模型部署资源分配部分

src/main/java/com/bipt/intelligentapplicationorchestrationservice/deploy/deployment/ResourceAllocator.java

@@ -0,0 +1,74 @@
+package com.bipt.intelligentapplicationorchestrationservice.deploy.deployment;
+
+import com.bipt.intelligentapplicationorchestrationservice.deploy.entity.DeploymentResource;
+import com.bipt.intelligentapplicationorchestrationservice.gpu.model.entity.GpuResource;
+import com.bipt.intelligentapplicationorchestrationservice.utils.ConfigConstants;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.beans.factory.annotation.Value;
+import org.springframework.stereotype.Component;
+import org.springframework.web.client.ResourceAccessException;
+
+import java.util.Comparator;
+import java.util.List;
+
+@Component
+public class ResourceAllocator {
+
+    @Autowired
+    private ConfigConstants config;
+
+    //获取剩余内存
+    private int getRemainingMemory(GpuResource resource){
+        return resource.getGPUMaxMemory()-resource.getGPUMemorySize();
+    }
+
+    public DeploymentResource allocate(
+            List<GpuResource> resources,
+            int requiredMemory,
+            String modelId,
+            boolean isGray
+    ){
+        resources.sort(Comparator.comparingInt(GpuResource::getGPUMemorySize));
+
+        //第一轮分配
+        for(GpuResource resource:resources){
+            if(getRemainingMemory(resource) >= requiredMemory) {
+                return createResource(resource, modelId, isGray);
+            }
+        }
+
+        //第二轮分配
+        return defragmentation(resources,requiredMemory, modelId, isGray);
+    }
+
+
+    private DeploymentResource defragmentation(
+            List<GpuResource> resources,
+            int requiredMemory,
+            String modelId,
+            boolean isGray
+    ){
+        //按内存碎片大小排序（最小碎片优先）
+        resources.sort(Comparator.comparingDouble(
+                r -> (double)getRemainingMemory(r) / r.getGPUMaxMemory()));
+
+        for(GpuResource resource:resources){
+            if(getRemainingMemory(resource) >= requiredMemory){
+                return createResource(resource, modelId, isGray);
+            }
+        }
+        throw new ResourceAccessException("GPU资源不足");
+    }
+
+    private DeploymentResource createResource(GpuResource gpu, String modelId, boolean isGray){
+        String urlType = isGray ? "gray":"prod";
+        String url = String.format(
+                config.URL_TEMPLATE,
+                gpu.getIp(),
+                config.MODEL_PORT,
+                modelId,
+                urlType
+        );
+        return new DeploymentResource(gpu, url);
+    }
+}

src/main/java/com/bipt/intelligentapplicationorchestrationservice/deploy/entity/DeployRequest.java

@@ -0,0 +1,4 @@
+package com.bipt.intelligentapplicationorchestrationservice.deploy.entity;
+
+public class DeployRequest {
+}

src/main/java/com/bipt/intelligentapplicationorchestrationservice/deploy/entity/DeployResponse.java

@@ -0,0 +1,25 @@
+package com.bipt.intelligentapplicationorchestrationservice.deploy.entity;
+
+public class DeployResponse<T> {
+    private boolean isSuccess;
+    private String errorInfo;
+    private int status;
+    private T data;
+
+    public DeployResponse(boolean b, String s, int i, T data) {
+        isSuccess = b;
+        errorInfo = s;
+        status = i;
+        this.data = data;
+    }
+
+    // 成功响应
+    public static <T> DeployResponse<T> success(T data) {
+        return new DeployResponse<>(true, "", 200, data);
+    }
+
+    // 失败响应
+    public static <T> DeployResponse<T> fail(int status, String error) {
+        return new DeployResponse<>(false, error, status, null);
+    }
+}

src/main/java/com/bipt/intelligentapplicationorchestrationservice/deploy/entity/DeploymentResource.java

@@ -0,0 +1,12 @@
+package com.bipt.intelligentapplicationorchestrationservice.deploy.entity;
+
+import com.bipt.intelligentapplicationorchestrationservice.gpu.model.entity.GpuResource;
+import lombok.AllArgsConstructor;
+import lombok.Getter;
+
+@Getter
+@AllArgsConstructor
+public class DeploymentResource {
+    private final GpuResource gpu;
+    private final String url;
+}