logpile/internal/parser/vendors/inspur/hgx_gpu_status_test.go

package inspur

import (
	"testing"
	"time"

	"git.mchus.pro/mchus/logpile/internal/models"
)

func TestEnrichGPUsFromHGXHWInfo_UsesHGXLogicalMapping(t *testing.T) {
	hw := &models.HardwareConfig{
		GPUs: []models.GPU{
			{Slot: "#GPU6"},
			{Slot: "#GPU7"},
			{Slot: "#GPU0"},
			{Slot: "#CPU0_PE1_E_BMC", Model: "AST2500 VGA"},
		},
	}

	content := []byte(`
# curl -X GET http://127.0.0.1/redfish/v1/Chassis/HGX_GPU_SXM_1/Assembly
{"Name":"GPU Board Assembly","Model":"B200 180GB HBM3e","PartNumber":"PN1","SerialNumber":"SXM1SN"}
# curl -X GET http://127.0.0.1/redfish/v1/Chassis/HGX_GPU_SXM_3/Assembly
{"Name":"GPU Board Assembly","Model":"B200 180GB HBM3e","PartNumber":"PN3","SerialNumber":"SXM3SN"}
# curl -X GET http://127.0.0.1/redfish/v1/Chassis/HGX_GPU_SXM_5/Assembly
{"Name":"GPU Board Assembly","Model":"B200 180GB HBM3e","PartNumber":"PN5","SerialNumber":"SXM5SN"}
{"Id":"HGX_FW_GPU_SXM_1","Version":"FW1"}
{"Id":"HGX_FW_GPU_SXM_3","Version":"FW3"}
{"Id":"HGX_FW_GPU_SXM_5","Version":"FW5"}
{"Id":"HGX_InfoROM_GPU_SXM_3","Version":"IR3"}
`)

	enrichGPUsFromHGXHWInfo(content, hw)

	if hw.GPUs[0].SerialNumber != "SXM3SN" {
		t.Fatalf("expected #GPU6 to map to SXM3 serial, got %q", hw.GPUs[0].SerialNumber)
	}
	if hw.GPUs[1].SerialNumber != "SXM1SN" {
		t.Fatalf("expected #GPU7 to map to SXM1 serial, got %q", hw.GPUs[1].SerialNumber)
	}
	if hw.GPUs[2].SerialNumber != "SXM5SN" {
		t.Fatalf("expected #GPU0 to map to SXM5 serial, got %q", hw.GPUs[2].SerialNumber)
	}
	if hw.GPUs[0].Firmware != "FW3" {
		t.Fatalf("expected #GPU6 firmware FW3, got %q", hw.GPUs[0].Firmware)
	}
	if hw.GPUs[0].VideoBIOS != "IR3" {
		t.Fatalf("expected #GPU6 InfoROM in VideoBIOS IR3, got %q", hw.GPUs[0].VideoBIOS)
	}
	if hw.GPUs[2].Firmware != "FW5" {
		t.Fatalf("expected #GPU0 firmware FW5, got %q", hw.GPUs[2].Firmware)
	}
	for _, g := range hw.GPUs {
		if g.Slot == "#CPU0_PE1_E_BMC" {
			t.Fatalf("expected non-HGX BMC VGA entry to be filtered out")
		}
	}
}

func TestEnrichGPUsFromHGXHWInfo_AddsMissingLogicalGPU(t *testing.T) {
	hw := &models.HardwareConfig{
		GPUs: []models.GPU{
			{Slot: "#GPU0"},
			{Slot: "#GPU1"},
			{Slot: "#GPU2"},
			{Slot: "#GPU3"},
			{Slot: "#GPU4"},
			{Slot: "#GPU5"},
			{Slot: "#GPU7"},
		},
	}

	content := []byte(`
# curl -X GET http://127.0.0.1/redfish/v1/Chassis/HGX_GPU_SXM_3/Assembly
{"Name":"GPU Board Assembly","Model":"B200 180GB HBM3e","PartNumber":"PN3","SerialNumber":"SXM3SN"}
`)

	enrichGPUsFromHGXHWInfo(content, hw)

	found := false
	for _, g := range hw.GPUs {
		if g.Slot == "#GPU6" {
			found = true
			if g.SerialNumber != "SXM3SN" {
				t.Fatalf("expected synthesized #GPU6 serial SXM3SN, got %q", g.SerialNumber)
			}
		}
	}
	if !found {
		t.Fatalf("expected synthesized #GPU6 entry")
	}
}

