mirror of
https://github.com/henrygd/beszel.git
synced 2026-08-16 07:17:46 +02:00
feat: fan RPM monitoring (#2032)
Adds fan RPM monitoring as a peer to the existing temperature collection, addressing #1918. --------- Co-authored-by: henrygd <hank@henrygd.me>
This commit is contained in:
101
agent/fans.go
Normal file
101
agent/fans.go
Normal file
@@ -0,0 +1,101 @@
|
||||
package agent
|
||||
|
||||
import (
|
||||
"log/slog"
|
||||
"os"
|
||||
"path/filepath"
|
||||
"strings"
|
||||
"sync"
|
||||
|
||||
"github.com/henrygd/beszel/agent/utils"
|
||||
"github.com/henrygd/beszel/internal/entities/system"
|
||||
)
|
||||
|
||||
type fanSensor struct {
|
||||
key, path string
|
||||
}
|
||||
|
||||
var getFanSensors = newFanSensorCache(hwmonRoot)
|
||||
|
||||
func newFanSensorCache(root string) func() ([]fanSensor, error) {
|
||||
return sync.OnceValues(func() ([]fanSensor, error) {
|
||||
return discoverHwmonFans(root)
|
||||
})
|
||||
}
|
||||
|
||||
// updateFans populates systemStats.Fans from the host's hwmon sysfs tree.
|
||||
// No-op on platforms where hwmon isn't available (see fans_other.go).
|
||||
func (a *Agent) updateFans(systemStats *system.Stats) {
|
||||
if hwmonRoot == "" {
|
||||
return
|
||||
}
|
||||
sensors, err := getFanSensors()
|
||||
if err != nil {
|
||||
slog.Debug("Error reading fans", "err", err)
|
||||
return
|
||||
}
|
||||
fans := readFanSensors(sensors)
|
||||
if len(fans) == 0 {
|
||||
return
|
||||
}
|
||||
systemStats.Fans = fans
|
||||
// Note: Commented out because we don't currently use this value in the UI.
|
||||
// Compute the single "dashboard" value used by the FanSpeed alert.
|
||||
// Per-sensor RPMs live in Stats.Fans and drive the multi-line FanChart
|
||||
// in the UI; the alert path only needs one number to compare against
|
||||
// the user's threshold, so we use the highest RPM across all fans
|
||||
// a.systemInfo.DashboardFan = 0
|
||||
// for _, rpm := range fans {
|
||||
// if rpm > a.systemInfo.DashboardFan {
|
||||
// a.systemInfo.DashboardFan = rpm
|
||||
// }
|
||||
// }
|
||||
}
|
||||
|
||||
// readHwmonFans walks the given hwmon root (typically /sys/class/hwmon) and
|
||||
// returns a map of "<chip>_<label-or-fan-idx>" → RPM for every fan*_input
|
||||
// file it finds. Zero RPM is retained because it can represent a real fan that
|
||||
// has stopped; negative and malformed readings are ignored.
|
||||
func readHwmonFans(root string) (map[string]uint16, error) {
|
||||
sensors, err := discoverHwmonFans(root)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return readFanSensors(sensors), nil
|
||||
}
|
||||
|
||||
func discoverHwmonFans(root string) ([]fanSensor, error) {
|
||||
entries, err := os.ReadDir(root)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
var sensors []fanSensor
|
||||
for _, entry := range entries {
|
||||
chipDir := filepath.Join(root, entry.Name())
|
||||
chipName := utils.ReadStringFile(filepath.Join(chipDir, "name"))
|
||||
if chipName == "" {
|
||||
chipName = entry.Name()
|
||||
}
|
||||
inputs, _ := filepath.Glob(filepath.Join(chipDir, "fan*_input"))
|
||||
for _, inputPath := range inputs {
|
||||
base := strings.TrimSuffix(filepath.Base(inputPath), "_input")
|
||||
label := utils.ReadStringFile(filepath.Join(chipDir, base+"_label"))
|
||||
key := chipName + "_" + base
|
||||
if label != "" {
|
||||
key = chipName + "_" + label
|
||||
}
|
||||
sensors = append(sensors, fanSensor{key, inputPath})
|
||||
}
|
||||
}
|
||||
return sensors, nil
|
||||
}
|
||||
|
||||
func readFanSensors(sensors []fanSensor) map[string]uint16 {
|
||||
fans := make(map[string]uint16, len(sensors))
|
||||
for _, sensor := range sensors {
|
||||
if rpm, ok := utils.ReadUintFile(sensor.path); ok {
|
||||
fans[sensor.key] = uint16(rpm)
|
||||
}
|
||||
}
|
||||
return fans
|
||||
}
|
||||
8
agent/fans_linux.go
Normal file
8
agent/fans_linux.go
Normal file
@@ -0,0 +1,8 @@
|
||||
//go:build linux
|
||||
|
||||
package agent
|
||||
|
||||
// hwmonRoot is the sysfs entry point for hardware monitor chips. Each
|
||||
// subdirectory (hwmon0, hwmon1, …) is one chip; fan*_input files inside it
|
||||
// expose RPM readings.
