public virtual void TestEditLogFileNotExistsWhenReadingMetadata()
{
URI uri = BKJMUtil.CreateJournalURI("/hdfsjournal-editlogfile");
NamespaceInfo nsi = NewNSInfo();
BookKeeperJournalManager bkjm = new BookKeeperJournalManager(conf, uri, nsi);
bkjm.Format(nsi);
try
{
// start new inprogress log segment with txid=1
// and write transactions till txid=50
string zkpath1 = StartAndFinalizeLogSegment(bkjm, 1, 50);
// start new inprogress log segment with txid=51
// and write transactions till txid=100
string zkpath2 = StartAndFinalizeLogSegment(bkjm, 51, 100);
// read the metadata from ZK. Here simulating the situation
// when reading,the edit log metadata can be removed by purger thread.
ZooKeeper zkspy = Org.Mockito.Mockito.Spy(BKJMUtil.ConnectZooKeeper());
bkjm.SetZooKeeper(zkspy);
Org.Mockito.Mockito.DoThrow(new KeeperException.NoNodeException(zkpath2 + " doesn't exists"
)).When(zkspy).GetData(zkpath2, false, null);
IList <EditLogLedgerMetadata> ledgerList = bkjm.GetLedgerList(false);
NUnit.Framework.Assert.AreEqual("List contains the metadata of non exists path.",
1, ledgerList.Count);
NUnit.Framework.Assert.AreEqual("LogLedgerMetadata contains wrong zk paths.", zkpath1
, ledgerList[0].GetZkPath());
}
finally
{
bkjm.Close();
}
}
public virtual void TestTwoWriters()
{
long start = 1;
NamespaceInfo nsi = NewNSInfo();
BookKeeperJournalManager bkjm1 = new BookKeeperJournalManager(conf, BKJMUtil.CreateJournalURI
("/hdfsjournal-dualWriter"), nsi);
bkjm1.Format(nsi);
BookKeeperJournalManager bkjm2 = new BookKeeperJournalManager(conf, BKJMUtil.CreateJournalURI
("/hdfsjournal-dualWriter"), nsi);
EditLogOutputStream out1 = bkjm1.StartLogSegment(start, NameNodeLayoutVersion.CurrentLayoutVersion
);
try
{
bkjm2.StartLogSegment(start, NameNodeLayoutVersion.CurrentLayoutVersion);
NUnit.Framework.Assert.Fail("Shouldn't have been able to open the second writer");
}
catch (IOException ioe)
{
Log.Info("Caught exception as expected", ioe);
}
finally
{
out1.Close();
}
}
public virtual void TestEmptyInprogressLedger()
{
URI uri = BKJMUtil.CreateJournalURI("/hdfsjournal-emptyInprogressLedger");
NamespaceInfo nsi = NewNSInfo();
BookKeeperJournalManager bkjm = new BookKeeperJournalManager(conf, uri, nsi);
bkjm.Format(nsi);
EditLogOutputStream @out = bkjm.StartLogSegment(1, NameNodeLayoutVersion.CurrentLayoutVersion
);
for (long i = 1; i <= 100; i++)
{
FSEditLogOp op = FSEditLogTestUtil.GetNoOpInstance();
op.SetTransactionId(i);
@out.Write(op);
}
@out.Close();
bkjm.FinalizeLogSegment(1, 100);
@out = bkjm.StartLogSegment(101, NameNodeLayoutVersion.CurrentLayoutVersion);
@out.Close();
bkjm.Close();
bkjm = new BookKeeperJournalManager(conf, uri, nsi);
bkjm.RecoverUnfinalizedSegments();
@out = bkjm.StartLogSegment(101, NameNodeLayoutVersion.CurrentLayoutVersion);
for (long i_1 = 1; i_1 <= 100; i_1++)
{
FSEditLogOp op = FSEditLogTestUtil.GetNoOpInstance();
op.SetTransactionId(i_1);
@out.Write(op);
}
@out.Close();
bkjm.FinalizeLogSegment(101, 200);
bkjm.Close();
}
public virtual void TestFailoverWithBK()
{
MiniDFSCluster cluster = null;
try
{
Configuration conf = new Configuration();
conf.SetInt(DFSConfigKeys.DfsHaTaileditsPeriodKey, 1);
conf.Set(DFSConfigKeys.DfsNamenodeSharedEditsDirKey, 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);
NUnit.Framework.Assert.IsTrue(fs.Exists(p));
}
finally
{
if (cluster != null)
{
cluster.Shutdown();
}
}
}
public virtual void TestSimpleRecovery()
{
NamespaceInfo nsi = NewNSInfo();
BookKeeperJournalManager bkjm = new BookKeeperJournalManager(conf, BKJMUtil.CreateJournalURI
("/hdfsjournal-simplerecovery"), nsi);
bkjm.Format(nsi);
EditLogOutputStream @out = bkjm.StartLogSegment(1, NameNodeLayoutVersion.CurrentLayoutVersion
);
for (long i = 1; i <= 100; i++)
{
FSEditLogOp op = FSEditLogTestUtil.GetNoOpInstance();
op.SetTransactionId(i);
@out.Write(op);
}
@out.SetReadyToFlush();
@out.Flush();
@out.Abort();
@out.Close();
NUnit.Framework.Assert.IsNull(zkc.Exists(bkjm.FinalizedLedgerZNode(1, 100), false
));
NUnit.Framework.Assert.IsNotNull(zkc.Exists(bkjm.InprogressZNode(1), false));
bkjm.RecoverUnfinalizedSegments();
NUnit.Framework.Assert.IsNotNull(zkc.Exists(bkjm.FinalizedLedgerZNode(1, 100), false
));
NUnit.Framework.Assert.IsNull(zkc.Exists(bkjm.InprogressZNode(1), false));
}
public virtual void TestSimpleRead()
{
NamespaceInfo nsi = NewNSInfo();
BookKeeperJournalManager bkjm = new BookKeeperJournalManager(conf, BKJMUtil.CreateJournalURI
("/hdfsjournal-simpleread"), nsi);
bkjm.Format(nsi);
long numTransactions = 10000;
EditLogOutputStream @out = bkjm.StartLogSegment(1, NameNodeLayoutVersion.CurrentLayoutVersion
);
for (long i = 1; i <= numTransactions; i++)
{
FSEditLogOp op = FSEditLogTestUtil.GetNoOpInstance();
op.SetTransactionId(i);
@out.Write(op);
}
@out.Close();
bkjm.FinalizeLogSegment(1, numTransactions);
IList <EditLogInputStream> @in = new AList <EditLogInputStream>();
bkjm.SelectInputStreams(@in, 1, true);
try
{
NUnit.Framework.Assert.AreEqual(numTransactions, FSEditLogTestUtil.CountTransactionsInStream
(@in[0]));
}
finally
{
@in[0].Close();
}
}
/// <summary>Test speculative read feature supported by bookkeeper.</summary>
/// <remarks>
/// Test speculative read feature supported by bookkeeper. Keep one bookie
/// alive and sleep all the other bookies. Non spec client will hang for long
/// time to read the entries from the bookkeeper.
/// </remarks>
/// <exception cref="System.Exception"/>
public virtual void TestSpeculativeRead()
{
// starting 9 more servers
for (int i = 1; i < 10; i++)
{
bks.AddItem(bkutil.NewBookie());
}
NamespaceInfo nsi = NewNSInfo();
Configuration conf = new Configuration();
int ensembleSize = numLocalBookies + 9;
conf.SetInt(BookKeeperJournalManager.BkjmBookkeeperEnsembleSize, ensembleSize);
conf.SetInt(BookKeeperJournalManager.BkjmBookkeeperQuorumSize, ensembleSize);
conf.SetInt(BookKeeperJournalManager.BkjmBookkeeperSpeculativeReadTimeoutMs, 100);
// sets 60 minute
conf.SetInt(BookKeeperJournalManager.BkjmBookkeeperReadEntryTimeoutSec, 3600);
BookKeeperJournalManager bkjm = new BookKeeperJournalManager(conf, BKJMUtil.CreateJournalURI
("/hdfsjournal-specread"), nsi);
bkjm.Format(nsi);
long numTransactions = 1000;
EditLogOutputStream @out = bkjm.StartLogSegment(1, NameNodeLayoutVersion.CurrentLayoutVersion
);
for (long i_1 = 1; i_1 <= numTransactions; i_1++)
{
FSEditLogOp op = FSEditLogTestUtil.GetNoOpInstance();
op.SetTransactionId(i_1);
@out.Write(op);
}
@out.Close();
bkjm.FinalizeLogSegment(1, numTransactions);
IList <EditLogInputStream> @in = new AList <EditLogInputStream>();
bkjm.SelectInputStreams(@in, 1, true);
// sleep 9 bk servers. Now only one server is running and responding to the
// clients
CountDownLatch sleepLatch = new CountDownLatch(1);
foreach (BookieServer bookie in bks)
{
SleepBookie(sleepLatch, bookie);
}
try
{
NUnit.Framework.Assert.AreEqual(numTransactions, FSEditLogTestUtil.CountTransactionsInStream
(@in[0]));
}
finally
{
@in[0].Close();
sleepLatch.CountDown();
bkjm.Close();
}
}
/// <exception cref="System.Exception"/>
public virtual void TestDefaultAckQuorum()
{
newBookie = bkutil.NewBookie();
int ensembleSize = numBookies + 1;
int quorumSize = numBookies + 1;
// ensure that the journal manager has to use all bookies,
// so that a failure will fail the journal manager
Configuration conf = new Configuration();
conf.SetInt(BookKeeperJournalManager.BkjmBookkeeperEnsembleSize, ensembleSize);
conf.SetInt(BookKeeperJournalManager.BkjmBookkeeperQuorumSize, quorumSize);
// sets 2 secs
conf.SetInt(BookKeeperJournalManager.BkjmBookkeeperAddEntryTimeoutSec, 2);
NamespaceInfo nsi = NewNSInfo();
BookKeeperJournalManager bkjm = new BookKeeperJournalManager(conf, BKJMUtil.CreateJournalURI
("/hdfsjournal-onebookiefailure"), nsi);
bkjm.Format(nsi);
CountDownLatch sleepLatch = new CountDownLatch(1);
SleepBookie(sleepLatch, newBookie);
EditLogOutputStream @out = bkjm.StartLogSegment(1, NameNodeLayoutVersion.CurrentLayoutVersion
);
int numTransactions = 100;
for (long i = 1; i <= numTransactions; i++)
{
FSEditLogOp op = FSEditLogTestUtil.GetNoOpInstance();
op.SetTransactionId(i);
@out.Write(op);
}
try
{
@out.Close();
bkjm.FinalizeLogSegment(1, numTransactions);
IList <EditLogInputStream> @in = new AList <EditLogInputStream>();
bkjm.SelectInputStreams(@in, 1, true);
try
{
NUnit.Framework.Assert.AreEqual(numTransactions, FSEditLogTestUtil.CountTransactionsInStream
(@in[0]));
}
finally
{
@in[0].Close();
}
NUnit.Framework.Assert.Fail("Should throw exception as not enough non-faulty bookies available!"
);
}
catch (IOException)
{
}
}
// expected
/// <summary>Test ack quorum feature supported by bookkeeper.</summary>
/// <remarks>
/// Test ack quorum feature supported by bookkeeper. Keep ack quorum bookie
/// alive and sleep all the other bookies. Now the client would wait for the
/// acknowledgement from the ack size bookies and after receiving the success
/// response will continue writing. Non ack client will hang long time to add
/// entries.
/// </remarks>
/// <exception cref="System.Exception"/>
public virtual void TestAckQuorum()
{
// slow bookie
newBookie = bkutil.NewBookie();
// make quorum size and ensemble size same to avoid the interleave writing
// of the ledger entries
int ensembleSize = numBookies + 1;
int quorumSize = numBookies + 1;
int ackSize = numBookies;
// ensure that the journal manager has to use all bookies,
// so that a failure will fail the journal manager
Configuration conf = new Configuration();
conf.SetInt(BookKeeperJournalManager.BkjmBookkeeperEnsembleSize, ensembleSize);
conf.SetInt(BookKeeperJournalManager.BkjmBookkeeperQuorumSize, quorumSize);
conf.SetInt(BookKeeperJournalManager.BkjmBookkeeperAckQuorumSize, ackSize);
// sets 60 minutes
conf.SetInt(BookKeeperJournalManager.BkjmBookkeeperAddEntryTimeoutSec, 3600);
NamespaceInfo nsi = NewNSInfo();
BookKeeperJournalManager bkjm = new BookKeeperJournalManager(conf, BKJMUtil.CreateJournalURI
("/hdfsjournal-onebookiefailure"), nsi);
bkjm.Format(nsi);
CountDownLatch sleepLatch = new CountDownLatch(1);
SleepBookie(sleepLatch, newBookie);
EditLogOutputStream @out = bkjm.StartLogSegment(1, NameNodeLayoutVersion.CurrentLayoutVersion
);
int numTransactions = 100;
for (long i = 1; i <= numTransactions; i++)
{
FSEditLogOp op = FSEditLogTestUtil.GetNoOpInstance();
op.SetTransactionId(i);
@out.Write(op);
}
@out.Close();
bkjm.FinalizeLogSegment(1, numTransactions);
IList <EditLogInputStream> @in = new AList <EditLogInputStream>();
bkjm.SelectInputStreams(@in, 1, true);
try
{
NUnit.Framework.Assert.AreEqual(numTransactions, FSEditLogTestUtil.CountTransactionsInStream
(@in[0]));
}
finally
{
sleepLatch.CountDown();
@in[0].Close();
bkjm.Close();
}
}
public virtual void TestInitializeBKSharedEdits()
{
MiniDFSCluster cluster = null;
try
{
Configuration conf = new Configuration();
HAUtil.SetAllowStandbyReads(conf, true);
conf.SetInt(DFSConfigKeys.DfsHaTaileditsPeriodKey, 1);
MiniDFSNNTopology topology = MiniDFSNNTopology.SimpleHATopology();
cluster = new MiniDFSCluster.Builder(conf).NnTopology(topology).NumDataNodes(0).Build
();
cluster.WaitActive();
// Shutdown and clear the current filebased shared dir.
cluster.ShutdownNameNodes();
FilePath shareddir = new FilePath(cluster.GetSharedEditsDir(0, 1));
NUnit.Framework.Assert.IsTrue("Initial Shared edits dir not fully deleted", FileUtil
.FullyDelete(shareddir));
// Check namenodes should not start without shared dir.
