feat: support local GPU mode (#57)

* feat: support local GPU mode

* chore: changed the GPU field in TensorFusionConnectionSpec to GPUs. Multi-GPU mode will be supported in the future.

* chore: define the QosLevel enumeration

Author: 0x5457

.vscode/launch.json

@@ -12,7 +12,7 @@
             "env": {
                 "ENABLE_WEBHOOKS": "false"
             },
-            "program": "${workspaceFolder}/cmd/operator/main.go",
+            "program": "${workspaceFolder}/cmd/main.go",
         },
         {
             "name": "Debug Discovery",
@@ -35,7 +35,7 @@
                 "KUBECONFIG": "~/.kube/config-tf-dev",
                 "ENABLE_WEBHOOKS": "false"
             },
-            "program": "${workspaceFolder}/cmd/operator/main.go",
+            "program": "${workspaceFolder}/cmd/main.go",
         },
         {
             "name": "Debug Demo Env Operator",
@@ -47,7 +47,7 @@
                 "KUBECONFIG": "~/.kube/dev_us-east-1_demo",
                 "ENABLE_WEBHOOKS": "false"
             },
-            "program": "${workspaceFolder}/cmd/operator/main.go",
+            "program": "${workspaceFolder}/cmd/main.go",
         },
         {
             "name": "Debug Test Code",

Makefile

@@ -93,11 +93,11 @@ lint-fix: golangci-lint ## Run golangci-lint linter and perform fixes
 
 .PHONY: build
 build: manifests generate fmt vet ## Build manager binary.
-	go build -o bin/manager cmd/operator/main.go
+	go build -o bin/manager cmd/main.go
 
 .PHONY: run
 run: manifests generate fmt vet ## Run a controller from your host.
-	go run ./cmd/operator/main.go
+	go run ./cmd/main.go
 
 # If you wish to build the manager image targeting other platforms you can use the --platform flag.
 # (i.e. docker build --platform linux/arm64). However, you must enable docker buildKit for it.

PROJECT

@@ -71,4 +71,12 @@ resources:
   kind: SchedulingConfigTemplate
   path: github.com/NexusGPU/tensor-fusion-operator/api/v1
   version: v1
+- api:
+    crdVersion: v1
+    namespaced: true
+  controller: true
+  domain: tensor-fusion.ai
+  kind: ClientProfile
+  path: github.com/NexusGPU/tensor-fusion-operator/api/v1
+  version: v1
 version: "3"

api/v1/clientprofile_types.go

@@ -0,0 +1,75 @@
+/*
+Copyright 2024.
+
+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 v1
+
+import (
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+)
+
+// +kubebuilder:validation:Enum=low;medium;high;critical
+type QoSLevel string
+
+const (
+	QoSLow      QoSLevel = "low"
+	QoSMedium   QoSLevel = "medium"
+	QoSHigh     QoSLevel = "high"
+	QoSCritical QoSLevel = "critical"
+)
+
+// ClientProfileSpec defines the desired state of ClientProfile.
+type ClientProfileSpec struct {
+	// +optional
+	PoolName string `json:"poolName,omitempty"`
+
+	// +optional
+	Resources Resources `json:"resources,omitempty"`
+
+	// +optional
+	// Qos defines the quality of service level for the client.
+	Qos QoSLevel `json:"qos,omitempty"`
+
+	IsLocalGPU bool `json:"isLocalGPU"`
+}
+
+// ClientProfileStatus defines the observed state of ClientProfile.
+type ClientProfileStatus struct {
+}
+
+// +kubebuilder:object:root=true
+// +kubebuilder:subresource:status
+
+// ClientProfile is the Schema for the clientprofiles API.
+type ClientProfile struct {
+	metav1.TypeMeta   `json:",inline"`
+	metav1.ObjectMeta `json:"metadata,omitempty"`
+
+	Spec   ClientProfileSpec   `json:"spec,omitempty"`
+	Status ClientProfileStatus `json:"status,omitempty"`
+}
+
+// +kubebuilder:object:root=true
+
+// ClientProfileList contains a list of ClientProfile.
+type ClientProfileList struct {
+	metav1.TypeMeta `json:",inline"`
+	metav1.ListMeta `json:"metadata,omitempty"`
+	Items           []ClientProfile `json:"items"`
+}
+
+func init() {
+	SchemeBuilder.Register(&ClientProfile{}, &ClientProfileList{})
+}

