Lightweight Django middleware for APM-style request profiling
Measure DB vs App time and query count with near-zero configuration.

• 🔍 DevTools visibility: See DB vs app/serialization time in Chrome DevTools via Server-Timing.
• 🚀 Zero-agent: No daemon, no SaaS — just one Django middleware.
• 🧩 Drop-in: Near-zero configuration (add middleware and go).
• 🔒 Privacy-first: Exposes timing + query counts only (no query contents stored).

Goal: make performance bottlenecks “visible” (DB vs app/serialization) without heavyweight APM.

Here's how django-xbench exposes request timing breakdown using the Server-Timing header:

Adds Server-Timing and X-Bench-Queries headers and optionally logs per-request metrics.

• ✅ Measures total request time and DB time (via connection.execute_wrapper)
• ✅ Calculates app time (= total - db)
• ✅ Counts DB queries
• ✅ Adds response headers:
  • Server-Timing: xbench-total;dur=..., xbench-db;dur=..., xbench-app;dur=...
  • X-Bench-Queries: <int>
• ✅ Optional logging:
  • [XBENCH] GET /path | xbench_total=...ms xbench_db=...ms xbench_app=...ms q=...
• ✅ Slow endpoint aggregation (in-memory, per process) + simple dashboard (experimental)
• ✅ Tested with pytest + pytest-django

pip install django-xbench

For local development (this repository):

1. Add middleware in your settings.py:

MIDDLEWARE = [
    # Recommended: place near the top to approximate end-to-end server time
    # (includes other middleware overhead).
    "django_xbench.middleware.XBenchMiddleware",

    # ... other middleware ...
]

2. Run your server and hit any endpoint:

In your project:

python manage.py runserver
curl -I http://127.0.0.1:8000/<your-endpoint>/

In this repo (demo):

# macOS / Linux
export DJANGO_SECRET_KEY="dev"
python -m examples.manage runserver --noreload
curl -I http://127.0.0.1:8000/db-heavy/

# Windows PowerShell
$env:DJANGO_SECRET_KEY="dev"
python -m examples.manage runserver --noreload
curl -I http://127.0.0.1:8000/db-heavy/

You should see headers similar to:

Server-Timing: xbench-total;dur=12.345, xbench-db;dur=1.234, xbench-app;dur=11.111
X-Bench-Queries: 3

Example:

Server-Timing: xbench-total;dur=52.300, xbench-db;dur=14.100, xbench-app;dur=38.200

• xbench-total: whole request duration
• xbench-db: total DB time measured by wrapper
• xbench-app: max(0, total - db) (serialization/template/python time etc.)

You can inspect this in Chrome DevTools → Network → Timing
(or any browser that supports the Server-Timing spec).

django-xbench supports two configuration styles.

Use a single XBENCH dictionary to keep settings compact and grouped:

XBENCH = {
    "ENABLED": True,       # default: True
    "LOG": False,          # default: False
    "LOG_LEVEL": "info",   # "info" or "debug"
    "SLOW_AGG": False,     # default: False
}

Older flat settings are still supported:

XBENCH_ENABLED = True
XBENCH_LOG_ENABLED = True
XBENCH_LOG_LEVEL = "debug"
XBENCH_SLOW_AGG_ENABLED = True

This feature keeps an in-memory rolling window of endpoint timings (per process) and shows the slowest endpoints by "damage" (total accumulated latency).

XBENCH = {"SLOW_AGG": True}

In your project's urls.py:

from django.urls import include, path

urlpatterns = [
    # ... your urls ...
    path("__xbench__/", include("django_xbench.slowagg.urls")),
]

• JSON snapshot: GET /__xbench__/slow/?n=20
• HTML dashboard: GET /__xbench__/slow/ui/?n=20

• Aggregation is in-memory per process. If you run multiple workers/processes, each one has its own rolling window.
• Intended for debugging / internal visibility, not as a full distributed APM.
• DB%: db_total / total
• Avg Q: average DB queries per request
• Damage: total accumulated latency in the window (sum of durations)

The dashboard only shows data after requests occur.

If you see "No data yet":

1. Make sure SLOW_AGG is enabled
2. Hit some endpoints (e.g. /db-heavy/)
3. Refresh the dashboard

If using Django runserver with auto-reload, aggregation resets on reload.

XBENCH = {
    "SLOW_AGG": True,
    "SLOW_BUCKET_SECONDS": 10,   # bucket size in seconds
    "SLOW_BUCKET_COUNT": 60,     # number of buckets (window = bucket_seconds * bucket_count)
    "SLOW_ENDPOINT_CAP": 200,    # max unique endpoints per bucket (overflow goes to "__other__")
}

Note: this repo includes a bundled examples/ Django project used by pytest-django. In CI, we set PYTHONPATH=examples to ensure examples.config.settings can be imported reliably.

If you want to see logs while testing:

This repository includes an examples/ Django project for manual testing.

Run it from the repository root:

# macOS / Linux
export DJANGO_SECRET_KEY="dev"
python -m examples.manage runserver --noreload

# Windows PowerShell
$env:DJANGO_SECRET_KEY="dev"
python -m examples.manage runserver --noreload

Try a few endpoints:

curl -I http://127.0.0.1:8000/db-heavy/
curl -I http://127.0.0.1:8000/app-heavy/
curl -I http://127.0.0.1:8000/admin/login/

• Python: 3.9+
• Django: 3.2+ (tested on 5.2)

Issues and PRs are welcome.
If you propose new metrics, please include:

• minimal reproducible example
• tests
• documentation update

MIT