|
||||
const hwmonRoot = "/sys/class/hwmon"
|
||||
7
agent/fans_other.go
Normal file
7
agent/fans_other.go
Normal file
@@ -0,0 +1,7 @@
|
||||
//go:build !linux
|
||||
|
||||
package agent
|
||||
|
||||
// hwmonRoot is empty on non-Linux platforms — fan RPM reporting via sysfs
|
||||
// hwmon is Linux-specific. updateFans() short-circuits when this is empty.
|
||||
const hwmonRoot = ""
|
||||
87
agent/fans_test.go
Normal file
87
agent/fans_test.go
Normal file
@@ -0,0 +1,87 @@
|
||||
//go:build testing
|
||||
|
||||
package agent
|
||||
|
||||
import (
|
||||
"os"
|
||||
"path/filepath"
|
||||
"testing"
|
||||
|
||||
"github.com/stretchr/testify/assert"
|
||||
"github.com/stretchr/testify/require"
|
||||
)
|
||||
|
||||
// writeFile creates path with parents and writes contents.
|
||||
func writeFile(t *testing.T, path, contents string) {
|
||||
t.Helper()
|
||||
require.NoError(t, os.MkdirAll(filepath.Dir(path), 0o755))
|
||||
require.NoError(t, os.WriteFile(path, []byte(contents), 0o644))
|
||||
}
|
||||
|
||||
// TestReadHwmonFans verifies the /sys/class/hwmon walker:
|
||||
// - picks up fan*_input from every chip,
|
||||
// - keys entries by chip name + sensor label (or fan idx if no label),
|
||||
// - retains 0 RPM for stopped fans,
|
||||
// - tolerates chips with no fan files at all.
|
||||
func TestReadHwmonFans(t *testing.T) {
|
||||
root := t.TempDir()
|
||||
|
||||
// hwmon0: Raspberry Pi 5 active cooler — one fan, no label.
|
||||
writeFile(t, filepath.Join(root, "hwmon0", "name"), "pwmfan\n")
|
||||
writeFile(t, filepath.Join(root, "hwmon0", "fan1_input"), "6500\n")
|
||||
|
||||
// hwmon1: a thermal-only chip, no fan files. Must not error.
|
||||
writeFile(t, filepath.Join(root, "hwmon1", "name"), "cpu_thermal\n")
|
||||
writeFile(t, filepath.Join(root, "hwmon1", "temp1_input"), "55000\n")
|
||||
|
||||
// hwmon2: two fans — one stopped (0 RPM) and one labeled "chassis".
|
||||
writeFile(t, filepath.Join(root, "hwmon2", "name"), "nct6798\n")
|
||||
writeFile(t, filepath.Join(root, "hwmon2", "fan1_input"), "0\n")
|
||||
writeFile(t, filepath.Join(root, "hwmon2", "fan2_input"), "1200\n")
|
||||
writeFile(t, filepath.Join(root, "hwmon2", "fan2_label"), "chassis\n")
|
||||
|
||||
fans, err := readHwmonFans(root)
|
||||
require.NoError(t, err)
|
||||
|
||||
assert.Equal(t, map[string]uint16{
|
||||
"pwmfan_fan1": 6500,
|
||||
"nct6798_fan1": 0,
|
||||
"nct6798_chassis": 1200,
|
||||
}, fans)
|
||||
}
|
||||
|
||||
// TestReadHwmonFansMissingRoot returns an error rather than panicking when the
|
||||
// hwmon root doesn't exist (e.g. running on a kernel without hwmon support).
