test: cover nvml init error paths

Author: gyuho

components/accelerator/nvidia/clock-speed/init_error_test.go

@@ -0,0 +1,30 @@
+package clockspeed
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	apiv1 "github.com/leptonai/gpud/api/v1"
+	"github.com/leptonai/gpud/components"
+	nvidianvml "github.com/leptonai/gpud/pkg/nvidia-query/nvml"
+)
+
+func TestInitErrorUnhealthy(t *testing.T) {
+	initErr := errors.New("nvml init failed")
+	comp, err := New(&components.GPUdInstance{
+		RootCtx:      context.Background(),
+		NVMLInstance: nvidianvml.NewErrored(initErr),
+	})
+	if err != nil {
+		t.Fatalf("failed to create component: %v", err)
+	}
+
+	if !comp.IsSupported() {
+		t.Fatalf("expected IsSupported true when NVML init error is present")
+	}
+
+	if got := comp.Check().HealthStateType(); got != apiv1.HealthStateTypeUnhealthy {
+		t.Fatalf("expected unhealthy, got %v", got)
+	}
+}

components/accelerator/nvidia/ecc/init_error_test.go

@@ -0,0 +1,30 @@
+package ecc
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	apiv1 "github.com/leptonai/gpud/api/v1"
+	"github.com/leptonai/gpud/components"
+	nvidianvml "github.com/leptonai/gpud/pkg/nvidia-query/nvml"
+)
+
+func TestInitErrorUnhealthy(t *testing.T) {
+	initErr := errors.New("nvml init failed")
+	comp, err := New(&components.GPUdInstance{
+		RootCtx:      context.Background(),
+		NVMLInstance: nvidianvml.NewErrored(initErr),
+	})
+	if err != nil {
+		t.Fatalf("failed to create component: %v", err)
+	}
+
+	if !comp.IsSupported() {
+		t.Fatalf("expected IsSupported true when NVML init error is present")
+	}
+
+	if got := comp.Check().HealthStateType(); got != apiv1.HealthStateTypeUnhealthy {
+		t.Fatalf("expected unhealthy, got %v", got)
+	}
+}

components/accelerator/nvidia/fabric-manager/init_error_test.go

@@ -0,0 +1,30 @@
+package fabricmanager
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	apiv1 "github.com/leptonai/gpud/api/v1"
+	"github.com/leptonai/gpud/components"
+	nvidianvml "github.com/leptonai/gpud/pkg/nvidia-query/nvml"
+)
+
+func TestInitErrorUnhealthy(t *testing.T) {
+	initErr := errors.New("nvml init failed")
+	comp, err := New(&components.GPUdInstance{
+		RootCtx:      context.Background(),
+		NVMLInstance: nvidianvml.NewErrored(initErr),
+	})
+	if err != nil {
+		t.Fatalf("failed to create component: %v", err)
+	}
+
+	if !comp.IsSupported() {
+		t.Fatalf("expected IsSupported true when NVML init error is present")
+	}
+
+	if got := comp.Check().HealthStateType(); got != apiv1.HealthStateTypeUnhealthy {
+		t.Fatalf("expected unhealthy, got %v", got)
+	}
+}

components/accelerator/nvidia/gpm/init_error_test.go

@@ -0,0 +1,30 @@
+package gpm
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	apiv1 "github.com/leptonai/gpud/api/v1"
+	"github.com/leptonai/gpud/components"
+	nvidianvml "github.com/leptonai/gpud/pkg/nvidia-query/nvml"
+)
+
+func TestInitErrorUnhealthy(t *testing.T) {
+	initErr := errors.New("nvml init failed")
+	comp, err := New(&components.GPUdInstance{
+		RootCtx:      context.Background(),
+		NVMLInstance: nvidianvml.NewErrored(initErr),
+	})
+	if err != nil {
+		t.Fatalf("failed to create component: %v", err)
+	}
+
+	if !comp.IsSupported() {
+		t.Fatalf("expected IsSupported true when NVML init error is present")
+	}
+
+	if got := comp.Check().HealthStateType(); got != apiv1.HealthStateTypeUnhealthy {
+		t.Fatalf("expected unhealthy, got %v", got)
+	}
+}

components/accelerator/nvidia/gpu-counts/init_error_test.go

@@ -0,0 +1,30 @@
+package gpucounts
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	apiv1 "github.com/leptonai/gpud/api/v1"
+	"github.com/leptonai/gpud/components"
+	nvidianvml "github.com/leptonai/gpud/pkg/nvidia-query/nvml"
+)
+
+func TestInitErrorUnhealthy(t *testing.T) {
+	initErr := errors.New("nvml init failed")
+	comp, err := New(&components.GPUdInstance{
+		RootCtx:      context.Background(),
+		NVMLInstance: nvidianvml.NewErrored(initErr),
+	})
+	if err != nil {
+		t.Fatalf("failed to create component: %v", err)
+	}
+
+	if !comp.IsSupported() {
+		t.Fatalf("expected IsSupported true when NVML init error is present")
+	}
+
+	if got := comp.Check().HealthStateType(); got != apiv1.HealthStateTypeUnhealthy {
+		t.Fatalf("expected unhealthy, got %v", got)
+	}
+}

