@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);
}
/**
* Test if the region can be closed properly
*
* @throws IOException
* @throws NodeExistsException
* @throws KeeperException
* @throws org.apache.hadoop.hbase.exceptions.DeserializationException
*/
@Test
public void testCloseRegion()
throws IOException, NodeExistsException, KeeperException, DeserializationException {
final Server server = new MockServer(HTU);
final RegionServerServices rss = HTU.createMockRegionServerService();
HTableDescriptor htd = TEST_HTD;
HRegionInfo hri = TEST_HRI;
// open a region first so that it can be closed later
OpenRegion(server, rss, htd, hri);
// close the region
// Create it CLOSING, which is what Master set before sending CLOSE RPC
int versionOfClosingNode =
ZKAssign.createNodeClosing(server.getZooKeeper(), hri, server.getServerName());
// The CloseRegionHandler will validate the expected version
// Given it is set to correct versionOfClosingNode,
// CloseRegionHandlerit should be RS_ZK_REGION_CLOSED
CloseRegionHandler handler =
new CloseRegionHandler(server, rss, hri, false, true, versionOfClosingNode);
handler.process();
// Handler should have transitioned it to RS_ZK_REGION_CLOSED
RegionTransition rt =
RegionTransition.parseFrom(ZKAssign.getData(server.getZooKeeper(), hri.getEncodedName()));
assertTrue(rt.getEventType().equals(EventType.RS_ZK_REGION_CLOSED));
}
/**
* Test that if we fail a flush, abort gets set on close.
*
* @see <a href="https://issues.apache.org/jira/browse/HBASE-4270">HBASE-4270</a>
* @throws IOException
* @throws NodeExistsException
* @throws KeeperException
*/
@Test
public void testFailedFlushAborts() throws IOException, NodeExistsException, KeeperException {
final Server server = new MockServer(HTU, false);
final RegionServerServices rss = HTU.createMockRegionServerService();
HTableDescriptor htd = TEST_HTD;
final HRegionInfo hri =
new HRegionInfo(htd.getTableName(), HConstants.EMPTY_END_ROW, HConstants.EMPTY_END_ROW);
HRegion region = HTU.createLocalHRegion(hri, htd);
try {
assertNotNull(region);
// Spy on the region so can throw exception when close is called.
HRegion spy = Mockito.spy(region);
final boolean abort = false;
Mockito.when(spy.close(abort)).thenThrow(new RuntimeException("Mocked failed close!"));
// The CloseRegionHandler will try to get an HRegion that corresponds
// to the passed hri -- so insert the region into the online region Set.
rss.addToOnlineRegions(spy);
// Assert the Server is NOT stopped before we call close region.
assertFalse(server.isStopped());
CloseRegionHandler handler = new CloseRegionHandler(server, rss, hri, false, false, -1);
boolean throwable = false;
try {
handler.process();
} catch (Throwable t) {
throwable = true;
} finally {
assertTrue(throwable);
// Abort calls stop so stopped flag should be set.
assertTrue(server.isStopped());
}
} finally {
HRegion.closeHRegion(region);
}
}
@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());
}
/**
* Inspect the log directory to recover any log file without
* an active region server.
*/
void splitLogAfterStartup() {
boolean retrySplitting = !conf.getBoolean("hbase.hlog.split.skip.errors",
HLog.SPLIT_SKIP_ERRORS_DEFAULT);
Path logsDirPath = new Path(this.rootdir, HConstants.HREGION_LOGDIR_NAME);
do {
if (master.isStopped()) {
LOG.warn("Master stopped while splitting logs");
break;
}
List<ServerName> serverNames = new ArrayList<ServerName>();
try {
if (!this.fs.exists(logsDirPath)) return;
FileStatus[] logFolders = FSUtils.listStatus(this.fs, logsDirPath, null);
// Get online servers after getting log folders to avoid log folder deletion of newly
// checked in region servers . see HBASE-5916
Set<ServerName> onlineServers = ((HMaster) master).getServerManager().getOnlineServers()
.keySet();
if (logFolders == null || logFolders.length == 0) {
LOG.debug("No log files to split, proceeding...");
return;
}
for (FileStatus status : logFolders) {
String sn = status.getPath().getName();
// truncate splitting suffix if present (for ServerName parsing)
if (sn.endsWith(HLog.SPLITTING_EXT)) {
sn = sn.substring(0, sn.length() - HLog.SPLITTING_EXT.length());
}
ServerName serverName = ServerName.parseServerName(sn);
if (!onlineServers.contains(serverName)) {
LOG.info("Log folder " + status.getPath() + " doesn't belong "
+ "to a known region server, splitting");
serverNames.add(serverName);
} else {
LOG.info("Log folder " + status.getPath()
+ " belongs to an existing region server");
}
}
splitLog(serverNames);
retrySplitting = false;
} catch (IOException ioe) {
LOG.warn("Failed splitting of " + serverNames, ioe);
if (!checkFileSystem()) {
LOG.warn("Bad Filesystem, exiting");
Runtime.getRuntime().halt(1);
}
try {
if (retrySplitting) {
Thread.sleep(conf.getInt(
"hbase.hlog.split.failure.retry.interval", 30 * 1000));
}
} catch (InterruptedException e) {
LOG.warn("Interrupted, aborting since cannot return w/o splitting");
Thread.currentThread().interrupt();
retrySplitting = false;
Runtime.getRuntime().halt(1);
}
}
} while (retrySplitting);
}
private void OpenRegion(
Server server, RegionServerServices rss, HTableDescriptor htd, HRegionInfo hri)
throws IOException, NodeExistsException, KeeperException, DeserializationException {
// Create it OFFLINE node, which is what Master set before sending OPEN RPC
ZKAssign.createNodeOffline(server.getZooKeeper(), hri, server.getServerName());
OpenRegionHandler openHandler = new OpenRegionHandler(server, rss, hri, htd);
rss.getRegionsInTransitionInRS().put(hri.getEncodedNameAsBytes(), Boolean.TRUE);
openHandler.process();
// This parse is not used?
