[#2350] improvement(coordinator): Add metrics of active/lost server n…

…umber (#2351) ### What changes were proposed in this pull request? Add metrics of active/lost server number for #2350 ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? Needn't --------- Co-authored-by: Junfan Zhang <[email protected]>
apache · Jan 26, 2025 · 7e63bde · 7e63bde
1 parent deb5de3
commit 7e63bde
Show file tree

Hide file tree

Showing 2 changed files with 17 additions and 2 deletions.
diff --git a/coordinator/src/main/java/org/apache/uniffle/coordinator/SimpleClusterManager.java b/coordinator/src/main/java/org/apache/uniffle/coordinator/SimpleClusterManager.java
@@ -25,6 +25,7 @@
 import java.io.InputStreamReader;
 import java.io.OutputStreamWriter;
 import java.nio.charset.StandardCharsets;
+import java.util.HashSet;
 import java.util.List;
 import java.util.Map;
 import java.util.Set;
@@ -171,6 +172,16 @@ void nodesCheck() {
 
       CoordinatorMetrics.gaugeUnhealthyServerNum.set(unhealthyNodes.size());
       CoordinatorMetrics.gaugeTotalServerNum.set(servers.size());
+      CoordinatorMetrics.gaugeLostServerNum.set(lostNodes.size());
+
+      // get the active server num.
+      Set<String> allServers = new HashSet<>(servers.keySet());
+      allServers.removeAll(excludedNodes);
+      for (ServerNode unhealthyNode : unhealthyNodes) {
+        allServers.remove(unhealthyNode.getId());
+      }
+      CoordinatorMetrics.gaugeActiveServerNum.set(allServers.size());
+
     } catch (Exception e) {
       LOG.warn("Error happened in nodesCheck", e);
     }

diff --git a/coordinator/src/main/java/org/apache/uniffle/coordinator/metric/CoordinatorMetrics.java b/coordinator/src/main/java/org/apache/uniffle/coordinator/metric/CoordinatorMetrics.java
@@ -33,7 +33,8 @@
 import org.apache.uniffle.common.util.RssUtils;
 
 public class CoordinatorMetrics {
-
+  private static final String ACTIVE_SERVER_NUM = "active_server_num";
+  private static final String LOST_SERVER_NUM = "lost_server_num";
   private static final String TOTAL_SERVER_NUM = "total_server_num";
   private static final String RUNNING_APP_NUM = "running_app_num";
   private static final String TOTAL_APP_NUM = "total_app_num";
@@ -46,7 +47,8 @@ public class CoordinatorMetrics {
   public static final String REMOTE_STORAGE_IN_USED_PREFIX = "remote_storage_in_used_";
   public static final String APP_NUM_TO_USER = "app_num";
   public static final String USER_LABEL = "user_name";
-
+  public static Gauge gaugeLostServerNum;
+  public static Gauge gaugeActiveServerNum;
   public static Gauge gaugeTotalServerNum;
   public static Gauge gaugeExcludeServerNum;
   public static Gauge gaugeUnhealthyServerNum;
@@ -107,6 +109,8 @@ public static void updateDynamicGaugeForRemoteStorage(String storageHost, double
   }
 
   private static void setUpMetrics() {
+    gaugeLostServerNum = metricsManager.addGauge(LOST_SERVER_NUM);
+    gaugeActiveServerNum = metricsManager.addGauge(ACTIVE_SERVER_NUM);
     gaugeTotalServerNum = metricsManager.addGauge(TOTAL_SERVER_NUM);
     gaugeExcludeServerNum = metricsManager.addGauge(EXCLUDE_SERVER_NUM);
     gaugeUnhealthyServerNum = metricsManager.addGauge(UNHEALTHY_SERVER_NUM);