feat(analytics): typesafe analytics.Event and analytics.record

[constantinius]

Closes: https://linear.app/getsentry/issue/TET-804/analytics-allow-both-old-style-attributes-as-well-as-new-style

Supersedes: #92943

Making analytics.record and analytics.Event typesafer

Deprecate stringly typed analytics.record() in favor of version that uses analytics.Event objects.
Migrating analytics.Event to dataclasses. For convenience introduce the analytics.eventclass decorator, which sets the type and applies the dataclass decorator with common parameters underneath.

Allows "old-style" definition of Events as well as to record() them until all event classes are migrated.

[codecov]

Codecov Report

Attention: Patch coverage is 96.47059% with 3 lines in your changes missing coverage. Please review.

✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/sentry/analytics/event.py	94.28%	2 Missing ⚠️
src/sentry/analytics/base.py	92.30%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master   #94953      +/-   ##
==========================================
+ Coverage   87.87%   87.90%   +0.02%     
==========================================
  Files       10456    10438      -18     
  Lines      604784   604154     -630     
  Branches    23617    23498     -119     
==========================================
- Hits       531454   531064     -390     
+ Misses      72969    72725     -244     
- Partials      361      365       +4     

[shellmayr]

LGTM - and great job designing this, it will make our analytics system so much more robust! 🥳

[cursor]

Bug: Dataclass Construction Fails Across Styles

The from_instance method has multiple issues:

• Old-style path: It ignores kwargs for uuid_, datetime_, and other dataclass fields, leading to incorrect event data. The direct cls() call is also fragile, potentially failing if old-style event subclasses introduce required fields without defaults.
• New-style path: It fails to filter init=False fields when constructing the instance, which can cause a TypeError.
• General: The codebase mixes Pydantic's @dataclass with standard library dataclasses functions, risking subtle incompatibilities.

src/sentry/analytics/event.py#L102-L118

sentry/src/sentry/analytics/event.py

Lines 102 to 118 in bfe2996

	# we can remove this.
	if cls.attributes:
	items = {
	attr.name: kwargs.get(attr.name, getattr(instance, attr.name, None))
	for attr in cls.attributes
	}
	self = cls()
	self.data = get_data(cls.attributes, items)
	return self
	return cls(
	**{
	f.name: kwargs.get(f.name, getattr(instance, f.name, None))
	for f in fields(cls)
	if f.name not in ("type", "uuid_", "datetime_")
	}
	)

Fix in Cursor • Fix in Web

---

Was this report helpful? Give feedback by reacting with 👍 or 👎

[sentry-io]

Suspect Issues

This pull request was deployed and Sentry observed the following issues:

• ‼️ KeyError: 'trial_usage.end.sent' in de

• ‼️ KeyError: 'trial_usage.end.sent' in de

• ‼️ KeyError: 'trial_usage.end.sent' in de

• ‼️ KeyError: 'trial_usage.end.sent' in de

• ‼️ ValueError: user_id is required (cannot be None) in prod

_{Did you find this useful? React with a 👍 or 👎}

[asottile-sentry]

would it be possible to write this without pydantic

we're permastuck on version 1 and I'm looking to factor it out because it is a liability to upgrading python -- adding new usage of it makes that a lot lot harder

[shellmayr]

Yeah AFAIK @constantinius wants to refactor this without pydantic when the migration of analytics events over to the typed structure is done. Do you have a timeline in mind?

[asottile-sentry]

not urgent -- still have to factor it out of hybrid cloud 😭

[shellmayr]

Alright cool - I'll bring up that you want to get pydantic out of the codebase when he's back from vacation 👍