Java Code Examples for org.apache.hadoop.hdfs.MiniDFSCluster#shutdownNameNode()
The following examples show how to use
org.apache.hadoop.hdfs.MiniDFSCluster#shutdownNameNode() .
You can vote up the ones you like or vote down the ones you don't like,
and go to the original project or source file by following the links above each example. You may check out the related API usage on the sidebar.
Example 1
| Source File: HdfsSortedOplogOrganizerJUnitTest.java From gemfirexd-oss with Apache License 2.0 | 5 votes |
|
public void testActiveReaderWithNameNodeHA() throws Exception {
deleteMiniClusterDir();
int nn1port = AvailablePortHelper.getRandomAvailableTCPPort();
int nn2port = AvailablePortHelper.getRandomAvailableTCPPort();
MiniDFSCluster cluster = initMiniHACluster(nn1port, nn2port);
initClientHAConf(nn1port, nn2port);
HDFSStoreImpl store1 = (HDFSStoreImpl) hsf.create("Store-1");
regionfactory.setHDFSStoreName(store1.getName());
Region<Object, Object> region1 = regionfactory.create("region-1");
HdfsRegionManager regionManager1 = ((LocalRegion)region1).getHdfsRegionManager();
HdfsSortedOplogOrganizer organizer = new HdfsSortedOplogOrganizer(regionManager1, 0);
ArrayList<TestEvent> items = new ArrayList<TestEvent>();
for (int i = 100000; i < 101000; i++) {
items.add(new TestEvent(("" + i), (i + " some string " + i)));
}
organizer.flush(items.iterator(), items.size());
organizer.getSortedOplogs().get(0).get().getReader();
TestUtils.addExpectedException("java.io.EOFException");
NameNode nnode2 = cluster.getNameNode(1);
assertTrue(nnode2.isStandbyState());
cluster.shutdownNameNode(0);
cluster.transitionToActive(1);
assertFalse(nnode2.isStandbyState());
for (int i = 100000; i < 100500; i++) {
byte[] keyBytes1 = BlobHelper.serializeToBlob("" + i);
assertEquals(i + " some string " + i, organizer.read(keyBytes1).getValue());
}
TestUtils.removeExpectedException("java.io.EOFException");
region1.destroyRegion();
store1.destroy();
cluster.shutdown();
FileUtils.deleteDirectory(new File("hdfs-test-cluster"));
}
Example 2
| Source File: TestWebHDFSForHA.java From big-c with Apache License 2.0 | 5 votes |
|
@Test
public void testSecureHAToken() throws IOException, InterruptedException {
Configuration conf = DFSTestUtil.newHAConfiguration(LOGICAL_NAME);
conf.setBoolean(DFSConfigKeys
.DFS_NAMENODE_DELEGATION_TOKEN_ALWAYS_USE_KEY, true);
MiniDFSCluster cluster = null;
WebHdfsFileSystem fs = null;
try {
cluster = new MiniDFSCluster.Builder(conf).nnTopology(topo)
.numDataNodes(0).build();
HATestUtil.setFailoverConfigurations(cluster, conf, LOGICAL_NAME);
cluster.waitActive();
fs = spy((WebHdfsFileSystem) FileSystem.get(WEBHDFS_URI, conf));
FileSystemTestHelper.addFileSystemForTesting(WEBHDFS_URI, conf, fs);
cluster.transitionToActive(0);
Token<?> token = fs.getDelegationToken(null);
cluster.shutdownNameNode(0);
cluster.transitionToActive(1);
token.renew(conf);
token.cancel(conf);
verify(fs).renewDelegationToken(token);
verify(fs).cancelDelegationToken(token);
} finally {
IOUtils.cleanup(null, fs);
if (cluster != null) {
cluster.shutdown();
}
}
}
Example 3
| Source File: HdfsSortedOplogOrganizerJUnitTest.java From gemfirexd-oss with Apache License 2.0 | 5 votes |
|
public void testFlushWithNameNodeHA() throws Exception {
deleteMiniClusterDir();
int nn1port = AvailablePortHelper.getRandomAvailableTCPPort();
int nn2port = AvailablePortHelper.getRandomAvailableTCPPort();
MiniDFSCluster cluster = initMiniHACluster(nn1port, nn2port);
initClientHAConf(nn1port, nn2port);
HDFSStoreImpl store1 = (HDFSStoreImpl) hsf.create("Store-1");
regionfactory.setHDFSStoreName(store1.getName());
Region<Object, Object> region1 = regionfactory.create("region-1");
HdfsRegionManager regionManager1 = ((LocalRegion)region1).getHdfsRegionManager();
HoplogOrganizer<SortedHoplogPersistedEvent> organizer = new HdfsSortedOplogOrganizer(regionManager1, 0);
ArrayList<TestEvent> items = new ArrayList<TestEvent>();
items.add(new TestEvent(("1"), ("1-1")));
organizer.flush(items.iterator(), items.size());
TestUtils.addExpectedException("java.io.EOFException");
NameNode nnode2 = cluster.getNameNode(1);
assertTrue(nnode2.isStandbyState());
cluster.shutdownNameNode(0);
cluster.transitionToActive(1);
assertFalse(nnode2.isStandbyState());
items.add(new TestEvent(("4"), ("1-4")));
organizer.flush(items.iterator(), items.size());
byte[] keyBytes1 = BlobHelper.serializeToBlob("1");
byte[] keyBytes4 = BlobHelper.serializeToBlob("4");
assertEquals("1-1", organizer.read(keyBytes1).getValue());
assertEquals("1-4", organizer.read(keyBytes4).getValue());
TestUtils.removeExpectedException("java.io.EOFException");
region1.destroyRegion();
store1.destroy();
cluster.shutdown();
FileUtils.deleteDirectory(new File("hdfs-test-cluster"));
}
Example 4
| Source File: HdfsSortedOplogOrganizerJUnitTest.java From gemfirexd-oss with Apache License 2.0 | 5 votes |
|
public void testFlushWithNameNodeHA() throws Exception {
deleteMiniClusterDir();
int nn1port = AvailablePortHelper.getRandomAvailableTCPPort();
int nn2port = AvailablePortHelper.getRandomAvailableTCPPort();
MiniDFSCluster cluster = initMiniHACluster(nn1port, nn2port);
initClientHAConf(nn1port, nn2port);
