Costs

The Costs screen (route /costs) shows what you have actually spent, and what each declared game costs to run.

ECS + Fargate spend, per-game estimates, and trailing-window deltas.

The page splits cleanly in two, and the two halves come from different places:

• Total spend and the daily chart are actuals, pulled from AWS Cost Explorer.
• The per-game table is an estimate, computed from each game's declared CPU and memory against published Fargate rates. It is not billing data.

Both are fetched when you open the page; the page does not poll.

Range selector

Two buttons in the header: 7d and 30d, defaulting to 7d. Cost Explorer only exposes daily granularity, which is why there are no hourly options.

Total spend

The headline figure is your total ECS + Fargate spend across the selected window, titled Total spend · trailing 7 days.

The delta pill

Next to it, a pill compares that window against the window of the same length immediately before it. For a 7-day view, that is the previous 7 days; for 30d, the previous 30 days. The app fetches double the range in one call and splits it in half.

The percentage is computed against the prior window:

(current − prior) ÷ prior × 100

The pill shows the absolute change and percentage, e.g. ↓ $1.23 (14.7%) vs prior. The direction is carried by the arrow and colour: green with a down arrow means spend fell, red with an up arrow means it rose.

When there is no prior data to compare against, the pill reads no prior period.

The stacked daily chart

Daily spend, stacked by game

One column per day, each column split into a segment per game, with a legend in the card header. Hovering a segment shows the game and its value for that day.

The per-game split is an approximation

Cost Explorer returns daily totals only — no per-service or per-game breakdown. The chart divides each day's total evenly across your configured games, so every game shows exactly the same share of every day, regardless of whether it actually ran.

The page says so itself, under the chart:

Per-game split is a uniform approximation — Cost Explorer returns daily totals only.

Read the column heights as real data — those are genuine daily totals. Read the segments as decoration only. Do not use the chart to work out which game is expensive; use the estimates table below for that, or your AWS bill.

X-axis labels are formatted in UTC so timezones behind UTC do not display the previous calendar day. Over a 30-day range the labels get crowded and are truncated.

If the estimates call failed, the games list is empty — the legend disappears and the columns render with no segments at all, even though the totals are fine.

Per-game estimates

A table of what each game costs to run, computed from its declared Fargate CPU and memory.

Column	Contents
Game	Name, with its chart colour swatch
vCPU	e.g. 1 for 1024 CPU units
Memory	e.g. 4 GB
$/hour	Shown to four decimal places when under a dollar
$/day
$/month

Sort by clicking any column header; the default is $/hour descending. Clicking a new column sorts descending for numeric columns and ascending for Game; clicking the active column flips the direction.

The filter box in the card header (Filter games…) does a case-insensitive substring match on the game name.

On a narrow window the table becomes a stack of cards with the same six values. The card stack has no sort controls.

What the day and month figures assume

The footnote under the table states it plainly:

$/day assumes 24 hr/day. $/month assumes 4 hr/day × 30 days.

So:

Column	Formula
$/hour	vCPU × $0.04048 + GB × $0.004445
$/day	$/hour × 24
$/month	$/hour × 4 × 30 — that is 120 hours, not a full month

The two columns therefore answer different questions. $/day is the worst case if you leave a server up around the clock. $/month is a realistic monthly bill for a server used a few hours an evening — which is what the start/stop-on-demand design is built for. Neither is $/day × 30.

Two further caveats:

• These are Fargate compute rates only. They exclude EFS storage, data transfer, Route 53, Lambda, DynamoDB and CloudWatch. Your real bill will be higher.
• The rates are hardcoded us-east-1 on-demand prices. If you deploy to another region, the estimates will be off by that region's premium.

If a game's task definition cannot be read, the estimate falls back to 2 vCPU / 8 GB — with no indication in the table that it is a fallback. A row showing exactly 2 vCPU and 8 GB when you declared something else is the tell.

Estimates are fetched once and deliberately kept when you switch ranges — they do not depend on the date range, and refetching would mean an extra ECS call per game.

When there is nothing to show: No estimates available.

Requirements and failure modes

Cost Explorer must be enabled

Cost Explorer is opt-in per AWS account, and the app's IAM principal needs ce:* (included in the HyveonDeployAll policy — see the setup guide). There is also a lag: Cost Explorer data is not real-time, so today's figure will be incomplete and very recent activity may not appear at all.

Activate the Project cost-allocation tag

Every resource Hyveon creates is tagged Project=hyveon. Activate the Project tag under AWS Billing → Cost allocation tags so you can break your bill down by project in Cost Explorer and in the AWS console.

Do this even though this page does not use it: the actuals shown here are not scoped by that tag. The query filters on the service dimension — Amazon Elastic Container Service and AWS Fargate — so if you run other ECS or Fargate workloads in the same account, their spend is included in these figures. Activating the tag is how you get a Hyveon-only view, in the AWS console.

What you see when it is unavailable

If Cost Explorer errors or returns nothing, this page shows:

Element	Shows
Total spend	$0.00
Delta pill	no prior period
Next to the pill	A red Cost Explorer: <error> note with the raw AWS message
Chart	No cost data available.
Estimates table	Unaffected — it is a separate call

The $0.00 is a display artefact of the failure, not a measurement. The red error note next to it is the real signal.

The dashboard handles the same failure differently: its Spend today and Forecast MTD tiles show an em-dash rather than $0.00, precisely to avoid implying you spent nothing.