YARN-5601. Make the RM epoch base value configurable. Contributed by Subru Krishnan

(cherry picked from commit 9ca2aba9cc65090162b3517b194b5e655ee4a157)
This commit is contained in:
Jian He 2016-09-02 12:23:57 +08:00 committed by Carlo Curino
parent 1882bc10fb
commit 2797507d51
13 changed files with 32 additions and 7 deletions

View File

@ -293,7 +293,10 @@
</Match>
<Match>
<Class name="org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore" />
<Field name="resourceManager"/>
<Or>
<Field name="resourceManager"/>
<Field name="baseEpoch"/>
</Or>
<Bug pattern="IS2_INCONSISTENT_SYNC" />
</Match>
<Match>

View File

@ -143,6 +143,9 @@ private static void addDeprecatedKeys() {
public static final String RM_HOSTNAME = RM_PREFIX + "hostname";
public static final String RM_EPOCH = RM_PREFIX + "epoch";
public static final long DEFAULT_RM_EPOCH = 0L;
/** The address of the applications manager interface in the RM.*/
public static final String RM_ADDRESS =
RM_PREFIX + "address";

View File

@ -75,6 +75,8 @@ public void initializeMemberVariables() {
.add(YarnConfiguration.FEDERATION_FAILOVER_ENABLED);
configurationPropsToSkipCompare
.add(YarnConfiguration.FEDERATION_STATESTORE_HEARTBEAT_INTERVAL_SECS);
configurationPropsToSkipCompare
.add(YarnConfiguration.RM_EPOCH);
// Ignore blacklisting nodes for AM failures feature since it is still a
// "work in progress"

View File

@ -772,6 +772,13 @@ protected void serviceStart() throws Exception {
LOG.error("Failed to load/recover state", e);
throw e;
}
} else {
if (HAUtil.isFederationEnabled(conf)) {
long epoch = conf.getLong(YarnConfiguration.RM_EPOCH,
YarnConfiguration.DEFAULT_RM_EPOCH);
rmContext.setEpoch(epoch);
LOG.info("Epoch set for Federation: " + epoch);
}
}
super.serviceStart();

View File

@ -197,7 +197,7 @@ protected synchronized void storeVersion() throws Exception {
@Override
public synchronized long getAndIncrementEpoch() throws Exception {
Path epochNodePath = getNodePath(rootDirPath, EPOCH_NODE);
long currentEpoch = 0;
long currentEpoch = baseEpoch;
FileStatus status = getFileStatusWithRetries(epochNodePath);
if (status != null) {
// load current epoch

View File

@ -255,7 +255,7 @@ protected Version getCurrentVersion() {
@Override
public synchronized long getAndIncrementEpoch() throws Exception {
long currentEpoch = 0;
long currentEpoch = baseEpoch;
byte[] dbKeyBytes = bytes(EPOCH_NODE);
try {
byte[] data = db.get(dbKeyBytes);

View File

@ -83,6 +83,7 @@ public synchronized RMState loadState() throws Exception {
@Override
public synchronized void initInternal(Configuration conf) {
epoch = baseEpoch;
}
@Override

View File

@ -98,6 +98,7 @@ public abstract class RMStateStore extends AbstractService {
"ReservationSystemRoot";
protected static final String VERSION_NODE = "RMVersionNode";
protected static final String EPOCH_NODE = "EpochNode";
protected long baseEpoch;
protected ResourceManager resourceManager;
private final ReadLock readLock;
private final WriteLock writeLock;
@ -690,6 +691,9 @@ protected void serviceInit(Configuration conf) throws Exception{
dispatcher.register(RMStateStoreEventType.class,
rmStateStoreEventHandler);
dispatcher.setDrainEventsOnStop();
// read the base epoch value from conf
baseEpoch = conf.getLong(YarnConfiguration.RM_EPOCH,
YarnConfiguration.DEFAULT_RM_EPOCH);
initInternal(conf);
}

View File

@ -438,7 +438,7 @@ protected synchronized Version loadVersion() throws Exception {
@Override
public synchronized long getAndIncrementEpoch() throws Exception {
String epochNodePath = getNodePath(zkRootNodePath, EPOCH_NODE);
long currentEpoch = 0;
long currentEpoch = baseEpoch;
if (exists(epochNodePath)) {
// load current epoch

View File

@ -91,6 +91,8 @@ public class RMStateStoreTestBase {
public static final Log LOG = LogFactory.getLog(RMStateStoreTestBase.class);
protected final long epoch = 10L;
static class TestDispatcher implements Dispatcher, EventHandler<Event> {
ApplicationAttemptId attemptId;
@ -564,13 +566,13 @@ public void testEpoch(RMStateStoreHelper stateStoreHelper)
store.setRMDispatcher(new TestDispatcher());
long firstTimeEpoch = store.getAndIncrementEpoch();
Assert.assertEquals(0, firstTimeEpoch);
Assert.assertEquals(epoch, firstTimeEpoch);
long secondTimeEpoch = store.getAndIncrementEpoch();
Assert.assertEquals(1, secondTimeEpoch);
Assert.assertEquals(epoch + 1, secondTimeEpoch);
long thirdTimeEpoch = store.getAndIncrementEpoch();
Assert.assertEquals(2, thirdTimeEpoch);
Assert.assertEquals(epoch + 2, thirdTimeEpoch);
}
public void testAppDeletion(RMStateStoreHelper stateStoreHelper)

View File

@ -117,6 +117,7 @@ public RMStateStore getRMStateStore() throws Exception {
conf.setInt(YarnConfiguration.FS_RM_STATE_STORE_NUM_RETRIES, 8);
conf.setLong(YarnConfiguration.FS_RM_STATE_STORE_RETRY_INTERVAL_MS,
900L);
conf.setLong(YarnConfiguration.RM_EPOCH, epoch);
if (adminCheckEnable) {
conf.setBoolean(
YarnConfiguration.YARN_INTERMEDIATE_DATA_ENCRYPTION, true);

View File

@ -82,6 +82,7 @@ public void testVersion() throws Exception {
@Test(timeout = 60000)
public void testEpoch() throws Exception {
conf.setLong(YarnConfiguration.RM_EPOCH, epoch);
LeveldbStateStoreTester tester = new LeveldbStateStoreTester();
testEpoch(tester);
}

View File

@ -189,6 +189,7 @@ private RMStateStore createStore(Configuration conf) throws Exception {
conf.set(YarnConfiguration.RM_ZK_ADDRESS,
curatorTestingServer.getConnectString());
conf.set(YarnConfiguration.ZK_RM_STATE_STORE_PARENT_PATH, workingZnode);
conf.setLong(YarnConfiguration.RM_EPOCH, epoch);
this.store = new TestZKRMStateStoreInternal(conf, workingZnode);
return this.store;
}