BUG: healthcheck failed on Windows and errored with "Cluster is not available after multiple attempts"

[fffonion]

BUG: healthcheck failed on Windows and errored with "Cluster is not available after multiple attempts"

[aresnow1]

You can try to increase the value of environment variable XINFERENCE_HEALTH_CHECK_INTERVAL https://inference.readthedocs.io/en/latest/getting_started/environments.html#xinference-health-check-interval.

[fffonion]

Hi @aresnow1 that won't work because it's not because healthecheck starts too soon, but it's trying to connect to 0.0.0.0:port, which isn't supported on windows (WinError 1214).

[hao203]

xinference version 0.8.5, windows 11, python 3.11
同样的问题。

[jaydyi]

@fffonion @hao203
if you use xinference on win, the xinference site's doc : launch on local, use command xinference-local --host 0.0.0.0 --port 9997 will have error like uper's description, you can solve the problem simply just run command xinference-local , then xinference will launch on http://127.0.0.1:9997/ like below:

[qinxuye]

Hi @aresnow1 that won't work because it's not because healthecheck starts too soon, but it's trying to connect to 0.0.0.0:port, which isn't supported on windows (WinError 1214).

Indeed, the windows does not support 0.0.0.0, do you have any idea, maybe raise an error or try to get IP for user?

[qinxuye]

Same issue as #1053

[fffonion]

@qinxuye yeah we probably need to find an available local IP on any interface, and use that for healthcheck. If we just are
just healthchecking the service but not the port connectivity, we can also do some hack to listen both 0.0.0.0 and 127.0.0.1 at the same time if user specified 0.0.0.0, and healthcheck on 127.0.0.1 only.