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

2026-02-16 23:17:13 +03:00
parent 514da76ddb
commit b33cca5fcc
19 changed files with 2051 additions and 65 deletions
--- a/internal/parser/vendors/nvidia/component_status_time_test.go
+++ b/internal/parser/vendors/nvidia/component_status_time_test.go
@@ -0,0 +1,143 @@
+package nvidia
+
+import (
+	"testing"
+	"time"
+
+	"git.mchus.pro/mchus/logpile/internal/models"
+	"git.mchus.pro/mchus/logpile/internal/parser"
+)
+
+func TestParseVerboseRunTestStartTimes(t *testing.T) {
+	content := []byte(`
+2026-01-22 09:11:32,458 - Testing nvswitch
+2026-01-22 09:45:36,016 - Testing gpu_fieldiag
+`)
+	got := parseVerboseRunTestStartTimes(content)
+
+	nvs := got["nvswitch"]
+	if nvs.IsZero() {
+		t.Fatalf("expected nvswitch timestamp")
+	}
+	gpu := got["gpu_fieldiag"]
+	if gpu.IsZero() {
+		t.Fatalf("expected gpu_fieldiag timestamp")
+	}
+	if nvs.Format(time.RFC3339) != "2026-01-22T09:11:32Z" {
+		t.Fatalf("unexpected nvswitch timestamp: %s", nvs.Format(time.RFC3339))
+	}
+	if gpu.Format(time.RFC3339) != "2026-01-22T09:45:36Z" {
+		t.Fatalf("unexpected gpu_fieldiag timestamp: %s", gpu.Format(time.RFC3339))
+	}
+}
+
+func TestParseRunLogTestStartTimes(t *testing.T) {
+	content := []byte(`
+Start time               Thu, 22 Jan 2026 07:42:26
+Testing gpumem FAILED [ 26:12s ]
+Testing gpustress OK [ 7:10s ]
+Testing nvswitch OK [ 9:25s ]
+`)
+
+	got := parseRunLogTestStartTimes(content)
+	if got["gpumem"].Format(time.RFC3339) != "2026-01-22T07:42:26Z" {
+		t.Fatalf("unexpected gpumem start: %s", got["gpumem"].Format(time.RFC3339))
+	}
+	if got["gpustress"].Format(time.RFC3339) != "2026-01-22T08:08:38Z" {
+		t.Fatalf("unexpected gpustress start: %s", got["gpustress"].Format(time.RFC3339))
+	}
+	if got["nvswitch"].Format(time.RFC3339) != "2026-01-22T08:15:48Z" {
+		t.Fatalf("unexpected nvswitch start: %s", got["nvswitch"].Format(time.RFC3339))
+	}
+}
+
+func TestApplyGPUAndNVSwitchCheckTimes(t *testing.T) {
+	gpuTs := time.Date(2026, 1, 22, 9, 45, 36, 0, time.UTC)
+	nvsTs := time.Date(2026, 1, 22, 9, 11, 32, 0, time.UTC)
+
+	result := &models.AnalysisResult{
+		Hardware: &models.HardwareConfig{
+			GPUs: []models.GPU{
+				{Slot: "GPUSXM5", Status: "FAIL"},
+			},
+			PCIeDevices: []models.PCIeDevice{
+				{Slot: "NVSWITCH0", DeviceClass: "NVSwitch", Status: "PASS"},
+				{Slot: "NIC0", DeviceClass: "NetworkController", Status: "PASS"},
+			},
+		},
+	}
+
+	ApplyGPUAndNVSwitchCheckTimes(result, componentCheckTimes{
+		GPUBySlot:      map[string]time.Time{"GPUSXM5": gpuTs},
+		NVSwitchBySlot: map[string]time.Time{"NVSWITCH0": nvsTs},
+	})
+
+	if got := result.Hardware.GPUs[0].StatusCheckedAt; !got.Equal(gpuTs) {
+		t.Fatalf("expected gpu status_checked_at %s, got %s", gpuTs.Format(time.RFC3339), got.Format(time.RFC3339))
