Determine a list of workloads and deployments we're interested in supporting

[RaasAhsan]

workloads

• one-off job
• scheduled, recurring job
• replicated container

need to think about routing traffic eventually

deployments

• canary deployments
• blue-green deployments
• rolling deployments

[JosephMoniz]

What's an example of "one-off job"?

[JosephMoniz]

Do we want dependent jobs? IE: Job A needs to run, but before it runs it needs Job B to run? These are usually pretty good for ETL style pipelines where you need to build out a graph of dependent jobs. Very similar to how AirFlow and Azkaban work.

[RaasAhsan]

I was thinking when you submit a one-off job to the cluster, it immediately gets scheduled and executed -- don't need to specify a specific time or recurrence.

---

Yeah, I think that's a pretty good use case. Perhaps we can configure retry policies for those jobs as well, based on an exit status code? Also, is there ever a need replicate/scale up a job?

---

Autoscaling based on CPU/memory might be pretty cool for the replicated container deployment.