Skip to content

runbook.md

Latest commit

 

History

History
128 lines (88 loc) · 2.65 KB

runbook.md

File metadata and controls

128 lines (88 loc) · 2.65 KB
Raw

Runbook

This runbook collects the most common commands and troubleshooting steps for aerotrack.

Core commands

Start the stack:

docker-compose up --build

Stop the stack:

docker-compose down

Check service status:

docker-compose ps

Tail logs:

docker-compose logs -f

Health check:

curl http://localhost:8000/health

Metadata check:

curl http://localhost:8000/metadata

Common demo commands

Single-frame detection:

curl -X POST "http://localhost:8000/detect" \
  -H "accept: application/json" \
  -F "file=@/absolute/path/to/frame.jpg"

Clip-level tracking:

curl -X POST "http://localhost:8000/track" \
  -H "accept: application/json" \
  -F "file=@/absolute/path/to/clip.mp4"

Generate a smoke-test clip:

python scripts/make_smoke_clip.py \
  --image data/raw/VisDrone2019-DET-val/images/0000271_01401_d_0000380.jpg \
  --output outputs/smoke.mp4

Generate the smoke-test clip from the API container:

docker-compose exec api python scripts/make_smoke_clip.py \
  --image data/raw/VisDrone2019-DET-val/images/0000271_01401_d_0000380.jpg \
  --output outputs/smoke.mp4

Training commands

Local validation run inside Docker:

docker-compose exec api python -m src.train \
  --data data/visdrone/VisDrone.yaml \
  --epochs 1 \
  --imgsz 640 \
  --batch 2 \
  --mlflow-tracking-uri http://mlflow:5000

Higher-cost run:

docker-compose exec api python -m src.train \
  --data data/visdrone/VisDrone.yaml \
  --epochs 50 \
  --imgsz 1024 \
  --batch 8 \
  --mlflow-tracking-uri http://mlflow:5000

Troubleshooting

Docker build is huge or slow

  • Confirm .dockerignore exists
  • Make sure data/ is not being sent into the image build context

MLflow does not start

  • Check whether the host-side port is free
  • Confirm MLFLOW_BACKEND_STORE_URI uses an absolute SQLite path like sqlite:////mlflow/mlflow.db

API starts but /detect is slow on first request

  • That is expected if the base YOLO weights need to download on first use
  • For demos, preload the weights or point AEROTRACK_MODEL_PATH to a local model file

Training dies with exit code 137

  • That usually indicates the process was killed under memory pressure
  • Reduce --imgsz and --batch
  • Prefer GPU-backed or higher-memory machines for the full default training profile

Training cannot find the dataset in Docker

  • The runtime training code rewrites the dataset YAML path to the local/container path when needed
  • Confirm data/visdrone/VisDrone.yaml exists and the images / labels directories are populated