django-api-opentelemetry 0.1.0__tar.gz

django_api_opentelemetry-0.1.0/PKG-INFO

@@ -0,0 +1,298 @@
+Metadata-Version: 2.3
+Name: django-api-opentelemetry
+Version: 0.1.0
+Summary: OpenTelemetry traces, metrics and exception logs for Django and Django Ninja
+Classifier: Development Status :: 3 - Alpha
+Classifier: Framework :: Django
+Classifier: Framework :: Django :: 5.2
+Classifier: Framework :: Django :: 6.0
+Classifier: Framework :: Django :: 6.1
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
+Requires-Dist: django>=5.2,<6.2
+Requires-Dist: opentelemetry-api>=1.44,<2
+Requires-Dist: django-ninja>=1.7,<1.8 ; extra == 'ninja'
+Requires-Dist: opentelemetry-exporter-otlp-proto-http>=1.44,<2 ; extra == 'otlp'
+Requires-Dist: opentelemetry-sdk>=1.44,<2 ; extra == 'otlp'
+Requires-Python: >=3.12
+Provides-Extra: ninja
+Provides-Extra: otlp
+Description-Content-Type: text/markdown
+
+# django-api-opentelemetry
+
+OpenTelemetry traces, metrics and exception logs for Django, with deeper
+instrumentation for Django Ninja APIs.
+
+> Tagged releases are validated in CI, then published to PyPI by the project's
+> GitHub Actions workflow.
+
+## Features
+
+- One installed-app entry adds HTTP server spans and the
+  `http.server.request.duration` and `http.server.active_requests` metrics for
+  Django requests, using resolved route templates rather than raw paths.
+- Django Ninja input-validation failures produce one warning event with the
+  validation count and route. Unexpected exceptions produce one correlated
+  error event with type, message and stack trace.
+- Ninja operation spans cover input resolution, the bound endpoint and response
+  serialization.
+- OTLP/HTTP export is opt-in and separate from recording. Existing tracer, meter
+  and logger providers remain application-owned.
+- The standard OpenTelemetry Django instrumentation is detected so the package
+  does not create a second HTTP span or HTTP metric.
+
+The package does **not** forward arbitrary Python application logs and does not
+create an access log for every successful request. Its automatic log events are
+limited to validation warnings and unexpected request exceptions.
+
+## Installation
+
+The base package supports plain Django without Ninja or the SDK/exporters:
+
+```bash
+pip install django-api-opentelemetry
+```
+
+Add the optional extras when the application uses Django Ninja or the bundled
+OTLP export setup:
+
+```bash
+pip install "django-api-opentelemetry[ninja,otlp]"
+```
+
+Supported versions:
+
+- Python 3.12–3.14
+- Django 5.2–6.1 (`>=5.2,<6.2`)
+- Django Ninja 1.7 (`>=1.7,<1.8`, optional)
+- OpenTelemetry API/SDK 1.44–1.x (`>=1.44,<2`)
+
+## Quick start
+
+Add the app. The installed app prepends its middleware when it is not already
+present, registers Django's exception signal and activates supported Ninja
+hooks:
+
+```python
+INSTALLED_APPS = [
+    # Existing applications...
+    "django_api_opentelemetry",
+]
+```
+
+Configuration is optional:
+
+```python
+DJANGO_NINJA_OTEL = {
+    "exclude": lambda request: request.path_info == "/health",
+}
+```
+
+For reliable export flushing on ASGI servers, wrap the Django ASGI application:
+
+```python
+# asgi.py
+from django.core.asgi import get_asgi_application
+
+from django_api_opentelemetry import instrument_asgi
+
+application = instrument_asgi(get_asgi_application())
+```
+
+`instrument_asgi()` adds a minimal lifespan wrapper. Native Django ASGI
+applications do not implement lifespan, so the wrapper acknowledges startup
+without initializing exporters, forwards HTTP and WebSocket scopes unchanged,
+and flushes package-owned telemetry when the server sends lifespan shutdown.
+
+The installed app still activates HTTP telemetry and Ninja hooks on its own; the
+wrapper only adds a reliable shutdown path. This is recommended because Uvicorn
+0.54 re-raises `SIGTERM` after draining the server, which bypasses `atexit`.
