.equals("-families")) {
numFamilies = Integer.parseInt(args[++i]);
} else if (cmd.equals("-qualifiers")) {
numQualifiers = Integer.parseInt(args[++i]);
} else if (cmd.equals("-keySize")) {
keySize = Integer.parseInt(args[++i]);
} else if (cmd.equals("-valueSize")) {
valueSize = Integer.parseInt(args[++i]);
} else if (cmd.equals("-nosync")) {
noSync = true;
} else if (cmd.equals("-verify")) {
verify = true;
} else if (cmd.equals("-verbose")) {
verbose = true;
} else if (cmd.equals("-nocleanup")) {
cleanup = false;
} else if (cmd.equals("-noclosefs")) {
noclosefs = true;
} else if (cmd.equals("-roll")) {
roll = Long.parseLong(args[++i]);
} else if (cmd.equals("-h")) {
printUsageAndExit();
} else if (cmd.equals("--help")) {
printUsageAndExit();
} else {
System.err.println("UNEXPECTED: " + cmd);
printUsageAndExit();
}
} catch (Exception e) {
printUsageAndExit();
}
}
// Run HLog Performance Evaluation
// First set the fs from configs. In case we are on hadoop1
FSUtils.setFsDefault(getConf(), FSUtils.getRootDir(getConf()));
FileSystem fs = FileSystem.get(getConf());
LOG.info("FileSystem: " + fs);
try {
if (rootRegionDir == null) {
rootRegionDir = TEST_UTIL.getDataTestDirOnTestFS("HLogPerformanceEvaluation");
}
rootRegionDir = rootRegionDir.makeQualified(fs);
cleanRegionRootDir(fs, rootRegionDir);
// Initialize Table Descriptor
HTableDescriptor htd = createHTableDescriptor(numFamilies);
final long whenToRoll = roll;
HLog hlog =
new FSHLog(fs, rootRegionDir, "wals", getConf()) {
int appends = 0;
@Override
protected void doWrite(
HRegionInfo info, HLogKey logKey, WALEdit logEdit, HTableDescriptor htd)
throws IOException {
this.appends++;
if (this.appends % whenToRoll == 0) {
LOG.info("Rolling after " + appends + " edits");
rollWriter();
}
super.doWrite(info, logKey, logEdit, htd);
};
};
hlog.rollWriter();
HRegion region = null;
try {
region = openRegion(fs, rootRegionDir, htd, hlog);
long putTime =
runBenchmark(new HLogPutBenchmark(region, htd, numIterations, noSync), numThreads);
logBenchmarkResult(
"Summary: threads=" + numThreads + ", iterations=" + numIterations,
numIterations * numThreads,
putTime);
if (region != null) {
closeRegion(region);
region = null;
}
if (verify) {
Path dir = ((FSHLog) hlog).getDir();
long editCount = 0;
FileStatus[] fsss = fs.listStatus(dir);
if (fsss.length == 0) throw new IllegalStateException("No WAL found");
for (FileStatus fss : fsss) {
Path p = fss.getPath();
if (!fs.exists(p)) throw new IllegalStateException(p.toString());
editCount += verify(p, verbose);
}
long expected = numIterations * numThreads;
if (editCount != expected) {
throw new IllegalStateException("Counted=" + editCount + ", expected=" + expected);
}
}
} finally {
if (region != null) closeRegion(region);
// Remove the root dir for this test region
if (cleanup) cleanRegionRootDir(fs, rootRegionDir);
}
} finally {
// We may be called inside a test that wants to keep on using the fs.
if (!noclosefs) fs.close();
}
return (0);
}
private void doIncrementalLoadTest(boolean shouldChangeRegions, boolean shouldKeepLocality)
throws Exception {
util = new HBaseTestingUtility();
Configuration conf = util.getConfiguration();
conf.setBoolean(HFileOutputFormat2.LOCALITY_SENSITIVE_CONF_KEY, shouldKeepLocality);
int hostCount = 1;
int regionNum = 5;
if (shouldKeepLocality) {
// We should change host count higher than hdfs replica count when MiniHBaseCluster supports
// explicit hostnames parameter just like MiniDFSCluster does.