AssertCanNotStartNamenode(cluster, 0);
AssertCanNotStartNamenode(cluster, 1);
// Configure bkjm as new shared edits dir in both namenodes
Configuration nn1Conf = cluster.GetConfiguration(0);
Configuration nn2Conf = cluster.GetConfiguration(1);
nn1Conf.Set(DFSConfigKeys.DfsNamenodeSharedEditsDirKey, BKJMUtil.CreateJournalURI
("/initializeSharedEdits").ToString());
nn2Conf.Set(DFSConfigKeys.DfsNamenodeSharedEditsDirKey, BKJMUtil.CreateJournalURI
("/initializeSharedEdits").ToString());
BKJMUtil.AddJournalManagerDefinition(nn1Conf);
BKJMUtil.AddJournalManagerDefinition(nn2Conf);
// Initialize the BKJM shared edits.
NUnit.Framework.Assert.IsFalse(NameNode.InitializeSharedEdits(nn1Conf));
// NameNode should be able to start and should be in sync with BKJM as
// shared dir
AssertCanStartHANameNodes(cluster, conf, "/testBKJMInitialize");
}
finally
{
if (cluster != null)
{
cluster.Shutdown();
}
}
}
public override void SetupCluster()
{
Configuration conf = SetupCommonConfig();
conf.Set(DFSConfigKeys.DfsNamenodeSharedEditsDirKey, BKJMUtil.CreateJournalURI("/checkpointing"
+ journalCount++).ToString());
BKJMUtil.AddJournalManagerDefinition(conf);
MiniDFSNNTopology topology = new MiniDFSNNTopology().AddNameservice(new MiniDFSNNTopology.NSConf
("ns1").AddNN(new MiniDFSNNTopology.NNConf("nn1").SetHttpPort(10001)).AddNN(new
MiniDFSNNTopology.NNConf("nn2").SetHttpPort(10002)));
cluster = new MiniDFSCluster.Builder(conf).NnTopology(topology).NumDataNodes(1).ManageNameDfsSharedDirs
(false).Build();
cluster.WaitActive();
nn0 = cluster.GetNameNode(0);
nn1 = cluster.GetNameNode(1);
fs = HATestUtil.ConfigureFailoverFs(cluster, conf);
cluster.TransitionToActive(0);
}
public virtual void TestConcurrentFormat()
{
URI uri = BKJMUtil.CreateJournalURI("/hdfsjournal-concurrentformat");
NamespaceInfo nsi = NewNSInfo();
// populate with data first
BookKeeperJournalManager bkjm = new BookKeeperJournalManager(conf, uri, nsi);
bkjm.Format(nsi);
for (int i = 1; i < 100 * 2; i += 2)
{
bkjm.StartLogSegment(i, NameNodeLayoutVersion.CurrentLayoutVersion);
bkjm.FinalizeLogSegment(i, i + 1);
}
bkjm.Close();
int numThreads = 40;
IList <Callable <TestBookKeeperJournalManager.ThreadStatus> > threads = new AList <Callable
<TestBookKeeperJournalManager.ThreadStatus> >();
CyclicBarrier barrier = new CyclicBarrier(numThreads);
for (int i_1 = 0; i_1 < numThreads; i_1++)
{
threads.AddItem(new _Callable_784(uri, nsi, barrier));
}
ExecutorService service = Executors.NewFixedThreadPool(numThreads);
IList <Future <TestBookKeeperJournalManager.ThreadStatus> > statuses = service.InvokeAll
(threads, 60, TimeUnit.Seconds);
int numCompleted = 0;
foreach (Future <TestBookKeeperJournalManager.ThreadStatus> s in statuses)
{
NUnit.Framework.Assert.IsTrue(s.IsDone());
NUnit.Framework.Assert.IsTrue("Thread threw invalid exception", s.Get() == TestBookKeeperJournalManager.ThreadStatus
.Completed || s.Get() == TestBookKeeperJournalManager.ThreadStatus.Goodexception
);
if (s.Get() == TestBookKeeperJournalManager.ThreadStatus.Completed)
{
numCompleted++;
}
}
Log.Info("Completed " + numCompleted + " formats");
NUnit.Framework.Assert.IsTrue("No thread managed to complete formatting", numCompleted
> 0);
}
public virtual void TestNumberOfTransactionsWithGaps()
{
NamespaceInfo nsi = NewNSInfo();
BookKeeperJournalManager bkjm = new BookKeeperJournalManager(conf, BKJMUtil.CreateJournalURI
("/hdfsjournal-gaps"), nsi);
bkjm.Format(nsi);
long txid = 1;
for (long i = 0; i < 3; i++)
{
long start = txid;
EditLogOutputStream @out = bkjm.StartLogSegment(start, NameNodeLayoutVersion.CurrentLayoutVersion
);
for (long j = 1; j <= DefaultSegmentSize; j++)
{
FSEditLogOp op = FSEditLogTestUtil.GetNoOpInstance();
op.SetTransactionId(txid++);