HDFSStoreImpl store1 = (HDFSStoreImpl) hsf.create("Store-1");
regionfactory.setHDFSStoreName(store1.getName());
Region<Object, Object> region1 = regionfactory.create("region-1");
HdfsRegionManager regionManager1 = ((LocalRegion)region1).getHdfsRegionManager();
HoplogOrganizer<SortedHoplogPersistedEvent> organizer = new HdfsSortedOplogOrganizer(regionManager1, 0);
ArrayList<TestEvent> items = new ArrayList<TestEvent>();
items.add(new TestEvent(("1"), ("1-1")));
organizer.flush(items.iterator(), items.size());
TestUtils.addExpectedException("java.io.EOFException");
NameNode nnode2 = cluster.getNameNode(1);
assertTrue(nnode2.isStandbyState());
cluster.shutdownNameNode(0);
cluster.transitionToActive(1);
assertFalse(nnode2.isStandbyState());
items.add(new TestEvent(("4"), ("1-4")));
organizer.flush(items.iterator(), items.size());
byte[] keyBytes1 = BlobHelper.serializeToBlob("1");
byte[] keyBytes4 = BlobHelper.serializeToBlob("4");
assertEquals("1-1", organizer.read(keyBytes1).getValue());
assertEquals("1-4", organizer.read(keyBytes4).getValue());
TestUtils.removeExpectedException("java.io.EOFException");
region1.destroyRegion();
store1.destroy();
cluster.shutdown();
FileUtils.deleteDirectory(new File("hdfs-test-cluster"));
}
Example 5
| Source File: HdfsSortedOplogOrganizerJUnitTest.java From gemfirexd-oss with Apache License 2.0 | 5 votes |
|
public void testNewReaderWithNameNodeHA() throws Exception {
deleteMiniClusterDir();
int nn1port = AvailablePortHelper.getRandomAvailableTCPPort();
int nn2port = AvailablePortHelper.getRandomAvailableTCPPort();
MiniDFSCluster cluster = initMiniHACluster(nn1port, nn2port);
initClientHAConf(nn1port, nn2port);
HDFSStoreImpl store1 = (HDFSStoreImpl) hsf.create("Store-1");
regionfactory.setHDFSStoreName(store1.getName());
Region<Object, Object> region1 = regionfactory.create("region-1");
HdfsRegionManager regionManager1 = ((LocalRegion)region1).getHdfsRegionManager();
HoplogOrganizer<SortedHoplogPersistedEvent> organizer = doRead(regionManager1);
organizer.close();
TestUtils.addExpectedException("java.io.EOFException");
NameNode nnode2 = cluster.getNameNode(1);
assertTrue(nnode2.isStandbyState());
cluster.shutdownNameNode(0);
cluster.transitionToActive(1);
assertFalse(nnode2.isStandbyState());
organizer = new HdfsSortedOplogOrganizer(regionManager1, 0);
byte[] keyBytes1 = BlobHelper.serializeToBlob("1");
byte[] keyBytes3 = BlobHelper.serializeToBlob("3");
byte[] keyBytes4 = BlobHelper.serializeToBlob("4");
assertEquals("2-1", organizer.read(keyBytes1).getValue());
assertEquals("3-3", organizer.read(keyBytes3).getValue());
assertEquals("1-4", organizer.read(keyBytes4).getValue());
TestUtils.removeExpectedException("java.io.EOFException");
region1.destroyRegion();
store1.destroy();
cluster.shutdown();
FileUtils.deleteDirectory(new File("hdfs-test-cluster"));
}
Example 6
| Source File: TestBookKeeperAsHASharedDir.java From big-c with Apache License 2.0 | 5 votes |
|
/**
* Test simple HA failover usecase with BK
*/
@Test
public void testFailoverWithBK() throws Exception {
MiniDFSCluster cluster = null;
try {
Configuration conf = new Configuration();
conf.setInt(DFSConfigKeys.DFS_HA_TAILEDITS_PERIOD_KEY, 1);
conf.set(DFSConfigKeys.DFS_NAMENODE_SHARED_EDITS_DIR_KEY,
BKJMUtil.createJournalURI("/hotfailover").toString());
BKJMUtil.addJournalManagerDefinition(conf);
cluster = new MiniDFSCluster.Builder(conf)
.nnTopology(MiniDFSNNTopology.simpleHATopology())
.numDataNodes(0)
.manageNameDfsSharedDirs(false)
.build();
NameNode nn1 = cluster.getNameNode(0);
NameNode nn2 = cluster.getNameNode(1);
cluster.waitActive();
cluster.transitionToActive(0);
Path p = new Path("/testBKJMfailover");
FileSystem fs = HATestUtil.configureFailoverFs(cluster, conf);
fs.mkdirs(p);
cluster.shutdownNameNode(0);
cluster.transitionToActive(1);
assertTrue(fs.exists(p));
} finally {
if (cluster != null) {
cluster.shutdown();
}
}
}
Example 7
| Source File: TestWebHDFSForHA.java From hadoop with Apache License 2.0 | 5 votes |
|
@Test
public void testHA() throws IOException {
Configuration conf = DFSTestUtil.newHAConfiguration(LOGICAL_NAME);
MiniDFSCluster cluster = null;
FileSystem fs = null;
try {
cluster = new MiniDFSCluster.Builder(conf).nnTopology(topo)
.numDataNodes(0).build();
HATestUtil.setFailoverConfigurations(cluster, conf, LOGICAL_NAME);
cluster.waitActive();
fs = FileSystem.get(WEBHDFS_URI, conf);
cluster.transitionToActive(0);
final Path dir = new Path("/test");
Assert.assertTrue(fs.mkdirs(dir));
cluster.shutdownNameNode(0);
cluster.transitionToActive(1);
final Path dir2 = new Path("/test2");
Assert.assertTrue(fs.mkdirs(dir2));
} finally {
IOUtils.cleanup(null, fs);
if (cluster != null) {
cluster.shutdown();
}
}
}
Example 8
| Source File: TestDFSUpgradeWithHA.java From hadoop with Apache License 2.0 | 4 votes |
|
@Test
public void testRollbackWithJournalNodes() throws IOException,
URISyntaxException {
MiniQJMHACluster qjCluster = null;
FileSystem fs = null;
try {
Builder builder = new MiniQJMHACluster.Builder(conf);
builder.getDfsBuilder()
.numDataNodes(0);
qjCluster = builder.build();
MiniDFSCluster cluster = qjCluster.getDfsCluster();
// No upgrade is in progress at the moment.
