Job priority

reference · Reviewed 2026-07-30 · FEAT-0081 GA

Priority is the cost and reliability setting you choose at submission. It determines which kind of cloud capacity your job runs on, and whether that capacity can be reclaimed mid-run.

The three priorities

Priority	Name	Cheaper	Can be interrupted
ODP	On-Demand Priority	—	No
ODE	On-Demand Economy	Yes	Yes
ODR	On-Demand Reserved	Yes, via commitment	No

ODP is the default: standard cloud capacity, yours until the job finishes.

ODE uses interruptible capacity, which is cheaper because the cloud provider can reclaim it at any time.

ODR runs on the same reliable capacity as ODP, but priced against a commitment your organization has made. Availability depends on your agreement: check with your admin.

What interruption actually means on ODE

When capacity is reclaimed, Rescale does not treat it as a job failure. The job is restarted or requeued automatically:

1. Cloud resources for the interrupted hardware are released.
2. New hardware is requested.
3. Your application restarts, from the beginning, or from a checkpoint if one exists.
4. Input files are re-downloaded onto the new nodes.

Two consequences that matter:

Without checkpointing, an interruption means starting over

An interrupted job restarts your application from the beginning unless it can resume from a checkpoint. On a multi-hour run, the savings from ODE can be wiped out by a single interruption.

ODE and checkpointing go together. Use ODE for short jobs, or for long jobs that checkpoint. Don't use it for long jobs that can't resume.

Don't delete input files while a job is running

Restart re-downloads the original input files. If an input file has been deleted since the job started, the restart fails. This applies to any restart, not just ODE, but ODE makes restarts far more likely.

After repeated interruptions, ODE jobs can move to ODP

If a job keeps getting interrupted, Rescale eventually runs the final attempt on ODP capacity so it can actually complete. You get reliability rather than an endless retry loop: priced accordingly.

ODP jobs always stay on ODP. They're never moved down.

One exception: if the final requeue was caused by license unavailability rather than capacity, the job keeps its original priority.

Choosing

Situation	Use
Short job, cost matters	ODE
Long job that checkpoints	ODE
Long job that can't checkpoint	ODP
Deadline-driven work	ODP
Interactive workstation	ODP — ODE isn't supported
Your org has a reserved commitment	ODR where eligible

Where you set it

On the Hardware Settings panel when creating a job. Not every coretype offers every priority; the available options are filtered to what your selection and organization support.

Priority appears in the Priority column of the jobs list.

Workstations

Workstations don't support ODE, because an interrupted workstation loses your session with no way to save data. See workstation limits.

Related

• Checkpoint long simulations — what makes ODE safe
• Why a job is queued — requeuing and backoff
• Create and submit a job — where to set it
• Hardware & Coretypes — what's available