/**
* Verify that the current and previous directories exist. Verify that previous hasn't been
* modified by comparing the checksum of all it's containing files with their original checksum.
* It is assumed that the server has recovered and upgraded. nsLevelUpgrade specify if the upgrade
* is at top level or ns level nsLevelUpgrade=true, we search basedir/current/NS-id/previous
* =false, we search basedir/previous
*/
void checkResult(NodeType nodeType, String[] baseDirs, int nnIndex, boolean simulatedPrevious)
throws IOException {
switch (nodeType) {
case NAME_NODE:
for (int i = 0; i < baseDirs.length; i++) {
assertTrue(new File(baseDirs[i], "current").isDirectory());
assertTrue(new File(baseDirs[i], "current/VERSION").isFile());
assertTrue(new File(baseDirs[i], "current/edits").isFile());
assertTrue(new File(baseDirs[i], "current/fsimage").isFile());
assertTrue(new File(baseDirs[i], "current/fstime").isFile());
}
break;
case DATA_NODE:
for (int i = 0; i < baseDirs.length; i++) {
assertEquals(
UpgradeUtilities.checksumContents(nodeType, new File(baseDirs[i], "current")),
UpgradeUtilities.checksumMasterContents(nodeType));
File nsBaseDir =
NameSpaceSliceStorage.getNsRoot(
cluster.getNameNode(nnIndex).getNamespaceID(), new File(baseDirs[i], "current"));
assertEquals(
UpgradeUtilities.checksumContents(
nodeType, new File(nsBaseDir, MiniDFSCluster.FINALIZED_DIR_NAME)),
UpgradeUtilities.checksumDatanodeNSStorageContents(nnIndex));
}
break;
}
for (int i = 0; i < baseDirs.length; i++) {
switch (nodeType) {
case NAME_NODE:
assertTrue(new File(baseDirs[i], "previous").isDirectory());
assertEquals(
UpgradeUtilities.checksumContents(nodeType, new File(baseDirs[i], "previous")),
UpgradeUtilities.checksumMasterContents(nodeType));
break;
case DATA_NODE:
File nsBaseDir = null;
nsBaseDir =
NameSpaceSliceStorage.getNsRoot(
cluster.getNameNode(nnIndex).getNamespaceID(), new File(baseDirs[i], "current"));
// Top level upgrade should not exist.
assertFalse(new File(baseDirs[i], "previous").isDirectory() && !simulatedPrevious);
assertTrue(new File(nsBaseDir, "previous").isDirectory());
assertEquals(
UpgradeUtilities.checksumContents(
nodeType, new File(nsBaseDir, "previous/finalized")),
UpgradeUtilities.checksumDatanodeNSStorageContents(nnIndex));
}
}
}
/*
* This test attempts to upgrade the datanode from federation
* version -35 to upper version
* This test is for non-federation cluster with single namenode
*/
public void testNonFederationClusterUpgradeAfterFederationVersion() throws Exception {
File[] baseDirs;
UpgradeUtilities.initialize();
for (int numDirs = 1; numDirs <= 2; numDirs++) {
conf = new Configuration();
conf.setInt("dfs.datanode.scan.period.hours", -1);
conf = UpgradeUtilities.initializeStorageStateConf(numDirs, conf);
String[] nameNodeDirs = conf.getStrings("dfs.name.dir");
String[] dataNodeDirs = conf.getStrings("dfs.data.dir");
log("DataNode upgrade with federation layout version in current", numDirs);
UpgradeUtilities.createStorageDirs(NAME_NODE, nameNodeDirs, "current");
try {
cluster = new MiniDFSCluster(conf, 0, StartupOption.UPGRADE);
baseDirs = UpgradeUtilities.createStorageDirs(DATA_NODE, dataNodeDirs, "current");
UpgradeUtilities.createVersionFile(
DATA_NODE,
baseDirs,
new StorageInfo(
FSConstants.FEDERATION_VERSION,
UpgradeUtilities.getCurrentNamespaceID(cluster),
UpgradeUtilities.getCurrentFsscTime(cluster)),
cluster.getNameNode().getNamespaceID());
cluster.startDataNodes(conf, 1, false, StartupOption.REGULAR, null);
checkResult(DATA_NODE, dataNodeDirs, 0, false);
} finally {
if (cluster != null) cluster.shutdown();
UpgradeUtilities.createEmptyDirs(nameNodeDirs);
UpgradeUtilities.createEmptyDirs(dataNodeDirs);
}
}
}
/*
* This test attempts to upgrade the datanode from federation version -35 to
* upper version This test is for federation cluster with 2 namenodes. It
* changes the layout version and ctime.
