Is it a bad idea to upgrade your Dagster Agent version while a job is running (k8s op-per-pod execution)? Meaning, upgrade the Dagster version specified in Helm chart values and apply the change. I assume that would lead to brief Agent downtime, but I don't know if the job would be sensitive to that (lose state, have any timeouts).

The job should be totally isolated from the agent - as you say there will be a brief agent downtime but it shouldn't affect running jobs