fix(cloud-agent-next): show current health outcomes

eshurakov · eshurakov · commit 5a090eb03bc4 · 2026-05-29T10:18:30.000+02:00
diff --git a/apps/web/src/app/admin/api/cloud-agent-next/hooks.ts b/apps/web/src/app/admin/api/cloud-agent-next/hooks.ts
@@ -38,6 +38,8 @@ export function useCloudAgentNextHealthOverview(
   return useQuery({
     ...trpc.admin.cloudAgentNext.getHealthOverview.queryOptions(params),
     enabled: enabled && enabledForInterval(params),
+    refetchOnReconnect: false,
+    refetchOnWindowFocus: false,
   });
 }
 
diff --git a/apps/web/src/app/admin/components/CloudAgentNextTelemetry/CloudAgentNextOutcomesPage.tsx b/apps/web/src/app/admin/components/CloudAgentNextTelemetry/CloudAgentNextOutcomesPage.tsx
@@ -42,6 +42,7 @@ import {
 } from '@/components/ui/select';
 import { Skeleton } from '@/components/ui/skeleton';
 import { cn } from '@/lib/utils';
+import { rollingHealthInterval } from './health-interval';
 import { getOperationalFailureStats } from './health-summary';
 import {
   DEFAULT_HEALTH_PERIOD,
@@ -70,11 +71,11 @@ type RangeOption = {
 };
 
 const RANGE_OPTIONS = [
-  { value: '1h', label: 'Last complete hour', durationMs: 60 * 60 * 1000, bucket: 'hour' },
-  { value: '3h', label: 'Last 3 complete hours', durationMs: 3 * 60 * 60 * 1000, bucket: 'hour' },
+  { value: '1h', label: 'Last hour', durationMs: 60 * 60 * 1000, bucket: 'hour' },
+  { value: '3h', label: 'Last 3 hours', durationMs: 3 * 60 * 60 * 1000, bucket: 'hour' },
   {
     value: '24h',
-    label: 'Last 24 complete hours',
+    label: 'Last 24 hours',
     durationMs: 24 * 60 * 60 * 1000,
     bucket: 'hour',
   },
@@ -131,13 +132,8 @@ function intervalForRange(
 ): CloudAgentNextHealthFilters {
   const selectedRange = RANGE_OPTIONS.find(option => option.value === range) ?? RANGE_OPTIONS[3];
   const createdOnPlatform = createdOnPlatformForSelection(platformSelection);
-  const end = new Date();
-  if (selectedRange.bucket === 'day') end.setUTCHours(0, 0, 0, 0);
-  else end.setUTCMinutes(0, 0, 0);
   return {
-    startDate: new Date(end.getTime() - selectedRange.durationMs).toISOString(),
-    endDate: end.toISOString(),
-    bucket: selectedRange.bucket,
+    ...rollingHealthInterval(selectedRange),
     ...(createdOnPlatform === undefined ? {} : { createdOnPlatform }),
   };
 }
@@ -206,22 +202,16 @@ function DashboardSkeleton() {
   );
 }
 
