ECR Image Pulls: Proxy Bypass for S3 Layer Downloads
Follow image pulls beyond the registry host when a private ECR route still sends layer downloads toward a corporate proxy.
ECR login and manifest retrieval can work while an image layer stalls. Amazon ECR stores layers in S3, and its VPC endpoint guidance treats the ECR API, registry and S3 retrieval path separately.
- Find the failed layer hostname. Keep the hostname from the actual request rather than assuming the registry bypass covers the whole pull.
- Locate the sender. For a Docker pull, inspect the daemon's proxy configuration. A native Docker Engine daemon.json example keeps the private ECR and S3 layer paths together. Environment variables intended for a new application container cannot configure an image download that happens before that container exists.
- Check the intended S3 path. Match bypass entries to the regional endpoint names and the client's supported grammar. Private DNS, route tables, endpoint policy and permissions must also support that route.
| Observed failure | What it tells you |
|---|---|
| Proxy rejection | The layer request reached an intermediary. |
| Direct connection timeout | The bypass may have selected direct routing, but that route can still be unusable. |
Avoid replacing one missing layer host with an unreviewed exception for every service in the provider's domain. A direct exception cannot create a missing gateway route.
Use the actual image's requests and current provider instructions. Special contents, including foreign Windows layers, can introduce other sources; a historical two-host allowlist need not cover them.