private void testTruncateTable(final TableName tableName, boolean preserveSplits)
throws IOException {
byte[][] splitKeys = new byte[2][];
splitKeys[0] = Bytes.toBytes(4);
splitKeys[1] = Bytes.toBytes(8);
// Create & Fill the table
HTable table = TEST_UTIL.createTable(tableName, HConstants.CATALOG_FAMILY, splitKeys);
try {
TEST_UTIL.loadNumericRows(table, HConstants.CATALOG_FAMILY, 0, 10);
assertEquals(10, TEST_UTIL.countRows(table));
} finally {
table.close();
}
assertEquals(3, TEST_UTIL.getHBaseCluster().getRegions(tableName).size());
// Truncate & Verify
this.admin.disableTable(tableName);
this.admin.truncateTable(tableName, preserveSplits);
table = new HTable(TEST_UTIL.getConfiguration(), tableName);
try {
assertEquals(0, TEST_UTIL.countRows(table));
} finally {
table.close();
}
if (preserveSplits) {
assertEquals(3, TEST_UTIL.getHBaseCluster().getRegions(tableName).size());
} else {
assertEquals(1, TEST_UTIL.getHBaseCluster().getRegions(tableName).size());
}
}
@Test
public void testRollback() throws IOException {
final int rowcount = TEST_UTIL.loadRegion(this.parent, CF);
assertTrue(rowcount > 0);
int parentRowCount = TEST_UTIL.countRows(this.parent);
assertEquals(rowcount, parentRowCount);
// Start transaction.
HRegion spiedRegion = spy(this.parent);
SplitTransactionImpl st = prepareGOOD_SPLIT_ROW(spiedRegion);
SplitTransactionImpl spiedUponSt = spy(st);
doNothing()
.when(spiedUponSt)
.assertReferenceFileCount(
anyInt(), eq(parent.getRegionFileSystem().getSplitsDir(st.getFirstDaughter())));
when(spiedRegion.createDaughterRegionFromSplits(spiedUponSt.getSecondDaughter()))
.thenThrow(new MockedFailedDaughterCreation());
// Run the execute. Look at what it returns.
boolean expectedException = false;
Server mockServer = Mockito.mock(Server.class);
when(mockServer.getConfiguration()).thenReturn(TEST_UTIL.getConfiguration());
try {
spiedUponSt.execute(mockServer, null);
} catch (MockedFailedDaughterCreation e) {
expectedException = true;
}
assertTrue(expectedException);
// Run rollback
assertTrue(spiedUponSt.rollback(null, null));
// Assert I can scan parent.
int parentRowCount2 = TEST_UTIL.countRows(this.parent);
assertEquals(parentRowCount, parentRowCount2);
// Assert rollback cleaned up stuff in fs
assertTrue(!this.fs.exists(HRegion.getRegionDir(this.testdir, st.getFirstDaughter())));
assertTrue(!this.fs.exists(HRegion.getRegionDir(this.testdir, st.getSecondDaughter())));
assertTrue(!this.parent.lock.writeLock().isHeldByCurrentThread());
// Now retry the split but do not throw an exception this time.
assertTrue(st.prepare());
PairOfSameType<Region> daughters = st.execute(mockServer, null);
// Count rows. daughters are already open
int daughtersRowCount = 0;
for (Region openRegion : daughters) {
try {
int count = TEST_UTIL.countRows(openRegion);
assertTrue(count > 0 && count != rowcount);
daughtersRowCount += count;
} finally {
HBaseTestingUtility.closeRegionAndWAL(openRegion);
}
}
assertEquals(rowcount, daughtersRowCount);
// Assert the write lock is no longer held on parent
assertTrue(!this.parent.lock.writeLock().isHeldByCurrentThread());
assertTrue("Rollback hooks should be called.", wasRollBackHookCalled());
}
@Test
public void testWholesomeSplit() throws IOException {
final int rowcount = TEST_UTIL.loadRegion(this.parent, CF, true);
assertTrue(rowcount > 0);
int parentRowCount = TEST_UTIL.countRows(this.parent);
assertEquals(rowcount, parentRowCount);
// Pretend region's blocks are not in the cache, used for
// testWholesomeSplitWithHFileV1
CacheConfig cacheConf = new CacheConfig(TEST_UTIL.getConfiguration());
((LruBlockCache) cacheConf.getBlockCache()).clearCache();
// Start transaction.
SplitTransactionImpl st = prepareGOOD_SPLIT_ROW();
// Run the execute. Look at what it returns.