+	}
+	if result.Hardware.GPUs[0].StatusAtCollect == nil || !result.Hardware.GPUs[0].StatusAtCollect.At.Equal(gpuTs) {
+		t.Fatalf("expected gpu status_at_collection.at %s", gpuTs.Format(time.RFC3339))
+	}
+	if got := result.Hardware.PCIeDevices[0].StatusCheckedAt; !got.Equal(nvsTs) {
+		t.Fatalf("expected nvswitch status_checked_at %s, got %s", nvsTs.Format(time.RFC3339), got.Format(time.RFC3339))
+	}
+	if result.Hardware.PCIeDevices[0].StatusAtCollect == nil || !result.Hardware.PCIeDevices[0].StatusAtCollect.At.Equal(nvsTs) {
+		t.Fatalf("expected nvswitch status_at_collection.at %s", nvsTs.Format(time.RFC3339))
+	}
+	if !result.Hardware.PCIeDevices[1].StatusCheckedAt.IsZero() {
+		t.Fatalf("expected non-nvswitch device status_checked_at to stay zero")
+	}
+}
+
+func TestCollectGPUAndNVSwitchCheckTimes_FromVerboseRun(t *testing.T) {
+	files := []parser.ExtractedFile{
+		{
+			Path: "verbose_run.log",
+			Content: []byte(`
+2026-01-22 09:11:32,458 - Testing nvswitch
+2026-01-22 09:45:36,016 - Testing gpu_fieldiag
+`),
+		},
+	}
+
+	got := CollectGPUAndNVSwitchCheckTimes(files)
+	if got.GPUDefault.Format(time.RFC3339) != "2026-01-22T09:45:36Z" {
+		t.Fatalf("unexpected GPU check time: %s", got.GPUDefault.Format(time.RFC3339))
+	}
+	if got.NVSwitchDefault.Format(time.RFC3339) != "2026-01-22T09:11:32Z" {
+		t.Fatalf("unexpected NVSwitch check time: %s", got.NVSwitchDefault.Format(time.RFC3339))
+	}
+}
+
+func TestCollectGPUAndNVSwitchCheckTimes_FromComponentOutputLogs(t *testing.T) {
+	files := []parser.ExtractedFile{
+		{
+			Path: "gpu_fieldiag/SXM5_SN_1653925025497/output.log",
+			Content: []byte(`
+$ some command
+MODS start: Thu Jan 22 09:45:36 2026 
+`),
+		},
+		{
+			Path: "nvswitch/output.log",
+			Content: []byte(`
+$ cmd devname=0000:08:00.0,NVSWITCH3 devname=0000:07:00.0,NVSWITCH2 devname=0000:06:00.0,NVSWITCH1 devname=0000:05:00.0,NVSWITCH0
+MODS start: Thu Jan 22 09:11:32 2026 
+`),
+		},
+	}
+
+	got := CollectGPUAndNVSwitchCheckTimes(files)
+	if got.GPUBySerial["1653925025497"].Format(time.RFC3339) != "2026-01-22T09:45:36Z" {
+		t.Fatalf("unexpected GPU serial check time: %s", got.GPUBySerial["1653925025497"].Format(time.RFC3339))
+	}
+	if got.GPUBySlot["GPUSXM5"].Format(time.RFC3339) != "2026-01-22T09:45:36Z" {
+		t.Fatalf("unexpected GPU slot check time: %s", got.GPUBySlot["GPUSXM5"].Format(time.RFC3339))
+	}
+	if got.NVSwitchBySlot["NVSWITCH0"].Format(time.RFC3339) != "2026-01-22T09:11:32Z" {
+		t.Fatalf("unexpected NVSwitch0 check time: %s", got.NVSwitchBySlot["NVSWITCH0"].Format(time.RFC3339))
+	}
+	if got.NVSwitchBySlot["NVSWITCH3"].Format(time.RFC3339) != "2026-01-22T09:11:32Z" {
+		t.Fatalf("unexpected NVSwitch3 check time: %s", got.NVSwitchBySlot["NVSWITCH3"].Format(time.RFC3339))
+	}
+}