Anomalies & forecast

spendlens anomalies

Finds days when a service's (or account's, or tag's) spend jumped well above its normal level.

spendlens anomalies
spendlens anomalies -g account --days 7 --threshold 4
spendlens anomalies --drops                   # also sudden decreases, e.g. a workload that stopped
spendlens anomalies --fail-on-anomaly         # exit 3 for alerting pipelines

How it decides. For each recent day it takes the trailing anomaly.lookback_days (default 30) as a baseline and computes a robust z-score from the median and median absolute deviation. Robust statistics aren't dragged around by earlier spikes the way a mean and standard deviation are.

A day is an anomaly only when all three hold: - the z-score is at least anomaly.threshold (default 3); - the increase is at least anomaly.min_increase per day (default 10), so tiny services don't trigger alerts; - the day's cost is at least anomaly.min_daily_cost.

Services with too little history (brand new ones) are skipped. cost diff reports them as new.

Severity comes from the z-score: low, medium (≥ 4), high (≥ 6) and critical (≥ 10).

spendlens forecast

Projects month-end spend and checks it against your budgets.

spendlens forecast
spendlens forecast --set budgets.monthly_total=55000 --fail-on-risk
  • Projection. Spend so far this month plus Cost Explorer's forecast for the remaining days. When the forecast API can't answer (for example with too little history), it falls back to a linear run-rate projection and says so.
  • Your budget. The projection is checked against budgets.monthly_total.
  • AWS Budgets. Every monthly cost budget in AWS Budgets is listed, using AWS's own forecasted spend for it.
  • Statuses:
  • over budget
  • forecast over budget
  • at risk (above budgets.warn_pct, default 80%)
  • on track
  • --fail-on-risk exits with code 3 if anything is forecast over its limit.