Apache
/
hadoop
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fixing HDFS state-store. Contributed by Arun Suresh. 

(cherry picked from commit 9a2a9553ee)
This commit is contained in:
Vinod Kumar Vavilapalli 2015-05-14 16:13:51 -07:00
parent 87862970f1
commit ad3196e016
4 changed files with 207 additions and 45 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

75
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/FileSystemRMStateStore.java
View File

@ -25,6 +25,7 @@ import java.io.DataOutputStream;
import java.io.FileNotFoundException;
import java.io.IOException;
import java.util.ArrayList;
import java.util.EnumSet;
import java.util.List;
import org.apache.commons.logging.Log;
@ -38,6 +39,7 @@ import org.apache.hadoop.fs.FileStatus;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.fs.PathFilter;
import org.apache.hadoop.fs.XAttrSetFlag;
import org.apache.hadoop.io.IOUtils;
import org.apache.hadoop.security.token.delegation.DelegationKey;
import org.apache.hadoop.yarn.api.records.ApplicationAttemptId;
@ -84,7 +86,8 @@ public class FileSystemRMStateStore extends RMStateStore {
protected static final String AMRMTOKEN_SECRET_MANAGER_NODE =
"AMRMTokenSecretManagerNode";
@VisibleForTesting
private static final String UNREADABLE_BY_SUPERUSER_XATTRIB =
"security.hdfs.unreadable.by.superuser";
protected FileSystem fs;
@VisibleForTesting
protected Configuration fsConf;
@ -97,6 +100,7 @@ public class FileSystemRMStateStore extends RMStateStore {
private Path dtSequenceNumberPath = null;
private int fsNumRetries;
private long fsRetryInterval;
private boolean isHDFS;
@VisibleForTesting
Path fsWorkingPath;
@ -141,11 +145,17 @@ public class FileSystemRMStateStore extends RMStateStore {
}
fs = fsWorkingPath.getFileSystem(fsConf);
isHDFS = fs.getScheme().toLowerCase().contains("hdfs");
mkdirsWithRetries(rmDTSecretManagerRoot);
mkdirsWithRetries(rmAppRoot);
mkdirsWithRetries(amrmTokenSecretManagerRoot);
}
@VisibleForTesting
void setIsHDFS(boolean isHDFS) {
this.isHDFS = isHDFS;
}
@Override
protected synchronized void closeInternal() throws Exception {
closeWithRetries();
@ -175,9 +185,9 @@ public class FileSystemRMStateStore extends RMStateStore {
byte[] data =
((VersionPBImpl) CURRENT_VERSION_INFO).getProto().toByteArray();
if (existsWithRetries(versionNodePath)) {
updateFile(versionNodePath, data);
updateFile(versionNodePath, data, false);
} else {
writeFileWithRetries(versionNodePath, data);
writeFileWithRetries(versionNodePath, data, false);
}
}
@ -194,12 +204,12 @@ public class FileSystemRMStateStore extends RMStateStore {
// increment epoch and store it
byte[] storeData = Epoch.newInstance(currentEpoch + 1).getProto()
.toByteArray();
updateFile(epochNodePath, storeData);
updateFile(epochNodePath, storeData, false);
} else {
// initialize epoch file with 1 for the next time.
byte[] storeData = Epoch.newInstance(currentEpoch + 1).getProto()
.toByteArray();
writeFileWithRetries(epochNodePath, storeData);
writeFileWithRetries(epochNodePath, storeData, false);
}
return currentEpoch;
}
@ -254,6 +264,8 @@ public class FileSystemRMStateStore extends RMStateStore {
}
byte[] childData = readFileWithRetries(childNodeStatus.getPath(),
childNodeStatus.getLen());
// Set attribute if not already set
setUnreadableBySuperuserXattrib(childNodeStatus.getPath());
if (childNodeName.startsWith(ApplicationId.appIdStrPrefix)) {
// application
if (LOG.isDebugEnabled()) {
@ -394,7 +406,7 @@ public class FileSystemRMStateStore extends RMStateStore {
try {
// currently throw all exceptions. May need to respond differently for HA
// based on whether we have lost the right to write to FS
writeFileWithRetries(nodeCreatePath, appStateData);
writeFileWithRetries(nodeCreatePath, appStateData, true);
} catch (Exception e) {
LOG.info("Error storing info for app: " + appId, e);
throw e;
@ -412,7 +424,7 @@ public class FileSystemRMStateStore extends RMStateStore {
try {
// currently throw all exceptions. May need to respond differently for HA
// based on whether we have lost the right to write to FS
updateFile(nodeCreatePath, appStateData);
updateFile(nodeCreatePath, appStateData, true);
} catch (Exception e) {
LOG.info("Error updating info for app: " + appId, e);
throw e;
@ -433,7 +445,7 @@ public class FileSystemRMStateStore extends RMStateStore {
try {
// currently throw all exceptions. May need to respond differently for HA
// based on whether we have lost the right to write to FS
writeFileWithRetries(nodeCreatePath, attemptStateData);
writeFileWithRetries(nodeCreatePath, attemptStateData, true);
} catch (Exception e) {
LOG.info("Error storing info for attempt: " + appAttemptId, e);
