Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Decision: Drive-wake posture

Principle: HDD defaults

Context

braid previously treated all pool disks as potentially asleep at any time. The TUI reflected that blanket anti-wake posture by keeping the expensive pool probe manual-only: smartctl -H -A --json per disk plus btrfs state was run only on startup or r.

That posture was broader than the actual ownership boundary. Today braid does not park drives with hdparm -S or any equivalent per-drive standby timer. While the pool is mounted, btrfs and normal NAS access already make member drives active. A future opt-in braid.autoSpinDown may park drives, but that feature belongs to the locked state and will own the “do not wake a parked locked member” rule.

Decision

While the pool filesystem is mounted (PoolStatus::Mounted in the TUI, pool.mounted in status/doctor), braid treats member disks as awake and reads/refreshes them freely. The anti-wake concern applies only to the locked state and is owned by the future opt-in braid.autoSpinDown feature, which will gate on braid-online.service. braid adds no online-side standby detection.

The automatic read boundary is narrow: the TUI pool auto-refresh loop is the only automatic disk-read loop added here, and it re-arms only while the live pool is mounted. It does not run while the pool is locked, not mounted, or in an error state.

The user-invoked read paths keep their existing semantics:

  • braid status live SMART is mounted-only; it returns the not-mounted status before spawning per-disk smartctl.
  • braid doctor SMART self-test diagnostics may target a member’s persisted by_id path even when the pool is unmounted or locked.
  • TUI Browse-tab SMART commands are explicit user actions and may target a member’s persisted by_id path regardless of mount state.

“Online” in this decision means the mounted live pool (PoolStatus::Mounted or pool.mounted). braid-online.service is a correlated systemd lifecycle marker, not something these read paths consult; it is the handle the future braid.autoSpinDown feature will gate on.

Locked-state TUI probes are not claimed to be non-waking. The pre-existing TUI startup/manual probe builds LUKS state before the mount gate and reads on-disk LUKS2 metadata with cryptsetup luksDump --dump-json-metadata <by-id> for locked members. That read can wake a parked drive. This decision leaves that behavior unchanged and only ensures the new automatic pool loop is mount-gated.

Non-goals

  • No smartctl -n standby.
  • No Standby SMART health state.
  • No braid.autoSpinDown implementation.
  • No hdparm integration.
  • No NixOS module changes.
  • No smartd configuration changes.
  • No change to status’s mounted-only SMART probe.
  • No change to explicit doctor or Browse-tab SMART reads.
  • No auto-refresh while locked.

Alternatives considered

Online-side standby detection

Rejected. Adding smartctl -n standby and a Standby health state creates parser and UI state complexity for a state braid does not create while the pool is mounted. If an operator has configured an out-of-band standby timer, the cost of being wrong is a single wake-on-read, matching today’s explicit reads.

Blanket anti-wake posture

Rejected. Keeping every pool probe manual-only forces interactive monitoring to behave like locked-state recovery. The locked state is the only state where braid-managed drive parking belongs, so the locked-state feature should own that rule instead of leaking it into mounted-pool UX.

See