Add cost views and workload rightsizing by nadaverell · Pull Request #1101 · skyhook-io/radar

nadaverell · 2026-07-05T14:16:47Z

Summary

Adds OpenCost-backed cost visibility for clusters, applications, and workloads, plus an on-demand Rightsizing view for finding CPU and memory requests worth reviewing.

Workload Cost tabs for Deployments, StatefulSets, and DaemonSets.
Application Cost tabs with contributor drill-through to workload cost.
Shared historical allocation charts with consistent ranges, ticks, and hover details.
A Rightsizing scan across the currently visible namespaces, plus per-workload guidance.
Provider-aware cloud-console links when provider identity is reliable.

Rightsizing methodology

Rightsizing is advisory evidence, not an efficiency grade, automatic YAML change, capacity plan, or guaranteed savings estimate.

Uses seven days of 5-minute Prometheus samples.
CPU demand target: P95 usage plus 15% headroom.
Memory demand target: maximum observed working set plus 15% headroom.
Rounds to practical Kubernetes values with wider steps at larger scales.
Keeps CPU at or above 10m and memory at or above 64Mi.
Separates the demand-based target from a conservative next step.
Stages large reductions: memory and CPU requests of at least one core are reduced by at most 50% per review window; smaller CPU requests have proportional floors.
Applies the conservative reduction limit to bursty or throttled CPU workloads.
Keeps CPU and memory independent and surfaces HPA, OOM, throttling, scaled-to-zero, ownership, and history-quality signals.
Withholds unsafe conclusions when evidence or ownership is insufficient.

The fleet scan runs only when the user starts it. Queries are concurrency-bounded and use exact kube-state-metrics ownership; workload detail can fall back to exact current pods when historical ownership is unavailable.

Cost behavior

RBAC-preflighted workload and application OpenCost APIs.
Exact controller attribution: Pod → ReplicaSet → Deployment, or direct Pod ownership for StatefulSets and DaemonSets.
Application totals, stacked trends, contributor rows, partial-coverage messaging, and scaled-to-zero handling.
Monthly-first projections with hourly rates explaining the current composition.
Explicit unavailable states for missing Prometheus/OpenCost data and query failures.

Scope

Batch/job cost, storage/PVC attribution, network attribution, VPA integration, and automatic request changes remain out of scope. Capacity reductions are requested resources across replicas, not guaranteed reductions in the cloud bill.

Validation

make tsc
make test
make build
git diff --check
Focused Go and Vitest coverage for Rightsizing calculations, safety gates, scan presentation, and workload/application cost states.
Live GKE validation against real Prometheus and OpenCost data across 136 workloads.
Visual review at desktop and narrow widths, including loading, results, expanded guidance, chart hover behavior, and console cleanliness.
Screenshots: /Users/nadaverell/workspace/sky/ws10/radar/.playwright-mcp/visual-test/20260713-rightsizing-conservative/

Note

Medium Risk
Large new surface area (Prometheus-heavy rightsizing math, batched scans, OpenCost RBAC APIs) with behavior changes on discovery failures; mistakes could mis-rank recommendations or mis-report cost, but changes are mostly additive behind new routes and optional UI tabs.

Overview
Introduces OpenCost-backed cost at cluster-adjacent APIs and in the UI: optional Cost tabs on applications and workloads (Deployments/StatefulSets/DaemonSets), plus server routes for application and per-workload current cost and trends with RBAC preflight, partial coverage, and scaled-to-zero handling.

Rightsizing is rebuilt around a 7-day Prometheus window: rightsizing classification (balanced/oversized/under-requested, etc.), confidence and sample coverage, HPA/OOM/throttle signals, and conservative staged reduction rules. A new fleet scan (POST /prometheus/rightsizing/scan) batches workloads (50 at a time) with kube-state-metrics ownership vectors and namespace-scoped RBAC.

Prometheus client changes improve reliability: serialized discovery, a short TTL cache for repeated discovery failures, ErrPrometheusNotFound for empty discovery, and preserving cached connections when probes fail due to context cancellation. OpenCost handlers map connection errors to ReasonNoPrometheus vs ReasonQueryError, and node cost responses can include ProviderID from the node informer.

pkg/opencost adds application cost/trend aggregation and hardens Prom queries (deduped node/PV series, shared cost expr helpers). UI adds SelectMenu and workload/application hooks for host-rendered cost bodies.

^{Reviewed by Cursor Bugbot for commit ae29b5c. Bugbot is set up for automated code reviews on this repo. Configure here.}

# Conflicts: # packages/k8s-ui/src/components/applications/ApplicationDetail.test.tsx # packages/k8s-ui/src/components/applications/ApplicationDetail.tsx # packages/k8s-ui/src/components/workload/WorkloadView.tsx # web/src/components/applications/ApplicationsView.tsx

cursor

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit ae29b5c. Configure here.}

cursor · 2026-07-12T22:34:03Z

+		unique = append(unique, prom.EscapeRegexMeta(value))
+		prev = value
+	}
+	return strings.Join(unique, "|")


Unsafe workload names in trends

Medium Severity

Application cost trend queries build owner_name=~ alternations with EscapeRegexMeta only, unlike rightsizing scan helpers that also run SanitizeLabelValue. Workload names containing PromQL-special characters can break or skew trend queries.

