We're seeing a high rate of internal build flakes last several days (50%+ across some services):

[internal] load metadata for docker.io/oven/bun:1.1.30
failed to do request: Head "https://registry-1.docker.io/v2/oven/bun/manifests/1.1.30": dial tcp: lookup registry-1.docker.io: i/o timeout
--

This appears to be either an error connecting to docker.io or a failure to set up network for the build environment.

Generally, if users are seeing build failures marked [internal] that could be retried and expect a different result, it is my opinion that railway builder should be retrying them rather than requiring manual intervention (we click "redeploy" and it succeeds, i.e. it's a flake, not a failure).

Alternatively, a configurable build retry number (similar to k8s) could be exposed similar to running containers.