YARN-4087. Followup fixes after YARN-2019 regarding RM behavior when state-store error occurs. Contributed by Jian He

(cherry picked from commit 9f7fcb54e798cf4fda1ea7972dd96491976e1857)
This commit is contained in:
Xuan 2015-09-07 18:09:35 -07:00 committed by Sangjin Lee
parent d27f09c936
commit 1828ba00be
4 changed files with 17 additions and 8 deletions

View File

@ -21,6 +21,9 @@ Release 2.6.2 - UNRELEASED
BUG FIXES BUG FIXES
YARN-4087. Followup fixes after YARN-2019 regarding RM behavior when
state-store error occurs. (Jian He via xgong)
Release 2.6.1 - 2015-09-23 Release 2.6.1 - 2015-09-23
INCOMPATIBLE CHANGES INCOMPATIBLE CHANGES

View File

@ -381,7 +381,7 @@ public class YarnConfiguration extends Configuration {
public static final boolean DEFAULT_RM_RECOVERY_ENABLED = false; public static final boolean DEFAULT_RM_RECOVERY_ENABLED = false;
public static final String YARN_FAIL_FAST = YARN_PREFIX + "fail-fast"; public static final String YARN_FAIL_FAST = YARN_PREFIX + "fail-fast";
public static final boolean DEFAULT_YARN_FAIL_FAST = true; public static final boolean DEFAULT_YARN_FAIL_FAST = false;
public static final String RM_FAIL_FAST = RM_PREFIX + "fail-fast"; public static final String RM_FAIL_FAST = RM_PREFIX + "fail-fast";

View File

@ -286,9 +286,12 @@
<property> <property>
<description>Should YARN fail fast if it encounters any errors. <description>Should YARN fail fast if it encounters any errors.
This is a global config for all other components including RM,NM etc.
If no value is set for component-specific config (e.g yarn.resourcemanager.fail-fast),
this value will be the default.
</description> </description>
<name>yarn.fail-fast</name> <name>yarn.fail-fast</name>
<value>true</value> <value>false</value>
</property> </property>
<property> <property>

View File

@ -42,6 +42,7 @@ import org.apache.hadoop.yarn.api.records.Container;
import org.apache.hadoop.yarn.api.records.ContainerExitStatus; import org.apache.hadoop.yarn.api.records.ContainerExitStatus;
import org.apache.hadoop.yarn.api.records.FinalApplicationStatus; import org.apache.hadoop.yarn.api.records.FinalApplicationStatus;
import org.apache.hadoop.yarn.api.records.impl.pb.ApplicationSubmissionContextPBImpl; import org.apache.hadoop.yarn.api.records.impl.pb.ApplicationSubmissionContextPBImpl;
import org.apache.hadoop.yarn.conf.HAUtil;
import org.apache.hadoop.yarn.conf.YarnConfiguration; import org.apache.hadoop.yarn.conf.YarnConfiguration;
import org.apache.hadoop.yarn.event.AsyncDispatcher; import org.apache.hadoop.yarn.event.AsyncDispatcher;
import org.apache.hadoop.yarn.event.Dispatcher; import org.apache.hadoop.yarn.event.Dispatcher;
@ -822,17 +823,19 @@ public abstract class RMStateStore extends AbstractService {
*/ */
protected void notifyStoreOperationFailed(Exception failureCause) { protected void notifyStoreOperationFailed(Exception failureCause) {
LOG.error("State store operation failed ", failureCause); LOG.error("State store operation failed ", failureCause);
if (failureCause instanceof StoreFencedException) { if (HAUtil.isHAEnabled(getConfig())) {
LOG.warn("State-store fenced ! Transitioning RM to standby");
Thread standByTransitionThread = Thread standByTransitionThread =
new Thread(new StandByTransitionThread()); new Thread(new StandByTransitionThread());
standByTransitionThread.setName("StandByTransitionThread Handler"); standByTransitionThread.setName("StandByTransitionThread Handler");
standByTransitionThread.start(); standByTransitionThread.start();
} else { } else if (YarnConfiguration.shouldRMFailFast(getConfig())) {
if (YarnConfiguration.shouldRMFailFast(getConfig())) { LOG.fatal("Fail RM now due to state-store error!");
rmDispatcher.getEventHandler().handle( rmDispatcher.getEventHandler().handle(
new RMFatalEvent(RMFatalEventType.STATE_STORE_OP_FAILED, new RMFatalEvent(RMFatalEventType.STATE_STORE_OP_FAILED,
failureCause)); failureCause));
} } else {
LOG.warn("Skip the state-store error.");
} }
} }