[vllm] test out ray v2 executor #1660
Conversation
There was a problem hiding this comment.
Code Review
This pull request enables the Ray V2 executor backend for vLLM by setting the VLLM_USE_RAY_V2_EXECUTOR_BACKEND environment variable in both the runtime environment utility and the test configuration. The reviewer suggests avoiding hardcoding this value to allow users to override it via environment variables and recommends moving the assignment into a backend-specific block to prevent environment pollution for other backends.
| # manually set this for testing everywhere | ||
| env_vars["VLLM_USE_RAY_V2_EXECUTOR_BACKEND"] = "1" |
There was a problem hiding this comment.
Hardcoding VLLM_USE_RAY_V2_EXECUTOR_BACKEND to "1" unconditionally prevents users from overriding this setting via environment variables. It is better to check if the variable is already set in os.environ and only apply the default if it is missing. This ensures that users can explicitly disable the V2 executor if they encounter issues. Additionally, since this is a vLLM-specific setting, it would ideally be placed within the vLLM backend check block (around line 629) to avoid polluting the environment for other backends.
| # manually set this for testing everywhere | |
| env_vars["VLLM_USE_RAY_V2_EXECUTOR_BACKEND"] = "1" | |
| # Use Ray V2 executor for vLLM by default, but allow override from environment | |
| env_vars["VLLM_USE_RAY_V2_EXECUTOR_BACKEND"] = os.environ.get("VLLM_USE_RAY_V2_EXECUTOR_BACKEND", "1") |
|
regression - train gpu ci all passes but runs a little slower on main at commit this commit - 2 hours 31 mins: https://console.anyscale.com/cld_hxkifz7xa22mwicp21nzkds1lw/prj_4b6c498rypyq6g7yhk6vzgjevt/jobs/prodjob_6uukbqbz2xe4ss84446xlm1p7g?job-logs-section-tabs=application_logs&job-tab=overview maybe new ray executor takes longer to spin down compared to before? megatron models ci passes as before |
Testing for #1653 to see if we need to fix anything on vllm side