components/accelerator/nvidia/hw-slowdown/init_error_test.go

@@ -0,0 +1,30 @@
+package hwslowdown
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	apiv1 "github.com/leptonai/gpud/api/v1"
+	"github.com/leptonai/gpud/components"
+	nvidianvml "github.com/leptonai/gpud/pkg/nvidia-query/nvml"
+)
+
+func TestInitErrorUnhealthy(t *testing.T) {
+	initErr := errors.New("nvml init failed")
+	comp, err := New(&components.GPUdInstance{
+		RootCtx:      context.Background(),
+		NVMLInstance: nvidianvml.NewErrored(initErr),
+	})
+	if err != nil {
+		t.Fatalf("failed to create component: %v", err)
+	}
+
+	if !comp.IsSupported() {
+		t.Fatalf("expected IsSupported true when NVML init error is present")
+	}
+
+	if got := comp.Check().HealthStateType(); got != apiv1.HealthStateTypeUnhealthy {
+		t.Fatalf("expected unhealthy, got %v", got)
+	}
+}

components/accelerator/nvidia/infiniband/init_error_test.go

@@ -0,0 +1,39 @@
+package infiniband
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	apiv1 "github.com/leptonai/gpud/api/v1"
+	"github.com/leptonai/gpud/components"
+	ibtypes "github.com/leptonai/gpud/components/accelerator/nvidia/infiniband/types"
+	nvidianvml "github.com/leptonai/gpud/pkg/nvidia-query/nvml"
+)
+
+func TestInitErrorUnhealthy(t *testing.T) {
+	initErr := errors.New("nvml init failed")
+	comp, err := New(&components.GPUdInstance{
+		RootCtx:      context.Background(),
+		NVMLInstance: nvidianvml.NewErrored(initErr),
+	})
+	if err != nil {
+		t.Fatalf("failed to create component: %v", err)
+	}
+
+	if !comp.IsSupported() {
+		t.Fatalf("expected IsSupported true when NVML init error is present")
+	}
+
+	ibComp, ok := comp.(*component)
+	if !ok {
+		t.Fatalf("expected infiniband component type, got %T", comp)
+	}
+	ibComp.getThresholdsFunc = func() ibtypes.ExpectedPortStates {
+		return ibtypes.ExpectedPortStates{AtLeastPorts: 1, AtLeastRate: 1}
+	}
+
+	if got := ibComp.Check().HealthStateType(); got != apiv1.HealthStateTypeUnhealthy {
+		t.Fatalf("expected unhealthy, got %v", got)
+	}
+}

components/accelerator/nvidia/memory/init_error_test.go

@@ -0,0 +1,30 @@
+package memory
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	apiv1 "github.com/leptonai/gpud/api/v1"
+	"github.com/leptonai/gpud/components"
+	nvidianvml "github.com/leptonai/gpud/pkg/nvidia-query/nvml"
+)
+
+func TestInitErrorUnhealthy(t *testing.T) {
+	initErr := errors.New("nvml init failed")
+	comp, err := New(&components.GPUdInstance{
+		RootCtx:      context.Background(),
+		NVMLInstance: nvidianvml.NewErrored(initErr),
+	})
+	if err != nil {
+		t.Fatalf("failed to create component: %v", err)
+	}
+
+	if !comp.IsSupported() {
+		t.Fatalf("expected IsSupported true when NVML init error is present")
+	}
+
+	if got := comp.Check().HealthStateType(); got != apiv1.HealthStateTypeUnhealthy {
+		t.Fatalf("expected unhealthy, got %v", got)
+	}
+}

components/accelerator/nvidia/nccl/init_error_test.go

@@ -0,0 +1,30 @@
+package nccl
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	apiv1 "github.com/leptonai/gpud/api/v1"
+	"github.com/leptonai/gpud/components"
+	nvidianvml "github.com/leptonai/gpud/pkg/nvidia-query/nvml"
+)
+
+func TestInitErrorUnhealthy(t *testing.T) {
+	initErr := errors.New("nvml init failed")
+	comp, err := New(&components.GPUdInstance{
+		RootCtx:      context.Background(),
+		NVMLInstance: nvidianvml.NewErrored(initErr),
+	})
+	if err != nil {
+		t.Fatalf("failed to create component: %v", err)
+	}
+
+	if !comp.IsSupported() {
+		t.Fatalf("expected IsSupported true when NVML init error is present")
+	}
+
+	if got := comp.Check().HealthStateType(); got != apiv1.HealthStateTypeUnhealthy {
+		t.Fatalf("expected unhealthy, got %v", got)
+	}
+}

components/accelerator/nvidia/nvlink/init_error_test.go

@@ -0,0 +1,30 @@
+package nvlink
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	apiv1 "github.com/leptonai/gpud/api/v1"
+	"github.com/leptonai/gpud/components"
+	nvidianvml "github.com/leptonai/gpud/pkg/nvidia-query/nvml"
+)
+
+func TestInitErrorUnhealthy(t *testing.T) {
+	initErr := errors.New("nvml init failed")
+	comp, err := New(&components.GPUdInstance{
+		RootCtx:      context.Background(),
+		NVMLInstance: nvidianvml.NewErrored(initErr),
+	})
+	if err != nil {
+		t.Fatalf("failed to create component: %v", err)
+	}
+
+	if !comp.IsSupported() {
+		t.Fatalf("expected IsSupported true when NVML init error is present")
+	}
+
+	if got := comp.Check().HealthStateType(); got != apiv1.HealthStateTypeUnhealthy {
+		t.Fatalf("expected unhealthy, got %v", got)
+	}
+}