Estimated Billing

AWSim ships with a billing meter that watches every request flowing through the gateway, multiplies usage by canonical AWS pricing, and surfaces a rolling estimated monthly bill at /billing in the admin console.

The dashboard answers questions like:

• How much would this workload cost on real AWS?
• Which service is dominating my bill?
• At my current burn rate, how long until I hit my budget?
• Is my cost trajectory accelerating, flat, or trending down?

It's a projection, not a contract — see Approximations below.

Quick start

Open the admin console (cd ui && bun run dev, then http://localhost:5173/billing). Hit some AWS APIs and watch the bill tick up:

export AWS_ENDPOINT_URL=http://localhost:4566
aws s3 mb s3://test-bucket
for i in $(seq 1 1000); do
  aws s3api put-object --bucket test-bucket --key obj-$i --body /etc/hostname
done

The Cost Trajectory panel shows the running bill over the last 30 minutes, the Per-Service Breakdown lists each service's contribution, and the Time to Bankruptcy widget projects when your set budget would be exhausted at the current rate.

What's metered

Axis	Description	Services
Per-request	Flat $/call rate (PUT, GET, etc.)	S3, DynamoDB, SQS, SNS, KMS, Secrets Manager, EventBridge, API Gateway, SES, CloudWatch Metrics, Route 53, Kinesis, CloudFront, Cognito
Per state-transition	$0.000025 / transition	Step Functions
Per character	$/character of input text	Polly, Comprehend
Data transfer out	$0.09/GB egress to internet	All metered services
Data ingest	$/GB of incoming payload	Firehose, CloudWatch Logs
Storage (GB-month)	$/GB stored, sampled point-in-time	S3, DynamoDB, ECR, CloudWatch Logs
Compute (GB-second)	duration × memory × rate	Lambda
Instance-hours	Running instances × hourly rate	EC2, RDS

Pricing data lives in crates/awsim-billing/pricing/<service>.json, vendored from the AWS Pricing Bulk JSON. Most files are populated by a refresh tool (see Refreshing pricing); a few (EC2, RDS, Polly, Comprehend) are hand-written because their AWS rates are per-instance-type or per-100-character units that need translation.

How the meter works

The gateway publishes a RequestEvent for every request it handles. A spawned tokio task in awsim-billing subscribes to the broadcast channel and routes each event to the right counter:

1. Per-request services — OpCounter::record(units, bytes_in, bytes_out, error) increments the operation counter. units is normally 1, but services can override it via metadata response headers (see Per-call metadata).
2. Storage / instance-hour services — a separate 30-second poll loop in the awsim binary samples on-disk byte counts (S3 BodyStore, DynamoDB SQLite file) and running instance counts (EC2/RDS) and feeds them into trapezoidal-integration accumulators.
3. Compute (GB-seconds) — Lambda emits X-Awsim-Memory-MB on its response headers; the meter multiplies that × event.duration_ms × the rate.

All accumulated cost is stored in pico-USD (1e-12 USD) integer atomics — coarse-grained units truncated tiny per-sample accruals to zero.

Per-call metadata

Services can attach X-Awsim-* response headers that the gateway peels off before the response leaves the building. The billing meter reads them off the resulting RequestEvent:

Header	Type	Used by
X-Awsim-Memory-MB	u32	Lambda compute (GB-seconds)
X-Awsim-State-Transitions	u32	Step Functions per-transition billing
X-Awsim-Char-Count	u64	Polly / Comprehend per-character billing

If a service grows a new billable axis, the typical fix is: have the handler emit a header with the right count, optionally extend RequestEvent, then read it off in the meter.

Persistence

When --data-dir is set, the billing state survives restarts via data/snapshots/billing.json. The poll loop also fires inside this block (storage / instance counts can't be sampled without data-dir). In-memory mode still meters per-request charges but skips storage / instance-hour billing.

Refreshing pricing

AWS publishes pricing as huge per-service JSON files at pricing.us-east-1.amazonaws.com. The billing crate ships a feature-gated refresh binary that pulls the relevant SKUs and writes the slim per-service files we vendor:

cargo run -p awsim-billing --bin refresh-pricing --features refresh

Run it after AWS bumps a rate, then commit the diff under crates/awsim-billing/pricing/. The tool handles the per-region offer files (most services), the global bulk file (CloudFront), and the AWSDataTransfer offer (egress rates). Hand-written files for EC2 / RDS / Polly / Comprehend stay untouched.

Approximations

The bill is honest about what it does not model precisely:

• Lambda memory defaults to 128 MB when an invocation didn't carry the function's configured memory through. Functions on bigger memory tiers underbill in proportion.
• Step Functions transitions are counted per StateEntered event in the ASL interpreter's history. Parallel/Map state branches may slightly over- or under-count vs. real AWS.
• EC2/RDS instance-hours charge at a baseline rate (t3.micro / db.t3.micro) regardless of the actual instance type. AWSim's request event doesn't carry the instance type through. Workloads on bigger types underbill in proportion.
• Polly charges at the Standard voice tier ($4 / million chars). Neural ($16 / M), Long-form ($100 / M), and Generative ($30 / M) tiers underbill.
• Cognito MAU is informational — the per-MAU rate appears on the dashboard but count stays at 0 because AWSim doesn't track unique-principal activity over time yet.
• Athena per-TB scanned is not metered — AWSim's Athena emulator returns SUCCEEDED immediately without scanning any data.

Admin endpoint

GET /_awsim/billing returns the JSON the dashboard renders:

{
  "currency": "USD",
  "elapsed_secs": 423,
  "running_cost_usd": 0.00128,
  "projected_monthly_cost_usd": 7.85,
  "services": [
    {
      "service": "s3",
      "display_name": "Amazon S3",
      "region": "us-east-1",
      "total_cost_usd": 0.00075,
      "request_count": 150,
      "storage_cost_usd": 0.00012,
      "storage_bytes": 5_242_880,
      "dimensions": [
        {
          "description": "PUT/COPY/POST/LIST requests",
          "price_per_request": 0.000005,
          "request_count": 150,
          "cost_usd": 0.00075
        },
        ...
      ]
    },
    ...
  ]
}

Useful for piping into other dashboards, alerting, or scripted budget checks.