Workload 0x93f99d05ef2a9abae0fd84653471bddfd32a460c98be42326cdc665f9d76b156

TDX measurements

Measured workload inputs

These exact files are inputs to the TDX measurement calculation for this workload.

• Phala app configuration — cdi-phala-app-compose.json

  {"allowed_envs":["MOAI_GCS_SERVICE_ACCOUNT_JSON_B64","DSTACK_DOCKER_USERNAME","DSTACK_DOCKER_PASSWORD","HF_TOKEN","PAYMENT_ADDRESS"],"docker_compose_file":"name: ritual-prod-shared-proxy\n\n# Production shared inference appliance for one 8xH200 Phala CVM.\n# Chain-agnostic by design: chains attach later through the proxy /register API.\n# Required encrypted deployment environment:\n#   HF_TOKEN\n#   MOAI_GCS_SERVICE_ACCOUNT_JSON_B64\n\nx-logging: &default-logging\n  driver: json-file\n  options:\n    max-size: \"100m\"\n    max-file: \"5\"\n\nx-restart: &default-restart\n  restart: unless-stopped\n  logging: *default-logging\n\nx-hf-volume: &hf-volume\n  hf-model-cache:/root/.cache/huggingface\n\nx-sglang-env: &sglang-env\n  HF_HOME: /root/.cache/huggingface\n  HF_HUB_OFFLINE: \"1\"\n\nservices:\n  # Fetch immutable model revisions into the app-scoped persistent cache.\n  model-bootstrap:\n    image: lmsysorg/sglang:v0.5.17@sha256:3be8803490a8b899a44f7ab2e22d8f6a1fb877cab52faeb400769a1555317db4\n    restart: \"no\"\n    volumes:\n      - *hf-volume\n    secrets:\n      - hf-token\n    environment:\n      HF_HOME: /root/.cache/huggingface\n    entrypoint: [\"/bin/bash\", \"-ceu\"]\n    command:\n      - |\n        python3 - <<'PY'\n        from huggingface_hub import snapshot_download\n        from pathlib import Path\n\n        token = Path(\"/run/secrets/hf-token\").read_text().strip()\n        if not token:\n            raise RuntimeError(\"HF token secret is empty\")\n\n        snapshots = (\n            (\"deepseek-ai/DeepSeek-V4-Flash-0731\", \"7872f01b1d1fe23eabc4c98b48bffcef5a386062\"),\n            (\"black-forest-labs/FLUX.2-klein-4B\", \"e7b7dc27f91deacad38e78976d1f2b499d76a294\"),\n            (\"Wan-AI/Wan2.2-T2V-A14B-Diffusers\", \"5be7df9619b54f4e2667b2755bc6a756675b5cd7\"),\n        )\n        for repo_id, revision in snapshots:\n            snapshot_download(repo_id=repo_id, revision=revision, token=token)\n        PY\n\n  dynamo-etcd:\n    <<: *default-restart\n    image: quay.io/coreos/etcd:v3.5.16@sha256:d967d98a12dc220a1a290794711dba7eba04b8ce465e12b02383d1bfbb33e159\n    network_mode: host\n    command:\n      - etcd\n      - --advertise-client-urls=http://127.0.0.1:2379\n      - --listen-client-urls=http://127.0.0.1:2379\n      - --data-dir=/etcd-data\n    volumes:\n      - dynamo-etcd:/etcd-data\n    healthcheck:\n      test: [\"CMD\", \"etcdctl\", \"--endpoints=http://127.0.0.1:2379\", \"endpoint\", \"health\"]\n      interval: 10s\n      timeout: 5s\n      retries: 12\n\n  dynamo-nats:\n    <<: *default-restart\n    image: nats:2.10-alpine@sha256:b83efabe3e7def1e0a4a31ec6e078999bb17c80363f881df35edc70fcb6bb927\n    network_mode: host\n    configs:\n      - source: dynamo-nats-config\n        target: /etc/nats/nats-server.conf\n    command: [\"--config\", \"/etc/nats/nats-server.conf\"]\n    healthcheck:\n      test: [\"CMD-SHELL\", \"wget -q -O /dev/null 'http://127.0.0.1:8222/healthz?js-enabled-only=true'\"]\n      interval: 10s\n      timeout: 5s\n      retries: 12\n      start_period: 10s\n\n  dynamo-frontend:\n    <<: *default-restart\n    image: ritualnetwork/dynamo-sglang:v0.5.17-dsv4@sha256:9cc856356107915be6028fb693c600012d570a6d736d21c622aff0a89995089e\n    user: \"0:0\"\n    working_dir: /workspace\n    network_mode: host\n    depends_on:\n      dynamo-etcd:\n        condition: service_healthy\n      dynamo-nats:\n        condition: service_healthy\n      model-bootstrap:\n        condition: service_completed_successfully\n    volumes:\n      - *hf-volume\n    environment:\n      HF_HOME: /root/.cache/huggingface\n      HF_HUB_OFFLINE: \"1\"\n      ETCD_ENDPOINTS: http://127.0.0.1:2379\n      NATS_SERVER: nats://127.0.0.1:4222\n      DYN_HTTP_PORT: \"30010\"\n      OTEL_SERVICE_NAME: dynamo-frontend\n    command:\n      - python3\n      - -m\n      - dynamo.frontend\n      - --http-host\n      - 127.0.0.1\n      - --http-port\n      - \"30010\"\n      - --router-mode\n      - round-robin\n      - --request-plane\n      - tcp\n      - --dyn-chat-processor\n      - sglang\n    healthcheck:\n      test: [\"CMD-SHELL\", \"curl -sf http://127.0.0.1:30010/v1/models >/dev/null\"]\n      interval: 15s\n      timeout: 5s\n      retries: 120\n      start_period: 30s\n\n  # DeepSeek text plane: host GPUs 0-3.\n  dynamo-worker:\n    <<: *default-restart\n    image: ritualnetwork/dynamo-sglang:v0.5.17-dsv4@sha256:9cc856356107915be6028fb693c600012d570a6d736d21c622aff0a89995089e\n    user: \"0:0\"\n    working_dir: /workspace\n    network_mode: host\n    ipc: host\n    depends_on:\n      dynamo-frontend:\n        condition: service_started\n      model-bootstrap:\n        condition: service_completed_successfully\n    volumes:\n      - *hf-volume\n      - sglang-cache:/root/.cache/sglang\n      - deep-gemm-cache:/root/.cache/deep_gemm\n      - flashinfer-cache:/root/.cache/flashinfer\n    environment:\n      HF_HOME: /root/.cache/huggingface\n      HF_HUB_OFFLINE: \"1\"\n      PYTORCH_CUDA_ALLOC_CONF: expandable_segments:True\n      SGLANG_JIT_DEEPGEMM_PRECOMPILE: \"0\"\n      SGLANG_JIT_DEEPGEMM_FAST_WARMUP: \"1\"\n      NCCL_CUMEM_ENABLE: \"1\"\n      ETCD_ENDPOINTS: http://127.0.0.1:2379\n      NATS_SERVER: nats://127.0.0.1:4222\n      DYN_SYSTEM_PORT: \"8081\"\n      OTEL_SERVICE_NAME: dynamo-worker\n    deploy:\n      resources:\n        reservations:\n          devices:\n            - driver: nvidia\n              device_ids: [\"0\", \"1\", \"2\", \"3\"]\n              capabilities: [gpu]\n    ulimits:\n      memlock: -1\n      stack: 67108864\n      nofile:\n        soft: 65536\n        hard: 65536\n    cap_add: [SYS_PTRACE]\n    command:\n      - python3\n      - -m\n      - dynamo.sglang\n      - --trust-remote-code\n      - --model-path\n      - /root/.cache/huggingface/hub/models--deepseek-ai--DeepSeek-V4-Flash-0731/snapshots/7872f01b1d1fe23eabc4c98b48bffcef5a386062\n      - --served-model-name\n      - deepseek-ai/DeepSeek-V4-Flash-0731\n      - --tp\n      - \"4\"\n      - --moe-runner-backend\n      - marlin\n      - --speculative-algorithm\n      - DSPARK\n      - --mem-fraction-static\n      - \"0.88\"\n      - --context-length\n      - \"1048576\"\n      - --max-running-requests\n      - \"16\"\n      - --incremental-streaming-output\n      - --dyn-reasoning-parser\n      - deepseek-v4\n      - --dyn-tool-call-parser\n      - deepseekv4\n      - --enable-metrics\n      - --enable-cache-report\n      - --host\n      - 127.0.0.1\n    healthcheck:\n      test: [\"CMD-SHELL\", \"curl -sf http://127.0.0.1:8081/metrics >/dev/null\"]\n      interval: 30s\n      timeout: 10s\n      retries: 10\n      start_period: 45m\n\n  # Wan video plane: host GPUs 4-5.\n  wan-t2v:\n    <<: *default-restart\n    image: ritualnetwork/sglang:v0.5.17-diffusion-accelerate1.14.0-tdx1@sha256:9fbe8c2f36ed46b3ec042b81de80d1eba344517decae86d781f20dc4b88cdfd5\n    network_mode: host\n    ipc: host\n    shm_size: \"128g\"\n    depends_on:\n      model-bootstrap:\n        condition: service_completed_successfully\n    volumes:\n      - *hf-volume\n      - wan-video-outputs:/outputs\n    environment:\n      <<: *sglang-env\n      SGLANG_CACHE_DIT_ENABLED: \"true\"\n      SGLANG_CACHE_DIT_TAYLORSEER: \"true\"\n      SGLANG_CACHE_DIT_TS_ORDER: \"2\"\n    deploy:\n      resources:\n        reservations:\n          devices:\n            - driver: nvidia\n              device_ids: [\"4\", \"5\"]\n              capabilities: [gpu]\n    command:\n      - sglang\n      - serve\n      - --model-path\n      - /root/.cache/huggingface/hub/models--Wan-AI--Wan2.2-T2V-A14B-Diffusers/snapshots/5be7df9619b54f4e2667b2755bc6a756675b5cd7\n      - --host\n      - 127.0.0.1\n      - --port\n      - \"30002\"\n      # Diffusion services otherwise share fixed defaults (master 30005,\n      # scheduler 5555) even when their HTTP ports differ.\n      - --master-port\n      - \"31000\"\n      - --scheduler-port\n      - \"31100\"\n      - --num-gpus\n      - \"2\"\n      - --tp-size\n      - \"1\"\n      - --sp-degree\n      - \"2\"\n      - --ulysses-degree\n      - \"2\"\n      - --ring-degree\n      - \"1\"\n      - --attention-backend\n      - fa\n      - --dit-cpu-offload\n      - \"false\"\n      - --dit-layerwise-offload\n      - \"false\"\n      - --text-encoder-cpu-offload\n      - \"false\"\n      - --image-encoder-cpu-offload\n      - \"false\"\n      - --vae-cpu-offload\n      - \"false\"\n      - --output-path\n      - /outputs\n    healthcheck:\n      test:\n        - CMD\n        - python3\n        - -c\n        - \"import urllib.request; urllib.request.urlopen('http://127.0.0.1:30002/health', timeout=10).read()\"\n      interval: 30s\n      timeout: 15s\n      retries: 10\n      start_period: 30m\n\n  # FLUX image plane: host GPU 6.\n  flux2-klein:\n    <<: *default-restart\n    image: ritualnetwork/sglang:v0.5.17-diffusion-accelerate1.14.0-tdx1@sha256:9fbe8c2f36ed46b3ec042b81de80d1eba344517decae86d781f20dc4b88cdfd5\n    network_mode: host\n    ipc: host\n    depends_on:\n      model-bootstrap:\n        condition: service_completed_successfully\n    volumes:\n      - *hf-volume\n      - flux-image-outputs:/outputs\n    environment:\n      <<: *sglang-env\n      SGLANG_CACHE_DIT_ENABLED: \"true\"\n      # device_ids exposes host GPU 6 as GPU 0 inside this container.\n      CUDA_VISIBLE_DEVICES: \"0\"\n    deploy:\n      resources:\n        reservations:\n          devices:\n            - driver: nvidia\n              device_ids: [\"6\"]\n              capabilities: [gpu]\n    command:\n      - sglang\n      - serve\n      - --model-path\n      - /root/.cache/huggingface/hub/models--black-forest-labs--FLUX.2-klein-4B/snapshots/e7b7dc27f91deacad38e78976d1f2b499d76a294\n      - --host\n      - 127.0.0.1\n      - --port\n      - \"30003\"\n      # Keep diffusion control planes disjoint from Wan and each other.\n      - --master-port\n      - \"32000\"\n      - --scheduler-port\n      - \"32100\"\n      - --output-path\n      - /outputs\n    healthcheck:\n      test:\n        - CMD\n        - python3\n        - -c\n        - \"import urllib.request; urllib.request.urlopen('http://127.0.0.1:30003/health', timeout=10).read()\"\n      interval: 30s\n      timeout: 15s\n      retries: 10\n      start_period: 15m\n\n  # Audio shares host GPU 6 with FLUX, matching the current qualified topology.\n  rai-audio:\n    <<: *default-restart\n    image: ritualnetwork/rai-simple:pt1-release-01-8bd2899@sha256:de96e7daac3e0b07e608d189fd51442d9961798032cf03c7bbbe4a3f7bd5c643\n    network_mode: host\n    environment:\n      RAI_PORT: \"30004\"\n      RAI_HOST: 127.0.0.1\n      RAI_WORKERS: \"1\"\n      RAI_MODEL: LiquidAI/LFM2.5-Audio-1.5B\n      RAI_DEVICE: cuda\n      RAI_DTYPE: bfloat16\n      # device_ids exposes host GPU 6 as GPU 0 inside this container.\n      CUDA_VISIBLE_DEVICES: \"0\"\n      TORCH_COMPILE_DISABLE: \"1\"\n      TORCHDYNAMO_DISABLE: \"1\"\n    deploy:\n      resources:\n        reservations:\n          devices:\n            - driver: nvidia\n              device_ids: [\"6\"]\n              capabilities: [gpu]\n    command: [\"python3\", \"-m\", \"rai.simple_server\"]\n    healthcheck:\n      test: [\"CMD-SHELL\", \"curl -sf http://127.0.0.1:30004/health >/dev/null\"]\n      interval: 15s\n      timeout: 5s\n      retries: 20\n      start_period: 120s\n\n  # MOAI Full-E2E-only FHE plane: host GPU 7. The client secret key never enters GCS.