Run Lifecycle

How jobs are scheduled?

konduktor launch my_job.yaml

What happens next? We manage the infrastructure and orchestration of your workloads. You can check the stage of your run using:

(konduktor) Andrews-MacBook-Air:konduktor asai$ konduktor status
User: asai-c41a
Jobs
NAME            STATUS     RESOURCES                  SUBMITTED   START TIME         END TIME
batch-job-160a  FAILED     1x(2CPU, 2GiMEM, H100:8)   22 mins     08/08/25 04:11PM   08/08/25 04:20PM
batch-job-6fc9  PENDING    1x(2CPU, 2GiMEM, H100:8)   27 mins     08/08/25 04:06PM   -
batch-job-873f  COMPLETED  1x(2CPU, 2GiMEM, H100:8)   33 mins     08/08/25 04:00PM   08/08/25 04:25PM

Shown above are the states of the job lifecycle. For on-demand users, billing is only compute when the workload is considered ACTIVE.

CREATED

At this stage, the workload has been defined and all the pre-requisites to run the code have been uploaded to object storage from the workstation requesting the workload.

PENDING

The workload has been accepted is either waiting for compute to finish being provisioned to fullfill the job requirements or for the containers to be pulled onto a new machine.

SUSPENDED (by system)

The workload has either been newly created or pre-empted by higher priority workloads and is awaiting quota before being rescheduled.

SUSPENDED (by user)

The workload was manually stopped by a user using konduktor stop <JOB_ID>. It can be unsuspended with konduktor start <JOB_ID>.

ACTIVE

The workload is running and performing work. At the beginning of this stage, the workload will synchronize files defined under file_mounts and workdir definitions in the workload definition to the runner. Afterwards it will begin executing, the shell script defined under run

COMPLETED

A terminal state indicating that the all workers/nodes in a workload have completed successfully.

FAILED

A terminal state indicating that at least one worker/node in a workload has failed, either due to infrastructure/hardware faults are application errors.