4장. JobSet Operator

Use the JobSet Operator on OpenShift Container Platform to manage large, distributed, and coordinated computing workloads, such as high-performance computing (HPC) or artificial intelligence (AI) training, and gain automatic stability, coordination, and failure recovery.

The JobSet Operator is based on the JobSet open source project.

JobSet Operator is designed to manage a group of jobs as a single, coordinated unit. This is especially useful for fields like HPC and training massive AI models where you need a team of machines to run for hours or days.

You can use the JobSet Operator to solve problems that are too big or too complex for a standard OpenShift Container Platform job. The JobSet Operator provides coordination, stability, and recovery.

The JobSet Operator automatically sets up stable headless service to get an IP address so workers can find and communicate with each other, even after a failure and restart. It also provides automatic failure recovery. If one small part of a large training job fails, the Operator can be configured to restart the entire group of workers from a saved checkpoint. This saves time and computing costs.

The JobSet Operator offers startup control, allowing you to define a specific startup sequence to ensure dependencies are met. For example, making sure the leader is running before any workers attempt to connect.

JobSet Operator makes managing large, distributed, and coordinated computing tasks on OpenShift Container Platform easier, turning many individual components into one resilient and manageable system.