RegionTransition.parseFrom(ZKAssign.getData(server.getZooKeeper(), hri.getEncodedName()));
// delete the node, which is what Master do after the region is opened
ZKAssign.deleteNode(
server.getZooKeeper(),
hri.getEncodedName(),
EventType.RS_ZK_REGION_OPENED,
server.getServerName());
}
DaughterOpener(final Server s, final Region r) {
super(
(s == null ? "null-services" : s.getServerName())
+ "-daughterOpener="
+ r.getRegionInfo().getEncodedName());
setDaemon(true);
this.server = s;
this.r = r;
}
/**
* Perform time consuming opening of the merged region.
*
* @param server Hosting server instance. Can be null when testing
* @param services Used to online/offline regions.
* @param merged the merged region
* @throws IOException If thrown, transaction failed. Call {@link #rollback(Server,
* RegionServerServices)}
*/
void openMergedRegion(final Server server, final RegionServerServices services, HRegion merged)
throws IOException {
boolean stopped = server != null && server.isStopped();
boolean stopping = services != null && services.isStopping();
if (stopped || stopping) {
LOG.info(
"Not opening merged region "
+ merged.getRegionNameAsString()
+ " because stopping="
+ stopping
+ ", stopped="
+ stopped);
return;
}
HRegionInfo hri = merged.getRegionInfo();
LoggingProgressable reporter =
server == null
? null
: new LoggingProgressable(
hri,
server
.getConfiguration()
.getLong("hbase.regionserver.regionmerge.open.log.interval", 10000));
merged.openHRegion(reporter);
if (services != null) {
try {
if (useCoordinationForAssignment) {
services.postOpenDeployTasks(merged);
} else if (!services.reportRegionStateTransition(
TransitionCode.MERGED,
mergedRegionInfo,
region_a.getRegionInfo(),
region_b.getRegionInfo())) {
throw new IOException(
"Failed to report merged region to master: " + mergedRegionInfo.getShortNameToLog());
}
services.addToOnlineRegions(merged);
} catch (KeeperException ke) {
throw new IOException(ke);
}
}
}
public HLogSplitterHandler(
final Server server,
String curTask,
final MutableInt curTaskZKVersion,
CancelableProgressable reporter,
AtomicInteger inProgressTasks,
TaskExecutor splitTaskExecutor,
RecoveryMode mode) {
super(server, EventType.RS_LOG_REPLAY);
this.curTask = curTask;
this.wal = ZKSplitLog.getFileName(curTask);
this.reporter = reporter;
this.inProgressTasks = inProgressTasks;
this.inProgressTasks.incrementAndGet();
this.serverName = server.getServerName();
this.zkw = server.getZooKeeper();
this.curTaskZKVersion = curTaskZKVersion;
this.splitTaskExecutor = splitTaskExecutor;
this.mode = mode;
}
/**
* Checks to see if the file system is still accessible.
* If not, sets closed
* @return false if file system is not available
*/
public boolean checkFileSystem() {
if (this.fsOk) {
try {
FSUtils.checkFileSystemAvailable(this.fs);
FSUtils.checkDfsSafeMode(this.conf);
} catch (IOException e) {
master.abort("Shutting down HBase cluster: file system not available", e);
this.fsOk = false;
}
}
return this.fsOk;
}
/**
* Perform time consuming opening of the daughter regions.
*
* @param server Hosting server instance. Can be null when testing
* @param services Used to online/offline regions.
* @param a first daughter region
* @param a second daughter region
* @throws IOException If thrown, transaction failed. Call {@link #rollback(Server,
* RegionServerServices)}
*/
@VisibleForTesting
void openDaughters(final Server server, final RegionServerServices services, Region a, Region b)
throws IOException {
boolean stopped = server != null && server.isStopped();
boolean stopping = services != null && services.isStopping();
// TODO: Is this check needed here?
if (stopped || stopping) {
LOG.info(
"Not opening daughters "
+ b.getRegionInfo().getRegionNameAsString()
+ " and "
+ a.getRegionInfo().getRegionNameAsString()
+ " because stopping="
+ stopping
+ ", stopped="
+ stopped);
} else {
// Open daughters in parallel.
DaughterOpener aOpener = new DaughterOpener(server, a);
DaughterOpener bOpener = new DaughterOpener(server, b);
aOpener.start();
bOpener.start();
try {
aOpener.join();
if (aOpener.getException() == null) {
transition(SplitTransactionPhase.OPENED_REGION_A);
}
bOpener.join();
if (bOpener.getException() == null) {
transition(SplitTransactionPhase.OPENED_REGION_B);
}
} catch (InterruptedException e) {
throw (InterruptedIOException) new InterruptedIOException().initCause(e);
}
if (aOpener.getException() != null) {
throw new IOException("Failed " + aOpener.getName(), aOpener.getException());
}
if (bOpener.getException() != null) {
throw new IOException("Failed " + bOpener.getName(), bOpener.getException());
}
if (services != null) {
if (!services.reportRegionStateTransition(
TransitionCode.SPLIT, parent.getRegionInfo(), hri_a, hri_b)) {
throw new IOException(
"Failed to report split region to master: "
+ parent.getRegionInfo().getShortNameToLog());
}
// Should add it to OnlineRegions
services.addToOnlineRegions(b);
services.addToOnlineRegions(a);
}
}
}
/** Test SplitTransactionListener */
@Test
public void testSplitTransactionListener() throws IOException {
SplitTransactionImpl st = new SplitTransactionImpl(this.parent, GOOD_SPLIT_ROW);
SplitTransaction.TransactionListener listener =
Mockito.mock(SplitTransaction.TransactionListener.class);
st.registerTransactionListener(listener);
st.prepare();
Server mockServer = Mockito.mock(Server.class);
when(mockServer.getConfiguration()).thenReturn(TEST_UTIL.getConfiguration());
PairOfSameType<Region> daughters = st.execute(mockServer, null);
verify(listener)
.transition(
st,
SplitTransaction.SplitTransactionPhase.STARTED,
SplitTransaction.SplitTransactionPhase.PREPARED);
verify(listener, times(15))
.transition(
any(SplitTransaction.class),
any(SplitTransaction.SplitTransactionPhase.class),
any(SplitTransaction.SplitTransactionPhase.class));
verifyNoMoreInteractions(listener);
}
/**
* Run the transaction.
