|
@@ -93,13 +93,20 @@ func RunWaitingCluster() {
|
|
}
|
|
}
|
|
|
|
|
|
// 取出但不移除
|
|
// 取出但不移除
|
|
- firstTaskCacheJson, err := infra.GlobalRedisClient.LIndex(global.KeyTaskQueueWaitingCluster, 0).Result()
|
|
|
|
- if err != nil {
|
|
|
|
- infra.GlobalLogger.Error("取出集群等待队列中的头元素报错,错误信息为:", err)
|
|
|
|
- global.GpuNodeListMutex.Unlock()
|
|
|
|
- continue
|
|
|
|
|
|
+ {
|
|
|
|
+ firstTaskCacheJson, err := infra.GlobalRedisClient.LIndex(global.KeyTaskQueueWaitingCluster, 0).Result()
|
|
|
|
+ if err != nil {
|
|
|
|
+ infra.GlobalLogger.Error("取出集群等待队列中的头元素报错,错误信息为:", err)
|
|
|
|
+ global.GpuNodeListMutex.Unlock()
|
|
|
|
+ continue
|
|
|
|
+ }
|
|
|
|
+ firstTaskCache, err = JsonToTaskCache(firstTaskCacheJson)
|
|
|
|
+ if err != nil {
|
|
|
|
+ infra.GlobalLogger.Error(err)
|
|
|
|
+ global.GpuNodeListMutex.Unlock()
|
|
|
|
+ continue
|
|
|
|
+ }
|
|
}
|
|
}
|
|
-
|
|
|
|
// --------------- 下载算法 ---------------
|
|
// --------------- 下载算法 ---------------
|
|
infra.GlobalLogger.Infof("开始下载算法 %v。", firstTaskCache.AlgorithmObjectKey)
|
|
infra.GlobalLogger.Infof("开始下载算法 %v。", firstTaskCache.AlgorithmObjectKey)
|
|
algorithmTarName = filepath.Base(firstTaskCache.AlgorithmObjectKey)
|
|
algorithmTarName = filepath.Base(firstTaskCache.AlgorithmObjectKey)
|
|
@@ -114,12 +121,6 @@ func RunWaitingCluster() {
|
|
}
|
|
}
|
|
infra.GlobalLogger.Infof("下载算法 %v 成功。", firstTaskCache.AlgorithmObjectKey)
|
|
infra.GlobalLogger.Infof("下载算法 %v 成功。", firstTaskCache.AlgorithmObjectKey)
|
|
|
|
|
|
- firstTaskCache, err = JsonToTaskCache(firstTaskCacheJson)
|
|
|
|
- if err != nil {
|
|
|
|
- infra.GlobalLogger.Error(err)
|
|
|
|
- global.GpuNodeListMutex.Unlock()
|
|
|
|
- continue
|
|
|
|
- }
|
|
|
|
err = domain.AddRunningCluster(firstTaskCache, gpuNode.Hostname)
|
|
err = domain.AddRunningCluster(firstTaskCache, gpuNode.Hostname)
|
|
if err != nil {
|
|
if err != nil {
|
|
infra.GlobalLogger.Error(err)
|
|
infra.GlobalLogger.Error(err)
|