# Storage monitoring requirements

Storage is a first-class domain with strict read-only behavior.

## Array

Display:
- current array state;
- parity presence/state;
- data disk membership;
- missing/disabled/emulated disks;
- read/write activity;
- parity check state, progress, speed, errors and history;
- last/next check when available.

No start/stop/check/correct action is offered in v1.

## Disks

For every disk:
- stable identity and role;
- model/serial with privacy-aware display;
- size and filesystem;
- used/free/inodes;
- temperature and trend;
- read/write throughput, IOPS and latency when available;
- spin state when available;
- SMART overall and selected attributes;
- self-test age/result;
- reallocated, pending and offline-uncorrectable sectors;
- CRC/interface errors;
- SSD wear/percentage used where relevant.

Rules must avoid declaring a disk healthy when SMART data is unavailable/stale.

## Pools

Support cache and ZFS/Btrfs pool concepts discovered in the environment:

- members;
- usable/used/free;
- profile/redundancy;
- degraded/faulted/offline member state;
- scrub status/results;
- filesystem errors;
- write/read performance;
- SSD wear;
- mover-related signals when available.

Do not assume all pools use the same filesystem.

## Shares

- name and configured policy;
- used size and growth;
- participating storage;
- cache/pool relationship;
- recent growth;
- forecast to thresholds;
- unavailable/stale state.

Potentially expensive size calculation must be rate-limited/cached and optional.

## Forecasting

Capacity forecast:
- uses configurable historical window;
- reports method and confidence/uncertainty;
- handles irregular bulk imports;
- never presents a precise date without qualification;
- can be disabled when insufficient history exists.

## Baseline alert classes

- array degraded/missing disk;
- parity errors;
- SMART critical attribute;
- temperature sustained above policy;
- pool degraded/faulted;
- capacity thresholds;
- read-only filesystem;
- scrub/check errors;
- stale storage data.

Availability, device health, capacity pressure and thermal pressure are separate
signals. An online device can therefore be capacity-critical or thermally
critical without being presented as offline, and a stale observation makes all
current signal severities Unknown. Pulse uses the stable source device ID as
the canonical physical identity. Repeated identical observations are
idempotent; conflicting roles or facts for one physical ID fail closed instead
of producing duplicate or contradictory topology nodes.

## Safety tests

Storage fault behavior is tested through fixtures/simulator or read-only captured data. Never trigger real SMART damage, pool degradation, disk removal or array operations.