[ericcurtin]

Summary

The agent trace relay now accepts OTLP/HTTP JSON as well as protobuf.

Related Issue

Closes #5319

Changes

• Transcode JSON to protobuf, then reuse the existing enrichment and limits
• Answer in the request's content type, {} on success
• Cap JSON bodies at 32 levels and 100,000 values to bound decode memory
• Enable with-serde on opentelemetry-proto for the supervisor only
• Update the relay docs

Testing

• Checks appropriate to the affected code and behavior pass
• Unit tests added/updated (if applicable)
• E2E tests added/updated (if applicable)

Checklist

• Follows Conventional Commits
• Commits are signed off (DCO)
• Architecture docs updated (if applicable)

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[ericcurtin]

recheck

[ericcurtin]

@rhuss @krishicks PTAL when you get a chance. Could you add an area label and /ok to test 1e86d87de483ee46270fd12087849f7647209f20 if it looks good? Thank you!

[rhuss]

cc-review Summary

What Went Well

• Encoding::of (lines 455-465) parses the media type once, and the same value picks the decoder and the response encoding.
• JSON is transcoded to protobuf and then goes through the same enrichment::enrich call inside the existing enrichment semaphore, so attribution has one implementation and MAX_ENRICHED_BYTES still applies.
• reply reaches every rejection, so 404, 408, 413, 415 and 503 (with Retry-After) all answer in the request's content type.
• json_request_is_transcoded_enriched_and_answered_in_json checks the exact hex-decoded trace and span ID bytes and both attribution attributes, and other_content_types_are_unsupported was updated rather than deleted.

Findings

Severity	File	Description	Source
Minor	Cargo.toml:73	with-serde is turned on in the workspace-wide opentelemetry-proto entry, but only openshell...	architecture
Minor	crates/openshell-supervisor/src/otlp_relay/receiver.rs:198	success(encoding.unwrap_or(Encoding::Protobuf)) repeats the fallback that line 187 already sto...	architecture
Minor	crates/openshell-supervisor/src/otlp_relay/receiver.rs:284	The JSON transcode (serde_json::from_slice plus encode_to_vec) runs inline in the async hand...	production
Minor	crates/openshell-supervisor/src/otlp_relay/receiver.rs:478	The JSON path decodes the whole request, but the crate docs still describe a protobuf-only relay...	architecture
Critical	crates/openshell-supervisor/src/otlp_relay/receiver.rs:483	The { count doesn't bound what serde_json::from_slice allocates. In opentelemetry-proto 0....	security
Minor	crates/openshell-supervisor/src/otlp_relay/receiver.rs:483	body.split(|byte| *byte == b'{').count() returns the number of braces plus one, so a body with...	architecture
Minor	crates/openshell-supervisor/src/otlp_relay/receiver.rs:484	The JSON cap returns EnrichError::TooManyResourceSpans, whose message is "more than 512 resour...	architecture
Minor	crates/openshell-supervisor/src/otlp_relay/receiver.rs:486	Some valid OTLP/JSON gets a 400. {"arrayValue":{}} and {"kvlistValue":{}} (an empty list wit...	correctness
Notable	crates/openshell-supervisor/src/otlp_relay/receiver.rs:486	Issue #5319 lists "Claude Code with detailed beta tracing, pointed at the relay endpoint, produc...	goal-alignment
Minor	crates/openshell-supervisor/src/otlp_relay/receiver.rs:488	The 400 body carries error.to_string() from serde_json, which includes the whole offending str...	security
Notable	crates/openshell-supervisor/src/otlp_relay/receiver.rs:513	rejected() still logs 400, 413 and 415 responses only at debug level, and RelayCounters coun...	production
Minor	crates/openshell-supervisor/src/otlp_relay/receiver.rs:838	json_body() sends "kind":1 and string-encoded startTimeUnixNano/endTimeUnixNano, but js...	test-quality
Minor	crates/openshell-supervisor/src/otlp_relay/receiver.rs:888	json_body_of_tiny_objects_is_too_large_and_never_buffered checks the status, the content type...	test-quality
Minor	crates/openshell-supervisor/src/otlp_relay/receiver.rs:899	metrics_and_logs_json_are_acknowledged_with_an_empty_object posts only to /v1/logs and, unli...	test-quality

Review Details

These findings were produced by cc-review's reviewer agents and posted without a per-finding review. Most failure scenarios were derived from reading the code; the measured figures (memory, timings, error sizes) come from small probe programs that the reviewer agents built and ran against this PR's transcode_json and locked dependency versions, not from running the relay itself.

• Findings posted: 14
• Findings reviewed and not posted: 0
• Gate outcome: FAIL
• Participating reviewers: architecture, correctness, goal-alignment, production, security, test-quality, gator

[ericcurtin]

@rhuss @TaylorMutch All threads addressed in 11422cd. PTAL when you get a chance. It needs an area label and /ok to test 11422cd5c5eae1a06e34fa0b9db611c767420c17. Thank you!