Structured Logging¶

File: app/core/logging.py

All log output is JSON, written to stdout. Configured once at startup via setup_logging().

Base format¶

Every log line:

{
  "timestamp": "2026-04-27T15:00:00+00:00",
  "level": "INFO",
  "message": "inference_success"
}

Inference success fields¶

{
  "timestamp": "...",
  "level": "INFO",
  "message": "inference_success",
  "request_id": "550e8400-...",
  "job_id": "7f3e1a2b-...",
  "model": "echo",
  "version": "v1",
  "tenant_id": "tenant_dev",
  "latency_ms": 3.2
}

Request ID propagation¶

Every request gets an X-Request-ID: - Client sends one → echoed back in response header and included in logs. - Client omits it → UUID generated, same propagation.

Use this to correlate a client request → server log → job record.

Shipping logs¶

Pipe stdout to your log collector. The JSON format is already structured — no additional parsing needed for Datadog, Loki, CloudWatch, etc.