HDFS-14329. RBF: Add maintenance nodes to federation metrics. Contributed by Ayush Saxena.

hadoop-hdfs-project/hadoop-hdfs-rbf/src/main/java/org/apache/hadoop/hdfs/server/federation/metrics/FederationMBean.java

@@ -130,6 +130,24 @@ public interface FederationMBean {
    */
   int getNumDecomDeadNodes();
 
+  /**
+   * Get the number of live datanodes which are under maintenance.
+   * @return Number of live datanodes which are under maintenance.
+   */
+  int getNumInMaintenanceLiveDataNodes();
+
+  /**
+   * Get the number of dead datanodes which are under maintenance.
+   * @return Number of dead datanodes which are under maintenance.
+   */
+  int getNumInMaintenanceDeadDataNodes();
+
+  /**
+   * Get the number of datanodes which are entering maintenance.
+   * @return Number of datanodes which are entering maintenance.
+   */
+  int getNumEnteringMaintenanceDataNodes();
+
   /**
    * Get Max, Median, Min and Standard Deviation of DataNodes usage.
    * @return the DataNode usage information, as a JSON string.

hadoop-hdfs-project/hadoop-hdfs-rbf/src/main/java/org/apache/hadoop/hdfs/server/federation/metrics/FederationMetrics.java

@@ -437,6 +437,24 @@ public int getNumDecomDeadNodes() {
         MembershipStats::getNumOfDecomDeadDatanodes);
   }
 
+  @Override
+  public int getNumInMaintenanceLiveDataNodes() {
+    return getNameserviceAggregatedInt(
+        MembershipStats::getNumOfInMaintenanceLiveDataNodes);
+  }
+
+  @Override
+  public int getNumInMaintenanceDeadDataNodes() {
+    return getNameserviceAggregatedInt(
+        MembershipStats::getNumOfInMaintenanceDeadDataNodes);
+  }
+
+  @Override
+  public int getNumEnteringMaintenanceDataNodes() {
+    return getNameserviceAggregatedInt(
+        MembershipStats::getNumOfEnteringMaintenanceDataNodes);
+  }
+
   @Override // NameNodeMXBean
   public String getNodeUsage() {
     float median = 0;

hadoop-hdfs-project/hadoop-hdfs-rbf/src/main/java/org/apache/hadoop/hdfs/server/federation/metrics/NamenodeBeanMetrics.java

@@ -697,16 +697,34 @@ public int getNumDecommissioningDataNodes() {
 
   @Override
   public int getNumInMaintenanceLiveDataNodes() {
+    try {
+      return getFederationMetrics().getNumInMaintenanceLiveDataNodes();
+    } catch (IOException e) {
+      LOG.debug("Failed to get number of live in maintenance nodes",
+          e.getMessage());
+    }
     return 0;
   }
 
   @Override
   public int getNumInMaintenanceDeadDataNodes() {
+    try {
+      return getFederationMetrics().getNumInMaintenanceDeadDataNodes();
+    } catch (IOException e) {
+      LOG.debug("Failed to get number of dead in maintenance nodes",
+          e.getMessage());
+    }
     return 0;
   }
 
   @Override
   public int getNumEnteringMaintenanceDataNodes() {
+    try {
+      return getFederationMetrics().getNumEnteringMaintenanceDataNodes();
+    } catch (IOException e) {
+      LOG.debug("Failed to get number of entering maintenance nodes",
+          e.getMessage());
+    }
     return 0;
   }
 

hadoop-hdfs-project/hadoop-hdfs-rbf/src/main/java/org/apache/hadoop/hdfs/server/federation/resolver/MembershipNamenodeResolver.java

@@ -283,6 +283,12 @@ public boolean registerNamenode(NamenodeStatusReport report)
       stats.setNumOfStaleDatanodes(report.getNumStaleDatanodes());
       stats.setNumOfDecomActiveDatanodes(report.getNumDecomLiveDatanodes());
       stats.setNumOfDecomDeadDatanodes(report.getNumDecomDeadDatanodes());
+      stats.setNumOfInMaintenanceLiveDataNodes(
+          report.getNumInMaintenanceLiveDataNodes());
+      stats.setNumOfInMaintenanceDeadDataNodes(
+          report.getNumInMaintenanceDeadDataNodes());
+      stats.setNumOfEnteringMaintenanceDataNodes(
+          report.getNumEnteringMaintenanceDataNodes());
       record.setStats(stats);
     }
 

hadoop-hdfs-project/hadoop-hdfs-rbf/src/main/java/org/apache/hadoop/hdfs/server/federation/resolver/NamenodeStatusReport.java

