E-Commerce Cloud Costs: Pre-Scaling for Peak Sales Days
A retail calendar is known months ahead. Capacity should be scheduled against it, then handed back the morning after.
Retail traffic has a property most workloads lack: the biggest spikes are on a calendar. A flash sale, a holiday weekend or a product drop is announced in advance. That turns peak capacity from a guessing game into a scheduling problem, and the cost question becomes how briefly you can hold that capacity.
Why reactive autoscaling is not enough on a sale day
Metric-based scaling reacts after load arrives, and new instances are not useful the moment they launch. Google’s own guidance for managed instance groups says VMs “aren’t ready immediately at the given start time”, so a schedule should start early enough for them to boot and start the application. Each cloud offers a native way to set capacity in advance:
- AWS Auto Scaling groups take scheduled actions that set desired, minimum and maximum capacity at a given time, in an IANA time zone. A group can hold up to 125 scheduled actions, and an action can run up to two minutes after its scheduled time.
- GCP managed instance groups accept up to 128 scaling schedules, each at least 5 minutes long, with a minimum instance count that applies while the schedule is active.
- Azure autoscale supports fixed date profiles, which are evaluated before recurring and default profiles. Microsoft’s own example is a profile that scales out on Black Friday.
The risk on the cost side is the other end of the event. A scheduled scale-out that nobody scales back in keeps peak capacity running into a quiet week.
How ZopNight Event Readiness handles the whole event
Event Readiness is a wizard for pre-scaling infrastructure for planned traffic spikes such as sales and launches:
- Targets: AWS Auto Scaling groups, AWS ECS services, GCP managed instance groups and Azure VM scale sets, through their existing autoscaler policies.
- Sizing: give a multiplier on baseline or an expected request count, and the capacity engine proposes scaled minimum, maximum and desired values per target, which an admin can override.
- Native schedules, no polling: ZopNight writes the provider’s own scheduled action:
PutScheduledUpdateGroupActionfor Auto Scaling groups,PutScheduledActionfor ECS, a fixed date profile for Azure, andscalingSchedulesfor GCP. - Automatic revert: original minimum, maximum and desired values are recorded when the event is scheduled and restored when it completes or is cancelled. Cancelling an AWS event removes every ZopNight scheduled action on the group, so none are left behind.
- Cost before commit: the wizard shows an estimated event cost, clearly marked as an estimate, before anything is saved.
- Databases are advisory only: attach a database to see connection-pool math and a suggested tier. ZopNight never changes the database; tier changes stay with your DBA team.
A readiness check scores each target against current capacity, autoscaler policy posture and recent activity, so gaps show up days ahead rather than at the start of the sale. Walk through a full plan in pre-scaling for Black Friday or the broader event readiness playbook.
The quiet months
Outside peak season, the same estate is judged per order. Unit economics lets you define “completed orders” as a denominator and plots infrastructure cost per order on the cost trend chart. See cost per order tracking. Autoscaler schedule triggers also include a Weekend Scale-Down preset for recurring troughs.
Questions we get a lot.
If yours isn't here, email us and we'll answer directly.
What happens if a sale is cancelled after the event is scheduled?
Cancel the event and ZopNight restores the original capacity values it recorded when the event was scheduled. On AWS it also removes every event scheduled action it created on the group or service.
Will Event Readiness resize our database for the sale?
No. Databases can be attached for sizing guidance only. ZopNight shows connection-pool math and a suggested tier but never modifies a customer database.