\n  moai-fhe:\n    <<: *default-restart\n    image: ritualnetwork/ritual-moai-gpu:8102c66-amd64@sha256:63113fdf2bf457b84e53e325611e65f488783cd2e39f24ff65d1bf1e5c308543\n    network_mode: host\n    environment:\n      MOAI_NUM_LAYERS: \"12\"\n      MOAI_HIDDEN_SIZE: \"768\"\n      MOAI_NUM_HEADS: \"12\"\n      MOAI_HEAD_SIZE: \"64\"\n      MOAI_INTERMEDIATE_SIZE: \"3072\"\n      MOAI_MAX_SEQ_LENGTH: \"128\"\n      MOAI_BATCH_SIZE: \"256\"\n      MOAI_MULT_DEPTH: \"29\"\n      BERT_BINARY_PATH: /app/build/fhe_b_inference_server\n      BERT_WEIGHTS_DIR: /app/build\n      MOAI_HTTP_HOST: 127.0.0.1\n      MOAI_HTTP_PORT: \"30008\"\n      MOAI_GRPC_HOST: 127.0.0.1\n      MOAI_GRPC_PORT: \"30009\"\n      MOAI_MODEL_NAME: bert-base-uncased\n      MOAI_MAX_CONCURRENT_INFERENCE: \"1\"\n      GCS_BUCKET: ritual-moai-fhe-pt1-e2e\n      MOAI_GCS_CACHE: \"1\"\n      MOAI_GCS_CACHE_DIR: /app/gcs_cache\n      OMP_NUM_THREADS: \"8\"\n      PYTHONUNBUFFERED: \"1\"\n      # device_ids exposes host GPU 7 as GPU 0 inside this container.\n      CUDA_VISIBLE_DEVICES: \"0\"\n    deploy:\n      resources:\n        reservations:\n          devices:\n            - driver: nvidia\n              device_ids: [\"7\"]\n              capabilities: [gpu]\n    volumes:\n      - moai-keys:/app/keys\n      - moai-logs:/app/logs\n      - moai-gcs-cache:/app/gcs_cache\n    tmpfs:\n      - /run/moai-secrets:rw,noexec,nosuid,nodev,mode=0700,size=1m\n    secrets:\n      - moai-gcs-sa-b64\n    entrypoint: [\"/bin/bash\", \"-ceu\"]\n    command:\n      - |\n        umask 077\n        credential=/run/moai-secrets/gcs-service-account.json\n        child_pid=\n        cleanup() { rm -f \"$${credential}\"; }\n        terminate() {\n          trap - HUP INT TERM\n          if [ -n \"$${child_pid}\" ]; then\n            kill -TERM \"$${child_pid}\" 2>/dev/null || true\n            wait \"$${child_pid}\" 2>/dev/null || true\n          fi\n          exit 143\n        }\n        trap cleanup EXIT\n        trap terminate HUP INT TERM\n        base64 -d /run/secrets/moai-gcs-sa-b64 >\"$${credential}\"\n        chmod 0600 \"$${credential}\"\n        export GOOGLE_APPLICATION_CREDENTIALS=\"$${credential}\"\n        python3 /app/python/server.py &\n        child_pid=$$!\n        wait \"$${child_pid}\"\n    healthcheck:\n      test:\n        - CMD\n        - python3\n        - -c\n        - \"import urllib.request; urllib.request.urlopen('http://127.0.0.1:30008/health', timeout=5).read()\"\n      interval: 30s\n      timeout: 10s\n      retries: 20\n      start_period: 180s\n\n  redis:\n    <<: *default-restart\n    image: redis:7.4-alpine@sha256:ff02b58f971e7d7d156a1267e283fcbbeee91773b6aa36c49dac28ecfe28eadf\n    network_mode: host\n    command:\n      - redis-server\n      - --maxmemory\n      - 1gb\n      - --maxmemory-policy\n      - allkeys-lru\n      - --bind\n      - 127.0.0.1\n      - --protected-mode\n      - \"yes\"\n      - --save\n      - \"\"\n      - --appendonly\n      - \"no\"\n    healthcheck:\n      test: [\"CMD-SHELL\", \"redis-cli ping | grep -q PONG\"]\n      interval: 10s\n      timeout: 5s\n      retries: 12\n\n  bifrost:\n    <<: *default-restart\n    image: maximhq/bifrost:v1.5.7-amd64@sha256:410de09d5b96b15bda204dc73ac4a7d1712d598d0b9e67a8e8917b964968ab79\n    network_mode: host\n    environment:\n      APP_HOST: 127.0.0.1\n    volumes:\n      - bifrost-data:/app/data\n    configs:\n      - source: bifrost-config\n        target: /app/data/config.json\n    healthcheck:\n      test: [\"CMD\", \"wget\", \"--no-verbose\", \"--tries=1\", \"--spider\", \"http://127.0.0.1:8080/health\"]\n      interval: 15s\n      timeout: 5s\n      retries: 12\n\n  # Chain-agnostic attested front door. Chains register after deployment via /register.\n  shared-proxy:\n    <<: *default-restart\n    image: ritualnetwork/ritual-vllm-proxy-go:cf17d76-0e65f02-amd64@sha256:d9e9bdd3adde92f5b2217981782bb8201a5be4bb6057a4fb044d32d1e0f92dbd\n    network_mode: host\n    privileged: true\n    depends_on:\n      dynamo-frontend:\n        condition: service_healthy\n      dynamo-worker:\n        condition: service_healthy\n      wan-t2v:\n        condition: service_healthy\n      flux2-klein:\n        condition: service_healthy\n      bifrost:\n        condition: service_healthy\n      redis:\n        condition: service_healthy\n      rai-audio:\n        condition: service_healthy\n      moai-fhe:\n        condition: service_healthy\n    devices:\n      - /dev/tdx_guest:/dev/tdx_guest\n      - /dev/nvidia0:/dev/nvidia0\n      - /dev/nvidia1:/dev/nvidia1\n      - /dev/nvidia2:/dev/nvidia2\n      - /dev/nvidia3:/dev/nvidia3\n      - /dev/nvidia4:/dev/nvidia4\n      - /dev/nvidia5:/dev/nvidia5\n      - /dev/nvidia6:/dev/nvidia6\n      - /dev/nvidia7:/dev/nvidia7\n      - /dev/nvidiactl:/dev/nvidiactl\n      - /dev/nvidia-uvm:/dev/nvidia-uvm\n    deploy:\n      resources:\n        reservations:\n          devices:\n            - driver: nvidia\n              count: all\n              capabilities: [gpu, compute, utility]\n    volumes:\n      - /etc/tdx-attest.conf:/etc/tdx-attest.conf:ro\n      - /var/run/dstack.sock:/var/run/dstack.sock\n      - /usr/bin/nvidia-smi:/usr/bin/nvidia-smi:ro\n    environment:\n      PORT: \"8020\"\n      OPS_PORT: \"8021\"\n      LOG_JSON: \"true\"\n      HTTP_REQUEST_LOGGING_ENABLED: \"false\"\n      MODEL_NAME: deepseek-ai/DeepSeek-V4-Flash-0731\n      VLLM_BASE_URL: http://127.0.0.1:30010\n      LLM_ADMISSION_ENABLED: \"true\"\n      LLM_QUEUE_DEPTH_LIMIT: \"100\"\n      LLM_ADMISSION_POLL_INTERVAL_MS: \"2000\"\n      LLM_ADMISSION_METRICS_URL: http://127.0.0.1:8081/metrics\n      LLM_ADMISSION_METRIC_NAME: sglang:num_queue_reqs\n      LLM_MAX_OUTPUT_TOKENS: \"8192\"\n      LLM_DEFAULT_OUTPUT_TOKENS: \"4096\"\n      LLM_TRACE_LOGGING_ENABLED: \"false\"\n      LLM_TRACE_LOG_REQUEST_BODY: \"false\"\n      LLM_TRACE_REQUEST_BODY_MAX_BYTES: \"4096\"\n      IMAGE_BACKEND_URL: http://127.0.0.1:8080/openai\n      AUDIO_BACKEND_URL: http://127.0.0.1:8080/openai\n      VIDEO_BACKEND_URL: http://127.0.0.1:30002\n      MOAI_BASE_URL: http://127.0.0.1:30008\n      MOAI_TIMEOUT_SECONDS: \"600\"\n      IMAGE_MODEL_NAME: black-forest-labs/FLUX.2-klein-4B\n      AUDIO_MODEL_NAME: LiquidAI/LFM2.5-Audio-1.5B\n      VIDEO_MODEL_NAME: Wan-AI/Wan2.2-T2V-A14B-Diffusers\n      IMAGE_MAX_CONCURRENT: \"8\"\n      IMAGE_MAX_QUEUED: \"20\"\n      AUDIO_MAX_CONCURRENT: \"5\"\n      AUDIO_MAX_QUEUED: \"10\"\n      VIDEO_MAX_CONCURRENT: \"2\"\n      VIDEO_MAX_QUEUED: \"4\"\n      IMAGE_TIMEOUT_SECONDS: \"300\"\n      AUDIO_TIMEOUT_SECONDS: \"300\"\n      VIDEO_TIMEOUT_SECONDS: \"1800\"\n      REDIS_HOST: 127.0.0.1\n      REDIS_PORT: \"6379\"\n      CHAT_CACHE_EXPIRATION: \"1200\"\n      EXECUTOR_REGISTRY_CACHE_TTL: 3h\n      TEE_TYPE: TDX\n      MOCK_GPU_ATTESTATION: \"false\"\n      MOCK_VLLM_MODE: \"false\"\n      EVIDENCE_GETTER_PATH: /usr/local/bin/evidence_getter\n      EVIDENCE_GETTER_USE_SUDO: \"false\"\n      REGISTRATION_API_ENABLED: \"true\"\n      REGISTRATION_ENABLED: \"false\"\n      PAYMENT_ADDRESS: ${PAYMENT_ADDRESS:?PAYMENT_ADDRESS is required}\n      SERVICE_ENDPOINT: https://${DSTACK_APP_ID:?DSTACK_APP_ID is required}-8020s.