Tolerate missing metrics, use alternate metric from replica count (#475)

Signed-off-by: Michael Edgar <medgar@redhat.com>
streamshub · Feb 14, 2024 · cbf691d · cbf691d
1 parent 63a9fe3
commit cbf691d
Show file tree

Hide file tree

Showing 5 changed files with 41 additions and 33 deletions.
diff --git a/ui/api/kafka/kpi.promql.ts b/ui/api/kafka/kpi.promql.ts
@@ -65,7 +65,7 @@ or
 sum by (__console_metric_name__, nodeId) (
   label_replace(
     label_replace(
-      kafka_cluster_partition_replicascount{namespace="${namespace}",pod=~"${cluster}-.+-\\\\d+",strimzi_io_kind="Kafka"} > 0,
+      kafka_server_replicamanager_partitioncount{namespace="${namespace}",pod=~"${cluster}-.+-\\\\d+",strimzi_io_kind="Kafka"} > 0,
       "nodeId",
       "$1",
       "pod",

diff --git a/ui/api/kafka/schema.ts b/ui/api/kafka/schema.ts
@@ -59,28 +59,28 @@ export const ClusterResponse = z.object({
 export type ClusterDetail = z.infer<typeof ClusterDetailSchema>;
 
 export const ClusterKpisSchema = z.object({
-  broker_state: z.record(z.number()),
-  total_topics: z.number(),
-  total_partitions: z.number(),
-  underreplicated_topics: z.number(),
+  broker_state: z.record(z.number()).optional(),
+  total_topics: z.number().optional(),
+  total_partitions: z.number().optional(),
+  underreplicated_topics: z.number().optional(),
   replica_count: z.object({
-    byNode: z.record(z.number()),
-    total: z.number(),
-  }),
+    byNode: z.record(z.number()).optional(),
+    total: z.number().optional(),
+  }).optional(),
   leader_count: z.object({
-    byNode: z.record(z.number()),
-    total: z.number(),
-  }),
+    byNode: z.record(z.number()).optional(),
+    total: z.number().optional(),
+  }).optional(),
   volume_stats_capacity_bytes: z.object({
-    byNode: z.record(z.number()),
-    total: z.number(),
-  }),
+    byNode: z.record(z.number()).optional(),
+    total: z.number().optional(),
+  }).optional(),
   volume_stats_used_bytes: z.object({
-    byNode: z.record(z.number()),
-    total: z.number(),
-  }),
+    byNode: z.record(z.number()).optional(),
+    total: z.number().optional(),
+  }).optional(),
 });
 export type ClusterKpis = z.infer<typeof ClusterKpisSchema>;
 
-export const MetricRangeSchema = z.record(z.string(), z.record(z.number()));
+export const MetricRangeSchema = z.record(z.string(), z.record(z.number()).optional());
 export type MetricRange = z.infer<typeof MetricRangeSchema>;
diff --git a/ui/app/[locale]/kafka/[kafkaId]/nodes/page.tsx b/ui/app/[locale]/kafka/[kafkaId]/nodes/page.tsx
@@ -8,6 +8,10 @@ import {
 import { PageSection } from "@/libs/patternfly/react-core";
 import { redirect } from "@/navigation";
 
+function nodeMetric(metrics: Record<string, number> | undefined, nodeId: number): number {
+    return metrics ? (metrics[nodeId.toString()] ?? 0) : 0;
+}
+
 export default async function NodesPage({ params }: { params: KafkaParams }) {
   const res = await getKafkaClusterKpis(params.kafkaId);
   if (!res) {
@@ -20,11 +24,11 @@ export default async function NodesPage({ params }: { params: KafkaParams }) {
   }
 
