UPSTREAM PR #17878: server : run child server on localhost#496
UPSTREAM PR #17878: server : run child server on localhost#496
Conversation
|
Explore the complete analysis inside the Version Insights Performance Analysis Summary: PR #496OverviewThis PR implements a networking configuration fix for the llama.cpp server router, forcing child server instances to bind exclusively to localhost ( Performance ImpactNo measurable performance impact detected. Power consumption analysis across all binaries shows changes below 0.001%:
Inference Performance: No impact on tokens per second. The modified code paths ( Code ChangesThe PR adds a
These changes resolve a routing failure where child servers bound to Security improvement: Child servers are no longer exposed on external network interfaces, reducing attack surface while maintaining functional equivalence for local routing scenarios. |
4f731df to
8e6f6e8
Compare
b9ba67d to
320a1fc
Compare
Mirrored from ggml-org/llama.cpp#17878
When passing in
--host 0.0.0.0, the child runs on host0.0.0.0and the router tries to access it at0.0.0.0. I can't think of why the child should not always run on127.0.0.1.get_free_port()binds toINADDR_ANY, which should select a port that is available across all interfaces. This can be changed toINADDR_LOOPBACKif we ensure the child will only ever bind to127.0.0.1. If not, thenINADDR_ANYis a safe choice.fixes #17862