Cloud Cost Is an Engineering Problem

The most expensive bug I've chased didn't crash anything — it just quietly inflated the Cloud Monitoring bill every month. The culprit was noisy Ops Agent metric collection: high-volume metrics nobody actually read.

Observability has a unit cost. Once you attribute spend to its source, an 'unexplained' bill becomes a fixable engineering problem. Tuning the metrics that nobody consumed recovered the waste without losing a single alert.

To stop the next leak before month-end, I put billing-anomaly detection on top of the BigQuery billing export: a scheduled query compares spend to baseline and alerts on a breach within hours.

The same data feeds an automated monthly report. Detection and reporting share one source of truth, so the numbers never argue — and cost stops being a surprise.