checkClusterPreviousDirExistence(cluster, false);
assertCTimesEqual(cluster);
checkJnPreviousDirExistence(qjCluster, false);
// Transition NN0 to active and do some FS ops.
cluster.transitionToActive(0);
fs = HATestUtil.configureFailoverFs(cluster, conf);
assertTrue(fs.mkdirs(new Path("/foo1")));
final long cidBeforeUpgrade = getCommittedTxnIdValue(qjCluster);
// Do the upgrade. Shut down NN1 and then restart NN0 with the upgrade
// flag.
cluster.shutdownNameNode(1);
cluster.getNameNodeInfos()[0].setStartOpt(StartupOption.UPGRADE);
cluster.restartNameNode(0, false);
checkNnPreviousDirExistence(cluster, 0, true);
checkNnPreviousDirExistence(cluster, 1, false);
checkJnPreviousDirExistence(qjCluster, true);
// NN0 should come up in the active state when given the -upgrade option,
// so no need to transition it to active.
assertTrue(fs.mkdirs(new Path("/foo2")));
final long cidDuringUpgrade = getCommittedTxnIdValue(qjCluster);
assertTrue(cidDuringUpgrade > cidBeforeUpgrade);
// Now bootstrap the standby with the upgraded info.
int rc = BootstrapStandby.run(
new String[]{"-force"},
cluster.getConfiguration(1));
assertEquals(0, rc);
cluster.restartNameNode(1);
checkNnPreviousDirExistence(cluster, 0, true);
checkNnPreviousDirExistence(cluster, 1, true);
checkJnPreviousDirExistence(qjCluster, true);
assertCTimesEqual(cluster);
// Shut down the NNs, but deliberately leave the JNs up and running.
Collection<URI> nn1NameDirs = cluster.getNameDirs(0);
cluster.shutdown();
conf.setStrings(DFSConfigKeys.DFS_NAMENODE_NAME_DIR_KEY, Joiner.on(",").join(nn1NameDirs));
NameNode.doRollback(conf, false);
final long cidAfterRollback = getCommittedTxnIdValue(qjCluster);
assertTrue(cidBeforeUpgrade < cidAfterRollback);
// make sure the committedTxnId has been reset correctly after rollback
assertTrue(cidDuringUpgrade > cidAfterRollback);
// The rollback operation should have rolled back the first NN's local
// dirs, and the shared dir, but not the other NN's dirs. Those have to be
// done by bootstrapping the standby.
checkNnPreviousDirExistence(cluster, 0, false);
checkJnPreviousDirExistence(qjCluster, false);
} finally {
if (fs != null) {
fs.close();
}
if (qjCluster != null) {
qjCluster.shutdown();
}
}
}
Example 9
| Source File: TestDFSUpgradeWithHA.java From big-c with Apache License 2.0 | 4 votes |
|
/**
* Make sure that even if the NN which initiated the upgrade is in the standby
* state that we're allowed to finalize.
*/
@Test
public void testFinalizeFromSecondNameNodeWithJournalNodes()
throws IOException, URISyntaxException {
MiniQJMHACluster qjCluster = null;
FileSystem fs = null;
try {
Builder builder = new MiniQJMHACluster.Builder(conf);
builder.getDfsBuilder()
.numDataNodes(0);
qjCluster = builder.build();
MiniDFSCluster cluster = qjCluster.getDfsCluster();
// No upgrade is in progress at the moment.
checkJnPreviousDirExistence(qjCluster, false);
checkClusterPreviousDirExistence(cluster, false);
assertCTimesEqual(cluster);
// Transition NN0 to active and do some FS ops.
cluster.transitionToActive(0);
fs = HATestUtil.configureFailoverFs(cluster, conf);
assertTrue(fs.mkdirs(new Path("/foo1")));
// Do the upgrade. Shut down NN1 and then restart NN0 with the upgrade
// flag.
cluster.shutdownNameNode(1);
cluster.getNameNodeInfos()[0].setStartOpt(StartupOption.UPGRADE);
cluster.restartNameNode(0, false);
checkNnPreviousDirExistence(cluster, 0, true);
checkNnPreviousDirExistence(cluster, 1, false);
checkJnPreviousDirExistence(qjCluster, true);
// Now bootstrap the standby with the upgraded info.
int rc = BootstrapStandby.run(
new String[]{"-force"},
cluster.getConfiguration(1));
assertEquals(0, rc);
cluster.restartNameNode(1);
// Make the second NN (not the one that initiated the upgrade) active when
// the finalize command is run.
cluster.transitionToStandby(0);
cluster.transitionToActive(1);
runFinalizeCommand(cluster);
checkClusterPreviousDirExistence(cluster, false);
checkJnPreviousDirExistence(qjCluster, false);
assertCTimesEqual(cluster);
} finally {
if (fs != null) {
fs.close();
}
if (qjCluster != null) {
qjCluster.shutdown();
}
}
}
Example 10
| Source File: TestDFSUpgradeWithHA.java From big-c with Apache License 2.0 | 4 votes |
|
/**
* Ensure that an admin cannot finalize an HA upgrade without at least one NN
* being active.