func TestApplyGPUStatusFromEvents_MarksFaultedGPU(t *testing.T) {
	hw := &models.HardwareConfig{
		GPUs: []models.GPU{
			{Slot: "#GPU6"},
			{Slot: "#GPU5"},
		},
	}

	events := []models.Event{
		{
			ID:          "17FFB002",
			Timestamp:   time.Now(),
			Description: "PCIe Present mismatch BIOS miss F_GPU6",
		},
	}

	applyGPUStatusFromEvents(hw, events)

	if hw.GPUs[0].Status != "Critical" {
		t.Fatalf("expected #GPU6 status Critical, got %q", hw.GPUs[0].Status)
	}
	if hw.GPUs[1].Status != "OK" {
		t.Fatalf("expected healthy GPU status OK, got %q", hw.GPUs[1].Status)
	}
}

func TestApplyGPUStatusFromEvents_UsesLatestEventAsCurrentStatusAndKeepsHistory(t *testing.T) {
	hw := &models.HardwareConfig{
		GPUs: []models.GPU{
			{Slot: "#GPU1"},
			{Slot: "#GPU3"},
			{Slot: "#GPU6"},
		},
	}

	events := []models.Event{
		{
			ID:          "17FFB002",
			Timestamp:   time.Date(2026, 1, 12, 22, 51, 16, 0, time.FixedZone("UTC+8", 8*3600)),
			Description: "PCIe Present mismatch BIOS miss F_GPU1 F_GPU3 F_GPU6",
		},
		{
			ID:          "17FFB002",
			Timestamp:   time.Date(2026, 1, 12, 23, 5, 18, 0, time.FixedZone("UTC+8", 8*3600)),
			Description: "PCIe Present mismatch BIOS miss F_GPU6",
		},
	}

	applyGPUStatusFromEvents(hw, events)

	if hw.GPUs[0].Status != "OK" {
		t.Fatalf("expected #GPU1 to recover to OK on latest event, got %q", hw.GPUs[0].Status)
	}
	if hw.GPUs[1].Status != "OK" {
		t.Fatalf("expected #GPU3 to recover to OK on latest event, got %q", hw.GPUs[1].Status)
	}
	if hw.GPUs[2].Status != "Critical" {
		t.Fatalf("expected #GPU6 to remain Critical, got %q", hw.GPUs[2].Status)
	}
	if len(hw.GPUs[0].StatusHistory) == 0 {
		t.Fatalf("expected #GPU1 status history to be populated")
	}
}

func TestApplyGPUStatusFromEvents_DeassertClearsAllGPUs(t *testing.T) {
	hw := &models.HardwareConfig{
		GPUs: []models.GPU{
			{Slot: "#GPU1"},
			{Slot: "#GPU3"},
			{Slot: "#GPU5"},
			{Slot: "#GPU6"},
		},
	}

	events := []models.Event{
		{
			ID:          "17FFB002",
			EventType:   "Assert",
			Timestamp:   time.Date(2026, 5, 27, 13, 6, 56, 0, time.FixedZone("UTC+8", 8*3600)),
			Description: "PCIe Present mismatch BIOS Scan, BIOS miss F_GPU1 F_GPU3 F_GPU5 F_GPU6",
		},
		{
			ID:          "17FFB002",
			EventType:   "Deassert",
			Timestamp:   time.Date(2026, 5, 27, 13, 15, 56, 0, time.FixedZone("UTC+8", 8*3600)),
			Description: "PCIe Present mismatch BIOS Scan, BIOS miss F_GPU1 F_GPU3 F_GPU5 F_GPU6",
		},
	}

	applyGPUStatusFromEvents(hw, events)

	for _, gpu := range hw.GPUs {
		if gpu.Status != "OK" {
			t.Fatalf("expected %s to recover to OK after Deassert, got %q", gpu.Slot, gpu.Status)
		}
	}
}

func TestParseIDLLog_ParsesStructuredJSONLine(t *testing.T) {
	content := []byte(`{ "MESSAGE": "|2026-01-12T23:05:18+08:00|PCIE|Assert|Critical|17FFB002|PCIe Present mismatch BIOS miss F_GPU6 - Assert|" }`)

	events := ParseIDLLog(content)
	if len(events) != 1 {
		t.Fatalf("expected 1 event from JSON line, got %d", len(events))
	}
	if events[0].ID != "17FFB002" {
		t.Fatalf("expected event ID 17FFB002, got %q", events[0].ID)
	}
	if events[0].Source != "BMC" {
		t.Fatalf("expected BMC source for IDL event, got %q", events[0].Source)
	}
	if events[0].SensorType != "pcie" {
		t.Fatalf("expected component type pcie, got %#v", events[0])
	}
}