hostCount = 3;
regionNum = 20;
}
byte[][] splitKeys = generateRandomSplitKeys(regionNum - 1);
String[] hostnames = new String[hostCount];
for (int i = 0; i < hostCount; ++i) {
hostnames[i] = "datanode_" + i;
}
util.startMiniCluster(1, hostCount, hostnames);
Table table = util.createTable(TABLE_NAME, FAMILIES, splitKeys);
Path testDir = util.getDataTestDirOnTestFS("testLocalMRIncrementalLoad");
try (RegionLocator r = util.getConnection().getRegionLocator(TABLE_NAME);
Admin admin = util.getConnection().getAdmin(); ) {
assertEquals("Should start with empty table", 0, util.countRows(table));
int numRegions = r.getStartKeys().length;
assertEquals("Should make " + regionNum + " regions", numRegions, regionNum);
// Generate the bulk load files
runIncrementalPELoad(conf, table.getTableDescriptor(), r, testDir);
// This doesn't write into the table, just makes files
assertEquals("HFOF should not touch actual table", 0, util.countRows(table));
// Make sure that a directory was created for every CF
int dir = 0;
for (FileStatus f : testDir.getFileSystem(conf).listStatus(testDir)) {
for (byte[] family : FAMILIES) {
if (Bytes.toString(family).equals(f.getPath().getName())) {
++dir;
}
}
}
assertEquals("Column family not found in FS.", FAMILIES.length, dir);
// handle the split case
if (shouldChangeRegions) {
LOG.info("Changing regions in table");
admin.disableTable(table.getName());
while (util.getMiniHBaseCluster()
.getMaster()
.getAssignmentManager()
.getRegionStates()
.isRegionsInTransition()) {
Threads.sleep(200);
LOG.info("Waiting on table to finish disabling");
}
util.deleteTable(table.getName());
byte[][] newSplitKeys = generateRandomSplitKeys(14);
table = util.createTable(TABLE_NAME, FAMILIES, newSplitKeys);
while (util.getConnection().getRegionLocator(TABLE_NAME).getAllRegionLocations().size()
!= 15
|| !admin.isTableAvailable(table.getName())) {
Thread.sleep(200);
LOG.info("Waiting for new region assignment to happen");
}
}
// Perform the actual load
new LoadIncrementalHFiles(conf).doBulkLoad(testDir, admin, table, r);
// Ensure data shows up
int expectedRows = NMapInputFormat.getNumMapTasks(conf) * ROWSPERSPLIT;
assertEquals(
"LoadIncrementalHFiles should put expected data in table",
expectedRows,
util.countRows(table));
Scan scan = new Scan();
ResultScanner results = table.getScanner(scan);
for (Result res : results) {
assertEquals(FAMILIES.length, res.rawCells().length);
Cell first = res.rawCells()[0];
for (Cell kv : res.rawCells()) {
assertTrue(CellUtil.matchingRow(first, kv));
assertTrue(Bytes.equals(CellUtil.cloneValue(first), CellUtil.cloneValue(kv)));
}
}
results.close();
String tableDigestBefore = util.checksumRows(table);
// Check region locality
HDFSBlocksDistribution hbd = new HDFSBlocksDistribution();
for (HRegion region : util.getHBaseCluster().getRegions(TABLE_NAME)) {
hbd.add(region.getHDFSBlocksDistribution());
}
for (String hostname : hostnames) {
float locality = hbd.getBlockLocalityIndex(hostname);
LOG.info("locality of [" + hostname + "]: " + locality);
assertEquals(100, (int) (locality * 100));
}
// Cause regions to reopen
admin.disableTable(TABLE_NAME);
while (!admin.isTableDisabled(TABLE_NAME)) {
Thread.sleep(200);
LOG.info("Waiting for table to disable");
}
admin.enableTable(TABLE_NAME);
util.waitTableAvailable(TABLE_NAME);
assertEquals(
"Data should remain after reopening of regions",
tableDigestBefore,
util.checksumRows(table));
} finally {
testDir.getFileSystem(conf).delete(testDir, true);
util.deleteTable(TABLE_NAME);
util.shutdownMiniCluster();
}
}
@Test
public void testExcludeMinorCompaction() throws Exception {
Configuration conf = util.getConfiguration();
conf.setInt("hbase.hstore.compaction.min", 2);
generateRandomStartKeys(5);
try {
util.startMiniCluster();
Path testDir = util.getDataTestDirOnTestFS("testExcludeMinorCompaction");
final FileSystem fs = util.getDFSCluster().getFileSystem();