*/
@Test
public void testCannotFinalizeIfNoActive() throws IOException,
URISyntaxException {
MiniDFSCluster cluster = null;
FileSystem fs = null;
try {
cluster = new MiniDFSCluster.Builder(conf)
.nnTopology(MiniDFSNNTopology.simpleHATopology())
.numDataNodes(0)
.build();
File sharedDir = new File(cluster.getSharedEditsDir(0, 1));
// No upgrade is in progress at the moment.
checkClusterPreviousDirExistence(cluster, false);
assertCTimesEqual(cluster);
checkPreviousDirExistence(sharedDir, false);
// Transition NN0 to active and do some FS ops.
cluster.transitionToActive(0);
fs = HATestUtil.configureFailoverFs(cluster, conf);
assertTrue(fs.mkdirs(new Path("/foo1")));
// Do the upgrade. Shut down NN1 and then restart NN0 with the upgrade
// flag.
cluster.shutdownNameNode(1);
cluster.getNameNodeInfos()[0].setStartOpt(StartupOption.UPGRADE);
cluster.restartNameNode(0, false);
checkNnPreviousDirExistence(cluster, 0, true);
checkNnPreviousDirExistence(cluster, 1, false);
checkPreviousDirExistence(sharedDir, true);
// NN0 should come up in the active state when given the -upgrade option,
// so no need to transition it to active.
assertTrue(fs.mkdirs(new Path("/foo2")));
// Restart NN0 without the -upgrade flag, to make sure that works.
cluster.getNameNodeInfos()[0].setStartOpt(StartupOption.REGULAR);
cluster.restartNameNode(0, false);
// Make sure we can still do FS ops after upgrading.
cluster.transitionToActive(0);
assertTrue(fs.mkdirs(new Path("/foo3")));
// Now bootstrap the standby with the upgraded info.
int rc = BootstrapStandby.run(
new String[]{"-force"},
cluster.getConfiguration(1));
assertEquals(0, rc);
// Now restart NN1 and make sure that we can do ops against that as well.
cluster.restartNameNode(1);
cluster.transitionToStandby(0);
cluster.transitionToActive(1);
assertTrue(fs.mkdirs(new Path("/foo4")));
assertCTimesEqual(cluster);
// Now there's no active NN.
cluster.transitionToStandby(1);
try {
runFinalizeCommand(cluster);
fail("Should not have been able to finalize upgrade with no NN active");
} catch (IOException ioe) {
GenericTestUtils.assertExceptionContains(
"Cannot finalize with no NameNode active", ioe);
}
} finally {
if (fs != null) {
fs.close();
}
if (cluster != null) {
cluster.shutdown();
}
}
}
Example 11
| Source File: TestBookKeeperAsHASharedDir.java From hadoop with Apache License 2.0 | 4 votes |
|
/**
* Test HA failover, where BK, as the shared storage, fails.
* Once it becomes available again, a standby can come up.
* Verify that any write happening after the BK fail is not
* available on the standby.
*/
@Test
public void testFailoverWithFailingBKCluster() throws Exception {
int ensembleSize = numBookies + 1;
BookieServer newBookie = bkutil.newBookie();
assertEquals("New bookie didn't start",
ensembleSize, bkutil.checkBookiesUp(ensembleSize, 10));
BookieServer replacementBookie = null;
MiniDFSCluster cluster = null;
try {
Configuration conf = new Configuration();
conf.setInt(DFSConfigKeys.DFS_HA_TAILEDITS_PERIOD_KEY, 1);
conf.set(DFSConfigKeys.DFS_NAMENODE_SHARED_EDITS_DIR_KEY,
BKJMUtil.createJournalURI("/hotfailoverWithFail").toString());
conf.setInt(BookKeeperJournalManager.BKJM_BOOKKEEPER_ENSEMBLE_SIZE,
ensembleSize);
conf.setInt(BookKeeperJournalManager.BKJM_BOOKKEEPER_QUORUM_SIZE,
ensembleSize);
BKJMUtil.addJournalManagerDefinition(conf);
cluster = new MiniDFSCluster.Builder(conf)
.nnTopology(MiniDFSNNTopology.simpleHATopology())
.numDataNodes(0)
.manageNameDfsSharedDirs(false)
.checkExitOnShutdown(false)
.build();
NameNode nn1 = cluster.getNameNode(0);
NameNode nn2 = cluster.getNameNode(1);
cluster.waitActive();
cluster.transitionToActive(0);
Path p1 = new Path("/testBKJMFailingBKCluster1");
Path p2 = new Path("/testBKJMFailingBKCluster2");
FileSystem fs = HATestUtil.configureFailoverFs(cluster, conf);
fs.mkdirs(p1);
newBookie.shutdown(); // will take down shared storage
assertEquals("New bookie didn't stop",
numBookies, bkutil.checkBookiesUp(numBookies, 10));
try {
fs.mkdirs(p2);
fail("mkdirs should result in the NN exiting");
} catch (RemoteException re) {
assertTrue(re.getClassName().contains("ExitException"));
}
cluster.shutdownNameNode(0);
try {
cluster.transitionToActive(1);
fail("Shouldn't have been able to transition with bookies down");
} catch (ExitException ee) {
assertTrue("Should shutdown due to required journal failure",
ee.getMessage().contains(
"starting log segment 3 failed for required journal"));
}
replacementBookie = bkutil.newBookie();
assertEquals("Replacement bookie didn't start",
ensembleSize, bkutil.checkBookiesUp(ensembleSize, 10));
cluster.transitionToActive(1); // should work fine now
assertTrue(fs.exists(p1));
assertFalse(fs.exists(p2));
} finally {
newBookie.shutdown();
if (replacementBookie != null) {
replacementBookie.shutdown();
}
if (cluster != null) {
cluster.shutdown();
}
}
}
Example 12
| Source File: TestDFSUpgradeWithHA.java From hadoop with Apache License 2.0 | 4 votes |
|
/**
* Make sure that starting a second NN with the -upgrade flag fails if the
* other NN has already done that.
*/
@Test
public void testCannotUpgradeSecondNameNode() throws IOException,
URISyntaxException {
MiniDFSCluster cluster = null;
FileSystem fs = null;
try {
cluster = new MiniDFSCluster.Builder(conf)
.nnTopology(MiniDFSNNTopology.simpleHATopology())
.numDataNodes(0)
.build();
File sharedDir = new File(cluster.getSharedEditsDir(0, 1));
// No upgrade is in progress at the moment.
checkClusterPreviousDirExistence(cluster, false);
assertCTimesEqual(cluster);
checkPreviousDirExistence(sharedDir, false);
// Transition NN0 to active and do some FS ops.
cluster.transitionToActive(0);
fs = HATestUtil.configureFailoverFs(cluster, conf);
assertTrue(fs.mkdirs(new Path("/foo1")));
// Do the upgrade. Shut down NN1 and then restart NN0 with the upgrade
// flag.
