Currently, all dask workflows are executed with autoscaler. However, while experimenting with several examples, autoscaler behaves sometimes weirdly by scheduling more than the maximum number of workers or schedulers not being able to detect some workers, although restarting of the workers solve the issue. Therefore, it is better to decide if this option is enabled to cluster administrators.