HDDS-11272. Statistics some node status information

hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/node/DatanodeInfo.java

@@ -227,6 +227,15 @@ public int getHealthyVolumeCount() {
     }
   }
 
+  public int getFailedVolumeCount() {
+    try {
+      lock.readLock().lock();
+      return failedVolumeCount;
+    } finally {
+      lock.readLock().unlock();
+    }
+  }
+
   /**
    * Returns count of healthy metadata volumes reported from datanode.
    * @return count of healthy metdata log volumes

hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/node/NodeStateManager.java

@@ -22,6 +22,7 @@
 import java.util.HashMap;
 import java.util.HashSet;
 import java.util.List;
+import java.util.ArrayList;
 import java.util.Map;
 import java.util.Set;
 import java.util.UUID;
@@ -477,6 +478,22 @@ public List<DatanodeInfo> getDeadNodes() {
     return getNodes(null, NodeState.DEAD);
   }
 
+  public List<DatanodeInfo> getDecommissioningNodes() {
+    return getNodes(NodeOperationalState.DECOMMISSIONING, null);
+  }
+
+  public int getDecommissioningNodeCount() {
+    return getDecommissioningNodes().size();
+  }
+
+  public List<DatanodeInfo> getEnteringMaintenanceNodes() {
+    return getNodes(NodeOperationalState.ENTERING_MAINTENANCE, null);
+  }
+
+  public int getEnteringMaintenanceNodeCount() {
+    return getEnteringMaintenanceNodes().size();
+  }
+
   /**
    * Returns all the nodes with the specified status.
    *
@@ -501,6 +518,26 @@ public List<DatanodeInfo> getNodes(
     return nodeStateMap.getDatanodeInfos(opState, health);
   }
 
+  public List<DatanodeInfo> getVolumeFailuresNodes() {
+    List<DatanodeInfo> allNodes = nodeStateMap.getAllDatanodeInfos();
+    if (allNodes.size() < 1) {
+      return allNodes;
+    }
+
+    List<DatanodeInfo> failedVolumeNodes = new ArrayList<>();
+    for (DatanodeInfo dn : allNodes) {
+      if (dn.getFailedVolumeCount() > 0) {
+        failedVolumeNodes.add(dn);
+      }
+    }
+
+    return failedVolumeNodes;
+  }
+
+  public int getVolumeFailuresNodeCount() {
+    return getVolumeFailuresNodes().size();
+  }
+
   /**
    * Returns all the nodes which have registered to NodeStateManager.
    *

hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/node/SCMNodeManager.java

@@ -1223,6 +1223,8 @@ public Map<String, String> getNodeStatistics() {
     Map<String, String> nodeStatistics = new HashMap<>();
     // Statistics node usaged
     nodeUsageStatistics(nodeStatistics);
+    // Statistics node states
+    nodeStateStatistics(nodeStatistics);
     // todo: Statistics of other instances
     return nodeStatistics;
   }
@@ -1265,6 +1267,19 @@ private void nodeUsageStatistics(Map<String, String> nodeStatics) {
     nodeStatics.put(UsageStatics.STDEV.getLabel(), decimalFormat.format(dev));
   }
 
+  private void nodeStateStatistics(Map<String, String> nodeStatics) {
+    int healthyNodeCount = nodeStateManager.getHealthyNodeCount();
+    int deadNodeCount = nodeStateManager.getDeadNodeCount();
+    int decommissioningNodeCount = nodeStateManager.getDecommissioningNodeCount();
+    int enteringMaintenanceNodeCount = nodeStateManager.getEnteringMaintenanceNodeCount();
+    int volumeFailuresNodeCount = nodeStateManager.getVolumeFailuresNodeCount();
+    nodeStatics.put(StateStatics.HEALTHY.getLabel(), String.valueOf(healthyNodeCount));
+    nodeStatics.put(StateStatics.DEAD.getLabel(), String.valueOf(deadNodeCount));
+    nodeStatics.put(StateStatics.DECOMMISSIONING.getLabel(), String.valueOf(decommissioningNodeCount));
+    nodeStatics.put(StateStatics.ENTERING_MAINTENANCE.getLabel(), String.valueOf(enteringMaintenanceNodeCount));
+    nodeStatics.put(StateStatics.VOLUME_FAILURES.getLabel(), String.valueOf(volumeFailuresNodeCount));
+  }
+
   /**
    * Based on the current time and the last heartbeat, calculate the time difference
    * and get a string of the relative value. E.g. "2s ago", "1m 2s ago", etc.
@@ -1346,6 +1361,21 @@ public String getLabel() {
     }
   }
 
