Problem description

• non-billable deployment,
• the autoscaling auto-chooses a machine based on the job required resources,
  -> if the resources are very near to the limit of a machine, there is a chance the job never runs because the autoscaling does not take in account the fact that a dask-sidecar automatically removes some of the resources to account for OPS containers and other 3rd party needed resources.

Resolution

• make sure the autoscaling takes in account the reserved resources when choosing a machine type