@@ -50,6 +50,15 @@ public class NamenodeStatusReport {
   /** Dead decommissioned datanodes. */
   private int deadDecomDatanodes = -1;
 
+  /** Live in maintenance datanodes. */
+  private int inMaintenanceLiveDataNodes = -1;
+
+  /** Dead in maintenance datanodes. */
+  private int inMaintenanceDeadDataNodes = -1;
+
+  /** Entering maintenance datanodes. */
+  private int enteringMaintenanceDataNodes = -1;
+
   /** Space stats. */
   private long availableSpace = -1;
   private long numOfFiles = -1;
@@ -228,15 +237,23 @@ public boolean getSafemode() {
    * @param numDecom Number of decommissioning nodes.
    * @param numLiveDecom Number of decommissioned live nodes.
    * @param numDeadDecom Number of decommissioned dead nodes.
+   * @param numInMaintenanceLive Number of in maintenance live nodes.
+   * @param numInMaintenanceDead Number of in maintenance dead nodes.
+   * @param numEnteringMaintenance Number of entering maintenance nodes.
    */
   public void setDatanodeInfo(int numLive, int numDead, int numStale,
-      int numDecom, int numLiveDecom, int numDeadDecom) {
+      int numDecom, int numLiveDecom, int numDeadDecom,
+      int numInMaintenanceLive, int numInMaintenanceDead,
+      int numEnteringMaintenance) {
     this.liveDatanodes = numLive;
     this.deadDatanodes = numDead;
     this.staleDatanodes = numStale;
     this.decomDatanodes = numDecom;
     this.liveDecomDatanodes = numLiveDecom;
     this.deadDecomDatanodes = numDeadDecom;
+    this.inMaintenanceLiveDataNodes = numInMaintenanceLive;
+    this.inMaintenanceDeadDataNodes = numInMaintenanceDead;
+    this.enteringMaintenanceDataNodes = numEnteringMaintenance;
     this.statsValid = true;
   }
 
@@ -294,6 +311,33 @@ public int getNumDecomDeadDatanodes() {
     return this.deadDecomDatanodes;
   }
 
+  /**
+   * Get the number of live in maintenance nodes.
+   *
+   * @return The number of live in maintenance nodes.
+   */
+  public int getNumInMaintenanceLiveDataNodes() {
+    return this.inMaintenanceLiveDataNodes;
+  }
+
+  /**
+   * Get the number of dead in maintenance nodes.
+   *
+   * @return The number of dead in maintenance nodes.
+   */
+  public int getNumInMaintenanceDeadDataNodes() {
+    return this.inMaintenanceDeadDataNodes;
+  }
+
+  /**
+   * Get the number of entering maintenance nodes.
+   *
+   * @return The number of entering maintenance nodes.
+   */
+  public int getNumEnteringMaintenanceDataNodes() {
+    return this.enteringMaintenanceDataNodes;
+  }
+
   /**
    * Set the filesystem information.
    *

hadoop-hdfs-project/hadoop-hdfs-rbf/src/main/java/org/apache/hadoop/hdfs/server/federation/router/NamenodeHeartbeatService.java

@@ -341,7 +341,10 @@ private void updateJMXParameters(
                 jsonObject.getInt("NumStaleDataNodes"),
                 jsonObject.getInt("NumDecommissioningDataNodes"),
                 jsonObject.getInt("NumDecomLiveDataNodes"),
-                jsonObject.getInt("NumDecomDeadDataNodes"));
+                jsonObject.getInt("NumDecomDeadDataNodes"),
+                jsonObject.getInt("NumInMaintenanceLiveDataNodes"),
+                jsonObject.getInt("NumInMaintenanceDeadDataNodes"),
+                jsonObject.getInt("NumEnteringMaintenanceDataNodes"));
           } else if (name.equals(
               "Hadoop:service=NameNode,name=FSNamesystem")) {
             report.setNamesystemInfo(

hadoop-hdfs-project/hadoop-hdfs-rbf/src/main/java/org/apache/hadoop/hdfs/server/federation/store/records/MembershipStats.java

@@ -97,6 +97,18 @@ public static MembershipStats newInstance() throws IOException {
 
   public abstract int getNumOfDecomDeadDatanodes();
 