Server mockServer = Mockito.mock(Server.class);
when(mockServer.getConfiguration()).thenReturn(TEST_UTIL.getConfiguration());
PairOfSameType<Region> daughters = st.execute(mockServer, null);
// Do some assertions about execution.
assertTrue(this.fs.exists(this.parent.getRegionFileSystem().getSplitsDir()));
// Assert the parent region is closed.
assertTrue(this.parent.isClosed());
// Assert splitdir is empty -- because its content will have been moved out
// to be under the daughter region dirs.
assertEquals(0, this.fs.listStatus(this.parent.getRegionFileSystem().getSplitsDir()).length);
// Check daughters have correct key span.
assertTrue(
Bytes.equals(
parent.getRegionInfo().getStartKey(),
daughters.getFirst().getRegionInfo().getStartKey()));
assertTrue(Bytes.equals(GOOD_SPLIT_ROW, daughters.getFirst().getRegionInfo().getEndKey()));
assertTrue(Bytes.equals(daughters.getSecond().getRegionInfo().getStartKey(), GOOD_SPLIT_ROW));
assertTrue(
Bytes.equals(
parent.getRegionInfo().getEndKey(), daughters.getSecond().getRegionInfo().getEndKey()));
// Count rows. daughters are already open
int daughtersRowCount = 0;
for (Region openRegion : daughters) {
try {
int count = TEST_UTIL.countRows(openRegion);
assertTrue(count > 0 && count != rowcount);
daughtersRowCount += count;
} finally {
HBaseTestingUtility.closeRegionAndWAL(openRegion);
}
}
assertEquals(rowcount, daughtersRowCount);
// Assert the write lock is no longer held on parent
assertTrue(!this.parent.lock.writeLock().isHeldByCurrentThread());
}
@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();
}
}
}
@Test
public void testCountReferencesFailsSplit() throws IOException {
final int rowcount = TEST_UTIL.loadRegion(this.parent, CF);
assertTrue(rowcount > 0);
int parentRowCount = TEST_UTIL.countRows(this.parent);
assertEquals(rowcount, parentRowCount);
// Start transaction.
HRegion spiedRegion = spy(this.parent);
SplitTransactionImpl st = prepareGOOD_SPLIT_ROW(spiedRegion);
SplitTransactionImpl spiedUponSt = spy(st);
doThrow(new IOException("Failing split. Expected reference file count isn't equal."))
.when(spiedUponSt)
.assertReferenceFileCount(
anyInt(),
eq(
new Path(
this.parent.getRegionFileSystem().getTableDir(),
st.getSecondDaughter().getEncodedName())));
// Run the execute. Look at what it returns.
boolean expectedException = false;
Server mockServer = Mockito.mock(Server.class);
when(mockServer.getConfiguration()).thenReturn(TEST_UTIL.getConfiguration());
try {
spiedUponSt.execute(mockServer, null);
} catch (IOException e) {
expectedException = true;
}
assertTrue(expectedException);
}
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);
}
/*
* Take a snapshot of a table, add data, and verify that deleting the snapshot does not affect
* either table.
* @param online - Whether the table is online or not during the snapshot
*/
private void runTestSnapshotDeleteIndependent(boolean online) throws Exception {
FileSystem fs = UTIL.getHBaseCluster().getMaster().getMasterFileSystem().getFileSystem();
Path rootDir = UTIL.getHBaseCluster().getMaster().getMasterFileSystem().getRootDir();
final Admin admin = UTIL.getHBaseAdmin();
final long startTime = System.currentTimeMillis();
final TableName localTableName = TableName.valueOf(STRING_TABLE_NAME + startTime);
try (Table original = UTIL.createTable(localTableName, TEST_FAM)) {
UTIL.loadTable(original, TEST_FAM);
}
// Take a snapshot
final String snapshotNameAsString = "snapshot_" + localTableName;
byte[] snapshotName = Bytes.toBytes(snapshotNameAsString);
SnapshotTestingUtils.createSnapshotAndValidate(
admin, localTableName, TEST_FAM_STR, snapshotNameAsString, rootDir, fs, online);
if (!online) {
admin.enableTable(localTableName);
}
TableName cloneTableName = TableName.valueOf("test-clone-" + localTableName);
admin.cloneSnapshot(snapshotName, cloneTableName);
// Ensure the original table does not reference the HFiles anymore
admin.majorCompact(localTableName);
// Deleting the snapshot used to break the cloned table by deleting in-use HFiles
admin.deleteSnapshot(snapshotName);
// Wait for cleaner run and DFS heartbeats so that anything that is deletable is fully deleted
Thread.sleep(10000);
try (Table original = UTIL.getConnection().getTable(localTableName)) {
try (Table clonedTable = UTIL.getConnection().getTable(cloneTableName)) {
// Verify that all regions of both tables are readable
final int origTableRowCount = UTIL.countRows(original);
final int clonedTableRowCount = UTIL.countRows(clonedTable);