cluster.shutdownNameNode(1);
cluster.getNameNodeInfos()[0].setStartOpt(StartupOption.UPGRADE);
cluster.restartNameNode(0, false);
checkNnPreviousDirExistence(cluster, 0, true);
checkNnPreviousDirExistence(cluster, 1, false);
checkPreviousDirExistence(sharedDir, true);
// NN0 should come up in the active state when given the -upgrade option,
// so no need to transition it to active.
assertTrue(fs.mkdirs(new Path("/foo2")));
// Restart NN0 without the -upgrade flag, to make sure that works.
cluster.getNameNodeInfos()[0].setStartOpt(StartupOption.REGULAR);
cluster.restartNameNode(0, false);
// Make sure we can still do FS ops after upgrading.
cluster.transitionToActive(0);
assertTrue(fs.mkdirs(new Path("/foo3")));
// Make sure that starting the second NN with the -upgrade flag fails.
cluster.getNameNodeInfos()[1].setStartOpt(StartupOption.UPGRADE);
try {
cluster.restartNameNode(1, false);
fail("Should not have been able to start second NN with -upgrade");
} catch (IOException ioe) {
GenericTestUtils.assertExceptionContains(
"It looks like the shared log is already being upgraded", ioe);
}
} finally {
if (fs != null) {
fs.close();
}
if (cluster != null) {
cluster.shutdown();
}
}
}
Example 13
| Source File: TestDFSUpgradeWithHA.java From hadoop with Apache License 2.0 | 4 votes |
|
/**
* Test rollback with NFS shared dir.
*/
@Test
public void testRollbackWithNfs() throws Exception {
MiniDFSCluster cluster = null;
FileSystem fs = null;
try {
cluster = new MiniDFSCluster.Builder(conf)
.nnTopology(MiniDFSNNTopology.simpleHATopology())
.numDataNodes(0)
.build();
File sharedDir = new File(cluster.getSharedEditsDir(0, 1));
// No upgrade is in progress at the moment.
checkClusterPreviousDirExistence(cluster, false);
assertCTimesEqual(cluster);
checkPreviousDirExistence(sharedDir, false);
// Transition NN0 to active and do some FS ops.
cluster.transitionToActive(0);
fs = HATestUtil.configureFailoverFs(cluster, conf);
assertTrue(fs.mkdirs(new Path("/foo1")));
// Do the upgrade. Shut down NN1 and then restart NN0 with the upgrade
// flag.
cluster.shutdownNameNode(1);
cluster.getNameNodeInfos()[0].setStartOpt(StartupOption.UPGRADE);
cluster.restartNameNode(0, false);
checkNnPreviousDirExistence(cluster, 0, true);
checkNnPreviousDirExistence(cluster, 1, false);
checkPreviousDirExistence(sharedDir, true);
// NN0 should come up in the active state when given the -upgrade option,
// so no need to transition it to active.
assertTrue(fs.mkdirs(new Path("/foo2")));
// Now bootstrap the standby with the upgraded info.
int rc = BootstrapStandby.run(
new String[]{"-force"},
cluster.getConfiguration(1));
assertEquals(0, rc);
cluster.restartNameNode(1);
checkNnPreviousDirExistence(cluster, 0, true);
checkNnPreviousDirExistence(cluster, 1, true);
checkPreviousDirExistence(sharedDir, true);
assertCTimesEqual(cluster);
// Now shut down the cluster and do the rollback.
Collection<URI> nn1NameDirs = cluster.getNameDirs(0);
cluster.shutdown();
conf.setStrings(DFSConfigKeys.DFS_NAMENODE_NAME_DIR_KEY, Joiner.on(",").join(nn1NameDirs));
NameNode.doRollback(conf, false);
// The rollback operation should have rolled back the first NN's local
// dirs, and the shared dir, but not the other NN's dirs. Those have to be
// done by bootstrapping the standby.
checkNnPreviousDirExistence(cluster, 0, false);
checkPreviousDirExistence(sharedDir, false);
} finally {
if (fs != null) {
fs.close();
}
if (cluster != null) {
cluster.shutdown();
}
}
}
Example 14
| Source File: TestDFSUpgradeWithHA.java From hadoop with Apache License 2.0 | 4 votes |
|
/**
* Make sure that even if the NN which initiated the upgrade is in the standby
* state that we're allowed to finalize.
*/
@Test
public void testFinalizeFromSecondNameNodeWithJournalNodes()
throws IOException, URISyntaxException {
MiniQJMHACluster qjCluster = null;
FileSystem fs = null;
try {
Builder builder = new MiniQJMHACluster.Builder(conf);
builder.getDfsBuilder()
.numDataNodes(0);
qjCluster = builder.build();
MiniDFSCluster cluster = qjCluster.getDfsCluster();
// No upgrade is in progress at the moment.
checkJnPreviousDirExistence(qjCluster, false);
checkClusterPreviousDirExistence(cluster, false);
assertCTimesEqual(cluster);
// Transition NN0 to active and do some FS ops.
cluster.transitionToActive(0);
fs = HATestUtil.configureFailoverFs(cluster, conf);
assertTrue(fs.mkdirs(new Path("/foo1")));
// Do the upgrade. Shut down NN1 and then restart NN0 with the upgrade
// flag.
cluster.shutdownNameNode(1);
cluster.getNameNodeInfos()[0].setStartOpt(StartupOption.UPGRADE);
cluster.restartNameNode(0, false);
checkNnPreviousDirExistence(cluster, 0, true);
checkNnPreviousDirExistence(cluster, 1, false);
checkJnPreviousDirExistence(qjCluster, true);
// Now bootstrap the standby with the upgraded info.
int rc = BootstrapStandby.run(
new String[]{"-force"},
cluster.getConfiguration(1));
assertEquals(0, rc);
cluster.restartNameNode(1);
// Make the second NN (not the one that initiated the upgrade) active when
// the finalize command is run.
cluster.transitionToStandby(0);
cluster.transitionToActive(1);
runFinalizeCommand(cluster);
checkClusterPreviousDirExistence(cluster, false);
checkJnPreviousDirExistence(qjCluster, false);
assertCTimesEqual(cluster);
} finally {
if (fs != null) {
fs.close();
}
if (qjCluster != null) {
qjCluster.shutdown();
}
}
}
Example 15
| Source File: TestHAAppend.java From big-c with Apache License 2.0 | 4 votes |
|
/**
* Test to verify the processing of PendingDataNodeMessageQueue in case of
* append. One block will marked as corrupt if the OP_ADD, OP_UPDATE_BLOCKS
* comes in one edit log segment and OP_CLOSE edit comes in next log segment
* which is loaded during failover. Regression test for HDFS-3605.
