cron

Overview ¶

Package cron is the occurrence engine behind a deployment's schedule: the reference's 5-field POSIX cron dialect, matched literally against a wall clock in an IANA timezone.

Due, Next and Upcoming share one generator on purpose. The list a user reads in a deployment's upcoming_runs_at and the instant the scheduler actually fires come from the same walk, so they cannot disagree — plan 37 §3.1, and the property test in cron_test.go asserts it.

The dialect is a strict subset of what a general cron library accepts, which is why this is hand-rolled: seconds and year fields, the special characters L, W, # and ?, and predefined shortcuts like @daily are all excluded, so a library would need a rejection layer in front of it and would still not supply the two daylight-saving rules below.

Timezones ¶

The tzdata import is deliberate and belongs here rather than in a main. The server image is debian:stable-slim with ca-certificates and nothing else, so without an embedded database every non-UTC deployment would be refused at create time in the image and never in the test gate, whose host has a zoneinfo tree. Costing ~450 KB in the binary buys the guarantee that no binary or test can forget it.

Daylight saving ¶

A wall clock that does not exist on the spring-forward day is skipped; one that occurs twice on the fall-back day fires twice, as two distinct instants. Neither rule is published by the reference — both are registered inferences (plan 37 §8.1 entry 19) — and both are the highest-risk code here, because a wrong answer looks exactly like a right one.