+  public abstract void setNumOfInMaintenanceLiveDataNodes(int nodes);
+
+  public abstract int getNumOfInMaintenanceLiveDataNodes();
+
+  public abstract void setNumOfInMaintenanceDeadDataNodes(int nodes);
+
+  public abstract int getNumOfInMaintenanceDeadDataNodes();
+
+  public abstract void setNumOfEnteringMaintenanceDataNodes(int nodes);
+
+  public abstract int getNumOfEnteringMaintenanceDataNodes();
+
   @Override
   public SortedMap<String, String> getPrimaryKeys() {
     // This record is not stored directly, no key needed

hadoop-hdfs-project/hadoop-hdfs-rbf/src/main/java/org/apache/hadoop/hdfs/server/federation/store/records/impl/pb/MembershipStatsPBImpl.java

@@ -208,4 +208,37 @@ public void setNumOfDecomDeadDatanodes(int nodes) {
   public int getNumOfDecomDeadDatanodes() {
     return this.translator.getProtoOrBuilder().getNumOfDecomDeadDatanodes();
   }
+
+  @Override
+  public void setNumOfInMaintenanceLiveDataNodes(int nodes) {
+    this.translator.getBuilder().setNumOfInMaintenanceLiveDataNodes(nodes);
+  }
+
+  @Override
+  public int getNumOfInMaintenanceLiveDataNodes() {
+    return this.translator.getProtoOrBuilder()
+        .getNumOfInMaintenanceLiveDataNodes();
+  }
+
+  @Override
+  public void setNumOfInMaintenanceDeadDataNodes(int nodes) {
+    this.translator.getBuilder().setNumOfInMaintenanceDeadDataNodes(nodes);
+  }
+
+  @Override
+  public int getNumOfInMaintenanceDeadDataNodes() {
+    return this.translator.getProtoOrBuilder()
+        .getNumOfInMaintenanceDeadDataNodes();
+  }
+
+  @Override
+  public void setNumOfEnteringMaintenanceDataNodes(int nodes) {
+    this.translator.getBuilder().setNumOfEnteringMaintenanceDataNodes(nodes);
+  }
+
+  @Override
+  public int getNumOfEnteringMaintenanceDataNodes() {
+    return this.translator.getProtoOrBuilder()
+        .getNumOfEnteringMaintenanceDataNodes();
+  }
 }

hadoop-hdfs-project/hadoop-hdfs-rbf/src/main/proto/FederationProtocol.proto

@@ -46,6 +46,9 @@ message NamenodeMembershipStatsRecordProto {
   optional uint32 numOfDecomActiveDatanodes = 23;
   optional uint32 numOfDecomDeadDatanodes = 24;
   optional uint32 numOfStaleDatanodes = 25;
+  optional uint32 numOfInMaintenanceLiveDataNodes = 26;
+  optional uint32 numOfInMaintenanceDeadDataNodes = 27;
+  optional uint32 numOfEnteringMaintenanceDataNodes = 28;
 }
 
 message NamenodeMembershipRecordProto {

hadoop-hdfs-project/hadoop-hdfs-rbf/src/test/java/org/apache/hadoop/hdfs/server/federation/metrics/TestFederationMetrics.java

@@ -145,6 +145,12 @@ public void testNamenodeStatsDataSource() throws IOException, JSONException {
           stats.getNumOfDecomActiveDatanodes());
       assertEquals(json.getLong("numOfDecomDeadDatanodes"),
           stats.getNumOfDecomDeadDatanodes());
+      assertEquals(json.getLong("numOfInMaintenanceLiveDataNodes"),
+          stats.getNumOfInMaintenanceLiveDataNodes());
+      assertEquals(json.getLong("numOfInMaintenanceDeadDataNodes"),
+          stats.getNumOfInMaintenanceDeadDataNodes());
+      assertEquals(json.getLong("numOfEnteringMaintenanceDataNodes"),
+          stats.getNumOfEnteringMaintenanceDataNodes());
       assertEquals(json.getLong("numOfBlocks"), stats.getNumOfBlocks());
       assertEquals(json.getString("rpcAddress"), mockEntry.getRpcAddress());
       assertEquals(json.getString("webAddress"), mockEntry.getWebAddress());
@@ -197,6 +203,12 @@ public void testNameserviceStatsDataSource()
           json.getLong("numOfDecomActiveDatanodes"));
       assertEquals(stats.getNumOfDecomDeadDatanodes(),
           json.getLong("numOfDecomDeadDatanodes"));
+      assertEquals(stats.getNumOfInMaintenanceLiveDataNodes(),
+          json.getLong("numOfInMaintenanceLiveDataNodes"));
+      assertEquals(stats.getNumOfInMaintenanceDeadDataNodes(),
+          json.getLong("numOfInMaintenanceDeadDataNodes"));
+      assertEquals(stats.getNumOfStaleDatanodes(),
+          json.getLong("numOfEnteringMaintenanceDataNodes"));
       assertEquals(stats.getProvidedSpace(),
           json.getLong("providedSpace"));
       nameservicesFound++;
@@ -268,6 +280,9 @@ private void validateClusterStatsBean(FederationMBean bean)
     long numDecom = 0;
     long numDecomLive = 0;
     long numDecomDead = 0;
+    long numInMaintenanceLive = 0;
+    long numInMaintenanceDead = 0;
+    long numEnteringMaintenance = 0;
     long numFiles = 0;
     for (MembershipState mock : getActiveMemberships()) {
       MembershipStats stats = mock.getStats();
@@ -278,6 +293,9 @@ private void validateClusterStatsBean(FederationMBean bean)
       numDecom += stats.getNumOfDecommissioningDatanodes();
       numDecomLive += stats.getNumOfDecomActiveDatanodes();
       numDecomDead += stats.getNumOfDecomDeadDatanodes();
+      numInMaintenanceLive += stats.getNumOfInMaintenanceLiveDataNodes();
+      numInMaintenanceDead += stats.getNumOfInMaintenanceLiveDataNodes();
+      numEnteringMaintenance += stats.getNumOfEnteringMaintenanceDataNodes();
     }
 