+For other worker types, call the synchronous `shutdown()` from a worker-exit hook
+after requests have drained:
+
+```python
+from django_api_opentelemetry import shutdown
+
+shutdown()
+```
+
+`shutdown()` waits for a bounded time, is safe to call more than once, and only
+shuts down resources owned by this package. It does not intercept signals, and
+the package cannot guarantee a flush on every possible server-termination path.
+Application-owned providers are never closed.
+
+## Configuration
+
+All keys are optional. Invalid key names or value types raise
+`django.core.exceptions.ImproperlyConfigured`; an invalid
+`DJANGO_NINJA_OTEL_AUTO_CONFIGURE` value raises the same exception.
+
+| Key | Default | Meaning |
+| --- | --- | --- |
+| `tracing` | `True` | Enable package-created spans, including operation spans. |
+| `metrics` | `True` | Enable HTTP measurements. |
+| `logs` | `True` | Emit automatic validation and exception log events. |
+| `operation_spans` | `True` | Add Ninja operation spans when tracing is enabled. |
+| `http_instrumentation` | `True` | Allow package-owned HTTP server spans and metrics. Set `False` when another wrapper owns them. |
+| `auto_configure` | Environment value, otherwise `False` | Configure OTLP/HTTP exporters. |
+| `exclude` | `None` | Callable receiving `HttpRequest`; true suppresses all package telemetry for that request. |
+| `tracer_provider` | `None` | Explicit tracer provider, otherwise the selected global provider. |
+| `meter_provider` | `None` | Explicit meter provider, otherwise the selected global provider. |
+| `logger_provider` | `None` | Explicit logger provider, otherwise the selected global provider. |
+
+`DJANGO_NINJA_OTEL_AUTO_CONFIGURE=true` enables OTLP exporter setup unless the
+Django setting sets `auto_configure` explicitly. Export setup is deliberately
+opt-in: recording telemetry and deciding where it is sent are separate concerns,
+matching the upstream FastAPI 0.143 change.
+
+```bash
+export DJANGO_NINJA_OTEL_AUTO_CONFIGURE=true
+export OTEL_SERVICE_NAME=my-api
+export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4318
+```
+
+Inside Compose, use the collector service address
+`http://collector:4318` instead of `localhost`.
+
+Exporter behavior:
+
+- With no endpoint configured, the package installs no OTLP exporter and does
+  not guess a default endpoint.
+- A base endpoint gains the signal path (`/v1/traces`, `/v1/metrics`,
+  `/v1/logs`); a signal-specific endpoint is used as written.
+- Only the OTLP HTTP/protobuf protocol is supported.
+- `OTEL_TRACES_EXPORTER`, `OTEL_METRICS_EXPORTER` or `OTEL_LOGS_EXPORTER` set to
+  `none` disables that signal; other non-OTLP values are skipped with a warning.
+- `OTEL_SDK_DISABLED=true` disables package telemetry.
+- Standard endpoint, header, timeout and resource environment variables are
+  delegated to the official OpenTelemetry exporters.
+
+### Providers and coexistence
+
+Explicit providers in `DJANGO_NINJA_OTEL` take precedence. Otherwise the package
+uses the global providers. With `auto_configure` enabled, it adds processors,
+readers or exporters to an existing provider when that provider supports
+registration, or installs an SDK provider when the global provider is still
+unconfigured. Only package-owned components and providers are flushed and shut
+down.
+
+If the standard OpenTelemetry Django middleware is installed, the package detects
+it and leaves HTTP spans and metrics to that integration while retaining Ninja
+operation spans, validation warnings and exception events. For other HTTP
+wrappers, set `http_instrumentation=False`; do not assume detection of every
+vendor wrapper.
+
+## Recording boundaries and privacy
+
+With the default prepended middleware position, a buffered response is finished
+after response middleware completes. For streaming responses, it is finished at
+iterator exhaustion, failure or close. These are Django-observable boundaries,
+not FastAPI's ASGI `send` boundary.
+
+Validation warnings contain only `django_ninja.validation.error_count` and the
+resolved `http.route`; invalid input is not serialized into the event. Unexpected
+exception events can contain the exception message and stack trace. Redact
+sensitive data at the OpenTelemetry processor/exporter layer or set
+`logs=False` if those values must not leave the process. Error logs are emitted
+independently of trace sampling.
+
+## Development in Docker
+
+Docker Engine/Desktop with Compose v2 and BuildKit is the only host requirement.
+No host Python, `uv`, Ruff or ty installation is needed. Dependencies live in
+`/opt/venv`; the checkout is mounted at `/workspace`.
