feat(request-trace): native S3 sink for request-trace records

[chethanuk]

Overview:

Add a native S3 RequestTraceSink so deployments can write dynamo.request.trace.v1 records straight to a bucket without an OpenTelemetry collector hop. Same schema as existing file/OTel/NATS sinks; new destination only.

Details:

• Extract SegmentSink from JsonlGzipWriter so roll/shutdown can finalize segments that cannot be appended after create (S3 objects).
• Add RequestTraceSinkKind::S3 + S3RequestTraceSink / S3SegmentSink, gated by default-on request-trace-s3.
• Object keys: {prefix}/YYYY/MM/DD/HH/{instance}-{startup}-{seq:06}.jsonl.gz.
• Env: DYN_REQUEST_TRACE_SINKS=s3, DYN_REQUEST_TRACE_S3_{BUCKET,REGION,PREFIX,ROLL_BYTES,FLUSH_INTERVAL_MS}.
• Credentials: AWS SDK default provider chain (no operator/CRD changes).
• Docs: docs/observability/request-tracing.md.
• Tests: unit replay (StaticReplayClient) + lib/llm/tests/request_trace_s3_replay.rs.

Where should the reviewer start?

1. lib/llm/src/telemetry/jsonl_gz.rs — SegmentSink trait + roll/shutdown await close
2. lib/llm/src/telemetry/s3_segment_sink.rs — PutObject segment sink
3. lib/llm/src/request_trace/sink.rs — S3RequestTraceSink + tests
4. lib/llm/src/request_trace/config.rs — kind parse + policy fields

Related Issues

🔗 This PR is linked to an issue:

• Relates to feat(request-trace): native S3 sink for request-trace records ai-dynamo/dynamo#11768

How to test

cargo test -p dynamo-llm --lib --no-default-features --features request-trace-s3 request_trace::
cargo test -p dynamo-llm --lib --no-default-features --features request-trace-s3 telemetry::jsonl_gz::
cargo test -p dynamo-llm --test request_trace_s3_replay --no-default-features --features request-trace-s3
cargo check -p dynamo-llm --no-default-features
cargo clippy -p dynamo-llm --lib --no-default-features --features request-trace-s3 -- -D warnings

Note on competing work

Upstream already has open PR ai-dynamo#11806 (YiqiuLiu) for the same issue. This fork PR is our independent implementation (SegmentSink refactor + roll-path close + StaticReplayClient tests). Do not open a second upstream PR without human review of the overlap.

[codeant-ai]

Skipping CodeAnt AI review — this PR changes more than 100 files, which usually means a migration, codemod, or vendored drop. Line-level review on diffs this large produces duplicate findings on the same rewrite pattern and drowns out anything that actually matters.

If you still want a review, comment @codeant-ai : review. For better signal, consider splitting the PR into smaller chunks.

[github-advanced-security]

CodeQL found more than 20 potential problems in the proposed changes. Check the Files changed tab for more details.

[gemini-code-assist]

Code Review

This pull request introduces significant updates to the Dynamo project, including improvements to the CI/CD pipeline, the introduction of a structured DGD (DynamoGraphDeployment) materialization and override protocol, and enhancements to the multimodal request handling and profiler. The review comments identify a missing import in sglang_processor.py, a potential crash in codeowners_match.py due to unhandled subprocess errors, and a missing finite check for the prefill_threshold in router_args.py. All review comments provide actionable feedback and should be addressed.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

The copy module is used on line 647 (schema = copy.deepcopy(schema)) but it is not imported in this file. Please import copy to avoid a NameError at runtime.

Suggested change

	from sglang.srt.parser.conversation import chat_template_exists
	import copy
	from sglang.srt.parser.conversation import chat_template_exists

[gemini-code-assist]

If repo is not a valid git repository, subprocess.check_output will raise an unhandled subprocess.CalledProcessError and crash the execution. Consider wrapping the call in a try-except block to gracefully handle the error and exit with a clear message, similar to the implementation in changed_paths.

Suggested change

	def load_tree(repo: Path) -> list[str]:
	"""Return tracked files under ``repo`` via ``git ls-files``."""
	out = subprocess.check_output(["git", "-C", str(repo), "ls-files"], text=True)
	return [p for p in out.splitlines() if p.strip()]
	def load_tree(repo: Path) -> list[str]:
	"""Return tracked files under ``repo`` via ``git ls-files``."""
	try:
	out = subprocess.check_output(["git", "-C", str(repo), "ls-files"], text=True)
	except subprocess.CalledProcessError as err:
	raise SystemExit(f"git ls-files failed in {repo!r}: {err}") from err
	return [p for p in out.splitlines() if p.strip()]

[gemini-code-assist]

For defensive programming, check math.isfinite on prefill_threshold to ensure it is a valid finite number, preventing non-finite values (like NaN or Infinity) from bypassing the < 0 check.

Suggested change

	prefill_threshold = self.active_prefill_tokens_threshold
	if prefill_threshold is not None and prefill_threshold < 0:
	raise ValueError("--active-prefill-tokens-threshold must be >= 0")
	prefill_threshold = self.active_prefill_tokens_threshold
	if prefill_threshold is not None and not (
	math.isfinite(prefill_threshold) and prefill_threshold >= 0
	):
	raise ValueError("--active-prefill-tokens-threshold must be a finite value >= 0")