YARN-4101. RM should print alert messages if Zookeeper and Resourcemanager gets connection issue. Contributed by Xuan Gong

hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/ha/ActiveStandbyElector.java

@@ -1141,4 +1141,8 @@ public class ActiveStandbyElector implements StatCallback, StringCallback {
       ((appData == null) ? "null" : StringUtils.byteToHexString(appData)) + 
       " cb=" + appClient;
   }
+
+  public String getHAZookeeperConnectionState() {
+    return this.zkConnectionState.name();
+  }
 }

hadoop-yarn-project/CHANGES.txt

@@ -871,6 +871,9 @@ Release 2.7.2 - UNRELEASED
     YARN-3893. Both RM in active state when Admin#transitionToActive failure 
     from refeshAll() (Bibin A Chundatt via rohithsharmaks)
 
+    YARN-4101. RM should print alert messages if Zookeeper and Resourcemanager
+    gets connection issue. (Xuan Gong via jianhe)
+
 Release 2.7.1 - 2015-07-06
 
   INCOMPATIBLE CHANGES

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/AdminService.java

@@ -782,4 +782,13 @@ public class AdminService extends CompositeService implements
       throw logAndWrapException(e, user.getShortUserName(), argName, msg);
     }
   }
+
+  public String getHAZookeeperConnectionState() {
+    if (!rmContext.isHAEnabled()) {
+      return "ResourceManager HA is not enabled.";
+    } else if (!autoFailoverEnabled) {
+      return "Auto Failover is not enabled.";
+    }
+    return this.embeddedElector.getHAZookeeperConnectionState();
+  }
 }

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/EmbeddedElectorService.java

@@ -205,4 +205,8 @@ public class EmbeddedElectorService extends AbstractService
     elector.quitElection(false);
     elector.joinElection(localActiveNodeInfo);
   }
+
+  public String getHAZookeeperConnectionState() {
+    return elector.getHAZookeeperConnectionState();
+  }
 }

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/webapp/AboutBlock.java

@@ -44,6 +44,8 @@ public class AboutBlock extends HtmlBlock {
       _("Cluster ID:", cinfo.getClusterId()).
       _("ResourceManager state:", cinfo.getState()).
       _("ResourceManager HA state:", cinfo.getHAState()).
+      _("ResourceManager HA zookeeper connection state:",
+          cinfo.getHAZookeeperConnectionState()).
       _("ResourceManager RMStateStore:", cinfo.getRMStateStore()).
       _("ResourceManager started on:", Times.format(cinfo.getStartedOn())).
       _("ResourceManager version:", cinfo.getRMBuildVersion() +

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/webapp/RMWebApp.java

@@ -117,4 +117,9 @@ public class RMWebApp extends WebApp implements YarnWebParams {
     }
     return path;
   }
+
+  public String getHAZookeeperConnectionState() {
+    return rm.getRMContext().getRMAdminService()
+      .getHAZookeeperConnectionState();
+  }
 }

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/webapp/RMWebAppFilter.java

@@ -58,6 +58,7 @@ public class RMWebAppFilter extends GuiceContainer {
   private String path;
   private static final int BASIC_SLEEP_TIME = 5;
   private static final int MAX_SLEEP_TIME = 5 * 60;
+  private static final Random randnum = new Random();
 
   @Inject
   public RMWebAppFilter(Injector injector, Configuration conf) {
@@ -126,6 +127,8 @@ public class RMWebAppFilter extends GuiceContainer {
         String redirectMsg =
             doRetry ? "Can not find any active RM. Will retry in next " + next
                 + " seconds." : "There is no active RM right now.";
+        redirectMsg += "\nHA Zookeeper Connection State: "
+            + rmWebApp.getHAZookeeperConnectionState();
         PrintWriter out = response.getWriter();
         out.println(redirectMsg);
         if (doRetry) {
@@ -172,6 +175,6 @@ public class RMWebAppFilter extends GuiceContainer {
 
   private static int calculateExponentialTime(int retries) {
     long baseTime = BASIC_SLEEP_TIME * (1L << retries);
-    return (int) (baseTime * ((new Random()).nextDouble() + 0.5));
+    return (int) (baseTime * (randnum.nextDouble() + 0.5));
   }
 }

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/webapp/dao/ClusterInfo.java

@@ -43,6 +43,7 @@ public class ClusterInfo {
   protected String hadoopVersion;
   protected String hadoopBuildVersion;
   protected String hadoopVersionBuiltOn;
+  protected String haZooKeeperConnectionState;
 
   public ClusterInfo() {
   } // JAXB needs this
@@ -62,6 +63,8 @@ public class ClusterInfo {
     this.hadoopVersion = VersionInfo.getVersion();
     this.hadoopBuildVersion = VersionInfo.getBuildVersion();
     this.hadoopVersionBuiltOn = VersionInfo.getDate();
+    this.haZooKeeperConnectionState =
+        rm.getRMContext().getRMAdminService().getHAZookeeperConnectionState();
   }
 
   public String getState() {
@@ -108,4 +111,7 @@ public class ClusterInfo {
     return this.startedOn;
   }
 
+  public String getHAZookeeperConnectionState() {
+    return this.haZooKeeperConnectionState;
+  }
 }

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/webapp/TestRMWebServices.java

@@ -285,6 +285,8 @@ public class TestRMWebServices extends JerseyTestBase {
           WebServicesTestUtils.getXmlLong(element, "startedOn"),
           WebServicesTestUtils.getXmlString(element, "state"),
           WebServicesTestUtils.getXmlString(element, "haState"),
+          WebServicesTestUtils.getXmlString(
+              element, "haZooKeeperConnectionState"),
           WebServicesTestUtils.getXmlString(element, "hadoopVersionBuiltOn"),
           WebServicesTestUtils.getXmlString(element, "hadoopBuildVersion"),
           WebServicesTestUtils.getXmlString(element, "hadoopVersion"),
@@ -300,9 +302,10 @@ public class TestRMWebServices extends JerseyTestBase {
       Exception {
     assertEquals("incorrect number of elements", 1, json.length());
     JSONObject info = json.getJSONObject("clusterInfo");
-    assertEquals("incorrect number of elements", 11, info.length());
+    assertEquals("incorrect number of elements", 12, info.length());
     verifyClusterGeneric(info.getLong("id"), info.getLong("startedOn"),
         info.getString("state"), info.getString("haState"),
+        info.getString("haZooKeeperConnectionState"),
         info.getString("hadoopVersionBuiltOn"),
         info.getString("hadoopBuildVersion"), info.getString("hadoopVersion"),
         info.getString("resourceManagerVersionBuiltOn"),
@@ -312,7 +315,8 @@ public class TestRMWebServices extends JerseyTestBase {
   }
 
   public void verifyClusterGeneric(long clusterid, long startedon,
-      String state, String haState, String hadoopVersionBuiltOn,
+      String state, String haState, String haZooKeeperConnectionState,
+      String hadoopVersionBuiltOn,
       String hadoopBuildVersion, String hadoopVersion,
       String resourceManagerVersionBuiltOn, String resourceManagerBuildVersion,
       String resourceManagerVersion) {