${DSTACK_GATEWAY_DOMAIN:?DSTACK_GATEWAY_DOMAIN is required}\n      NVIDIA_VISIBLE_DEVICES: all\n      NVIDIA_DRIVER_CAPABILITIES: compute,utility\n    healthcheck:\n      test: [\"CMD-SHELL\", \"curl -sf http://127.0.0.1:8021/health >/dev/null\"]\n      interval: 15s\n      timeout: 5s\n      retries: 12\n      start_period: 45s\n\n  output-cleanup:\n    <<: *default-restart\n    image: alpine:3@sha256:28bd5fe8b56d1bd048e5babf5b10710ebe0bae67db86916198a6eec434943f8b\n    volumes:\n      - wan-video-outputs:/outputs/video\n      - flux-image-outputs:/outputs/image\n    environment:\n      VIDEO_TTL_MIN: \"30\"\n      IMAGE_TTL_MIN: \"15\"\n      CLEANUP_INTERVAL_SEC: \"60\"\n    entrypoint: [\"/bin/sh\", \"-ceu\"]\n    command:\n      - |\n        while true; do\n          find /outputs/video -type f -mmin +$${VIDEO_TTL_MIN} -delete 2>/dev/null || true\n          find /outputs/image -type f -mmin +$${IMAGE_TTL_MIN} -delete 2>/dev/null || true\n          sleep \"$${CLEANUP_INTERVAL_SEC}\"\n        done\n\nconfigs:\n  dynamo-nats-config:\n    content: |\n      host: \"127.0.0.1\"\n      port: 4222\n      http: \"127.0.0.1:8222\"\n      jetstream {}\n  bifrost-config:\n    content: |\n      {\n        \"$$schema\": \"https://www.getbifrost.ai/schema\",\n        \"providers\": {\n          \"deepseek-ai\": {\n            \"keys\": [{\n              \"name\": \"deepseek-ai-dynamo-local\",\n              \"value\": \"not-needed\",\n              \"models\": [\"DeepSeek-V4-Flash-0731\", \"deepseek-ai/DeepSeek-V4-Flash-0731\"],\n              \"weight\": 1\n            }],\n            \"network_config\": {\n              \"base_url\": \"http://127.0.0.1:30010\",\n              \"default_request_timeout_in_seconds\": 120,\n              \"max_retries\": 3\n            },\n            \"custom_provider_config\": {\n              \"is_key_less\": true,\n              \"base_provider_type\": \"openai\"\n            }\n          },\n          \"flux\": {\n            \"keys\": [{\n              \"name\": \"flux-sglang-local\",\n              \"value\": \"not-needed\",\n              \"models\": [\"black-forest-labs/FLUX.2-klein-4B\"],\n              \"weight\": 1\n            }],\n            \"network_config\": {\n              \"base_url\": \"http://127.0.0.1:30003\",\n              \"default_request_timeout_in_seconds\": 300,\n              \"max_retries\": 3\n            },\n            \"custom_provider_config\": {\n              \"is_key_less\": true,\n              \"base_provider_type\": \"openai\"\n            }\n          },\n          \"audio\": {\n            \"keys\": [{\n              \"name\": \"audio-rai-local\",\n              \"value\": \"not-needed\",\n              \"models\": [\"LiquidAI/LFM2.5-Audio-1.5B\"],\n              \"weight\": 1\n            }],\n            \"network_config\": {\n              \"base_url\": \"http://127.0.0.1:30004\",\n              \"default_request_timeout_in_seconds\": 300,\n              \"max_retries\": 3\n            },\n            \"custom_provider_config\": {\n              \"is_key_less\": true,\n              \"base_provider_type\": \"openai\"\n            }\n          },\n          \"video\": {\n            \"keys\": [{\n              \"name\": \"video-sglang-local\",\n              \"value\": \"not-needed\",\n              \"models\": [\"Wan-AI/Wan2.2-T2V-A14B-Diffusers\"],\n              \"weight\": 1\n            }],\n            \"network_config\": {\n              \"base_url\": \"http://127.0.0.1:30002\",\n              \"default_request_timeout_in_seconds\": 1800,\n              \"max_retries\": 1\n            },\n            \"custom_provider_config\": {\n              \"is_key_less\": true,\n              \"base_provider_type\": \"openai\"\n            }\n          }\n        },\n        \"config_store\": {\n          \"enabled\": true,\n          \"type\": \"sqlite\",\n          \"config\": {\"path\": \"/app/data/config.db\"}\n        },\n        \"logs_store\": {\n          \"enabled\": true,\n          \"type\": \"sqlite\",\n          \"config\": {\"path\": \"/app/data/logs.db\"}\n        },\n        \"client\": {\n          \"initial_pool_size\": 300,\n          \"allowed_origins\": [],\n          \"enable_logging\": true,\n          \"enable_governance\": false,\n          \"allow_direct_keys\": false,\n          \"max_request_body_size_mb\": 100\n        }\n      }\n\nsecrets:\n  # Values come from Phala's encrypted deployment env and are mounted as files,\n  # not exposed in long-lived container environments or process arguments.\n  hf-token:\n    environment: HF_TOKEN\n  moai-gcs-sa-b64:\n    environment: MOAI_GCS_SERVICE_ACCOUNT_JSON_B64\n\nvolumes:\n  hf-model-cache:\n  dynamo-etcd:\n  sglang-cache:\n  deep-gemm-cache:\n  flashinfer-cache:\n  wan-video-outputs:\n  flux-image-outputs:\n  bifrost-data:\n  moai-keys:\n  moai-logs:\n  moai-gcs-cache:\n","features":["kms","tproxy-net"],"gateway_enabled":true,"kms_enabled":true,"local_key_provider_enabled":false,"manifest_version":2,"name":"app_53b1afccdae26112c17e890b6c3586cfae7717ce","no_instance_id":false,"port_policy":{"ports":[{"port":8020,"pp":false},{"port":8021,"pp":false},{"port":8090,"pp":false}],"restrict_mode":true},"pre_launch_script":"#!