feat: support configuring init container image

config/default/configmap.yaml

@@ -5,4 +5,4 @@ metadata:
 data:
   config.data: |
     scheduler-name: default-scheduler
-    # init-container-image: inftyai/model-loader:v0.0.10
+    init-container-image: inftyai/model-loader:v0.0.10

pkg/controller/inference/service_controller.go

@@ -116,10 +116,11 @@ func (r *ServiceReconciler) Reconcile(ctx context.Context, req ctrl.Request) (ct
 		return ctrl.Result{}, err
 	}
 
-	workloadApplyConfiguration, err := buildWorkloadApplyConfiguration(service, models)
+	workloadApplyConfiguration, err := buildWorkloadApplyConfiguration(service, models, configs)
 	if err != nil {
 		return ctrl.Result{}, err
 	}
+
 	if err := setControllerReferenceForWorkload(service, workloadApplyConfiguration, r.Scheme); err != nil {
 		return ctrl.Result{}, err
 	}
@@ -162,7 +163,7 @@ func (r *ServiceReconciler) SetupWithManager(mgr ctrl.Manager) error {
 		Complete(r)
 }
 
-func buildWorkloadApplyConfiguration(service *inferenceapi.Service, models []*coreapi.OpenModel) (*applyconfigurationv1.LeaderWorkerSetApplyConfiguration, error) {
+func buildWorkloadApplyConfiguration(service *inferenceapi.Service, models []*coreapi.OpenModel, configs *helper.GlobalConfigs) (*applyconfigurationv1.LeaderWorkerSetApplyConfiguration, error) {
 	workload := applyconfigurationv1.LeaderWorkerSet(service.Name, service.Namespace)
 
 	leaderWorkerTemplate := applyconfigurationv1.LeaderWorkerTemplate()
@@ -193,7 +194,7 @@ func buildWorkloadApplyConfiguration(service *inferenceapi.Service, models []*co
 	leaderWorkerTemplate.WithWorkerTemplate(&podTemplateSpecApplyConfiguration)
 
 	// The core logic to inject additional configurations.
-	injectModelProperties(leaderWorkerTemplate, models, service)
+	injectModelProperties(leaderWorkerTemplate, models, service, configs)
 
 	spec := applyconfigurationv1.LeaderWorkerSetSpec()
 	spec.WithLeaderWorkerTemplate(leaderWorkerTemplate)
@@ -215,17 +216,17 @@ func buildWorkloadApplyConfiguration(service *inferenceapi.Service, models []*co
 	return workload, nil
 }
 
-func injectModelProperties(template *applyconfigurationv1.LeaderWorkerTemplateApplyConfiguration, models []*coreapi.OpenModel, service *inferenceapi.Service) {
+func injectModelProperties(template *applyconfigurationv1.LeaderWorkerTemplateApplyConfiguration, models []*coreapi.OpenModel, service *inferenceapi.Service, configs *helper.GlobalConfigs) {
 	isMultiNodesInference := template.LeaderTemplate != nil
 
 	for i, model := range models {
 		source := modelSource.NewModelSourceProvider(model)
 		// Skip model-loader initContainer if llmaz.io/skip-model-loader annotation is set.
 		if !helper.SkipModelLoader(service) {
 			if isMultiNodesInference {
-				source.InjectModelLoader(template.LeaderTemplate, i)
+				source.InjectModelLoader(template.LeaderTemplate, i, configs.InitContainerImage)
 			}
-			source.InjectModelLoader(template.WorkerTemplate, i)
+			source.InjectModelLoader(template.WorkerTemplate, i, configs.InitContainerImage)
 		} else {
 			if isMultiNodesInference {
 				source.InjectModelEnvVars(template.LeaderTemplate)

pkg/controller_helper/configmap.go

@@ -40,5 +40,16 @@ func ParseGlobalConfigmap(cm *corev1.ConfigMap) (*GlobalConfigs, error) {
 		return nil, fmt.Errorf("failed to unmarshal config.data: %v", err)
 	}
 
