HBASE-27036 Displays the number of decommissioned region server for status command (#4431)

add decommissionedServers to cluster metric

Signed-off-by: Pankaj Kumar <pankajkumar@apache.org>
Signed-off-by: Duo Zhang <zhangduo@apache.org>
Signed-off-by: Viraj Jasani <vjasani@apache.org

hbase-client/src/main/java/org/apache/hadoop/hbase/ClusterMetrics.java

@@ -80,6 +80,11 @@ public interface ClusterMetrics {
    */
   List<ServerName> getDeadServerNames();
 
+  /**
+   * @return the names of region servers on the decommissioned list
+   */
+  List<ServerName> getDecommissionedServerNames();
+
   /**
    * @return the names of region servers on the live list
    */
@@ -222,5 +227,9 @@ public interface ClusterMetrics {
      * metrics about monitored tasks
      */
     TASKS,
+    /**
+     * metrics about decommissioned region servers
+     */
+    DECOMMISSIONED_SERVERS,
   }
 }

hbase-client/src/main/java/org/apache/hadoop/hbase/ClusterMetricsBuilder.java

@@ -69,7 +69,9 @@ public final class ClusterMetricsBuilder {
           .map(status -> ClusterStatusProtos.TableRegionStatesCount.newBuilder()
             .setTableName(ProtobufUtil.toProtoTableName((status.getKey())))
             .setRegionStatesCount(ProtobufUtil.toTableRegionStatesCount(status.getValue())).build())
-          .collect(Collectors.toList()));
+          .collect(Collectors.toList()))
+        .addAllDecommissionedServers(metrics.getDecommissionedServerNames().stream()
+          .map(ProtobufUtil::toServerName).collect(Collectors.toList()));
     if (metrics.getMasterName() != null) {
       builder.setMaster(ProtobufUtil.toServerName((metrics.getMasterName())));
     }
@@ -111,7 +113,9 @@ public final class ClusterMetricsBuilder {
         .collect(Collectors.toMap(e -> ProtobufUtil.toTableName(e.getTableName()),
           e -> ProtobufUtil.toTableRegionStatesCount(e.getRegionStatesCount()))))
       .setMasterTasks(proto.getMasterTasksList().stream().map(t -> ProtobufUtil.getServerTask(t))
-        .collect(Collectors.toList()));
+        .collect(Collectors.toList()))
+      .setDecommissionedServerNames(proto.getDecommissionedServersList().stream()
+        .map(ProtobufUtil::toServerName).collect(Collectors.toList()));
     if (proto.hasClusterId()) {
       builder.setClusterId(ClusterId.convert(proto.getClusterId()).toString());
     }
@@ -167,6 +171,8 @@ public final class ClusterMetricsBuilder {
         return ClusterMetrics.Option.TABLE_TO_REGIONS_COUNT;
       case TASKS:
         return ClusterMetrics.Option.TASKS;
+      case DECOMMISSIONED_SERVERS:
+        return ClusterMetrics.Option.DECOMMISSIONED_SERVERS;
       // should not reach here
       default:
         throw new IllegalArgumentException("Invalid option: " + option);
@@ -206,6 +212,8 @@ public final class ClusterMetricsBuilder {
         return ClusterStatusProtos.Option.TABLE_TO_REGIONS_COUNT;
       case TASKS:
         return ClusterStatusProtos.Option.TASKS;
+      case DECOMMISSIONED_SERVERS:
+        return ClusterStatusProtos.Option.DECOMMISSIONED_SERVERS;
       // should not reach here
       default:
         throw new IllegalArgumentException("Invalid option: " + option);
@@ -253,6 +261,7 @@ public final class ClusterMetricsBuilder {
   private Map<TableName, RegionStatesCount> tableRegionStatesCount = Collections.emptyMap();
   @Nullable
   private List<ServerTask> masterTasks;
+  private List<ServerName> decommissionedServerNames = Collections.emptyList();
 
   private ClusterMetricsBuilder() {
   }
@@ -317,6 +326,11 @@ public final class ClusterMetricsBuilder {
     return this;
   }
 
+  public ClusterMetricsBuilder setDecommissionedServerNames(List<ServerName> value) {
+    this.decommissionedServerNames = value;
+    return this;
+  }
+
   public ClusterMetricsBuilder
     setTableRegionStatesCount(Map<TableName, RegionStatesCount> tableRegionStatesCount) {
     this.tableRegionStatesCount = tableRegionStatesCount;
@@ -326,13 +340,14 @@ public final class ClusterMetricsBuilder {
   public ClusterMetrics build() {
     return new ClusterMetricsImpl(hbaseVersion, deadServerNames, liveServerMetrics, masterName,
       backupMasterNames, regionsInTransition, clusterId, masterCoprocessorNames, balancerOn,
-      masterInfoPort, serversName, tableRegionStatesCount, masterTasks);
+      masterInfoPort, serversName, tableRegionStatesCount, masterTasks, decommissionedServerNames);
   }
 
