Writing

Notes on running things that break

Post-incident write-ups, the occasional opinion about on-call, and what we changed in the product because of both.

The quiet hour: what a stalled queue looked like from the outside

For fifty-one minutes our writes succeeded and our reads returned yesterday. Nothing alerted, because every check we had was asking the wrong question. Here is the timeline, the wrong assumption, and the four checks that exist now.

14 July 2026 - 11 minutes - by Priya Raman

- On-call
The handover note nobody reads, and how we fixed it

Six lines, written by the person leaving the shift, in the same place every time. It cut our repeat pages roughly in half.

30 June 2026 - 6 minutes
- Product
Why subscribers now choose services, not pages

One customer told us they had muted us entirely because a region they do not use kept emailing them. That was the whole brief.

11 June 2026 - 4 minutes
- Post-incident
A certificate expired on a Sunday and we found out from a customer

The renewal job had been failing quietly for eleven weeks. The interesting part is not the job, it is why nobody saw eleven weeks of failures.

25 May 2026 - 9 minutes
- Engineering
Hosting a status page nowhere near the thing it reports on

Separate provider, separate region, separate registrar. What that costs, and the two places the independence still leaks.

2 May 2026 - 7 minutes
- Writing
The first update is the one that matters

Nobody needs a cause in the first ten minutes. They need to know you have seen it, and when you will speak again.

18 April 2026 - 3 minutes
- Engineering
Two teams, one outage, two different uptime figures

Partial degradation is where every uptime definition falls apart. We picked one, wrote it down, and stopped arguing.

3 April 2026 - 8 minutes

The quiet hour: what a stalled queue looked like from the outside

The handover note nobody reads, and how we fixed it

Why subscribers now choose services, not pages

A certificate expired on a Sunday and we found out from a customer

Hosting a status page nowhere near the thing it reports on

The first update is the one that matters

Two teams, one outage, two different uptime figures