thermograph/infra/deploy/forgejo/docker-stack.yml
Emi Griffith 4e97d8e5dc
All checks were successful
PR build (required check) / changes (pull_request) Successful in 6s
secrets-guard / encrypted (pull_request) Successful in 5s
PR build (required check) / build-backend (pull_request) Has been skipped
shell-lint / shellcheck (pull_request) Successful in 7s
PR build (required check) / build-frontend (pull_request) Has been skipped
PR build (required check) / validate-observability (pull_request) Successful in 18s
PR build (required check) / gate (pull_request) Successful in 2s
infra: split the estate into vps1/vps2, moving beta next to prod and dev onto vps1
Environments stop being machines. Until now each box WAS an environment --
"beta" named both a deploy target and a host, "the desktop" named both the
operator's computer and the dev server -- so every path could assume one
environment per host and hardcode /opt/thermograph, /etc/thermograph.env and
ports 8137/8080. That assumption ends here:

  vps1  75.119.132.91  Forgejo, Grafana/Loki, the portfolio site, and DEV
                       (own Postgres, mesh-only on 10.10.0.2:8137)
  vps2  169.58.46.181  PROD and BETA as two Swarm stacks sharing one
                       TimescaleDB instance, plus Centralis, Postfix, backups
  desktop              AI model hosting + flex Swarm capacity, no environment

Nothing live has moved; the ordered cutover is in
infra/deploy/RUNBOOK-vps1-vps2-cutover.md.

deploy/env-topology.sh is the single source of truth: env -> host, checkout,
branch, deploy mode, stack name, env file, LB ports, DB role/database, service
prefix. THERMOGRAPH_ENV is the input, and on vps2 it is the only thing
distinguishing a beta deploy from a prod one -- deploy.sh refuses to run when it
disagrees with the checkout it was invoked from. The host-wide secrets-env and
deploy-mode markers survive only as a fallback for a single-environment box.

Beta gets its own stack file rather than an overlay (a merge cannot REMOVE the
db service, and beta having no database of its own is the point). Its services
are prefixed beta-*: Swarm registers a service's short name as a DNS alias on
every network it joins, so two stacks both calling a service `web` on the shared
data network would let prod's frontend resolve a beta task. Prod's stack, env
and LB config are untouched.

One Postgres, two databases with two roles: deploy/db/provision-env-db.sh
creates thermograph_beta (NOSUPERUSER, owns only its own database, CONNECT
revoked from PUBLIC) plus a <role>_ro for ad-hoc queries, and refuses to run for
the environment that owns the instance -- doing so would demote prod's bootstrap
superuser.

Fixes that co-residency would otherwise have broken silently:

- CI secrets are keyed by host (VPS1_SSH_*, VPS2_SSH_*). SSH_* meant "beta" and
  also "the Forgejo box" because those shared a machine; that conflation is what
  once had the prod backup dumping beta.
- The nightly backup dumps BOTH application databases to separate off-box
  prefixes, and fails loudly on a missing one instead of skipping it.
- Alloy stopped deriving the `host` label from the node -- on vps2 that would
  have filed every beta line as prod, feeding prod's alert rules with beta's
  traffic. It is now derived per source, with a new `node` label for the machine,
  and beta's log volume is mounted via a vps2-only overlay.
- ops/dbq.sh, ops/iceberg.sh and the secrets seed scripts derived their SSH
  target and env-file path from the topology instead of hardcoding beta to
  75.119.132.91 -- which is vps1's address now.
- Dev's overlay binds DEV_BIND_ADDR (loopback by default, the mesh address on
  vps1) instead of 0.0.0.0: correct for a home LAN box, a public exposure of
  unreviewed branches on a VPS.

Terraform's hosts variable is keyed by machine with a nested environments map,
and main.tf flattens (host, environment) pairs so two environments on one box
cannot share a checkout. Caddy's single reference config is split per host.

