Serve run cli to remote

Ray Libraries (Data, Train, Tune, Serve) Ray Serve
1

How severe does this issue affect your experience of using Ray?

  • High: It blocks me to complete my task.

I tried running

serve run --address 10.184.26.214:8265 capitalizer:capitalizer_app

, where capitalizer:capitalizer_app is a very simple ray serve app that capitalizes. However, I get the following error:

Traceback (most recent call last):
  File "/Users/ammar.alrashed/code/ray_versions/ray_271/lib/python3.8/site-packages/ray/_private/worker.py", line 1598, in init
    _global_node = ray._private.node.Node(
  File "/Users/ammar.alrashed/code/ray_versions/ray_271/lib/python3.8/site-packages/ray/_private/node.py", line 156, in __init__
    self._init_gcs_client()
  File "/Users/ammar.alrashed/code/ray_versions/ray_271/lib/python3.8/site-packages/ray/_private/node.py", line 729, in _init_gcs_client
    raise RuntimeError(
RuntimeError: Failed to connect to GCS.

Does anyone know what the problem is?

2

serve run uses the Ray client address, instead of the dashboard address. The Ray client address format is ray://[hostname]:[head node port]. By default, the head node port is 10001.

In your example, the address is likely:

ray://10.184.26.214:10001

By the way, are you trying to connect to a remote cluster? That ip address (10…) points to a node on the local network, so it won’t work if you’re trying to a connect to a remote cluster. In that case, you would need to use a public ip or hostname.