YARN-3742. YARN RM will shut down if ZKClient creation times out. (Daniel Templeton via kasha)

This commit is contained in:
Karthik Kambatla 2017-05-09 14:44:16 -07:00 committed by Xiaoyu Yao
parent d6d1e2438b
commit a819433b3f
9 changed files with 127 additions and 31 deletions

View File

@ -54,6 +54,8 @@ import org.apache.hadoop.yarn.server.resourcemanager.HATestUtil;
import org.apache.hadoop.yarn.server.resourcemanager.ResourceManager;
import org.apache.hadoop.yarn.server.resourcemanager.RMCriticalThreadUncaughtExceptionHandler;
import org.apache.hadoop.yarn.server.resourcemanager.MockRM;
import org.apache.hadoop.yarn.server.resourcemanager.RMFatalEvent;
import org.apache.hadoop.yarn.server.resourcemanager.RMFatalEventType;
import org.apache.hadoop.yarn.server.webproxy.WebAppProxyServer;
import org.apache.hadoop.yarn.webapp.YarnWebParams;
import org.junit.After;
@ -200,7 +202,8 @@ public class TestRMFailover extends ClientBaseWithFixes {
// so it transitions to standby.
ResourceManager rm = cluster.getResourceManager(
cluster.getActiveRMIndex());
rm.handleTransitionToStandByInNewThread();
rm.getRMContext().getDispatcher().getEventHandler().handle(
new RMFatalEvent(RMFatalEventType.STATE_STORE_FENCED, "test"));
verifyRMTransitionToStandby(rm);
verifyConnections();
}

View File

@ -108,8 +108,9 @@ public class ActiveStandbyElectorBasedElectorService extends AbstractService
elector.ensureParentZNode();
if (!isParentZnodeSafe(clusterId)) {
notifyFatalError(electionZNode + " znode has invalid data! "+
"Might need formatting!");
notifyFatalError(String.format("invalid data in znode, %s, " +
"which may require the state store to be reformatted",
electionZNode));
}
super.serviceInit(conf);

View File

@ -304,13 +304,12 @@ public class AdminService extends CompositeService implements
// call all refresh*s for active RM to get the updated configurations.
refreshAll();
} catch (Exception e) {
LOG.error("RefreshAll failed so firing fatal event", e);
rmContext
.getDispatcher()
.getEventHandler()
.handle(
new RMFatalEvent(RMFatalEventType.TRANSITION_TO_ACTIVE_FAILED,
e));
e, "failure to refresh configuration settings"));
throw new ServiceFailedException(
"Error on refreshAll during transition to Active", e);
}

View File

@ -23,7 +23,7 @@ import java.lang.Thread.UncaughtExceptionHandler;
import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;
import org.apache.hadoop.classification.InterfaceAudience.Private;
import org.apache.hadoop.yarn.conf.HAUtil;
import org.apache.hadoop.yarn.exceptions.YarnException;
/**
* This class either shuts down {@link ResourceManager} or transitions the
@ -45,14 +45,19 @@ public class RMCriticalThreadUncaughtExceptionHandler
@Override
public void uncaughtException(Thread t, Throwable e) {
LOG.fatal("Critical thread " + t.getName() + " crashed!", e);
Exception ex;
if (HAUtil.isHAEnabled(rmContext.getYarnConfiguration())) {
rmContext.getResourceManager().handleTransitionToStandByInNewThread();
if (e instanceof Exception) {
ex = (Exception)e;
} else {
rmContext.getDispatcher().getEventHandler().handle(
new RMFatalEvent(RMFatalEventType.CRITICAL_THREAD_CRASH,
new Exception(e)));
ex = new YarnException(e);
}
RMFatalEvent event =
new RMFatalEvent(RMFatalEventType.CRITICAL_THREAD_CRASH, ex,
String.format("a critical thread, %s, that exited unexpectedly",
t.getName()));
rmContext.getDispatcher().getEventHandler().handle(event);
}
}

View File

@ -20,18 +20,73 @@ package org.apache.hadoop.yarn.server.resourcemanager;
import org.apache.hadoop.util.StringUtils;
import org.apache.hadoop.yarn.event.AbstractEvent;
/**
* Event that indicates a non-recoverable error for the resource manager.
*/
public class RMFatalEvent extends AbstractEvent<RMFatalEventType> {
private String cause;
private final Exception cause;
private final String message;
public RMFatalEvent(RMFatalEventType rmFatalEventType, String cause) {
/**
* Create a new event of the given type with the given cause.
* @param rmFatalEventType The {@link RMFatalEventType} of the event
* @param message a text description of the reason for the event
*/
public RMFatalEvent(RMFatalEventType rmFatalEventType, String message) {
this(rmFatalEventType, null, message);
}
/**
* Create a new event of the given type around the given source
* {@link Exception}.
* @param rmFatalEventType The {@link RMFatalEventType} of the event
* @param cause the source exception
*/
public RMFatalEvent(RMFatalEventType rmFatalEventType, Exception cause) {
this(rmFatalEventType, cause, null);
}
/**
* Create a new event of the given type around the given source
* {@link Exception} with the given cause.
* @param rmFatalEventType The {@link RMFatalEventType} of the event
* @param cause the source exception
* @param message a text description of the reason for the event
*/
public RMFatalEvent(RMFatalEventType rmFatalEventType, Exception cause,
String message) {
super(rmFatalEventType);
this.cause = cause;
this.message = message;
}
public RMFatalEvent(RMFatalEventType rmFatalEventType, Exception cause) {
super(rmFatalEventType);
this.cause = StringUtils.stringifyException(cause);
/**
* Get a text description of the reason for the event. If a cause was, that
* {@link Exception} will be converted to a {@link String} and included in
* the result.
* @return a text description of the reason for the event
*/
public String getExplanation() {
StringBuilder sb = new StringBuilder();
if (message != null) {
sb.append(message);
if (cause != null) {
sb.append(": ");
}
}
if (cause != null) {
sb.append(StringUtils.stringifyException(cause));
}
return sb.toString();
}
public String getCause() {return this.cause;}
@Override
public String toString() {
return String.format("RMFatalEvent of type %s, caused by %s",
getType().name(), getExplanation());
}
}

