W3C trace context propagation fails W3C Trace Context Level 2 compliance tests

[Kielek]

Description

System.Diagnostics.W3CPropagator and the related Activity propagation path do not fully comply with the W3C Trace Context Level 2 test suite.

The most direct issue is in W3CPropagator.ExtractTraceIdAndState: after traceparent is missing or rejected, it still reads and validates tracestate. W3C Trace Context Level 2 says that if parsing traceparent fails, vendors must not attempt to parse tracestate.

Observed source shape:

getter(carrier, TraceParent, out traceId, out _);
if (IsInvalidTraceParent(traceId))
{
    traceId = null;
}

getter(carrier, TraceState, out string? traceStateValue, out _);
traceState = ValidateTraceState(traceStateValue);

This allows a request with only tracestate to propagate tracestate into a newly generated trace context.

Running the full upstream W3C HTTP compliance suite also shows additional Level 2 failures around future-version traceparent propagation and tracestate validation.

Reproduction Steps

I added a standalone repro under:

src/tests/W3CTraceContextCompliance

The test harness is available here:

https://github.com/Kielek/runtime/commit/9f88d19ffa843faf7a29c258ac2b77795f88bea9

The repro starts a small ASP.NET Core endpoint that uses DistributedContextPropagator.CreateW3CPropagator(), then runs the embedded upstream w3c/trace-context test suite against it.

From the dotnet/runtime repository root:

pwsh -File .\src\tests\W3CTraceContextCompliance\run-w3c-suite.ps1

The runner uses:

W3C trace-context commit: 34b10ac5af7f0caeb28efe35fe51cd4763ec5771
SPEC_LEVEL=2
STRICT_LEVEL=2

Test-suite note: the embedded W3C commit above contains tracestate key grammar expectations that were reported in w3c/trace-context#583 and corrected by w3c/trace-context#584. For validation with the corrected suite, use the PR w3c/trace-context#584 head commit:

W3C trace-context PR #584 head: 6261328fd701ca967f13ca0543504db879c3b6e0

If System.Diagnostics.DiagnosticSource is already built:

pwsh -File .\src\tests\W3CTraceContextCompliance\run-w3c-suite.ps1 `
  -SkipDiagnosticSourceBuild

Expected behavior

The W3C Trace Context Level 2 suite should pass.

In particular:

• tracestate should be ignored when traceparent is missing or invalid.
• Valid future-version traceparent values should be accepted according to the forward-compatibility rules.
• Invalid tracestate values should not be propagated.
• Valid tracestate values with Level 2 key/member/OWS rules should be preserved.

Actual behavior

The full suite runs, but fails:

STRICT_LEVEL: 2
SPEC_LEVEL:   2
harness listening on http://127.0.0.1:7777

Ran 41 tests in 0.978s

FAILED (failures=5, errors=2)
W3C suite exit code: 7

Failing tests:

FAIL  test_traceparent_version_0xCC
FAIL  test_tracestate_included_traceparent_missing
ERROR test_tracestate_key_illegal_characters
ERROR test_tracestate_key_illegal_vendor_format
FAIL  test_tracestate_key_length_limit
FAIL  test_tracestate_member_count_limit
FAIL  test_tracestate_ows_handling

The test_tracestate_key_illegal_vendor_format and part of test_tracestate_key_length_limit failures are affected by the W3C suite grammar issue tracked by w3c/trace-context#583 and fixed in w3c/trace-context#584. The remaining failures are runtime behavior issues independent of that test-suite correction.

The focused ExtractTraceIdAndState behavior can be reduced to:

missing traceparent + tracestate=foo=1
expected: traceId=null, traceState=null
actual:   traceId=null, traceState=foo=1

and:

invalid traceparent + tracestate=foo=1
expected: traceId=null, traceState=null
actual:   traceId=null, traceState=foo=1

Regression?

Probably no, IMO still worth to include into .NET 11.

Known Workarounds

N/ACallers can manually discard the returned traceState whenever traceId is null, but that does not fix the built-in propagator behavior or the HTTP propagation path.

Configuration

Observed locally with:

Repository: dotnet/runtime
Target framework: net11.0
OS: Windows 10.0.26200
Architecture: x64
.NET SDK: 11.0.100-preview.5.26227.104
W3C trace-context suite commit: 34b10ac5af7f0caeb28efe35fe51cd4763ec5771
Python: 3.11 venv with the pinned OpenTelemetry W3C test requirements

Other information

Reported w3c/trace-context#583 and proposed fix w3c/trace-context#584 under W3C.

[dotnet-policy-service]

Tagging subscribers to this area: @steveisok, @dotnet/area-system-diagnostics-tracing
See info in area-owners.md if you want to be subscribed.