Harbour Commerce Platform

Introduction

Harbour lost its two best sales days a year to a checkout that folded under load. The fix was less about servers than about how much the browser was being asked to do before a customer could pay.

Measuring the Real Failure

The infrastructure was never the constraint. Replaying the previous peak showed the critical path pulling in scripts that had nothing to do with completing an order, each one adding a place for the transaction to stall.

Shortening the Critical Path

We stripped the critical path back to what the transaction needed, deferred the rest, and rewrote the checkout as a sequence that survives a dropped connection.

• Anything not required to take payment loads after the order can be completed.
• Each checkout step commits server-side, so a lost connection resumes instead of restarting.
• Third-party scripts run in a sandbox that cannot block the page.

Proving It Under Load

None of that counts until it has been tested at the volume that broke the old build, so load testing ran against production-shaped data rather than a synthetic catalogue.

• Tests replayed the previous peak weekend, then the same again at triple the volume.
• Failure modes were rehearsed deliberately, including a payment provider timing out mid-order.
• The results set the alerting thresholds that are still in use.

Conclusion

The following peak weekend completed without an outage, at roughly triple the previous concurrent load, and the checkout has not been the cause of an incident since.