   const nodes: Node[] = cluster.attributes.nodes.map((node) => {
-    const status = kpis.broker_state[node.id] === 3 ? "Stable" : "Unstable";
-    const leaders = kpis.leader_count.byNode[node.id];
-    const followers = kpis.replica_count.byNode[node.id] - leaders;
-    const diskCapacity = kpis.volume_stats_capacity_bytes.byNode[node.id];
-    const diskUsage = kpis.volume_stats_used_bytes.byNode[node.id];
+    const status = nodeMetric(kpis.broker_state, node.id) === 3 ? "Stable" : "Unstable";
+    const leaders = nodeMetric(kpis.leader_count?.byNode, node.id);
+    const followers = nodeMetric(kpis.replica_count?.byNode, node.id) - leaders;
+    const diskCapacity = nodeMetric(kpis.volume_stats_capacity_bytes?.byNode, node.id);
+    const diskUsage = nodeMetric(kpis.volume_stats_used_bytes?.byNode, node.id);
     return {
       id: node.id,
       status,

diff --git a/ui/app/[locale]/kafka/[kafkaId]/overview/ChartIncomingOutgoing.tsx b/ui/app/[locale]/kafka/[kafkaId]/overview/ChartIncomingOutgoing.tsx
@@ -18,8 +18,8 @@ import { useFormatter } from "next-intl";
 import { useChartWidth } from "./useChartWidth";
 
 type ChartIncomingOutgoingProps = {
-  incoming: Record<string, TimeSeriesMetrics>;
-  outgoing: Record<string, TimeSeriesMetrics>;
+  incoming: Record<string, TimeSeriesMetrics | undefined>;
+  outgoing: Record<string, TimeSeriesMetrics | undefined>;
 };
 
 type Datum = {
@@ -42,7 +42,7 @@ export function ChartIncomingOutgoing({
   const hasMetrics =
     Object.keys(incoming).length > 0 && Object.keys(outgoing).length > 0;
   if (!hasMetrics) {
-    return <div>TODO</div>;
+    return <div><i>Not available</i></div>;
   }
   // const showDate = shouldShowDate(duration);
   const CursorVoronoiContainer = createContainer("voronoi", "cursor");
@@ -124,7 +124,7 @@ export function ChartIncomingOutgoing({
         />
         <ChartGroup>
           {Object.entries(incoming).map(([name, entries], idx) => {
-            const entriesArray = Object.entries(entries);
+            const entriesArray = Object.entries(entries ?? {});
             return (
               <ChartArea
                 key={`incoming-line-${name}}`}
@@ -140,8 +140,8 @@ export function ChartIncomingOutgoing({
             );
           })}
           {Object.entries(outgoing).map(([name, entries], idx) => {
-            const entriesArray = Object.entries(entries);
-            const incomingArray = Object.keys(incoming[name]);
+            const entriesArray = Object.entries(entries ?? {});
+            const incomingArray = Object.keys(incoming[name] ?? {});
             return (
               <ChartArea
                 key={`outgoing-line-${name}}`}

diff --git a/ui/app/[locale]/kafka/[kafkaId]/overview/page.tsx b/ui/app/[locale]/kafka/[kafkaId]/overview/page.tsx
@@ -99,6 +99,10 @@ async function ConnectedTopicsPartitionsCard({
   );
 }
 
+function timeSeriesMetrics(ranges: Record<ClusterMetric, MetricRange> | undefined, rangeName: ClusterMetric) : TimeSeriesMetrics[] {
+    return ranges ? Object.values(ranges[rangeName] ?? {}).map(val => val ?? {}) : [];
+}
+
 async function ConnectedClusterChartsCard({
   data,
 }: {
@@ -112,10 +116,10 @@ async function ConnectedClusterChartsCard({
     <>
       <ClusterChartsCard
         isLoading={false}
-        usedDiskSpace={Object.values(res?.ranges["volumeUsed"] || {})}
-        availableDiskSpace={Object.values(res?.ranges["volumeCapacity"] || {})}
-        memoryUsage={Object.values(res?.ranges["memory"] || {})}
-        cpuUsage={Object.values(res?.ranges["cpu"] || {})}
+        usedDiskSpace={ timeSeriesMetrics(res?.ranges, "volumeUsed") }
+        availableDiskSpace={ timeSeriesMetrics(res?.ranges, "volumeCapacity") }
+        memoryUsage={ timeSeriesMetrics(res?.ranges, "memory") }
+        cpuUsage={ timeSeriesMetrics(res?.ranges, "cpu") }
       />
     </>
   );