   private static class ClusterMetricsImpl implements ClusterMetrics {
     @Nullable
     private final String hbaseVersion;
     private final List<ServerName> deadServerNames;
+    private final List<ServerName> decommissionedServerNames;
     private final Map<ServerName, ServerMetrics> liveServerMetrics;
     @Nullable
     private final ServerName masterName;
@@ -353,9 +368,10 @@ public final class ClusterMetricsBuilder {
       List<ServerName> backupMasterNames, List<RegionState> regionsInTransition, String clusterId,
       List<String> masterCoprocessorNames, Boolean balancerOn, int masterInfoPort,
       List<ServerName> serversName, Map<TableName, RegionStatesCount> tableRegionStatesCount,
-      List<ServerTask> masterTasks) {
+      List<ServerTask> masterTasks, List<ServerName> decommissionedServerNames) {
       this.hbaseVersion = hbaseVersion;
       this.deadServerNames = Preconditions.checkNotNull(deadServerNames);
+      this.decommissionedServerNames = Preconditions.checkNotNull(decommissionedServerNames);
       this.liveServerMetrics = Preconditions.checkNotNull(liveServerMetrics);
       this.masterName = masterName;
       this.backupMasterNames = Preconditions.checkNotNull(backupMasterNames);
@@ -379,6 +395,11 @@ public final class ClusterMetricsBuilder {
       return Collections.unmodifiableList(deadServerNames);
     }
 
+    @Override
+    public List<ServerName> getDecommissionedServerNames() {
+      return Collections.unmodifiableList(decommissionedServerNames);
+    }
+
     @Override
     public Map<ServerName, ServerMetrics> getLiveServerMetrics() {
       return Collections.unmodifiableMap(liveServerMetrics);

hbase-protocol-shaded/src/main/protobuf/server/ClusterStatus.proto

@@ -349,6 +349,7 @@ message ClusterStatus {
   repeated ServerName servers_name = 11;
   repeated TableRegionStatesCount table_region_states_count = 12;
   repeated ServerTask master_tasks = 13;
+  repeated ServerName decommissioned_servers = 14;
 }
 
 enum Option {
@@ -365,4 +366,5 @@ enum Option {
   SERVERS_NAME = 10;
   TABLE_TO_REGIONS_COUNT = 11;
   TASKS = 12;
+  DECOMMISSIONED_SERVERS = 13;
 }

hbase-server/src/main/java/org/apache/hadoop/hbase/master/HMaster.java

@@ -2903,6 +2903,12 @@ public class HMaster extends HBaseServerBase<MasterRpcServices> implements Maste
           }
           break;
         }
+        case DECOMMISSIONED_SERVERS: {
+          if (serverManager != null) {
+            builder.setDecommissionedServerNames(serverManager.getDrainingServersList());
+          }
+          break;
+        }
       }
     }
 

hbase-server/src/test/java/org/apache/hadoop/hbase/TestClientClusterMetrics.java

@@ -132,6 +132,12 @@ public class TestClientClusterMetrics {
     Assert.assertEquals(origin.getMasterInfoPort(), defaults.getMasterInfoPort());
     Assert.assertEquals(origin.getServersName().size(), defaults.getServersName().size());
     Assert.assertEquals(ADMIN.getRegionServers().size(), defaults.getServersName().size());
+    // We decommission the first online region server and verify the metrics.
+    List<ServerName> serverNames = origin.getServersName().subList(0, 1);
+    ADMIN.decommissionRegionServers(serverNames, false);
+    Assert.assertEquals(1, ADMIN.getClusterMetrics().getDecommissionedServerNames().size());
+    Assert.assertEquals(ADMIN.getClusterMetrics().getDecommissionedServerNames().get(0),
+      serverNames.get(0));
   }
 
   @Test

hbase-server/src/test/java/org/apache/hadoop/hbase/master/TestRegionsRecoveryChore.java

@@ -227,6 +227,11 @@ public class TestRegionsRecoveryChore {
         return null;
       }
 
+      @Override
+      public List<ServerName> getDecommissionedServerNames() {
+        return null;
+      }
+
       @Override
       public Map<ServerName, ServerMetrics> getLiveServerMetrics() {
         Map<ServerName, ServerMetrics> liveServerMetrics = new HashMap<>();

hbase-shell/src/main/ruby/hbase/admin.rb

@@ -1043,6 +1043,7 @@ module Hbase
       else
         puts "1 active master, #{cluster_metrics.getBackupMasterNames.size} backup masters,
               #{cluster_metrics.getLiveServerMetrics.size} servers,
+              #{cluster_metrics.getDecommissionedServerNames.size} decommissioned,
               #{cluster_metrics.getDeadServerNames.size} dead,
               #{format('%.4f', cluster_metrics.getAverageLoad)} average load"
       end