fiftyone-teams-app-deploy/docker/legacy-auth/compose.delegated-operators.gpu.yaml at e49f76f1f426b60afaf7b9c02f1ef92c1d050759 · voxel51/fiftyone-teams-app-deploy · 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
95
96
97
98
99
100
101
102
---
# GPU-enabled delegated-operator worker + telemetry sidecar. Requires a
# Linux host with the NVIDIA driver, nvidia-container-toolkit, and the
# `nvidia` runtime configured in dockerd. Docker Desktop on macOS /
# Windows does not support GPU passthrough.
#
# Gated behind the `gpu` profile. Activate with:
#
#   docker compose --profile gpu \
#     -f compose.yaml \
#     -f compose.delegated-operators.yaml \
#     -f compose.delegated-operators.gpu.yaml \
#     up -d
#
# Telemetry (redis + paired sidecars) is bundled by default in the base
# compose.yaml and compose.delegated-operators.yaml files.
#
# For Proxy Server instructions please see
#  https://github.com/voxel51/fiftyone-teams-app-deploy/tree/main/docker#environment-proxies
services:
  teams-do-gpu:
    image: voxel51/fiftyone-teams-cv-full:v2.22.0
    profiles: ["gpu"]
    shm_size: "8g"
    command: >
      /bin/sh -c "fiftyone delegated launch -t remote -m -n teams-do-gpu"
    environment:
      API_URL: ${API_URL}
      FIFTYONE_DATABASE_ADMIN: "false"
      FIFTYONE_DATABASE_NAME: ${FIFTYONE_DATABASE_NAME}
      FIFTYONE_DATABASE_URI: ${FIFTYONE_DATABASE_URI}
      FIFTYONE_ENCRYPTION_KEY: ${FIFTYONE_ENCRYPTION_KEY}
      FIFTYONE_INTERNAL_SERVICE: "true"
      FIFTYONE_MEDIA_CACHE_SIZE_BYTES: -1
      FIFTYONE_PLUGINS_DIR: /opt/plugins
      FIFTYONE_TELEMETRY_REDIS_URL: ${FIFTYONE_TELEMETRY_REDIS_URL:-redis://telemetry-redis:6379}
      NVIDIA_VISIBLE_DEVICES: ${NVIDIA_VISIBLE_DEVICES:-all}
      NVIDIA_DRIVER_CAPABILITIES: ${NVIDIA_DRIVER_CAPABILITIES:-compute,utility}
      TELEMETRY_SOCKET: /tmp/telemetry/agent.sock
    deploy:
      replicas: 1
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: ${NVIDIA_GPU_COUNT:-1}
              capabilities: ["gpu"]
    volumes:
      - plugins-vol:/opt/plugins:ro
      - telemetry-socket-gpu:/tmp/telemetry
    restart: always

  teams-do-gpu-telemetry:
    image: voxel51/telemetry-sidecar:v2.22.0
    pid: "service:teams-do-gpu"
    profiles: ["gpu"]
    user: "1000:1000"
    cap_add:
      - SYS_PTRACE  # required for profiling metrics (sampled stacks via py-spy)
    deploy:
      resources:
        limits:
          cpus: "0.10"
          memory: 512M
        reservations:
          cpus: "0.10"
          memory: 512M
          devices:
            - driver: nvidia
              count: ${NVIDIA_GPU_COUNT:-1}
              capabilities: ["gpu"]
    environment:
      NVIDIA_VISIBLE_DEVICES: ${NVIDIA_VISIBLE_DEVICES:-all}
      NVIDIA_DRIVER_CAPABILITIES: ${NVIDIA_DRIVER_CAPABILITIES:-compute,utility}
      POD_NAME: teams-do-gpu
      POD_NAMESPACE: ${TELEMETRY_NAMESPACE:-docker}
      SERVICE_TYPE: delegated-operator
      TARGET_NAME: ${TEAMS_DO_TARGET_NAME:-fiftyone delegated}
      EXECUTOR_SIDECAR: "true"
      TELEMETRY_SOCKET: /tmp/telemetry/agent.sock
      FIFTYONE_TELEMETRY_REDIS_URL: ${FIFTYONE_TELEMETRY_REDIS_URL:-redis://telemetry-redis:6379}
      FIFTYONE_DATABASE_URI: ${FIFTYONE_DATABASE_URI}
      FIFTYONE_DATABASE_NAME: ${FIFTYONE_DATABASE_NAME}
    volumes:
      - telemetry-socket-gpu:/tmp/telemetry
    depends_on:
      teams-do-gpu:
        condition: service_started
        restart: true
      telemetry-redis:
        condition: service_started
    restart: always

# Tmpfs-backed named volume with uid/gid set at mount time so the
# non-root sidecar can bind agent.sock.
volumes:
  plugins-vol:
  telemetry-socket-gpu:
    driver_opts:
      type: tmpfs
      device: tmpfs
      o: "uid=1000,gid=1000,mode=0755"