/*
* Provoke sufficient log cleaning to move the entire VLSN right
* sufficiently that the new VLSN range no longer overlaps the VLSN
* range upon entry thus guaranteeing a LogFileRefreshException.
*/
private void shiftVLSNRight(ReplicatedEnvironment menv) {
/* Shift the vlsn range window. */
RepImpl menvImpl = repEnvInfo[0].getRepImpl();
final VLSNIndex vlsnIndex = menvImpl.getRepNode().getVLSNIndex();
VLSN masterHigh = vlsnIndex.getRange().getLast();
CheckpointConfig checkpointConfig = new CheckpointConfig();
checkpointConfig.setForce(true);
do {
/*
* Populate just the master, leaving the replica behind Re-populate
* with the same keys to create Cleaner fodder.
*/
populateDB(menv, TEST_DB_NAME, 100);
/*
* Sleep to permit the cbvlsn on the master to be updated. It's
* done with the period: FeederManager.MASTER_CHANGE_CHECK_TIMEOUT
*/
try {
Thread.sleep(1000);
} catch (InterruptedException e) {
fail("unexpected interrupt");
}
menv.cleanLog();
menv.checkpoint(checkpointConfig);
} while (masterHigh.compareTo(vlsnIndex.getRange().getFirst()) > 0);
}
public void setLastAppliedVLSN(VLSN justApplied) {
if (justApplied.compareTo(lastApplied) <= 0) {
throw EnvironmentFailureException.unexpectedState(
"Txn "
+ getId()
+ " attempted VLSN = "
+ justApplied
+ " txnLastApplied = "
+ lastApplied);
}
this.lastApplied = justApplied;
}
/**
* Check the RepNode.currentCommitVLSN difference to see if there is any dirty data between two
* actions. We only do the flush when there exists dirty data.
*
* <p>The reason that why we only cares about the commit VLSN is those unlogged
* uncommitted/abort transaction will be aborted during recovery. It's useless to keep track of
* those VLSNs.
*/
@Override
public void run() {
final VLSN newTxnEndVLSN = repNode.getCurrentTxnEndVLSN();
/* Do nothing if no updates. */
if (newTxnEndVLSN == null) {
return;
}
if (lastTxnEndVLSN == null || newTxnEndVLSN.compareTo(lastTxnEndVLSN) == 1) {
lastTxnEndVLSN = newTxnEndVLSN;
repNode.getRepImpl().getLogManager().flush();
}
}
/**
* This is really multiple tests in one. It tests network restore with a replica in each of the
* following three states:
*
* <p>1) A brand new node joining the group and needing a network restore.
*
* <p>2) An existing node with its own unique log needing a network restore.
*
* <p>3) Repeated network restores, reflecting a mature node.
*/
@Test
public void testBasic() throws DatabaseException, Exception {
/*
* The cleaner thread can see InsufficientLogExceptions so just stifle
* those exceptions from stderr.
*/
DaemonThread.stifleExceptionChatter = true;
configureForMaxCleaning(2);
final RepEnvInfo info1 = repEnvInfo[0];
RepEnvInfo info2 = repEnvInfo[1];
ReplicatedEnvironment masterRep = info1.openEnv();
Environment menv = masterRep;
EnvironmentMutableConfig mconfig = menv.getMutableConfig();
mconfig.setConfigParam(EnvironmentParams.ENV_RUN_CLEANER.getName(), "false");
menv.setMutableConfig(mconfig);
/*
* Have just the master join first. We do this to test the special case
* of a brand new node joining a group and needing VLSN 1. The same
* node then rejoins with its VLSN > 1 to test subsequent rejoins
* where the node has already participated in the replication.
*/
populateDB(masterRep, TEST_DB_NAME, 100);
mconfig = menv.getMutableConfig();
mconfig.setConfigParam(EnvironmentParams.ENV_RUN_CLEANER.getName(), "true");
menv.setMutableConfig(mconfig);
File cenvDir = info2.getEnvHome();
final int cid = 2;
for (int i = 0; i < RESTORE_CYCLES; i++) {
leaveGroupAllButMaster();
shiftVLSNRight(masterRep);
RepNodeImpl memberPrev =
info1.getRepNode().getGroup().getMember(info2.getRepConfig().getNodeName());
/* Node1 is not known on the first iteration. */
final VLSN prevSync = (i == 0) ? null : memberPrev.getBarrierState().getLastCBVLSN();
try {
/* Should force a network restore. */
setExceptionListener(info2);
info2.openEnv();
fail("exception expected");
} catch (InsufficientLogException e) {
RepNodeImpl member =
info1.getRepNode().getGroup().getMember(info2.getRepConfig().getNodeName());
/*
* The sync state should have been advanced to help contribute
* to the global CBVLSN and prevent it from advancing.
*/
final VLSN currSync = member.getBarrierState().getLastCBVLSN();
assertTrue((i == 0) || currSync.compareTo(prevSync) >= 0);
NetworkRestore networkRestore = new NetworkRestore();
networkRestore.execute(e, new NetworkRestoreConfig());
final NetworkBackupStats stats = networkRestore.getNetworkBackupStats();
assertThat(stats.getExpectedBytes(), greaterThan(0));
assertThat(stats.getTransferredBytes(), greaterThan(0));
/* Create a replacement replicator. */
info2 = RepTestUtils.setupEnvInfo(cenvDir, RepTestUtils.DEFAULT_DURABILITY, cid, info1);
setExceptionListener(info2);
info2.openEnv();
}
/* Verify that we can continue with the "restored" log files. */
populateDB(masterRep, TEST_DB_NAME, 100, 100);
VLSN commitVLSN = RepTestUtils.syncGroupToLastCommit(repEnvInfo, 2);
RepTestUtils.checkNodeEquality(commitVLSN, false, repEnvInfo);
info2.closeEnv();
}
}