备份yaml方便测试

simulation-resource-scheduler/src/main/java/com/css/simulation/resource/scheduler/domain/service/ProjectDomainService.java

@@ -108,30 +108,31 @@ public class ProjectDomainService {
             String replace4 = replace3.replace("kafka-partition", "\"" + kafkaPartition + "\"");
             String replace5 = replace4.replace("kafka-offset", "\"" + kafkaOffset + "\"");
             String replace6 = replace5.replace("minio-ip", minioConfiguration.getEndpointWithoutHttp());
-            String replace7 = replace6.replace("minio-access-key", minioConfiguration.getAccessKey());
-            String replace8 = replace7.replace("minio-secret-key", minioConfiguration.getSecretKey());
+            String replace7 = replace6.replace("minio-bucket", minioConfiguration.getBucketName());
+            String replace8 = replace7.replace("minio-access-key", minioConfiguration.getAccessKey());
+            String replace9 = replace8.replace("minio-secret-key", minioConfiguration.getSecretKey());
 
-            String replace9 = replace8.replace("algorithm-container", "algorithm-" + projectId);
-            String replace10 = replace9.replaceAll("algorithm-image", algorithmDockerImage);
+            String replace10 = replace9.replace("algorithm-container", "algorithm-" + projectId);
+            String replace11 = replace10.replaceAll("algorithm-image", algorithmDockerImage);
 
-            String replace11 = replace10.replace("pod-name", podName); // pod 名称包括 projectId 和 随机字符串
-            String replace12 = replace11.replace("namespace-name", kubernetesConfiguration.getNamespace()); // pod 名称包括 projectId 和 随机字符串
-            String replace13 = replace12.replace("node-name", nodeName);     // 指定 pod 运行节点
+            String replace12 = replace11.replace("pod-name", podName); // pod 名称包括 projectId 和 随机字符串
+            String replace13 = replace12.replace("namespace-name", kubernetesConfiguration.getNamespace()); // pod 名称包括 projectId 和 随机字符串
+            String replace14 = replace13.replace("node-name", nodeName);     // 指定 pod 运行节点
 
-            String replace14;
+            String replace15;
             if (cpuOrder != null) {
-                replace14 = replace13.replace("cpu-order", "\"" + cpuOrder + "\"");     // 指定 cpu 编号
+                replace15 = replace14.replace("cpu-order", "\"" + cpuOrder + "\"");     // 指定 cpu 编号
                 stringRedisTemplate.opsForValue().set("project:" + projectId + ":pod:" + podName + ":cpu", String.valueOf(cpuOrder));    //  pod 运行使用的 cpu编号
             } else {
-                replace14 = replace13;
+                replace15 = replace14;
             }
 
             if (DictConstants.USE_GPU.equals(isChoiceGpu)) {
-                String replace15 = replace14.replace("vtd-image", kubernetesConfiguration.getImageVtdGpu());
-                finalYaml = replace15.replace("vtd-command", kubernetesConfiguration.getCommandVtdGpu());
+                String replace16 = replace15.replace("vtd-image", kubernetesConfiguration.getImageVtdGpu());
+                finalYaml = replace16.replace("vtd-command", kubernetesConfiguration.getCommandVtdGpu());
             } else if (DictConstants.NOT_USE_GPU.equals(isChoiceGpu)) {
-                String replace15 = replace14.replace("vtd-image", kubernetesConfiguration.getImageVtdNogpu());
-                finalYaml = replace15.replace("vtd-command", kubernetesConfiguration.getCommandVtdNogpu());
+                String replace16 = replace15.replace("vtd-image", kubernetesConfiguration.getImageVtdNogpu());
+                finalYaml = replace16.replace("vtd-command", kubernetesConfiguration.getCommandVtdNogpu());
             } else {
                 throw new RuntimeException("是否使用 gpu：" + isChoiceGpu);
             }

simulation-resource-scheduler/src/main/resources/kubernetes/template/pod/carsim-pod-template.yaml

@@ -20,10 +20,10 @@ spec:
       hostnames:
         - gpu001
   initContainers:
-    - name: initAlgorithmImageContainer
+    - name: init
       image: algorithm-image
       imagePullPolicy: Always
-      command: [ 'sh', '-c', 'echo algorithm image downloaded && sleep 2' ]
+      command: ['sh', '-c', 'echo algorithm image downloaded && sleep 2']
   containers:
     - name: vtd-container
       image: vtd-image
@@ -61,7 +61,7 @@ spec:
         privileged: true
     - name: algorithm-container
       image: algorithm-image
-      imagePullPolicy: Always
+      imagePullPolicy: Never
       command: [ "/bin/sh", "-c", "/run.sh; touch /tmp/hello.txt;while true;do /bin/echo $(date +%T) >> /tmp/hello.txt; sleep 600; done;" ]
     - name: carsim-container
       image: carsim-image
@@ -88,13 +88,15 @@ spec:
           value: par-path
         - name: LOG_PATH
           value: log/carsim.log
+        - name: MINIO_BUCKET_NAME
+          value: minio-bucket
       ports:
         - name: carsimport
           containerPort: 8888
-#      resources:
-#        requests:
-#          cpu: "1"
-#          memory: "2Gi"
+  #      resources:
+  #        requests:
+  #          cpu: "1"
+  #          memory: "2Gi"
   restartPolicy: Never
   volumes:
     - name: nvidia0
@@ -102,4 +104,4 @@ spec:
         path: /dev/nvidia0
     - name: nvidiactl
       hostPath:
-        path: /dev/nvidiactl
+        path: /dev/nvidiactl

simulation-resource-scheduler/src/main/resources/kubernetes/template/pod/vtd-pod-template.yaml

@@ -22,15 +22,16 @@ spec:
       hostnames:
         - gpu001
   initContainers:
-    - name: initAlgorithmImageContainer
+    - name: init
       image: algorithm-image
       imagePullPolicy: Always
       command: ['sh', '-c', 'echo algorithm image downloaded && sleep 2']
   containers:
     - name: vtd-container
       image: vtd-image
-      imagePullPolicy: Never
+      imagePullPolicy: Always
       command: [ "/Controller/VTDController", "vtd-command", "kafka-topic" ]
+      #command: [ "/Controller/VTDController", "/Controller/config/docker_cloud_algContest.ini", "kafka-topic" ]
       resources:
         limits:
           nvidia.com/gpu: 1
@@ -60,6 +61,8 @@ spec:
           value: kafka-offset
         - name: CPU_ORDER
           value: cpu-order
+        - name: MINIO_BUCKET_NAME
+          value: minio-bucket
     - name: algorithm-container
       image: algorithm-image
       imagePullPolicy: Never