/bin/bash\necho \"----------------------------------------------\"\necho \"Running Phala Cloud Pre-Launch Script v0.0.19\"\necho \"----------------------------------------------\"\nset -e\n\n# Function: notify host\n\nnotify_host() {\n    if command -v dstack-util >/dev/null 2>&1; then\n        dstack-util notify-host -e \"$1\" -d \"$2\"\n    else\n        tdxctl notify-host -e \"$1\" -d \"$2\"\n    fi\n}\n\nnotify_host_hoot_info() {\n    notify_host \"boot.progress\" \"$1\"\n}\n\nnotify_host_hoot_error() {\n    notify_host \"boot.error\" \"$1\"\n}\n\n# Function: Perform Docker cleanup\nperform_cleanup() {\n    echo \"Pruning unused images\"\n    docker image prune -af\n    echo \"Pruning unused volumes\"\n    docker volume prune -f\n    notify_host_hoot_info \"docker cleanup completed\"\n}\n\n# Function: Check Docker login status without exposing credentials\ncheck_docker_login() {\n    local registry=\"$1\"\n\n    # When registry is specified, check auth entry for that registry in Docker config\n    if [[ -n \"$registry\" ]]; then\n        local docker_config_path=\"${DOCKER_CONFIG:-$HOME/.docker}/config.json\"\n        if [[ -f \"$docker_config_path\" ]] && grep -q \"$registry\" \"$docker_config_path\"; then\n            return 0\n        else\n            return 1\n        fi\n    fi\n\n    # Fallback check when no explicit registry is provided\n    if docker info 2>/dev/null | grep -q \"Username\"; then\n        return 0\n    else\n        return 1\n    fi\n}\n\n# Function: print the SHA256 fingerprint of every authorized key\n#\n# Output matches `ssh-keygen -lf` and the fingerprints GitHub and Phala Cloud\n# display, so an operator can match the boot log against the account's key list.\n# The guest image ships neither ssh-keygen nor a base64 applet (BusyBox has\n# base32 only), so openssl does both the decode and the digest.\nprint_authorized_key_fingerprints() {\n    local file=\"$1\"\n    if ! command -v openssl >/dev/null 2>&1; then\n        echo \"openssl not available; skipping SSH key fingerprints\"\n        return 0\n    fi\n\n    local key_type blob comment decoded_bytes fingerprint\n    # `|| [[ -n \"$key_type\" ]]` keeps the final line when the file does not end\n    # with a newline; read returns non-zero there but still fills the variables.\n    while read -r key_type blob comment || [[ -n \"$key_type\" ]]; do\n        if [[ -z \"$blob\" ]]; then\n            continue\n        fi\n        # openssl exits 0 on undecodable input and emits nothing, which would\n        # otherwise print the SHA256 of the empty string as a real fingerprint.\n        decoded_bytes=$(printf '%s' \"$blob\" | openssl base64 -d -A 2>/dev/null | wc -c || true)\n        if [[ \"$decoded_bytes\" -eq 0 ]]; then\n            echo \"  $key_type <unreadable key>\"\n            continue\n        fi\n        fingerprint=$(\n            printf '%s' \"$blob\"                 | openssl base64 -d -A 2>/dev/null                 | openssl dgst -sha256 -binary 2>/dev/null                 | openssl base64 -A 2>/dev/null                 | tr -d '='                 || true\n        )\n        echo \"  $key_type SHA256:$fingerprint\"\n    done < \"$file\"\n}\n\n# Main logic starts here\necho \"Starting login process...\"\n\n# Check if Docker credentials exist\nif [[ -n \"$DSTACK_DOCKER_USERNAME\" && -n \"$DSTACK_DOCKER_PASSWORD\" ]]; then\n    echo \"Docker credentials found\"\n    DOCKER_REGISTRY_TARGET=\"${DSTACK_DOCKER_REGISTRY:-docker.io}\"\n    echo \"Target Docker registry: $DOCKER_REGISTRY_TARGET\"\n\n    # Check if already logged in\n    if check_docker_login \"$DSTACK_DOCKER_REGISTRY\"; then\n        echo \"Already logged in to Docker registry: $DOCKER_REGISTRY_TARGET\"\n    else\n        echo \"Logging in to Docker registry: $DOCKER_REGISTRY_TARGET\"\n        # Login without exposing password in process list\n        if [[ -n \"$DSTACK_DOCKER_REGISTRY\" ]]; then\n            if ! echo \"$DSTACK_DOCKER_PASSWORD\" | docker login -u \"$DSTACK_DOCKER_USERNAME\" --password-stdin \"$DSTACK_DOCKER_REGISTRY\"; then\n                echo \"Docker login failed: $DOCKER_REGISTRY_TARGET\"\n                notify_host_hoot_error \"docker login failed\"\n                exit 1\n            fi\n        elif ! echo \"$DSTACK_DOCKER_PASSWORD\" | docker login -u \"$DSTACK_DOCKER_USERNAME\" --password-stdin; then\n            echo \"Docker login failed: $DOCKER_REGISTRY_TARGET\"\n            notify_host_hoot_error \"docker login failed\"\n            exit 1\n        fi\n        echo \"Docker login successful: $DOCKER_REGISTRY_TARGET\"\n    fi\n# Check if AWS ECR credentials exist\nelif [[ -n \"$DSTACK_AWS_ACCESS_KEY_ID\" && -n \"$DSTACK_AWS_SECRET_ACCESS_KEY\" && -n \"$DSTACK_AWS_REGION\" && -n \"$DSTACK_AWS_ECR_REGISTRY\" ]]; then\n    echo \"AWS ECR credentials found\"\n\n    # Check if AWS CLI is installed\n    if [ ! -f \"./aws/dist/aws\" ]; then\n        notify_host_hoot_info \"awscli not installed, installing...\"\n        echo \"AWS CLI not installed, installing...\"\n        if ! curl \"https://awscli.amazonaws.com/awscli-exe-linux-x86_64-2.24.14.zip\" -o \"awscliv2.zip\"; then\n            echo \"AWS CLI download failed\"\n            notify_host_hoot_error \"awscli install failed\"\n            exit 1\n        fi\n        if ! echo \"6ff031a26df7daebbfa3ccddc9af1450 awscliv2.zip\" | md5sum -c; then\n            echo \"MD5 checksum failed\"\n            notify_host_hoot_error \"awscli install failed\"\n            exit 1\n        fi\n        unzip awscliv2.zip &> /dev/null\n    else\n        echo \"AWS CLI is already installed: ./aws/dist/aws\"\n    fi\n\n    # Set AWS credentials as environment variables\n    export AWS_ACCESS_KEY_ID=\"$DSTACK_AWS_ACCESS_KEY_ID\"\n    export AWS_SECRET_ACCESS_KEY=\"$DSTACK_AWS_SECRET_ACCESS_KEY\"\n    export AWS_DEFAULT_REGION=\"$DSTACK_AWS_REGION\"\n\n    # Set session token if provided (for temporary credentials)\n    if [[ -n \"$DSTACK_AWS_SESSION_TOKEN\" ]]; then\n        echo \"AWS session token found, using temporary credentials\"\n        export AWS_SESSION_TOKEN=\"$DSTACK_AWS_SESSION_TOKEN\"\n    fi\n\n    # Test AWS credentials before attempting ECR login\n    echo \"Testing AWS credentials...