fix(otel): reconnect async traces (e.g. LROs)

[dbolduc]

Fixes #13141

See the linked issue for the problem write up. The changes in this PR generated the seemingly correct, second screenshot.

The ABI update was necessary because we stop defining some templated future<StatusOr<std::chrono::...>> type, which triggered a false positive in check-api.

Something is buggy with (at least) opentelemetry-cpp + gcc 7.3.1. So patch the Dockerfile used in that build to work around it.

---

This change is 

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (ecb9dcc) 93.00% compared to head (1a577cb) 92.99%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #13147      +/-   ##
==========================================
- Coverage   93.00%   92.99%   -0.01%     
==========================================
  Files        2137     2137              
  Lines      185877   185898      +21     
==========================================
+ Hits       172868   172884      +16     
- Misses      13009    13014       +5     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[dbolduc]

Ugh, the gcc-7-3 error is out of our control. I need to think about how we can avoid it in our repo.

The following code reproduces the same error. It does not use any google-cloud-cpp helpers. Some bug in the compiler causes this basic opentelemetry-cpp test to fail.

TEST(OpenTelemetry, DarrenOTelOnly) {
  auto span_catcher = InstallSpanCatcher();

  auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  auto tracer = provider->GetTracer("gcc-7-3 test");

  auto parent = tracer->StartSpan("parent");
  opentelemetry::trace::Scope parent_scope(parent);

  auto child = tracer->StartSpan("child");
  child->End();

  parent->End();

  EXPECT_THAT(
      span_catcher->GetSpans(),
      ElementsAre(AllOf(SpanNamed("child"),
                        SpanWithParentSpanId(parent->GetContext().span_id())),
                  SpanNamed("parent")));
}

[ RUN      ] OpenTelemetry.DarrenOTelOnly
../google/cloud/internal/grpc_opentelemetry_test.cc:279: Failure
Value of: span_catcher->GetSpans()
Expected: has 2 elements where
element #0 (has name: child) and (has parent span id: 5d48e34bfafc4edb),
element #1 has name: parent
  Actual: { (ptr = 0x1f422a0, value = Span {name=child, kind=INTERNAL, instrumentation_scope {gcc-7-3 test, },
			parent_span_id=0000000000000000
			attributes=[],
			events=[],
			links=[]}), (ptr = 0x1f42c20, value = Span {name=parent, kind=INTERNAL, instrumentation_scope {gcc-7-3 test, },
			parent_span_id=0000000000000000
			attributes=[],
			events=[],
			links=[]}) }, whose element #0 doesn't match, has parent span id: 0000000000000000

[  FAILED  ] OpenTelemetry.DarrenOTelOnly (0 ms)

[coryan]

Does OTel claim to support GCC 7.x? Is there a bug in OTel we can reference? Can we detect this with an OTel version?

[dbolduc]

Does OTel claim to support GCC 7.x?

My reading is they support compilers that implement the standard correctly.

Is there a bug in OTel we can reference?

good call. yes: open-telemetry/opentelemetry-cpp#1014

[dbolduc]

There is a patch linked in the issue. I will try to apply it to our centos-7 dockerfile.

[coryan]

According to some comments on that bug, GCC 8.x also trouble. You may need to apply that patch to several other builds.

[coryan]

My reading is they support compilers that implement the standard correctly.

There are no compilers without defects. That implies they support no compilers at all.

[dbolduc]

Patched.

According to some comments on that bug, GCC 8.x also trouble. You may need to apply that patch to several other builds.

Thanks for pointing that out. I forgot that the demo-install builds do not run these unit tests. I assumed they all worked.

Opened #13159 because I am not sure if we should apply the patch by default, or just call out the issue in a comment.

There are no compilers without defects. That implies they support no compilers at all.

I realize that and my comment was tongue-in-cheek. Here is their actual text: https://github.com/open-telemetry/opentelemetry-cpp/blob/cb603ad97f33e52340e627e1cb43ba73bb1d7ef0/README.md?plain=1#L48-L49

[devbww]

std::move(span)

[dbolduc]

Done

[dbolduc]

Ping.

(I am not blocked on this PR. It would just be nice to have the fix in.)