[flyingnobita]

Description

Fixes #3806: a disk whose smartctl --scan type pySMART does not accept (USB bridges such as -d usbsunplus, usbjmicron) no longer takes out SMART data for every disk.

On develop this is not a crash: the per-plugin guard in GlancesPluginBase.update() catches the ValueError, but get_smart_data() then returns nothing at all, healthy disks included, and logs Plugin smart update failed every cycle. On v4 (4.5.8, as reported in the issue) there is no such guard and Glances exits at startup. A backport to support/glancesv4 will follow if you want it.

 get_smart_data(hide_attributes)
+  attach _SkippedDeviceFilter to the root logger
   try
-    DeviceList()
+    DeviceList(catch_errors=True)   # pySMART skips the bad device, keeps the others
-  except (TypeError, FileNotFoundError)
+  except (TypeError, FileNotFoundError, ValueError)
     import_error_tag = True
+  finally
+    detach the filter

Why the filter: with catch_errors=True, pySMART calls logging.exception("Error parsing device ...") on the root logger (it has no named logger) on every refresh. In the Docker image (LOG_CFG=/app/docker-logger.json) that is a full traceback in docker logs every few seconds. The filter turns the first such record per device into one WARNING without traceback and drops the repeats. It only matches pySMART's Error parsing device record (by pathname and raw message), never formats other records, and is attached only around the DeviceList() call.

Alternative, your call: drop the filter and keep only the 2-line catch_errors=True change. The plugin then works, at the cost of the repeated tracebacks shown below.

Evidence

Real hardware: Sunplus USB bridge (1bcf:0c31) reported as /dev/sda -d usbsunplus, plus an NVMe. Image nicolargo/glances:dev (5.0.0b1, pySMART 1.4.3, smartctl 7.5), patched data.py bind-mounted, --enable-plugin smart.

	Before	catch_errors only	This PR
/api/5/smart	null	nvme0 (23 attributes), no sda	nvme0 (23 attributes), no sda
Plugin smart update failed	17 in 68 s	0	0
Error parsing device tracebacks	0	18 in 68 s	0 in 96 s
Smart plugin - Skipping	-	-	1

docker logs with this PR:

2026-10-21 15:44:27,169 -- WARNING -- Smart plugin - Skipping /dev/sda: Unknown interface: usbsunplus specified for /dev/sda

Tests in tests/test_smart_data.py (written first, red on develop):

• test_unknown_interface_skips_device_keeps_others: drives the real pySMART.DeviceList.initialize() with an nvme + usbsunplus scan; returns the NVMe, plugin stays enabled.
• test_device_list_value_error_disables_plugin: whole-list ValueError returns [] and sets import_error_tag.
• test_skipped_device_logs_one_warning_without_traceback: two refreshes, one WARNING per device, no traceback.
• test_root_logger_filters_restored, test_unrelated_root_errors_not_suppressed, test_filter_never_raises: the filter is scoped and cannot break other logging.
• test_missing_smartctl_returns_no_device: its fake now accepts **kwargs, otherwise it would pass through the TypeError branch instead of FileNotFoundError.

make test: 4162 passed, 20 skipped. ruff check, ruff format --check and pre-commit run --all-files pass.

Notes

• catch_errors exists since pySMART 1.2.3. On older versions the call raises TypeError, which is caught, so SMART is disabled rather than crashing. pyproject.toml does not pin pySMART; pinning pySMART>=1.2.3 is left to you.
• A per-device error of any type now skips that device instead of disabling the plugin for all of them.
• Out of scope, already present: a scan line without -d <type> makes pySMART raise AttributeError outside its try.

Resume

• Bug fix: yes
• New feature: no
• Fixed tickets: Smart plugin: a USB-bridged disk (e.g. -d usbsunplus) crashes the web server at startup #3806