*
* @param server Hosting server instance. Can be null when testing
* @param services Used to online/offline regions.
* @throws IOException If thrown, transaction failed. Call {@link #rollback(Server,
* RegionServerServices)}
* @return merged region
* @throws IOException
* @see #rollback(Server, RegionServerServices)
*/
public HRegion execute(final Server server, final RegionServerServices services)
throws IOException {
useCoordinationForAssignment =
server == null ? true : ConfigUtil.useZKForAssignment(server.getConfiguration());
if (rmd == null) {
rmd =
server != null && server.getCoordinatedStateManager() != null
? ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())
.getRegionMergeCoordination()
.getDefaultDetails()
: null;
}
if (rsCoprocessorHost == null) {
rsCoprocessorHost =
server != null ? ((HRegionServer) server).getRegionServerCoprocessorHost() : null;
}
HRegion mergedRegion = createMergedRegion(server, services);
if (rsCoprocessorHost != null) {
rsCoprocessorHost.postMergeCommit(this.region_a, this.region_b, mergedRegion);
}
return stepsAfterPONR(server, services, mergedRegion);
}
/**
* Open daughter regions, add them to online list and update meta.
*
* @param server
* @param daughter
* @throws IOException
* @throws KeeperException
*/
@VisibleForTesting
void openDaughterRegion(final Server server, final Region daughter)
throws IOException, KeeperException {
HRegionInfo hri = daughter.getRegionInfo();
LoggingProgressable reporter =
server == null
? null
: new LoggingProgressable(
hri,
server
.getConfiguration()
.getLong("hbase.regionserver.split.daughter.open.log.interval", 10000));
((HRegion) daughter).openHRegion(reporter);
}
@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));
}
public MasterFileSystem(Server master, MasterServices services,
MasterMetrics metrics, boolean masterRecovery)
throws IOException {
this.conf = master.getConfiguration();
this.master = master;
this.services = services;
this.metrics = metrics;
// Set filesystem to be that of this.rootdir else we get complaints about
// mismatched filesystems if hbase.rootdir is hdfs and fs.defaultFS is
// default localfs. Presumption is that rootdir is fully-qualified before
// we get to here with appropriate fs scheme.
//设置HBaseæ ¹ç›®å½•
this.rootdir = FSUtils.getRootDir(conf);
//HBaseè¡¨åˆ›å»ºå’Œåˆ é™¤çš„ä¸´æ—¶ç›®å½•/${hbase.rootdir}/.tmp
this.tempdir = new Path(this.rootdir, HConstants.HBASE_TEMP_DIRECTORY);
// Cover both bases, the old way of setting default fs and the new.
// We're supposed to run on 0.20 and 0.21 anyways.
this.fs = this.rootdir.getFileSystem(conf);
String fsUri = this.fs.getUri().toString();
conf.set("fs.default.name", fsUri);
conf.set("fs.defaultFS", fsUri);
// make sure the fs has the same conf
fs.setConf(conf);
this.distributedLogSplitting =
conf.getBoolean("hbase.master.distributed.log.splitting", true);
if (this.distributedLogSplitting) {
this.splitLogManager = new SplitLogManager(master.getZooKeeper(),
master.getConfiguration(), master, master.getServerName().toString());
this.splitLogManager.finishInitialization(masterRecovery);
} else {
this.splitLogManager = null;
}
// setup the filesystem variable
// set up the archived logs path
this.oldLogDir = createInitialFileSystemLayout();
}
public HRegion stepsAfterPONR(
final Server server, final RegionServerServices services, HRegion mergedRegion)
throws IOException {
openMergedRegion(server, services, mergedRegion);
if (useCoordination(server)) {
((BaseCoordinatedStateManager) server.getCoordinatedStateManager())
.getRegionMergeCoordination()
.completeRegionMergeTransaction(
services, mergedRegionInfo, region_a, region_b, rmd, mergedRegion);
}
if (rsCoprocessorHost != null) {
rsCoprocessorHost.postMerge(this.region_a, this.region_b, mergedRegion);
}
return mergedRegion;
}
public WALSplitterHandler(
final Server server,
SplitLogWorkerCoordination coordination,
SplitLogWorkerCoordination.SplitTaskDetails splitDetails,
CancelableProgressable reporter,
AtomicInteger inProgressTasks,
TaskExecutor splitTaskExecutor,
RecoveryMode mode) {
super(server, EventType.RS_LOG_REPLAY);
this.splitTaskDetails = splitDetails;
this.coordination = coordination;
this.reporter = reporter;
this.inProgressTasks = inProgressTasks;
this.inProgressTasks.incrementAndGet();
this.serverName = server.getServerName();
this.splitTaskExecutor = splitTaskExecutor;
this.mode = mode;
}
/**
* Reproduce locking up that happens when we get an inopportune sync during setup for zigzaglatch
* wait. See HBASE-14317. If below is broken, we will see this test timeout because it is locked
* up.
*
* <p>First I need to set up some mocks for Server and RegionServerServices. I also need to set up
* a dodgy WAL that will throw an exception when we go to append to it.
*/
@Test(timeout = 20000)
public void testLockupWhenSyncInMiddleOfZigZagSetup() throws IOException {
// A WAL that we can have throw exceptions when a flag is set.
class DodgyFSLog extends FSHLog {
// Set this when want the WAL to start throwing exceptions.
volatile boolean throwException = false;
// Latch to hold up processing until after another operation has had time to run.