@out.Write(op);
}
@out.Close();
bkjm.FinalizeLogSegment(start, txid - 1);
NUnit.Framework.Assert.IsNotNull(zkc.Exists(bkjm.FinalizedLedgerZNode(start, txid
- 1), false));
}
zkc.Delete(bkjm.FinalizedLedgerZNode(DefaultSegmentSize + 1, DefaultSegmentSize *
2), -1);
long numTrans = bkjm.GetNumberOfTransactions(1, true);
NUnit.Framework.Assert.AreEqual(DefaultSegmentSize, numTrans);
try
{
numTrans = bkjm.GetNumberOfTransactions(DefaultSegmentSize + 1, true);
NUnit.Framework.Assert.Fail("Should have thrown corruption exception by this point"
);
}
catch (JournalManager.CorruptionException)
{
}
// if we get here, everything is going good
numTrans = bkjm.GetNumberOfTransactions((DefaultSegmentSize * 2) + 1, true);
NUnit.Framework.Assert.AreEqual(DefaultSegmentSize, numTrans);
}
public virtual void TestMultiplePrimariesStarted()
{
Path p1 = new Path("/testBKJMMultiplePrimary");
MiniDFSCluster cluster = null;
try
{
Configuration conf = new Configuration();
conf.SetInt(DFSConfigKeys.DfsHaTaileditsPeriodKey, 1);
conf.Set(DFSConfigKeys.DfsNamenodeSharedEditsDirKey, BKJMUtil.CreateJournalURI("/hotfailoverMultiple"
).ToString());
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);
FileSystem fs = HATestUtil.ConfigureFailoverFs(cluster, conf);
fs.Mkdirs(p1);
nn1.GetRpcServer().RollEditLog();
cluster.TransitionToActive(1);
fs = cluster.GetFileSystem(0);
// get the older active server.
try
{
fs.Delete(p1, true);
NUnit.Framework.Assert.Fail("Log update on older active should cause it to exit");
}
catch (RemoteException re)
{
NUnit.Framework.Assert.IsTrue(re.GetClassName().Contains("ExitException"));
}
}
finally
{
if (cluster != null)
{
cluster.Shutdown();
}
}
}
public virtual void TestCorruptInprogressNode()
{
URI uri = BKJMUtil.CreateJournalURI("/hdfsjournal-corruptInprogress");
NamespaceInfo nsi = NewNSInfo();
BookKeeperJournalManager bkjm = new BookKeeperJournalManager(conf, uri, nsi);
bkjm.Format(nsi);
EditLogOutputStream @out = bkjm.StartLogSegment(1, NameNodeLayoutVersion.CurrentLayoutVersion
);
for (long i = 1; i <= 100; i++)
{
FSEditLogOp op = FSEditLogTestUtil.GetNoOpInstance();
op.SetTransactionId(i);
@out.Write(op);
}
@out.Close();
bkjm.FinalizeLogSegment(1, 100);
@out = bkjm.StartLogSegment(101, NameNodeLayoutVersion.CurrentLayoutVersion);
@out.Close();
bkjm.Close();
string inprogressZNode = bkjm.InprogressZNode(101);
zkc.SetData(inprogressZNode, Sharpen.Runtime.GetBytesForString("WholeLottaJunk"),
-1);
bkjm = new BookKeeperJournalManager(conf, uri, nsi);
try
{
bkjm.RecoverUnfinalizedSegments();
NUnit.Framework.Assert.Fail("Should have failed. There should be no way of creating"
+ " an empty inprogess znode");
}
catch (IOException e)
{
// correct behaviour
NUnit.Framework.Assert.IsTrue("Exception different than expected", e.Message.Contains
("has no field named"));
}
finally
{
bkjm.Close();
}
}
public virtual void TestNumberOfTransactionsWithInprogressAtEnd()
{
NamespaceInfo nsi = NewNSInfo();
BookKeeperJournalManager bkjm = new BookKeeperJournalManager(conf, BKJMUtil.CreateJournalURI
("/hdfsjournal-inprogressAtEnd"), nsi);
bkjm.Format(nsi);
long txid = 1;
for (long i = 0; i < 3; i++)
{
long start = txid;
EditLogOutputStream @out = bkjm.StartLogSegment(start, NameNodeLayoutVersion.CurrentLayoutVersion
);
for (long j = 1; j <= DefaultSegmentSize; j++)
{
FSEditLogOp op = FSEditLogTestUtil.GetNoOpInstance();
op.SetTransactionId(txid++);
@out.Write(op);
}
@out.Close();
bkjm.FinalizeLogSegment(start, (txid - 1));
NUnit.Framework.Assert.IsNotNull(zkc.Exists(bkjm.FinalizedLedgerZNode(start, (txid
- 1)), false));
}
long start_1 = txid;
EditLogOutputStream out_1 = bkjm.StartLogSegment(start_1, NameNodeLayoutVersion.CurrentLayoutVersion
);
