fankes/beszel
2
0
Fork 0
mirror of https://github.com/fankes/beszel.git synced 2025-10-19 01:39:34 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Probable fix for Jetson gpu issue (#895)

Browse Source
This commit is contained in:
henrygd
2025-06-26 22:11:48 -04:00
parent 8c52f30a71
commit 4395520a28
2 changed files with 106 additions and 45 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
beszel/internal/agent/gpu.go
View File

@@ -243,21 +243,26 @@ func (gm *GPUManager) GetCurrentData() map[string]system.GPUData {
// copy / reset the data
gpuData := make(map[string]system.GPUData, len(gm.GpuDataMap))
for id, gpu := range gm.GpuDataMap {
// sum the data
gpu.Temperature = twoDecimals(gpu.Temperature)
gpu.MemoryUsed = twoDecimals(gpu.MemoryUsed)
gpu.MemoryTotal = twoDecimals(gpu.MemoryTotal)
gpu.Usage = twoDecimals(gpu.Usage / gpu.Count)
gpu.Power = twoDecimals(gpu.Power / gpu.Count)
// reset the count
gpu.Count = 1
// dereference to avoid overwriting anything else
gpuCopy := *gpu
var gpuAvg system.GPUData
gpuAvg.Temperature = twoDecimals(gpu.Temperature)
gpuAvg.MemoryUsed = twoDecimals(gpu.MemoryUsed)
gpuAvg.MemoryTotal = twoDecimals(gpu.MemoryTotal)
// avoid division by zero
if gpu.Count > 0 {
gpuAvg.Usage = twoDecimals(gpu.Usage / gpu.Count)
gpuAvg.Power = twoDecimals(gpu.Power / gpu.Count)
}
// reset accumulators in the original
gpu.Usage, gpu.Power, gpu.Count = 0, 0, 0
// append id to the name if there are multiple GPUs with the same name
if nameCounts[gpu.Name] > 1 {
gpuCopy.Name = fmt.Sprintf("%s %s", gpu.Name, id)
gpuAvg.Name = fmt.Sprintf("%s %s", gpu.Name, id)
}
gpuData[id] = gpuCopy
gpuData[id] = gpuAvg
}
slog.Debug("GPU", "data", gpuData)
return gpuData

64
beszel/internal/agent/gpu_test.go
View File

@@ -279,6 +279,19 @@ func TestParseJetsonData(t *testing.T) {
Count: 1,
},
},
{
name: "orin nano",
input: "06-18-2025 11:25:24 RAM 3452/7620MB (lfb 25x4MB) SWAP 1518/16384MB (cached 174MB) CPU [1%@1420,2%@1420,0%@1420,2%@1420,2%@729,1%@729] GR3D_FREQ 0% cpu@50.031C soc2@49.031C soc0@50C gpu@49.031C tj@50.25C soc1@50.25C VDD_IN 4824mW/4824mW VDD_CPU_GPU_CV 518mW/518mW VDD_SOC 1475mW/1475mW",
wantMetrics: &system.GPUData{
Name: "GPU",
MemoryUsed: 3452.0,
MemoryTotal: 7620.0,
Usage: 0.0,
Temperature: 50.25,
Power: 0.518,
Count: 1,
},
},
{
name: "missing temperature",
input: "11-14-2024 22:54:33 RAM 4300/30698MB GR3D_FREQ 45% VDD_GPU_SOC 2171mW",
@@ -318,6 +331,7 @@ func TestParseJetsonData(t *testing.T) {
}
func TestGetCurrentData(t *testing.T) {
t.Run("calculates averages and resets accumulators", func(t *testing.T) {
gm := &GPUManager{
GpuDataMap: map[string]*system.GPUData{
"0": {
@@ -347,15 +361,45 @@ func TestGetCurrentData(t *testing.T) {
assert.Equal(t, "GPU1 0", result["0"].Name)
assert.Equal(t, "GPU1 1", result["1"].Name)
// Check averaged values
// Check averaged values in the result
assert.InDelta(t, 50.0, result["0"].Usage, 0.01)
assert.InDelta(t, 100.0, result["0"].Power, 0.01)
assert.InDelta(t, 30.0, result["1"].Usage, 0.01)
assert.InDelta(t, 60.0, result["1"].Power, 0.01)
// Verify reset counts
assert.Equal(t, float64(1), gm.GpuDataMap["0"].Count)
assert.Equal(t, float64(1), gm.GpuDataMap["1"].Count)
// Verify that accumulators in the original map are reset
assert.Equal(t, float64(0), gm.GpuDataMap["0"].Count, "GPU 0 Count should be reset")
assert.Equal(t, float64(0), gm.GpuDataMap["0"].Usage, "GPU 0 Usage should be reset")
assert.Equal(t, float64(0), gm.GpuDataMap["0"].Power, "GPU 0 Power should be reset")
assert.Equal(t, float64(0), gm.GpuDataMap["1"].Count, "GPU 1 Count should be reset")
assert.Equal(t, float64(0), gm.GpuDataMap["1"].Usage, "GPU 1 Usage should be reset")
assert.Equal(t, float64(0), gm.GpuDataMap["1"].Power, "GPU 1 Power should be reset")
})
t.Run("handles zero count without panicking", func(t *testing.T) {
gm := &GPUManager{
GpuDataMap: map[string]*system.GPUData{
"0": {
Name: "TestGPU",
Count: 0,
Usage: 0,
Power: 0,
},
},
}
var result map[string]system.GPUData
assert.NotPanics(t, func() {
result = gm.GetCurrentData()
})
// Check that usage and power are 0
assert.Equal(t, 0.0, result["0"].Usage)
assert.Equal(t, 0.0, result["0"].Power)
// Verify reset count
assert.Equal(t, float64(0), gm.GpuDataMap["0"].Count)
})
}
func TestDetectGPUs(t *testing.T) {
@@ -722,6 +766,18 @@ func TestAccumulation(t *testing.T) {
assert.InDelta(t, expected.avgUsage, gpu.Usage, 0.01, "Average usage in GetCurrentData should match")
assert.InDelta(t, expected.avgPower, gpu.Power, 0.01, "Average power in GetCurrentData should match")
}
// Verify that accumulators in the original map are reset
for id := range tt.expectedValues {
gpu, exists := gm.GpuDataMap[id]
assert.True(t, exists, "GPU with ID %s should still exist after GetCurrentData", id)
if !exists {
continue
}
assert.Equal(t, float64(0), gpu.Count, "Count should be reset for GPU ID %s", id)
assert.Equal(t, float64(0), gpu.Usage, "Usage should be reset for GPU ID %s", id)
assert.Equal(t, float64(0), gpu.Power, "Power should be reset for GPU ID %s", id)
}
})
}
}