CountDownLatch latch = new CountDownLatch(1);
public DodgyFSLog(FileSystem fs, Path root, String logDir, Configuration conf)
throws IOException {
super(fs, root, logDir, conf);
}
@Override
protected void afterCreatingZigZagLatch() {
// If throwException set, then append will throw an exception causing the WAL to be
// rolled. We'll come in here. Hold up processing until a sync can get in before
// the zigzag has time to complete its setup and get its own sync in. This is what causes
// the lock up we've seen in production.
if (throwException) {
try {
LOG.info("LATCHED");
// So, timing can have it that the test can run and the bad flush below happens
// before we get here. In this case, we'll be stuck waiting on this latch but there
// is nothing in the WAL pipeline to get us to the below beforeWaitOnSafePoint...
// because all WALs have rolled. In this case, just give up on test.
if (!this.latch.await(5, TimeUnit.SECONDS)) {
LOG.warn("GIVE UP! Failed waiting on latch...Test is ABORTED!");
}
} catch (InterruptedException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
}
}
@Override
protected void beforeWaitOnSafePoint() {
if (throwException) {
LOG.info("COUNTDOWN");
// Don't countdown latch until someone waiting on it otherwise, the above
// afterCreatingZigZagLatch will get to the latch and no one will ever free it and we'll
// be stuck; test won't go down
while (this.latch.getCount() <= 0) Threads.sleep(1);
this.latch.countDown();
}
}
@Override
protected Writer createWriterInstance(Path path) throws IOException {
final Writer w = super.createWriterInstance(path);
return new Writer() {
@Override
public void close() throws IOException {
w.close();
}
@Override
public void sync() throws IOException {
if (throwException) {
throw new IOException("FAKE! Failed to replace a bad datanode...SYNC");
}
w.sync();
}
@Override
public void append(Entry entry) throws IOException {
if (throwException) {
throw new IOException("FAKE! Failed to replace a bad datanode...APPEND");
}
w.append(entry);
}
@Override
public long getLength() {
return w.getLength();
}
};
}
}
// Mocked up server and regionserver services. Needed below.
Server server = Mockito.mock(Server.class);
Mockito.when(server.getConfiguration()).thenReturn(CONF);
Mockito.when(server.isStopped()).thenReturn(false);
Mockito.when(server.isAborted()).thenReturn(false);
RegionServerServices services = Mockito.mock(RegionServerServices.class);
// OK. Now I have my mocked up Server & RegionServerServices and dodgy WAL, go ahead with test.
FileSystem fs = FileSystem.get(CONF);
Path rootDir = new Path(dir + getName());
DodgyFSLog dodgyWAL = new DodgyFSLog(fs, rootDir, getName(), CONF);
Path originalWAL = dodgyWAL.getCurrentFileName();
// I need a log roller running.
LogRoller logRoller = new LogRoller(server, services);
logRoller.addWAL(dodgyWAL);
// There is no 'stop' once a logRoller is running.. it just dies.
logRoller.start();
// Now get a region and start adding in edits.
HTableDescriptor htd = new HTableDescriptor(TableName.META_TABLE_NAME);
final HRegion region = initHRegion(tableName, null, null, dodgyWAL);
byte[] bytes = Bytes.toBytes(getName());
NavigableMap<byte[], Integer> scopes = new TreeMap<byte[], Integer>(Bytes.BYTES_COMPARATOR);
scopes.put(COLUMN_FAMILY_BYTES, 0);
MultiVersionConcurrencyControl mvcc = new MultiVersionConcurrencyControl();
try {
// First get something into memstore. Make a Put and then pull the Cell out of it. Will
// manage append and sync carefully in below to manufacture hang. We keep adding same
// edit. WAL subsystem doesn't care.
Put put = new Put(bytes);
put.addColumn(COLUMN_FAMILY_BYTES, Bytes.toBytes("1"), bytes);
WALKey key =
new WALKey(
region.getRegionInfo().getEncodedNameAsBytes(),
htd.getTableName(),
System.currentTimeMillis(),
mvcc,
scopes);
WALEdit edit = new WALEdit();
CellScanner CellScanner = put.cellScanner();
assertTrue(CellScanner.advance());
edit.add(CellScanner.current());
// Put something in memstore and out in the WAL. Do a big number of appends so we push
// out other side of the ringbuffer. If small numbers, stuff doesn't make it to WAL
for (int i = 0; i < 1000; i++) {
region.put(put);
}
// Set it so we start throwing exceptions.
LOG.info("SET throwing of exception on append");
dodgyWAL.throwException = true;
// This append provokes a WAL roll request
dodgyWAL.append(region.getRegionInfo(), key, edit, true);
boolean exception = false;
try {
dodgyWAL.sync();
} catch (Exception e) {
exception = true;
}
assertTrue("Did not get sync exception", exception);
// Get a memstore flush going too so we have same hung profile as up in the issue over
// in HBASE-14317. Flush hangs trying to get sequenceid because the ringbuffer is held up
// by the zigzaglatch waiting on syncs to come home.
Thread t =
new Thread("Flusher") {
public void run() {
try {
if (region.getMemstoreSize() <= 0) {
throw new IOException("memstore size=" + region.getMemstoreSize());
}
region.flush(false);
} catch (IOException e) {
// Can fail trying to flush in middle of a roll. Not a failure. Will succeed later
// when roll completes.
LOG.info("In flush", e);
}
LOG.info("Exiting");
};
};
t.setDaemon(true);
t.start();
// Wait until
while (dodgyWAL.latch.getCount() > 0) Threads.sleep(1);
// Now assert I got a new WAL file put in place even though loads of errors above.
assertTrue(originalWAL != dodgyWAL.getCurrentFileName());
// Can I append to it?
dodgyWAL.throwException = false;
try {
region.put(put);
} catch (Exception e) {
LOG.info("In the put", e);
}
} finally {
// To stop logRoller, its server has to say it is stopped.
