Skip to content

configuration.md

Latest commit

 

History

History
121 lines (96 loc) · 5.83 KB

configuration.md

File metadata and controls

121 lines (96 loc) · 5.83 KB

Configuration

← Back to README

Settings are read from environment variables, with optional fallback to a .env file in the repository root. pydantic-settings is used when available; a lightweight dotenv parser handles the case where dependencies are not yet installed.

Safe Local .env

HOST=127.0.0.1
PORT=8765
LOG_LEVEL=INFO
DEBUG=false

REDIS_HOST=127.0.0.1
REDIS_PORT=6379
REDIS_DB=0
CACHE_TTL_SECONDS=3600
CACHE_MAX_ENTRIES=256

DEPTHLENS_PRELOAD_MODEL=false
DEPTHLENS_WARMUP_MODEL=MiDaS_small
DEPTHLENS_WARMUP_DEVICE=auto
DEPTHLENS_MAX_DIM=1536
DEPTHLENS_DEFAULT_METRICS=fast
DEPTHLENS_DEFAULT_OUTPUTS=color
DEPTHLENS_OBSERVABILITY_ENABLED=true
DEPTHLENS_PROMETHEUS_ENABLED=true
DEPTHLENS_TELEMETRY_MAX_EVENTS=200
DEPTHLENS_TRACE_HISTORY_LIMIT=200
DEPTHLENS_CRASH_HISTORY_LIMIT=100
DEPTHLENS_BENCHMARK_HISTORY_LIMIT=50
DEPTHLENS_TRACE_SAMPLE_RATE=1.0

Server

Variable Default Description
HOST 127.0.0.1 locally, 0.0.0.0 in Docker ASGI bind host
PORT 8765 ASGI port
LOG_LEVEL INFO DEBUG, INFO, WARNING, ERROR, or CRITICAL
DEBUG false Enables FastAPI debug mode (detailed error responses)
WEB_CONCURRENCY 1 Uvicorn worker count (Docker only; use 1 for single-GPU inference)

Cache

Variable Default Description
REDIS_URL unset Full Redis URL override (takes precedence over individual fields)
REDIS_HOST 127.0.0.1 Redis host
REDIS_PORT 6379 Redis port
REDIS_DB 0 Redis logical database
REDIS_PASSWORD unset Optional Redis password
REDIS_SOCKET_TIMEOUT_SECONDS 1.5 Connect/read timeout; keep low to fail fast and fall back to in-memory
REDIS_MAX_CONNECTIONS 20 Connection pool maximum
CACHE_TTL_SECONDS 3600 Cache entry lifetime
CACHE_MAX_ENTRIES 256 In-memory LRU entry cap

Inference

Variable Default Description
DEPTHLENS_PRELOAD_MODEL false Warm a model in the background after startup
DEPTHLENS_WARMUP_MODEL MiDaS_small Model to pre-warm when preload is enabled
DEPTHLENS_WARMUP_DEVICE auto Device to pre-warm on
DEPTHLENS_SKIP_WARMUP unset Set to 1 to skip warmup (used in CI/testing)
DEPTHLENS_MAX_DIM 1536 Maximum long image edge before down-scaling
DEPTHLENS_DEFAULT_METRICS fast none, fast, or full
DEPTHLENS_DEFAULT_OUTPUTS color color, gray, or color,gray
INFERENCE_MAX_CONCURRENCY 2 Max concurrent inference operations (asyncio semaphore)
DEPTHLENS_OBSERVABILITY_ENABLED true Enable local observability snapshots and instrumentation
DEPTHLENS_PROMETHEUS_ENABLED true Enable /metrics Prometheus exposition when prometheus-client is available
DEPTHLENS_TELEMETRY_MAX_EVENTS 200 Bounded recent HTTP/inference event history size
DEPTHLENS_TRACE_HISTORY_LIMIT 200 Bounded trace span history size
DEPTHLENS_CRASH_HISTORY_LIMIT 100 Bounded sanitized crash history size
DEPTHLENS_BENCHMARK_HISTORY_LIMIT 50 Bounded benchmark history size
DEPTHLENS_TRACE_SAMPLE_RATE 1.0 Trace sampling ratio from 0.0 to 1.0
ORT_INTRA_OP_NUM_THREADS CPU-dependent / Docker 2 ONNX Runtime intra-op thread pool
ORT_INTER_OP_NUM_THREADS 1 ONNX Runtime inter-op thread pool

Paths

Variable Default Description
DEPTHLENS_BACKEND_PORT 8765 Electron backend port hint (read before spawning uvicorn)
DEPTHLENSPRO_MODEL_DIR unset Custom model directory; ONNX files searched in {dir}/onnx/
DEPTHLENS_ONNX_DIR unset Direct ONNX directory override
ONNX_WEIGHTS_DIR unset Legacy ONNX directory (lowest priority)
DEPTHLENS_AUTO_EXPORT_ONNX false Auto-export ONNX during benchmark when weights are missing

Observability Privacy

Telemetry is local-only: DepthLens Pro does not send analytics to cloud services or external telemetry endpoints. Histories are bounded in process memory, Prometheus labels intentionally avoid high-cardinality user data, and telemetry avoids raw images, base64 payloads, uploaded filenames, image hashes, cache keys, local full paths, and private exception details.

CI / Test Flags

Variable Purpose
TESTING=1 Lightweight test mode; skips warmup and model downloads
CI=1 CI marker used by test fixtures
CODEX_ENV=1 Automation/sandboxed environment marker
DEPTHLENS_DISABLE_MODEL_DOWNLOADS=1 Prevents torch.hub from downloading weights (used in offline CI)

Security and timeout settings

Variable Default Description
DEPTHLENS_ROUTE_TIMEOUT_SECONDS 300 Generous timeout for estimate, compare, detect, and reconstruct route work. Timeouts return REQUEST_TIMEOUT except /reconstruct, which returns RECONSTRUCTION_TIMEOUT.
DEPTHLENS_BATCH_ITEM_TIMEOUT_SECONDS 300 Per-item timeout for batch processing so one slow image produces a structured per-item REQUEST_TIMEOUT without altering the batch response contract.
DEPTHLENS_CORS_ALLOWED_ORIGINS empty Comma-separated additional browser origins allowed by CORS. Localhost/127.0.0.1/file-null flows are allowed by default.
DEPTHLENS_CORS_ALLOW_ALL false Development-only wildcard CORS escape hatch; credentials remain disabled.

Runtime installs use backend/requirements.txt. Development and CI quality checks use backend/requirements-dev.txt, which includes runtime dependencies plus Black, Ruff, mypy, and pytest.

See resource path contract for packaged/development resource names shared by backend, Electron, and verification scripts.