mortred_model_server/docker-compose.yml at main · MaybeShewill-CV/mortred_model_server · 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
# Mortred Model Server - container orchestration for BOTH deployment profiles.
#
# GPU profile (default, full CUDA/TensorRT stack):
#   docker compose --profile gpu up -d
# CPU profile (no NVIDIA GPU required):
#   docker compose --profile cpu up -d
# Both: MORTRED_API_TOKEN + MORTRED_GATEWAY_AUTH_TOKEN + MORTRED_METRICS_TOKEN
# must be set (`mortredctl init-trust`). weights dir mount defaults to ./weights,
# override with WEIGHTS_DIR.
#
# Two host ports: supervisor (management) 8787 + gateway (inference) 8080.
# Both bind 127.0.0.1 on the host. Autostart uses conf/packs/demo.toml
# (MORTRED_PACK), not every conf/server model. LAN/WAN TLS is Nginx on the
# host network (mortredctl init-edge, deploy/nginx). Use "8080:8080" /
# "8787:8787" only if you accept plaintext Bearer tokens on that interface —
# then MORTRED_EXPOSE=unsafe and a distinct MORTRED_METRICS_TOKEN are required.
# Do not publish model ports (9001+).
services:
  mortred:
    profiles: ["gpu"]
    build:
      context: .
      target: mortred-gpu
    image: mortred_model_server:gpu
    container_name: mortred
    environment:
      MORTRED_EXPOSE: docker
      MORTRED_API_TOKEN: "${MORTRED_API_TOKEN:?set MORTRED_API_TOKEN to protect the supervisor API}"
      MORTRED_GATEWAY_AUTH_TOKEN: "${MORTRED_GATEWAY_AUTH_TOKEN:?set MORTRED_GATEWAY_AUTH_TOKEN to protect inference}"
      MORTRED_METRICS_TOKEN: "${MORTRED_METRICS_TOKEN:?set MORTRED_METRICS_TOKEN (distinct scrape secret)}"
      MORTRED_AUTOSTART: "true"
      MORTRED_PACK: "/opt/mortred/conf/packs/demo.toml"
    ports:
      - "127.0.0.1:8787:8787"
      - "127.0.0.1:8080:8080"
    volumes:
      - "${WEIGHTS_DIR:-./weights}:/opt/mortred/weights"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    healthcheck:
      test: ["CMD", "curl", "-fs", "http://localhost:8787/api/v1/health"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 60s
    restart: unless-stopped

  mortred-cpu:
    profiles: ["cpu"]
    build:
      context: .
      target: mortred-cpu
    image: mortred_model_server:cpu
    container_name: mortred-cpu
    environment:
      MORTRED_EXPOSE: docker
      MORTRED_API_TOKEN: "${MORTRED_API_TOKEN:?set MORTRED_API_TOKEN to protect the supervisor API}"
      MORTRED_GATEWAY_AUTH_TOKEN: "${MORTRED_GATEWAY_AUTH_TOKEN:?set MORTRED_GATEWAY_AUTH_TOKEN to protect inference}"
      MORTRED_METRICS_TOKEN: "${MORTRED_METRICS_TOKEN:?set MORTRED_METRICS_TOKEN (distinct scrape secret)}"
      MORTRED_PROFILE: "cpu"
      MORTRED_AUTOSTART: "true"
      MORTRED_PACK: "/opt/mortred/conf/packs/demo.toml"
    ports:
      - "127.0.0.1:8787:8787"
      - "127.0.0.1:8080:8080"
    volumes:
      - "${WEIGHTS_DIR:-./weights}:/opt/mortred/weights"
    healthcheck:
      test: ["CMD", "curl", "-fs", "http://localhost:8787/api/v1/health"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 30s
    restart: unless-stopped

  # Linux host network only. Generate first:
  #   mortredctl init-edge --mode lan --out conf/local/edge
  # A bridge-network Nginx cannot reach Mortred's 127.0.0.1 listeners.
  mortred-edge:
    profiles: ["edge"]
    image: nginx:1.26-alpine
    network_mode: host
    volumes:
      - "${MORTRED_EDGE_DIR:-./conf/local/edge}:/data/edge:ro"
    command: ["nginx", "-p", "/data/edge", "-c", "nginx.conf", "-g", "daemon off;"]
    restart: unless-stopped