Add host temperature telemetry and history

This commit is contained in:
2026-10-13 23:39:04 -04:00
parent 09396ab584
commit 171f657876
28 changed files with 1466 additions and 32 deletions
+34
View File
@@ -354,6 +354,40 @@ journalctl -u electric-cherry-update.service -n 100
## Placement and resource capacity
### Host temperature
Agent **1.10.0** reports temperature sensors exposed by supported Linux hwmon
drivers without installing additional packages. **Hosts** shows CPU temperature
beneath CPU usage. A host's **Overview → Inventory** shows the selected reading
and expandable **Temperature sensors**, including per-core and storage details.
Stale readings are marked as last reported; absent or failed readings are
**Unavailable**, never zero or a readiness failure.
**Fleet → Capacity → Temperature history** uses the existing host selector and
Last hour / 24 hours / 7 days controls. **All hosts** displays a separate CPU line
for each host. Select one host to add its NVMe composite and chipset lines using
the legend. Hover, tap, or use the arrow keys on the chart to inspect a period.
Each host page's **Capacity** button opens the view filtered to that host.
The game filter affects launch estimates, not temperatures.
The CPU reading is the hottest reported package, falling back to a physical
AMD die/CCD and then a core reading. AMD Tctl cooling-control values are excluded
because they may include an offset. Supported device drivers are `coretemp`,
`k10temp`, `nvme` and `pch_*`; sensor availability depends on hardware and drivers.
Virtual machines and older agents may have no supported readings. Missing,
faulty and disabled sensors are skipped individually.
Temperature history starts after the agent upgrade. The controller stores one
peak per sensor per minute, limited to the CPU summary and up to sixteen NVMe
composite/chipset inputs per host. Longer chart periods show the maximum of
those peaks; a period without readings remains a gap. Brief gaps within an
aggregated period may still leave a peak for that period. **Host resource history
retention** applies to CPU, memory and temperature samples (seven days by default;
zero keeps them indefinitely). Series colors identify hosts/sensors and do not
indicate thermal safety. Temperature does not change placement or trigger alerts.
### CPU and memory capacity
### Host inventory and eligibility
The Inventory panel shows CPU usage with agent 1.4.2 and later. It measures