Assert.assertEquals(origTableRowCount, clonedTableRowCount);
}
}
}
/*
* Take a snapshot of a table, do a split, and verify that this only affects one table
* @param online - Whether the table is online or not during the snapshot
*/
private void runTestRegionOperationsIndependent(boolean online) throws Exception {
FileSystem fs = UTIL.getHBaseCluster().getMaster().getMasterFileSystem().getFileSystem();
Path rootDir = UTIL.getHBaseCluster().getMaster().getMasterFileSystem().getRootDir();
// Create a table
Admin admin = UTIL.getHBaseAdmin();
final long startTime = System.currentTimeMillis();
final TableName localTableName = TableName.valueOf(STRING_TABLE_NAME + startTime);
Table original = UTIL.createTable(localTableName, TEST_FAM);
UTIL.loadTable(original, TEST_FAM);
final int loadedTableCount = UTIL.countRows(original);
System.out.println("Original table has: " + loadedTableCount + " rows");
final String snapshotNameAsString = "snapshot_" + localTableName;
// Create a snapshot
SnapshotTestingUtils.createSnapshotAndValidate(
admin, localTableName, TEST_FAM_STR, snapshotNameAsString, rootDir, fs, online);
if (!online) {
admin.enableTable(localTableName);
}
TableName cloneTableName = TableName.valueOf("test-clone-" + localTableName);
// Clone the snapshot
byte[] snapshotName = Bytes.toBytes(snapshotNameAsString);
admin.cloneSnapshot(snapshotName, cloneTableName);
// Verify that region information is the same pre-split
((ClusterConnection) UTIL.getConnection()).clearRegionCache();
List<HRegionInfo> originalTableHRegions = admin.getTableRegions(localTableName);
final int originalRegionCount = originalTableHRegions.size();
final int cloneTableRegionCount = admin.getTableRegions(cloneTableName).size();
Assert.assertEquals(
"The number of regions in the cloned table is different than in the original table.",
originalRegionCount,
cloneTableRegionCount);
// Split a region on the parent table
admin.splitRegion(originalTableHRegions.get(0).getRegionName());
waitOnSplit(UTIL.getConnection(), original, originalRegionCount);
// Verify that the cloned table region is not split
final int cloneTableRegionCount2 = admin.getTableRegions(cloneTableName).size();
Assert.assertEquals(
"The number of regions in the cloned table changed though none of its regions were split.",
cloneTableRegionCount,
cloneTableRegionCount2);
}
@Test
public void testMetaScanner() throws Exception {
LOG.info("Starting testMetaScanner");
setUp();
final TableName TABLENAME = TableName.valueOf("testMetaScanner");
final byte[] FAMILY = Bytes.toBytes("family");
TEST_UTIL.createTable(TABLENAME, FAMILY);
Configuration conf = TEST_UTIL.getConfiguration();
HTable table = (HTable) connection.getTable(TABLENAME);
TEST_UTIL.createMultiRegions(
conf,
table,
FAMILY,
new byte[][] {
HConstants.EMPTY_START_ROW, Bytes.toBytes("region_a"), Bytes.toBytes("region_b")
});
// Make sure all the regions are deployed
TEST_UTIL.countRows(table);
MetaScanner.MetaScannerVisitor visitor = mock(MetaScanner.MetaScannerVisitor.class);
doReturn(true).when(visitor).processRow((Result) anyObject());
// Scanning the entire table should give us three rows
MetaScanner.metaScan(connection, visitor, TABLENAME);
verify(visitor, times(3)).processRow((Result) anyObject());
// Scanning the table with a specified empty start row should also
// give us three hbase:meta rows
reset(visitor);
doReturn(true).when(visitor).processRow((Result) anyObject());
MetaScanner.metaScan(connection, visitor, TABLENAME, HConstants.EMPTY_BYTE_ARRAY, 1000);
verify(visitor, times(3)).processRow((Result) anyObject());
// Scanning the table starting in the middle should give us two rows:
// region_a and region_b
reset(visitor);
doReturn(true).when(visitor).processRow((Result) anyObject());
MetaScanner.metaScan(connection, visitor, TABLENAME, Bytes.toBytes("region_ac"), 1000);
verify(visitor, times(2)).processRow((Result) anyObject());
// Scanning with a limit of 1 should only give us one row
reset(visitor);
doReturn(true).when(visitor).processRow((Result) anyObject());
MetaScanner.metaScan(connection, visitor, TABLENAME, Bytes.toBytes("region_ac"), 1);
verify(visitor, times(1)).processRow((Result) anyObject());
table.close();
}
/**
* 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);
}
}
@Test
public void testFailAfterPONR() throws IOException, KeeperException {
final int rowcount = TEST_UTIL.loadRegion(this.parent, CF);
assertTrue(rowcount > 0);
int parentRowCount = TEST_UTIL.countRows(this.parent);
assertEquals(rowcount, parentRowCount);
// Start transaction.