+
+Build the development image with host-matching ownership:
+
+```bash
+docker compose build \
+  --build-arg LOCAL_UID="$(id -u)" \
+  --build-arg LOCAL_GID="$(id -g)" dev
+```
+
+Start the demo server on `127.0.0.1:8000` only:
+
+```bash
+docker compose up dev
+```
+
+The demo command selects `tests.project.demo_settings` and runs the shared ASGI
+project with reload enabled. `tests.project.asgi` already wraps the Django
+application with `instrument_asgi(get_asgi_application())`, so the demo exercises
+the lifespan shutdown path. Ordinary pytest runs use the baseline
+`tests.project.settings`, which deliberately keeps the package out of
+`INSTALLED_APPS`, so plain-Django tests remain a real condition. The demo
+settings are not applied to pytest.
+
+Run tests and checks:
+
+```bash
+docker compose run --rm dev python -m pytest
+docker compose run --rm dev bash scripts/check.sh all
+docker compose run --rm dev python -m ruff check .
+docker compose run --rm dev python -m ruff format --check .
+docker compose run --rm dev ty check --python /opt/venv src
+```
+
+`scripts/check.sh all` runs Ruff lint, Ruff format check, ty and pytest. The image
+is built with a locked `uv sync` for the `ninja`/`otlp` extras and development
+group. Rebuild after changing `pyproject.toml` or `uv.lock`; normal test commands
+do not resync dependencies. Compatibility jobs build isolated environments with
+`scripts/compat.sh` rather than modifying the committed lockfile. Final local
+verification passed Ruff, format, ty and 227 tests in the default environment,
+and the nine Python/Django matrix jobs plus four compatibility jobs. See
+[task status](docs/tasks.md) for the evidence.
+
+Build and verify distributions:
+
+```bash
+docker build --target dist --output dist .
+docker compose run --rm dev python scripts/check-dist.py --self-test
+docker compose run --rm dev python scripts/check-dist.py dist
+```
+
+The `dist` target exports exactly the wheel and sdist built inside the image.
+`check-dist.py` runs `twine check`, installs the wheel into a clean temporary
+environment outside `/workspace`, verifies the import resolves to the installed
+distribution, rebuilds a wheel from the sdist and repeats the import check. The
+directory must contain exactly one wheel and one sdist; remove stale artifacts
+first.
+
+Run the collector smoke test against the built wheel:
+
+```bash
+docker compose --profile telemetry down -v
+docker compose --profile telemetry up -d --wait collector
+docker compose --profile telemetry run --rm smoke
+docker compose --profile telemetry down -v
+```
+
+The `telemetry` profile runs a one-shot volume initializer, then a non-root
+collector writing all three signals to a shared volume. The smoke service sees
+that output read-only, installs the wheel from `./dist` into a fresh temporary
+virtual environment, starts Uvicorn and requests a success route, a Ninja
+validation failure and an uncaught error. It asserts all three signals, trace/log
+correlation, validation-input omission and both HTTP metrics.
+
+The smoke check requires a fresh collector volume. If it reports stale output,
+run `docker compose --profile telemetry down -v` before starting the collector
+again. A repeated smoke run without that reset is expected to fail rather than
+reuse old records. Collector diagnostics are available with:
+
+```bash
+docker compose --profile telemetry logs collector
+```
+
+See [development and CI](docs/development-and-ci.md) for the full container,
+compatibility-matrix and release workflow, and [task status](docs/tasks.md) for
+current verification evidence.
+
+## Deliberate limits
+
+The first release does not include DRF, Channels/WebSockets, background-job
+instrumentation, automatic forwarding of application logs, per-chunk spans or a
+dashboard stack. Ninja hooks require Ninja 1.7 and are disabled with one
+diagnostic on an unsupported version while ordinary Django telemetry continues.

django_api_opentelemetry-0.1.0/README.md

@@ -0,0 +1,275 @@
+# django-api-opentelemetry
+
+OpenTelemetry traces, metrics and exception logs for Django, with deeper
+instrumentation for Django Ninja APIs.
+
+> Tagged releases are validated in CI, then published to PyPI by the project's
+> GitHub Actions workflow.
+
+## Features
+
+- One installed-app entry adds HTTP server spans and the
+  `http.server.request.duration` and `http.server.active_requests` metrics for
+  Django requests, using resolved route templates rather than raw paths.
