Based on real production needs, I recently made several important updates to https://github.com/shaowenchen/ops and released version v1.0.0. Here I mainly introduce the key features of this release. 1. Support for Multi-Cluster Task Execution In practice, the recommendation is: Create the hosts of the current cluster as Hosts You can create multiple Clusters, and the Cluster objects you own are the clusters under management Task and Pipeline objects are automatically synchronized to all Clusters under the cluster, with no manual trigger required.

1. Jindo Mounting OBS Set the environment variables 1 2 3 4 export ENDPOINT=obs.cn-north-4.myhuaweicloud.com export BUCKET= export AK= export SK= Create the credentials 1 2 3 4 5 6 7 8 9 10 kubectl apply -f - <<EOF apiVersion: v1 kind: Secret metadata: name: myobssecret type: Opaque stringData: fs.obs.accessKeyId: ${AK} fs.

1. Data Processing Architecture It is mainly divided into four layers: Processing capability layer: Spark on Kubernetes provides streaming data processing capability Data management layer: Iceberg provides dataset access operations such as ACID and tables Storage layer: Hive MetaStore manages Iceberg table metadata, PostgreSQL serves as the storage backend for

The DLRover version used in this article is 0.3.7 1. DLRover Operator 1.1 Starting the ElasticJob and ScalePlan Controllers Implementation code: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 // 创建 ElasticJob 的控制器 if err = controllers.NewElasticJobReconciler(mgr, masterImage).SetupWithManager(mgr); err

1. Problems Facing Distributed Training Estimating training resources is difficult and cannot be automated How much compute, how much time, how much bandwidth, how many CPUs, how much memory — without enough accumulated experience it is hard to estimate accurately. The result is over-requesting and over-allocation, causing enormous resource waste.