*/
@Test
public void testMultipleAppendsDuringCatchupTailing() throws Exception {
Configuration conf = new Configuration();
// Set a length edits tailing period, and explicit rolling, so we can
// control the ingest of edits by the standby for this test.
conf.set(DFSConfigKeys.DFS_HA_TAILEDITS_PERIOD_KEY, "5000");
conf.setInt(DFSConfigKeys.DFS_HA_LOGROLL_PERIOD_KEY, -1);
MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf)
.nnTopology(MiniDFSNNTopology.simpleHATopology())
.numDataNodes(3).build();
FileSystem fs = null;
try {
cluster.transitionToActive(0);
fs = HATestUtil.configureFailoverFs(cluster, conf);
Path fileToAppend = new Path("/FileToAppend");
Path fileToTruncate = new Path("/FileToTruncate");
final byte[] data = new byte[1 << 16];
DFSUtil.getRandom().nextBytes(data);
final int[] appendPos = AppendTestUtil.randomFilePartition(
data.length, COUNT);
final int[] truncatePos = AppendTestUtil.randomFilePartition(
data.length, 1);
// Create file, write some data, and hflush so that the first
// block is in the edit log prior to roll.
FSDataOutputStream out = createAndHflush(
fs, fileToAppend, data, appendPos[0]);
FSDataOutputStream out4Truncate = createAndHflush(
fs, fileToTruncate, data, data.length);
// Let the StandbyNode catch the creation of the file.
cluster.getNameNode(0).getRpcServer().rollEditLog();
cluster.getNameNode(1).getNamesystem().getEditLogTailer().doTailEdits();
out.close();
out4Truncate.close();
// Append and re-close a few time, so that many block entries are queued.
for (int i = 0; i < COUNT; i++) {
int end = i < COUNT - 1? appendPos[i + 1]: data.length;
out = fs.append(fileToAppend);
out.write(data, appendPos[i], end - appendPos[i]);
out.close();
}
boolean isTruncateReady = fs.truncate(fileToTruncate, truncatePos[0]);
// Ensure that blocks have been reported to the SBN ahead of the edits
// arriving.
cluster.triggerBlockReports();
// Failover the current standby to active.
cluster.shutdownNameNode(0);
cluster.transitionToActive(1);
// Check the FSCK doesn't detect any bad blocks on the SBN.
int rc = ToolRunner.run(new DFSck(cluster.getConfiguration(1)),
new String[] { "/", "-files", "-blocks" });
assertEquals(0, rc);
assertEquals("CorruptBlocks should be empty.", 0, cluster.getNameNode(1)
.getNamesystem().getCorruptReplicaBlocks());
AppendTestUtil.checkFullFile(fs, fileToAppend, data.length, data,
fileToAppend.toString());
if (!isTruncateReady) {
TestFileTruncate.checkBlockRecovery(fileToTruncate,
cluster.getFileSystem(1));
}
AppendTestUtil.checkFullFile(fs, fileToTruncate, truncatePos[0], data,
fileToTruncate.toString());
} finally {
if (null != cluster) {
cluster.shutdown();
}
if (null != fs) {
fs.close();
}
}
}
Example 16
| Source File: TestDFSUpgradeWithHA.java From hadoop with Apache License 2.0 | 4 votes |
|
/**
* Make sure that an HA NN can successfully upgrade when configured using
* JournalNodes.
*/
@Test
public void testUpgradeWithJournalNodes() throws IOException,
URISyntaxException {
MiniQJMHACluster qjCluster = null;
FileSystem fs = null;
try {
Builder builder = new MiniQJMHACluster.Builder(conf);
builder.getDfsBuilder()
.numDataNodes(0);
qjCluster = builder.build();
MiniDFSCluster cluster = qjCluster.getDfsCluster();
// No upgrade is in progress at the moment.
checkJnPreviousDirExistence(qjCluster, false);
checkClusterPreviousDirExistence(cluster, false);
assertCTimesEqual(cluster);
// Transition NN0 to active and do some FS ops.
cluster.transitionToActive(0);
fs = HATestUtil.configureFailoverFs(cluster, conf);
assertTrue(fs.mkdirs(new Path("/foo1")));
// get the value of the committedTxnId in journal nodes
final long cidBeforeUpgrade = getCommittedTxnIdValue(qjCluster);
// Do the upgrade. Shut down NN1 and then restart NN0 with the upgrade
// flag.
cluster.shutdownNameNode(1);
cluster.getNameNodeInfos()[0].setStartOpt(StartupOption.UPGRADE);
cluster.restartNameNode(0, false);
checkNnPreviousDirExistence(cluster, 0, true);
checkNnPreviousDirExistence(cluster, 1, false);
checkJnPreviousDirExistence(qjCluster, true);
assertTrue(cidBeforeUpgrade <= getCommittedTxnIdValue(qjCluster));
// NN0 should come up in the active state when given the -upgrade option,
// so no need to transition it to active.
assertTrue(fs.mkdirs(new Path("/foo2")));
// Restart NN0 without the -upgrade flag, to make sure that works.
cluster.getNameNodeInfos()[0].setStartOpt(StartupOption.REGULAR);
cluster.restartNameNode(0, false);
// Make sure we can still do FS ops after upgrading.
cluster.transitionToActive(0);
assertTrue(fs.mkdirs(new Path("/foo3")));
assertTrue(getCommittedTxnIdValue(qjCluster) > cidBeforeUpgrade);
// Now bootstrap the standby with the upgraded info.
int rc = BootstrapStandby.run(
new String[]{"-force"},
cluster.getConfiguration(1));
assertEquals(0, rc);
// Now restart NN1 and make sure that we can do ops against that as well.
cluster.restartNameNode(1);
cluster.transitionToStandby(0);
cluster.transitionToActive(1);
assertTrue(fs.mkdirs(new Path("/foo4")));
assertCTimesEqual(cluster);
} finally {
if (fs != null) {
fs.close();
}
if (qjCluster != null) {
qjCluster.shutdown();
}
}
}
Example 17
| Source File: TestBookKeeperAsHASharedDir.java From big-c with Apache License 2.0 | 4 votes |
|
/**
* Test HA failover, where BK, as the shared storage, fails.
