benchmark: add micro-benchmarks for TraceID hex formatting and parsing

[AneesPatel]

Description

Adds benchmark/trace_id_bench.cpp to introduce targeted Google Benchmark coverage for the 128-bit and 64-bit TraceID formatting paths. Also updates benchmark/CMakeLists.txt to include the new benchmark executable.

Baseline Metrics (Ubuntu 22.04 / GCC 11.4):

Benchmark	Time (ns/op)	Iterations	Notes
BM_TraceID_HexPadded	204 ns	3.6M	Full 128-bit ID to 32-char hex. Cost reflects two underlying std::string allocations and concatenation.
BM_TraceID_ParseHex_128bit	108 ns	6.5M	W3C traceparent extraction path.
BM_TraceID_ParseHex_64bit	60 ns	11.8M	Legacy 64-bit extraction path.
BM_HexPadded_uint64	46 ns	17.6M	Single to_chars + std::string alloc.
BM_Hex_uint64	24 ns	29.5M	No padding.

Motivation

TraceID::hex_padded() is called on every outgoing W3C traceparent header injection, and TraceID::parse_hex() on every incoming extraction. Because these functions sit in the hottest path of distributed tracing, their latency and allocation overhead multiply across millions of spans.

Currently, the underlying hex_padded<uint64_t>() and hex<uint64_t>() functions utilize a stack-buffer-to-std::string pattern. While the existing BM_TraceTinyCCSource benchmark provides excellent end-to-end throughput data, it cannot isolate micro-regressions in these specific serialization operations.

This baseline provides the data needed to safely prevent future regressions and evaluate allocation-reduction strategies in the critical path.

Jira ticket

N/A

Additional Notes

• Validated locally via Linux (WSL2 Ubuntu 22.04) with GCC 11.4.

How to run

mkdir .build && cd .build
cmake .. -DDD_TRACE_BUILD_BENCHMARK=1 -DCMAKE_BUILD_TYPE=Release
make -j
./benchmark/dd_trace_cpp-benchmark --benchmark_filter='^BM_TraceID|^BM_Hex'

[xlamorlette-datadog]

Please, what is you personal objective in submitting this PR?

[AneesPatel]

Hi @xlamorlette-datadog, thanks for the review and the great catch on the padding paths!

To answer your question, I'm currently an undergraduate studying computer science with an interest in low-latency systems and performance-oriented C++. My goal with this PR is to add measurable coverage for these serialization paths so that future allocation-reduction work (e.g., using std::to_chars with stack-allocated buffers to avoid the std::string heap allocation) can be evaluated with concrete latency data instead of guesswork.

I completely agree with testing the edge cases, and I've just pushed a commit that factorizes the tests using BENCHMARK_CAPTURE exactly as you suggested.

One interesting finding from my local runs is that BM_TraceID_HexPadded/WithPadding ({1, 0}) actually ran faster than the NoPadding case (~88 ns vs ~131 ns on my machine). It looks like when high == 0 the implementation skips formatting the high half entirely, which is why the WithPadding case ends up faster despite the name.

Let me know if the new structure looks good, thanks again for the review.

[datadog-prod-us1-3]

✨ Fix all issues with BitsAI

⚠️ Warnings

🚦 13 Pipeline jobs failed

Development | build-linux-cmake (gnu, arm64)     

🛟 This job is unlikely to succeed on retry. Please review your pipeline configuration.

Internal Error: API key is missing

Development | build-linux-cmake (gnu, x64)     

🛟 This job is unlikely to succeed on retry. Please review your pipeline configuration.

API key is missing for datadog-ci upload.

Development | build-linux-cmake (llvm, arm64)     

🛟 This job is unlikely to succeed on retry. Please review your pipeline configuration.

API key is missing during execution.

View all 13 failed jobs.

Useful? React with 👍 / 👎

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 3e8fece | Docs | Datadog PR Page | Give us feedback!}

[xlamorlette-datadog]

Some CI jobs are failing because this comes from a fork. I'm going to mirror this PR in our main repository.

[xlamorlette-datadog]

All CI jobs are green on the mirrored (draft) PR.

[xlamorlette-datadog]

@AneesPatel Would you please be able to sign your commits?
(This is blocking the PR merge.)

[xlamorlette-datadog]

@AneesPatel Thanks for your contribution!