Admin admin = util.getHBaseAdmin();
HTable table = util.createTable(TABLE_NAME, FAMILIES);
assertEquals("Should start with empty table", 0, util.countRows(table));
// deep inspection: get the StoreFile dir
final Path storePath =
HStore.getStoreHomedir(
FSUtils.getTableDir(FSUtils.getRootDir(conf), TABLE_NAME),
admin.getTableRegions(TABLE_NAME).get(0),
FAMILIES[0]);
assertEquals(0, fs.listStatus(storePath).length);
// put some data in it and flush to create a storefile
Put p = new Put(Bytes.toBytes("test"));
p.add(FAMILIES[0], Bytes.toBytes("1"), Bytes.toBytes("1"));
table.put(p);
admin.flush(TABLE_NAME);
assertEquals(1, util.countRows(table));
quickPoll(
new Callable<Boolean>() {
public Boolean call() throws Exception {
return fs.listStatus(storePath).length == 1;
}
},
5000);
// Generate a bulk load file with more rows
conf.setBoolean("hbase.mapreduce.hfileoutputformat.compaction.exclude", true);
util.startMiniMapReduceCluster();
runIncrementalPELoad(conf, table, testDir);
// Perform the actual load
new LoadIncrementalHFiles(conf).doBulkLoad(testDir, table);
// Ensure data shows up
int expectedRows = NMapInputFormat.getNumMapTasks(conf) * ROWSPERSPLIT;
assertEquals(
"LoadIncrementalHFiles should put expected data in table",
expectedRows + 1,
util.countRows(table));
// should have a second StoreFile now
assertEquals(2, fs.listStatus(storePath).length);
// minor compactions shouldn't get rid of the file
admin.compact(TABLE_NAME);
try {
quickPoll(
new Callable<Boolean>() {
public Boolean call() throws Exception {
return fs.listStatus(storePath).length == 1;
}
},
5000);
throw new IOException("SF# = " + fs.listStatus(storePath).length);
} catch (AssertionError ae) {
// this is expected behavior
}
// a major compaction should work though
admin.majorCompact(TABLE_NAME);
quickPoll(
new Callable<Boolean>() {
public Boolean call() throws Exception {
return fs.listStatus(storePath).length == 1;
}
},
5000);
} finally {
util.shutdownMiniMapReduceCluster();
util.shutdownMiniCluster();
}
}
private void doIncrementalLoadTest(boolean shouldChangeRegions) throws Exception {
util = new HBaseTestingUtility();
Configuration conf = util.getConfiguration();
byte[][] splitKeys = generateRandomSplitKeys(4);
util.startMiniCluster();
try {
HTable table = util.createTable(TABLE_NAME, FAMILIES, splitKeys);
Admin admin = table.getConnection().getAdmin();
Path testDir = util.getDataTestDirOnTestFS("testLocalMRIncrementalLoad");
assertEquals("Should start with empty table", 0, util.countRows(table));
int numRegions = -1;
try (RegionLocator r = table.getRegionLocator()) {
numRegions = r.getStartKeys().length;
}
assertEquals("Should make 5 regions", numRegions, 5);
// Generate the bulk load files
util.startMiniMapReduceCluster();
runIncrementalPELoad(conf, table.getTableDescriptor(), table.getRegionLocator(), testDir);
// This doesn't write into the table, just makes files
assertEquals("HFOF should not touch actual table", 0, util.countRows(table));
// Make sure that a directory was created for every CF
int dir = 0;
for (FileStatus f : testDir.getFileSystem(conf).listStatus(testDir)) {
for (byte[] family : FAMILIES) {
if (Bytes.toString(family).equals(f.getPath().getName())) {
++dir;
}
}
}
assertEquals("Column family not found in FS.", FAMILIES.length, dir);
// handle the split case
if (shouldChangeRegions) {
LOG.info("Changing regions in table");
admin.disableTable(table.getName());
while (util.getMiniHBaseCluster()
.getMaster()
.getAssignmentManager()
.getRegionStates()
.isRegionsInTransition()) {
Threads.sleep(200);
LOG.info("Waiting on table to finish disabling");
}
util.deleteTable(table.getName());
byte[][] newSplitKeys = generateRandomSplitKeys(14);
table = util.createTable(TABLE_NAME, FAMILIES, newSplitKeys);
while (table.getRegionLocator().getAllRegionLocations().size() != 15
|| !admin.isTableAvailable(table.getName())) {
