Skip to content

Diagnostics dashboard

The Diagnostics dashboard (Grafana folder Advanced) is for the person who maintains a Smoking Pi, not for the question "is my connection fine". It answers the questions that come up when another dashboard looks wrong: is every collector still writing, which target series stopped, did the copy in InfluxDB have a gap, and what do the services record about their own runs. It is the InfluxDB tree's only dashboard about the monitor itself; the ClickHouse backend has no version of it.

Open it from the Overview's Where to look next list or from the SmokePing dashboards dropdown. Its default range is the last 24 hours.

What it shows

Row Panel What it answers
Collectors SmokePing probes, by layer How old the newest ICMP, DNS, TCP, HTTP and ISP-gateway point is. Orange past two steps and a minute (the rule the web admin applies per target), red past half an hour
Host collectors The same for the Wi-Fi sampler, the uplink byte meter, the public-address and resolver-identity collectors and the uplink heartbeat, each against its own cadence
Services The same for the inference module's passes, the budget report, the DNS wizard's hourly snapshot and the house's DNS timing
Series that stopped Silent targets Every target that wrote in the last 24 hours and has been silent longer than its window, longest silence first. Click a target for every layer it is measured on
Silent targets (count), Series with data (24 h) The table's size and the population it is drawn from. 0 silent is the normal reading; a target you just removed, paused or renamed reads silent for up to 24 h, until its last point ages out of the window
Points written per 15 minutes, by layer A flat line per layer is normal. A dip every layer shares is a gap in the InfluxDB copy (the database down, the card full, the exporter restarting); a dip in one layer is that probe
The services, about themselves Inference passes Seconds and failures per hourly pass, moved here from Connectivity & Quality (it is the module's health, not the connection's)
What the watchdog did Driver reloads, reboots and SmokePing restarts the host watchdog recorded (watchdog_event). Empty is the healthy reading
Loss incidents the alerter diagnosed One row per closed incident: class, summary, minutes, confidence (low, medium or high), targets affected
Measurement budget report Whether the budget exporter priced every probe, how many it could not, and whether the estimate is over the ceiling. These flags were written since v2.16.0 and shown nowhere until now
Text Empty on purpose; where the rest lives Why a panel is blank on this hardware, and which command or card holds the health that never reaches InfluxDB

What it cannot show

The dashboard reads InfluxDB, so it only knows about collectors that write there. A target that was configured but never measured has no series to be silent: that is the web admin's Measurements card and the doctor's targets-measured and silent-series checks, which compare against the configured targets. The DNS observer, the mDNS responder and the alerter keep no heartbeat in the database; their state is in smoking-pi dns status, the container healthchecks (smoking-pi status) and the web admin's cards. The alerter's dead-man check-in lives off the host on purpose (alerting).

The watchdog panel is empty until the watchdog has had to act. Events are recorded in InfluxDB since v2.32.0; smoking-pi watchdog status shows the full log, including the enable and disable marks, which are not written to the database.

Freshness rule

"Silent" means the last point is older than two steps plus a minute, the same rule config-manager's freshness module applies to the RRD files for the web admin's card (one missed step is a slow probe round, two is a problem). The step comes from the step field the exporter writes with every point; the ISP gateway's high-frequency probe writes about once a minute and has its own thresholds. The per-target query runs over the last 24 hours and took 0.4 s on the reference Pi with 247 targets.

See also