[Starosdev]

Summary

WP-D (collector robustness) for the 2.0.0 release. The collector stops registering devices it could not identify, gains an exclude: list plus a built-in virtual-disk exclusion, and can collect SMART data from several devices in parallel (default unchanged: one at a time).

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)
• Documentation update
• Refactoring (no functional changes)
• CI/CD or infrastructure change

Related Issues

Related to #965. Addresses #913 and #912.

Changes Made

• Detection (Linux, darwin, windows, freebsd) does not register a device when smartctl --info fails: fatal exit status, non-exit errors such as timeout or missing binary, empty output, or undecodable JSON. Non-fatal exit bits (for example exit 4, fix(collector): preserve jmb39x-q device type and accept usable SMART JSON #663) still register. Previously these devices registered without identity and were a likely source of duplicate devices. Each skipped device is reported every run as a collector scan error (ReportScanError("info", ...)), which notifies through notify_on_collector_error like before, so a hard-failing drive still alerts.
• New exclude: list of regexes matched against device path, model name, and device type; invalid patterns are skipped with a WARN. Checked before --info (path, type) and after it (model).
• New exclude_virtual_disks (default true): skips QEMU HARDDISK, VBOX HARDDISK, and Virtual Disk / Msft Virtual Disk models, case-insensitive, with one INFO line. Devices listed explicitly under devices: or in allow_listed_devices are never skipped by the built-in list; user exclude: patterns still apply. Patterns compile once per scan.
• New commands.metrics_parallelism (default 1, today's behavior; values below 1 are treated as 1). metrics_smartctl_wait now applies per worker.
• Fixed a data race in config.GetDeviceOverrides() memoization found by -race.
• Documented in example.collector.yaml and docs/TROUBLESHOOTING_DEVICE_COLLECTOR.md.

Testing

• I have tested this locally
• I have added/updated tests that prove my fix/feature works
• All existing tests pass

go build ./..., go vet ./..., go test -race ./collector/..., go test ./..., GOOS=windows|freebsd|darwin go vet ./collector/... clean, GOOS=linux golangci-lint run --new-from-rev=origin/beta ./... (v2.14.0) 0 issues. New tests cover fatal, timeout, empty, and bad-JSON --info (unregistered, recorded, one scan error each) versus exit-4, virtual-disk exclusion on and off, user patterns including an invalid regex, type exclusion, and explicit devices exempt from built-ins, and parallelism 1/3/0 with a channel barrier for peak concurrency.

Checklist

• My code follows the project's code style
• I have performed a self-review of my code
• I have commented my code where necessary (particularly complex areas)
• I have updated the documentation if needed
• My changes generate no new warnings

Notes

• Breaking (footer on 0a452cd): devices that fail identification are reported as scan errors instead of registered, and virtual-disk models are skipped by default unless listed explicitly or exclude_virtual_disks: false. Previously registered devices in either group stop updating; they are not archived.
• Difference from 1.x: an unidentified device's failure is a scan-level error, so its existing dashboard row is not marked with a collector error.
• The virtual-disk model strings come from [FEAT] Detect virtual disks and skip or mark them unsupported instead of failed #912; not verified against real smartctl output.
• exclude must be a YAML list; the COLLECTOR_EXCLUDE env var splits on whitespace.
• With metrics_parallelism above 1, collector config is read from several goroutines; avoid Set() on it after startup.
• Pre-existing and unrelated: collector/pkg/mdadm/detect/mount_linux.go:98 fails GOOS=linux go vet from arm64 hosts.

Review: not required (collector change on the beta integration branch; no production effect until a release, but an independent read-only review is being run per release coordination)