The first hour
What happens after docker compose up -d, and what is worth tuning in the first days.
Right away
- The first sweep runs within a minute: every device in the inventory is probed, and the
dashboard fills. Until then a grey bar says it is under way; it turns red only if no sweep
has come three minutes after the start. A device is shown down only after two missed sweeps in a row
(
cadence.debounce_fails), so a single lost ping never turns anything red. - The internet gets its own loop: the addresses in
wan.targetsare pinged every 15 seconds. The dashboard's Internet card fills as the record grows. - The owl's first audit (with a model set, The model) runs at the first sweep, then every hour
(
cadence.llm_interval_s). It sends a digest only if it has something new to say; on a healthy network you hear nothing.
With a MikroTik
- The router's DHCP is read every five minutes (
mikrotik.discovery_interval_s). The first read is a baseline: what is already on your network is not announced as new. - Under each site, the dashboard lists the devices on its DHCP that nobody watches, with two
buttons. Watch writes one into
inventory.yamland pings it at once; Known marks it as belonging there, so it no longer counts as unknown. The family's phones are Known; the printer is worth a Watch.
In the morning
With the update check on (updates.enabled), lanowl reads every managed machine's waiting
updates at 06:30 (updates.at), then the owl reviews what each machine exposes
(exposure.enabled). If lanowl starts after that time, the first check runs at once. Only
what is new and matters reaches Telegram; everything is on the Security tab.
The weekly review arrives on Sundays at 10:00 (weekly).
Tuning in the first days
The first days show what is noise on your network. Most of it is settled on the device's page
(Settings → Devices), or in inventory.yaml by hand:
| You see | Change |
|---|---|
| A page for something that can wait until morning | Lower its criticality: high and warning ride the next digest, low and info are told once and never page |
| Solar gear "down" every night, a dusk-to-dawn light every day | expect_offline: sun or day |
| A plug on weak Wi-Fi going down and up | Give its check more patience: {type: icmp, timeout_ms: 2500, count: 3} |
| Five cameras, five messages, when their switch restarts | depends_on: <the switch's address> on each camera, or majority_down_critical: true on their group |
| Everything at a remote site paging when its tunnel drops | depends_on the tunnel's endpoint, and a larger debounce_fails for it |
| A device you switch off on purpose (the TV, holiday gear) | Pause it: /pause tv on Telegram, or its sheet on the dashboard |
| A diagnosis that misses something only you know | Tell the owl: /remember the NAS sleeps from 01:00 to 07:00 |
Every one of these is explained in Devices and How alerts work.
Edit this page on GitHub.