Personal file share server on AWS — Samba for macOS Finder integration and FileBrowser for web-based file management. Accessible exclusively over the site-to-site VPN.
The live share is still the management-account CDK stack until cutover proof. The prod replacement is HCP Terraform in `terraform/`, workspace `file-share-prod`, trigger `terraform/**`. CDK deploy on push to main is frozen.
| Path | Role |
|---|---|
| `terraform/` | Prod EC2, subnet in the syslog VPC, DLM, and HCP roles |
| `lib/file-share-stack.ts` | Management-account CDK stack, still the live path until decommission |
cdk.context.json # Cached VPC/subnet lookups and the resolved AL2023 AMI
```
`cdk.json` is the CDK app manifest. Its `app` command (`npx tsx bin/app.ts`) runs the TypeScript entry point directly via [tsx](https://github.com/privatenumber/tsx) — no separate compile step is needed for synth or deploy. `bin/app.ts` instantiates `FileShareStack` with an explicit `stackName: "file-share"` (kebab-case, per convention) and the target account/region.
`aws-cdk-lib` is pinned to an exact version (`2.261.0`); the `aws-cdk` CLI is available as a dev dependency and via `npx cdk`.
Common commands (also exposed as npm scripts):
| Command | Purpose |
|---|---|
| `npx cdk synth` (`npm run synth`) | Synthesize the CloudFormation template to `cdk.out/` |
| `npx cdk diff` (`npm run diff`) | Diff the synthesized stack against what is deployed |
| `npx cdk deploy` (`npm run deploy`) | Deploy the stack |
- **EC2** — `t4g.small` (ARM64, Amazon Linux 2023) in the private subnet. The AMI is cached in the committed `cdk.context.json` (`cachedInContext: true`), so deploys never pick up a new AL2023 release implicitly — an AMI change forces instance replacement and must be deliberate (`cdk context --reset <ami key> && cdk synth`).
- **SFTP** — password auth for user `adam` (`ForceCommand internal-sftp`, same password as SMB); used by Hazel for automated uploads
- **EBS data volume** — `vol-04d951cccacc435b5`, 500 GiB gp3 encrypted, mounted at `/data`. **Unmanaged import**: the stack references it by ID (`Volume.fromVolumeAttributes` + `CfnVolumeAttachment`), so CloudFormation can attach it but can never create, replace, or delete it — the data survives instance replacement and even stack deletion. UserData waits for the attachment, then mounts the existing filesystem; a `blkid` guard ensures a disk that already has a filesystem is never formatted.
- **DLM** — Daily EBS snapshots at 06:00 UTC, 30-day retention (targets the `file-share-backup=true` tag, set directly on the volume)
> **History:** the data volume was originally an inline `blockDevice`, which destroyed data on instance replacement (2026-05-27 incident), then a stack-managed standalone volume, which was orphaned when an uncommitted deploy got reverted by CD (2026-06-05 incident). The unmanaged-import design ends that failure class.
Prod changes go through HCP Terraform workspace `file-share-prod` (manual apply until the move is sealed). The bootstrap apply creates the HCP roles and the instance boundary. The following apply, using `hcptf-file-share`, creates the subnet, security group, and DLM policy. `data_volume_id` stays empty until the snapshot copy exists, so those applies do not boot an instance.
The management-account CDK workflow no longer runs on push. `workflow_dispatch` remains for an explicit rollback of that stack.
The instance has no public IP. Its route table sends `10.10.0.0/16` and `10.30.0.0/16` through the VPN gateway in workspace variable `vpn_gateway_id` and everything else through a NAT gateway in the syslog public subnet. `10.10.0.0/16` is the Ronkonkoma office LAN and `10.30.0.0/16` is the Locust office LAN, the same pair the syslog VPN already routes. `10.20.0.0/16` is the management VPC and is not routed here.
Clients use the private IP. Office routing must include `10.40.20.0/24` on the existing syslog IPsec before SMB from the office will work. A check from `10.10.70.0/24` on 2026-09-28 reached the gateway for `10.40.10.254` and got no hop-1 reply for `10.40.20.1`. The nightly DLM policy targets volumes tagged `file-share-backup=true`. Tag the copied volume with that key at cutover.
The data volume is **not managed by CloudFormation** (imported by ID), so changing a size in `lib/file-share-stack.ts` has no effect. Expand it directly — no downtime:
