- Type:
- Internal measurement
- Version:
- ZeroOS internal development build, control-plane high-availability test stage
- Date:
- August 2026
- Environment:
- Two virtual machines in separate physical failure domains, PostgreSQL 16. Test environment with no production traffic.
- Method:
- With a standby controller already running, we force-killed the main controller process and timed how long the standby took to take over. We also checked that every late write from the old controller was rejected and that task records stayed consistent.
- Sample / runs:
- One recorded on-site failover
- What it does not show:
- Does not cover power-level isolation, whole-machine recovery, model-gateway recovery or wide-area network latency. Not re-measured in a customer environment.