[Autoscaler] Autoscaler on ray 1.3 with minikube does not scale down


I started minikube with 24 CPUs and 24GB of ram.The cluster was started with 2 workers excluding head node with max of 12 workers set in example-full.yaml. I ran 1500 tasks on ray head node and I see that ray scaled up the nodes

ray.init(address=“auto” )
2021-04-23 07:21:56,873 INFO worker.py:655 – Connecting to existing Ray cluster at address:
{‘node_ip_address’: ‘’, ‘raylet_ip_address’: ‘’, ‘redis_address’: ‘’, ‘object_store_address’: ‘/tmp/ray/session_2021-04-23_07-16-41_226731_155/sockets/plasma_store’, ‘raylet_socket_name’: ‘/tmp/ray/session_2021-04-23_07-16-41_226731_155/sockets/raylet’, ‘webui_url’: ‘’, ‘session_dir’: ‘/tmp/ray/session_2021-04-23_07-16-41_226731_155’, ‘metrics_export_port’: 52943, ‘node_id’: ‘40a5e6554a7026e0d029ad05ba02b430d9b8070d6f719ab850f5e58d’}

res = ray.get([f.remote() for _ in range(1500)])
(autoscaler +3m2s) Tip: use ray status to view detailed autoscaling status. To disable autoscaler event messages, you can set AUTOSCALER_EVENTS=0.
(autoscaler +3m2s) Adding 5 nodes of type worker_node.

I did exit from the head node and as I understand all the tasks will be purged and I was hoping that the workers would be scaled down too but I don’t see that happening:

example-cluster-ray-head-gfvt6 1/1 Running 0 36m
example-cluster-ray-worker-4wcds 1/1 Running 0 29m
example-cluster-ray-worker-8kbcb 1/1 Running 0 29m
example-cluster-ray-worker-hkzkf 1/1 Running 0 29m
example-cluster-ray-worker-tkxm9 1/1 Running 0 29m
example-cluster-ray-worker-tsmbd 1/1 Running 0 29m
example-cluster-ray-worker-v6ffd 1/1 Running 0 35m
example-cluster-ray-worker-vppcr 1/1 Running 0 35m

I cannot reproduce this issue. maybe I did ray.init(num_cpus = 12) instead of ray.init(address= “auto”) and since the minikube cluster didn’t have enough resources it caused the autoscaler to break???

Hey @asm582 , sorry for the long wait. This is due to your question not being categorized. Could you make sure that for any new posts you add a category (e.g. “Ray Core” or “RLlib”) to your question? This helps us assign the right person to respond more quickly.

Are you still seeing this problem or did this get resolved (you said, you “cannot reproduce”)?