for (long j_1 = 1; j_1 <= DefaultSegmentSize / 2; j_1++)
{
FSEditLogOp op = FSEditLogTestUtil.GetNoOpInstance();
op.SetTransactionId(txid++);
out_1.Write(op);
}
out_1.SetReadyToFlush();
out_1.Flush();
out_1.Abort();
out_1.Close();
long numTrans = bkjm.GetNumberOfTransactions(1, true);
NUnit.Framework.Assert.AreEqual((txid - 1), numTrans);
}
public virtual void TestRefinalizeAlreadyFinalizedInprogress()
{
URI uri = BKJMUtil.CreateJournalURI("/hdfsjournal-refinalizeInprogressLedger");
NamespaceInfo nsi = NewNSInfo();
BookKeeperJournalManager bkjm = new BookKeeperJournalManager(conf, uri, nsi);
bkjm.Format(nsi);
EditLogOutputStream @out = bkjm.StartLogSegment(1, NameNodeLayoutVersion.CurrentLayoutVersion
);
for (long i = 1; i <= 100; i++)
{
FSEditLogOp op = FSEditLogTestUtil.GetNoOpInstance();
op.SetTransactionId(i);
@out.Write(op);
}
@out.Close();
bkjm.Close();
string inprogressZNode = bkjm.InprogressZNode(1);
string finalizedZNode = bkjm.FinalizedLedgerZNode(1, 100);
NUnit.Framework.Assert.IsNotNull("inprogress znode doesn't exist", zkc.Exists(inprogressZNode
, null));
NUnit.Framework.Assert.IsNull("finalized znode exists", zkc.Exists(finalizedZNode
, null));
byte[] inprogressData = zkc.GetData(inprogressZNode, false, null);
// finalize
bkjm = new BookKeeperJournalManager(conf, uri, nsi);
bkjm.RecoverUnfinalizedSegments();
bkjm.Close();
NUnit.Framework.Assert.IsNull("inprogress znode exists", zkc.Exists(inprogressZNode
, null));
NUnit.Framework.Assert.IsNotNull("finalized znode doesn't exist", zkc.Exists(finalizedZNode
, null));
zkc.Create(inprogressZNode, inprogressData, ZooDefs.Ids.OpenAclUnsafe, CreateMode
.Persistent);
// should work fine
bkjm = new BookKeeperJournalManager(conf, uri, nsi);
bkjm.RecoverUnfinalizedSegments();
bkjm.Close();
}
public virtual void TestEmptyInputStream()
{
ZooKeeper zk = BKJMUtil.ConnectZooKeeper();
BookKeeper bkc = new BookKeeper(new ClientConfiguration(), zk);
try
{
LedgerHandle lh = bkc.CreateLedger(BookKeeper.DigestType.Crc32, Sharpen.Runtime.GetBytesForString
("foobar"));
lh.Close();
EditLogLedgerMetadata metadata = new EditLogLedgerMetadata("/foobar", HdfsConstants
.NamenodeLayoutVersion, lh.GetId(), unchecked ((int)(0x1234)));
try
{
new BookKeeperEditLogInputStream(lh, metadata, -1);
NUnit.Framework.Assert.Fail("Shouldn't get this far, should have thrown");
}
catch (IOException ioe)
{
NUnit.Framework.Assert.IsTrue(ioe.Message.Contains("Invalid first bk entry to read"
));
}
metadata = new EditLogLedgerMetadata("/foobar", HdfsConstants.NamenodeLayoutVersion
, lh.GetId(), unchecked ((int)(0x1234)));
try
{
new BookKeeperEditLogInputStream(lh, metadata, 0);
NUnit.Framework.Assert.Fail("Shouldn't get this far, should have thrown");
}
catch (IOException ioe)
{
NUnit.Framework.Assert.IsTrue(ioe.Message.Contains("Invalid first bk entry to read"
));
}
}
finally
{
bkc.Close();
zk.Close();
}
}
public virtual void TestNumberOfTransactions()
{
NamespaceInfo nsi = NewNSInfo();
BookKeeperJournalManager bkjm = new BookKeeperJournalManager(conf, BKJMUtil.CreateJournalURI
("/hdfsjournal-txncount"), nsi);
bkjm.Format(nsi);
EditLogOutputStream @out = bkjm.StartLogSegment(1, NameNodeLayoutVersion.CurrentLayoutVersion
);
for (long i = 1; i <= 100; i++)
{
FSEditLogOp op = FSEditLogTestUtil.GetNoOpInstance();
op.SetTransactionId(i);
@out.Write(op);
}
@out.Close();
bkjm.FinalizeLogSegment(1, 100);
long numTrans = bkjm.GetNumberOfTransactions(1, true);
NUnit.Framework.Assert.AreEqual(100, numTrans);
}
public virtual void SetUp()
{
Configuration conf = new Configuration();
conf.SetInt(DFSConfigKeys.DfsNamenodeCheckpointCheckPeriodKey, 1);
conf.SetInt(DFSConfigKeys.DfsNamenodeCheckpointTxnsKey, 5);
conf.SetInt(DFSConfigKeys.DfsHaTaileditsPeriodKey, 1);
