Skip to content

Example 12 — Infrastructure Cost Modeling

File: files/examples/cost-modeling.toml Industry: Engineering / Finance / FinOps Tags: finops, cost, modeling, deployment

Features Demonstrated

  • register — captures cost estimates, actual measurements, and computed deltas per environment
  • if conditional — gates deployment on budget approval and triggers variance alert conditionally
  • matrix — runs cost estimation and post-deployment measurement across three environments (dev/staging/prod) in parallel
  • on_failure (task-level) — wires a forensic snapshot task to deployment, preserving cost state for rollback analysis
  • retries + retry_delay — cloud cost API calls may be transient; all measurement tasks retry automatically
  • timeout — cost API and Terraform calls are time-bounded
  • ignore_failure — cost reporting tasks must not block the deployment record
  • env vars — AWS region, cost granularity, FinOps dashboard URL, budget threshold

Why This Pattern Matters

Infrastructure deployments change your cost structure. Without explicit cost modeling in the deployment workflow itself, teams routinely discover cost surprises in the next billing cycle — after it's too late to act. Embedding cost estimation and post-deployment measurement directly into the deployment workflow means:

  • Budget approval happens automatically, based on actual IaC plan output
  • Actual vs estimated variance is measured and stored with the run record
  • Finance and engineering are alerted to overruns immediately, not at month-end
  • The full cost timeline is part of the audit trail alongside the technical deployment record

Pipeline Structure

[snapshot-cost-baseline]
         ↓
[estimate-deployment-cost] × {dev, staging, prod}  ← matrix: IaC plan per env
         ↓
[check-budget-threshold]   ← register: budget_decision ("approved" / "requires_approval")
         ↓  (if budget_decision == "approved")
[deploy-infrastructure]    ← on_failure → [preserve-cost-snapshot]
         ↓
[measure-actual-cost] × {dev, staging, prod}        ← matrix: actual cost delta per env
         ↓
[compute-variance]         ← register: variance_status
         ↓  (if variance_status == "over_threshold")
[alert-cost-variance]
         ↓
[publish-finops-report]

[preserve-cost-snapshot]   ← forensic, fires only if deployment fails

Run Commands

# Standard run (work-stealing for fastest cost API parallelism)
wf run cost-modeling --work-stealing --print-output

# Override the budget threshold at runtime (default: 15%)
wf run cost-modeling --var BUDGET_THRESHOLD=10 --work-stealing --print-output

# Strict threshold — require sign-off on anything over 5%
wf run cost-modeling --var BUDGET_THRESHOLD=5 --work-stealing --print-output

# With overall timeout ceiling
wf run cost-modeling --work-stealing --timeout 1h --print-output

# Visualise the DAG (note matrix nodes for three environments)
wf graph cost-modeling
wf graph cost-modeling --format html --export cost-modeling.html   # writes HTML file

What to Observe

  • estimate-deployment-cost expands into three matrix nodes — [env=dev], [env=staging], [env=prod] — all running in parallel; each registers estimated_delta_dev, estimated_delta_staging, estimated_delta_prod
  • check-budget-threshold waits for all three and computes the total delta as a percentage of baseline; registers budget_decision
  • deploy-infrastructure has if = 'budget_decision == "approved"' — if the threshold is exceeded, the task is skipped entirely (no deployment), which is visible in wf audit
  • measure-actual-cost is another three-node matrix — runs post-deployment to measure the real delta
  • compute-variance compares estimates to actuals and registers variance_status
  • alert-cost-variance only runs if variance_status == "over_threshold" — check wf audit to see whether it fired
  • preserve-cost-snapshot is a type = "forensic" task — it only runs if deploy-infrastructure fails; it preserves cost context for rollback analysis

Inspect After Running

RUN_ID=$(wf runs --tag finops --limit 1 | awk 'NR==2{print $1}')

# See all registered cost variables
wf inspect $RUN_ID

# Chronological event trail — see what ran, what was skipped, if alert fired
wf audit $RUN_ID

# Per-task output
wf logs $RUN_ID check-budget-threshold     # see the approval decision
wf logs $RUN_ID compute-variance           # see the variance table
wf logs $RUN_ID alert-cost-variance        # present only if variance exceeded threshold

# Compare two deployment runs (e.g. before and after a cost optimisation)
wf diff <prev-run-id> $RUN_ID

Adapting for Production

  1. Cost estimation: Replace the simulated case blocks in estimate-deployment-cost with real tooling:
  2. Infracost — infracost breakdown --path ./infra/{{.env}} --format json
  3. OpenCost — query the API post-plan
  4. Terraform cost estimation via Terraform Cloud / HCP Terraform API
  5. Actual cost measurement: Replace the simulated deltas in measure-actual-cost with:
  6. AWS Cost Explorer API (aws ce get-cost-and-usage)
  7. Azure Cost Management API
  8. GCP Cloud Billing API
  9. Budget threshold: Expose BUDGET_THRESHOLD as a per-environment parameter; different thresholds for prod vs non-prod are common
  10. Variance alert routing: Update alert-cost-variance to call your PagerDuty, Slack, or JIRA API
  11. FinOps report: Update publish-finops-report to push data to your cost allocation system (CloudHealth, Apptio, etc.)
  12. Currency: All dollar amounts in this example are USD monthly deltas; adjust the arithmetic for hourly costs or non-USD currencies