\"\n    if ! ./aws/dist/aws sts get-caller-identity &> /dev/null; then\n        echo \"AWS credentials test failed\"\n        # For session token credentials, this might be expected if they're expired\n        # Log warning but don't fail startup\n        if [[ -n \"$DSTACK_AWS_SESSION_TOKEN\" ]]; then\n            echo \"Warning: AWS temporary credentials may have expired, continuing startup\"\n            notify_host_hoot_info \"AWS temporary credentials may have expired\"\n        else\n            echo \"AWS credentials test failed\"\n            notify_host_hoot_error \"Invalid AWS credentials\"\n            exit 1\n        fi\n    else\n        echo \"Logging in to AWS ECR...\"\n        if ./aws/dist/aws ecr get-login-password --region \"$DSTACK_AWS_REGION\" | docker login --username AWS --password-stdin \"$DSTACK_AWS_ECR_REGISTRY\"; then\n            echo \"AWS ECR login successful\"\n            notify_host_hoot_info \"AWS ECR login successful\"\n        else\n            echo \"AWS ECR login failed\"\n            # For session token credentials, don't fail startup if login fails\n            if [[ -n \"$DSTACK_AWS_SESSION_TOKEN\" ]]; then\n                echo \"Warning: AWS ECR login failed with temporary credentials, continuing startup\"\n                notify_host_hoot_info \"AWS ECR login failed with temporary credentials\"\n            else\n                notify_host_hoot_error \"AWS ECR login failed\"\n                exit 1\n            fi\n        fi\n    fi\nfi\n\nperform_cleanup\n\n#\n# GHCR image pull access verification (pure HTTP, no docker daemon)\n#\nif [[ \"$DOCKER_REGISTRY_TARGET\" == \"ghcr.io\" && -n \"$DSTACK_DOCKER_USERNAME\" && -n \"$DSTACK_DOCKER_PASSWORD\" ]]; then\n    COMPOSE_IMAGES=$(grep 'image:' /dstack/docker-compose.yaml 2>/dev/null | awk '{print $2}' | tr -d '\"'\"'\" || true)\n    HEADER=\"Accept: application/vnd.oci.image.index.v1+json,application/vnd.oci.image.manifest.v1+json,application/vnd.docker.distribution.manifest.list.v2+json,application/vnd.docker.distribution.manifest.v2+json\"\n    for img in $COMPOSE_IMAGES; do\n        [[ \"$img\" != ghcr.io/* ]] && continue\n        ref=\"${img#ghcr.io/}\"\n        # Digest-pinned refs (repo@sha256:...) use the digest as the manifest\n        # reference; tag refs (repo:tag) use the tag, defaulting to latest.\n        if [[ \"$ref\" == *@* ]]; then\n            repo=\"${ref%%@*}\"\n            reference=\"${ref#*@}\"\n        else\n            repo=\"${ref%%:*}\"\n            reference=\"${ref##*:}\"\n            [[ \"$reference\" == \"$ref\" ]] && reference=\"latest\"\n        fi\n        echo \"Verifying GHCR pull access: $img\"\n        token=$(curl -sf -u \"$DSTACK_DOCKER_USERNAME:$DSTACK_DOCKER_PASSWORD\" \"https://ghcr.io/token?service=ghcr.io&scope=repository:${repo}:pull\" | jq -r '.token // empty' || true)\n        if [[ -z \"$token\" ]]; then\n            echo \"ERROR: GHCR token exchange failed for $img\"\n            notify_host_hoot_error \"GHCR token exchange failed: $img\"\n            exit 1\n        fi\n        http_code=$(curl -s -o /dev/null -w \"%{http_code}\" -H \"Authorization: Bearer $token\" -H \"$HEADER\" \"https://ghcr.io/v2/${repo}/manifests/${reference}\")\n        if [[ \"$http_code\" != \"200\" ]]; then\n            echo \"ERROR: GHCR pull access denied for $img (HTTP $http_code)\"\n            notify_host_hoot_error \"GHCR pull access denied: $img (HTTP $http_code)\"\n            exit 1\n        fi\n        echo \"GHCR pull access OK: $img\"\n    done\nfi\n\n#\n# Pull latest images from docker-compose.yaml so existing CVMs pick up new tags.\n# Pull is fail-soft: short-lived tokens may expire between login and pull, but\n# falling back to already-cached images is preferable to blocking boot.\n#\necho \"Images before pull:\"\ndocker images --format '{{.Repository}}:{{.Tag}} {{.ID}} ({{.CreatedSince}})'\n\necho \"Pulling latest images from /dstack/docker-compose.yaml...\"\nif docker compose -f /dstack/docker-compose.yaml pull; then\n    echo \"docker compose pull completed\"\n    notify_host_hoot_info \"docker compose pull completed\"\nelse\n    echo \"WARNING: docker compose pull failed; continuing with existing images\"\n    notify_host_hoot_info \"docker compose pull failed; using existing images\"\nfi\n\necho \"Images after pull:\"\ndocker images --format '{{.Repository}}:{{.Tag}} {{.ID}} ({{.CreatedSince}})'\n\n#\n# Set root password.\n#\necho \"Setting root password..\"\n\n# Check if password files are writable\nPASSWD_WRITABLE=true\nif [ ! -w /etc/passwd ]; then\n    echo \"Warning: /etc/passwd is read-only\"\n    PASSWD_WRITABLE=false\nfi\nif [ ! -w /etc/shadow ]; then\n    echo \"Warning: /etc/shadow is read-only\"\n    PASSWD_WRITABLE=false\nfi\n\nif [ \"$PASSWD_WRITABLE\" = \"false\" ]; then\n    echo \"Skipping password setup due to read-only file system\"\nelse\n    # Check if chpasswd is available\n    if command -v chpasswd >/dev/null 2>&1; then\n        echo \"Using chpasswd method\"\n\n        if [ -n \"$DSTACK_ROOT_PASSWORD\" ]; then\n            echo \"Setting root password from user..\"\n            echo \"root:$DSTACK_ROOT_PASSWORD\" | chpasswd\n            unset DSTACK_ROOT_PASSWORD\n            echo \"Root password set/updated from DSTACK_ROOT_PASSWORD\"\n        elif [ -z \"$(grep '^root:' /etc/shadow 2>/dev/null | cut -d: -f2)\" ]; then\n            echo \"Setting random root password..\"\n            # dstack guest busybox lacks head -c (FANCY_HEAD off); dd is the\n            # last pipeline command, so its 0 status masks tr's SIGPIPE exit.