vllm-project / vllm-project/aibrix

Parallelize initContainer and main container image downloading

Open
#184 0 comments 0 reactions 0 assignees View on GitHub
area/acceleration kind/feature priority/important-soon
Dominant language
Go
Stars
5.1k
Forks
694
Avg merge
1d 19h
Merged PRs (30d)
98

Description

### 🚀 Feature Description and Motivation

![image](https://github.com/user-attachments/assets/5db49371-71f3-4f12-a882-2823c82dbc46)

Main container image is pulled after initContainer is done. If the initContainer runs time consuming work, there's an opportunity to download the image at the same time. In this case, it's faster that the workflow that pull image -> starts the engine -> pull model from remote.

### Use Case

_No response_

### Proposed Solution

_No response_

Contributor guide

Open the contributing guide

Research direction

Start by locating the workflow that pulls the initContainer and main container images, then trace how the engine starts and the model is fetched. Define how image downloads can overlap without changing container startup ordering, and verify that the resulting workflow preserves the intended engine and model initialization behavior.

Written by the indexing model from the issue text.

Assessment

Tech stack
kubernetes
Domain
infrastructure
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
30/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.