throw e;
@ -454,7 +466,7 @@ public class FileSystemRMStateStore extends RMStateStore {
try {
// currently throw all exceptions. May need to respond differently for HA
// based on whether we have lost the right to write to FS
updateFile(nodeCreatePath, attemptStateData);
updateFile(nodeCreatePath, attemptStateData, true);
} catch (Exception e) {
LOG.info("Error updating info for attempt: " + appAttemptId, e);
throw e;
@ -505,10 +517,10 @@ public class FileSystemRMStateStore extends RMStateStore {
new RMDelegationTokenIdentifierData(identifier, renewDate);
if (isUpdate) {
LOG.info("Updating RMDelegationToken_" + identifier.getSequenceNumber());
updateFile(nodeCreatePath, identifierData.toByteArray());
updateFile(nodeCreatePath, identifierData.toByteArray(), true);
} else {
LOG.info("Storing RMDelegationToken_" + identifier.getSequenceNumber());
writeFileWithRetries(nodeCreatePath, identifierData.toByteArray());
writeFileWithRetries(nodeCreatePath, identifierData.toByteArray(), true);
// store sequence number
Path latestSequenceNumberPath = getNodePath(rmDTSecretManagerRoot,
@ -539,7 +551,7 @@ public class FileSystemRMStateStore extends RMStateStore {
try (DataOutputStream fsOut = new DataOutputStream(os)) {
LOG.info("Storing RMDelegationKey_" + masterKey.getKeyId());
masterKey.write(fsOut);
writeFileWithRetries(nodeCreatePath, os.toByteArray());
writeFileWithRetries(nodeCreatePath, os.toByteArray(), true);
}
}
@ -572,6 +584,16 @@ public class FileSystemRMStateStore extends RMStateStore {
return getNodePath(root, appId.toString());
}
@VisibleForTesting
protected Path getAppDir(ApplicationId appId) {
return getAppDir(rmAppRoot, appId);
}
@VisibleForTesting
protected Path getAppAttemptDir(ApplicationAttemptId appAttId) {
return getNodePath(getAppDir(appAttId.getApplicationId()), appAttId
.toString());
}
// FileSystem related code
private boolean checkAndRemovePartialRecordWithRetries(final Path record)
@ -594,12 +616,13 @@ public class FileSystemRMStateStore extends RMStateStore {
}.runWithRetries();
}
private void writeFileWithRetries(final Path outputPath,final byte[] data)
private void writeFileWithRetries(final Path outputPath, final byte[] data,
final boolean makeUnreadableByAdmin)
throws Exception {
new FSAction<Void>() {
@Override
public Void run() throws Exception {
writeFile(outputPath, data);
writeFile(outputPath, data, makeUnreadableByAdmin);
return null;
}
}.runWithRetries();
@ -746,7 +769,8 @@ public class FileSystemRMStateStore extends RMStateStore {
* data to .tmp file and then rename it. Here we are assuming that rename is
* atomic for underlying file system.
*/
private void writeFile(Path outputPath, byte[] data) throws Exception {
protected void writeFile(Path outputPath, byte[] data, boolean
makeUnradableByAdmin) throws Exception {
Path tempPath =
new Path(outputPath.getParent(), outputPath.getName() + ".tmp");
FSDataOutputStream fsOut = null;
@ -754,6 +778,9 @@ public class FileSystemRMStateStore extends RMStateStore {
// final status.
try {
fsOut = fs.create(tempPath, true);
if (makeUnradableByAdmin) {
setUnreadableBySuperuserXattrib(tempPath);
}
fsOut.write(data);
fsOut.close();
fsOut = null;
@ -768,10 +795,11 @@ public class FileSystemRMStateStore extends RMStateStore {
* data to .new file and then rename it. Here we are assuming that rename is
* atomic for underlying file system.
*/
protected void updateFile(Path outputPath, byte[] data) throws Exception {
protected void updateFile(Path outputPath, byte[] data, boolean
makeUnradableByAdmin) throws Exception {
Path newPath = new Path(outputPath.getParent(), outputPath.getName() + ".new");
// use writeFileWithRetries to make sure .new file is created atomically
writeFileWithRetries(newPath, data);
writeFileWithRetries(newPath, data, makeUnradableByAdmin);
replaceFile(newPath, outputPath);
}
@ -810,9 +838,9 @@ public class FileSystemRMStateStore extends RMStateStore {
AMRMTokenSecretManagerState.newInstance(amrmTokenSecretManagerState);
byte[] stateData = data.getProto().toByteArray();
if (isUpdate) {
updateFile(nodeCreatePath, stateData);
updateFile(nodeCreatePath, stateData, true);
} else {
writeFileWithRetries(nodeCreatePath, stateData);
writeFileWithRetries(nodeCreatePath, stateData, true);
}
}
@ -825,4 +853,13 @@ public class FileSystemRMStateStore extends RMStateStore {
public long getRetryInterval() {
return fsRetryInterval;
}
private void setUnreadableBySuperuserXattrib(Path p)
throws IOException {
if (isHDFS &&
!fs.getXAttrs(p).containsKey(UNREADABLE_BY_SUPERUSER_XATTRIB)) {
fs.setXAttr(p, UNREADABLE_BY_SUPERUSER_XATTRIB, null,
EnumSet.of(XAttrSetFlag.CREATE));
}
}
}

