Agent Schedules
"Every night, summarize what changed and file it." A schedule is that standing order: a cron expression, read in a time zone, that starts a durable run of an agent or a workflow each time it falls due. It runs under an API key, with that key's grants, owner and quotas, never as the server.
1Make a schedule#
Open Admin > Build > Schedules and choose New schedule:
| Field | What it sets |
|---|---|
| When | A cron expression: minute, hour, day of month, month, day of week. The form reads it back in words with its next three times |
| Time zone | The zone the expression is read in (Europe/Paris, America/New_York); empty reads it in UTC |
| Starts | A published agent, or a published workflow (which also takes variables) |
| Asked each time | The run's input |
| Runs as | The API key whose grants, owner and quotas the runs use |
| Catch up after downtime | What happens to an occurrence missed while the server was down |
| Expression | Runs |
|---|---|
0 2 * * * |
Every day at 02:00 |
30 9 * * 1-5 |
Monday to Friday at 09:30 |
0 9 * * 1 |
On Monday at 09:00 |
*/15 * * * * |
Every 15 minutes |
0 6 1 * * |
On day 1 of the month at 06:00 |
A schedule counts from the moment it is saved: an edit never fires an occurrence that fell before it. Wall times the clock skips when daylight saving begins fire when the hour resumes.
2Once, and only once#
One node of a farm fires the schedules (the user-schedules job holds a lease, listed on
Admin > Operate > Jobs), every thirty seconds. Each occurrence is claimed in the shared store
before its run starts, so a node taking over never fires it again.
When the server was down past an occurrence (more than five minutes late), the schedule either skips
it and records skipped with the time it was due, or, with Catch up after downtime, runs once
on return. Several missed occurrences run once, never one after another.
3Follow it#
Each row shows the schedule in words, when it runs next and what its last occurrence came to:
started, skipped, or refused with the reason (the key was revoked, its quota is spent, the
agent or workflow is no longer published). Runs lists the runs it started, each opening its
timeline; Run now starts one at once without moving the next occurrence.
A run that cannot start, or that fails, raises an alert on Admin > Operate > Alerts: nobody waits on a scheduled run to notice it.