Set up Arena scheduled jobs

Add a cron schedule to an ArenaJob so Arena runs the same evaluation automatically, such as a nightly regression suite.

Prerequisites

• Enterprise features enabled in your Omnia install.
• An ArenaProject that holds the scenarios to run.

Create the schedule

An ArenaJob with a schedule is a template. Each firing creates an ArenaRun from a copy of the job’s spec.

1. Add a schedule section to the ArenaJob. sourceRef.name is the ArenaProject:

   apiVersion: omnia.altairalabs.ai/v1alpha1
   kind: ArenaJob
   metadata:
     name: nightly-regression
     namespace: arena
   spec:
     sourceRef:
       name: regression-suite
     type: evaluation
     workers:
       replicas: 2
     schedule:
       cron: "0 2 * * *"        # 2:00 AM daily
       timezone: "America/New_York"
       concurrencyPolicy: Forbid

2. Apply it:

   kubectl apply -f nightly-regression.yaml

To create the schedule from the dashboard instead, open the project, select Run this project, and tick Repeat on a schedule. Fill in Cron, Timezone and If a run is still going. The button then reads Create schedule, and nothing runs until the cron first fires. The dialog defaults the timezone to your browser’s, whereas an ArenaJob that names none uses UTC.

cron takes five space-separated fields: minute, hour, day of month, month and day of week (0-6, Sunday is 0). timezone takes an IANA name such as UTC or Europe/London. A cron expression or timezone that does not parse stops the job scheduling (see Troubleshooting).

Schedule	Cron
Every day at 2:00	0 2 * * *
Every hour	0 * * * *
Every 6 hours	0 */6 * * *
Sundays at 3:00	0 3 * * 0
Monday to Friday at 8:00	0 8 * * 1-5
1st of the month at 4:00	0 4 1 * *

A daily schedule in a timezone with daylight saving fires twice on the fall-back day and not at all on the spring-forward day, as a Kubernetes CronJob does. Make scheduled work safe to repeat.

Choose what happens on overlap

concurrencyPolicy decides what a tick does while a previous run is still going. The default is Forbid.

Value	Dashboard label	Behavior
Forbid	Skip the new one	Skips the tick; the job records a RunSkipped event
Allow	Run both	Starts the new run alongside the old one
Replace	Cancel the old one	Sets spec.cancelled on the running runs, then starts the new one

• Pick Forbid when runs compete for the same providers or write results that must not overlap.
• Pick Allow when every tick must run and the cluster has capacity for the overlap.
• Pick Replace when only the latest result matters. It cancels work in progress.

Control how long runs are kept

Set the history limits so a nightly job does not accumulate ArenaRuns. The defaults keep the 3 newest successful runs and 1 failed run:

spec:
  successfulRunsHistoryLimit: 7
  failedRunsHistoryLimit: 3

ttlSecondsAfterFinished sets how long Kubernetes keeps each run’s finished worker Job and pods (default 3600). It does not delete ArenaRuns. See ArenaJob for every schedule and retention field.

To copy each run’s output files to S3, add an output section as described in Configure Arena S3 result storage.

Run the job once now

An ArenaJob with no schedule runs once when the controller first sees it. To get an extra run of a scheduled job’s configuration, create a second job without a schedule:

kubectl create -f - <<EOF
apiVersion: omnia.altairalabs.ai/v1alpha1
kind: ArenaJob
metadata:
  generateName: manual-regression-
  namespace: arena
spec:
  sourceRef:
    name: regression-suite
  type: evaluation
  workers:
    replicas: 2
EOF

Suspend and resume

Suspending stops the job starting new runs and keeps the job and its run history. Runs already going are left alone; to stop one, set spec.cancelled on the ArenaRun.

kubectl patch arenajob nightly-regression -n arena --type=merge -p '{"spec":{"suspend":true}}'

Resume with "suspend":false.

While the job is suspended, status.nextScheduleTime is cleared. After you resume, the job fires once on its next due tick and does not replay the ticks it missed.

From the dashboard, open Runs. Schedules are listed above the runs, with Fires, State, Last, Next and If one is still running columns and a Suspend or Resume button per row. The list appears only when the workspace has a schedule.

Deleting the job deletes its ArenaRuns, because the job owns them.

Verify

1. Check the scheduling status. -o wide adds the next schedule time:

   kubectl get arenajob nightly-regression -n arena -o wide

   status.lastScheduleTime is the last tick that fired and status.nextScheduleTime the next one.

2. List the runs the job created:

   kubectl get arenaruns -n arena -l omnia.altairalabs.ai/arena-job=nightly-regression

3. Read the job’s events. A tick that fires records RunCreated; a tick skipped by Forbid records RunSkipped:

   kubectl describe arenajob nightly-regression -n arena

Phase, progress and results are on each ArenaRun, not on the ArenaJob. See Monitor Arena jobs.

Troubleshooting

The job never fires

Run kubectl describe arenajob nightly-regression -n arena. If the Ready condition is False with reason InvalidSchedule, the cron expression or timezone does not parse. The controller stops until you correct the spec. Also check that spec.suspend is not true.

A tick produced no run

With concurrencyPolicy: Forbid, a tick that arrives while a run is active is skipped and recorded as a RunSkipped event. The skipped tick is not retried when the run finishes.

Related pages

• ArenaJob reference: every schedule field
• Monitor Arena jobs: track runs and results
• Configure Arena S3 result storage: copy scheduled runs’ output files to a bucket