Mockito.when(server.isStopped()).thenReturn(true);
if (logRoller != null) logRoller.close();
try {
if (region != null) region.close();
if (dodgyWAL != null) dodgyWAL.close();
} catch (Exception e) {
LOG.info("On way out", e);
}
}
}
@Test
public void testLogCleaning() throws Exception {
Configuration conf = TEST_UTIL.getConfiguration();
// set TTL
long ttl = 10000;
conf.setLong("hbase.master.logcleaner.ttl", ttl);
conf.setBoolean(HConstants.REPLICATION_ENABLE_KEY, HConstants.REPLICATION_ENABLE_DEFAULT);
Replication.decorateMasterConfiguration(conf);
Server server = new DummyServer();
ReplicationQueues repQueues =
ReplicationFactory.getReplicationQueues(server.getZooKeeper(), conf, server);
repQueues.init(server.getServerName().toString());
final Path oldLogDir = new Path(TEST_UTIL.getDataTestDir(), HConstants.HREGION_OLDLOGDIR_NAME);
String fakeMachineName = URLEncoder.encode(server.getServerName().toString(), "UTF8");
final FileSystem fs = FileSystem.get(conf);
// Create 2 invalid files, 1 "recent" file, 1 very new file and 30 old files
long now = System.currentTimeMillis();
fs.delete(oldLogDir, true);
fs.mkdirs(oldLogDir);
// Case 1: 2 invalid files, which would be deleted directly
fs.createNewFile(new Path(oldLogDir, "a"));
fs.createNewFile(new Path(oldLogDir, fakeMachineName + "." + "a"));
// Case 2: 1 "recent" file, not even deletable for the first log cleaner
// (TimeToLiveLogCleaner), so we are not going down the chain
System.out.println("Now is: " + now);
for (int i = 1; i < 31; i++) {
// Case 3: old files which would be deletable for the first log cleaner
// (TimeToLiveLogCleaner), and also for the second (ReplicationLogCleaner)
Path fileName = new Path(oldLogDir, fakeMachineName + "." + (now - i));
fs.createNewFile(fileName);
// Case 4: put 3 old log files in ZK indicating that they are scheduled
// for replication so these files would pass the first log cleaner
// (TimeToLiveLogCleaner) but would be rejected by the second
// (ReplicationLogCleaner)
if (i % (30 / 3) == 1) {
repQueues.addLog(fakeMachineName, fileName.getName());
System.out.println("Replication log file: " + fileName);
}
}
// sleep for sometime to get newer modifcation time
Thread.sleep(ttl);
fs.createNewFile(new Path(oldLogDir, fakeMachineName + "." + now));
// Case 2: 1 newer file, not even deletable for the first log cleaner
// (TimeToLiveLogCleaner), so we are not going down the chain
fs.createNewFile(new Path(oldLogDir, fakeMachineName + "." + (now + 10000)));
for (FileStatus stat : fs.listStatus(oldLogDir)) {
System.out.println(stat.getPath().toString());
}
assertEquals(34, fs.listStatus(oldLogDir).length);
LogCleaner cleaner = new LogCleaner(1000, server, conf, fs, oldLogDir);
cleaner.chore();
// We end up with the current log file, a newer one and the 3 old log
// files which are scheduled for replication
TEST_UTIL.waitFor(
1000,
new Waiter.Predicate<Exception>() {
@Override
public boolean evaluate() throws Exception {
return 5 == fs.listStatus(oldLogDir).length;
}
});
for (FileStatus file : fs.listStatus(oldLogDir)) {
System.out.println("Kept log files: " + file.getPath().getName());
}
}
/**
* Prepare the merged region and region files.
*
* @param server Hosting server instance. Can be null when testing
* @param services Used to online/offline regions.
* @return merged region
* @throws IOException If thrown, transaction failed. Call {@link #rollback(Server,
* RegionServerServices)}
*/
HRegion createMergedRegion(final Server server, final RegionServerServices services)
throws IOException {
LOG.info(
"Starting merge of "
+ region_a
+ " and "
+ region_b.getRegionNameAsString()
+ ", forcible="
+ forcible);
if ((server != null && server.isStopped()) || (services != null && services.isStopping())) {
throw new IOException("Server is stopped or stopping");
}
if (rsCoprocessorHost != null) {
if (rsCoprocessorHost.preMerge(this.region_a, this.region_b)) {
throw new IOException(
"Coprocessor bypassing regions " + this.region_a + " " + this.region_b + " merge.");
}
}
// If true, no cluster to write meta edits to or to use coordination.
boolean testing =
server == null
? true
: server.getConfiguration().getBoolean("hbase.testing.nocluster", false);
HRegion mergedRegion = stepsBeforePONR(server, services, testing);
@MetaMutationAnnotation List<Mutation> metaEntries = new ArrayList<Mutation>();
if (rsCoprocessorHost != null) {
if (rsCoprocessorHost.preMergeCommit(this.region_a, this.region_b, metaEntries)) {
throw new IOException(
"Coprocessor bypassing regions " + this.region_a + " " + this.region_b + " merge.");
}
try {
for (Mutation p : metaEntries) {
HRegionInfo.parseRegionName(p.getRow());
}
} catch (IOException e) {
LOG.error(
"Row key of mutation from coprocessor is not parsable as region name."
+ "Mutations from coprocessor should only be for hbase:meta table.",
e);
throw e;
}
}
// This is the point of no return. Similar with SplitTransaction.
// IF we reach the PONR then subsequent failures need to crash out this
// regionserver
this.journal.add(JournalEntry.PONR);
// Add merged region and delete region_a and region_b
// as an atomic update. See HBASE-7721. This update to hbase:meta makes the region
// will determine whether the region is merged or not in case of failures.
