/* Insert 100 records begins with the beginKey. */
private void doWork(Environment master, String dbName, int beginKey) throws Exception {
DatabaseConfig dbConfig = new DatabaseConfig();
dbConfig.setAllowCreate(true);
dbConfig.setTransactional(true);
/* Insert/Update the records of the database. */
Database db = master.openDatabase(null, dbName, dbConfig);
DatabaseEntry key = new DatabaseEntry();
DatabaseEntry data = new DatabaseEntry();
for (int i = 0; i < 100; i++) {
IntegerBinding.intToEntry(beginKey + i, key);
StringBinding.stringToEntry("herococo", data);
db.put(null, key, data);
}
db.close();
/*
* Do a sync at the end of the stage to make sure master and
* replica have the same data set.
*/
VLSN commitVLSN = RepTestUtils.syncGroupToLastCommit(repEnvInfo, repEnvInfo.length);
RepTestUtils.checkNodeEquality(commitVLSN, false, repEnvInfo);
}
/**
* Bounce the master, causing replica1 to switch roles with the master. If a higher appVersion is
* specified, the bounced node will also be upgraded.
*/
private void bounceMaster(final int appVersion) throws Exception {
/* Disable updates to RepGroupDB due to LocalCBVLSN updates. */
LocalCBVLSNUpdater.setSuppressGroupDBUpdates(true);
for (RepEnvInfo info : repEnvInfo) {
if (info.getEnv() == masterEnv) {
/*
* Sync up the replication group so that node2 doesn't do
* hard recovery.
*/
RepTestUtils.syncGroupToLastCommit(repEnvInfo, repEnvInfo.length);
/* Disable replay on replicas. */
shutdownFeeder(info.getRepNode(), replicaEnv1);
shutdownFeeder(info.getRepNode(), replicaEnv2);
/* Close the master. */
masterApp.close();
masterApp = null;
info.closeEnv();
masterEnv = null;
/* Force repEnvInfo[2] to the master. */
WaitForMasterListener masterWaiter = new WaitForMasterListener();
replicaEnv2.setStateChangeListener(masterWaiter);
RepNode repNode = repEnvInfo[2].getRepNode();
repNode.forceMaster(true);
/* Enable the LocalCBVLSN updates. */
LocalCBVLSNUpdater.setSuppressGroupDBUpdates(false);
masterWaiter.awaitMastership();
assertTrue(repNode.isMaster());
masterEnv = replicaEnv2;
/* Replica2 was elected, swap names with replica1. */
final ReplicatedEnvironment tmpEnv = replicaEnv1;
replicaEnv1 = replicaEnv2;
replicaEnv2 = tmpEnv;
final AppInterface tmpApp = replicaApp1;
replicaApp1 = replicaApp2;
replicaApp2 = tmpApp;
/* Replica1 (or 2, see above) has been elected master. */
masterApp = newAppObject(appVersion);
masterApp.adopt(replicaApp1);
/* Former master (just upgraded) becomes replica1. */
replicaEnv1 = info.openEnv();
replicaApp1.open(replicaEnv1);
break;
}
}
assertNotNull(masterApp);
assertSame(masterEnv.getState(), ReplicatedEnvironment.State.MASTER);
}
/**
* [#18882] Before this bug fix, this test would result in a java.io.FileNotFoundException out of
* FeederReader$SwitchWindow.fillNext.
*/
@Test
public void testDataInWriteQueue() throws Exception {
openGroup();
ExecutorService appThreads = Executors.newFixedThreadPool(numThreads);
int opsPerThread = numRecords / numThreads;
for (int i = 0; i < numThreads; i++) {
appThreads.execute(new AppWork(i, opsPerThread));
}
appThreads.shutdown();
appThreads.awaitTermination(6000, TimeUnit.SECONDS);
VLSN vlsn = RepTestUtils.syncGroupToLastCommit(repEnvInfo, repEnvInfo.length);
RepTestUtils.checkNodeEquality(vlsn, verbose, repEnvInfo);
closeGroup();
}
/** Simulates the scenario where an entire group goes down and is restarted. */
@Test
public void testAllJoinLeaveJoinGroup() throws DatabaseException, InterruptedException {
createGroup();
ReplicatedEnvironment masterRep = repEnvInfo[0].getEnv();
populateDB(masterRep, TEST_DB_NAME, 100);
RepTestUtils.syncGroupToLastCommit(repEnvInfo, repEnvInfo.length);
/* Shutdown the entire group. */
closeNodes(repEnvInfo);
/*
* Restart the group, using a longer join wait time to allow the
* secondary to query the primaries a second time after the election is
* complete. See RepNode.MASTER_QUERY_INTERVAL.
*/
final long masterQueryInterval = 10000;
restartNodes(JOIN_WAIT_TIME + masterQueryInterval, repEnvInfo);
}
@Test
public void testNoQuorum() throws DatabaseException, InterruptedException {
for (int i = 0; i < 3; i++) {
ReplicatedEnvironment rep = repEnvInfo[i].openEnv();
State state = rep.getState();
assertEquals((i == 0) ? State.MASTER : State.REPLICA, state);
}
RepTestUtils.syncGroupToLastCommit(repEnvInfo, 3);
repEnvInfo[1].closeEnv();
repEnvInfo[2].closeEnv();
// A new node joining in the absence of a quorum must fail
try {
repEnvInfo[3].openEnv();
fail("Expected exception");
} catch (UnknownMasterException e) {
/* Expected. */
}
}
/**
* 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();
}
}