+	if err := configs.validate(); err != nil {
+		return nil, fmt.Errorf("invalid global config: %v", err)
+	}
+
 	return &configs, nil
 }
+
+func (c *GlobalConfigs) validate() error {
+	if c.InitContainerImage == "" {
+		return fmt.Errorf("init-container-image is required")
+	}
+	return nil
+}

pkg/controller_helper/configmap_test.go

@@ -0,0 +1,64 @@
+/*
+Copyright 2025 The InftyAI Team.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+package helper
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestGlobalConfigs_validate(t *testing.T) {
+	tests := []struct {
+		name        string
+		config      *GlobalConfigs
+		expectError bool
+		errorMsg    string
+	}{
+		{
+			name: "valid config",
+			config: &GlobalConfigs{
+				SchedulerName:      "custom-scheduler",
+				InitContainerImage: "inftyai/model-loader:v0.0.10",
+			},
+			expectError: false,
+		},
+		{
+			name: "empty init container image",
+			config: &GlobalConfigs{
+				SchedulerName:      "custom-scheduler",
+				InitContainerImage: "",
+			},
+			expectError: true,
+			errorMsg:    "init-container-image is required",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			err := tt.config.validate()
+
+			if tt.expectError {
+				require.Error(t, err)
+				assert.Contains(t, err.Error(), tt.errorMsg)
+			} else {
+				require.NoError(t, err)
+			}
+		})
+	}
+}

pkg/controller_helper/modelsource/modelhub.go

@@ -21,8 +21,6 @@ import (
 	"strings"
 
 	coreapplyv1 "k8s.io/client-go/applyconfigurations/core/v1"
-
-	"github.com/inftyai/llmaz/pkg"
 )
 
 var _ ModelSourceProvider = &ModelHubProvider{}
@@ -62,7 +60,7 @@ func (p *ModelHubProvider) ModelPath(skipModelLoader bool) string {
 	return CONTAINER_MODEL_PATH + "models--" + strings.ReplaceAll(p.modelID, "/", "--")
 }
 
-func (p *ModelHubProvider) InjectModelLoader(template *coreapplyv1.PodTemplateSpecApplyConfiguration, index int) {
+func (p *ModelHubProvider) InjectModelLoader(template *coreapplyv1.PodTemplateSpecApplyConfiguration, index int, initContainerImage string) {
 	initContainerName := MODEL_LOADER_CONTAINER_NAME
 	if index != 0 {
 		initContainerName += "-" + strconv.Itoa(index)
@@ -71,7 +69,7 @@ func (p *ModelHubProvider) InjectModelLoader(template *coreapplyv1.PodTemplateSp
 	// Handle initContainer.
 	initContainer := coreapplyv1.Container().
 		WithName(initContainerName).
-		WithImage(pkg.LOADER_IMAGE).
+		WithImage(initContainerImage).
 		WithVolumeMounts(coreapplyv1.VolumeMount().WithName(MODEL_VOLUME_NAME).WithMountPath(CONTAINER_MODEL_PATH))
 
 	// We have exactly one container in the template.Spec.Containers.

pkg/controller_helper/modelsource/modelhub_test.go

@@ -27,8 +27,6 @@ import (
 	"github.com/stretchr/testify/assert"
 	corev1 "k8s.io/api/core/v1"
 	coreapplyv1 "k8s.io/client-go/applyconfigurations/core/v1"
-
-	"github.com/inftyai/llmaz/pkg"
 )
 
 func Test_ModelHubProvider_InjectModelLoader(t *testing.T) {
@@ -38,10 +36,11 @@ func Test_ModelHubProvider_InjectModelLoader(t *testing.T) {
 	ignorePatterns := []string{"*.tmp"}
 
 	tests := []struct {
-		name            string
-		provider        *ModelHubProvider
-		index           int
-		expectMainModel bool
+		name               string
+		provider           *ModelHubProvider
+		index              int
+		expectMainModel    bool
+		initContainerImage string
 	}{
 		{
 			name: "inject full modelhub with fileName, revision, allow/ignore",
@@ -54,8 +53,9 @@ func Test_ModelHubProvider_InjectModelLoader(t *testing.T) {
 				modelAllowPatterns:  allowPatterns,
 				modelIgnorePatterns: ignorePatterns,
 			},
-			index:           0,
-			expectMainModel: true,
+			index:              0,
+			expectMainModel:    true,
+			initContainerImage: "model-loader:latest",
 		},
 		{
 			name: "inject with index > 0 skips volume/container mount",
@@ -64,8 +64,20 @@ func Test_ModelHubProvider_InjectModelLoader(t *testing.T) {
 				modelID:   "some/model",
 				modelHub:  "Huggingface",
 			},
-			index:           1,
-			expectMainModel: false,
+			index:              1,
+			expectMainModel:    false,
+			initContainerImage: "model-loader:latest",
+		},
+		{
+			name: "inject with custom initContainerImage",
+			provider: &ModelHubProvider{
+				modelName: "llama3",
+				modelID:   "meta/llama-3",
+				modelHub:  "Huggingface",
+			},
+			index:              0,
+			expectMainModel:    true,
+			initContainerImage: "custom-model-loader:latest",
 		},
 	}
 
