Scheduling configuration

Resource usage constraints

Configure the scope of resource quota usage in the current workspace, such as which modules, roles, and quotas are accessible. If you don't need fine-grained control, skip this configuration — all modules are available by default.

• Accessible modules: Restrict which modules (DSW, DLC, or EAS) a resource quota can be used for.

• Accessible roles: Restrict access to a resource quota to specific roles only.

• Usage limits: Set the maximum GPUs, CPU cores, and memory that different roles can use from a resource quota when submitting jobs within a specified priority range.

  Important

  If a user matches multiple constraints, the strictest one applies. For example, if user A matches both the admin constraint (priority 1–7, up to 100 GPUs) and the member-A constraint (priority 7–9, up to 10 GPUs) for a given resource quota, user A can use at most 10 GPUs when submitting a priority-7 job on that quota.

• Resource spec templates: When a resource quota has spec templates configured, users must select a spec template to use that quota when creating DSW instances or DLC jobs.

DLC

• Maximum run duration: Set the maximum time a job can run. Jobs that exceed this limit are terminated automatically.

• Job priority: Restrict the maximum priority that different roles or members can use when submitting jobs.

• Allow public resource usage: When disabled, on-demand DLC jobs can't be created in the current workspace, preventing unexpected charges.

• Maximum wait duration: Set timeout thresholds for job wait time, queue time, and environment preparation time within a resource quota. A notification is sent when any threshold is exceeded.

  Note

  To receive timeout alerts, configure notification rules for DLC jobs on the Event Notification Configuration page. Without that configuration, setting thresholds here alone won't trigger any alerts.

DSW

• Auto-shutdown policy:

  • Shutdown policy: Trigger automatic shutdown based on instance runtime or resource utilization (such as idle duration or CPU, memory, and GPU usage) to reduce costs.

  • Exclusion policy: Exempt specific DSW instances from auto-shutdown based on instance priority or instance name.

  • Environment save policy: Instances using cloud disks save their environment automatically — no configuration is needed here. For non-cloud-disk instances, administrators can enable Save environment image before auto-shutdown as needed.

    Note: If the image is large, saving may take a long time or fail. This can lower the shutdown success rate and delay resource release.

• Enable auto-migration for abnormal node instances: When enabled, if a node in the underlying Lingjun computing cluster fails or experiences a fault, the system automatically shuts down and restarts affected instances to support the node self-healing process and maintain full resource availability. The environment image is saved during the restart, but any running processes can't be recovered.

• Instance priority: Restrict the maximum priority that different roles or members can use when creating instances.

• Allow public resource usage: When disabled, on-demand DSW instances can't be created in the current workspace, preventing unexpected charges.

• Maximum wait duration: Set timeout thresholds for resource quota usage. Instances are stopped when queue time or environment preparation time exceeds its threshold.

EAS

• Service priority: Set the resource usage priority by workspace role or member. Priority ranges from 1 to 9, where 1 is the lowest and 9 is the highest.

• Allow public resource usage: When disabled, on-demand EAS services can't be created in the current workspace, preventing unexpected charges.