*/
public void testFederationClusterUpgradeAfterFederationVersionWithCTimeChange() throws Exception {
File[] baseDirs;
Configuration baseConf = new Configuration();
UpgradeUtilities.initialize(2, baseConf, true);
for (int numDirs = 1; numDirs <= 2; numDirs++) {
conf = new Configuration();
conf.setInt("dfs.datanode.scan.period.hours", -1);
conf = UpgradeUtilities.initializeStorageStateConf(numDirs, conf);
String[] nameNodeDirs = conf.getStrings("dfs.name.dir");
String[] dataNodeDirs = conf.getStrings("dfs.data.dir");
log("DataNode upgrade with federation layout version in current and ctime change", numDirs);
UpgradeUtilities.createFederatedNameNodeStorageDirs(nameNodeDirs);
conf.set(
FSConstants.DFS_FEDERATION_NAMESERVICES,
baseConf.get(FSConstants.DFS_FEDERATION_NAMESERVICES));
try {
cluster = new MiniDFSCluster(conf, 0, StartupOption.UPGRADE, false, 2);
baseDirs = UpgradeUtilities.createStorageDirs(DATA_NODE, dataNodeDirs, "current");
for (int i = 0; i < 2; i++) {
UpgradeUtilities.createVersionFile(
DATA_NODE,
baseDirs,
new StorageInfo(
FSConstants.FEDERATION_VERSION,
cluster.getNameNode(i).getNamespaceID(),
cluster.getNameNode(i).versionRequest().getCTime() - 1),
cluster.getNameNode(i).getNamespaceID());
}
cluster.startDataNodes(conf, 1, false, StartupOption.REGULAR, null);
for (int i = 0; i < 2; i++) {
checkResult(DATA_NODE, dataNodeDirs, i, false);
}
} finally {
if (cluster != null) cluster.shutdown();
UpgradeUtilities.createEmptyDirs(nameNodeDirs);
UpgradeUtilities.createEmptyDirs(dataNodeDirs);
}
}
}
/**
* Initialize the data structures used by this class. IMPORTANT NOTE: This method must be called
* once before calling any other public method on this class.
*
* <p>Creates a singleton master populated storage directory for a Namenode (contains edits,
* fsimage, version, and time files) and a Datanode (contains version and block files). This can
* be a lengthy operation.
*/
public static void initialize() throws Exception {
createEmptyDirs(new String[] {TEST_ROOT_DIR.toString()});
Configuration config = new Configuration();
config.set("dfs.name.dir", namenodeStorage.toString());
config.set("dfs.data.dir", datanodeStorage.toString());
MiniDFSCluster cluster = null;
try {
// format data-node
createEmptyDirs(new String[] {datanodeStorage.toString()});
// format and start NameNode and start DataNode
NameNode.format(config);
cluster = new MiniDFSCluster(config, 1, StartupOption.REGULAR);
NameNode namenode = cluster.getNameNode();
namenodeStorageNamespaceID = namenode.versionRequest().getNamespaceID();
namenodeStorageFsscTime = namenode.versionRequest().getCTime();
FileSystem fs = FileSystem.get(config);
Path baseDir = new Path("/TestUpgrade");
fs.mkdirs(baseDir);
// write some files
int bufferSize = 4096;
byte[] buffer = new byte[bufferSize];
for (int i = 0; i < bufferSize; i++) buffer[i] = (byte) ('0' + i % 50);
writeFile(fs, new Path(baseDir, "file1"), buffer, bufferSize);
writeFile(fs, new Path(baseDir, "file2"), buffer, bufferSize);
// save image
namenode.setSafeMode(SafeModeAction.SAFEMODE_ENTER);
namenode.saveNamespace();
namenode.setSafeMode(SafeModeAction.SAFEMODE_LEAVE);
// write more files
writeFile(fs, new Path(baseDir, "file3"), buffer, bufferSize);
writeFile(fs, new Path(baseDir, "file4"), buffer, bufferSize);
} finally {
// shutdown
if (cluster != null) cluster.shutdown();
FileUtil.fullyDelete(new File(namenodeStorage, "in_use.lock"));
FileUtil.fullyDelete(new File(datanodeStorage, "in_use.lock"));
}
namenodeStorageChecksum = checksumContents(NAME_NODE, new File(namenodeStorage, "current"));
datanodeStorageChecksum = checksumContents(DATA_NODE, new File(datanodeStorage, "current"));
}
public void testBlocksScheduledCounter() throws IOException {
MiniDFSCluster cluster = new MiniDFSCluster(new Configuration(), 1, true, null);
cluster.waitActive();
FileSystem fs = cluster.getFileSystem();
// open a file an write a few bytes:
FSDataOutputStream out = fs.create(new Path("/testBlockScheduledCounter"));
for (int i = 0; i < 1024; i++) {
out.write(i);
}
// flush to make sure a block is allocated.
((DFSOutputStream) (out.getWrappedStream())).sync();
ArrayList<DatanodeDescriptor> dnList = new ArrayList<DatanodeDescriptor>();
cluster.getNameNode().namesystem.DFSNodesStatus(dnList, dnList);
DatanodeDescriptor dn = dnList.get(0);
assertEquals(1, dn.getBlocksScheduled());
// close the file and the counter should go to zero.
out.close();
assertEquals(0, dn.getBlocksScheduled());
}
/**
* Test case that stops a writer after finalizing a block but before calling completeFile,
* recovers a file from another writer, starts writing from that writer, and then has the old
* lease holder call completeFile
*/
@Test(timeout = 60000)
public void testCompleteOtherLeaseHoldersFile() throws Throwable {
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(3).build();
try {
cluster.waitActive();
NameNode preSpyNN = cluster.getNameNode();
NameNode spyNN = spy(preSpyNN);
// Delay completeFile
DelayAnswer delayer = new DelayAnswer();
doAnswer(delayer).when(spyNN).complete(anyString(), anyString(), (Block) anyObject());
DFSClient client = new DFSClient(null, spyNN, conf, null);
file1 = new Path("/testCompleteOtherLease");
final OutputStream stm = client.create("/testCompleteOtherLease", true);
// write 1/2 block
AppendTestUtil.write(stm, 0, 4096);
final AtomicReference<Throwable> err = new AtomicReference<Throwable>();
Thread t =
new Thread() {
public void run() {
try {
stm.close();
} catch (Throwable t) {
err.set(t);
}
}
};
t.start();
LOG.info("Waiting for close to get to latch...");
delayer.waitForCall();
// At this point, the block is finalized on the DNs, but the file
// has not been completed in the NN.
// Lose the leases
LOG.info("Killing lease checker");
client.leasechecker.interruptAndJoin();
FileSystem fs1 = cluster.getFileSystem();
FileSystem fs2 = AppendTestUtil.createHdfsWithDifferentUsername(fs1.getConf());
LOG.info("Recovering file");
recoverFile(fs2);
LOG.info("Opening file for append from new fs");
FSDataOutputStream appenderStream = fs2.append(file1);
LOG.info("Writing some data from new appender");
AppendTestUtil.write(appenderStream, 0, 4096);
LOG.info("Telling old close to proceed.");
delayer.proceed();
LOG.info("Waiting for close to finish.");
t.join();
LOG.info("Close finished.");
// We expect that close will get a "Lease mismatch"
// error.
Throwable thrownByClose = err.get();
assertNotNull(thrownByClose);
assertTrue(thrownByClose instanceof IOException);
if (!thrownByClose.getMessage().contains("Lease mismatch")) throw thrownByClose;
// The appender should be able to close properly
appenderStream.close();
} finally {
cluster.shutdown();
}
}
static boolean hasLease(MiniDFSCluster cluster, Path src) {
return cluster.getNameNode().namesystem.leaseManager.getLeaseByPath(src.toString()) != null;
}
/**
* This test attempts to upgrade the NameNode and DataNode under a number of valid and invalid
* conditions.
*/
public void testUpgrade() throws Exception {
File[] baseDirs;
UpgradeUtilities.initialize();
for (int numDirs = 1; numDirs <= 2; numDirs++) {
conf = new Configuration();
conf.setInt("dfs.datanode.scan.period.hours", -1);
conf = UpgradeUtilities.initializeStorageStateConf(numDirs, conf);
String[] nameNodeDirs = conf.getStrings("dfs.name.dir");
String[] dataNodeDirs = conf.getStrings("dfs.data.dir");
log("Normal NameNode upgrade", numDirs);
UpgradeUtilities.createStorageDirs(NAME_NODE, nameNodeDirs, "current");
cluster = new MiniDFSCluster(conf, 0, StartupOption.UPGRADE);
checkResult(NAME_NODE, nameNodeDirs);
cluster.shutdown();
UpgradeUtilities.createEmptyDirs(nameNodeDirs);
log("Normal DataNode upgrade", numDirs);
UpgradeUtilities.createStorageDirs(NAME_NODE, nameNodeDirs, "current");
cluster = new MiniDFSCluster(conf, 0, StartupOption.UPGRADE);
UpgradeUtilities.createStorageDirs(DATA_NODE, dataNodeDirs, "current");
cluster.startDataNodes(conf, 1, false, StartupOption.REGULAR, null);
checkResult(DATA_NODE, dataNodeDirs);
cluster.shutdown();
UpgradeUtilities.createEmptyDirs(nameNodeDirs);
UpgradeUtilities.createEmptyDirs(dataNodeDirs);
log("NameNode upgrade with existing previous dir", numDirs);
UpgradeUtilities.createStorageDirs(NAME_NODE, nameNodeDirs, "current");
UpgradeUtilities.createStorageDirs(NAME_NODE, nameNodeDirs, "previous");
startNameNodeShouldFail(StartupOption.UPGRADE);
UpgradeUtilities.createEmptyDirs(nameNodeDirs);
log("DataNode upgrade with existing previous dir", numDirs);
UpgradeUtilities.createStorageDirs(NAME_NODE, nameNodeDirs, "current");
cluster = new MiniDFSCluster(conf, 0, StartupOption.UPGRADE);
UpgradeUtilities.createStorageDirs(DATA_NODE, dataNodeDirs, "current");
UpgradeUtilities.createStorageDirs(DATA_NODE, dataNodeDirs, "previous");
startDataNodeShouldFail(StartupOption.REGULAR);
cluster.shutdown();
UpgradeUtilities.createEmptyDirs(nameNodeDirs);
UpgradeUtilities.createEmptyDirs(dataNodeDirs);
log("DataNode upgrade with future stored layout version in current", numDirs);
UpgradeUtilities.createStorageDirs(NAME_NODE, nameNodeDirs, "current");
cluster = new MiniDFSCluster(conf, 0, StartupOption.UPGRADE);
baseDirs = UpgradeUtilities.createStorageDirs(DATA_NODE, dataNodeDirs, "current");
UpgradeUtilities.createVersionFile(
DATA_NODE,
baseDirs,
new StorageInfo(
Integer.MIN_VALUE,
UpgradeUtilities.getCurrentNamespaceID(cluster),
UpgradeUtilities.getCurrentFsscTime(cluster)),
cluster.getNameNode().getNamespaceID());
startDataNodeShouldFail(StartupOption.REGULAR);
