Scale up for the
sale, before
the sale.
Pre-scaling for a launch is usually a calendar reminder and somebody's nerve. Tell ZopNight the date, the window and the multiple, and it writes the cloud's own scheduled actions on your scaling groups, then puts the capacity back when the window closes.
A wizard, then the cloud does the rest.
No agent watching a clock, nothing to fail at 2am.
Describe the event
A name, a window, and either a multiplier on current baseline or the requests per second you expect. The capacity engine returns a scaled min, max and desired per target. Override any of it at step two.
Preview the cost
A live estimate before anything is written, clearly badged as an estimate. Databases can be attached as monitor-only targets: you get connection-pool maths, headroom and a suggested tier, and ZopNight never modifies your database. That call stays with your DBA.
Schedule it
ZopNight writes the cloud's own scheduled actions. AWS scheduled update group actions, ECS scheduled actions through Application Auto Scaling, an Azure fixed-date profile, GCP scaling schedules. There is no polling loop that can miss its window.
Come back down
The original min, max and desired are snapshotted when you schedule. On complete or cancel the executor restores them, and a cancel sweeps every scheduled action it wrote, so nothing is left orphaned on a partial failure.
Designed around what goes wrong.
Each of these is a specific bug class, handled rather than hoped away.
- Lifecycle
- Draft, scheduling, scheduled, scaling up, active, scaling down, completed. Cancelled is terminal. Concurrent schedule requests on the same event are rejected instead of racing.
- Idempotent cancel
- Schedule IDs are derived deterministically from the event ID, so a re-queued cancel is a no-op rather than a second operation. AWS cancel sweeps every scheduled action for the group, so a partial failure leaves no orphans.
- Provider correctness
- GCP cancel forces the scaling schedules field, because a PATCH merges map keys and would silently leave stale schedules behind. ECS cancel deletes every returned schedule ID and names each one that failed.
- Databases stay yours
- Attached as monitor-only. ZopNight surfaces sizing recommendations from real connection-pool maths, never hardcoded specs, and never issues a tier change or a replica scale.
Plan the spike instead of surviving it.
Set up one event against a real scaling group and see the cost estimate before anything is written.