Environment:Vllm project Vllm AWS ECR
| Knowledge Sources | |
|---|---|
| Domains | Container_Registry, AWS |
| Last Updated | 2026-02-08 00:00 GMT |
Overview
AWS Elastic Container Registry (ECR) environment for storing and distributing vLLM Docker images used in release pipelines, CI/CD testing, and production deployments.
Description
This environment defines the AWS ECR container registry infrastructure that vLLM's release pipeline uses to publish Docker images. ECR serves as the primary container registry for vLLM's official Docker images, which include pre-built binaries for CUDA and ROCm backends along with all required dependencies. The release pipeline builds multi-architecture Docker images (typically x86_64 with CUDA support), tags them with the release version and git SHA, and pushes them to ECR repositories. These images are then used by downstream consumers for production deployments and by CI pipelines as base images for testing. ECR provides image scanning for vulnerability detection, lifecycle policies for automatic cleanup of old images, and IAM-based access control for push/pull authorization.
Usage
The release pipeline (configured in Buildkite) authenticates to ECR using AWS credentials (IAM role or access keys), builds the Docker image from vLLM's Dockerfile, and pushes to the configured ECR repository. Image tags follow the convention vllm/vllm-openai:vX.Y.Z for releases and vllm/vllm-openai:latest for the most recent stable build. CI pipelines pull images from ECR as base images for GPU test execution. Access to the ECR repository requires appropriate AWS IAM permissions for ecr:GetAuthorizationToken, ecr:BatchGetImage, and ecr:PutImage.
Requirements
| Requirement | Value |
|---|---|
| AWS Account | AWS account with ECR access |
| ECR Repository | Configured repository for vLLM Docker images |
| AWS Credentials | IAM role or access key/secret key with ECR permissions |
| Docker | Docker Engine with docker push capability
|
| AWS CLI | aws-cli v2 for ECR authentication (aws ecr get-login-password)
|
| IAM Permissions | ecr:GetAuthorizationToken, ecr:BatchGetImage, ecr:PutImage, ecr:InitiateLayerUpload, ecr:UploadLayerPart, ecr:CompleteLayerUpload |
| Image Format | OCI or Docker v2 manifest |
| Region | AWS region where ECR repository is hosted |