cluster.shutdown();
UpgradeUtilities.createEmptyDirs(nameNodeDirs);
UpgradeUtilities.createEmptyDirs(dataNodeDirs);
log("DataNode upgrade with newer fsscTime in current", numDirs);
UpgradeUtilities.createStorageDirs(NAME_NODE, nameNodeDirs, "current");
cluster = new MiniDFSCluster(conf, 0, StartupOption.UPGRADE);
baseDirs = UpgradeUtilities.createStorageDirs(DATA_NODE, dataNodeDirs, "current");
UpgradeUtilities.createVersionFile(
DATA_NODE,
baseDirs,
new StorageInfo(
UpgradeUtilities.getCurrentLayoutVersion(),
UpgradeUtilities.getCurrentNamespaceID(cluster),
Long.MAX_VALUE),
cluster.getNameNode().getNamespaceID());
startDataNodeShouldFail(StartupOption.REGULAR);
cluster.shutdown();
UpgradeUtilities.createEmptyDirs(nameNodeDirs);
UpgradeUtilities.createEmptyDirs(dataNodeDirs);
log("NameNode upgrade with no edits file", numDirs);
baseDirs = UpgradeUtilities.createStorageDirs(NAME_NODE, nameNodeDirs, "current");
for (File f : baseDirs) {
FileUtil.fullyDelete(new File(f, "edits"));
}
startNameNodeShouldFail(StartupOption.UPGRADE);
UpgradeUtilities.createEmptyDirs(nameNodeDirs);
log("NameNode upgrade with no image file", numDirs);
baseDirs = UpgradeUtilities.createStorageDirs(NAME_NODE, nameNodeDirs, "current");
for (File f : baseDirs) {
FileUtil.fullyDelete(new File(f, "fsimage"));
}
startNameNodeShouldFail(StartupOption.UPGRADE);
UpgradeUtilities.createEmptyDirs(nameNodeDirs);
log("NameNode upgrade with corrupt version file", numDirs);
baseDirs = UpgradeUtilities.createStorageDirs(NAME_NODE, nameNodeDirs, "current");
for (File f : baseDirs) {
UpgradeUtilities.corruptFile(new File(f, "VERSION"));
}
startNameNodeShouldFail(StartupOption.UPGRADE);
UpgradeUtilities.createEmptyDirs(nameNodeDirs);
log("NameNode upgrade with old layout version in current", numDirs);
baseDirs = UpgradeUtilities.createStorageDirs(NAME_NODE, nameNodeDirs, "current");
UpgradeUtilities.createVersionFile(
NAME_NODE,
baseDirs,
new StorageInfo(
Storage.LAST_UPGRADABLE_LAYOUT_VERSION + 1,
UpgradeUtilities.getCurrentNamespaceID(null),
UpgradeUtilities.getCurrentFsscTime(null)),
0);
startNameNodeShouldFail(StartupOption.UPGRADE);
UpgradeUtilities.createEmptyDirs(nameNodeDirs);
log("NameNode upgrade with future layout version in current", numDirs);
baseDirs = UpgradeUtilities.createStorageDirs(NAME_NODE, nameNodeDirs, "current");
UpgradeUtilities.createVersionFile(
NAME_NODE,
baseDirs,
new StorageInfo(
Integer.MIN_VALUE,
UpgradeUtilities.getCurrentNamespaceID(null),
UpgradeUtilities.getCurrentFsscTime(null)),
0);
startNameNodeShouldFail(StartupOption.UPGRADE);
UpgradeUtilities.createEmptyDirs(nameNodeDirs);
log("Normal Datanode upgrade after datanode format", numDirs);
UpgradeUtilities.createEmptyDirs(dataNodeDirs);
baseDirs = UpgradeUtilities.createStorageDirs(NAME_NODE, nameNodeDirs, "current");
UpgradeUtilities.createVersionFile(
NAME_NODE,
baseDirs,
new StorageInfo(
FSConstants.LAYOUT_VERSION, UpgradeUtilities.getCurrentNamespaceID(null), 0),
0);
cluster =
new MiniDFSCluster(
0,
conf,
1,
false,
false,
false,
StartupOption.UPGRADE,
null,
null,
null,
false,
false,
1,