+  private enum StateStatics {
+    HEALTHY("Healthy"),
+    DEAD("Dead"),
+    DECOMMISSIONING("Decommissioning"),
+    ENTERING_MAINTENANCE("EnteringMaintenance"),
+    VOLUME_FAILURES("VolumeFailures");
+    private String label;
+    public String getLabel() {
+      return label;
+    }
+    StateStatics(String label) {
+      this.label = label;
+    }
+  }
+
   /**
    * Returns the min of no healthy volumes reported out of the set
    * of datanodes constituting the pipeline.

hadoop-hdds/server-scm/src/main/resources/webapps/scm/scm-overview.html

@@ -51,6 +51,30 @@ <h2>Statistics</h2>
         <td>Standard Deviation</td>
         <td>{{statistics.nodes.usages.stdev}}</td>
     </tr>
+    <tr>
+        <th>Datanode State</th>
+        <th>Count</th>
+    </tr>
+    <tr>
+        <td>Healthy Nodes</td>
+        <td>{{statistics.nodes.state.healthy}}</td>
+    </tr>
+    <tr>
+        <td>Dead Nodes</td>
+        <td>{{statistics.nodes.state.dead}}</td>
+    </tr>
+    <tr>
+        <td>Decommissioning Nodes</td>
+        <td>{{statistics.nodes.state.decommissioning}}</td>
+    </tr>
+    <tr>
+        <td>Entering Maintenance Nodes</td>
+        <td>{{statistics.nodes.state.enteringmaintenance}}</td>
+    </tr>
+    <tr>
+        <td>Volume Failures Nodes</td>
+        <td>{{statistics.nodes.state.volumefailures}}</td>
+    </tr>
     </tbody>
 </table>
 

hadoop-hdds/server-scm/src/main/resources/webapps/scm/scm.js

@@ -39,6 +39,13 @@
                         max : "N/A",
                         median : "N/A",
                         stdev : "N/A"
+                    },
+                    state : {
+                        healthy : "N/A",
+                        dead : "N/A",
+                        decommissioning : "N/A",
+                        enteringmaintenance : "N/A",
+                        volumefailures : "N/A"
                     }
                 }
             }
@@ -92,15 +99,25 @@
                     $scope.lastIndex = Math.ceil(nodeStatusCopy.length / $scope.RecordsToDisplay);
                     $scope.nodeStatus = nodeStatusCopy.slice(0, $scope.RecordsToDisplay);
 
-                    ctrl.nodemanagermetrics.NodeStatistics.forEach(function(obj) {
-                        if(obj.key == "Min") {
-                            $scope.statistics.nodes.usages.min = obj.value;
-                        } else if(obj.key == "Max") {
-                            $scope.statistics.nodes.usages.max = obj.value;
-                        } else if(obj.key == "Median") {
-                            $scope.statistics.nodes.usages.median = obj.value;
-                        } else if(obj.key == "Stdev") {
-                            $scope.statistics.nodes.usages.stdev = obj.value;
+                    ctrl.nodemanagermetrics.NodeStatistics.forEach(({key, value}) => {
+                        if(key == "Min") {
+                            $scope.statistics.nodes.usages.min = value;
+                        } else if(key == "Max") {
+                            $scope.statistics.nodes.usages.max = value;
+                        } else if(key == "Median") {
+                            $scope.statistics.nodes.usages.median = value;
+                        } else if(key == "Stdev") {
+                            $scope.statistics.nodes.usages.stdev = value;
+                        } else if(key == "Healthy") {
+                            $scope.statistics.nodes.state.healthy = value;
+                        } else if(key == "Dead") {
+                            $scope.statistics.nodes.state.dead = value;
+                        } else if(key == "Decommissioning") {
+                            $scope.statistics.nodes.state.decommissioning = value;
+                        } else if(key == "EnteringMaintenance") {
+                            $scope.statistics.nodes.state.enteringmaintenance = value;
+                        } else if(key == "VolumeFailures") {
+                            $scope.statistics.nodes.state.volumefailures = value;
                         }
                     });
                 });