-function HealthSummary({
-  summary,
-  bucket,
-}: {
-  summary: HealthData['summary'];
-  bucket: HealthBucket;
-}) {
+function HealthSummary({ summary }: { summary: HealthData['summary'] }) {
   const operationalFailures = getOperationalFailureStats(summary);
   const failureRate = operationalFailures.failureRatePercent;
   return (
     <Card>
       <CardHeader>
         <CardTitle>Observed health</CardTitle>
         <CardDescription>
-          Events observed during complete UTC {bucket === 'day' ? 'days' : 'hours'} in the selected
-          period. Interruptions are excluded from failure rate.
+          Events observed in the selected rolling period. Interruptions are excluded from failure
+          rate.
         </CardDescription>
       </CardHeader>
       <CardContent className="grid gap-3 sm:grid-cols-2 xl:grid-cols-5">
@@ -304,8 +294,8 @@ function OutcomeTrendChart({
       <CardHeader>
         <CardTitle>{label} outcomes</CardTitle>
         <CardDescription>
-          Completed, failed, setup-failed, and interrupted events in complete UTC-
-          {bucket === 'day' ? 'day' : 'hour'} buckets.
+          Completed, failed, setup-failed, and interrupted events in UTC-
+          {bucket === 'day' ? 'day' : 'hour'} buckets. Edge buckets may be partial.
         </CardDescription>
       </CardHeader>
       <CardContent>
@@ -646,8 +636,8 @@ export default function CloudAgentNextOutcomesPage() {
           </div>
         </div>
         <p className="text-muted-foreground text-xs">
-          Reporting is best-effort, so totals can undercount execution. Periods use complete UTC
-          {bucket === 'day' ? ' days' : ' hours'}.
+          Reporting is best-effort, so totals can undercount execution. Periods end at refresh time;
+          edge UTC {bucket === 'day' ? 'days' : 'hours'} may be partial.
         </p>
         {health.error && (
           <Alert variant="destructive">
@@ -665,7 +655,7 @@ export default function CloudAgentNextOutcomesPage() {
           <DashboardSkeleton />
         ) : health.data ? (
           <>
-            <HealthSummary summary={health.data.summary} bucket={bucket} />
+            <HealthSummary summary={health.data.summary} />
             <OutcomeTrendChart data={health.data.series} range={range} bucket={bucket} />
             <TopErrors errors={health.data.topErrors} interval={interval} />
           </>
diff --git a/apps/web/src/app/admin/components/CloudAgentNextTelemetry/health-interval.test.ts b/apps/web/src/app/admin/components/CloudAgentNextTelemetry/health-interval.test.ts
@@ -0,0 +1,16 @@
+import { rollingHealthInterval } from './health-interval';
+
+describe('rollingHealthInterval', () => {
+  it('ends an hourly range at the exact refresh time', () => {
+    expect(
+      rollingHealthInterval(
+        { durationMs: 3 * 60 * 60 * 1000, bucket: 'hour' },
+        new Date('2035-01-10T12:34:56.789Z')
+      )
+    ).toEqual({
+      startDate: '2035-01-10T09:34:56.789Z',
+      endDate: '2035-01-10T12:34:56.789Z',
+      bucket: 'hour',
+    });
+  });
+});
diff --git a/apps/web/src/app/admin/components/CloudAgentNextTelemetry/health-interval.ts b/apps/web/src/app/admin/components/CloudAgentNextTelemetry/health-interval.ts
@@ -0,0 +1,20 @@
+export type HealthBucket = 'hour' | 'day';
+
+type HealthRange = {
+  durationMs: number;
+  bucket: HealthBucket;
+};
+
+type HealthInterval = {
+  startDate: string;
+  endDate: string;
+  bucket: HealthBucket;
+};
+
+export function rollingHealthInterval(range: HealthRange, now = new Date()): HealthInterval {
+  return {
+    startDate: new Date(now.getTime() - range.durationMs).toISOString(),
+    endDate: now.toISOString(),
+    bucket: range.bucket,
+  };
+}
diff --git a/apps/web/src/routers/admin-cloud-agent-next-router.test.ts b/apps/web/src/routers/admin-cloud-agent-next-router.test.ts
@@ -361,22 +361,90 @@ describe('adminCloudAgentNextRouter', () => {
     ]);
   });
 
-  it('rejects health intervals that do not align to UTC bucket boundaries', async () => {
+  it('summarizes rolling health intervals across partial daily UTC buckets', async () => {
     const caller = await createCallerForUser(adminUser.id);
-    await expect(
-      caller.admin.cloudAgentNext.getHealthOverview({
-        startDate: at(0, 30),
-        endDate: at(1, 30),
-        bucket: 'hour',
-      })
-    ).rejects.toMatchObject({ code: 'BAD_REQUEST' });
-    await expect(
-      caller.admin.cloudAgentNext.getHealthOverview({
-        startDate: START_DATE,
-        endDate: at(23),
-        bucket: 'day',
-      })
-    ).rejects.toMatchObject({ code: 'BAD_REQUEST' });
+    const health = await caller.admin.cloudAgentNext.getHealthOverview({
+      startDate: '2035-01-09T12:00:00.000Z',
+      endDate: at(5, 30),
+      bucket: 'day',
+    });
+
+    expect(health.series).toEqual([
+      {
+        bucketStart: '2035-01-09T00:00:00.000Z',
+        completedRuns: 0,
+        failedRuns: 0,
+        interruptedRuns: 0,
+        setupFailures: 0,
+      },
+      {
+        bucketStart: START_DATE,
+        completedRuns: 1,
+        failedRuns: 2,
+        interruptedRuns: 1,
+        setupFailures: 2,
+      },
+    ]);
+  });
+
+  it('summarizes rolling health intervals across partial hourly UTC buckets', async () => {
+    const caller = await createCallerForUser(adminUser.id);
+    const health = await caller.admin.cloudAgentNext.getHealthOverview({
+      startDate: at(0, 30),
+      endDate: at(5, 30),
+      bucket: 'hour',
+    });
+
+    expect(health.summary).toEqual({
+      completedRuns: 1,
+      failedRuns: 2,
+      interruptedRuns: 1,
+      setupFailures: 1,
+    });
+    expect(health.series).toEqual([
+      {
+        bucketStart: at(0),
+        completedRuns: 0,
+        failedRuns: 0,
+        interruptedRuns: 0,
+        setupFailures: 0,
+      },
+      {
+        bucketStart: at(1),
+        completedRuns: 1,
+        failedRuns: 0,
+        interruptedRuns: 0,
+        setupFailures: 0,
+      },
+      {
+        bucketStart: at(2),
+        completedRuns: 0,
+        failedRuns: 1,
+        interruptedRuns: 0,
+        setupFailures: 0,
+      },
+      {
+        bucketStart: at(3),
+        completedRuns: 0,
+        failedRuns: 1,
+        interruptedRuns: 0,
+        setupFailures: 0,
+      },
+      {
+        bucketStart: at(4),
+        completedRuns: 0,
+        failedRuns: 0,
+        interruptedRuns: 1,
+        setupFailures: 0,
+      },
+      {
+        bucketStart: at(5),
+        completedRuns: 0,
+        failedRuns: 0,
+        interruptedRuns: 0,
+        setupFailures: 1,
+      },
+    ]);
   });
 