\n            DSTACK_ROOT_PASSWORD=$(\n                LC_ALL=C tr -dc 'A-Za-z0-9' < /dev/urandom | dd bs=1 count=32 2>/dev/null\n            )\n            echo \"root:$DSTACK_ROOT_PASSWORD\" | chpasswd\n            unset DSTACK_ROOT_PASSWORD\n            echo \"Root password set (random auto-init)\"\n        else\n            echo \"Root password already set; no changes.\"\n        fi\n    else\n        # No chpasswd: do not pipe into interactive passwd(1). It often reads\n        # /dev/tty and fails with \"password for root is unchanged\" during boot.\n        echo \"Using passwd method\"\n\n        if [ -n \"$DSTACK_ROOT_PASSWORD\" ]; then\n            echo \"Setting root password from user..\"\n            if command -v passwd >/dev/null 2>&1 && passwd --help 2>&1 | grep -q -- '--stdin'; then\n                echo \"$DSTACK_ROOT_PASSWORD\" | passwd --stdin root\n            elif command -v openssl >/dev/null 2>&1; then\n                # busybox passwd has no --stdin; write the SHA-512 crypt hash\n                # into /etc/shadow directly instead of driving interactive passwd.\n                HASHED_ROOT_PASSWORD=$(openssl passwd -6 \"$DSTACK_ROOT_PASSWORD\")\n                sed -i \"s|^root:[^:]*:|root:${HASHED_ROOT_PASSWORD}:|\" /etc/shadow\n                unset HASHED_ROOT_PASSWORD\n            else\n                echo \"Error: cannot set DSTACK_ROOT_PASSWORD non-interactively\"\n                echo \"Need chpasswd, passwd --stdin, or openssl\"\n                exit 1\n            fi\n            unset DSTACK_ROOT_PASSWORD\n            echo \"Root password set/updated from DSTACK_ROOT_PASSWORD\"\n        elif [ -z \"$(grep '^root:' /etc/shadow 2>/dev/null | cut -d: -f2)\" ]; then\n            # Never lock the account here: sshd built without PAM (dstack dev\n            # images) treats a locked account as \"deny all authentication\",\n            # which breaks public-key SSH too. Set a discarded random password\n            # by writing its crypt hash into /etc/shadow directly.\n            if command -v openssl >/dev/null 2>&1; then\n                echo \"Setting random root password..\"\n                DSTACK_ROOT_PASSWORD=$(\n                    LC_ALL=C tr -dc 'A-Za-z0-9' < /dev/urandom | dd bs=1 count=32 2>/dev/null\n                )\n                HASHED_ROOT_PASSWORD=$(openssl passwd -6 \"$DSTACK_ROOT_PASSWORD\")\n                sed -i \"s|^root:[^:]*:|root:${HASHED_ROOT_PASSWORD}:|\" /etc/shadow\n                unset DSTACK_ROOT_PASSWORD HASHED_ROOT_PASSWORD\n                echo \"Root password set (random auto-init)\"\n            else\n                echo \"Warning: no openssl to hash a random password; leaving root password empty\"\n            fi\n        else\n            echo \"Root password already set; no changes.\"\n        fi\n    fi\nfi\n\n#\n# Set SSH authorized keys\n#\nif mkdir -p /home/root/.ssh 2>/dev/null; then\n    if [[ -n \"$DSTACK_ROOT_PUBLIC_KEY\" ]]; then\n        echo \"$DSTACK_ROOT_PUBLIC_KEY\" > /home/root/.ssh/authorized_keys\n        unset DSTACK_ROOT_PUBLIC_KEY\n        echo \"Root public key set\"\n    fi\n    if [[ -n \"$DSTACK_AUTHORIZED_KEYS\" ]]; then\n        echo \"$DSTACK_AUTHORIZED_KEYS\" > /home/root/.ssh/authorized_keys\n        unset DSTACK_AUTHORIZED_KEYS\n        echo \"Root authorized_keys set\"\n    fi\n\n    if [[ -f /dstack/user_config ]] && jq empty /dstack/user_config 2>/dev/null; then\n        if [[ $(jq 'has(\"ssh_authorized_keys\")' /dstack/user_config 2>/dev/null) == \"true\" ]]; then\n            # jq -r terminates the value with a newline; jq -j does not, which\n            # both concatenated the next appended key onto the same line and\n            # made line-based counting report one key fewer than the file holds.\n            jq -r '.ssh_authorized_keys' /dstack/user_config >> /home/root/.ssh/authorized_keys\n            # Drop duplicates and the blank line an empty key list produces.\n            sort -u /home/root/.ssh/authorized_keys > /home/root/.ssh/authorized_keys.tmp\n            mv /home/root/.ssh/authorized_keys.tmp /home/root/.ssh/authorized_keys\n            KEY_COUNT=$(grep -c '^[^[:space:]]' /home/root/.ssh/authorized_keys || true)\n            echo \"Set root authorized_keys from user preferences, total $KEY_COUNT keys\"\n            print_authorized_key_fingerprints /home/root/.ssh/authorized_keys\n        fi\n    fi\nelse\n    echo \"Warning: Cannot create /home/root/.ssh directory (read-only file system?)\"\n    echo \"Skipping SSH key setup\"\nfi\n\nif [[ -S /var/run/dstack.sock ]]; then\n    export DSTACK_APP_ID=$(curl -s --unix-socket /var/run/dstack.sock http://dstack/Info | jq -j .app_id)\nelif [[ -S /var/run/tappd.sock ]]; then\n    export DSTACK_APP_ID=$(curl -s --unix-socket /var/run/tappd.sock http://dstack/prpc/Tappd.Info | jq -j .app_id)\nfi\n# Check if DSTACK_GATEWAY_DOMAIN is not set, try to get it from user_config or app-compose.json\n# Priority: user_config > app-compose.json\nif [[ -z \"$DSTACK_GATEWAY_DOMAIN\" ]]; then\n    # First try to get from /dstack/user_config if it exists and is valid JSON\n    if [[ -f /dstack/user_config ]] && jq empty /dstack/user_config 2>/dev/null; then\n        if [[ $(jq 'has(\"default_gateway_domain\")' /dstack/user_config 2>/dev/null) == \"true\" ]]; then\n            export DSTACK_GATEWAY_DOMAIN=$(jq -j '.default_gateway_domain' /dstack/user_config)\n        fi\n    fi\n\n    # If still not set, try to get from app-compose.json\n    if [[ -z \"$DSTACK_GATEWAY_DOMAIN\" ]] && [[ $(jq 'has(\"default_gateway_domain\")' app-compose.json) == \"true\" ]]; then\n        export DSTACK_GATEWAY_DOMAIN=$(jq -j '.default_gateway_domain' app-compose.json)\n    fi\nfi\nif [[ -n \"$DSTACK_GATEWAY_DOMAIN\" ]]; then\n    export DSTACK_APP_DOMAIN=$DSTACK_APP_ID\".\"$DSTACK_GATEWAY_DOMAIN\nfi\n\necho \"----------------------------------------------\"\necho \"Script execution completed\"\necho \"----------------------------------------------\"","public_logs":false,"public_sysinfo":false,"public_tcbinfo":true,"runner":"docker-compose","secure_time":false,"storage_fs":"zfs","tproxy_enabled":true}

