Work Intent: Stop the disks query from spinning up array disks (#2018) #2105

Overview

Fix #2018. Any query that goes through DisksService.getDisks() wakes every spun-down array disk: disks, disk(id), assignableDisks and the onboarding Internal Boot step (getAssignableDisks() / getInternalBootDevices()). Even { disks { id } } is enough. Anything that polls the disk list keeps large drives from staying in standby.

The cause is the same one #2090 / #2091 found for temperatures: getDisks() builds its list from diskLayout() in systeminformation, which on Linux runs smartctl -a -j and smartctl -H on every disk without -n standby.

I've been running the approach below as a runtime patch on my own server (Unraid 7.3, mixed array with 16 and 18 TB drives) for a few months. With the disks spun down, { disks { id } } spins up every large drive on stock, and none with the patch. The query also goes from several seconds to under 500 ms.

Technical Approach

• Replace diskLayout() in getDisks() with a listing from lsblk -d -b -J -O, which only reads kernel metadata. Ideally this builds on getPhysicalDisks() from fix(api): stop temperature metrics from spinning up array disks #2091 once it lands, adding the columns getDisks() needs (vendor, firmware revision, rotational) so there is a single place that lists disks.
• Map each device to the shape diskLayout() returns today, so parseDisk() stays as it is: device, name (model), vendor, serialNum, firmwareRevision, interfaceType (from tran), size in bytes, and type as HD / SSD / NVMe (from rota and tran, same values diskLayout() returns). lsblk reports vendor: "ATA" for SATA drives, so in that case the vendor is derived from the model string.
• Get smartStatus from smartctl -n standby -H -j <device>. Exit code 2 means the disk is in standby and was not queried, which maps to UNKNOWN.
• Keep blockDevices() for partitions.
• Tests: unit tests for the lsblk mapping (size in bytes, drive type, vendor fallback, filtering out loop/ram devices) and for the smartctl status mapping, plus a test that getDisks() no longer calls diskLayout().

Out of scope: the geometry fields on Disk (bytesPerSector, totalSectors, etc.) are non-nullable but diskLayout() already returns null for them on Linux, so selecting them errors today. This change would leave them as they are; happy to look at that separately.

Scope

• API
• Plugin
• Web UI
• Build/Deploy Process
• Documentation

Timeline & Impact

• Estimated time needed: a few days, since the logic already runs as a patch and mostly needs porting and tests. I'd start after fix(api): stop temperature metrics from spinning up array disks #2091 is merged to avoid conflicts in disks.service.ts.
• Potential impacts: smartStatus becomes UNKNOWN for disks in standby instead of waking them to read it. No schema changes.

Pre-submission Checklist

• I have searched for similar work/issues
• I understand this needs approval before starting
• I am willing to make adjustments based on feedback