Yi Cheng

@SongGuyang they are orthogonal Please take [this](https://docs.google.com/document/d/1yYSDuuek_IvZ-cR7HOHzNy_Yn7LIHAka_zpPxzx69oo/edit) doc another look. I think the high-level direction is to make sure what we build fit into the ray world, and not a...

Synced with @SongGuyang offline and we agreed the following plan: - Once we introduced the new fields, we should use this to setup the default working dir for py/cpp/java and...

cc @scv119 the release blocker

@simon-mo @edoakes could you take a look at this ASAP? Please check the PR to get some ideas of what might be wrong. Please let me know if there is...

@sihanwang41 Unfortunately, the nightly tests script still hasn't been merged yet https://github.com/ray-project/ray/pull/27413/files But you can give it a try by running run_gcs_ft_on_k8s.py in ray/release/k8s_tests You need a k8s cluster to...

I run it by killing nodes (run_gcs_ft_on_k8s.py) and With the PR: Without the PR:

Sure, let me give it a try!

@simon-mo Here is the result - SERVE_DEPLOYMENT_HANDLE_IS_SYNC=1 - the failure is around 5%-10% - SERVE_DEPLOYMENT_HANDLE_IS_SYNC=0 - the failure is very high (25%) and the QPS drops - With that PR...

Hmmm, my bad, I run 1) for a longer time and it looks very similar as 3). Maybe we can just disable it temporarily for now? But how is the...

Let me run it for a more longer time