Docs, onboarding and the runbooks are updated throughout, including the security
rationales that co-residency makes false: separate SSH credentials no longer put
a host boundary between beta and prod, and the boundary that remains is the
database and the filesystem.
2026-07-25 15:01:29 -07:00

157 lines
7.3 KiB
YAML

# Forgejo (self-hosted Git + CI/CD) as a Docker Swarm stack — the only workload
# this Swarm cluster runs (see deploy/swarm/README.md). Deliberately separate
# from the Terraform-managed docker-compose.yml that runs the Thermograph app
# itself: this stack's only job is Forgejo and its container registry.
#
# The Actions RUNNER is deliberately NOT a service in this stack. Per
# thermograph-docs/runbooks/implementation-handoff.md (Track B step 5), the canonical
# design registers the runner on the desktop node as a plain systemd service
# (deploy/forgejo/register-lan-runner.sh) — the same place the pre-Forgejo
# GitHub self-hosted runner already lived, not a Swarm-scheduled container.
#
# No Traefik here. Forgejo is pinned to vps1 (node.labels.role == forge)
# because that's what was chosen, and vps1 is ALSO the emigriffith.dev portfolio
# host — Caddy already owns its ports 80/443 (see /etc/caddy/Caddyfile on that
# box). A second reverse proxy binding those same ports would either fail to
# start or fight Caddy. Instead: forgejo's web port publishes to
# 127.0.0.1:3080 only (host-local, mode: host), and vps1's existing Caddy gets
# a new site block reverse-proxying git.thermograph.org -> 127.0.0.1:3080,
# same pattern as its other blocks. TLS is Caddy's existing automatic-HTTPS
# (HTTP-01), not a second ACME flow.
#
# Deploy from the manager node (vps2, which runs prod), after all three nodes
# (vps1, vps2, desktop) have joined the swarm and vps1 is labeled role=forge:
#
# docker stack deploy -c deploy/forgejo/docker-stack.yml forgejo
#
# Requires this Swarm secret to exist first (see deploy/forgejo/README.md):
# forgejo_db_password
#
# Registry exposure (hazard #15 in thermograph-docs/runbooks/hop1-forgejo-registry-cutover.md):
# resolved as the runbook's first listed option — serve Git/UI publicly but
# firewall the /v2/ registry API paths so only WireGuard-mesh clients can
# reach them. That's implemented in the Caddy site block (see
# deploy/forgejo/README.md), not here — nothing in this stack file is
# registry-specific, the restriction lives entirely in Caddy's config on vps1.
services:
db:
image: postgres:16-alpine
environment:
POSTGRES_USER: forgejo
POSTGRES_DB: forgejo
POSTGRES_PASSWORD_FILE: /run/secrets/forgejo_db_password
secrets: [forgejo_db_password]
volumes:
- forgejo_db:/var/lib/postgresql/data
networks: [forgejo_net]
deploy:
placement:
constraints: [node.labels.role == forge]
resources:
limits:
cpus: "${FORGEJO_DB_CPUS:-1}"
memory: ${FORGEJO_DB_MEMORY:-1g}
restart_policy:
condition: on-failure
forgejo:
image: codeberg.org/forgejo/forgejo:9-rootless
depends_on: [db]
environment:
FORGEJO__database__DB_TYPE: postgres
FORGEJO__database__HOST: db:5432
FORGEJO__database__NAME: forgejo
FORGEJO__database__USER: forgejo
FORGEJO__server__DOMAIN: "${FORGEJO_DOMAIN:-git.thermograph.org}"
FORGEJO__server__ROOT_URL: "https://${FORGEJO_DOMAIN:-git.thermograph.org}/"
FORGEJO__server__SSH_PORT: "2222"
# Without this, Forgejo starts no SSH server at all despite SSH_PORT
# being set — git@ clones get "connection refused", not a slow failure.
FORGEJO__server__START_SSH_SERVER: "true"
