devices

Imported from rknightion/meraki-dashboard-exporter (src/meraki_dashboard_exporter/collectors/devices/AGENTS.md). Install upstream with npx skills add rknightion/meraki-dashboard-exporter --skill devices. Copyright stays with the author.

Device collectors

Two sub-collector shapes

• Per-device-type collectors extend BaseDeviceCollector (base.py) and expose async collect(device), which DeviceCollector calls once per device.
• Metric-domain sub-collectors that a device-type collector composes internally (mx_firewall.py, mx_vpn.py, ms_stack.py, and the mr/ submodules) do not extend BaseDeviceCollector. They take a parent in __init__, mostly mix in SubCollectorMixin, and expose their own collect/collect_for_network/collect_* methods called directly by the owning coordinator. There is no uniform interface across them, so do not assume one when wiring a new call site.

mr/ is the only device type split into its own subpackage.

Gauges

Create through self.parent._create_gauge(), never a direct Gauge(). Set through self.parent._set_metric() so expiration tracking works. This differs from network_health_collectors/ and organization_collectors/, where the gauge is created on the coordinator and set by attribute-name string.

_set_metric buckets a series under the calling collector's own class name, and because device sub-collectors delegate to the parent, every device series is tracked as DeviceCollector.

Ownership is not what the filename suggests

DeviceCollector.__init__ instantiates the MX org-wide sub-collectors directly (mx_ha_collector, mx_uplink_usage_collector, mx_uplink_health_collector) alongside ms_stack_collector and ms_power_collector. They are not owned by mx.py or ms.py despite the filenames, so a change to the MX coordinator does not reach them.

Never clear a shared gauge's label series

An org-wide sub-collector method runs once per org, concurrently across orgs, against one shared gauge instance. Clearing the gauge's series there wipes every other org's series mid-cycle. Stale label series after a status transition are removed by the metric expiration manager via parent._set_metric tracking instead.

Device family quirks

• MS: the org-level getOrganizationSwitchPortsStatusesBySwitch endpoint is absent on older SDK versions. MSCollector probes for it once (_org_port_status_supported, cached) and falls back to per-device getDeviceSwitchPortsStatuses. PoE budget is not exposed by the API at all; the gauge exists but stays unpopulated and would need a per-model lookup table.
• MS stacks: PSU, fan and temperature hardware health is deliberately deferred pending Meraki API investigation, not missing by accident.
• MX: mx_vpn.py merges Meraki and third-party peers out of one org-level response. A third-party peer carries no networkId and is keyed by its public IP instead, so a path that assumes networkId is present drops or mislabels them.
• MX performance score: mx.py records each physical MX's request attempt before calling getDeviceAppliancePerformance. Errors, including an unsupported-feature response, must remain gated by the scheduler-solved mx_performance interval (floor 1800s), or they retry every device collection cycle.
• MT: MTCollector has two constructors, as_subcollector (under DeviceCollector) and as_standalone (used by ../mt_sensor.py), so a change to mt.py lands in two collectors. Responses can carry both temperature and rawTemperature. Only temperature is used; rawTemperature is undocumented and deliberately skipped.
• MG: per-device collect() is a no-op by design. The real cellular metrics come from the org-wide collect_uplink_statuses(), so do not add a per-device cellular fetch on the assumption one is missing.
• MG and MV: response contracts for surfaces the available hardware cannot exercise are modelled deliberately lenient and marked unverified in the source. Verify against the live API before tightening one.