|
||||
func TestReadHwmonFansMissingRoot(t *testing.T) {
|
||||
_, err := readHwmonFans(filepath.Join(t.TempDir(), "does-not-exist"))
|
||||
assert.Error(t, err)
|
||||
}
|
||||
|
||||
// TestReadHwmonFansEmpty returns an empty map (not nil error) when the root
|
||||
// exists but contains no chips at all.
|
||||
func TestReadHwmonFansEmpty(t *testing.T) {
|
||||
root := t.TempDir()
|
||||
fans, err := readHwmonFans(root)
|
||||
require.NoError(t, err)
|
||||
assert.Empty(t, fans)
|
||||
}
|
||||
|
||||
func TestFanDiscoveryCache(t *testing.T) {
|
||||
root := t.TempDir()
|
||||
input := filepath.Join(root, "hwmon0", "fan1_input")
|
||||
writeFile(t, filepath.Join(root, "hwmon0", "name"), "chip\n")
|
||||
writeFile(t, input, "1000\n")
|
||||
|
||||
getSensors := newFanSensorCache(root)
|
||||
sensors, err := getSensors()
|
||||
require.NoError(t, err)
|
||||
fans := readFanSensors(sensors)
|
||||
assert.Equal(t, uint16(1000), fans["chip_fan1"])
|
||||
|
||||
writeFile(t, input, "1200\n")
|
||||
writeFile(t, filepath.Join(root, "hwmon0", "fan1_label"), "case\n")
|
||||
sensors, err = getSensors()
|
||||
require.NoError(t, err)
|
||||
fans = readFanSensors(sensors)
|
||||
assert.Equal(t, map[string]uint16{"chip_fan1": 1200}, fans)
|
||||
}
|
||||
@@ -210,6 +210,9 @@ func (a *Agent) getSystemStats(cacheTimeMs uint16) system.Stats {
|
||||
// TODO: maybe refactor to methods on systemStats
|
||||
a.updateTemperatures(&systemStats)
|
||||
|
||||
// fan speeds (Linux-only; sysfs hwmon)
|
||||
a.updateFans(&systemStats)
|
||||
|
||||
// GPU data
|
||||
if a.gpuManager != nil {
|
||||
// reset high gpu percent
|
||||
|
||||
@@ -33,6 +33,7 @@ type Stats struct {
|
||||
MaxNetworkSent float64 `json:"nsm,omitempty" cbor:"-"`
|
||||
MaxNetworkRecv float64 `json:"nrm,omitempty" cbor:"-"`
|
||||
Temperatures map[string]float64 `json:"t,omitempty" cbor:"20,keyasint,omitempty"`
|
||||
Fans map[string]uint16 `json:"f,omitempty" cbor:"36,keyasint,omitempty"`
|
||||
ExtraFs map[string]*FsStats `json:"efs,omitempty" cbor:"21,keyasint,omitempty"`
|
||||
GPUData map[string]GPUData `json:"g,omitempty" cbor:"22,keyasint,omitempty"`
|
||||
// LoadAvg1 float64 `json:"l1,omitempty" cbor:"23,keyasint,omitempty"`
|
||||
|
||||
@@ -191,6 +191,8 @@ func AverageSystemStatsSlice(records []system.Stats) system.Stats {
|
||||
// accumulate cpu breakdown [user, system, iowait, steal, idle]
|
||||
var cpuBreakdownSums []float64
|
||||
tempCount := float64(0)
|
||||
var fanSums map[string]uint64
|
||||
fanCount := uint64(0)
|
||||
|
||||
// Accumulate totals
|
||||
for i := range records {
|
||||
@@ -282,6 +284,17 @@ func AverageSystemStatsSlice(records []system.Stats) system.Stats {
|
||||
}
|
||||
}
|
||||
|
||||
// Accumulate fan speeds
|
||||
if stats.Fans != nil {
|
||||
if fanSums == nil {
|
||||
fanSums = make(map[string]uint64, len(stats.Fans))
|
||||
}
|
||||
fanCount++
|
||||
for key, value := range stats.Fans {
|
||||
fanSums[key] += uint64(value)
|
||||
}
|
||||
}
|
||||
|
||||
// Accumulate extra filesystem stats
|
||||
if stats.ExtraFs != nil {
|
||||
if sum.ExtraFs == nil {
|
||||