View File

@ -23,6 +23,7 @@ import org.apache.hadoop.classification.InterfaceAudience;
@InterfaceAudience.Private
public enum RMFatalEventType {
// Source <- Store
STATE_STORE_FENCED,
STATE_STORE_OP_FAILED,
// Source <- Embedded Elector

View File

@ -811,15 +811,45 @@ public class ResourceManager extends CompositeService implements Recoverable {
}
@Private
public static class RMFatalEventDispatcher
implements EventHandler<RMFatalEvent> {
private class RMFatalEventDispatcher implements EventHandler<RMFatalEvent> {
@Override
public void handle(RMFatalEvent event) {
LOG.fatal("Received a " + RMFatalEvent.class.getName() + " of type " +
event.getType().name() + ". Cause:\n" + event.getCause());
LOG.error("Received " + event);
ExitUtil.terminate(1, event.getCause());
if (HAUtil.isHAEnabled(getConfig())) {
// If we're in an HA config, the right answer is always to go into
// standby.
LOG.warn("Transitioning the resource manager to standby.");
handleTransitionToStandByInNewThread();
} else {
// If we're stand-alone, we probably want to shut down, but the if and
// how depends on the event.
switch(event.getType()) {
case STATE_STORE_FENCED:
LOG.fatal("State store fenced even though the resource manager " +
"is not configured for high availability. Shutting down this " +
"resource manager to protect the integrity of the state store.");
ExitUtil.terminate(1, event.getExplanation());
break;
case STATE_STORE_OP_FAILED:
if (YarnConfiguration.shouldRMFailFast(getConfig())) {
LOG.fatal("Shutting down the resource manager because a state " +
"store operation failed, and the resource manager is " +
"configured to fail fast. See the yarn.fail-fast and " +
"yarn.resourcemanager.fail-fast properties.");
ExitUtil.terminate(1, event.getExplanation());
} else {
LOG.warn("Ignoring state store operation failure because the " +
"resource manager is not configured to fail fast. See the " +
"yarn.fail-fast and yarn.resourcemanager.fail-fast " +
"properties.");
}
break;
default:
LOG.fatal("Shutting down the resource manager.");
ExitUtil.terminate(1, event.getExplanation());
}
}
}
}
@ -827,7 +857,7 @@ public class ResourceManager extends CompositeService implements Recoverable {
* Transition to standby state in a new thread. The transition operation is
* asynchronous to avoid deadlock caused by cyclic dependency.
*/
public void handleTransitionToStandByInNewThread() {
private void handleTransitionToStandByInNewThread() {
Thread standByTransitionThread =
new Thread(activeServices.standByTransitionRunnable);
standByTransitionThread.setName("StandByTransitionThread");

View File

@ -1129,18 +1129,18 @@ public abstract class RMStateStore extends AbstractService {
Exception failureCause) {
boolean isFenced = false;
LOG.error("State store operation failed ", failureCause);
if (HAUtil.isHAEnabled(getConfig())) {
LOG.warn("State-store fenced ! Transitioning RM to standby");
rmDispatcher.getEventHandler().handle(
new RMFatalEvent(RMFatalEventType.STATE_STORE_FENCED,
failureCause));
isFenced = true;
resourceManager.handleTransitionToStandByInNewThread();
} else if (YarnConfiguration.shouldRMFailFast(getConfig())) {
LOG.fatal("Fail RM now due to state-store error!");
} else {
rmDispatcher.getEventHandler().handle(
new RMFatalEvent(RMFatalEventType.STATE_STORE_OP_FAILED,
failureCause));
} else {
LOG.warn("Skip the state-store error.");
}
return isFenced;
}

View File

@ -43,6 +43,7 @@ public class TestMemoryRMStateStore {
store.init(conf);
ResourceManager mockRM = mock(ResourceManager.class);
store.setResourceManager(mockRM);
store.setRMDispatcher(new RMStateStoreTestBase.TestDispatcher());
RMDelegationTokenIdentifier mockTokenId =
mock(RMDelegationTokenIdentifier.class);
store.removeRMDelegationToken(mockTokenId);
@ -58,6 +59,7 @@ public class TestMemoryRMStateStore {
};
store.init(conf);
store.setResourceManager(mockRM);
store.setRMDispatcher(new RMStateStoreTestBase.TestDispatcher());
store.removeRMDelegationToken(mockTokenId);
assertTrue("RMStateStore should have been in fenced state",
store.isFencedState());