Additional Locations (1)

pkg/opencost/application_trend.go#L139-L162

^{Reviewed by Cursor Bugbot for commit ae29b5c. Configure here.}

nadaverell requested a review from hisco as a code owner July 5, 2026 14:16

cursor Bot reviewed Jul 5, 2026

View reviewed changes

Comment thread web/src/components/cost/WorkloadCostTab.tsx

Comment thread web/src/components/cost/WorkloadCostTab.tsx Outdated

github-advanced-security AI found potential problems Jul 5, 2026

View reviewed changes

cursor Bot reviewed Jul 5, 2026

View reviewed changes

Comment thread web/src/components/cost/WorkloadCostTab.tsx

cursor Bot reviewed Jul 5, 2026

View reviewed changes

Comment thread web/src/api/client.ts

nadaverell added 4 commits July 10, 2026 02:27

Add workload cost tab

ab4bdcb

Fix workload cost trends with duplicate metrics

e76d12c

Improve OpenCost discovery UX

a62bc0b

Clarify workload cost metrics

9cf2aa5

nadaverell force-pushed the add-workload-cost-tab branch from 9fe6e3c to 9cf2aa5 Compare July 9, 2026 23:43

nadaverell added 2 commits July 10, 2026 02:46

Fix workload cost stale loading states

8995715

Add application cost tab

87f4acd

nadaverell changed the title ~~Add workload cost tab~~ Add workload and application cost tabs Jul 10, 2026

github-advanced-security AI found potential problems Jul 10, 2026

View reviewed changes

Comment thread internal/server/opencost_application.go Fixed

Comment thread internal/server/opencost_application.go Fixed

Comment thread pkg/opencost/application_trend.go Fixed

Comment thread pkg/opencost/application_trend.go Fixed

Polish application cost UX

9fa8ad2

cursor Bot reviewed Jul 10, 2026

View reviewed changes

Comment thread internal/server/opencost_application.go Outdated

Make cost views monthly-first

ae6ac0c

cursor Bot reviewed Jul 10, 2026

View reviewed changes

Comment thread web/src/api/client.ts

nadaverell added 3 commits July 10, 2026 23:43

Polish cost rows and static pod costs

1d30072

Add cloud console links to cost nodes

03cff90

Fix app cost partial review findings

cdcb372

cursor Bot reviewed Jul 11, 2026

View reviewed changes

Comment thread web/src/components/cost/ApplicationCostTab.tsx

nadaverell added 2 commits July 11, 2026 16:59

Handle unavailable app cost states

b5cc97f

Fix cost review edge cases

7d628c6

cursor Bot reviewed Jul 12, 2026

View reviewed changes

Comment thread web/src/api/client.ts Outdated

Polish cost rate presentation

cfd0666

cursor Bot reviewed Jul 12, 2026

View reviewed changes

Comment thread internal/prometheus/client.go

Comment thread pkg/opencost/workload_trend.go

Fix cost unavailable state handling

7f7f6fc

nadaverell force-pushed the add-workload-cost-tab branch from 5654415 to 7f7f6fc Compare July 12, 2026 09:32

Unify cost trend charts

e1f7a4d

nadaverell added 2 commits July 12, 2026 14:50

Replace cost efficiency grades with allocation use

110ec11

Add evidence-based workload request fit

2e3e9db

nadaverell changed the title ~~Add workload and application cost tabs~~ Add cost views and workload request fit Jul 12, 2026

cursor Bot reviewed Jul 12, 2026

View reviewed changes

Comment thread pkg/opencost/workloads.go Outdated

Comment thread internal/server/opencost_workload.go

Fix cost evidence availability states

0adca60

cursor Bot reviewed Jul 12, 2026

View reviewed changes

Comment thread internal/prometheus/rightsizing.go

Clarify workload request fit scope

32f53b4

cursor Bot reviewed Jul 12, 2026

View reviewed changes

Comment thread internal/opencost/handlers.go

Comment thread pkg/opencost/workloads.go

Comment thread pkg/opencost/application_cost.go Outdated

Handle incomplete cost metrics safely

d35d15d

cursor Bot reviewed Jul 12, 2026

View reviewed changes

Comment thread internal/prometheus/rightsizing.go Outdated

nadaverell added 2 commits July 12, 2026 17:54

Validate rounded request recommendations

a997f35

Add fleet workload request fit scan

4e3b5ac

nadaverell mentioned this pull request Jul 12, 2026

Add actionable workload rightsizing guidance #1154

Merged

nadaverell added 7 commits July 12, 2026 19:38

Preserve per-namespace HPA evidence

074e61a

Limit unknown HPA gate to reductions

a43c174

Turn request fit into actionable cost guidance

1cd5f4d

Make rightsizing recommendations conservative

fb260c9

Merge rightsizing scan into cost views

656ed09

Use rightsizing terminology consistently

ca91f0c

nadaverell changed the title ~~Add cost views and workload request fit~~ Add cost views and workload rightsizing Jul 12, 2026

cursor Bot reviewed Jul 12, 2026

View reviewed changes

Comment thread packages/k8s-ui/src/components/applications/ApplicationDetail.tsx

nadaverell added 2 commits July 13, 2026 01:19

Clear application cost selection on tab change

60f00c1

Rename rightsizing routes

ae29b5c

cursor Bot reviewed Jul 12, 2026

View reviewed changes

Uh oh!

Conversation

nadaverell commented Jul 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Rightsizing methodology

Cost behavior

Scope

Validation

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cursor Bot left a comment

Choose a reason for hiding this comment

Uh oh!

cursor Bot Jul 12, 2026

Choose a reason for hiding this comment

Unsafe workload names in trends

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

nadaverell commented Jul 5, 2026 •

edited

Loading