+- Django Ninja input-validation failures produce one warning event with the
+  validation count and route. Unexpected exceptions produce one correlated
+  error event with type, message and stack trace.
+- Ninja operation spans cover input resolution, the bound endpoint and response
+  serialization.
+- OTLP/HTTP export is opt-in and separate from recording. Existing tracer, meter
+  and logger providers remain application-owned.
+- The standard OpenTelemetry Django instrumentation is detected so the package
+  does not create a second HTTP span or HTTP metric.
+
+The package does **not** forward arbitrary Python application logs and does not
+create an access log for every successful request. Its automatic log events are
+limited to validation warnings and unexpected request exceptions.
+
+## Installation
+
+The base package supports plain Django without Ninja or the SDK/exporters:
+
+```bash
+pip install django-api-opentelemetry
+```
+
+Add the optional extras when the application uses Django Ninja or the bundled
+OTLP export setup:
+
+```bash
+pip install "django-api-opentelemetry[ninja,otlp]"
+```
+
+Supported versions:
+
+- Python 3.12–3.14
+- Django 5.2–6.1 (`>=5.2,<6.2`)
+- Django Ninja 1.7 (`>=1.7,<1.8`, optional)
+- OpenTelemetry API/SDK 1.44–1.x (`>=1.44,<2`)
+
+## Quick start
+
+Add the app. The installed app prepends its middleware when it is not already
+present, registers Django's exception signal and activates supported Ninja
+hooks:
+
+```python
+INSTALLED_APPS = [
+    # Existing applications...
+    "django_api_opentelemetry",
+]
+```
+
+Configuration is optional:
+
+```python
+DJANGO_NINJA_OTEL = {
+    "exclude": lambda request: request.path_info == "/health",
+}
+```
+
+For reliable export flushing on ASGI servers, wrap the Django ASGI application:
+
+```python
+# asgi.py
+from django.core.asgi import get_asgi_application
+
+from django_api_opentelemetry import instrument_asgi
+
+application = instrument_asgi(get_asgi_application())
+```
+
+`instrument_asgi()` adds a minimal lifespan wrapper. Native Django ASGI
+applications do not implement lifespan, so the wrapper acknowledges startup
+without initializing exporters, forwards HTTP and WebSocket scopes unchanged,
+and flushes package-owned telemetry when the server sends lifespan shutdown.
+
+The installed app still activates HTTP telemetry and Ninja hooks on its own; the
+wrapper only adds a reliable shutdown path. This is recommended because Uvicorn
+0.54 re-raises `SIGTERM` after draining the server, which bypasses `atexit`.
+For other worker types, call the synchronous `shutdown()` from a worker-exit hook
+after requests have drained:
+
+```python
+from django_api_opentelemetry import shutdown
+
+shutdown()
+```
+
+`shutdown()` waits for a bounded time, is safe to call more than once, and only
+shuts down resources owned by this package. It does not intercept signals, and
+the package cannot guarantee a flush on every possible server-termination path.
+Application-owned providers are never closed.
+
+## Configuration
+
+All keys are optional. Invalid key names or value types raise
+`django.core.exceptions.ImproperlyConfigured`; an invalid
+`DJANGO_NINJA_OTEL_AUTO_CONFIGURE` value raises the same exception.
+
+| Key | Default | Meaning |
+| --- | --- | --- |
+| `tracing` | `True` | Enable package-created spans, including operation spans. |
+| `metrics` | `True` | Enable HTTP measurements. |
+| `logs` | `True` | Emit automatic validation and exception log events. |
+| `operation_spans` | `True` | Add Ninja operation spans when tracing is enabled. |
+| `http_instrumentation` | `True` | Allow package-owned HTTP server spans and metrics. Set `False` when another wrapper owns them. |
+| `auto_configure` | Environment value, otherwise `False` | Configure OTLP/HTTP exporters. |
+| `exclude` | `None` | Callable receiving `HttpRequest`; true suppresses all package telemetry for that request. |
+| `tracer_provider` | `None` | Explicit tracer provider, otherwise the selected global provider. |
+| `meter_provider` | `None` | Explicit meter provider, otherwise the selected global provider. |
+| `logger_provider` | `None` | Explicit logger provider, otherwise the selected global provider. |
+
+`DJANGO_NINJA_OTEL_AUTO_CONFIGURE=true` enables OTLP exporter setup unless the
+Django setting sets `auto_configure` explicitly. Export setup is deliberately
+opt-in: recording telemetry and deciding where it is sent are separate concerns,
+matching the upstream FastAPI 0.143 change.
