Add ForceFlush to TracerProvider in SDK

[ThomsonTan]

Fixes #587

Changes

Add ForceFlush to TracerProvider as per the spec PR open-telemetry/opentelemetry-specification#1452

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 #588 (aa73133) into main (3848295) will increase coverage by 0.04%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main     #588      +/-   ##
==========================================
+ Coverage   94.41%   94.46%   +0.04%     
==========================================
  Files         200      200              
  Lines        9080     9087       +7     
==========================================
+ Hits         8573     8584      +11     
+ Misses        507      503       -4     

Impacted Files	Coverage Δ
.../include/opentelemetry/sdk/trace/tracer_provider.h	100.00% <ø> (ø)
sdk/src/logs/logger.cc	65.90% <ø> (ø)
sdk/src/trace/tracer_provider.cc	68.18% <100.00%> (+3.18%)	⬆️
sdk/test/trace/tracer_provider_test.cc	100.00% <100.00%> (ø)
sdk/test/common/circular_buffer_test.cc	100.00% <0.00%> (+1.04%)	⬆️
sdk/test/metrics/counter_aggregator_test.cc	100.00% <0.00%> (+1.78%)	⬆️
...include/opentelemetry/sdk/trace/simple_processor.h	100.00% <0.00%> (+11.11%)	⬆️

[reyang]

Curious - do we want to call ForceFlush or Shutdown? (I thought the latter)

[lalitb]

As per the specs (https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/trace/sdk.md#forceflush):
ForceFlush MUST invoke ForceFlush on all registered SpanProcessors.

So ForceFlush() method needs to be defined both for SpanProcessor and Exporter. This can be done as separate PR and in that case Shutdown() call can be removed for now from above.

Note: PR to update specs for adding ForceFlush in Exporter is still open ( open-telemetry/opentelemetry-specification#1467 )

[ThomsonTan]

Thanks for the catch, @reyang

@lalitb I think we only support one SpanProcessor now. I'd prefer to add ForceFlush to SpanExporter in a separate PR.

[reyang]

I think we only support one SpanProcessor now.

That's fine, I think we could support multiple processors by using a de-mux processor (so from the provider perspective, there is only one processor).

Here are some examples:

• https://github.com/open-telemetry/opentelemetry-dotnet/blob/main/src/OpenTelemetry/CompositeProcessor.cs
• https://github.com/open-telemetry/opentelemetry-python/blob/99128b35d1a9cf2ecb04097c6a87d320a552a3e4/opentelemetry-sdk/src/opentelemetry/sdk/trace/__init__.py#L127

[reyang]

LGTM.