SplitTransactionImpl st = prepareGOOD_SPLIT_ROW();
SplitTransactionImpl spiedUponSt = spy(st);
Mockito.doThrow(new MockedFailedDaughterOpen())
.when(spiedUponSt)
.openDaughterRegion((Server) Mockito.anyObject(), (HRegion) Mockito.anyObject());
// Run the execute. Look at what it returns.
boolean expectedException = false;
Server mockServer = Mockito.mock(Server.class);
when(mockServer.getConfiguration()).thenReturn(TEST_UTIL.getConfiguration());
try {
spiedUponSt.execute(mockServer, null);
} catch (IOException e) {
if (e.getCause() != null && e.getCause() instanceof MockedFailedDaughterOpen) {
expectedException = true;
}
}
assertTrue(expectedException);
// Run rollback returns that we should restart.
assertFalse(spiedUponSt.rollback(null, null));
// Make sure that region a and region b are still in the filesystem, that
// they have not been removed; this is supposed to be the case if we go
// past point of no return.
Path tableDir = this.parent.getRegionFileSystem().getTableDir();
Path daughterADir = new Path(tableDir, spiedUponSt.getFirstDaughter().getEncodedName());
Path daughterBDir = new Path(tableDir, spiedUponSt.getSecondDaughter().getEncodedName());
assertTrue(TEST_UTIL.getTestFileSystem().exists(daughterADir));
assertTrue(TEST_UTIL.getTestFileSystem().exists(daughterBDir));
}
protected int countRows(final Table table, final byte[]... families) throws IOException {
return UTIL.countRows(table, families);
}
/**
* 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();
}
}
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();
}
}
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();
}
}
/*
* Take a snapshot of a table, add data, and verify that this only
* affects one table
* @param online - Whether the table is online or not during the snapshot
*/
private void runTestSnapshotAppendIndependent(boolean online) throws Exception {
FileSystem fs = UTIL.getHBaseCluster().getMaster().getMasterFileSystem().getFileSystem();
Path rootDir = UTIL.getHBaseCluster().getMaster().getMasterFileSystem().getRootDir();
Admin admin = UTIL.getHBaseAdmin();
final long startTime = System.currentTimeMillis();
final TableName localTableName = TableName.valueOf(STRING_TABLE_NAME + startTime);
try (Table original = UTIL.createTable(localTableName, TEST_FAM)) {
UTIL.loadTable(original, TEST_FAM);
final int origTableRowCount = UTIL.countRows(original);
// Take a snapshot
final String snapshotNameAsString = "snapshot_" + localTableName;
byte[] snapshotName = Bytes.toBytes(snapshotNameAsString);
SnapshotTestingUtils.createSnapshotAndValidate(
admin, localTableName, TEST_FAM_STR, snapshotNameAsString, rootDir, fs, online);
if (!online) {
admin.enableTable(localTableName);
}
TableName cloneTableName = TableName.valueOf("test-clone-" + localTableName);
admin.cloneSnapshot(snapshotName, cloneTableName);
try (Table clonedTable = UTIL.getConnection().getTable(cloneTableName)) {
final int clonedTableRowCount = UTIL.countRows(clonedTable);
Assert.assertEquals(
"The line counts of original and cloned tables do not match after clone. ",
origTableRowCount,
clonedTableRowCount);
// Attempt to add data to the test
final String rowKey = "new-row-" + System.currentTimeMillis();
Put p = new Put(Bytes.toBytes(rowKey));
p.add(TEST_FAM, Bytes.toBytes("someQualifier"), Bytes.toBytes("someString"));
original.put(p);
// Verify that it is not present in the original table
Assert.assertEquals(
"The row count of the original table was not modified by the put",
origTableRowCount + 1,
UTIL.countRows(original));
Assert.assertEquals(
"The row count of the cloned table changed as a result of addition to the original",
clonedTableRowCount,
UTIL.countRows(clonedTable));
p = new Put(Bytes.toBytes(rowKey));
p.add(TEST_FAM, Bytes.toBytes("someQualifier"), Bytes.toBytes("someString"));
clonedTable.put(p);
// Verify that the new family is not in the restored table's description
Assert.assertEquals(
"The row count of the original table was modified by the put to the clone",
origTableRowCount + 1,
UTIL.countRows(original));
Assert.assertEquals(
"The row count of the cloned table was not modified by the put",
clonedTableRowCount + 1,
UTIL.countRows(clonedTable));
}
}
}
@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();
}
}