Show cluster healthy/unhealthy status on dashboard

Author: prakhar10

gateway-ha/src/main/java/io/trino/gateway/ha/domain/response/DistributionResponse.java

@@ -23,6 +23,8 @@ public class DistributionResponse
     private Integer totalBackendCount;
     private Integer offlineBackendCount;
     private Integer onlineBackendCount;
+    private Integer healthyBackendCount;
+    private Integer unhealthyBackendCount;
 
     /**
      * Total number of queries.
@@ -55,6 +57,28 @@ public class DistributionResponse
     private Map<String, List<LineChart>> lineChart;
     private String startTime;
 
+    @JsonProperty
+    public Integer getHealthyBackendCount()
+    {
+        return healthyBackendCount;
+    }
+
+    public void setHealthyBackendCount(Integer healthyBackendCount)
+    {
+        this.healthyBackendCount = healthyBackendCount;
+    }
+
+    @JsonProperty
+    public Integer getUnhealthyBackendCount()
+    {
+        return unhealthyBackendCount;
+    }
+
+    public void setUnhealthyBackendCount(Integer unhealthyBackendCount)
+    {
+        this.unhealthyBackendCount = unhealthyBackendCount;
+    }
+
     @JsonProperty
     public String getStartTime()
     {

gateway-ha/src/main/java/io/trino/gateway/ha/resource/GatewayWebAppResource.java

@@ -16,6 +16,7 @@
 import com.google.common.base.Strings;
 import com.google.inject.Inject;
 import io.trino.gateway.ha.clustermonitor.ClusterStats;
+import io.trino.gateway.ha.clustermonitor.TrinoStatus;
 import io.trino.gateway.ha.config.HaGatewayConfiguration;
 import io.trino.gateway.ha.config.ProxyBackendConfiguration;
 import io.trino.gateway.ha.config.RoutingRulesConfiguration;
@@ -160,6 +161,11 @@ public Response getDistribution(QueryDistributionRequest query)
         Map<String, String> urlToNameMap = allBackends
                 .stream().collect(Collectors.toMap(ProxyBackendConfiguration::getProxyTo, ProxyBackendConfiguration::getName, (o, n) -> n));
         Map<Boolean, List<ProxyBackendConfiguration>> activeMap = allBackends.stream().collect(Collectors.groupingBy(ProxyBackendConfiguration::isActive));
+        Map<Boolean, Integer> statusCounts = allBackends.stream()
+                .map(backendStateManager::getBackendState)
+                .collect(Collectors.partitioningBy(
+                        state -> state.trinoStatus() == TrinoStatus.HEALTHY,
+                        Collectors.collectingAndThen(Collectors.counting(), Long::intValue)));
         Integer latestHour = query.latestHour();
         Long ts = System.currentTimeMillis() - (latestHour * 60 * 60 * 1000);
         List<DistributionResponse.LineChart> lineChart = queryHistoryManager.findDistribution(ts);
@@ -179,6 +185,8 @@ public Response getDistribution(QueryDistributionRequest query)
         distributionResponse.setTotalBackendCount(allBackends.size());
         distributionResponse.setOfflineBackendCount(requireNonNullElse(activeMap.get(false), Collections.emptyList()).size());
         distributionResponse.setOnlineBackendCount(requireNonNullElse(activeMap.get(true), Collections.emptyList()).size());
+        distributionResponse.setHealthyBackendCount(statusCounts.getOrDefault(true, 0));
+        distributionResponse.setUnhealthyBackendCount(statusCounts.getOrDefault(false, 0));
         distributionResponse.setLineChart(lineChartMap);
         distributionResponse.setDistributionChart(distributionChart);
         distributionResponse.setTotalQueryCount(totalQueryCount);

webapp/src/components/dashboard.tsx

@@ -49,6 +49,14 @@ export function Dashboard() {
       key: Locale.Dashboard.BackendsOffline,
       value: distributionDetail?.offlineBackendCount
     },
+    {
+      key: Locale.Dashboard.BackendsHealthy,
+      value: distributionDetail?.healthyBackendCount,
+    },
+    {
+      key: Locale.Dashboard.BackendsUnhealthy,
+      value: distributionDetail?.unhealthyBackendCount
+    },
     {
       key: (
         <div className={styles.tip}>

webapp/src/locales/en_US.ts

@@ -12,6 +12,8 @@ const en_US = {
     Backends: "Backends",
     BackendsOffline: "Backends offline",
     BackendsOnline: "Backends online",
+    BackendsHealthy: "Backends healthy",
+    BackendsUnhealthy: "Backends unhealthy",
     StartTime: "Started at",
     Summary: "Summary",
     QueryDistribution: "Query distribution in last hour",

webapp/src/types/dashboard.d.ts

@@ -2,6 +2,8 @@ export interface  DistributionDetail {
   totalBackendCount: number;
   offlineBackendCount: number;
   onlineBackendCount: number;
+  healthyBackendCount: number;
+  unhealthyBackendCount: number;
   totalQueryCount: number;
   averageQueryCountMinute: number;
   averageQueryCountSecond: number;