recotem/compose.yaml at main · codelibs/recotem · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
# compose.yaml — Recotem tutorial compose file
#
# This file is a runnable example for the getting-started tutorial. It is
# NOT for production use.
#
# Usage (3 commands from the repo root):
#
#   1. Generate a signing key + an API key
#      $ docker run --rm ghcr.io/codelibs/recotem:latest keygen --type signing --kid dev
#      # → copy the env_entry plaintext into RECOTEM_SIGNING_KEYS below
#      $ docker run --rm ghcr.io/codelibs/recotem:latest keygen --type api --kid dev
#      # → copy the env_entry hash into RECOTEM_API_KEYS below
#      # → keep the api plaintext for later (used as X-API-Key header)
#
#   2. Train (one-shot)
#      $ RECOTEM_SIGNING_KEYS="dev:..." \
#        RECOTEM_API_KEYS="dev:sha256:..." \
#        docker compose run --rm train
#
#   3. Serve + curl
#      $ RECOTEM_SIGNING_KEYS="..." RECOTEM_API_KEYS="..." \
#        docker compose up -d serve
#      $ curl -sX POST http://localhost:8080/v1/recipes/purchase_log:recommend \
#          -H "X-API-Key: <api plaintext>" \
#          -H "Content-Type: application/json" \
#          -d '{"user_id": "1", "limit": 5}'

x-recotem-image: &recotem-image
  image: ghcr.io/codelibs/recotem:latest
  # To build locally:
  # build:
  #   context: .
  #   dockerfile: Dockerfile

services:

  # ── train ──────────────────────────────────────────────────────────────────
  # One-shot training job. Run via `docker compose run --rm train`.
  # Exit codes: 0=success, 2=recipe error, 3=datasource error,
  #             4=training error, 5=artifact error.
  #
  # Note: train does NOT bind any network port — it reads data, trains,
  # and writes an artifact file. RECOTEM_ALLOWED_HOSTS, RECOTEM_API_KEYS,
  # RECOTEM_HOST/PORT, and RECOTEM_ALLOWED_ORIGINS are serve-only variables
  # and are intentionally omitted here.
  train:
    <<: *recotem-image
    command: ["train", "/recipes/recipe.yaml"]
    working_dir: /workspace
    volumes:
      - ./examples/tutorial-purchase-log:/recipes:ro
      - artifacts:/workspace/artifacts
    environment:
      RECOTEM_SIGNING_KEYS: "${RECOTEM_SIGNING_KEYS}"
      RECOTEM_LOG_FORMAT: "json"
    restart: "no"

  # ── serve ──────────────────────────────────────────────────────────────────
  # Long-running prediction server. Hot-swaps when train rewrites the artifact.
  serve:
    <<: *recotem-image
    command: ["serve", "--recipes", "/recipes"]
    working_dir: /workspace
    ports:
      - "8080:8080"
    volumes:
      - ./examples/tutorial-purchase-log:/recipes:ro
      - artifacts:/workspace/artifacts:ro
    environment:
      RECOTEM_SIGNING_KEYS: "${RECOTEM_SIGNING_KEYS}"
      RECOTEM_API_KEYS:     "${RECOTEM_API_KEYS}"
      RECOTEM_HOST: "0.0.0.0"
      RECOTEM_PORT: "8080"
      RECOTEM_WATCH_INTERVAL: "10"
      RECOTEM_LOG_FORMAT: "json"
      RECOTEM_ENV: "production"
    restart: unless-stopped
    healthcheck:
      test:
        - "CMD-SHELL"
        - "python -c \"import sys, urllib.request; sys.exit(0 if urllib.request.urlopen('http://127.0.0.1:8080/health', timeout=5).status == 200 else 1)\""
      interval: 30s
      timeout: 10s
      retries: 3
      # Initial model load can take longer than the regular interval; bump
      # this if your artifacts are large or stored on a slow object store.
      start_period: 60s

# ── volumes ────────────────────────────────────────────────────────────────────
# Shared between train (writes) and serve (reads). The recipe writes to
# ./artifacts/purchase_log.recotem (CWD-relative); the working_dir above plus
# this mount resolve that to /workspace/artifacts.
volumes:
  artifacts: