Skip to content
ZekaForge
Deploy

Scaling the device-worker fleet

The honest mechanics of scaling real-device capacity — hardware, managed services and autoscaling.

Scaling a web app means adding stateless servers. Scaling real-device testing is harder, because a session is pinned to a physical phone that someone has to own, power and maintain.

The constraint

A cloud server has no phone attached. Real-device access requires either a fleet of device workers — hosts with phones physically connected over USB — or a managed service such as AWS Device Farm that offers remote-access sessions.

Two paths

  • Self-hosted workers give full control and the lowest per-session cost at scale, but you own the hardware, USB hubs, and device health.
  • Managed service removes the hardware burden and is ideal early on, at a higher per-minute price and with the provider's device catalog.

ZekaForge's orchestration treats both as pluggable backends, so you can start managed and add owned hardware later.

Fleet topology

Workers register with the orchestrator, advertise their device model and Android version, and receive sessions. A queue (Redis-backed) holds pending reservations so demand spikes wait rather than fail.

Autoscaling trade-offs

You can autoscale worker hosts, but you cannot autoscale physical phones instantly. Practical approaches: keep a warm pool sized to typical peak, queue overflow, and for managed backends scale on demand.

Health and recovery

Phones get stuck. Workers run watchdogs that detect unresponsive devices, reboot them via ADB, and quarantine repeat offenders so sessions never land on a bad device.

Cost control

Idle phones still cost money. Concurrency limits, session timeouts, and S3 lifecycle rules on recordings keep spend predictable.

Honesty note

Until hardware or a managed backend is wired in, ZekaForge runs a labeled demo so the orchestration, quotas and UI are real and reviewable before a single phone is plugged in.

Partager

Articles liés

Commencez avec ZekaForge

Créez un compte gratuit et explorez Device Cloud et Cyber AI en quelques minutes.