conf.Set(DFSConfigKeys.DfsNamenodeSharedEditsDirKey, BKJMUtil.CreateJournalURI("/bootstrapStandby"
).ToString());
BKJMUtil.AddJournalManagerDefinition(conf);
conf.SetBoolean(DFSConfigKeys.DfsImageCompressKey, true);
conf.Set(DFSConfigKeys.DfsImageCompressionCodecKey, typeof(TestStandbyCheckpoints.SlowCodec
).GetCanonicalName());
CompressionCodecFactory.SetCodecClasses(conf, ImmutableList.Of <Type>(typeof(TestStandbyCheckpoints.SlowCodec
)));
MiniDFSNNTopology topology = new MiniDFSNNTopology().AddNameservice(new MiniDFSNNTopology.NSConf
("ns1").AddNN(new MiniDFSNNTopology.NNConf("nn1").SetHttpPort(10001)).AddNN(new
MiniDFSNNTopology.NNConf("nn2").SetHttpPort(10002)));
cluster = new MiniDFSCluster.Builder(conf).NnTopology(topology).NumDataNodes(1).ManageNameDfsSharedDirs
(false).Build();
cluster.WaitActive();
}
public virtual void TestNameNodeMultipleSwitchesUsingBKJM()
{
MiniDFSCluster cluster = null;
try
{
Configuration conf = new Configuration();
conf.SetInt(DFSConfigKeys.DfsHaTaileditsPeriodKey, 1);
conf.Set(DFSConfigKeys.DfsNamenodeSharedEditsDirKey, BKJMUtil.CreateJournalURI("/correctEditLogSelection"
).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);
nn1.GetRpcServer().RollEditLog();
// Roll Edits from current Active.
// Transition to standby current active gracefully.
cluster.TransitionToStandby(0);
// Make the other Active and Roll edits multiple times
cluster.TransitionToActive(1);
nn2.GetRpcServer().RollEditLog();
nn2.GetRpcServer().RollEditLog();
// Now One more failover. So NN1 should be able to failover successfully.
cluster.TransitionToStandby(1);
cluster.TransitionToActive(0);
}
finally
{
if (cluster != null)
{
cluster.Shutdown();
}
}
}
public virtual void TestFailoverWithFailingBKCluster()
{
int ensembleSize = numBookies + 1;
BookieServer newBookie = bkutil.NewBookie();
NUnit.Framework.Assert.AreEqual("New bookie didn't start", ensembleSize, bkutil.CheckBookiesUp
(ensembleSize, 10));
BookieServer replacementBookie = null;
MiniDFSCluster cluster = null;
try
{
Configuration conf = new Configuration();
conf.SetInt(DFSConfigKeys.DfsHaTaileditsPeriodKey, 1);
conf.Set(DFSConfigKeys.DfsNamenodeSharedEditsDirKey, BKJMUtil.CreateJournalURI("/hotfailoverWithFail"
).ToString());
conf.SetInt(BookKeeperJournalManager.BkjmBookkeeperEnsembleSize, ensembleSize);
conf.SetInt(BookKeeperJournalManager.BkjmBookkeeperQuorumSize, 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
NUnit.Framework.Assert.AreEqual("New bookie didn't stop", numBookies, bkutil.CheckBookiesUp
(numBookies, 10));
try
{
fs.Mkdirs(p2);
NUnit.Framework.Assert.Fail("mkdirs should result in the NN exiting");
}
catch (RemoteException re)
{
NUnit.Framework.Assert.IsTrue(re.GetClassName().Contains("ExitException"));
}
cluster.ShutdownNameNode(0);
try
{
cluster.TransitionToActive(1);
NUnit.Framework.Assert.Fail("Shouldn't have been able to transition with bookies down"
);
}
catch (ExitUtil.ExitException ee)
{
NUnit.Framework.Assert.IsTrue("Should shutdown due to required journal failure",
ee.Message.Contains("starting log segment 3 failed for required journal"));
}
replacementBookie = bkutil.NewBookie();
NUnit.Framework.Assert.AreEqual("Replacement bookie didn't start", ensembleSize,
bkutil.CheckBookiesUp(ensembleSize, 10));
cluster.TransitionToActive(1);
// should work fine now
NUnit.Framework.Assert.IsTrue(fs.Exists(p1));
NUnit.Framework.Assert.IsFalse(fs.Exists(p2));
}
finally
{
newBookie.Shutdown();
if (replacementBookie != null)
{
replacementBookie.Shutdown();
}
if (cluster != null)
{
cluster.Shutdown();
}
}
}
public virtual void TestWriteRestartFrom1()
{
NamespaceInfo nsi = NewNSInfo();
BookKeeperJournalManager bkjm = new BookKeeperJournalManager(conf, BKJMUtil.CreateJournalURI
("/hdfsjournal-restartFrom1"), nsi);
bkjm.Format(nsi);
long txid = 1;
long start = txid;
EditLogOutputStream @out = bkjm.StartLogSegment(txid, NameNodeLayoutVersion.CurrentLayoutVersion
);
for (long j = 1; j <= DefaultSegmentSize; j++)
{
FSEditLogOp op = FSEditLogTestUtil.GetNoOpInstance();
op.SetTransactionId(txid++);
@out.Write(op);
}
@out.Close();
bkjm.FinalizeLogSegment(start, (txid - 1));
txid = 1;
try
{
@out = bkjm.StartLogSegment(txid, NameNodeLayoutVersion.CurrentLayoutVersion);
NUnit.Framework.Assert.Fail("Shouldn't be able to start another journal from " +
txid + " when one already exists");
}
catch (Exception ioe)
{
Log.Info("Caught exception as expected", ioe);
}
// test border case
txid = DefaultSegmentSize;
try
{
@out = bkjm.StartLogSegment(txid, NameNodeLayoutVersion.CurrentLayoutVersion);
NUnit.Framework.Assert.Fail("Shouldn't be able to start another journal from " +
txid + " when one already exists");
}
catch (IOException ioe)
{
Log.Info("Caught exception as expected", ioe);
}
// open journal continuing from before
txid = DefaultSegmentSize + 1;
start = txid;
@out = bkjm.StartLogSegment(start, NameNodeLayoutVersion.CurrentLayoutVersion);
NUnit.Framework.Assert.IsNotNull(@out);
for (long j_1 = 1; j_1 <= DefaultSegmentSize; j_1++)
{
FSEditLogOp op = FSEditLogTestUtil.GetNoOpInstance();
op.SetTransactionId(txid++);
@out.Write(op);
}
@out.Close();
bkjm.FinalizeLogSegment(start, (txid - 1));
// open journal arbitarily far in the future
txid = DefaultSegmentSize * 4;
@out = bkjm.StartLogSegment(txid, NameNodeLayoutVersion.CurrentLayoutVersion);
NUnit.Framework.Assert.IsNotNull(@out);
}
public virtual void Setup()
{
zkc = BKJMUtil.ConnectZooKeeper();
}
public virtual void TestOneBookieFailure()
{
newBookie = bkutil.NewBookie();
BookieServer replacementBookie = null;
try
{
int ensembleSize = numBookies + 1;
NUnit.Framework.Assert.AreEqual("New bookie didn't start", ensembleSize, bkutil.CheckBookiesUp
(ensembleSize, 10));
// ensure that the journal manager has to use all bookies,
// so that a failure will fail the journal manager
Configuration conf = new Configuration();
conf.SetInt(BookKeeperJournalManager.BkjmBookkeeperEnsembleSize, ensembleSize);
conf.SetInt(BookKeeperJournalManager.BkjmBookkeeperQuorumSize, ensembleSize);
long txid = 1;
NamespaceInfo nsi = NewNSInfo();
BookKeeperJournalManager bkjm = new BookKeeperJournalManager(conf, BKJMUtil.CreateJournalURI
("/hdfsjournal-onebookiefailure"), nsi);
bkjm.Format(nsi);
EditLogOutputStream @out = bkjm.StartLogSegment(txid, NameNodeLayoutVersion.CurrentLayoutVersion
);
for (long i = 1; i <= 3; i++)
{
FSEditLogOp op = FSEditLogTestUtil.GetNoOpInstance();
op.SetTransactionId(txid++);
@out.Write(op);
}
@out.SetReadyToFlush();
@out.Flush();
replacementBookie = bkutil.NewBookie();
NUnit.Framework.Assert.AreEqual("replacement bookie didn't start", ensembleSize +
1, bkutil.CheckBookiesUp(ensembleSize + 1, 10));
newBookie.Shutdown();
NUnit.Framework.Assert.AreEqual("New bookie didn't die", ensembleSize, bkutil.CheckBookiesUp
(ensembleSize, 10));
for (long i_1 = 1; i_1 <= 3; i_1++)
{
FSEditLogOp op = FSEditLogTestUtil.GetNoOpInstance();
op.SetTransactionId(txid++);
@out.Write(op);
}
@out.SetReadyToFlush();
@out.Flush();
}
catch (Exception e)
{
Log.Error("Exception in test", e);
throw;
}
finally
{
if (replacementBookie != null)
{
replacementBookie.Shutdown();
}
newBookie.Shutdown();
if (bkutil.CheckBookiesUp(numBookies, 30) != numBookies)
{
Log.Warn("Not all bookies from this test shut down, expect errors");
}
}
}
public static void StartBK()
{
journalCount = 0;
bkutil = new BKJMUtil(numBookies);
bkutil.Start();
}
public static void SetupBookkeeper()
{
bkutil = new BKJMUtil(numBookies);
bkutil.Start();
}