@@ -387,6 +400,14 @@ func AverageSystemStatsSlice(records []system.Stats) system.Stats {
|
||||
}
|
||||
}
|
||||
|
||||
// Average fan speeds
|
||||
if fanSums != nil && fanCount > 0 {
|
||||
sum.Fans = make(map[string]uint16, len(fanSums))
|
||||
for key, value := range fanSums {
|
||||
sum.Fans[key] = uint16(value / fanCount)
|
||||
}
|
||||
}
|
||||
|
||||
// Average extra filesystem stats
|
||||
if sum.ExtraFs != nil {
|
||||
for key := range sum.ExtraFs {
|
||||
|
||||
@@ -291,6 +291,28 @@ func TestAverageSystemStatsSlice_Temperatures(t *testing.T) {
|
||||
assert.Equal(t, 80.0, result.Temperatures["gpu"])
|
||||
}
|
||||
|
||||
// Tests that fan speeds are averaged and records without fan data are excluded.
|
||||
func TestAverageSystemStatsSlice_Fans(t *testing.T) {
|
||||
input := []system.Stats{
|
||||
{
|
||||
Fans: map[string]uint16{"cpu": 60_000, "case": 1_000},
|
||||
},
|
||||
{
|
||||
Fans: map[string]uint16{"cpu": 50_000, "case": 2_000},
|
||||
},
|
||||
{
|
||||
// No fan data - should not affect fan averaging
|
||||
Cpu: 30.0,
|
||||
},
|
||||
}
|
||||
|
||||
result := records.AverageSystemStatsSlice(input)
|
||||
|
||||
require.NotNil(t, result.Fans)
|
||||
assert.Equal(t, uint16(55_000), result.Fans["cpu"])
|
||||
assert.Equal(t, uint16(1_500), result.Fans["case"])
|
||||
}
|
||||
|
||||
func TestAverageSystemStatsSlice_NetworkInterfaces(t *testing.T) {
|
||||
input := []system.Stats{
|
||||
{
|
||||
|
||||
@@ -9,7 +9,7 @@ import { CpuChart, ContainerCpuChart } from "./system/charts/cpu-charts"
|
||||
import { MemoryChart, ContainerMemoryChart, SwapChart } from "./system/charts/memory-charts"
|
||||
import { RootDiskCharts, ExtraFsCharts } from "./system/charts/disk-charts"
|
||||
import { BandwidthChart, ContainerNetworkChart } from "./system/charts/network-charts"
|
||||
import { TemperatureChart, BatteryChart } from "./system/charts/sensor-charts"
|
||||
import { TemperatureChart, FanChart, BatteryChart } from "./system/charts/sensor-charts"
|
||||
import { GpuPowerChart, GpuDetailCharts } from "./system/charts/gpu-charts"
|
||||
import { LazyContainersTable, LazySmartTable, LazySystemdTable } from "./system/lazy-tables"
|
||||
import { LoadAverageChart } from "./system/charts/load-average-chart"
|
||||
@@ -123,6 +123,8 @@ export default memo(function SystemDetail({ id }: { id: string }) {
|
||||
|
||||
<TemperatureChart {...coreProps} />
|
||||
|
||||
<FanChart {...coreProps} />
|
||||
|
||||
<BatteryChart {...coreProps} />
|
||||
|
||||
{hasGpuPowerData && <GpuPowerChart chartData={chartData} grid={grid} dataEmpty={dataEmpty} />}
|
||||
@@ -188,6 +190,7 @@ export default memo(function SystemDetail({ id }: { id: string }) {
|
||||
<LoadAverageChart chartData={chartData} grid={grid} dataEmpty={dataEmpty} />
|
||||
<BandwidthChart {...coreProps} systemStats={systemStats} />
|
||||
<TemperatureChart {...coreProps} setPageBottomExtraMargin={setPageBottomExtraMargin} />
|
||||
<FanChart {...coreProps} />
|
||||
<BatteryChart {...coreProps} />
|
||||
<SwapChart chartData={chartData} grid={grid} dataEmpty={dataEmpty} systemStats={systemStats} />