+
+```bash
+export DJANGO_NINJA_OTEL_AUTO_CONFIGURE=true
+export OTEL_SERVICE_NAME=my-api
+export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4318
+```
+
+Inside Compose, use the collector service address
+`http://collector:4318` instead of `localhost`.
+
+Exporter behavior:
+
+- With no endpoint configured, the package installs no OTLP exporter and does
+  not guess a default endpoint.
+- A base endpoint gains the signal path (`/v1/traces`, `/v1/metrics`,
+  `/v1/logs`); a signal-specific endpoint is used as written.
+- Only the OTLP HTTP/protobuf protocol is supported.
+- `OTEL_TRACES_EXPORTER`, `OTEL_METRICS_EXPORTER` or `OTEL_LOGS_EXPORTER` set to
+  `none` disables that signal; other non-OTLP values are skipped with a warning.
+- `OTEL_SDK_DISABLED=true` disables package telemetry.
+- Standard endpoint, header, timeout and resource environment variables are
+  delegated to the official OpenTelemetry exporters.
+
+### Providers and coexistence
+
+Explicit providers in `DJANGO_NINJA_OTEL` take precedence. Otherwise the package
+uses the global providers. With `auto_configure` enabled, it adds processors,
+readers or exporters to an existing provider when that provider supports
+registration, or installs an SDK provider when the global provider is still
+unconfigured. Only package-owned components and providers are flushed and shut
+down.
+
+If the standard OpenTelemetry Django middleware is installed, the package detects
+it and leaves HTTP spans and metrics to that integration while retaining Ninja
+operation spans, validation warnings and exception events. For other HTTP
+wrappers, set `http_instrumentation=False`; do not assume detection of every
+vendor wrapper.
+
+## Recording boundaries and privacy
+
+With the default prepended middleware position, a buffered response is finished
+after response middleware completes. For streaming responses, it is finished at
+iterator exhaustion, failure or close. These are Django-observable boundaries,
+not FastAPI's ASGI `send` boundary.
+
+Validation warnings contain only `django_ninja.validation.error_count` and the
+resolved `http.route`; invalid input is not serialized into the event. Unexpected
+exception events can contain the exception message and stack trace. Redact
+sensitive data at the OpenTelemetry processor/exporter layer or set
+`logs=False` if those values must not leave the process. Error logs are emitted
+independently of trace sampling.
+
+## Development in Docker
+
+Docker Engine/Desktop with Compose v2 and BuildKit is the only host requirement.
+No host Python, `uv`, Ruff or ty installation is needed. Dependencies live in
+`/opt/venv`; the checkout is mounted at `/workspace`.
+
+Build the development image with host-matching ownership:
+
+```bash
+docker compose build \
+  --build-arg LOCAL_UID="$(id -u)" \
+  --build-arg LOCAL_GID="$(id -g)" dev
+```
+
+Start the demo server on `127.0.0.1:8000` only:
+
+```bash
+docker compose up dev
+```
+
+The demo command selects `tests.project.demo_settings` and runs the shared ASGI
+project with reload enabled. `tests.project.asgi` already wraps the Django
+application with `instrument_asgi(get_asgi_application())`, so the demo exercises
+the lifespan shutdown path. Ordinary pytest runs use the baseline
+`tests.project.settings`, which deliberately keeps the package out of
+`INSTALLED_APPS`, so plain-Django tests remain a real condition. The demo
+settings are not applied to pytest.
+
+Run tests and checks:
+
+```bash
+docker compose run --rm dev python -m pytest
+docker compose run --rm dev bash scripts/check.sh all
+docker compose run --rm dev python -m ruff check .
+docker compose run --rm dev python -m ruff format --check .
+docker compose run --rm dev ty check --python /opt/venv src
+```
+
+`scripts/check.sh all` runs Ruff lint, Ruff format check, ty and pytest. The image
+is built with a locked `uv sync` for the `ninja`/`otlp` extras and development
+group. Rebuild after changing `pyproject.toml` or `uv.lock`; normal test commands
+do not resync dependencies. Compatibility jobs build isolated environments with
+`scripts/compat.sh` rather than modifying the committed lockfile. Final local
+verification passed Ruff, format, ty and 227 tests in the default environment,
+and the nine Python/Django matrix jobs plus four compatibility jobs. See
+[task status](docs/tasks.md) for the evidence.
