HDDS-11196. Improve SCM WebUI Display.

apache · Sep 14, 2024 · 3ce5a53 · 3ce5a53
1 parent e573701
commit 3ce5a53
Show file tree

Hide file tree

Showing 7 changed files with 141 additions and 5 deletions.
diff --git a/hadoop-hdds/framework/src/main/resources/webapps/static/ozone.css b/hadoop-hdds/framework/src/main/resources/webapps/static/ozone.css
@@ -91,3 +91,7 @@ body {
 .om-roles-background {
     background-color: #dcfbcd!important;
 }
+
+.scm-roles-background {
+    background-color: #dcfbcd!important;
+}
diff --git a/hadoop-hdds/framework/src/main/resources/webapps/static/templates/jvm.html b/hadoop-hdds/framework/src/main/resources/webapps/static/templates/jvm.html
@@ -21,6 +21,6 @@
     </tr>
     <tr>
         <th>Input arguments:</th>
-        <td>{{$ctrl.jmx.InputArguments}}</td>
+        <td><pre>{{$ctrl.jmx.InputArguments.join('\n')}}</pre></td>
     </tr>
 </table>
diff --git a/hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/node/SCMNodeManager.java b/hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/node/SCMNodeManager.java
@@ -146,6 +146,8 @@ public class SCMNodeManager implements NodeManager {
   private static final String LASTHEARTBEAT = "LASTHEARTBEAT";
   private static final String USEDSPACEPERCENT = "USEDSPACEPERCENT";
   private static final String TOTALCAPACITY = "CAPACITY";
+  private static final String DNUUID = "UUID";
+  private static final String VERSION = "VERSION";
   /**
    * Constructs SCM machine Manager.
    */
@@ -447,6 +449,8 @@ public RegisteredCommand register(
           processNodeReport(datanodeDetails, nodeReport);
           LOG.info("Updated datanode to: {}", dn);
           scmNodeEventPublisher.fireEvent(SCMEvents.NODE_ADDRESS_UPDATE, dn);
+        } else if (isVersionChange(oldNode.getVersion(), datanodeDetails.getVersion())) {
+          nodeStateManager.updateNode(datanodeDetails, layoutInfo);
         }
       } catch (NodeNotFoundException e) {
         LOG.error("Cannot find datanode {} from nodeStateManager",
@@ -508,6 +512,18 @@ private boolean updateDnsToUuidMap(
     return ipChanged || hostNameChanged;
   }
 
+  /**
+   * Check if the version has been updated.
+   *
+   * @param oldVersion datanode oldVersion
+   * @param newVersion datanode newVersion
+   * @return true means replacement is needed, while false means replacement is not needed.
+   */
+  private boolean isVersionChange(String oldVersion, String newVersion) {
+    final boolean ipChanged = !Objects.equals(oldVersion, newVersion);
+    return ipChanged;
+  }
+
   /**
    * Send heartbeat to indicate the datanode is alive and doing well.
    *
@@ -1136,6 +1152,8 @@ public Map<String, Map<String, String>> getNodeStatusInfo() {
       String nonScmUsedPerc = storagePercentage[1];
       map.put(USEDSPACEPERCENT,
           "Ozone: " + scmUsedPerc + "%, other: " + nonScmUsedPerc + "%");
+      map.put(DNUUID, dni.getUuidString());
+      map.put(VERSION, dni.getVersion());
       nodes.put(hostName, map);
     }
     return nodes;

diff --git a/hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/server/SCMMXBean.java b/hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/server/SCMMXBean.java
@@ -18,6 +18,7 @@
 
 package org.apache.hadoop.hdds.scm.server;
 
+import java.util.List;
 import java.util.Map;
 
 import org.apache.hadoop.hdds.annotation.InterfaceAudience;
@@ -74,6 +75,8 @@ public interface SCMMXBean extends ServiceRuntimeInfo {
 
   String getScmRatisRoles();
 
