Repository navigation
Device capability probing
Metric definitions now span five MIB families, and no device implements them all. Configuration → Metrics → Probe device asks each device what it actually supports:
- Scalar metrics tested in one multi-varbind SNMP GET; anything that answers is supported.
- Indexed metrics have their instance name column walked, so sensors, disks, storage and interfaces arrive named. You tick "temp-cpu", not index 4.
- The configuration screen lists only supported metrics and collapses the rest, which stays expandable in case a firmware update adds an OID. Each row shows the reading seen while probing, so a dead sensor is obvious.
- Enable all switches on every supported scalar in one click.
Probing takes 0.1 to 0.7 seconds on UniFi hardware.
New metrics
15 definitions from the UCD-SNMP tree (load at 1/5/15 minutes, CPU idle/user/system, memory total/available/buffers/cached, swap, disk, TCP connections, uptime) plus Sensor Temperature from the lm-sensors table. Together these fill the gap on UDM Pro and other Net-SNMP hosts, which implement none of the CPU or temperature OIDs shipped previously.
Fixes
- Saving the Metrics tab no longer resets the node's other metrics; rows keep their IDs so values, rates, alert state and history survive edits.
- A node with SNMP switched off now explains why there is no probe button, and SNMP metrics configured on such a node are flagged instead of silently showing dashes.
- Metric values round to two decimals. Uptime renders as "11d 18h", millidegree sensors as °C.
- The Metrics page lists ping-only nodes, so ICMP latency and packet loss are visible.
Upgrade
git pull && docker compose up -d --build --force-recreate
Migration v4 runs automatically. Existing metric configuration is untouched and an unprobed node behaves exactly as before. Afterwards, press Probe device on each SNMP node.