Argo CD Cost Intelligence: Per-App Costs Next to Every Sync

Sync an application in Argo CD and you see at once whether it’s healthy, what changed and which revision is running. What you don’t see is whether that sync just made it more expensive. That shows up weeks later on the cloud bill, when nobody can tell which change caused it. Octopus Deploy’s new open source extension, Cost Intelligence for Argo CD, puts each application’s cost in a tab next to everything else.

What Happened

Octopus’s announcement explains why bills can’t answer the question. Cloud cost allocation can split a cluster by namespace or pod label, if someone turned it on, but it doesn’t know about Argo CD applications or syncs. Argo CD 3 tracks an application’s resources with an annotation by default, not a label, so there’s usually nothing to split by. Shared namespaces make it worse.

The plugin runs beside Argo CD and gets its numbers from OpenCost. After each successful sync, Argo CD Notifications calls it; it asks OpenCost for the application’s cost over the previous 24 hours and projects that onto a 30-day month. Figures cover only the pods of workloads Argo CD manages for that application, so apps sharing a namespace see only their own costs. Because each measurement looks back a day, you see a change’s full cost about a day after the sync. The plugin keeps each day’s cost itself, 400 days by default.

The tab has three views:

• Actual costs: projected monthly cost, change since last sync, idle spend on requested but unused CPU and memory, and a breakdown by CPU, memory, storage, network, load balancers and GPU. A What changed panel lists workloads whose cost moved. If pods use under half their requested CPU over a day, it suggests a smaller request, or shows Vertical Pod Autoscaler suggestions if you run it. If CI runs Infracost, out-of-cluster services like databases can be added.
• Resources: cost per workload and pod, suggested requests as YAML, and orphaned pods such as evicted or pending ones.
• Trends: daily cost over up to a year, with a month-end forecast after seven days of data.

Platform teams get a sidebar page with cost per namespace, apps ranked by idle spend, chargeback by any label and the control-plane fee, plus Prometheus metrics that join with argocd_app_info.

Why It Matters

Cost becomes part of the deploy feedback loop rather than a monthly forensic exercise. Octopus’s example: a release adds a second replica on Monday, and by the next sync the What changed panel shows the checkout workload’s cost before and after.

The limits are stated plainly. It measures only the cluster OpenCost and Argo CD both run in; apps deployed elsewhere show no costs. Without your provider’s prices, OpenCost uses public list prices. It’s an early 0.1.0 release whose Helm values and schema may change. And the platform page is visible to anyone who can view its Application, including role:readonly if that’s your default, so check access.

What This Means for Builders

It needs Argo CD 3.4 to 3.6, OpenCost 1.121.1 or newer and Helm 3.8 or newer, installs with one Helm chart plus PostgreSQL, and is Apache 2.0 licensed. A mock OpenCost lets you try the tab first. It isn’t affiliated with the Argo project or CNCF. Stack Archive has covered Octopus’s remote MCP server for agents.

Source: Octopus blog

Content created with AI assistance and reviewed for accuracy.