On-host crash probe: test scope

This page summarizes a historical internal v6 test plan. It is not an end-user procedure or a completed on-host crash result. Use the PoolFS README for current usage and the recovery test record for measured outcomes.

What the probe measures

The internal probe was designed to create a fresh two-member v6 pool, write successive 4 KiB values, and record each completed write in a separately flushed acknowledgement log. The probe itself does not crash or reboot Windows. Its verification phase compares the pool after an interruption with the last complete acknowledgement. It accepts that generation or one additional complete write that could have committed before its acknowledgement was recorded, then checks the unmounted member set with the offline scrubber and checker.

The verifier refuses missing or resized members and rejects malformed complete acknowledgement records. A log stored on the same host could itself lose or tear its newest record during a physical power loss; without the run state and both member images, the trial cannot support a recovery claim.

What this does not prove

A normal Windows restart flushes file buffers and is a different event from an abrupt host interruption. Microsoft documents Windows shutdown flushing. Even a successful abrupt host trial would cover one interruption timing, not all controller-cache behavior or physical-disk failure modes. The source note records a dry-run method for the probe but no completed on-host crash result. The separate Hyper-V guest reset result kept the storage host powered and does not establish physical host power-loss safety.