I run MinIO in a development environment on a single EC2 instance managed by an Auto Scaling Group. After OS patching, MinIO becomes unavailable, so the usual recovery process is to terminate the instance and let the ASG launch a replacement. The replacement runs user-data/cloud-init, downloads the MinIO binary, applies the configuration, and starts the service.
This worked after the August patch cycle, but after the September cycle the replacement instance still returns HTTP 505 from the application. Cloud-init logs show that the MinIO binary download is failing with HTTP 413. The download URL in the bootstrap script also appears to have stopped working when opened directly.
I understand MinIO's binary distribution or download process may have changed. Did MinIO stop providing or change access to these binaries between the two patch cycles? Could that explain why the older replacement worked while the newer one fails? What is the recommended way to install and pin a specific supported MinIO release in an EC2 launch template or bootstrap script?
3 Answers
There have also been changes around the availability and maintenance of some MinIO distribution options, so an older release may still be retrievable while a newer or generic endpoint is not. If you use a container instead, select an explicit image tag rather than latest; otherwise download an explicitly versioned release artifact from a dependable source, validate it, and test the bootstrap script independently of the ASG.
Do not depend on a moving download URL in user-data. Pin a known, supported MinIO release in the launch template or bootstrap process, verify the downloaded file’s SHA-256 checksum, and make the script stop if the response is anything other than HTTP 200. Comparing the resolved URL and headers from the August and September instances should confirm whether the endpoint changed.
The failure is probably related to the download endpoint or a proxy in front of it, rather than MinIO returning an application error. An HTTP 413 means the request was rejected before the binary was successfully retrieved. Check the exact URL and response headers from the new instance with commands such as curl -I and curl -v, and capture the response body in the cloud-init logs.
The URL in the user-data script appears to be the issue; it no longer works even when pasted into a browser. Updating the endpoint or using a controlled internal artifact location should make the replacement process more reliable.