HBASE-25460 : Expose drainingServers as cluster metric (#2994) (#2995)

Signed-off-by: Viraj Jasani <[email protected]>

Author: rahulLiving

hbase-hadoop-compat/src/main/java/org/apache/hadoop/hbase/master/MetricsMasterSource.java

@@ -55,8 +55,10 @@ public interface MetricsMasterSource extends BaseSource {
   String AVERAGE_LOAD_NAME = "averageLoad";
   String LIVE_REGION_SERVERS_NAME = "liveRegionServers";
   String DEAD_REGION_SERVERS_NAME = "deadRegionServers";
+  String DRAINING_REGION_SERVER_NAME = "draininigRegionServers";
   String NUM_REGION_SERVERS_NAME = "numRegionServers";
   String NUM_DEAD_REGION_SERVERS_NAME = "numDeadRegionServers";
+  String NUM_DRAINING_REGION_SERVERS_NAME = "numDrainingRegionServers";
   String ZOOKEEPER_QUORUM_NAME = "zookeeperQuorum";
   String SERVER_NAME_NAME = "serverName";
   String CLUSTER_ID_NAME = "clusterId";
@@ -73,6 +75,8 @@ public interface MetricsMasterSource extends BaseSource {
   String NUMBER_OF_REGION_SERVERS_DESC = "Number of RegionServers";
   String DEAD_REGION_SERVERS_DESC = "Names of dead RegionServers";
   String NUMBER_OF_DEAD_REGION_SERVERS_DESC = "Number of dead RegionServers";
+  String DRAINING_REGION_SERVER_DESC = "Names of draining RegionServers";
+  String NUMBER_OF_DRAINING_REGION_SERVERS_DESC = "Number of draining RegionServers";
   String ZOOKEEPER_QUORUM_DESC = "ZooKeeper Quorum";
   String SERVER_NAME_DESC = "Server Name";
   String CLUSTER_ID_DESC = "Cluster Id";

hbase-hadoop-compat/src/main/java/org/apache/hadoop/hbase/master/MetricsMasterWrapper.java

@@ -117,6 +117,20 @@ public interface MetricsMasterWrapper {
    */
   int getNumDeadRegionServers();
 
+  /**
+   * Get the draining region servers
+   *
+   * @return Draining region server
+   */
+  String getDrainingRegionServers();
+
+  /**
+   * Get the number of draining region servers
+   *
+   * @return number of draining region servers
+   */
+  int getNumDrainingRegionServers();
+
   /**
    * Get the number of master WAL files.
    */

hbase-hadoop2-compat/src/main/java/org/apache/hadoop/hbase/master/MetricsMasterSourceImpl.java

@@ -108,6 +108,10 @@ public void getMetrics(MetricsCollector metricsCollector, boolean all) {
           .addGauge(Interns.info(NUM_DEAD_REGION_SERVERS_NAME,
               NUMBER_OF_DEAD_REGION_SERVERS_DESC),
               masterWrapper.getNumDeadRegionServers())
+          .tag(Interns.info(DRAINING_REGION_SERVER_NAME, DRAINING_REGION_SERVER_DESC),
+              masterWrapper.getDrainingRegionServers())
+          .addGauge(Interns.info(NUM_DRAINING_REGION_SERVERS_NAME, NUMBER_OF_REGION_SERVERS_DESC),
+              masterWrapper.getNumDrainingRegionServers())
           .tag(Interns.info(ZOOKEEPER_QUORUM_NAME, ZOOKEEPER_QUORUM_DESC),
               masterWrapper.getZookeeperQuorum())
           .tag(Interns.info(SERVER_NAME_NAME, SERVER_NAME_DESC), masterWrapper.getServerName())

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

@@ -133,6 +133,24 @@ public int getNumDeadRegionServers() {
     return !(master.isStopped() || master.isStopping());
   }
 
+  @Override
+  public String getDrainingRegionServers() {
+    ServerManager serverManager = this.master.getServerManager();
+    if (serverManager == null) {
+        return "";
+    }
+    return StringUtils.join(serverManager.getDrainingServersList()  , ";");
+  }
+
+  @Override
+  public int getNumDrainingRegionServers() {
+    ServerManager serverManager = this.master.getServerManager();
+    if (serverManager == null) {
+        return 0;
+    }
+    return serverManager.getDrainingServersList().size();
+  }
+
   @Override
   public String getServerName() {
     ServerName serverName = master.getServerName();

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

@@ -139,6 +139,7 @@ public void testDefaultMasterMetrics() throws Exception {
     metricsHelper.assertGauge( "numRegionServers",1 + (tablesOnMaster? 1: 0), masterSource);
     metricsHelper.assertGauge( "averageLoad", 1 + (tablesOnMaster? 0: 1), masterSource);
     metricsHelper.assertGauge( "numDeadRegionServers", 0, masterSource);
+    metricsHelper.assertGauge("numDrainingRegionServers", 0, masterSource);
 
     metricsHelper.assertGauge("masterStartTime", master.getMasterStartTime(), masterSource);
     metricsHelper.assertGauge("masterActiveTime", master.getMasterActiveTime(), masterSource);