/**
* Make a thread allocate a vlsn, but then fail before it's tracked by the vlsn index. This
* happened in [#20919] when 1.rep environment close was called 2.the repNode was nulled out 3.a
* concurrent writing thread got a NPE within its call to LogManager.log because the repNode was
* null. This thread exited after it had bumped the vlsn, but before it had entered the vlsn in
* the vlsnIndex 4.rep environment close tried to do a checkpoint, but the checkpoint hung. This
* fix works by having (3) invalidate the environment, and by having (4) check for an invalidated
* environment.
*/
@Test
public void testLoggingFailure() throws DatabaseException, IOException {
/* Make a single replicated environment. */
RepEnvInfo[] repEnvInfo = RepTestUtils.setupEnvInfos(envRoot, 1);
RepTestUtils.joinGroup(repEnvInfo);
/*
* Disable cleaning and CBVLSN updating, to control vlsn creation
* explicitly.
*/
Environment env = repEnvInfo[0].getEnv();
EnvironmentMutableConfig config = env.getMutableConfig();
config.setConfigParam("je.env.runCleaner", "false");
env.setMutableConfig(config);
LocalCBVLSNUpdater.setSuppressGroupDBUpdates(false);
DatabaseConfig dbConfig = new DatabaseConfig();
dbConfig.setTransactional(true);
dbConfig.setAllowCreate(true);
Database db = env.openDatabase(null, "foo", dbConfig);
DatabaseEntry value = new DatabaseEntry(new byte[4]);
EnvironmentImpl envImpl = DbInternal.getEnvironmentImpl(env);
LogManager logManager = DbInternal.getEnvironmentImpl(env).getLogManager();
/*
* Inject an exception into the next call to log() that is made
* for a replicated log entry.
*/
logManager.setDelayVLSNRegisterHook(new ForceException());
VLSNIndex vlsnIndex = ((RepImpl) envImpl).getVLSNIndex();
try {
db.put(null, value, value);
fail("Should throw exception");
} catch (Exception expected) {
assertTrue(
"latest="
+ vlsnIndex.getLatestAllocatedVal()
+ " last mapped="
+ vlsnIndex.getRange().getLast().getSequence(),
vlsnIndex.getLatestAllocatedVal() > vlsnIndex.getRange().getLast().getSequence());
}
try {
VLSNIndex.AWAIT_CONSISTENCY_MS = 1000;
envImpl.awaitVLSNConsistency();
fail("Should throw and break out");
} catch (DatabaseException expected) {
}
/* Before the fix, this test hung. */
}
/**
* 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();
}
}