MASAKA / Operations / Autoscaling
Operations

Capacity follows
regional demand.

Workers are launched from queue demand and host headroom. No user configures a fixed production worker count.

Scaling signals

DemandQueued + activeMeasured independently for China and overseas.
SafetyCPU + memoryHost pressure pauses expansion before overload.
ReadinessHeartbeats + capacityOnly healthy workers claim regional sessions.

Control loop

Demand and host headroom feed the autoscaler, which creates healthy warm workers.QUEUE DEMANDHOST HEADROOMREGIONAL AUTOSCALERbounded launch batcheswarm spare targetWORKERShealthy capacityautomatic repair

100+ peak capacity

There is no configured numeric fleet ceiling. A sufficiently provisioned host can plan beyond 100 workers, while actual concurrency remains bounded by browser memory, CPU, network, launch rate, upstream provider quotas, and account-level session limits.