Skip to content

Chargeback by team, label or namespace

Chargeback is the Cost tab’s one cost table: Cost by namespace, workload, team or label on FinOps → Cost & savings → Cost. It answers “what did each namespace, or each team, spend?” for one cluster over the last hour, 24 hours, 7 days or 30 days (7 days to start). It reads the allocation the cost engine computes in your cluster (OpenCost, in the seylo-observability chart), through the agent, and shows the split a finance sheet uses, with efficiency, and lets you download the same rows as CSV.

Chargeback asks OpenCost’s own API, which is new in two places at once:

Component Version Why
The Seylo agent 0.2.11 or later The agent is what asks OpenCost, from inside your cluster
The seylo-observability chart 0.2.11 or later, with OpenCost on Earlier charts let only VictoriaMetrics reach OpenCost’s port

Upgrade the agent (Helm 3.14 or later). When the agent is too old, the card shows:

Terminal window
helm upgrade seylo-agent oci://ghcr.io/seylo-inc/charts/seylo-agent --namespace seylo-agent --reset-then-reuse-values --version 0.2.22

If your install ever set image.tag or image.digest — every upgrade command on this site does — --reset-then-reuse-values keeps the old ones, so pass the new ones too:

Terminal window
helm upgrade seylo-agent oci://ghcr.io/seylo-inc/charts/seylo-agent \
--version 0.2.22 --namespace seylo-agent --reset-then-reuse-values \
--set image.tag=0.2.22 \
--set image.digest=sha256:ed9222f8b6e0beb0fb12873388439ce270304bc5b54dc41a80a96707c552661f

Upgrade the chart, turning OpenCost on if it is not already:

Terminal window
helm upgrade seylo-observability oci://ghcr.io/seylo-inc/charts/seylo-observability --version 0.2.22 --namespace seylo-observability --reset-then-reuse-values --set opencost.enabled=true

On Helm older than 3.14, --reset-then-reuse-values does not exist. Save the values you set and pass them back as a file instead:

Terminal window
helm get values seylo-agent -n seylo-agent -o yaml > agent-values.yaml
helm upgrade seylo-agent oci://ghcr.io/seylo-inc/charts/seylo-agent \
--version 0.2.22 --namespace seylo-agent -f agent-values.yaml \
--set image.tag=0.2.22 \
--set image.digest=sha256:ed9222f8b6e0beb0fb12873388439ce270304bc5b54dc41a80a96707c552661f
helm get values seylo-observability -n seylo-observability -o yaml > obs-values.yaml
helm upgrade seylo-observability oci://ghcr.io/seylo-inc/charts/seylo-observability \
--version 0.2.22 --namespace seylo-observability -f obs-values.yaml \
--set opencost.enabled=true

Do not use --reuse-values: it keeps the old chart’s defaults, so a block the new chart added renders empty. See upgrade the agent.

Until both are in place the table says which is missing:

You see Cause Do
Cost by namespace needs agent 0.2.11 and the 0.2.11 observability chart, The agent on "<cluster>" is version <version>; Seylo's cost figures come from OpenCost's API, which needs agent 0.2.11 and seylo-observability chart 0.2.11 or later. Run both upgrades on that cluster, in this order: The agent is older Run both upgrades, in the order shown
Cost by namespace needs agent 0.2.11 and the 0.2.11 observability chart, The agent could not reach OpenCost: the seylo-observability chart on this cluster predates 0.2.11, whose NetworkPolicy closes OpenCost's port to everything but VictoriaMetrics. Run both upgrades on that cluster, in this order: The agent is new enough, the chart is not Run both upgrades, in the order shown
No cost by namespace for this cluster, OpenCost is not running in this cluster. Turn it on with: OpenCost is off Run the chart command above

Choose up to three levels, each one of:

  • Namespace
  • Workload — OpenCost’s controller: the Deployment, StatefulSet, Job and so on a pod belongs to
  • Label key — the card offers the label keys your cluster’s pods actually carry, team-like keys first; Another label… takes any key, such as app.kubernetes.io/part-of
  • An annotation… — any annotation key, such as owner

The table opens grouped by namespace alone, the question most people open it with; the cluster’s team-like label keys are listed first in the picker, one choice away. Pick a key another level already uses and the two swap, so “namespace, then team” becomes “team, then namespace” in one step. Clearing a level clears the levels after it.

