Skip to content

Deploy / Landing Zone

The most common friction in a migration is “where do I run the target?”. flareover ships a one-command landing zone so you don’t have to assemble the stack by hand.

The deploy/ directory is a docker compose project that stands up the full target stack and consumes the prepare --out artifacts directly:

  • PowerDNS: authoritative DNS (+ DNSSEC)
  • Caddy: reverse proxy / TLS / HTTP/3, built with caddy-waf + souin
  • CertMate: certificates via DNS-01 (wildcard; Let’s Encrypt or Actalis)
  • Garage: S3-compatible object storage (replaces MinIO, which is archived upstream)
  • secure-proxy-manager: the optional egress shield
Terminal window
cd deploy
cp .env.example .env # set PDNS_API_KEY, CERTMATE_TOKEN, GARAGE_RPC_SECRET, GARAGE_ADMIN_TOKEN
docker compose up -d --build # --build compiles the custom Caddy once
./garage-init.sh # once: Garage cluster layout + the `flareover` key

GARAGE_RPC_SECRET must be 32 bytes as hex (openssl rand -hex 32). A fresh Garage node serves nothing until it has a cluster layout: until ./garage-init.sh [capacity] has run (default capacity 100G), its /health answers 503. The script applies the layout, creates the key flareover (override the name with GARAGE_KEY) with permission to create buckets, and prints the exports the generated garage/provision.sh needs. It is safe to re-run. The key’s secret is read with docker compose exec -T garage /garage key info flareover --show-secret.

Then point provision at it, loading the secrets from the environment (never the command line):

Terminal window
set -a; . ./.env; set +a
flareover provision --snapshot snap.json --decisions decisions.lock \
--pdns-url http://localhost:8081 \
--certmate-url http://localhost:8000
# object storage, if any (after exporting what garage-init.sh printed):
flareover storage buckets.json --minio-endpoint http://127.0.0.1:3900 --out ./out
bash ./out/garage/provision.sh

Garage’s config is deploy/garage.toml: one node, replication_factor = 1, S3 signing region garage. Buckets confirmed public are served by Garage’s web endpoint as <bucket>.<web root domain>; the shipped web root domain is .web.garage.localhost, so change it to your own domain and route that domain to port 3902 from Caddy. See Object Storage.

  • Caddy (80/443) and PowerDNS (53, authoritative: it must be publicly reachable to serve the zone) are internet-facing.
  • Every admin/API surface (PowerDNS 8081, CertMate 8000, Garage 3900 S3 API / 3902 web endpoint / 3903 admin, secure-proxy-manager 3128) binds to 127.0.0.1.
  • The PowerDNS control API additionally refuses any source outside the compose network: the restriction is in the setting, not only in the port binding, so exposing 8081 later does not silently remove it.
  • Every service runs with no-new-privileges and cap_drop: ALL, with capabilities added back only where the image’s entrypoint needs them.

Put your own firewall in front regardless. See Security.

The eight volumes are not equal. The five in this table hold state that cannot be regenerated:

Volume Holds If you lose it
pdns-data the authoritative zone and the DNSSEC signing keys the zone stops resolving, and the DS record at your registrar points at keys that no longer exist
certmate-data + certs issued certificates and issuance state re-issuance, and rate limits at the CA
garage-meta + garage-data the migrated objects and the index that finds them (back up both, together) possibly the only remaining copy, once the source bucket is gone

caddy-data, caddy-config and spm-data are disposable — re-running prepare plus a restart rebuilds them. The commands, and what a good restore looks like (the zone back and its original DNSSEC key), are in deploy/README.md.

Image versions are pinned in .env rather than floating on :latest: these services hold the authoritative zone, the DNSSEC keys and the migrated objects, so an upgrade should be a reviewed change and a rollback should be possible.

A CI job brings this stack up on every change to deploy/ and asserts that the PowerDNS API answers, that a zone can be created and signed, and that it survives a backup/restore round trip with the same signing key. It also runs garage-init.sh, waits for Garage’s /health to answer 200, and does an S3 put/get round trip with the key the script created.

prepare --edge-provider <key> emits a cloud-init that installs and configures a Caddy + WireGuard edge on a specific provider (see Sovereignty Tiers). For Scaleway and OVHcloud it also emits a script that creates and boots the instance from that cloud-init.

Terminal window
flareover prepare snap.json --decisions decisions.lock \
--edge-provider hetzner --mesh-edge 203.0.113.10:51820 --out ./out
# → ./out/edge/cloud-init.yaml (installs Caddy + WireGuard, writes the config)

The cloud-init carries the mesh WireGuard private key, so treat ./out/ as a secret.

terraform/hetzner/ is a module that boots the edge on Hetzner Cloud straight from the generated cloud-init (hcloud_server + firewall + SSH key; outputs the edge IP). Point cloud_init_path at out/edge/cloud-init.yaml, terraform apply, and feed the printed IP back into prepare --edge-ip / present.

Keep your origin exactly where it is (WireGuard mesh)

Section titled “Keep your origin exactly where it is (WireGuard mesh)”

The lowest-risk shape leaves your origin unchanged and just re-tunnels it: flareover stands up your own edge node(s) and a WireGuard tunnel, and the origin only swaps its managed tunnel daemon for wg-quick. Add --mesh-edge (repeat it for an HA edge front). The origin keeps zero public inbound. Full walkthrough: Keep your origin.

The deploy/ compose file is the fast path; the Hardened Proxmox landing zone covers the hardened bare-metal / Proxmox blueprint: isolated origin bridge, edge on the routable bridge, origin reachable only through the edge/tunnel.

Privacy & legal notice