Bill horror story
$72,000 in two hours: the Cloud Run and Firestore test that ran away
October 21, 2026 · 3 min read
A scraper bug, Cloud Run's scaling and a $7 budget that only alerted: how a startup's overnight test turned into a bill just under $72,000.
Retold from The Register, by Tim Anderson, December 10, 2020. Summary in our own words; every figure is the source's.
What happened
Milkie Way, a small startup founded by Sudeep Chauhan, was testing a web scraper on Cloud Run that saved what it found to Firestore. They had set a budget of $7. A bug let the scraper recurse forever when pages linked to each other, and Cloud Run's defaults at the time let it scale out to many instances serving many requests each.
Over about two hours one night it made roughly 116 billion Firestore reads and 33 million writes. The bill first showed around $5,000 and ended just under $72,000. Chauhan's lesson, as The Register quoted it: "Fail fast, learn fast with cloud is a bad idea."
How it ended
After the company sent Google a detailed incident report, Google waived the bill as a one-time gesture. The original blog post, "We Burnt $72K testing Firebase + Cloud Run and almost went Bankrupt", is no longer online.
What would have stopped it
- A cap on scaling. A low max instances setting on the Cloud Run service would have bounded how fast the loop could spend.
- Acting on the budget, not just alerting. The $7 budget existed; it only sent alerts, which arrive hours late. An automatic kill switch would have cut billing on the first report past the limit, at a fraction of $72,000.
- Watching usage, not cost. Billions of Firestore reads in an hour is a spike Cloud Monitoring shows within minutes. A usage spike rule on Firestore reads would have fired long before the first cost report.