|
||||
{pageBottomExtraMargin > 0 && <div style={{ marginBottom: pageBottomExtraMargin }}></div>}
|
||||
|
||||
@@ -1,7 +1,7 @@
|
||||
import { t } from "@lingui/core/macro"
|
||||
import AreaChartDefault from "@/components/charts/area-chart"
|
||||
import { batteryStateTranslations } from "@/lib/i18n"
|
||||
import { $temperatureFilter, $userSettings } from "@/lib/stores"
|
||||
import { $fanFilter, $temperatureFilter, $userSettings } from "@/lib/stores"
|
||||
import { cn, decimalString, formatTemperature, toFixedFloat } from "@/lib/utils"
|
||||
import type { ChartData, SystemStatsRecord } from "@/types"
|
||||
import { ChartCard, FilterBar } from "../chart-card"
|
||||
@@ -209,3 +209,102 @@ export function TemperatureChart({
|
||||
</div>
|
||||
)
|
||||
}
|
||||
|
||||
export function FanChart({
|
||||
chartData,
|
||||
grid,
|
||||
dataEmpty,
|
||||
}: {
|
||||
chartData: ChartData
|
||||
grid: boolean
|
||||
dataEmpty: boolean
|
||||
}) {
|
||||
const showFanChart = chartData.systemStats.at(-1)?.stats.f
|
||||
|
||||
const filter = useStore($fanFilter)
|
||||
|
||||
const statsRef = useRef(chartData.systemStats)
|
||||
statsRef.current = chartData.systemStats
|
||||
|
||||
// Stable key derived from current sensor names (sorted) so the memo only
|
||||
// recomputes when the set of fans changes — not on every data point.
|
||||
let sensorNamesKey = ""
|
||||
for (let i = chartData.systemStats.length - 1; i >= 0; i--) {
|
||||
const f = chartData.systemStats[i].stats?.f
|
||||
if (f) {
|
||||
sensorNamesKey = Object.keys(f).sort().join("\0")
|
||||
break
|
||||
}
|
||||
}
|
||||
|
||||
const { colorMap, dataKeys, sortedKeys } = useMemo(() => {
|
||||
const stats = statsRef.current
|
||||
const sums = {} as Record<string, number>
|
||||
for (const data of stats) {
|
||||
const f = data.stats?.f
|
||||
if (!f) continue
|
||||
for (const key of Object.keys(f)) {
|
||||
sums[key] = (sums[key] ?? 0) + f[key]
|
||||
}
|
||||
}
|
||||
const sorted = Object.keys(sums).sort((a, b) => sums[b] - sums[a])
|
||||
const colorMap = {} as Record<string, string>
|
||||
const dataKeys = {} as Record<string, (d: SystemStatsRecord) => number | undefined>
|
||||
for (let i = 0; i < sorted.length; i++) {
|
||||
const key = sorted[i]
|
||||
colorMap[key] = `hsl(${((i * 360) / sorted.length) % 360}, 60%, 55%)`
|
||||
dataKeys[key] = (d: SystemStatsRecord) => d.stats?.f?.[key]
|
||||
}
|
||||
return { colorMap, dataKeys, sortedKeys: sorted }
|
||||
}, [sensorNamesKey])
|
||||
|
||||
const dataPoints = useMemo(() => {
|
||||
return sortedKeys.map((key) => {
|
||||
const filterTerms = filter
|
||||
? filter
|
||||
.toLowerCase()
|
||||
.split(" ")
|
||||
.filter((term) => term.length > 0)
|
||||
: []
|
||||
const filtered = filterTerms.length > 0 && !filterTerms.some((term) => key.toLowerCase().includes(term))
|
||||
const strokeOpacity = filtered ? 0.1 : 1
|
||||
return {
|
||||
label: key,
|
||||
dataKey: dataKeys[key],
|
||||
color: colorMap[key],
|
||||
strokeOpacity,
|
||||
activeDot: !filtered,
|
||||
}
|
||||
})
|
||||
}, [sortedKeys, filter, dataKeys, colorMap])
|
||||
|