• dstack reconstruction configuration — cdi-phala-dstack-reconstruction-config.json

  {"schema_version":1,"app_id":"53b1afccdae26112c17e890b6c3586cfae7717ce","instance_id":null,"compose_hash":"1ddda8f6c2219063efe067c7ea4839c0e0b8a3bf2c0a85d6eed2f3f6a838cfd1","os_image_hash":"806a352e16175d90568de97dff563f31f680239e6b90e9b5b2e9141d0955b0d9","vm_config":{"spec_version":1,"os_image_hash":"806a352e16175d90568de97dff563f31f680239e6b90e9b5b2e9141d0955b0d9","cpu_count":192,"memory_size":1649267441664,"qemu_single_pass_add_pages":null,"pic":null,"qemu_version":"9.2.1","pci_hole64_size":0,"hugepages":false,"num_gpus":0,"num_nvswitches":0,"hotplug_off":false,"image":"dstack-nvidia-0.5.9","host_share_mode":""}}
  

Docker container images and source

• alpine: docker.io/library/alpine:3@sha256:28bd5fe8b56d1bd048e5babf5b10710ebe0bae67db86916198a6eec434943f8b
  Source/build provenance not recorded
• nats: docker.io/library/nats:2.10-alpine@sha256:b83efabe3e7def1e0a4a31ec6e078999bb17c80363f881df35edc70fcb6bb927
  Source/build provenance not recorded
• redis: docker.io/library/redis:7.4-alpine@sha256:ff02b58f971e7d7d156a1267e283fcbbeee91773b6aa36c49dac28ecfe28eadf
  Source/build provenance not recorded
• sglang: docker.io/lmsysorg/sglang:v0.5.17@sha256:3be8803490a8b899a44f7ab2e22d8f6a1fb877cab52faeb400769a1555317db4
  Source/build provenance not recorded
• bifrost: docker.io/maximhq/bifrost:v1.5.7-amd64@sha256:410de09d5b96b15bda204dc73ac4a7d1712d598d0b9e67a8e8917b964968ab79
  Source/build provenance not recorded
• dynamo-sglang: docker.io/ritualnetwork/dynamo-sglang:v0.5.17-dsv4@sha256:9cc856356107915be6028fb693c600012d570a6d736d21c622aff0a89995089e
  Source/build provenance not recorded
• rai-simple: docker.io/ritualnetwork/rai-simple:pt1-release-01-8bd2899@sha256:de96e7daac3e0b07e608d189fd51442d9961798032cf03c7bbbe4a3f7bd5c643
  Source/build provenance not recorded
• ritual-moai-gpu: docker.io/ritualnetwork/ritual-moai-gpu:8102c66-amd64@sha256:63113fdf2bf457b84e53e325611e65f488783cd2e39f24ff65d1bf1e5c308543
  Source/build provenance not recorded
• ritual-vllm-proxy-go: docker.io/ritualnetwork/ritual-vllm-proxy-go:cf17d76-0e65f02-amd64@sha256:d9e9bdd3adde92f5b2217981782bb8201a5be4bb6057a4fb044d32d1e0f92dbd
  ritual-net/ritual-vllm-proxy-go@0e65f027af2d48bd31c5d48313e970a46b931611 · GitHub attestation run
• image_10: docker.io/ritualnetwork/sglang:v0.5.17-diffusion-accelerate1.14.0-tdx1@sha256:9fbe8c2f36ed46b3ec042b81de80d1eba344517decae86d781f20dc4b88cdfd5
  Source/build provenance not recorded
• etcd: quay.io/coreos/etcd:v3.5.16@sha256:d967d98a12dc220a1a290794711dba7eba04b8ce465e12b02383d1bfbb33e159
  Source/build provenance not recorded

dstack OS release source

Dstack-TEE/meta-dstack@e3655d1390feee3736476f4bda35c4354b4a12fc · dstack reproduce script referenced by the platform profile

Platform profile

This workload records the opaque ID of the authority-accepted platform profile required to verify it.

pp_b06ed351-1955-476e-885a-883c5d9086d1
Download platform-profile bundle (.tar)

Follow the links from the workload ID to its measurements, images, and source code. Expand a check to see what it establishes and inspect the supporting evidence.

published dstack OVMF firmware                    → MRTD
VM launch configuration + captured OVMF events    → RTMR0
published Linux kernel image                       → RTMR1
kernel command line + initrd + rootfs hash         → RTMR2
offline-approved VMM launch-field profile           → MRCONFIGID format, XFAM, TDATTRIBUTES

verified OS image + Docker configuration + TDX boot profile
                              ↓
                   pinned measurement tool
                              ↓
              recomputed eight-field TDX tuple
                              ↓
          canonical output digest matches manifest