Skip to content

terraform/aws: example for routing one S3 bucket over PrivateLink #52

Description

@maxiscoding28

Pointing an app connector at a public S3 endpoint does not work well. S3 regional endpoints resolve to a large pool of addresses that rotates. The connector keeps discovering new addresses, so the route set grows and never settles. Clients see the churn as ERR_NETWORK_CHANGED in Chrome.

People hit this because the guidance points both ways. Tailscale ships an aws-s3-<region> preset app, while the best-practices guide recommends serving multi-tenant content outside app connectors. S3 regional endpoints are shared multi-tenant address space, but nothing connects those two pages.

The fix is to give the bucket a stable private address first, then advertise that. Three pieces have to line up:

  1. An S3 interface VPC endpoint with private_dns_only_for_inbound_resolver_endpoint = false. Resolving the bucket domain inside the VPC now returns the endpoint ENI.
  2. A split DNS entry on the exact bucket FQDN. Split DNS does not match siblings, so other buckets stay public. This is what scopes the change to one bucket.
  3. The connector advertises the VPC resolver and the endpoint ENI. The resolver is the only one that knows the private DNS mapping, and a client outside the VPC cannot reach it directly.

The same shape applies to any interface-endpoint service, not just S3.

I have this built as terraform/aws/aws-app-connector-s3-privatelink. It passes tflint, fmt, and validate, and I have applied it and confirmed the private path end to end.

Opened as a draft PR in #51.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions