# AirStack remote developer workflow on OSMO.
#
# Submits a single GPU task ("workspace") that runs Docker-in-Docker
# (DinD) and brings up the regular AirStack three-container stack (Isaac Sim
# with WebRTC livestream, robot-desktop, GCS) on the inner Docker daemon. The
# task also runs sshd so a student can attach VS Code or Cursor over Remote-SSH
# from their laptop (Mac, Windows, or Linux — no local Docker / NVIDIA driver
# required).
#
# To submit (replace <gpu-pool> and substitute your actual pubkey):
#
#   osmo workflow submit osmo/workflows/airstack-dev.yaml \
#     --pool <gpu-pool> \
#     --set-env "SSH_PUB_KEY=$(cat ~/.ssh/id_ed25519.pub)"
#
# To stream the IDE, Isaac Sim, and Foxglove back to the laptop:
#
#   # SSH so VS Code / Cursor Remote-SSH can attach (1 terminal):
#   osmo workflow port-forward <wf-id> workspace --port 2200:22 --connect-timeout 86400
#
#   # Isaac Sim WebRTC (2 terminals — TCP + UDP):
#   osmo workflow port-forward <wf-id> workspace \
#     --port 47995-48012,49000-49007,49100 --connect-timeout 86400
#   osmo workflow port-forward <wf-id> workspace \
#     --port 49099 --udp --connect-timeout 86400
#
#   # GCS Foxglove websocket (1 terminal):
#   osmo workflow port-forward <wf-id> workspace --port 8766:8766 --connect-timeout 86400
#
# See docs/tutorials/airstack_on_osmo.md for the full walkthrough and
# osmo/README.md for the lab-admin setup (pool prerequisites, OSMO credential
# registration, workspace image build).

workflow:
  name: airstack-dev
  groups:                              # `groups:` keeps room to add a separate
                                       # dind sidecar or split out isaac-sim /
                                       # robot-desktop / gcs as their own
                                       # tasks later. For now, one lead task.
  - name: airstack
    tasks:
    - name: workspace
      lead: true
      image: airlab-docker.andrew.cmu.edu/airstack/airstack-osmo-workspace:latest
      # Required so the inner dockerd can run. hostNetwork is intentionally
      # NOT set — osmo workflow port-forward reaches the pod NS, where the
      # inner dockerd publishes ports via standard NAT.
      privileged: true
      command: ["bash"]
      args: ["/usr/local/bin/entrypoint.sh"]
      environment:
        # Behaviour switches consumed by entrypoint.sh and airstack.sh:
        AUTOLAUNCH: "true"             # boot AirStack on startup
        ISAAC_SIM_LIVESTREAM: "true"   # use the isaac-sim-livestream profile
        NUM_ROBOTS: "1"
        AIRSTACK_BRANCH: "main"        # branch entrypoint.sh clones
        AIRSTACK_REPO_URL: "https://github.com/castacks/AirStack.git"
        # SSH_PUB_KEY is supplied at submit time:
        #   --set-env "SSH_PUB_KEY=$(cat ~/.ssh/id_ed25519.pub)"
      credentials:
        # Each student registers these in their own OSMO profile once.
        # See docs/tutorials/airstack_on_osmo.md "Step 0".
        #
        # airlab-nucleus (GENERIC) — materialized into omni_pass.env by
        # entrypoint.sh so Compose env_files it into the Isaac Sim ctnr.
        airlab-nucleus:
          OMNI_USER: omni_user
          OMNI_PASS: omni_pass
          OMNI_SERVER: omni_server
        # airlab-docker-login (GENERIC) — exposed as env vars so the
        # inner dockerd can authenticate to airlab-docker.andrew.cmu.edu
        # when `airstack up` triggers an AirStack image-pull. (The
        # *outer* pod's image-pull of airstack-osmo-workspace is handled
        # automatically by a sibling REGISTRY-type credential which does
        # not need a reference here — OSMO auto-attaches it on submit.)
        airlab-docker-login:
          AIRLAB_REGISTRY_USER: username
          AIRLAB_REGISTRY_PASS: password

  resources:
    default:
      cpu: 16
      gpu: 1
      memory: 64Gi
      # AirStack image set is large: airstack-dev-9 (2026-05-14) hit the
      # 100Gi container ephemeral cap during inner Compose's first image
      # extract, before the second image even started downloading. So the
      # full set of inner images alone exceeds 100Gi extracted. Going to
      # 500Gi to comfortably hold isaac-sim + robot-desktop + gcs images,
      # plus the AirStack source clone, colcon build output, and bag
      # recordings. The airstack pool's workers have 4.2Ti of ephemeral
      # capacity each (root disks resized 2026-05-14), so this leaves
      # plenty of room for other tenants on the shared workers.
      storage: 500Gi

  timeout:
    # 8h covers a normal dev session. Bump for longer runs; cancel manually
    # before the timeout if you want to free the GPU early.
    exec_timeout: 8h