14
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/RMStateStoreTestBase.java
View File

@ -112,6 +112,12 @@ public class RMStateStoreTestBase extends ClientBaseWithFixes{
}
public static class StoreStateVerifier {
void afterStoreApp(RMStateStore store, ApplicationId appId) {}
void afterStoreAppAttempt(RMStateStore store, ApplicationAttemptId
appAttId) {}
}
interface RMStateStoreHelper {
RMStateStore getRMStateStore() throws Exception;
boolean isFinalStateValid() throws Exception;
@ -182,6 +188,12 @@ public class RMStateStoreTestBase extends ClientBaseWithFixes{
void testRMAppStateStore(RMStateStoreHelper stateStoreHelper)
throws Exception {
testRMAppStateStore(stateStoreHelper, new StoreStateVerifier());
}
void testRMAppStateStore(RMStateStoreHelper stateStoreHelper,
StoreStateVerifier verifier)
throws Exception {
long submitTime = System.currentTimeMillis();
long startTime = System.currentTimeMillis() + 1234;
Configuration conf = new YarnConfiguration();
@ -205,6 +217,7 @@ public class RMStateStoreTestBase extends ClientBaseWithFixes{
.toApplicationAttemptId("appattempt_1352994193343_0001_000001");
ApplicationId appId1 = attemptId1.getApplicationId();
storeApp(store, appId1, submitTime, startTime);
verifier.afterStoreApp(store, appId1);
// create application token and client token key for attempt1
Token<AMRMTokenIdentifier> appAttemptToken1 =
@ -236,6 +249,7 @@ public class RMStateStoreTestBase extends ClientBaseWithFixes{
storeApp(store, appIdRemoved, submitTime, startTime);
storeAttempt(store, attemptIdRemoved,
"container_1352994193343_0002_01_000001", null, null, dispatcher);
verifier.afterStoreAppAttempt(store, attemptIdRemoved);
RMApp mockRemovedApp = mock(RMApp.class);
RMAppAttemptMetrics mockRmAppAttemptMetrics =

127
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/TestFSRMStateStore.java
View File