Column What it is
CPU, Memory, GPU What the rows were allocated, priced as described below
Storage, Network, Load balancer Always a dash here, with the reason: OpenCost’s own rates for these are defaults unless a billing integration prices them. Storage, load balancers and network, under the table, prices disks, public IPs and classified egress from Seylo’s price book. See the cost table
Shared Shared cost OpenCost attributes. Shown only when every node’s price is trusted
Idle The cost of capacity nobody requested: its own row, or shared into the rows (below)
Total The priced part of the row. “+ not priced” beside it means part of the row could not be priced; hover it for why
Efficiency Usage divided by request over the window, as three figures: CPU / memory / both, weighted by their cost

The totals add only what is priced. A dash is never zero, and always carries its reason on hover.

Share idle spreads the cost of capacity nobody requested across the rows, in proportion to each row’s cost, so the rows add up to what the nodes cost. Off, idle is its own row and the rows show only what was allocated.

With it on, each row’s CPU, memory and GPU include its share of idle, and there is no separate idle row. OpenCost does not report the share separately, so Seylo does not split it out with an estimate: the Idle column says “Shared into CPU, memory and GPU.”

Use it off to see what each team asked for; on to recharge the whole bill.

Share idle is offered only to people who see the whole cluster, because idle is the whole cluster’s cost.

Below the table, Where the prices come from names each source. Every node’s CPU, memory and GPU dollars are judged against Seylo’s price book before they are shown, and get one of three treatments:

Label When
Trusted — “OpenCost, agreeing with …” OpenCost’s node price agrees with the price book within 1%
Trusted — “OpenCost, from your Azure Price Sheet (your negotiated rates)” OpenCost reports your Azure price sheet as enabled, available and without error. See make the figures match your invoice
Rescaled — “OpenCost’s split, rescaled from its $a/h to $b/h …” The book has a rate and OpenCost’s differs. The split between CPU, memory and GPU stays OpenCost’s; the total is the book’s
Unpriced, with the reason No rate in the book, a spot node, no cloud provider ID, or no price from OpenCost for the node

When OpenCost has no customer price source, the line says so, for example OpenCost reports no customer price source (no billing integration), so its prices are list or built-in default prices.

Spot nodes and nodes Seylo refuses for their shape stay unpriced even with a working price sheet.

With your Azure bill connected, the book’s rate for an Azure machine type is what your bill says it cost, and it is named beside the figure as your Azure bill (Cost Management, amortised, in US dollars as Azure states them).

Download CSV saves exactly the rows and totals on screen, generated by the server from the same answer: the same columns and rounding, the price source and the reason on each row, and empty cells — never 0 — where a figure is not priced. Label values that a spreadsheet would run as a formula are defused. The file name carries the cluster, the window, the grouping and whether idle was shared.

OpenCost’s API has no notion of who is asking: it answers for every namespace to anyone who reaches it. So for someone whose access is limited to some namespaces, Seylo narrows the answer twice:

  • the request asks OpenCost only for those namespaces, and
  • every row is checked in the agent, before it leaves your cluster, and again on Seylo’s server. A row that does not belong to one of those namespaces is dropped, never shown, and not counted on screen either.

What such a person sees:

  • their namespaces only, with a note above the totals saying so;
  • a team’s figure covers only their namespaces. A team label that also runs in namespaces they cannot see shows only its cost in theirs. Seylo groups by team itself, after the namespace check, for exactly this reason;
  • no idle row, no unattributed row, and no share idle: those are the whole cluster’s cost.

The cost_allocation tool takes the same grouping as aggregate (for example label:team,namespace) and shareIdle, and goes through the same checks under the token’s own access.