Andrew Richards

08/22/2022, 3:03 PM
I'm experiencing a strange issue with version 2.1.1 where a scheduled flow run will try to run a task on all three of my Kubernetes pods simultaneously. This results in one of the tasks succeeding and two of them failing. Each pod is running on its own node. I am using SequentialTaskRunner with Prefect Shell Task and Process Infrastructure. I do not want to use Kubernetes Infrastructure because I want control over using my own kubectl commands. I'd prefer to spin up Prefect Agent pods like I would with any other process I run in Kubernetes. When I manually run the flow through the UI, it works perfectly and only schedules the task to one pod in the cluster. When I auto-schedule the flow run, it attempts to run the task on all pods in the cluster simultaneously despite them running on their own nodes. The UI thinks it is only running one task which can be seen in the screenshots. Anyone know why this might be?
This seems like a bug in the scheduler since this behavior doesn't occur with manual flow runs

Jeff Hale

08/22/2022, 3:47 PM
I agree, @Andrew Richards. Strange that the behavior is different. Are you able to open an issue?