sf-veritas 0.14.0a0

Middleware for Django, Flask, and FastAPI to intercept requests, logs, and exceptions.

SF3 Middleware - Currently transmissions are diabled

Middleware for Python (Django, Flask, and FastAPI) to intercept requests, print statements, logs, and exceptions while persisting tracing.

Installation

Use Poetry to install:

poetry add sf-veritas

Usage

Django, etc

Switch the following:

python manage.py runserver 0.0.0.0:8000

To:

sf-veritas API-KEY python manage.py runserver 0.0.0.0:8000

TODO - Rename all Sailfish artifacts to Sailfish.ai

Debug Mode

Sailfish supports a server-triggered debug mode that turns on full function-span capture (arguments, return values, timings) for a time-bounded, budget-bounded window — for live backend services that have no HTTP surface (Celery workers, Kafka consumers, cron jobs).

A daemon thread holds a WebSocket connection to the Sailfish backend. When the UI enables debug for a service identifier, the rule list is broadcast to every pod of that service simultaneously via Redis pub/sub on the backend, applied through a process-wide override on the C profiler's hot path. Capture stops when TTL expires, span budget is exhausted, byte budget is exhausted, or the UI cancels.

Disable via SF_DEBUG_MODE_DISABLED=true. See docs/docs-enterprise/debug-mode.md for the full user-facing description, rule schema, and limitations.

Network Hop Calculation Time

To evaluate the performance impact of this package, we benchmarked 1000 HTTP requests with and without the package enabled.

Configuration	Mean (ms)	Median (ms)	Std Dev (ms)
✅ With Package	79.12	54.00	111.18
❌ Without Package	69.70	52.00	73.78

⚠️ Note: The package introduces a slight increase in mean response time and variance. This trade-off may be acceptable depending on the value the package provides (e.g., additional logging, monitoring, or security features).

---

Optimized Entrypoint Capture (Post-Refactor)

After optimizing how the user-code entrypoint is captured (via faster stack inspection), we observed improved stability and performance across 1015 analyzed requests:

Configuration	Mean (ms)	Median (ms)	Std Dev (ms)
✅ With Package	142.45	138.50	80.78
❌ Without Package	131.07	127.00	35.75

⚠️ The optimized implementation added a slight increase in mean latency (~8.7%), but this tradeoff is offset by improved accuracy of entrypoint capture.

---