+  List<List<String>> getRatisRoles();
+
   /**
    * Primordial node is the node on which scm init operation is performed.
    * @return hostname of primordialNode

diff --git a/...s/server-scm/src/main/java/org/apache/hadoop/hdds/scm/server/StorageContainerManager.java b/...s/server-scm/src/main/java/org/apache/hadoop/hdds/scm/server/StorageContainerManager.java
@@ -170,7 +170,12 @@
 import org.apache.hadoop.security.UserGroupInformation.AuthenticationMethod;
 import org.apache.hadoop.security.authentication.client.AuthenticationException;
 import org.apache.hadoop.util.ReflectionUtils;
+import org.apache.ratis.proto.RaftProtos;
+import org.apache.ratis.protocol.RaftPeer;
 import org.apache.ratis.protocol.RaftPeerId;
+import org.apache.ratis.server.DivisionInfo;
+import org.apache.ratis.server.RaftServer;
+import org.apache.ratis.thirdparty.com.google.protobuf.ByteString;
 import org.apache.ratis.util.ExitUtils;
 import org.apache.ratis.util.JvmPauseMonitor;
 import org.slf4j.Logger;
@@ -2137,6 +2142,74 @@ public String getScmRatisRoles() {
         HddsUtils.format(server.getRatisRoles()) : "STANDALONE";
   }
 
+  @Override
+  public List<List<String>> getRatisRoles() {
+    final SCMRatisServer server = getScmHAManager().getRatisServer();
+
+    // If Ratis is enabled
+    if(server != null) {
+
+      // To attempt to find the SCM Leader,
+      // and if the Leader is not found
+      // return Leader is not found message.
+      RaftServer.Division division = server.getDivision();
+      RaftPeer leader = getLeader(division);
+      if (leader == null) {
+        return getRatisRolesException("No leader found");
+      }
+
+      // If the SCMRatisServer is stopped, return a service stopped message.
+      if (server.isStopped()) {
+        return getRatisRolesException("Server is shutting down");
+      }
+
+      // Attempt to retrieve role information.
+      try {
+        List<String> ratisRoles = server.getRatisRoles();
+        List<List<String>> result = new ArrayList<>();
+        for (String role : ratisRoles) {
+          String[] roleArr = role.split(":");
+          List<String> scmInfo = new ArrayList<>();
+          // Host Name
+          scmInfo.add(roleArr[0]);
+          // Node ID
+          scmInfo.add(roleArr[3]);
+          // Ratis Port
+          scmInfo.add(roleArr[1]);
+          // Role
+          scmInfo.add(roleArr[2]);
+          result.add(scmInfo);
+        }
+        return result;
+      } catch (Exception e) {
+        LOG.error("Failed to getRatisRoles.", e);
+        return getRatisRolesException("Exception Occurred, " + e.getMessage());
+      }
+    }
+
+    // If Ratis is not enabled, we will throw an exception directly.
+    return getRatisRolesException("Ratis is disabled");
+  }
+
+  public RaftPeer getLeader(RaftServer.Division division) {
+    if (division.getInfo().isLeader()) {
+      return division.getPeer();
+    } else {
+      DivisionInfo info = division.getInfo();
+      RaftProtos.RoleInfoProto roleInfoProto = info.getRoleInfoProto();
+      RaftProtos.FollowerInfoProto followerInfo = roleInfoProto.getFollowerInfo();
+      RaftProtos.ServerRpcProto leaderInfo = followerInfo.getLeaderInfo();
+      RaftProtos.RaftPeerProto peerLeaderId = leaderInfo.getId();
+      ByteString leaderId = peerLeaderId.getId();
+      return leaderId.isEmpty() ? null :
+          division.getRaftConf().getPeer(RaftPeerId.valueOf(leaderId));
+    }
+  }
+
+  private static List<List<String>> getRatisRolesException(String exceptionString) {
+    return Collections.singletonList(Collections.singletonList(exceptionString));
+  }
+
   /**
    * @return hostname of primordialNode
    */

diff --git a/hadoop-hdds/server-scm/src/main/resources/webapps/scm/scm-overview.html b/hadoop-hdds/server-scm/src/main/resources/webapps/scm/scm-overview.html
@@ -140,6 +140,10 @@ <h2>Node Status</h2>
                                         'sortdesc':(columnName == 'comstate' && !reverse)}">Commisioned State</span> </th>
             <th ng-click = "columnSort('lastheartbeat')" class="nodeStatusInfo">  <span ng-class="{'sorting' : (columnName != 'lastheartbeat'), 'sortasc' : (columnName == 'heartbeat' && !reverse),
                                         'sortdesc':(columnName == 'lastheartbeat' && !reverse)}">Last Heartbeat</span> </th>
+            <th ng-click = "columnSort('uuid')" class="uuId" ><span ng-class="{'sorting' : (columnName != 'uuid'), 'sortasc' : (columnName == 'uuid' && !reverse),
+                                        'sortdesc':(columnName == 'uuid' && !reverse)}">UUID</span></th>
+            <th ng-click = "columnSort('version')" class="version" ><span ng-class="{'sorting' : (columnName != 'version'), 'sortasc' : (columnName == 'version' && !reverse),
+                                        'sortdesc':(columnName == 'version' && !reverse)}">Version</span></th>
         </tr>
     </thead>
     <tbody>
@@ -157,6 +161,8 @@ <h2>Node Status</h2>
             <td>{{typestat.opstate}}</td>
             <td>{{typestat.comstate}}</td>
             <td>{{typestat.lastheartbeat}}</td>
+            <td>{{typestat.uuid}}</td>
+            <td>{{typestat.version}}</td>
        </tr>
     </tbody>
 </table>
@@ -210,10 +216,6 @@ <h2>Status</h2>
         <td> Force Exit Safe Mode </td>
         <td>{{$ctrl.overview.jmx.SafeModeExitForceful}}</td>
     </tr>
-    <tr>
-        <td> SCM Roles (HA) </td>
-        <td>{{$ctrl.overview.jmx.ScmRatisRoles}}</td>
-    </tr>
     <tr ng-hide="!$ctrl.overview.jmx.PrimordialNode">
         <td> Primordial Node (HA) </td>
         <td>{{$ctrl.overview.jmx.PrimordialNode}}</td>
@@ -235,6 +237,35 @@ <h2>Meta-Data Volume Information</h2>
     </tbody>
 </table>
 
+<h2>SCM Roles (HA)</h2>
+<h4 ng-show="$ctrl.overview.jmx.RatisRoles.length == 1 && $ctrl.overview.jmx.RatisRoles[0].length == 1">{{$ctrl.overview.jmx.RatisRoles[0][0]}}</h4>
+<div ng-show="$ctrl.overview.jmx.RatisRoles.length > 1">
+    <table class="table table-striped table-bordered" class="col-md-6">
+        <thead>
+        <tr>
+            <th>Host Name</th>
+            <th>Node ID</th>
+            <th>Ratis Port</th>
+            <th>Role</th>
+        </tr>
+        </thead>
+        <tbody ng-repeat="roles in $ctrl.overview.jmx.RatisRoles">
+        <tr class="scm-roles-background" ng-if="$ctrl.role.Id == roles[1]">
+            <td>{{roles[0]}}</td>
+            <td>{{roles[1]}}</td>
+            <td>{{roles[2]}}</td>
+            <td>{{roles[3]}}</td>
+        </tr>
+        <tr ng-if="$ctrl.role.Id != roles[1]">
+            <td>{{roles[0]}}</td>
+            <td>{{roles[1]}}</td>
+            <td>{{roles[2]}}</td>
+            <td>{{roles[3]}}</td>
+        </tr>
+        </tbody>
+    </table>
+</div>
+
 <h2>Safemode rules statuses</h2>
 
 <table class="table table-bordered table-striped" class="col-md-6">

diff --git a/hadoop-hdds/server-scm/src/main/resources/webapps/scm/scm.js b/hadoop-hdds/server-scm/src/main/resources/webapps/scm/scm.js
@@ -56,6 +56,11 @@
                 }
             }
 
+            $http.get("jmx?qry=Ratis:service=RaftServer,group=*,id=*")
+                .then(function (result) {
+                    ctrl.role = result.data.beans[0];
+            });
+
             function get_protocol(URLScheme, value, baseProto, fallbackProto) {
                 let protocol = "unknown"
                 let port = -1;
@@ -95,6 +100,8 @@
                                     capacity: value && value.find((element) => element.key === "CAPACITY").value,
                                     comstate: value && value.find((element) => element.key === "COMSTATE").value,
                                     lastheartbeat: value && value.find((element) => element.key === "LASTHEARTBEAT").value,
+                                    uuid: value && value.find((element) => element.key === "UUID").value,
+                                    version: value && value.find((element) => element.key === "VERSION").value,
                                     port: portSpec.port,
                                     protocol: portSpec.proto
                                 }