[ETW Exporter] Tail based sampling support

[lalitb]

Changes

Adding limited support for tail based sampling. Similar to head-sampling, TailSampler::ShouldSample() method needs to be overridden with sampling logic, which get's executed before the span is exporter.

class TailSampler
{
public:
  // convert to etw span if required for getters on span.
  // auto etw_span = static_cast<const opentelemetry::exporter::etw::Span*>(&span);
  // Decision based on
  //      Span::GetStatus()
  //      Span::GetProperties()
  //      Span::GetContext()
  virtual opentelemetry::sdk::trace::SamplingResult ShouldSample(
      const opentelemetry::trace::Span &span) noexcept = 0;
};

The inherited TailSampler class object needs to be passed as argument to TracerProvider constructor. The default is AlwaysOnTailSampler

Also, added use of macro OPENTELEMETRY_NOT_USE_RTTI to disable using RTTI in ETW exporter even if it is enabled in compiler. This is because the ETW exporter modified the attributes passed as argument if RTTI is enabled to reuse them for creating ETW event structure, which is sometime not desirable.

For significant contributions please make sure you have completed the following items:

• CHANGELOG.md updated for non-trivial changes
• Unit tests have been added
• Changes in public API reviewed

[codecov]

Codecov Report

Merging #1780 (e9423ee) into main (f23c4a2) will decrease coverage by 0.02%.
The diff coverage is n/a.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1780      +/-   ##
==========================================
- Coverage   85.73%   85.71%   -0.01%     
==========================================
  Files         171      171              
  Lines        5240     5240              
==========================================
- Hits         4492     4491       -1     
- Misses        748      749       +1     

Impacted Files	Coverage Δ
sdk/src/trace/batch_span_processor.cc	90.63% <0.00%> (-0.78%)	⬇️

[ThomsonTan]

Could we start a span and test it is indeed dropped?

[lalitb]

Yes I was thinking to mock the actual ETW transport layer and do the test. In current design, this is not easy to do, perhaps can be taken separately if it is fine?

[ThomsonTan]

Instead of making a virtual call to AlwaysOn sampler by default, perhaps check if there is a non-default sampler set, then make the virtual call? In this way, it doesn't introduce extra cost (except the null check) when tail based sampler is not used.

[lalitb]

Good comment. Actually we do have virtual method calls for head-sampler, id-generator both in ETW exporter, and the otel-sdk. There is no reported performance reported with them. Seemingly modern compilers can optimize these repeated calls by caching the address, and avoid virtual indirection. Perhaps we can look back to this if there are reported overhead internally and/or by community.