Thread.sleep(200);
LOG.info("Waiting for new region assignment to happen");
}
}
// Perform the actual load
new LoadIncrementalHFiles(conf).doBulkLoad(testDir, table);
// Ensure data shows up
int expectedRows = NMapInputFormat.getNumMapTasks(conf) * ROWSPERSPLIT;
assertEquals(
"LoadIncrementalHFiles should put expected data in table",
expectedRows,
util.countRows(table));
Scan scan = new Scan();
ResultScanner results = table.getScanner(scan);
for (Result res : results) {
assertEquals(FAMILIES.length, res.rawCells().length);
Cell first = res.rawCells()[0];
for (Cell kv : res.rawCells()) {
assertTrue(CellUtil.matchingRow(first, kv));
assertTrue(Bytes.equals(CellUtil.cloneValue(first), CellUtil.cloneValue(kv)));
}
}
results.close();
String tableDigestBefore = util.checksumRows(table);
// Cause regions to reopen
admin.disableTable(TABLE_NAME);
while (!admin.isTableDisabled(TABLE_NAME)) {
Thread.sleep(200);
LOG.info("Waiting for table to disable");
}
admin.enableTable(TABLE_NAME);
util.waitTableAvailable(TABLE_NAME);
assertEquals(
"Data should remain after reopening of regions",
tableDigestBefore,
util.checksumRows(table));
} finally {
util.shutdownMiniMapReduceCluster();
util.shutdownMiniCluster();
}
}
@Test(timeout = 30000)
public void testBulkLoad() throws IOException {
// Create table then get the single region for our new table.
LOG.debug("Creating test table");
HTableDescriptor hdt = HTU.createTableDescriptor("testBulkLoad");
hdt.setRegionReplication(NB_SERVERS);
hdt.addCoprocessor(SlowMeCopro.class.getName());
Table table = HTU.createTable(hdt, new byte[][] {f}, HTU.getConfiguration());
// create hfiles to load.
LOG.debug("Creating test data");
Path dir = HTU.getDataTestDirOnTestFS("testBulkLoad");
final int numRows = 10;
final byte[] qual = Bytes.toBytes("qual");
final byte[] val = Bytes.toBytes("val");
final List<Pair<byte[], String>> famPaths = new ArrayList<Pair<byte[], String>>();
for (HColumnDescriptor col : hdt.getColumnFamilies()) {
Path hfile = new Path(dir, col.getNameAsString());
TestHRegionServerBulkLoad.createHFile(
HTU.getTestFileSystem(), hfile, col.getName(), qual, val, numRows);
famPaths.add(new Pair<byte[], String>(col.getName(), hfile.toString()));
}
// bulk load HFiles
LOG.debug("Loading test data");
@SuppressWarnings("deprecation")
final HConnection conn = HTU.getHBaseAdmin().getConnection();
RegionServerCallable<Void> callable =
new RegionServerCallable<Void>(
conn, hdt.getTableName(), TestHRegionServerBulkLoad.rowkey(0)) {
@Override
public Void call(int timeout) throws Exception {
LOG.debug(
"Going to connect to server "
+ getLocation()
+ " for row "
+ Bytes.toStringBinary(getRow()));
byte[] regionName = getLocation().getRegionInfo().getRegionName();
BulkLoadHFileRequest request =
RequestConverter.buildBulkLoadHFileRequest(famPaths, regionName, true);
getStub().bulkLoadHFile(null, request);
return null;
}
};
RpcRetryingCallerFactory factory = new RpcRetryingCallerFactory(HTU.getConfiguration());
RpcRetryingCaller<Void> caller = factory.<Void>newCaller();
caller.callWithRetries(callable, 10000);
// verify we can read them from the primary
LOG.debug("Verifying data load");
for (int i = 0; i < numRows; i++) {
byte[] row = TestHRegionServerBulkLoad.rowkey(i);
Get g = new Get(row);
Result r = table.get(g);
Assert.assertFalse(r.isStale());
}
// verify we can read them from the replica
LOG.debug("Verifying replica queries");
try {
SlowMeCopro.cdl.set(new CountDownLatch(1));
for (int i = 0; i < numRows; i++) {
byte[] row = TestHRegionServerBulkLoad.rowkey(i);
Get g = new Get(row);
g.setConsistency(Consistency.TIMELINE);
Result r = table.get(g);
Assert.assertTrue(r.isStale());
}
SlowMeCopro.cdl.get().countDown();
} finally {
SlowMeCopro.cdl.get().countDown();
SlowMeCopro.sleepTime.set(0);
}
HTU.getHBaseAdmin().disableTable(hdt.getTableName());
HTU.deleteTable(hdt.getTableName());
}