Estimate total completion time for a Kubernetes Job given parallelism and completions.
A Kubernetes Job with N completions and parallelism P runs in ceil(N/P) sequential batches, since only P pods run at once. Total wall-clock time is that batch count multiplied by the average time for one pod to finish. Increasing parallelism reduces total time proportionally until you hit cluster capacity limits, at which point new pods queue as Pending instead of running concurrently.
totalTime = ceil(completions / parallelism) × avgPodDuration
Use a higher percentile (p90) rather than the mean for avgPodDurationSeconds, since the slowest pod in each batch determines when the next batch can start — variance directly extends total wall-clock time.
Only up to the point where the cluster has spare capacity to actually run that many pods concurrently — beyond that, extra parallelism just means more pods sit Pending waiting for resources.
A Job runs once until it reaches its target completions; a CronJob creates a new Job on a repeating schedule. Use the CronJob schedule calculator to plan recurring runs.