@ -21,8 +21,15 @@ package org.apache.hadoop.yarn.server.resourcemanager.recovery;
import static org.junit.Assert.assertTrue;
import java.io.IOException;
import java.security.PrivilegedExceptionAction;
import java.util.LinkedList;
import java.util.Queue;
import java.util.concurrent.atomic.AtomicBoolean;
import org.apache.hadoop.fs.permission.FsPermission;
import org.apache.hadoop.hdfs.DistributedFileSystem;
import org.apache.hadoop.security.AccessControlException;
import org.apache.hadoop.security.UserGroupInformation;
import org.junit.Assert;
import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;
@ -38,7 +45,6 @@ import org.apache.hadoop.yarn.api.records.ApplicationId;
import org.apache.hadoop.yarn.conf.YarnConfiguration;
import org.apache.hadoop.yarn.server.records.impl.pb.VersionPBImpl;
import org.apache.hadoop.yarn.server.records.Version;
import org.apache.hadoop.yarn.server.resourcemanager.recovery.TestZKRMStateStore.TestZKRMStateStoreTester;
import org.apache.hadoop.yarn.server.resourcemanager.recovery.records.ApplicationStateData;
import org.apache.hadoop.yarn.server.resourcemanager.rmapp.RMApp;
import org.apache.hadoop.yarn.server.resourcemanager.rmapp.RMAppState;
@ -56,6 +62,7 @@ public class TestFSRMStateStore extends RMStateStoreTestBase {
Path workingDirPathURI;
TestFileSystemRMStore store;
MiniDFSCluster cluster;
boolean adminCheckEnable;
class TestFileSystemRMStore extends FileSystemRMStateStore {
@ -83,8 +90,9 @@ public class TestFSRMStateStore extends RMStateStoreTestBase {
}
}
public TestFSRMStateStoreTester(MiniDFSCluster cluster) throws Exception {
Path workingDirPath = new Path("/Test");
public TestFSRMStateStoreTester(MiniDFSCluster cluster, boolean adminCheckEnable) throws Exception {
Path workingDirPath = new Path("/yarn/Test");
this.adminCheckEnable = adminCheckEnable;
this.cluster = cluster;
FileSystem fs = cluster.getFileSystem();
fs.mkdirs(workingDirPath);
@ -111,6 +119,11 @@ public class TestFSRMStateStore extends RMStateStoreTestBase {
store.startInternal();
Assert.assertTrue(store.fs != previousFs);
Assert.assertTrue(store.fs.getConf() == store.fsConf);
if (adminCheckEnable) {
store.setIsHDFS(true);
} else {
store.setIsHDFS(false);
}
return store;
}
@ -123,8 +136,9 @@ public class TestFSRMStateStore extends RMStateStoreTestBase {
@Override
public void writeVersion(Version version) throws Exception {
store.updateFile(store.getVersionNode(), ((VersionPBImpl) version)
.getProto().toByteArray());
store.updateFile(store.getVersionNode(), ((VersionPBImpl)
version)
.getProto().toByteArray(), false);
}
@Override
@ -146,7 +160,7 @@ public class TestFSRMStateStore extends RMStateStoreTestBase {
MiniDFSCluster cluster =
new MiniDFSCluster.Builder(conf).numDataNodes(1).build();
try {
fsTester = new TestFSRMStateStoreTester(cluster);
fsTester = new TestFSRMStateStoreTester(cluster, false);
// If the state store is FileSystemRMStateStore then add corrupted entry.
// It should discard the entry and remove it from file system.
FSDataOutputStream fsOut = null;
@ -178,13 +192,110 @@ public class TestFSRMStateStore extends RMStateStoreTestBase {
}
}
@Test(timeout = 60000)
public void testHDFSRMStateStore() throws Exception {
final HdfsConfiguration conf = new HdfsConfiguration();
UserGroupInformation yarnAdmin =
UserGroupInformation.createUserForTesting("yarn",
new String[]{"admin"});
final MiniDFSCluster cluster =
new MiniDFSCluster.Builder(conf).numDataNodes(1).build();
cluster.getFileSystem().mkdir(new Path("/yarn"),
FsPermission.valueOf("-rwxrwxrwx"));
cluster.getFileSystem().setOwner(new Path("/yarn"), "yarn", "admin");
final UserGroupInformation hdfsAdmin = UserGroupInformation.getCurrentUser();
final StoreStateVerifier verifier = new StoreStateVerifier() {
@Override
void afterStoreApp(final RMStateStore store, final ApplicationId appId) {
try {
// Wait for things to settle
Thread.sleep(5000);
hdfsAdmin.doAs(
new PrivilegedExceptionAction<Void>() {
@Override
public Void run() throws Exception {
verifyFilesUnreadablebyHDFS(cluster,
((FileSystemRMStateStore) store).getAppDir
(appId));
return null;
}
});
} catch (Exception e) {
throw new RuntimeException(e);
}
}
@Override
void afterStoreAppAttempt(final RMStateStore store,
final ApplicationAttemptId appAttId) {
try {
// Wait for things to settle
Thread.sleep(5000);
hdfsAdmin.doAs(
new PrivilegedExceptionAction<Void>() {
@Override
public Void run() throws Exception {
verifyFilesUnreadablebyHDFS(cluster,
((FileSystemRMStateStore) store)
.getAppAttemptDir(appAttId));
return null;
}
});
} catch (Exception e) {
throw new RuntimeException(e);
}
}
};
try {
yarnAdmin.doAs(new PrivilegedExceptionAction<Void>() {
@Override
public Void run() throws Exception {
fsTester = new TestFSRMStateStoreTester(cluster, true);
testRMAppStateStore(fsTester, verifier);
return null;
}
});
} finally {
cluster.shutdown();
}
}
private void verifyFilesUnreadablebyHDFS(MiniDFSCluster cluster,
Path root) throws Exception{
DistributedFileSystem fs = cluster.getFileSystem();
Queue<Path> paths = new LinkedList<>();
paths.add(root);
while (!paths.isEmpty()) {
Path p = paths.poll();
FileStatus stat = fs.getFileStatus(p);
if (!stat.isDirectory()) {
try {
LOG.warn("\n\n ##Testing path [" + p + "]\n\n");
fs.open(p);
Assert.fail("Super user should not be able to read ["+ UserGroupInformation.getCurrentUser() + "] [" + p.getName() + "]");
} catch (AccessControlException e) {
Assert.assertTrue(e.getMessage().contains("superuser is not allowed to perform this operation"));
} catch (Exception e) {
Assert.fail("Should get an AccessControlException here");
}
}
if (stat.isDirectory()) {
FileStatus[] ls = fs.listStatus(p);
for (FileStatus f : ls) {
paths.add(f.getPath());
}
}
}
}
@Test(timeout = 60000)
public void testCheckMajorVersionChange() throws Exception {
HdfsConfiguration conf = new HdfsConfiguration();
MiniDFSCluster cluster =
new MiniDFSCluster.Builder(conf).numDataNodes(1).build();
try {
fsTester = new TestFSRMStateStoreTester(cluster) {
fsTester = new TestFSRMStateStoreTester(cluster, false) {
Version VERSION_INFO = Version.newInstance(Integer.MAX_VALUE, 0);
@Override
@ -268,7 +379,7 @@ public class TestFSRMStateStore extends RMStateStoreTestBase {
new MiniDFSCluster.Builder(conf).numDataNodes(2).build();
cluster.waitActive();
try {
TestFSRMStateStoreTester fsTester = new TestFSRMStateStoreTester(cluster);
TestFSRMStateStoreTester fsTester = new TestFSRMStateStoreTester(cluster, false);
final RMStateStore store = fsTester.getRMStateStore();
store.setRMDispatcher(new TestDispatcher());
final AtomicBoolean assertionFailedInThread = new AtomicBoolean(false);

