nvidia: improve component mapping, firmware, statuses and check times

This commit is contained in:
2026-02-16 23:17:13 +03:00
parent 514da76ddb
commit b33cca5fcc
19 changed files with 2051 additions and 65 deletions

View File

@@ -44,3 +44,79 @@ func TestApplyGPUStatuses_FromSummaryCSV_FailAndPass(t *testing.T) {
t.Fatalf("expected serial 222 status PASS, got %q", bySerial["222"])
}
}
func TestApplyGPUFailureDetails_FromSummaryJSON_BySerial(t *testing.T) {
jsonData := []byte(`[
{
"Error Code": "005-000-1-000000000363",
"Test": "gpumem",
"Component ID": "SXM5_SN_1653925025497",
"Notes": "Row remapping failed",
"Virtual ID": "gpumem",
"Ignore Error": "False"
}
]`)
result := &models.AnalysisResult{
Hardware: &models.HardwareConfig{
GPUs: []models.GPU{
{Slot: "GPUSXM5", SerialNumber: "1653925025497"},
{Slot: "GPUSXM2", SerialNumber: "1653925024190"},
},
},
}
details := CollectGPUFailureDetailsFromSummaryJSON(jsonData)
ApplyGPUFailureDetails(result, details)
if got := result.Hardware.GPUs[0].ErrorDescription; got != "Row remapping failed" {
t.Fatalf("expected serial 1653925025497 error Row remapping failed, got %q", got)
}
if got := result.Hardware.GPUs[1].ErrorDescription; got != "" {
t.Fatalf("expected no error description for healthy GPU, got %q", got)
}
}
func TestApplyNVSwitchStatuses_FromSummaryJSON(t *testing.T) {
jsonData := []byte(`[
{
"Error Code": "0",
"Test": "inventory",
"Component ID": "NVSWITCH_NVSWITCH0_VendorID",
"Notes": "OK",
"Virtual ID": "inventory",
"Ignore Error": "False"
},
{
"Error Code": "1",
"Test": "inventory",
"Component ID": "NVSWITCH_NVSWITCH1_LinkState",
"Notes": "Link down",
"Virtual ID": "inventory",
"Ignore Error": "False"
}
]`)
result := &models.AnalysisResult{
Hardware: &models.HardwareConfig{
PCIeDevices: []models.PCIeDevice{
{Slot: "NVSWITCH0", Status: "Unknown"},
{Slot: "NVSWITCH1", Status: "Unknown"},
{Slot: "NVSWITCH2", Status: "Unknown"},
},
},
}
statuses := CollectNVSwitchStatusesFromSummaryJSON(jsonData)
ApplyNVSwitchStatuses(result, statuses)
if got := result.Hardware.PCIeDevices[0].Status; got != "PASS" {
t.Fatalf("expected NVSWITCH0 status PASS, got %q", got)
}
if got := result.Hardware.PCIeDevices[1].Status; got != "FAIL" {
t.Fatalf("expected NVSWITCH1 status FAIL, got %q", got)
}
if got := result.Hardware.PCIeDevices[2].Status; got != "Unknown" {
t.Fatalf("expected NVSWITCH2 status unchanged Unknown, got %q", got)
}
}