@@ -83,7 +95,7 @@ func Test_ModelHubProvider_InjectModelLoader(t *testing.T) {
 					),
 				)
 
-			tt.provider.InjectModelLoader(template, tt.index)
+			tt.provider.InjectModelLoader(template, tt.index, tt.initContainerImage)
 
 			assert.Len(t, template.Spec.InitContainers, 1)
 			initContainer := template.Spec.InitContainers[0]
@@ -92,8 +104,9 @@ func Test_ModelHubProvider_InjectModelLoader(t *testing.T) {
 			if tt.index != 0 {
 				expectedName += "-" + strconv.Itoa(tt.index)
 			}
+			expectedImage := tt.initContainerImage
 			assert.Equal(t, expectedName, *initContainer.Name)
-			assert.Equal(t, pkg.LOADER_IMAGE, *initContainer.Image)
+			assert.Equal(t, expectedImage, *initContainer.Image)
 
 			wantEnv := buildExpectedEnv(tt.provider)
 			if diff := cmp.Diff(wantEnv, initContainer.Env, envSortOpt); diff != "" {

pkg/controller_helper/modelsource/modelsource.go

@@ -59,7 +59,7 @@ type ModelSourceProvider interface {
 	ModelPath(skipModelLoader bool) string
 	// InjectModelLoader will inject the model loader to the spec,
 	// index refers to the suffix of the initContainer name, like model-loader, model-loader-1.
-	InjectModelLoader(spec *coreapplyv1.PodTemplateSpecApplyConfiguration, index int)
+	InjectModelLoader(spec *coreapplyv1.PodTemplateSpecApplyConfiguration, index int, initContainerImage string)
 	// InjectModelEnvVars will inject the model credentials env to the model-runner container.
 	// This is used when the model-loader initContainer is not injected, and the model loading is handled by the model-runner container.
 	InjectModelEnvVars(spec *coreapplyv1.PodTemplateSpecApplyConfiguration)

pkg/controller_helper/modelsource/modelsource_test.go

@@ -130,7 +130,7 @@ func TestEnvInjectModelLoader(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			tt.provider.InjectModelLoader(tt.template, 0)
+			tt.provider.InjectModelLoader(tt.template, 0, "model-loader:latest")
 			initContainer := tt.template.Spec.InitContainers[0]
 			assert.Subset(t, initContainer.Env, tt.template.Spec.Containers[0].Env)
 		})

pkg/controller_helper/modelsource/uri.go

@@ -21,8 +21,6 @@ import (
 	"strings"
 
 	coreapplyv1 "k8s.io/client-go/applyconfigurations/core/v1"
-
-	"github.com/inftyai/llmaz/pkg"
 )
 
 var _ ModelSourceProvider = &URIProvider{}
@@ -80,7 +78,7 @@ func (p *URIProvider) ModelPath(skipModelLoader bool) string {
 	return CONTAINER_MODEL_PATH + "models--" + splits[len(splits)-1]
 }
 
-func (p *URIProvider) InjectModelLoader(template *coreapplyv1.PodTemplateSpecApplyConfiguration, index int) {
+func (p *URIProvider) InjectModelLoader(template *coreapplyv1.PodTemplateSpecApplyConfiguration, index int, initContainerImage string) {
 	// We don't have additional operations for Ollama, just load in runtime.
 	if p.protocol == Ollama {
 		return
@@ -112,10 +110,11 @@ func (p *URIProvider) InjectModelLoader(template *coreapplyv1.PodTemplateSpecApp
 	if index != 0 {
 		initContainerName += "-" + strconv.Itoa(index)
 	}
+
 	// Handle initContainer.
 	initContainer := coreapplyv1.Container().
 		WithName(initContainerName).
-		WithImage(pkg.LOADER_IMAGE).
+		WithImage(initContainerImage).
 		WithVolumeMounts(
 			coreapplyv1.VolumeMount().
 				WithName(MODEL_VOLUME_NAME).