false);
cluster.shutdown();
UpgradeUtilities.createEmptyDirs(nameNodeDirs);
} // end numDir loop
}
public void testFederationClusterUpgradeAfterFederationVersionWithTopLevelLayout()
throws Exception {
File[] baseDirs;
Configuration baseConf = new Configuration();
UpgradeUtilities.initialize(2, baseConf, true);
for (int numDirs = 1; numDirs <= 2; numDirs++) {
conf = new Configuration();
conf.setInt("dfs.datanode.scan.period.hours", -1);
conf = UpgradeUtilities.initializeStorageStateConf(numDirs, conf);
String[] nameNodeDirs = conf.getStrings("dfs.name.dir");
String[] dataNodeDirs = conf.getStrings("dfs.data.dir");
log(
"DataNode upgrade with federation layout version in current and no ns level layout version",
numDirs);
UpgradeUtilities.createFederatedNameNodeStorageDirs(nameNodeDirs);
conf.set(
FSConstants.DFS_FEDERATION_NAMESERVICES,
baseConf.get(FSConstants.DFS_FEDERATION_NAMESERVICES));
try {
cluster = new MiniDFSCluster(conf, 0, StartupOption.UPGRADE, false, 2);
baseDirs = UpgradeUtilities.createStorageDirs(DATA_NODE, dataNodeDirs, "current");
for (int i = 0; i < 2; i++) {
UpgradeUtilities.createVersionFile(
DATA_NODE,
baseDirs,
new StorageInfo(
FSConstants.FEDERATION_VERSION,
cluster.getNameNode(i).getNamespaceID(),
cluster.getNameNode(i).versionRequest().getCTime()),
cluster.getNameNode(i).getNamespaceID(),
false);
}
cluster.startDataNodes(conf, 1, false, StartupOption.REGULAR, null);
for (int i = 0; i < 2; i++) {
checkResult(DATA_NODE, dataNodeDirs, i, false);
}
// Finalize upgrade.
for (int i = 0; i < 2; i++) {
cluster.getNameNode(i).finalizeUpgrade();
}
cluster.restartDataNodes();
// Wait for datanodes to finalize.
Thread.sleep(10000);
for (int nnIndex = 0; nnIndex < 2; nnIndex++) {
for (int i = 0; i < dataNodeDirs.length; i++) {
File nsBaseDir =
NameSpaceSliceStorage.getNsRoot(
cluster.getNameNode(nnIndex).getNamespaceID(),
new File(dataNodeDirs[i], "current"));
assertFalse(new File(nsBaseDir, "previous").exists());
}
}
} finally {
if (cluster != null) cluster.shutdown();
UpgradeUtilities.createEmptyDirs(nameNodeDirs);
UpgradeUtilities.createEmptyDirs(dataNodeDirs);
}
}
}
/**
* Return the File System State Creation Timestamp (FSSCTime) inherent in the currently running
* Namenode. If no Namenode is running, return the FSSCTime of the master Namenode storage
* directory.
*
* <p>The UpgradeUtilities.initialize() method must be called once before calling this method.
*/
public static long getCurrentFsscTime(MiniDFSCluster cluster) throws IOException {
if (cluster != null) {
return cluster.getNameNode().versionRequest().getCTime();
}
return namenodeStorageFsscTime;
}
/**
* Return the namespace ID inherent in the currently running Namenode. If no Namenode is running,
* return the namespace ID of the master Namenode storage directory.
*
* <p>The UpgradeUtilities.initialize() method must be called once before calling this method.
*/
public static int getCurrentNamespaceID(MiniDFSCluster cluster) throws IOException {
if (cluster != null) {
return cluster.getNameNode().versionRequest().getNamespaceID();
}
return namenodeStorageNamespaceID;
}
