Four settings before receipts-worker ships
I wrote the production values, alerts, deploy workflow and runbook from
what staging measured. Four settings are yours, each a
TODO(you) in its file. Each file saves to disk as you type,
and I render the chart against staging once they're set.
- Pods at the morning peak. One pod renders 46 receipts a minute; the 09:00 peak is 130. Three pods leave 6% headroom, four leave 41%. values-production.yaml
- Memory limit. Chromium peaked at 1.35Gi in staging at four renders a pod. I set 1536Mi; a pod over it is killed mid-render. values-production.yaml
- Who a backlog pages. I routed it to payments on-call, whose cluster runs the queue. alerts.yaml
- Who to call next. The runbook's escalation, for a backlog on-call can't clear in an hour. receipts-worker.md
The workflow and staging values are as staging ran them: 3,412 receipts on 12 November, none dead-lettered.