Why we rebuilt the order queue as a durable job table
Submitting an order to a supplier inside the checkout request looked fine until a supplier had a bad afternoon. Here is what we replaced it with, and why leases beat retries-in-a-loop.
Mostly engineering write-ups and operational post-mortems, occasionally a playbook. Written by the people who did the work.
Submitting an order to a supplier inside the checkout request looked fine until a supplier had a bad afternoon. Here is what we replaced it with, and why leases beat retries-in-a-loop.
Every multi-tenant platform says data is isolated. We made ours provable: a scan that discovers tables from the schema catalogue and fails the release if any business table lacks a tenant column.
Margin floors, rounding and market multipliers sound simple until costs move overnight. A walkthrough of how to set rules you will not have to babysit.
We publish every probe result for ninety days. Turning that into a public commitment changed which problems we prioritised — mostly for the better.
The people who most need to reach support are often the ones whose page just broke. So the form is a plain POST, and it accepts a ticket with no workspace attached.
Our launch board runs nineteen readiness checks and treats “unknown” as a failure. That single rule caught more real problems than the other eighteen checks combined.