sentry-python - Sentry SDK for Python

Sentry-Python is an experimental SDK for Sentry. For a stable one, use raven.

Getting started with the new Sentry SDK for Python

Install this package with pip install sentry-sdk. Then, in your code:

import sentry_sdk
sentry_sdk.init(dsn="https://foo@sentry.io/123")

After initialization, you can capture exceptions like this:

sentry_sdk.capture_exception(ValueError())

try:
    raise ValueError()
except Exception:
    sentry_sdk.capture_exception()

...or send messages:

sentry_sdk.capture_message("Hi Sentry!")

Scopes (contexts, tags)

You can create a scope to attach data to all events happening inside of it:

with sentry_sdk.get_current_hub().push_scope():
    with sentry_sdk.configure_scope() as scope:
        scope.transaction = "my_view_name"
        scope.set_tag("key", "value")
        scope.user = {"id": 123}

    # ValueError event will have all that data attached
    capture_exception(ValueError())

# This one not since it is outside of the context manager
capture_exception(ValueError())

Scopes can be nested. If you call push_scope inside of the with-statement again, that scope will be pushed onto a stack. It will also inherit all data from the outer scope.

Scopes in unconfigured environments

If you never call init, no data will ever get sent to any server. In such situations, code like this is essentially deadweight:

with sentry_sdk.configure_scope() as scope:
    scope.user = _get_user_data()

Sentry-Python supports an alternative syntax for configuring a scope that solves this problem:

@sentry_sdk.configure_scope
def _(scope):
    scope.user = _get_user_data()

Your function will just not be executed if there is no client configured.

In your testing and development environment you still might want to run that code without sending any events. In that case, simply call init without a DSN:

sentry_sdk.init()

Breadcrumbs

Breadcrumbs also live on the stack. By default any (non-debug) log message anywhere in your system ends up as a breadcrumb, see the logging docs for more information. You can, however, also create breadcrumbs manually:

sentry_sdk.add_breadcrumb(
    timestamp=datetime.datetime.now(),
    type="log",
    level="debug",
    # message="hi",
    # category="myapp.models",
})

You can also pass a callback to add_breadcrumb like so:

sentry_sdk.add_breadcrumb(lambda: {
    "timestamp": datetime.datetime.now(),
    "type": "log",
    "level": "debug",
    # "message": "hi",
    # "category": "myapp.models",
})

The callback will only be called if a sentry client is configured.

Concurrency

• Sentry-Python currently does not support gevent-based setups.
• On init, Sentry-Python spawns a thread on its own. That means if you use uwsgi, you currently need to enable threads.
• On Python 3.7, Sentry-Python supports and automatically uses ContextVars. This should effectively enable Sentry-Python to work with asyncio under that Python version.

PII

Currently Sentry-Python does not send any personally-identifiable user data with events by default. You need to explicitly enable this behavior with the send_default_pii option passed to init:

init(..., send_default_pii=True)

Integrations

Head over to the other pages to check out integrations, which use these low-level APIs so you don't have to.