# Config is fully supplied via env, so lock the install wizard rather
# than leave it open on the public internet waiting for someone to
# complete it first (DB config alone does NOT imply this).
FORGEJO__security__INSTALL_LOCK: "true"
# Actions on: repo/org/user-level runners register against this instance.
FORGEJO__actions__ENABLED: "true"
# --- Access model: OAuth-only login, multi-user via an approval gate ---
# Login is Google SSO only. Forgejo 9.0.3 has no setting to hide the
# password sign-in form, so "OAuth-only" is enforced operationally: every
# account is given an unusable (random) password and no NEW local account
# can be created (ALLOW_ONLY_EXTERNAL_REGISTRATION). ENABLE_PASSWORD_SIGNIN_FORM
# is set anyway — ignored on 9.0.3, effective if the instance is upgraded.
# Auto-registration stays ON so a Google login auto-links (by verified email)
# to a PRE-CREATED account (`forgejo admin user create --username X --email
# X@gmail.com --random-password`). A single required-claim-value can't
# allowlist more than one email, so instead of pinning one address the
# required-claim-value is CLEARED (a DB change on the auth source, persisted
# in the forgejo_data volume — not re-applied from here) and every NEW signup
# is gated: a stranger who signs in with Google lands INACTIVE
# (REGISTER_MANUAL_CONFIRM) and RESTRICTED (DEFAULT_USER_IS_RESTRICTED)
# pending admin approval, so an un-provisioned Google user gets zero access.
FORGEJO__oauth2_client__ACCOUNT_LINKING: "auto"
FORGEJO__oauth2_client__ENABLE_AUTO_REGISTRATION: "true"
FORGEJO__service__ALLOW_ONLY_EXTERNAL_REGISTRATION: "true"
FORGEJO__service__ENABLE_PASSWORD_SIGNIN_FORM: "false"
FORGEJO__service__REGISTER_MANUAL_CONFIRM: "true"
FORGEJO__service__DEFAULT_USER_IS_RESTRICTED: "true"
# Outbound mail via vps2's Postfix null client (prod's box) over the
# WireGuard mesh (10.10.0.1:25 — mynetworks permits vps1, 10.10.0.2,
# where this Forgejo runs); self-signed cert on :25, so trust it.
# Send-only; used for admin/approval and notification mail.
FORGEJO__mailer__ENABLED: "true"
FORGEJO__mailer__PROTOCOL: "smtp"
FORGEJO__mailer__SMTP_ADDR: "10.10.0.1"
FORGEJO__mailer__SMTP_PORT: "25"
FORGEJO__mailer__FORCE_TRUST_SERVER_CERT: "true"
FORGEJO__mailer__FROM: "Thermograph Git <no-reply@thermograph.org>"
# Gitea/Forgejo's app.ini env-mapping honors a __FILE suffix to read a
# value from a file instead of the literal env var — same convention as
# the official Postgres image's POSTGRES_PASSWORD_FILE above.
FORGEJO__database__PASSWD__FILE: /run/secrets/forgejo_db_password
secrets:
- source: forgejo_db_password
target: forgejo_db_password
volumes:
- forgejo_data:/var/lib/gitea
networks: [forgejo_net]
ports:
# SSH for git@ clones — published on whichever node the task lands on
# (pinned to vps1 by the placement constraint below, so effectively
# always vps1's public IP, port 2222).
- target: 2222
published: 2222
protocol: tcp
mode: host
# Web UI/API. Swarm's port schema has no host_ip scoping, so this binds
# 0.0.0.0:3080 on vps1 — NOT actually localhost-only by itself. A
# DOCKER-USER iptables rule (see deploy/forgejo/README.md) is what
# actually restricts it, since Docker's own iptables rules bypass ufw
# for published ports. vps1's Caddy reverse-proxies to 127.0.0.1:3080.
- target: 3000
published: 3080
protocol: tcp
mode: host
deploy:
placement:
constraints: [node.labels.role == forge]
resources:
limits:
cpus: "${FORGEJO_CPUS:-2}"
memory: ${FORGEJO_MEMORY:-2g}
restart_policy:
condition: on-failure
networks:
forgejo_net:
driver: overlay
attachable: false
volumes:
forgejo_db:
forgejo_data:
secrets:
forgejo_db_password:
external: true