* Once it becomes available again, a standby can come up.
* Verify that any write happening after the BK fail is not
* available on the standby.
*/
@Test
public void testFailoverWithFailingBKCluster() throws Exception {
int ensembleSize = numBookies + 1;
BookieServer newBookie = bkutil.newBookie();
assertEquals("New bookie didn't start",
ensembleSize, bkutil.checkBookiesUp(ensembleSize, 10));
BookieServer replacementBookie = null;
MiniDFSCluster cluster = null;
try {
Configuration conf = new Configuration();
conf.setInt(DFSConfigKeys.DFS_HA_TAILEDITS_PERIOD_KEY, 1);
conf.set(DFSConfigKeys.DFS_NAMENODE_SHARED_EDITS_DIR_KEY,
BKJMUtil.createJournalURI("/hotfailoverWithFail").toString());
conf.setInt(BookKeeperJournalManager.BKJM_BOOKKEEPER_ENSEMBLE_SIZE,
ensembleSize);
conf.setInt(BookKeeperJournalManager.BKJM_BOOKKEEPER_QUORUM_SIZE,
ensembleSize);
BKJMUtil.addJournalManagerDefinition(conf);
cluster = new MiniDFSCluster.Builder(conf)
.nnTopology(MiniDFSNNTopology.simpleHATopology())
.numDataNodes(0)
.manageNameDfsSharedDirs(false)
.checkExitOnShutdown(false)
.build();
NameNode nn1 = cluster.getNameNode(0);
NameNode nn2 = cluster.getNameNode(1);
cluster.waitActive();
cluster.transitionToActive(0);
Path p1 = new Path("/testBKJMFailingBKCluster1");
Path p2 = new Path("/testBKJMFailingBKCluster2");
FileSystem fs = HATestUtil.configureFailoverFs(cluster, conf);
fs.mkdirs(p1);
newBookie.shutdown(); // will take down shared storage
assertEquals("New bookie didn't stop",
numBookies, bkutil.checkBookiesUp(numBookies, 10));
try {
fs.mkdirs(p2);
fail("mkdirs should result in the NN exiting");
} catch (RemoteException re) {
assertTrue(re.getClassName().contains("ExitException"));
}
cluster.shutdownNameNode(0);
try {
cluster.transitionToActive(1);
fail("Shouldn't have been able to transition with bookies down");
} catch (ExitException ee) {
assertTrue("Should shutdown due to required journal failure",
ee.getMessage().contains(
"starting log segment 3 failed for required journal"));
}
replacementBookie = bkutil.newBookie();
assertEquals("Replacement bookie didn't start",
ensembleSize, bkutil.checkBookiesUp(ensembleSize, 10));
cluster.transitionToActive(1); // should work fine now
assertTrue(fs.exists(p1));
assertFalse(fs.exists(p2));
} finally {
newBookie.shutdown();
if (replacementBookie != null) {
replacementBookie.shutdown();
}
if (cluster != null) {
cluster.shutdown();
}
}
}
Example 18
| Source File: TestHAAppend.java From hadoop with Apache License 2.0 | 4 votes |
|
/**
* Test to verify the processing of PendingDataNodeMessageQueue in case of
* append. One block will marked as corrupt if the OP_ADD, OP_UPDATE_BLOCKS
* comes in one edit log segment and OP_CLOSE edit comes in next log segment
* which is loaded during failover. Regression test for HDFS-3605.
*/
@Test
public void testMultipleAppendsDuringCatchupTailing() throws Exception {
Configuration conf = new Configuration();
// Set a length edits tailing period, and explicit rolling, so we can
// control the ingest of edits by the standby for this test.
conf.set(DFSConfigKeys.DFS_HA_TAILEDITS_PERIOD_KEY, "5000");
conf.setInt(DFSConfigKeys.DFS_HA_LOGROLL_PERIOD_KEY, -1);
MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf)
.nnTopology(MiniDFSNNTopology.simpleHATopology())
.numDataNodes(3).build();
FileSystem fs = null;
try {
cluster.transitionToActive(0);
fs = HATestUtil.configureFailoverFs(cluster, conf);
Path fileToAppend = new Path("/FileToAppend");
Path fileToTruncate = new Path("/FileToTruncate");
final byte[] data = new byte[1 << 16];
DFSUtil.getRandom().nextBytes(data);
final int[] appendPos = AppendTestUtil.randomFilePartition(
data.length, COUNT);
final int[] truncatePos = AppendTestUtil.randomFilePartition(
data.length, 1);
// Create file, write some data, and hflush so that the first
// block is in the edit log prior to roll.
FSDataOutputStream out = createAndHflush(
fs, fileToAppend, data, appendPos[0]);
FSDataOutputStream out4Truncate = createAndHflush(
fs, fileToTruncate, data, data.length);
// Let the StandbyNode catch the creation of the file.
cluster.getNameNode(0).getRpcServer().rollEditLog();
cluster.getNameNode(1).getNamesystem().getEditLogTailer().doTailEdits();
out.close();
out4Truncate.close();
// Append and re-close a few time, so that many block entries are queued.
for (int i = 0; i < COUNT; i++) {
int end = i < COUNT - 1? appendPos[i + 1]: data.length;
out = fs.append(fileToAppend);
out.write(data, appendPos[i], end - appendPos[i]);
out.close();
}
boolean isTruncateReady = fs.truncate(fileToTruncate, truncatePos[0]);
// Ensure that blocks have been reported to the SBN ahead of the edits
// arriving.
cluster.triggerBlockReports();
// Failover the current standby to active.
cluster.shutdownNameNode(0);
cluster.transitionToActive(1);
// Check the FSCK doesn't detect any bad blocks on the SBN.
int rc = ToolRunner.run(new DFSck(cluster.getConfiguration(1)),
new String[] { "/", "-files", "-blocks" });
assertEquals(0, rc);
assertEquals("CorruptBlocks should be empty.", 0, cluster.getNameNode(1)
.getNamesystem().getCorruptReplicaBlocks());
AppendTestUtil.checkFullFile(fs, fileToAppend, data.length, data,
fileToAppend.toString());
if (!isTruncateReady) {
TestFileTruncate.checkBlockRecovery(fileToTruncate,
cluster.getFileSystem(1));
}
AppendTestUtil.checkFullFile(fs, fileToTruncate, truncatePos[0], data,
fileToTruncate.toString());
} finally {
if (null != cluster) {
cluster.shutdown();
}
if (null != fs) {
fs.close();
}
}
}
Example 19
| Source File: TestDFSUpgradeWithHA.java From big-c with Apache License 2.0 | 4 votes |
|
/**
* Test rollback with NFS shared dir.
*/
@Test
public void testRollbackWithNfs() throws Exception {
MiniDFSCluster cluster = null;
FileSystem fs = null;
try {
cluster = new MiniDFSCluster.Builder(conf)
.nnTopology(MiniDFSNNTopology.simpleHATopology())
.numDataNodes(0)
.build();
File sharedDir = new File(cluster.getSharedEditsDir(0, 1));
// No upgrade is in progress at the moment.
checkClusterPreviousDirExistence(cluster, false);
assertCTimesEqual(cluster);
checkPreviousDirExistence(sharedDir, false);
// Transition NN0 to active and do some FS ops.
cluster.transitionToActive(0);
fs = HATestUtil.configureFailoverFs(cluster, conf);
assertTrue(fs.mkdirs(new Path("/foo1")));
// Do the upgrade. Shut down NN1 and then restart NN0 with the upgrade
// flag.
cluster.shutdownNameNode(1);
cluster.getNameNodeInfos()[0].setStartOpt(StartupOption.UPGRADE);
cluster.restartNameNode(0, false);
checkNnPreviousDirExistence(cluster, 0, true);
checkNnPreviousDirExistence(cluster, 1, false);
checkPreviousDirExistence(sharedDir, true);
// NN0 should come up in the active state when given the -upgrade option,
// so no need to transition it to active.
assertTrue(fs.mkdirs(new Path("/foo2")));
// Now bootstrap the standby with the upgraded info.
int rc = BootstrapStandby.run(
new String[]{"-force"},
cluster.getConfiguration(1));
assertEquals(0, rc);
cluster.restartNameNode(1);
checkNnPreviousDirExistence(cluster, 0, true);
checkNnPreviousDirExistence(cluster, 1, true);
checkPreviousDirExistence(sharedDir, true);
assertCTimesEqual(cluster);
// Now shut down the cluster and do the rollback.
Collection<URI> nn1NameDirs = cluster.getNameDirs(0);
cluster.shutdown();
conf.setStrings(DFSConfigKeys.DFS_NAMENODE_NAME_DIR_KEY, Joiner.on(",").join(nn1NameDirs));
NameNode.doRollback(conf, false);
// The rollback operation should have rolled back the first NN's local
// dirs, and the shared dir, but not the other NN's dirs. Those have to be
// done by bootstrapping the standby.
checkNnPreviousDirExistence(cluster, 0, false);
checkPreviousDirExistence(sharedDir, false);
} finally {
if (fs != null) {
fs.close();
}
if (cluster != null) {
cluster.shutdown();
}
}
}
Example 20
| Source File: TestDFSUpgradeWithHA.java From big-c with Apache License 2.0 | 4 votes |
|
/**
* Make sure that an HA NN can successfully upgrade when configured using
* JournalNodes.
*/
@Test
public void testUpgradeWithJournalNodes() throws IOException,
URISyntaxException {
MiniQJMHACluster qjCluster = null;
FileSystem fs = null;
try {
Builder builder = new MiniQJMHACluster.Builder(conf);
builder.getDfsBuilder()
.numDataNodes(0);
qjCluster = builder.build();
MiniDFSCluster cluster = qjCluster.getDfsCluster();
// No upgrade is in progress at the moment.
checkJnPreviousDirExistence(qjCluster, false);
checkClusterPreviousDirExistence(cluster, false);
assertCTimesEqual(cluster);
// Transition NN0 to active and do some FS ops.
cluster.transitionToActive(0);
fs = HATestUtil.configureFailoverFs(cluster, conf);
assertTrue(fs.mkdirs(new Path("/foo1")));
// get the value of the committedTxnId in journal nodes
final long cidBeforeUpgrade = getCommittedTxnIdValue(qjCluster);
// Do the upgrade. Shut down NN1 and then restart NN0 with the upgrade
// flag.
cluster.shutdownNameNode(1);
cluster.getNameNodeInfos()[0].setStartOpt(StartupOption.UPGRADE);
cluster.restartNameNode(0, false);
checkNnPreviousDirExistence(cluster, 0, true);
checkNnPreviousDirExistence(cluster, 1, false);
checkJnPreviousDirExistence(qjCluster, true);
assertTrue(cidBeforeUpgrade <= getCommittedTxnIdValue(qjCluster));
// NN0 should come up in the active state when given the -upgrade option,
// so no need to transition it to active.
assertTrue(fs.mkdirs(new Path("/foo2")));
// Restart NN0 without the -upgrade flag, to make sure that works.
cluster.getNameNodeInfos()[0].setStartOpt(StartupOption.REGULAR);
cluster.restartNameNode(0, false);
// Make sure we can still do FS ops after upgrading.
cluster.transitionToActive(0);
assertTrue(fs.mkdirs(new Path("/foo3")));
assertTrue(getCommittedTxnIdValue(qjCluster) > cidBeforeUpgrade);
// Now bootstrap the standby with the upgraded info.
int rc = BootstrapStandby.run(
new String[]{"-force"},
cluster.getConfiguration(1));
assertEquals(0, rc);
// Now restart NN1 and make sure that we can do ops against that as well.
cluster.restartNameNode(1);
cluster.transitionToStandby(0);
cluster.transitionToActive(1);
assertTrue(fs.mkdirs(new Path("/foo4")));
assertCTimesEqual(cluster);
} finally {
if (fs != null) {
fs.close();
}
if (qjCluster != null) {
qjCluster.shutdown();
}
}
}