api/v1/gpunode_types.go

@@ -102,7 +102,7 @@ type GPUNodeAllocationDetails struct {
 
 	Requests GPUResourceUnit `json:"requests"`
 	Limits   GPUResourceUnit `json:"limits"`
-	QoS      string          `json:"qos,omitempty"`
+	QoS      QoSLevel        `json:"qos,omitempty"`
 }
 
 // +kubebuilder:validation:Enum=Pending;Provisioning;Migrating;Running;Succeeded;Failed;Unknown;Destroying

api/v1/gpupool_types.go

@@ -260,14 +260,14 @@ type AlertConfig struct {
 // Define different QoS and their price.
 type QosConfig struct {
 	Definitions []QosDefinition `json:"definitions,omitempty"`
-	DefaultQoS  string          `json:"defaultQoS,omitempty"`
+	DefaultQoS  QoSLevel        `json:"defaultQoS,omitempty"`
 	Pricing     []QosPricing    `json:"pricing,omitempty"`
 }
 
 type QosDefinition struct {
-	Name        string `json:"name,omitempty"`
-	Description string `json:"description,omitempty"`
-	Priority    int    `json:"priority,omitempty"` // Range from 1-100, reflects the scheduling priority when GPU is full and tasks are in the queue.
+	Name        QoSLevel `json:"name,omitempty"`
+	Description string   `json:"description,omitempty"`
+	Priority    int      `json:"priority,omitempty"` // Range from 1-100, reflects the scheduling priority when GPU is full and tasks are in the queue.
 }
 
 type GPUResourceUnit struct {
@@ -292,7 +292,7 @@ type GPUOrCPUResourceUnit struct {
 }
 
 type QosPricing struct {
-	Qos string `json:"qos,omitempty"`
+	Qos QoSLevel `json:"qos,omitempty"`
 
 	Requests GPUResourcePricingUnit `json:"requests,omitempty"`
 

api/v1/schedulingconfigtemplate_types.go

@@ -152,10 +152,10 @@ type ScaleToZero struct {
 }
 
 type AutoFreeze struct {
-	Qos             string `json:"qos,omitempty"`
-	FreezeToMemTTL  string `json:"freezeToMemTTL,omitempty"`
-	FreezeToDiskTTL string `json:"freezeToDiskTTL,omitempty"`
-	Enable          *bool  `json:"enable,omitempty"`
+	Qos             QoSLevel `json:"qos,omitempty"`
+	FreezeToMemTTL  string   `json:"freezeToMemTTL,omitempty"`
+	FreezeToDiskTTL string   `json:"freezeToDiskTTL,omitempty"`
+	Enable          *bool    `json:"enable,omitempty"`
 }
 
 type SmartSchedulerModelInput struct {

api/v1/tensorfusionconnection_types.go

@@ -35,6 +35,10 @@ type Resources struct {
 type TensorFusionConnectionSpec struct {
 	PoolName  string    `json:"poolName"`
 	Resources Resources `json:"resources"`
+
+	// +optional
+	// localGpu mode will schedule the GPU in advance
+	GPUs []string `json:"gpu"`
 }
 
 type TensorFusionConnectionPhase string
@@ -50,7 +54,7 @@ const (
 type TensorFusionConnectionStatus struct {
 	Phase         TensorFusionConnectionPhase `json:"phase"`
 	ConnectionURL string                      `json:"connectionURL"`
-	QosClass      string                      `json:"qosClass,omitempty"`
+	QoS           QoSLevel                    `json:"qos,omitempty"`
 	GPU           string                      `json:"gpu,omitempty"`
 }