Open teaglebuilt opened 3 years ago
Hi @teaglebuilt. Thanks for the question, it's a good one!
We've noticed spawning pod for every node is an overkill, but what we've considered is using kedro
node tags to allow grouping node executions within a single pod.
I'm keeping this ticket open until we provide a follow up issue with implementation guidelines.
@em-pe yep, and spawning per node is also a great solution that we can use as well. Although some nodes might be a process finishing faster than it would take to spin up a pod...but still want the visibility of that node, so you do not want to group it into another execution process.
For that reason, I have looked at either telling the Kubernetes Pod Operator to do the same logic as the reattach on restart
for directing a task back to the same pod for execution.
Or, also the possibility of mixing the logic of the Sub Dag Operator with the Kubernetes Pod Operator, which I anticipate will create bottle necks with the scheduler. Although, the scheduler can scale now, so it migiht not be a problem...
Instead of launching a pod per node....could i launch a pod for the entire dagrun with the localexecutor? If so, I was curious if anyone had tried this?
I have tasks that are too quick to spin up pods, i was hoping to have a pod per dag/ dag run.