✓ 50,000 events a day processed reliably
✓ Infra cost down from $1,900 to $210/month
✓ 0.02% error rate with retries and dead-letter queues
Before → after
| Metric | Before | After |
|---|---|---|
| Events processed | 3,000/day | 50,000/day |
| Monthly infra cost | $1,900 | $210 |
| Error rate | n/a | 0.02% |
An e-commerce brand was running order processing, webhooks, and image jobs on a small server that was constantly either idle or overloaded. Every flash sale was a gamble.
The challenge
- A single server that sat idle most of the day and melted during spikes.
- Manual retries when webhooks failed, and no dead-letter handling.
- Image resizing blocked the main request path, slowing checkout.
- A monthly bill that didn’t match the actual, tiny workload.
What we built
- Serverless functions replaced the server with event-driven Cloud Functions triggered by Pub/Sub and webhooks.
- Image pipeline moved resizing and optimization to functions that fire on upload.
- Retries & dead-letter queues added automatic retries and a dead-letter path for failed events.
- Orchestration chained functions for order confirmations, emails, and inventory updates.
- Monitoring wired error tracking and alerting so failures surface in minutes.
The results
The brand now processes 50,000 events a day with a 0.02% error rate, and the infrastructure bill fell from $1,900 to $210 a month. Flash sales stopped being a gamble.
The takeaway
Event-driven serverless is the right shape for bursty, small workloads: you pay for actual use, and “no servers to manage” means no 2am pages.
Stack Cloud FunctionsPub/SubCloud StorageCloud RunWebhooks
“
Fifty thousand events a day, and I haven't touched a server in months.