Skip to content
KayD
← All notes

A second way to run the clock

SLA timers on the Uniguru operations platform run as scheduled jobs, with a fallback for the clock itself.

, Reliability, Architecture

The Uniguru operations platform holds a consultancy's lead pipeline, admissions, partners and staff work in one place. Some of that work has deadlines: a lead should be followed up within its SLA, and a breach should be recorded when it is not.

Clocks as scheduled jobs

SLA clocks run as scheduled jobs on Vercel and record breaches. If the job that measures time stops running, the platform quietly stops noticing late work, which is the one thing it is there to notice.

So the SLA clock has a GitHub Actions fallback. One scheduler can fail without the clock stopping.

Checks in more than one place

The same thinking applies to access. Every user has a system role, roles map to permission groups, and each route and server action checks the permissions it needs, on top of row-level security in Postgres.