Shelly operator¶

Purpose¶

Reconciles Shelly smart devices from ShellyDevice custom resources in the home namespace, alongside the shelly-prometheus-exporter metrics scraper. app/prometheusrule.yaml ships fleet-health alerts on the operator's own shelly_device_* metrics; power and temperature alerts live in shelly-prometheus-exporter's separate PrometheusRule.

Exporter poll interval¶

exporterDeviceUpdateInterval is set to 60s, matching the shelly-prometheus-exporter ServiceMonitor's 60s scrape interval, instead of the chart's 30s default.

That poll accounts for about 98% of all RPC load on the fleet: the exporter makes about 6 calls per device per cycle, against about 1 per device per reconcileInterval from the operator itself. At the 30s default, half the samples were overwritten before Prometheus read them, so the extra polling bought no stored resolution and doubled device traffic.

Device firmware 2.0.0 also rate-limits under that load and starts answering HTTP 429, which left the operator unable to read config on roughly a third of the fleet, reporting InSync=Unknown for them (unverified, not the same as healthy). RPC dropped from about 12 to about 6 calls per minute per device after the change (upstream shelly-operator#54).

Raise the ServiceMonitor scrape interval and this setting together. Raising one alone brings back either the oversample or the rate limiting.

Restart-required calibration¶

A manual sweep before ShellyDeviceRestartRequired shipped found four devices pending a restart, the oldest for four and a half days, with no other record of it. The alert waits for: 6h, shorter than the 24h used for a pending firmware update, because nothing clears a pending restart on its own.