Update utility functions to handle potential null values for at-risk and watchdog data to avoid eror for historyAlertStatus leading to all chart returning unavailable

Author: JeanMarcMilletScality

ui/src/services/platformlibrary/metrics.ts

@@ -1,6 +1,7 @@
 import {
   PORT_NODE_EXPORTER,
   STATUS_CRITICAL,
+  STATUS_HEALTH,
   STATUS_WARNING,
 } from '../../constants';
 import { queryPromtheusMetrics } from '../prometheus/fetchMetrics';
@@ -1186,6 +1187,16 @@ function convertSegmentToAlert(segment) {
         description:
           'Alerting services were unavailable during this period of time',
       };
+
+    case STATUS_HEALTH:
+      return {
+        ...baseSegment,
+        severity: 'healthy',
+        labels: {
+          alertname: 'NoAlerts',
+        },
+        description: 'The cluster is healthy',
+      };
   }
 }
 
@@ -1201,62 +1212,86 @@ export const getClusterAlertSegmentQuery = (duration: number) => {
     startingTimeISO,
     currentTimeISO,
     frequency,
-  }: TimeSpanProps) =>
-    queryPrometheusRange(
+  }: TimeSpanProps) => {
+    const prometheusPromise = queryPrometheusRange(
       startingTimeISO,
       currentTimeISO,
       frequency,
       encodeURIComponent(query),
-    )?.then((resolve) => {
-      if (resolve.status === 'error') {
-        throw resolve.error;
-      }
+    );
 
-      if (resolve.data.resultType !== 'matrix') {
-        throw new Error('Failed to fetch data from Prometheus');
-      }
+    if (!prometheusPromise) {
+      throw new Error(
+        'queryPrometheusRange returned null/undefined - Prometheus might not be available',
+      );
+    }
 
-      const clusterAtRiskResult = resolve.data.result.find(
-        (result) => result.metric.alertname === 'ClusterAtRisk',
-      ) || {
-        values: [],
-      };
+    return prometheusPromise
+      ?.then((resolve) => {
+        if (resolve.status === 'error') {
+          throw new Error(resolve.error);
+        }
 
-      const clusterDegradedResult = resolve.data.result.find(
-        (result) => result.metric.alertname === 'ClusterDegraded',
-      ) || {
-        values: [],
-      };
+        const clusterAtRiskResult = resolve.data.result.find(
+          // @ts-expect-error - FIXME when you are working on it
+          (result) => result.metric.alertname === 'ClusterAtRisk',
+        ) || {
+          values: [],
+        };
 
-      const watchdogResult = resolve.data.result.find(
-        (result) => result.metric.alertname === 'Watchdog',
-      ) || {
-        values: [],
-      };
-      const pointsAtRisk = addMissingDataPoint(
-        clusterAtRiskResult.values,
-        Date.parse(startingTimeISO) / 1000,
-        Date.parse(currentTimeISO) / 1000 - Date.parse(startingTimeISO) / 1000,
-        frequency,
-      );
-      const pointsDegraded = addMissingDataPoint(
-        clusterDegradedResult.values,
-        Date.parse(startingTimeISO) / 1000,
-        Date.parse(currentTimeISO) / 1000 - Date.parse(startingTimeISO) / 1000,
-        frequency,
-      );
-      const pointsWatchdog = addMissingDataPoint(
-        watchdogResult.values,
-        Date.parse(startingTimeISO) / 1000,
-        Date.parse(currentTimeISO) / 1000 - Date.parse(startingTimeISO) / 1000,
-        frequency,
-      );
-      return getSegments({
-        pointsDegraded,
-        pointsAtRisk,
-        pointsWatchdog,
-      }).map(convertSegmentToAlert);
-    });
+        const clusterDegradedResult = resolve.data.result.find(
+          // @ts-expect-error - FIXME when you are working on it
+          (result) => result.metric.alertname === 'ClusterDegraded',
+        ) || {
+          values: [],
+        };
+
+        const watchdogResult = resolve.data.result.find(
+          // @ts-expect-error - FIXME when you are working on it
+          (result) => result.metric.alertname === 'Watchdog',
+        ) || {
+          values: [],
+        };
+        const pointsAtRisk = addMissingDataPoint(
+          // @ts-expect-error - FIXME when you are working on it
+          clusterAtRiskResult.values,
+          Date.parse(startingTimeISO) / 1000,
+          Date.parse(currentTimeISO) / 1000 -
+            Date.parse(startingTimeISO) / 1000,
+          frequency,
+        );
+
+        const pointsDegraded = addMissingDataPoint(
+          // @ts-expect-error - FIXME when you are working on it
+          clusterDegradedResult.values,
+          Date.parse(startingTimeISO) / 1000,
+          Date.parse(currentTimeISO) / 1000 -
+            Date.parse(startingTimeISO) / 1000,
+          frequency,
+        );
+
+        const pointsWatchdog = addMissingDataPoint(
+          // @ts-expect-error - FIXME when you are working on it
+          watchdogResult.values,
+          Date.parse(startingTimeISO) / 1000,
+          Date.parse(currentTimeISO) / 1000 -
+            Date.parse(startingTimeISO) / 1000,
+          frequency,
+        );
+
+        const segments = getSegments({
+          pointsDegraded,
+          pointsAtRisk,
+          pointsWatchdog,
+        });
+        console.log('DEBUG: segments', segments);
+        return segments.map(convertSegmentToAlert);
+      })
+      .catch((error) => {
+        // Re-throw the error to see the original error in React Query
+        throw error;
+      });
+  };
 
   return {
     queryKey: ['clusterAlertsNumber', duration],
@@ -1279,6 +1314,7 @@ export const getClusterAlertSegmentQuery = (duration: number) => {
     refetchOnWindowFocus: false,
   };
 };
+
 export const prometheusKey = {
   query: (query) => ['query', query],
   queryRange: ['queryRange'],

ui/src/services/utils.ts

@@ -324,9 +324,10 @@ export function getNaNSegments(points: [number, number | string | null][]): {
 }
 export function getSegments({ pointsAtRisk, pointsDegraded, pointsWatchdog }) {
   return pointsDegraded.reduce((agg, [timestamp, degradedValue], index) => {
-    const atRiskValue = pointsAtRisk[index][1];
+    const atRiskValue = pointsAtRisk[index]?.[1] || 0;
+    const watchdogValue = pointsWatchdog[index]?.[1] || '0';
     const currentType =
-      pointsWatchdog[index][1] !== '1'
+      watchdogValue !== '1'
         ? NAN_STRING
         : atRiskValue > 0
         ? STATUS_CRITICAL