Time zone scheduling
Plan for the burst at the start of the business day
Local-time schedules can concentrate work even when the daily total is small. Spread only the tasks whose timing promise allows it.
In this article
Measure the due-time distribution
Count expected occurrences by resolved time window and business zone. Many customers choosing 9 am can create a sharp peak that an average daily rate hides.
Include daylight-saving changes in the forecast because the UTC position of local peaks can move. Different zones may overlap differently across the year.
Use the actual recurrence rules rather than assuming every schedule repeats at a fixed UTC interval.
Separate exact and flexible timing
Some tasks must run at a specific time. Others only need to complete within a window. Make that distinction explicit in the product and scheduler.
Where users agree to flexibility, distribute execution within the allowed window to reduce concentration. Do not silently delay a precise appointment reminder because a spreading feature is convenient.
Managed schedulers may provide flexible-window features, but the application's promise still determines whether they are appropriate.
Estimate processing demand
Use measured work per occurrence and downstream constraints. For an illustrative 6,000 reports due within ten minutes, the system needs an average of ten completed reports per second during that window, before allowing for variation and retries.
If each report makes several provider calls, those limits may dominate capacity. More worker instances cannot remove an external quota.
Plan a safe backlog policy for overload or outage rather than leaving every missed occurrence to retry at once.
Keep catch-up bounded
After downtime, decide which occurrences remain useful and how they compete with current work. A catch-up-once policy can differ from executing every missed interval.
Preserve occurrence identity and record skipped or superseded work under the business rule.
The capacity plan should support the timing users were promised, including concentrated local peaks and recovery. A low total job count does not establish that the service can handle the moment when most jobs become due together.
Primary sources
AWS: EventBridge schedule typesReferences checked 11 September 2026.