Deployments keep a number of identical Pods running. Other controllers fit other shapes of work.
A DaemonSet runs one Pod on every eligible node, and on new nodes as they join: log agents, monitoring and network plugins. Eligibility comes from taints, node selectors and affinity, so a DaemonSet skips the control plane unless it tolerates its taint.
A Job runs Pods until enough of them succeed. completions says how many successes it needs and parallelism how many Pods run at once; a failed Pod counts against backoffLimit. Job Pods must use restartPolicy Never or OnFailure. A CronJob creates a Job on a cron schedule (minute, hour, day of month, month, day of week), in UTC unless timeZone says otherwise, and concurrencyPolicy decides what happens when a run is still going at the next tick: Allow, Forbid or Replace.
kubectl create job --from=cronjob/NAME runs a CronJob once now, which is the quickest way to test it.
Agents on every node, a finished Job, and a CronJob run on demand.