fix(tracing): Fix unsampled/deferred trace propagation in span streaming

[sentrivana]

Description

The SDK didn't behave correctly in some trace propagation scenarios in stream mode.

The current SDK design was emitting headers with a deferred sampling decision if the sampling decision was negative (whether inherited or not), because it was falling back to TwP (tracing without performance) behavior too eagerly.

Additionally, some edge cases concerning ignore_spans and its interaction with sampling were clarified and the necessary changes included in this PR.

How it should work

In the following, tracing enabled means either a traces_sample_rate or a traces_sampler is defined. Tracing disabled means both are None and tracing without performance is active.

There are three kinds of sampling decisions: sampled (1), unsampled (0), and deferred (``), meaning no decision.

1. No incoming trace headers
   • tracing enabled: make a sampling decision and propagate it
   • tracing disabled: do not make a decision, propagate a deferred decision (no sampling decision)
2. Incoming trace headers with a sampled/unsampled decision
   • tracing enabled: respect and propagate incoming decision
   • tracing disabled: respect and propagate incoming decision
3. Incoming trace headers with a deferred sampling decision (essentially same as 1)
   • tracing enabled: make a sampling decision and propagate it
   • tracing disabled: propagate deferred sampling decision

The problem and the fix

If a no-op span was active, we were falling back to the propagation context to generate propagation headers, which by default doesn't contain a sampling decision.

With this PR, NoOpStreamedSpans now carry their own trace and span ID information as well as the current propagation data (a snapshot of the propagation context, like StreamedSpans) and use it to emit the correct headers. This mirrors old behavior: spans would always be created, whether they'd be sampled or not, and they would be responsible for generating propagation headers; with the fallback to the propagation context only kicking in if there is no span.

I recommend checking out the new tests to see what behavior is now expected.

Edge cases

The following cases regarding ignore_spans were clarified and will be documented soon:

• An outgoing request is made from an ignored segment: sentry-trace should have an unsampled decision (-0), parent_span_id is the ID of the segment (or any random span ID, since for all intents and purposes, the ignored segment doesn't actually exist for any downstream service)
• An outgoing request is made from an ignored child span of a non-ignored segment: sentry-trace's sampling decision should be the segment's sampling decision; parent_span_id should be the first non-ignored span ancestor

Issues

Closes #6755

Reminders

• Please add tests to validate your changes, and lint your code using uv run ruff.
• Add GH Issue ID & Linear ID (if applicable)
• PR title should use conventional commit style (feat:, fix:, ref:, meta:)
• For external contributors: CONTRIBUTING.md, Sentry SDK development docs, Discord community

[github-actions]

Codecov Results 📊

✅ 91200 passed | ⏭️ 6302 skipped | Total: 97502 | Pass Rate: 93.54% | Execution Time: 312m 25s

📊 Comparison with Base Branch

Metric	Change
Total Tests	📈 +180
Passed Tests	📈 +180
Failed Tests	—
Skipped Tests	—

All tests are passing successfully.

✅ Patch coverage is 97.22%. Project has 2444 uncovered lines.
✅ Project coverage is 89.77%. Comparing base (base) to head (head).

Files with missing lines (1)

File	Patch %	Lines
sentry_sdk/traces.py	93.75%	⚠️ 1 Missing and 1 partials

Coverage diff

@@            Coverage Diff             @@
##          main       #PR       +/-##
==========================================
+ Coverage    89.76%    89.77%    +0.01%
==========================================
  Files          193       193         —
  Lines        23888     23898       +10
  Branches      8250      8256        +6
==========================================
+ Hits         21442     21454       +12
- Misses        2446      2444        -2
- Partials      1353      1352        -1

---

Generated by Codecov Action

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Want reviews to match your repository better? Bugbot Learning can learn team-specific rules from PR activity. A team admin can enable Learning in the Cursor dashboard.

^{Reviewed by Cursor Bugbot for commit 46b410a. Configure here.}

[sentrivana]

Converting to draft to address bot comments

[ericapisani]

Haven't had a chance to look at the tests yet, but noticed a couple things on my first pass

[ericapisani]

This can also be written as to remove the extra conditional:

Suggested change

	if sampled is False or sampled is None:
	if bool(sampled):

[ericapisani]

Of these, only _sampled needs to be added as the other 3 already exist on the parent 's __slots__ and will still be accessible by instances of NoOpStreamedSpan.

As an example:

>>> class Foo():
...     __slots__ = ("x")
...
>>> class Bar(Foo):
...     __slots__ = ("y")
...
>>> b = Bar()
>>> b.x = "hi"
>>> b.y = "world"
>>> print(b.x)
hi
>>> b.z
Traceback (most recent call last):
  File "<python-input-7>", line 1, in <module>
    b.z
AttributeError: 'Bar' object has no attribute 'z'

One thing to be wary of that I found when looking into the slots docs - duplicating the variable name in __slots__ as shown here creates a shadow of the parent value.

[ericapisani]

Other than the slots thing that I mentioned earlier, this LGTM