// If it is successful, master will roll-forward, if not, master will
// rollback
if (!testing && useCoordinationForAssignment) {
if (metaEntries.isEmpty()) {
MetaTableAccessor.mergeRegions(
server.getConnection(),
mergedRegion.getRegionInfo(),
region_a.getRegionInfo(),
region_b.getRegionInfo(),
server.getServerName(),
region_a.getTableDesc().getRegionReplication());
} else {
mergeRegionsAndPutMetaEntries(
server.getConnection(),
mergedRegion.getRegionInfo(),
region_a.getRegionInfo(),
region_b.getRegionInfo(),
server.getServerName(),
metaEntries,
region_a.getTableDesc().getRegionReplication());
}
} else if (services != null && !useCoordinationForAssignment) {
if (!services.reportRegionStateTransition(
TransitionCode.MERGE_PONR,
mergedRegionInfo,
region_a.getRegionInfo(),
region_b.getRegionInfo())) {
// Passed PONR, let SSH clean it up
throw new IOException(
"Failed to notify master that merge passed PONR: "
+ region_a.getRegionInfo().getRegionNameAsString()
+ " and "
+ region_b.getRegionInfo().getRegionNameAsString());
}
}
return mergedRegion;
}
private boolean useCoordination(final Server server) {
return server != null
&& useCoordinationForAssignment
&& server.getCoordinatedStateManager() != null;
}
/**
* Reproduce locking up that happens when there's no further syncs after append fails, and causing
* an isolated sync then infinite wait. See HBASE-16960. If below is broken, we will see this test
* timeout because it is locked up.
*
* <p>Steps for reproduce:<br>
* 1. Trigger server abort through dodgyWAL1<br>
* 2. Add a {@link DummyWALActionsListener} to dodgyWAL2 to cause ringbuffer event handler thread
* sleep for a while thus keeping {@code endOfBatch} false<br>
* 3. Publish a sync then an append which will throw exception, check whether the sync could
* return
*/
@Test(timeout = 20000)
public void testLockup16960() throws IOException {
// A WAL that we can have throw exceptions when a flag is set.
class DodgyFSLog extends FSHLog {
// Set this when want the WAL to start throwing exceptions.
volatile boolean throwException = false;
public DodgyFSLog(FileSystem fs, Path root, String logDir, Configuration conf)
throws IOException {
super(fs, root, logDir, conf);
}
@Override
protected Writer createWriterInstance(Path path) throws IOException {
final Writer w = super.createWriterInstance(path);
return new Writer() {
@Override
public void close() throws IOException {
w.close();
}
@Override
public void sync() throws IOException {
if (throwException) {
throw new IOException("FAKE! Failed to replace a bad datanode...SYNC");
}
w.sync();
}
@Override
public void append(Entry entry) throws IOException {
if (throwException) {
throw new IOException("FAKE! Failed to replace a bad datanode...APPEND");
}
w.append(entry);
}
@Override
public long getLength() {
return w.getLength();
}
};
}
@Override
protected long doReplaceWriter(Path oldPath, Path newPath, Writer nextWriter)
throws IOException {
if (throwException) {
throw new FailedLogCloseException("oldPath=" + oldPath + ", newPath=" + newPath);
}
long oldFileLen = 0L;
oldFileLen = super.doReplaceWriter(oldPath, newPath, nextWriter);
return oldFileLen;
}
}
// Mocked up server and regionserver services. Needed below.
Server server =
new DummyServer(CONF, ServerName.valueOf("hostname1.example.org", 1234, 1L).toString());
RegionServerServices services = Mockito.mock(RegionServerServices.class);
CONF.setLong("hbase.regionserver.hlog.sync.timeout", 10000);
// OK. Now I have my mocked up Server & RegionServerServices and dodgy WAL,
// go ahead with test.
FileSystem fs = FileSystem.get(CONF);
Path rootDir = new Path(dir + getName());
DodgyFSLog dodgyWAL1 = new DodgyFSLog(fs, rootDir, getName(), CONF);
Path rootDir2 = new Path(dir + getName() + "2");
final DodgyFSLog dodgyWAL2 = new DodgyFSLog(fs, rootDir2, getName() + "2", CONF);
// Add a listener to force ringbuffer event handler sleep for a while
dodgyWAL2.registerWALActionsListener(new DummyWALActionsListener());
// I need a log roller running.
LogRoller logRoller = new LogRoller(server, services);
logRoller.addWAL(dodgyWAL1);
logRoller.addWAL(dodgyWAL2);
// There is no 'stop' once a logRoller is running.. it just dies.
logRoller.start();
// Now get a region and start adding in edits.
HTableDescriptor htd = new HTableDescriptor(TableName.META_TABLE_NAME);
final HRegion region = initHRegion(tableName, null, null, dodgyWAL1);
byte[] bytes = Bytes.toBytes(getName());
NavigableMap<byte[], Integer> scopes = new TreeMap<byte[], Integer>(Bytes.BYTES_COMPARATOR);
scopes.put(COLUMN_FAMILY_BYTES, 0);
MultiVersionConcurrencyControl mvcc = new MultiVersionConcurrencyControl();
try {
Put put = new Put(bytes);
put.addColumn(COLUMN_FAMILY_BYTES, Bytes.toBytes("1"), bytes);
WALKey key =
new WALKey(
region.getRegionInfo().getEncodedNameAsBytes(),
htd.getTableName(),
System.currentTimeMillis(),
mvcc,
scopes);
WALEdit edit = new WALEdit();
CellScanner CellScanner = put.cellScanner();
assertTrue(CellScanner.advance());
edit.add(CellScanner.current());
LOG.info("SET throwing of exception on append");
dodgyWAL1.throwException = true;
// This append provokes a WAL roll request
dodgyWAL1.append(region.getRegionInfo(), key, edit, true);
boolean exception = false;
try {
dodgyWAL1.sync();
} catch (Exception e) {
exception = true;
}
assertTrue("Did not get sync exception", exception);
// LogRoller call dodgyWAL1.rollWriter get FailedLogCloseException and
// cause server abort.
try {
// wait LogRoller exit.
