Files
archie/docker-compose.yml
T
sam c54d328f0d feat(store): run services as a non-superuser role and name every connection (#1857)
Every service connected to PostgreSQL as the database owner, which is a
superuser, so anything that compromised an app process owned the cluster, and
every connection reached the server with an empty application_name, so
pg_stat_activity could not say which service a session belonged to.

Split the DB credential into two login roles and name every pool:

- archie_migrator owns public and its objects and runs goose. Migrations run
  on a short-lived pool opened from the new migration_database_url, before the
  serving pool opens. It is not a superuser.
- archie_runtime serves requests with SELECT, INSERT, UPDATE, DELETE, sequence
  USAGE/SELECT and function EXECUTE only. It has no CREATE privilege, so it
  cannot change the schema.
- Every pool sets application_name: archie-state-store, archie-gateway, and a
  -migrate / -recovery / -schema / -provision suffix on the transient
  administrative connections. Existing one-URL installs keep working: an
  absent migration_database_url falls back to database_url and the migration
  runs on the serving pool exactly as before.

Provisioning is a new idempotent `archie-state-store provision` subcommand: it
needs an admin URL the services never hold (database owner or superuser),
creates or reconciles both roles, hands every public-schema object to the
migration role, revokes PUBLIC's CREATE on public, sets row-level grants, and
sets the migration role's default privileges so tables future migrations
create are usable without a hand-run grant. It stores SCRAM verifiers, refuses
superuser roles, and verifies the boundary before committing.

install.sh now provisions the bundled path and separates the credentials: the
superuser password the Compose database is created with lives only in a
compose-only postgres.env, while the env file the units load holds the
non-superuser PGPASSWORD. A bundled install that predates the split is
upgraded in place: the roles are created, the superuser password is rotated
out of the env file, and config.toml is moved to the runtime and migration
URLs. External databases keep their operator-managed credential; the
deployments README documents the one-shot provision command.

setup writes both bundled URLs, dev:postgres provisions the dev database, the
e2e stack provisions its throwaway server, and the deployment templates point
at the two roles.
2026-11-11 03:12:54 +11:00

125 lines
5.3 KiB
YAML

# PostgreSQL for the State Store, optional external NATS, and the agent images
# the daemon spawns per task.
#
# archied itself runs on the host under systemd, not here -- its chat tools
# operate on host files, which a container cannot reach. See
# deployments/systemd-user-service.md.
#
# Embedded NATS is the normal deployment and needs no Compose service.
# State Store DB: docker compose up -d postgres
# External broker: docker compose up -d nats
# Build worker: task image:agent (stamps this checkout's version)
# Profile image: docker compose build agent-net
# Stop external: docker compose down
# Logs: journalctl --user -u archied -f (the daemon)
# docker compose logs -f nats (the bus)
services:
# ── PostgreSQL (State Store) ───────────────────────────────────
# The standalone archie-state-store process's database. PostgreSQL 18 is
# the required minimum and target; `postgres.Open` refuses anything older.
# `task dev` brings this up before the State Store via dev:postgres. The
# This superuser is the dev database's admin: `task dev` runs archie-state-store
# provision after first start to create the least-privilege roles the
# mirroring deployments/dev.toml URLs name -- change them together.
postgres:
image: postgres:18
environment:
POSTGRES_USER: archie
POSTGRES_PASSWORD: archie
POSTGRES_DB: archie
# postgres:18 moved its default PGDATA under a versioned path; pin the
# classic location so the named volume keeps working across image updates.
PGDATA: /var/lib/postgresql/data/pgdata
ports:
- "127.0.0.1:5432:5432"
volumes:
- pg_data:/var/lib/postgresql/data
restart: unless-stopped
healthcheck:
test: ["CMD-SHELL", "pg_isready -U archie -d archie"]
interval: 10s
timeout: 5s
retries: 5
# ── NATS JetStream ────────────────────────────────────────────────
nats:
image: nats:2.14-alpine3.22
# The client port is published on every interface (agent containers reach
# it over the bridge), so it refuses to start without a token. Set
# NATS_TOKEN here and name the same value in [nats] token_env.
entrypoint:
- sh
- -c
- '[ -n "$$NATS_TOKEN" ] || { echo "NATS_TOKEN is required" >&2; exit 1; }; exec nats-server -js -m 8222 --auth "$$NATS_TOKEN"'
environment:
NATS_TOKEN: ${NATS_TOKEN:-}
ports:
- "4222:4222" # client
- "127.0.0.1:8222:8222" # monitoring (http://localhost:8222)
volumes:
- nats_data:/data
restart: unless-stopped
healthcheck:
test: ["CMD", "wget", "-q", "-O", "-", "http://localhost:8222/healthz"]
interval: 10s
timeout: 5s
retries: 5
# ── agent image (built once, spawned per task by the daemon) ───────
# Not a long-running service: the daemon starts agent containers itself
# via the Docker socket. This stanza exists only so the image can be
# built locally.
#
# The build profile keeps `up -d` from starting it. Without it compose
# creates a container that exits 0 immediately and still reports
# "Started", which reads like a crash loop. Naming the service explicitly
# still works without activating the profile:
#
# docker compose build agent
# docker compose pull agent -- plain `pull` skips profiled services
#
# Routine freshness is handled by the daemon, which pulls this image per
# task: see containers.pull_policy in the config. Watchtower cannot do it
# -- it only restarts long-running containers, and this is not one.
agent:
profiles: ["build"]
build:
context: .
dockerfile: Dockerfile
# Stamped into the agent so the version it reports is the one the rest of
# the install reports. `task image:agent` passes this checkout's version;
# a bare `docker build` passes nothing and the agent answers dev.
args:
RUNTIME_VERSION: ${RUNTIME_VERSION:-dev}
# A tag, not the digest: Compose refuses to build an image it is told to
# name by digest. The digest a task container actually runs is pinned in
# the daemon's own [containers] image, so nothing here decides it. See
# deployments/systemd-user-service.md#6-pinning-the-archie-agent-image.
image: git.catlow.cloud/sam/archie-agent:latest
# ── example agent profile image ─────────────────────────────────────
# The agent image plus network lookup tools, for a workflow that names
# `profile: network-investigator` (archipelago: workflows/firewall).
# Build-only for the same reason as agent. It is local, so leave
# containers.pull_policy at "missing": "always" would try to pull it.
agent-net:
profiles: ["build"]
build:
context: examples/profiles
dockerfile: Dockerfile.agent-net
image: archie-agent-net:local
volumes:
nats_data:
pg_data:
# The subnet is pinned for the optional external-broker profile because its
# configured URL must be reachable from both the native daemon and workers.
# Embedded mode discovers its selected bridge gateway at startup instead.
networks:
default:
ipam:
config:
- subnet: 172.19.0.0/16