Error when deploying to Sagemaker starting from release 0.3.0

[miko7879]

Since release 0.3.0 we have been unable to deploy vLLM containers to Sagemaker. It has to do with the get_ip() function call in utils.py. In release 0.3.0 we got an OS error, and with 0.3.3 we are getting the following: socket.gaierror: [Errno -2] Name or service not known, when making the following call: s.connect(("dns.google", 80)). Could you please add something to make the get_ip() function fall back to what was there in release 0.2.7 if all else fails?

For reference, here is the function call in 0.2.7: https://github.com/vllm-project/vllm/blob/v0.2.7/vllm/utils.py#L60

In 0.3.0 it was changed to the following, and stopped working on Sagemaker: https://github.com/vllm-project/vllm/blob/v0.3.0/vllm/utils.py#L164

It is now as follows: https://github.com/vllm-project/vllm/blob/v0.3.3/vllm/utils.py#L172

[simon-mo]

This might be fixed by #3176

[miko7879]

I don't believe that #3176 will resolve this because it has been failing with release tag 0.3.0, which uses the same line as in #3176

[miko7879]

I believe this will be resolved by #2961

[a1164714]

I got the same error.

[DefTruth]

same error, even the 8.8.8.8 is unreachable in my env.

[miko7879]

@a1164714 @DefTruth I'm glad I'm not the only one haha. Could you please thumbs up my comment on the release tracker to try and get the fix pushed in to 0.3.4?

#3155 (comment)

[miko7879]

Confirmed that this issue is fixed starting 0.4.0.