Thread.sleep(50);
} catch (InterruptedException e) {
e.printStackTrace();
}
final CountDownLatch latch = new CountDownLatch(1);
// make RingBufferEventHandler sleep 1s, so the following sync
// endOfBatch=false
key =
new WALKey(
region.getRegionInfo().getEncodedNameAsBytes(),
TableName.valueOf("sleep"),
System.currentTimeMillis(),
mvcc,
scopes);
dodgyWAL2.append(region.getRegionInfo(), key, edit, true);
Thread t =
new Thread("Sync") {
public void run() {
try {
dodgyWAL2.sync();
} catch (IOException e) {
LOG.info("In sync", e);
}
latch.countDown();
LOG.info("Sync exiting");
};
};
t.setDaemon(true);
t.start();
try {
// make sure sync have published.
Thread.sleep(100);
} catch (InterruptedException e1) {
e1.printStackTrace();
}
// make append throw DamagedWALException
key =
new WALKey(
region.getRegionInfo().getEncodedNameAsBytes(),
TableName.valueOf("DamagedWALException"),
System.currentTimeMillis(),
mvcc,
scopes);
dodgyWAL2.append(region.getRegionInfo(), key, edit, true);
while (latch.getCount() > 0) {
Threads.sleep(100);
}
assertTrue(server.isAborted());
} finally {
if (logRoller != null) {
logRoller.close();
}
try {
if (region != null) {
region.close();
}
if (dodgyWAL1 != null) {
dodgyWAL1.close();
}
if (dodgyWAL2 != null) {
dodgyWAL2.close();
}
} catch (Exception e) {
LOG.info("On way out", e);
}
}
}
/**
* Prepare the regions and region files.
*
* @param server Hosting server instance. Can be null when testing (won't try and update in zk if
* a null server)
* @param services Used to online/offline regions.
* @param user
* @throws IOException If thrown, transaction failed. Call {@link #rollback(Server,
* RegionServerServices)}
* @return Regions created
*/
@VisibleForTesting
PairOfSameType<Region> createDaughters(
final Server server, final RegionServerServices services, User user) throws IOException {
LOG.info("Starting split of region " + this.parent);
if ((server != null && server.isStopped()) || (services != null && services.isStopping())) {
throw new IOException("Server is stopped or stopping");
}
assert !this.parent.lock.writeLock().isHeldByCurrentThread()
: "Unsafe to hold write lock while performing RPCs";
transition(SplitTransactionPhase.BEFORE_PRE_SPLIT_HOOK);
// Coprocessor callback
if (this.parent.getCoprocessorHost() != null) {
// TODO: Remove one of these
parent.getCoprocessorHost().preSplit(user);
parent.getCoprocessorHost().preSplit(splitrow, user);
}
transition(SplitTransactionPhase.AFTER_PRE_SPLIT_HOOK);
// If true, no cluster to write meta edits to or to update znodes in.
boolean testing =
server == null
? true
: server.getConfiguration().getBoolean("hbase.testing.nocluster", false);
this.fileSplitTimeout =
testing
? this.fileSplitTimeout
: server
.getConfiguration()
.getLong("hbase.regionserver.fileSplitTimeout", this.fileSplitTimeout);
PairOfSameType<Region> daughterRegions = stepsBeforePONR(server, services, testing);
final List<Mutation> metaEntries = new ArrayList<Mutation>();
boolean ret = false;
if (this.parent.getCoprocessorHost() != null) {
ret = parent.getCoprocessorHost().preSplitBeforePONR(splitrow, metaEntries, user);
if (ret) {
throw new IOException(
"Coprocessor bypassing region "
+ parent.getRegionInfo().getRegionNameAsString()
+ " split.");
}
try {
for (Mutation p : metaEntries) {
HRegionInfo.parseRegionName(p.getRow());
}
} catch (IOException e) {
LOG.error(
"Row key of mutation from coprossor is not parsable as region name."
+ "Mutations from coprocessor should only for hbase:meta table.");
throw e;
}
}
// This is the point of no return. Adding subsequent edits to .META. as we
// do below when we do the daughter opens adding each to .META. can fail in
// various interesting ways the most interesting of which is a timeout
// BUT the edits all go through (See HBASE-3872). IF we reach the PONR
// then subsequent failures need to crash out this regionserver; the
// server shutdown processing should be able to fix-up the incomplete split.
// The offlined parent will have the daughters as extra columns. If
// we leave the daughter regions in place and do not remove them when we
// crash out, then they will have their references to the parent in place
// still and the server shutdown fixup of .META. will point to these
// regions.
// We should add PONR JournalEntry before offlineParentInMeta,so even if
// OfflineParentInMeta timeout,this will cause regionserver exit,and then
// master ServerShutdownHandler will fix daughter & avoid data loss. (See
// HBase-4562).
transition(SplitTransactionPhase.PONR);
// Edit parent in meta. Offlines parent region and adds splita and splitb
// as an atomic update. See HBASE-7721. This update to META makes the region
// will determine whether the region is split or not in case of failures.
// If it is successful, master will roll-forward, if not, master will rollback
// and assign the parent region.
if (services != null
&& !services.reportRegionStateTransition(
TransitionCode.SPLIT_PONR, parent.getRegionInfo(), hri_a, hri_b)) {
// Passed PONR, let SSH clean it up
throw new IOException(
"Failed to notify master that split passed PONR: "
+ parent.getRegionInfo().getRegionNameAsString());
}
return daughterRegions;
}
public HRegion stepsBeforePONR(
final Server server, final RegionServerServices services, boolean testing)
throws IOException {
if (rmd == null) {
rmd =
server != null && server.getCoordinatedStateManager() != null
? ((BaseCoordinatedStateManager) server.getCoordinatedStateManager())
.getRegionMergeCoordination()
.getDefaultDetails()
: null;
}
// If server doesn't have a coordination state manager, don't do coordination actions.
if (useCoordination(server)) {
try {
((BaseCoordinatedStateManager) server.getCoordinatedStateManager())
.getRegionMergeCoordination()
.startRegionMergeTransaction(
mergedRegionInfo,
server.getServerName(),
region_a.getRegionInfo(),
region_b.getRegionInfo());
} catch (IOException e) {
throw new IOException(
"Failed to start region merge transaction for "
+ this.mergedRegionInfo.getRegionNameAsString(),
e);
}
} else if (services != null && !useCoordinationForAssignment) {
if (!services.reportRegionStateTransition(
TransitionCode.READY_TO_MERGE,
mergedRegionInfo,
region_a.getRegionInfo(),
region_b.getRegionInfo())) {
throw new IOException(
"Failed to get ok from master to merge "
+ region_a.getRegionInfo().getRegionNameAsString()
+ " and "
+ region_b.getRegionInfo().getRegionNameAsString());
}
}
this.journal.add(JournalEntry.SET_MERGING);
if (useCoordination(server)) {
// After creating the merge node, wait for master to transition it
// from PENDING_MERGE to MERGING so that we can move on. We want master
// knows about it and won't transition any region which is merging.