+
+Build and verify distributions:
+
+```bash
+docker build --target dist --output dist .
+docker compose run --rm dev python scripts/check-dist.py --self-test
+docker compose run --rm dev python scripts/check-dist.py dist
+```
+
+The `dist` target exports exactly the wheel and sdist built inside the image.
+`check-dist.py` runs `twine check`, installs the wheel into a clean temporary
+environment outside `/workspace`, verifies the import resolves to the installed
+distribution, rebuilds a wheel from the sdist and repeats the import check. The
+directory must contain exactly one wheel and one sdist; remove stale artifacts
+first.
+
+Run the collector smoke test against the built wheel:
+
+```bash
+docker compose --profile telemetry down -v
+docker compose --profile telemetry up -d --wait collector
+docker compose --profile telemetry run --rm smoke
+docker compose --profile telemetry down -v
+```
+
+The `telemetry` profile runs a one-shot volume initializer, then a non-root
+collector writing all three signals to a shared volume. The smoke service sees
+that output read-only, installs the wheel from `./dist` into a fresh temporary
+virtual environment, starts Uvicorn and requests a success route, a Ninja
+validation failure and an uncaught error. It asserts all three signals, trace/log
+correlation, validation-input omission and both HTTP metrics.
+
+The smoke check requires a fresh collector volume. If it reports stale output,
+run `docker compose --profile telemetry down -v` before starting the collector
+again. A repeated smoke run without that reset is expected to fail rather than
+reuse old records. Collector diagnostics are available with:
+
+```bash
+docker compose --profile telemetry logs collector
+```
+
+See [development and CI](docs/development-and-ci.md) for the full container,
+compatibility-matrix and release workflow, and [task status](docs/tasks.md) for
+current verification evidence.
+
+## Deliberate limits
+
+The first release does not include DRF, Channels/WebSockets, background-job
+instrumentation, automatic forwarding of application logs, per-chunk spans or a
+dashboard stack. Ninja hooks require Ninja 1.7 and are disabled with one
+diagnostic on an unsupported version while ordinary Django telemetry continues.

django_api_opentelemetry-0.1.0/pyproject.toml

@@ -0,0 +1,71 @@
+[build-system]
+requires = ["uv_build>=0.13,<0.14"]
+build-backend = "uv_build"
+
+[project]
+name = "django-api-opentelemetry"
+version = "0.1.0"
+description = "OpenTelemetry traces, metrics and exception logs for Django and Django Ninja"
+readme = "README.md"
+requires-python = ">=3.12"
+dependencies = [
+    "django>=5.2,<6.2",
+    "opentelemetry-api>=1.44,<2",
+]
+classifiers = [
+    "Development Status :: 3 - Alpha",
+    "Framework :: Django",
+    "Framework :: Django :: 5.2",
+    "Framework :: Django :: 6.0",
+    "Framework :: Django :: 6.1",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+    "Programming Language :: Python :: 3.14",
+]
+
+[project.optional-dependencies]
+ninja = ["django-ninja>=1.7,<1.8"]
+otlp = [
+    "opentelemetry-exporter-otlp-proto-http>=1.44,<2",
+    "opentelemetry-sdk>=1.44,<2",
+]
+
+[dependency-groups]
+dev = [
+    "build>=1.2",
+    "django-stubs>=5.2",
+    "httpx>=0.27",
+    "opentelemetry-instrumentation-django[asgi]>=0.65b0",
+    "pytest>=8",
+    "pytest-asyncio>=0.24",
+    "pytest-django>=4.9",
+    "ruff>=0.9",
+    "twine>=6",
+    "ty>=0.0.1",
+    "uv_build>=0.13,<0.14",
+    "uvicorn>=0.34",
+]
+
+[tool.pytest.ini_options]
+DJANGO_SETTINGS_MODULE = "tests.project.settings"
+minversion = "8.0"
+testpaths = ["tests"]
+addopts = "-ra"
+asyncio_mode = "auto"
+
+[tool.ruff]
+target-version = "py312"
+src = [
+    "src",
+    "tests",
+]
+
+[tool.ruff.lint]
+select = [
+    "E",
+    "F",
+    "I",
+    "UP",
+    "B",
+]