Skip to content

v0.4.0: Docker Compose Stack, Health Probes & Unified Async

Choose a tag to compare

@zjykzj zjykzj released this 22 Aug 06:16
· 45 commits to main since this release

Added

  • Health probe endpoints: GET /health (liveness, always 200) and GET /health/ready (readiness, 503 + code 6 until the predictor is loaded) β€” the only endpoints that carry meaning in the HTTP status, for orchestrator probes
  • Docker Compose full stack: Dockerfile + docker-compose.yml (web + worker + RabbitMQ + Redis) β€” one-command containerized startup; the model is bind-mounted from ./models, never baked into the image

Changed

  • Async is one deployment shape: INFERFORGE_ASYNC=1 registers both async apis (callback + query, requires celery + rabbitmq + redis) β€” callback vs query is now a per-request choice, not a deployment choice (breaking)
  • INFERFORGE_QUERY=1 kept as a deprecated alias with a startup warning; requirements-query.txt removed, redis merged into requirements-async.txt (breaking)

See CHANGELOG.md for the full change history.