2
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/ResourceManagerHA.md
View File

@ -53,7 +53,7 @@ When there are multiple RMs, the configuration (yarn-site.xml) used by clients a
### Recovering prevous active-RM's state
With the [ResourceManger Restart](./ResourceManagerRestart.html) enabled, the RM being promoted to an active state loads the RM internal state and continues to operate from where the previous active left off as much as possible depending on the RM restart feature. A new attempt is spawned for each managed application previously submitted to the RM. Applications can checkpoint periodically to avoid losing any work. The state-store must be visible from the both of Active/Standby RMs. Currently, there are two RMStateStore implementations for persistence - FileSystemRMStateStore and ZKRMStateStore. The `ZKRMStateStore` implicitly allows write access to a single RM at any point in time, and hence is the recommended store to use in an HA cluster. When using the ZKRMStateStore, there is no need for a separate fencing mechanism to address a potential split-brain situation where multiple RMs can potentially assume the Active role.
With the [ResourceManger Restart](./ResourceManagerRestart.html) enabled, the RM being promoted to an active state loads the RM internal state and continues to operate from where the previous active left off as much as possible depending on the RM restart feature. A new attempt is spawned for each managed application previously submitted to the RM. Applications can checkpoint periodically to avoid losing any work. The state-store must be visible from the both of Active/Standby RMs. Currently, there are two RMStateStore implementations for persistence - FileSystemRMStateStore and ZKRMStateStore. The `ZKRMStateStore` implicitly allows write access to a single RM at any point in time, and hence is the recommended store to use in an HA cluster. When using the ZKRMStateStore, there is no need for a separate fencing mechanism to address a potential split-brain situation where multiple RMs can potentially assume the Active role. When using the ZKRMStateStore, it is advisable to NOT set the "`zookeeper.DigestAuthenticationProvider.superDigest`" property on the Zookeeper cluster to ensure that the zookeeper admin does not have access to YARN application/user credential information.
Deployment
----------