Question on Practice Set 2

[PeterMescher]

– For the question starting: “Your company regularly executes a batch job process hosted in an on-premises server which takes around 33 hours in total to complete.”

GKE with pre-emptible nodes would seem to be the correct answer since the total job can be broken down into smaller tasks, which are probably better run in pods vs. entire VM’s. The solution says it’s incorrect because the nodes will be pre-empted at some point during the job, but the “correct” answer of using a MIG also says to use pre-emptible VM’s, which will have the same issue. (And I’m not sure what the “Target CPU Utilization” bit has to do with the answer, if anything.) I’d think the GKE solution would be better since the pods can be started up faster when the node underneath them is replaced.