test/util/validation/validate_service.go

@@ -42,7 +42,6 @@ import (
 
 	coreapi "github.com/inftyai/llmaz/api/core/v1alpha1"
 	inferenceapi "github.com/inftyai/llmaz/api/inference/v1alpha1"
-	"github.com/inftyai/llmaz/pkg"
 	helper "github.com/inftyai/llmaz/pkg/controller_helper"
 	modelSource "github.com/inftyai/llmaz/pkg/controller_helper/modelsource"
 	pkgUtil "github.com/inftyai/llmaz/pkg/util"
@@ -114,7 +113,7 @@ func ValidateService(ctx context.Context, k8sClient client.Client, service *infe
 			return err
 		}
 
-		if err := ValidateConfigmap(ctx, k8sClient, service); err != nil {
+		if err := ValidateConfigmap(ctx, k8sClient, service, &workload); err != nil {
 			return err
 		}
 
@@ -143,8 +142,8 @@ func ValidateModelLoader(model *coreapi.OpenModel, index int, template corev1.Po
 		if initContainer.Name != containerName {
 			return fmt.Errorf("unexpected initContainer name, want %s, got %s", modelSource.MODEL_LOADER_CONTAINER_NAME, initContainer.Name)
 		}
-		if initContainer.Image != pkg.LOADER_IMAGE {
-			return fmt.Errorf("unexpected initContainer image, want %s, got %s", pkg.LOADER_IMAGE, initContainer.Image)
+		if initContainer.Image == "" {
+			return fmt.Errorf("unexpected initContainer image, initContainer image should not be empty")
 		}
 
 		var envStrings []string
@@ -440,7 +439,7 @@ func CheckServiceAvaliable() error {
 	return nil
 }
 
-func ValidateConfigmap(ctx context.Context, k8sClient client.Client, service *inferenceapi.Service) error {
+func ValidateConfigmap(ctx context.Context, k8sClient client.Client, service *inferenceapi.Service, workload *lws.LeaderWorkerSet) error {
 	cm := corev1.ConfigMap{}
 	if err := k8sClient.Get(ctx, types.NamespacedName{Name: "llmaz-global-config", Namespace: "llmaz-system"}, &cm); err != nil {
 		return err
@@ -451,6 +450,7 @@ func ValidateConfigmap(ctx context.Context, k8sClient client.Client, service *in
 		return fmt.Errorf("failed to parse global configmap: %v", err)
 	}
 
+	// Validate scheduler name.
 	if service.Spec.WorkloadTemplate.LeaderTemplate != nil {
 		if service.Spec.WorkloadTemplate.LeaderTemplate.Spec.SchedulerName != data.SchedulerName {
 			return fmt.Errorf("unexpected scheduler name %s, want %s", service.Spec.WorkloadTemplate.LeaderTemplate.Spec.SchedulerName, data.SchedulerName)
@@ -461,5 +461,22 @@ func ValidateConfigmap(ctx context.Context, k8sClient client.Client, service *in
 		return fmt.Errorf("unexpected scheduler name %s, want %s", service.Spec.WorkloadTemplate.WorkerTemplate.Spec.SchedulerName, data.SchedulerName)
 	}
 
+	if !helper.SkipModelLoader(service) {
+		// Validate init container image.
+		if service.Spec.WorkloadTemplate.LeaderTemplate != nil {
+			for _, initContainer := range workload.Spec.LeaderWorkerTemplate.LeaderTemplate.Spec.InitContainers {
+				if initContainer.Image != data.InitContainerImage {
+					return fmt.Errorf("unexpected init container image %s in leader template, want %s", initContainer.Image, data.InitContainerImage)
+				}
+			}
+		}
+
+		for _, initContainer := range workload.Spec.LeaderWorkerTemplate.WorkerTemplate.Spec.InitContainers {
+			if initContainer.Image != data.InitContainerImage {
+				return fmt.Errorf("unexpected init container image %s in worker template, want %s", initContainer.Image, data.InitContainerImage)
+			}
+		}
+	}
+
 	return nil
 }