   it('lists affected sessions for an exact top-error source and occurrence interval', async () => {
diff --git a/apps/web/src/routers/admin-cloud-agent-next-router.ts b/apps/web/src/routers/admin-cloud-agent-next-router.ts
@@ -38,22 +38,6 @@ function hasBoundedInterval(input: { startDate: string; endDate: string }) {
   return new Date(input.endDate).getTime() - new Date(input.startDate).getTime() <= MAX_INTERVAL_MS;
 }
 
-function hasAlignedBucketBounds(input: {
-  startDate: string;
-  endDate: string;
-  bucket: z.infer<typeof healthBucketSchema>;
-}) {
-  const dates = [new Date(input.startDate), new Date(input.endDate)];
-  return dates.every(date =>
-    input.bucket === 'day'
-      ? date.getUTCHours() === 0 &&
-        date.getUTCMinutes() === 0 &&
-        date.getUTCSeconds() === 0 &&
-        date.getUTCMilliseconds() === 0
-      : date.getUTCMinutes() === 0 && date.getUTCSeconds() === 0 && date.getUTCMilliseconds() === 0
-  );
-}
-
 const HealthOverviewFilterSchema = z
   .object({
     ...intervalShape,
@@ -67,10 +51,6 @@ const HealthOverviewFilterSchema = z
   .refine(input => hasBoundedInterval(input), {
     message: 'Date interval cannot exceed 90 days',
     path: ['endDate'],
-  })
-  .refine(input => hasAlignedBucketBounds(input), {
-    message: 'Date interval must align to UTC bucket boundaries',
-    path: ['startDate'],
   });
 const HealthErrorSessionsFilterSchema = z
   .object({
@@ -173,11 +153,13 @@ type HealthError = {
 };
 
 function emptyHealthSeries(input: HealthOverviewFilter): HealthSeriesPoint[] {
-  const start = new Date(input.startDate).getTime();
+  const firstBucket = new Date(input.startDate);
+  if (input.bucket === 'day') firstBucket.setUTCHours(0, 0, 0, 0);
+  else firstBucket.setUTCMinutes(0, 0, 0);
   const end = new Date(input.endDate).getTime();
   const bucketMs = input.bucket === 'day' ? DAY_MS : HOUR_MS;
   const series: HealthSeriesPoint[] = [];
-  for (let timestamp = start; timestamp < end; timestamp += bucketMs) {
+  for (let timestamp = firstBucket.getTime(); timestamp < end; timestamp += bucketMs) {
     series.push({
       bucketStart: new Date(timestamp).toISOString(),
       completedRuns: 0,

Original file line number	Diff line number	Diff line change
`@@ -38,6 +38,8 @@ export function useCloudAgentNextHealthOverview(`
`38`	`38`	`return useQuery({`
`39`	`39`	`...trpc.admin.cloudAgentNext.getHealthOverview.queryOptions(params),`
`40`	`40`	`enabled: enabled && enabledForInterval(params),`
	`41`	`+ refetchOnReconnect: false,`
	`42`	`+ refetchOnWindowFocus: false,`
`41`	`43`	`});`
`42`	`44`	`}`
`43`	`45`