     assertEquals(numBlocks, bean.getNumBlocks());
@@ -287,6 +305,10 @@ private void validateClusterStatsBean(FederationMBean bean)
     assertEquals(numDecom, bean.getNumDecommissioningNodes());
     assertEquals(numDecomLive, bean.getNumDecomLiveNodes());
     assertEquals(numDecomDead, bean.getNumDecomDeadNodes());
+    assertEquals(numInMaintenanceLive, bean.getNumInMaintenanceLiveDataNodes());
+    assertEquals(numInMaintenanceDead, bean.getNumInMaintenanceDeadDataNodes());
+    assertEquals(numEnteringMaintenance,
+        bean.getNumEnteringMaintenanceDataNodes());
     assertEquals(numFiles, bean.getNumFiles());
     assertEquals(getActiveMemberships().size() + getStandbyMemberships().size(),
         bean.getNumNamenodes());

hadoop-hdfs-project/hadoop-hdfs-rbf/src/test/java/org/apache/hadoop/hdfs/server/federation/store/records/TestMembershipState.java

@@ -52,6 +52,9 @@ public class TestMembershipState {
   private static final int NUM_DECOM = 700;
   private static final int NUM_DECOM_ACTIVE = 800;
   private static final int NUM_DECOM_DEAD = 900;
+  private static final int NUM_MAIN_LIVE = 151;
+  private static final int NUM_MAIN_DEAD = 303;
+  private static final int NUM_ENTER_MAIN = 144;
   private static final long NUM_BLOCK_MISSING = 1000;
 
   private static final long TOTAL_SPACE = 1100;
@@ -78,6 +81,9 @@ private MembershipState createRecord() throws IOException {
     stats.setNumOfDecommissioningDatanodes(NUM_DECOM);
     stats.setNumOfDecomActiveDatanodes(NUM_DECOM_ACTIVE);
     stats.setNumOfDecomDeadDatanodes(NUM_DECOM_DEAD);
+    stats.setNumOfInMaintenanceLiveDataNodes(NUM_MAIN_LIVE);
+    stats.setNumOfInMaintenanceDeadDataNodes(NUM_MAIN_DEAD);
+    stats.setNumOfEnteringMaintenanceDataNodes(NUM_ENTER_MAIN);
     stats.setNumOfBlocksMissing(NUM_BLOCK_MISSING);
     stats.setTotalSpace(TOTAL_SPACE);
     stats.setAvailableSpace(AVAILABLE_SPACE);
@@ -107,6 +113,9 @@ private void validateRecord(MembershipState record) throws IOException {
     assertEquals(NUM_DECOM, stats.getNumOfDecommissioningDatanodes());
     assertEquals(NUM_DECOM_ACTIVE, stats.getNumOfDecomActiveDatanodes());
     assertEquals(NUM_DECOM_DEAD, stats.getNumOfDecomDeadDatanodes());
+    assertEquals(NUM_MAIN_LIVE, stats.getNumOfInMaintenanceLiveDataNodes());
+    assertEquals(NUM_MAIN_DEAD, stats.getNumOfInMaintenanceDeadDataNodes());
+    assertEquals(NUM_ENTER_MAIN, stats.getNumOfEnteringMaintenanceDataNodes());
     assertEquals(TOTAL_SPACE, stats.getTotalSpace());
     assertEquals(AVAILABLE_SPACE, stats.getAvailableSpace());
   }