Currently, spawning servers is often slow. It takes minutes to obtain a fresh EC2 instance and download the ECR image.
Add a placeholder pod (that uses the current singleuser container images but executes a no-op like sleep infinity) with a lowered preemption priority, that will keep a warm instance ready for the next singleuser pod to expropriate.
Could use KEDA cron to scale this deployment, to disengage startup-acceleration outside of work hours.
Currently, spawning servers is often slow. It takes minutes to obtain a fresh EC2 instance and download the ECR image.
Add a placeholder pod (that uses the current singleuser container images but executes a no-op like
sleep infinity) with a lowered preemption priority, that will keep a warm instance ready for the next singleuser pod to expropriate.Could use KEDA cron to scale this deployment, to disengage startup-acceleration outside of work hours.
Note, will be unneeded until #12 resolved.