Machine health checks

Fly.io reports whether a running machine is passing its configured health check. A Beam agent reads that status continuously, applies the account's rule for how many failed checks count as a real outage, and restarts the affected machine or notifies on-call once the threshold is crossed. This catches a stalled machine before it causes a wider outage a customer notices. Machines that fail to recover after a restart, or that fail in a pattern the rule doesn't cover, are escalated to an engineer to diagnose directly.

Multi-region deployment

Fly.io runs an app's machines across regions a team selects for latency or redundancy. A Beam agent reads the account's approved region list and current machine distribution, and provisions or removes a machine in a region to match that approved list whenever demand or an outage changes the picture. This keeps an app's footprint aligned with policy without an engineer adjusting it by hand. Requests to add a region outside the approved list, or unexpected regional cost spikes, are routed to an engineer for a decision.

Auto-restart policies

Fly.io can restart a machine automatically based on a policy an engineer sets, such as on crash or on a schedule. A Beam agent reads the account's approved restart policy, applies it consistently across an app's machines, and notifies the team when a restart occurs outside expected maintenance hours. This keeps restart behavior predictable rather than left to whatever a machine's own crash loop decides. Repeated restarts within a short window, which usually signal a deeper issue, are escalated to an engineer instead of being retried indefinitely.