((BaseCoordinatedStateManager) server.getCoordinatedStateManager())
.getRegionMergeCoordination()
.waitForRegionMergeTransaction(services, mergedRegionInfo, region_a, region_b, rmd);
}
this.region_a.getRegionFileSystem().createMergesDir();
this.journal.add(JournalEntry.CREATED_MERGE_DIR);
Map<byte[], List<StoreFile>> hstoreFilesOfRegionA =
closeAndOfflineRegion(services, this.region_a, true, testing);
Map<byte[], List<StoreFile>> hstoreFilesOfRegionB =
closeAndOfflineRegion(services, this.region_b, false, testing);
assert hstoreFilesOfRegionA != null && hstoreFilesOfRegionB != null;
//
// mergeStoreFiles creates merged region dirs under the region_a merges dir
// Nothing to unroll here if failure -- clean up of CREATE_MERGE_DIR will
// clean this up.
mergeStoreFiles(hstoreFilesOfRegionA, hstoreFilesOfRegionB);
if (useCoordination(server)) {
try {
// Do the final check in case any merging region is moved somehow. If so, the transition
// will fail.
((BaseCoordinatedStateManager) server.getCoordinatedStateManager())
.getRegionMergeCoordination()
.confirmRegionMergeTransaction(
this.mergedRegionInfo,
region_a.getRegionInfo(),
region_b.getRegionInfo(),
server.getServerName(),
rmd);
} catch (IOException e) {
throw new IOException(
"Failed setting MERGING on " + this.mergedRegionInfo.getRegionNameAsString(), e);
}
}
// Log to the journal that we are creating merged region. We could fail
// halfway through. If we do, we could have left
// stuff in fs that needs cleanup -- a storefile or two. Thats why we
// add entry to journal BEFORE rather than AFTER the change.
this.journal.add(JournalEntry.STARTED_MERGED_REGION_CREATION);
HRegion mergedRegion =
createMergedRegionFromMerges(this.region_a, this.region_b, this.mergedRegionInfo);
return mergedRegion;
}
/**
* @param server Hosting server instance (May be null when testing).
* @param services Services of regionserver, used to online regions.
* @throws IOException If thrown, rollback failed. Take drastic action.
* @return True if we successfully rolled back, false if we got to the point of no return and so
* now need to abort the server to minimize damage.
*/
@SuppressWarnings("deprecation")
public boolean rollback(final Server server, final RegionServerServices services)
throws IOException {
assert this.mergedRegionInfo != null;
// Coprocessor callback
if (rsCoprocessorHost != null) {
rsCoprocessorHost.preRollBackMerge(this.region_a, this.region_b);
}
boolean result = true;
ListIterator<JournalEntry> iterator = this.journal.listIterator(this.journal.size());
// Iterate in reverse.
while (iterator.hasPrevious()) {
JournalEntry je = iterator.previous();
switch (je) {
case SET_MERGING:
if (useCoordination(server)) {
((BaseCoordinatedStateManager) server.getCoordinatedStateManager())
.getRegionMergeCoordination()
.clean(this.mergedRegionInfo);
} else if (services != null
&& !useCoordinationForAssignment
&& !services.reportRegionStateTransition(
TransitionCode.MERGE_REVERTED,
mergedRegionInfo,
region_a.getRegionInfo(),
region_b.getRegionInfo())) {
return false;
}
break;
case CREATED_MERGE_DIR:
this.region_a.writestate.writesEnabled = true;
this.region_b.writestate.writesEnabled = true;
this.region_a.getRegionFileSystem().cleanupMergesDir();
break;
case CLOSED_REGION_A:
try {
// So, this returns a seqid but if we just closed and then reopened,
// we should be ok. On close, we flushed using sequenceid obtained
// from hosting regionserver so no need to propagate the sequenceid
// returned out of initialize below up into regionserver as we
// normally do.
this.region_a.initialize();
} catch (IOException e) {
LOG.error(
"Failed rollbacking CLOSED_REGION_A of region "
+ this.region_a.getRegionNameAsString(),
e);
throw new RuntimeException(e);
}
break;
case OFFLINED_REGION_A:
if (services != null) services.addToOnlineRegions(this.region_a);
break;
case CLOSED_REGION_B:
try {
this.region_b.initialize();
} catch (IOException e) {
LOG.error(
"Failed rollbacking CLOSED_REGION_A of region "
+ this.region_b.getRegionNameAsString(),
e);
throw new RuntimeException(e);
}
break;
case OFFLINED_REGION_B:
if (services != null) services.addToOnlineRegions(this.region_b);
break;
case STARTED_MERGED_REGION_CREATION:
this.region_a.getRegionFileSystem().cleanupMergedRegion(this.mergedRegionInfo);
break;
case PONR:
// We got to the point-of-no-return so we need to just abort. Return
// immediately. Do not clean up created merged regions.
return false;
default:
throw new RuntimeException("Unhandled journal entry: " + je);
}
}
// Coprocessor callback
if (rsCoprocessorHost != null) {
rsCoprocessorHost.postRollBackMerge(this.region_a, this.region_b);
}
return result;
}