||||
if (!showFanChart) {
|
||||
return null
|
||||
}
|
||||
|
||||
const legend = dataPoints.length < 12
|
||||
|
||||
return (
|
||||
<div className={cn("odd:last-of-type:col-span-full", { "col-span-full": !grid })}>
|
||||
<ChartCard
|
||||
empty={dataEmpty}
|
||||
grid={grid}
|
||||
title={t`Fans`}
|
||||
description={t`System fan speeds (RPM)`}
|
||||
cornerEl={<FilterBar store={$fanFilter} />}
|
||||
legend={legend}
|
||||
>
|
||||
<LineChartDefault
|
||||
chartData={chartData}
|
||||
itemSorter={(a, b) => b.value - a.value}
|
||||
domain={["auto", "auto"]}
|
||||
legend={legend}
|
||||
tickFormatter={(val) => `${toFixedFloat(val, 0)}`}
|
||||
contentFormatter={(item) => `${decimalString(item.value, 0)} RPM`}
|
||||
dataPoints={dataPoints}
|
||||
filter={filter}
|
||||
></LineChartDefault>
|
||||
</ChartCard>
|
||||
</div>
|
||||
)
|
||||
}
|
||||
|
||||
@@ -64,6 +64,9 @@ export const $containerFilter = atom("")
|
||||
/** Temperature chart filter */
|
||||
export const $temperatureFilter = atom("")
|
||||
|
||||
/** Fan-speed chart filter */
|
||||
export const $fanFilter = atom("")
|
||||
|
||||
/** Fallback copy to clipboard dialog content */
|
||||
export const $copyContent = atom("")
|
||||
|
||||
|
||||
2
internal/site/src/types.d.ts
vendored
2
internal/site/src/types.d.ts
vendored
@@ -143,6 +143,8 @@ export interface SystemStats {
|
||||
bm?: [number, number]
|
||||
/** temperatures */
|
||||
t?: Record<string, number>
|
||||
/** fan speeds (RPM) — keyed by `<chip>_<label-or-fan-idx>` */
|
||||
f?: Record<string, number>
|
||||
/** extra filesystems */
|
||||
efs?: Record<string, ExtraFsStats>
|
||||
/** GPU data */
|
||||
|
||||
@@ -16,7 +16,7 @@ It has a friendly web interface, simple configuration, and is ready to use out o
|
||||
- **Lightweight**: Smaller and less resource-intensive than leading solutions.
|
||||
- **Simple**: Easy setup with little manual configuration required.
|
||||
- **Docker stats**: Tracks CPU, memory, and network usage history for each container.
|
||||
- **Alerts**: Configurable alerts for CPU, memory, disk, bandwidth, temperature, load average, and status.
|
||||
- **Alerts**: Configurable alerts for CPU, memory, disk, bandwidth, temperature, fan speed, load average, and status.
|
||||
- **Multi-user**: Users manage their own systems. Admins can share systems across users.
|
||||
- **OAuth / OIDC**: Supports many OAuth2 providers. Password auth can be disabled.
|
||||
- **Automatic backups**: Save to and restore from disk or S3-compatible storage.
|
||||
@@ -48,6 +48,7 @@ The [quick start guide](https://beszel.dev/guide/getting-started) and other docu
|
||||
- **Network usage** - Host system and containers.
|
||||
- **Load average** - Host system.
|
||||
- **Temperature** - Host system sensors.
|
||||
- **Fan speed** - Host system sensors (Linux, via `/sys/class/hwmon`).
|
||||
- **GPU usage / power draw** - Nvidia, AMD, and Intel.
|
||||
- **Battery** - Host system battery charge.
|
||||
- **Containers** - Status and metrics of all running Docker / Podman containers.
|
||||
|
||||
Reference in New Issue
Block a user