vllm-project / vllm-project/aibrix
Using UDS to improve Inference GW plugin latency
- Dominant language
- Go
- Stars
- 5.1k
- Forks
- 697
- Avg merge
- 1d 19h
- Merged PRs (30d)
- 104
Description
### 🚀 Feature Description and Motivation
Currently we deploy plugin(ext-proc) as a deployment in cluster, which communicates with envoy through gprc call in pod-to-pod networking, which brings external cost for different NetNS, node data transfer.
Support deploying plugin as a sidecar of envoyproxy to use UDS communitcation between containers.
### Use Case
System performance improvemnt
### Proposed Solution
_No response_
Contributor guide
Research direction
The issue names no files, tests, or entry points. Start by tracing the current ext-proc plugin deployment and Envoy gRPC communication, then determine how sidecar deployment and UDS communication should be integrated; done means the plugin runs beside Envoy and the latency improvement is demonstrated.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- go, grpc, kubernetes
- Domain
- distributed-systems, infrastructure, networking
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100