/// <exception cref="System.IO.IOException"/>
private void CheckRecovery(MiniJournalCluster cluster, long segmentTxId, long expectedEndTxId
)
{
int numFinalized = 0;
for (int i = 0; i < cluster.GetNumNodes(); i++)
{
FilePath logDir = cluster.GetCurrentDir(i, QJMTestUtil.Jid);
FileJournalManager.EditLogFile elf = FileJournalManager.GetLogFile(logDir, segmentTxId
);
if (elf == null)
{
continue;
}
if (!elf.IsInProgress())
{
numFinalized++;
if (elf.GetLastTxId() != expectedEndTxId)
{
NUnit.Framework.Assert.Fail("File " + elf + " finalized to wrong txid, expected "
+ expectedEndTxId);
}
}
}
if (numFinalized < cluster.GetQuorumSize())
{
NUnit.Framework.Assert.Fail("Did not find a quorum of finalized logs starting at "
+ segmentTxId);
}
}
/// <summary>
/// Test the case where, at the beginning of a segment, transactions
/// have been written to one JN but not others.
/// </summary>
/// <exception cref="System.Exception"/>
public virtual void DoTestOutOfSyncAtBeginningOfSegment(int nodeWithOneTxn)
{
int nodeWithEmptySegment = (nodeWithOneTxn + 1) % 3;
int nodeMissingSegment = (nodeWithOneTxn + 2) % 3;
QJMTestUtil.WriteSegment(cluster, qjm, 1, 3, true);
WaitForAllPendingCalls(qjm.GetLoggerSetForTests());
cluster.GetJournalNode(nodeMissingSegment).StopAndJoin(0);
// Open segment on 2/3 nodes
EditLogOutputStream stm = qjm.StartLogSegment(4, NameNodeLayoutVersion.CurrentLayoutVersion
);
try
{
WaitForAllPendingCalls(qjm.GetLoggerSetForTests());
// Write transactions to only 1/3 nodes
FailLoggerAtTxn(spies[nodeWithEmptySegment], 4);
try
{
QJMTestUtil.WriteTxns(stm, 4, 1);
NUnit.Framework.Assert.Fail("Did not fail even though 2/3 failed");
}
catch (QuorumException qe)
{
GenericTestUtils.AssertExceptionContains("mock failure", qe);
}
}
finally
{
stm.Abort();
}
// Bring back the down JN.
cluster.RestartJournalNode(nodeMissingSegment);
// Make a new QJM. At this point, the state is as follows:
// A: nodeWithEmptySegment: 1-3 finalized, 4_inprogress (empty)
// B: nodeWithOneTxn: 1-3 finalized, 4_inprogress (1 txn)
// C: nodeMissingSegment: 1-3 finalized
GenericTestUtils.AssertGlobEquals(cluster.GetCurrentDir(nodeWithEmptySegment, QJMTestUtil
.Jid), "edits_.*", NNStorage.GetFinalizedEditsFileName(1, 3), NNStorage.GetInProgressEditsFileName
(4));
GenericTestUtils.AssertGlobEquals(cluster.GetCurrentDir(nodeWithOneTxn, QJMTestUtil
.Jid), "edits_.*", NNStorage.GetFinalizedEditsFileName(1, 3), NNStorage.GetInProgressEditsFileName
(4));
GenericTestUtils.AssertGlobEquals(cluster.GetCurrentDir(nodeMissingSegment, QJMTestUtil
.Jid), "edits_.*", NNStorage.GetFinalizedEditsFileName(1, 3));
// Stop one of the nodes. Since we run this test three
// times, rotating the roles of the nodes, we'll test
// all the permutations.
cluster.GetJournalNode(2).StopAndJoin(0);
qjm = CreateSpyingQJM();
qjm.RecoverUnfinalizedSegments();
if (nodeWithOneTxn == 0 || nodeWithOneTxn == 1)
{
// If the node that had the transaction committed was one of the nodes
// that responded during recovery, then we should have recovered txid
// 4.
CheckRecovery(cluster, 4, 4);
QJMTestUtil.WriteSegment(cluster, qjm, 5, 3, true);
}
else
{
// Otherwise, we should have recovered only 1-3 and should be able to
// start a segment at 4.
CheckRecovery(cluster, 1, 3);
QJMTestUtil.WriteSegment(cluster, qjm, 4, 3, true);
}
}