Skip to content

Conversation

@gjpower
Copy link
Contributor

@gjpower gjpower commented Oct 14, 2024

  • /v1/completions endpoint was missing an exit_stack.close() call

This was causing the llama proxy lock to be retained until exit_stack is cleaned up by garbage collector

@gjpower
Copy link
Contributor Author

gjpower commented Oct 15, 2024

#1798 supersedes this MR as I found thread starvation issue was not just due to improper closing of exit_stack but due to thread starvation from dependency pull of llama_proxy locking all worker threads

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant