Files
esh-pfi-infrastructure/stacks/arbo/compose.yaml
T
vh e0d1c44137 chore(fleet): repoint stale irv-ml1 refs (10.100.79.3 -> irv-ml1.nh3.internal)
The 2026-09-06 headscale cutover retired irv-ml1's wg0 tunnel IP 10.100.79.3
(now 10.6.110.50). Repointed all LIVE canonical refs to the DNS NAME so the next
move can't re-break them: homepage.href/siteMonitor labels across 25 stack
composes, load-bearing env defaults (asset-engine INFERENCE_HOST, open-webui
AUDIO_TTS_OPENAI_API_BASE_URL, skaldsong SKALDSONG_TTS_BASE_URL, zonos-gateway
ZONOS_URL, dia), homepage services.yaml manual cards (Voice Design Studio,
IRV-ML1), and servers/irv-ml1/ssh-target. Updated the stale 'WG tunnel' comment
to the mesh reality.

Left as-is: README curl-examples and .env.example comments (docs), and historical
mentions in CLAUDE.md/persistent-memory. NOTE: applying the label repoints to the
RUNNING irv-ml1 containers needs a recreate per service (labels read at creation);
deployed .env values are separate from these canonical defaults.
2026-09-07 15:08:56 -07:00

104 lines
5.1 KiB
YAML

# Arbo (catalog) engine — FastAPI sidecar that drives ComfyUI from a
# versioned workflow catalog. Serves the SPA + /workflows + /run.
#
# Co-located on irv-ml1 BESIDE the comfyui stack (ADR-0001 / D3). This
# collapses the engine<->ComfyUI link to the shared traefik-net + the
# local filesystem and ELIMINATES the old SSH/WireGuard dependency:
# - generation : COMFYUI_URL=http://comfyui:8188 (container DNS on
# traefik-net — the "localhost-equivalent"; no scp, no WG hop)
# - file ops : ARBO_COMFY_INPUT_DIR / ARBO_COMFY_OUTPUT_DIR bind the
# SAME host dirs comfyui mounts (/worktank/comfyui/basedir/{input,
# output}), so to-input / upload-input / artifact-wipe are local fs
# ops. ARBO_COMFY_SSH_HOST is left UNSET on purpose → the engine's
# local-cp/shutil path (comfy-dev's D3 follow-on) takes over.
#
# State that MUST survive restarts (named volumes, local disk — NOT NFS,
# per the DB-off-NFS rule; SQLite-on-NFS locking is a known foot-gun):
# - arbo_db -> the gallery/history SQLite (ARBO_DB)
# - arbo_heroes -> runtime hero images (ARBO_HEROES_DIR)
# The catalog is a comfy-dev git checkout bind-mounted read-only (D2):
# a workflow update = `git pull` on the host + restart, no image rebuild.
#
# IMAGE + CATALOG TARGET are owned by comfy-dev (they build the image +
# add /healthz). Two items marked CONFIRM below need their image layout.
# All tunables live in .env — edit that, not this file.
services:
engine:
image: ${ARBO_IMAGE} # CONFIRM: gitea.phasefinal.com/vh/arbo:<tag>, comfy-dev builds
container_name: arbo
restart: unless-stopped
# Run as the host owner of /worktank (1000:1000 = lkraven) so writes
# into comfyui's basedir/input + the named volumes land with the
# right ownership (matches the comfyui stack's WANTED_UID posture).
user: "${ARBO_UID:-1000}:${ARBO_GID:-1000}"
ports:
- "${ARBO_BIND:-0.0.0.0}:${ARBO_PORT:-8200}:8200"
env_file: .env
environment:
- ARBO_HOST=0.0.0.0
- ARBO_PORT=8200
- ARBO_DB=/data/gallery.sqlite
- ARBO_HEROES_DIR=/heroes
# engine<->ComfyUI over the shared network (no SSH):
- COMFYUI_URL=http://comfyui:8188
- ARBO_COMFY_INPUT_DIR=/comfy/input
- ARBO_COMFY_OUTPUT_DIR=/comfy/output
# ARBO_COMFY_SSH_HOST intentionally UNSET -> local-cp path (D3 follow-on)
# LiteLLM gateway for the granite hero/metadata calls:
- GRANITE_ENDPOINT=http://10.250.50.70:4000/v1
- GRANITE_KEY=${GRANITE_KEY} # arbo-prompt-enhance scoped vkey (.env)
# ENGINE_TOKEN intentionally NOT injected — bearer auth is OFF on prod by
# operator decision (2026-06-13); WireGuard is the access boundary. The
# app's protected-gate no-ops only when the var is ABSENT — an empty string
# still gates — so it must not be set here OR in .env. Re-enable: restore
# this line + set ENGINE_TOKEN in .env, then `compose up -d`. Reverses ADR-0001.
#- ENGINE_TOKEN=${ENGINE_TOKEN}
volumes:
- arbo_db:/data
- arbo_heroes:/heroes
# comfyui's input/output, shared on the host (rw: upload-input writes
# input, artifact-wipe deletes output):
- /worktank/comfyui/basedir/input:/comfy/input
- /worktank/comfyui/basedir/output:/comfy/output
# comfy-dev checkout on the host (D2 git-pull target), read-only. catalog +
# graphs + frontend all ride the mount, so a catalog OR SPA change reaches
# prod via `git pull + restart` — no image rebuild (ADR-0001 D2, extended to
# the frontend per the v0.11.2 prod fix). Baked image copies are the fallback.
- ${ARBO_CATALOG_DIR:-/worktank/arbo/repo}/catalog:/app/catalog:ro
- ${ARBO_CATALOG_DIR:-/worktank/arbo/repo}/graphs:/app/graphs:ro
- ${ARBO_CATALOG_DIR:-/worktank/arbo/repo}/frontend:/app/frontend:ro
# pyproject.toml rides the same checkout mount so /healthz catalog_version
# reports the MOUNTED repo version (read from /app/pyproject.toml), not the
# baked importlib.metadata one — otherwise a catalog/frontend-only webhook
# deploy (no image rebuild) leaves the reported version stale (comfy-dev
# v0.12.5 observability fix, 2026-06-16). Activated on the next engine
# recreate; the code falls back cleanly if the file is absent.
- ${ARBO_CATALOG_DIR:-/worktank/arbo/repo}/pyproject.toml:/app/pyproject.toml:ro
# python-based (the slim image ships no curl/wget; python is always present):
healthcheck:
test: ["CMD-SHELL", "python -c \"import urllib.request,sys; sys.exit(0 if urllib.request.urlopen('http://localhost:8200/healthz').status==200 else 1)\""]
interval: 30s
timeout: 10s
retries: 3
start_period: 30s
networks:
- tnet
labels:
- homepage.group=AI - Studios
- homepage.name=Arbo
- homepage.icon=mdi-image-multiple-outline
- homepage.description=Catalog-driven ComfyUI engine (irv-ml1)
- homepage.href=http://irv-ml1.nh3.internal:${ARBO_PORT:-8200}
volumes:
arbo_db:
name: arbo_db
arbo_heroes:
name: arbo_heroes
networks:
tnet:
name: traefik-net
external: true