feat(tracing): OTel improvements from go-fan review

[Copilot]

Implements the High/Medium recommendations from the go-fan OTel review: allocation guards for noop spans, explicit error-scrubbing helper, centralised semconv imports, batch processor tuning, and safer fanout shutdown.

Changes

span.IsRecording() guards — unified.go, proxy/handler.go

Wrap AddEvent calls and attribute slice construction with if span.IsRecording() to skip allocations entirely in noop mode (the common case when tracing is disabled).

RecordSpanErrorSafe helper — internal/tracing/span_helpers.go

New helper that records only a scrubbed public message on the span instead of the internal error, making the security intent explicit and preventing accidental leakage of transport/credential details to trace backends:

// Before — ad-hoc scrubbing, easy to get wrong at new call sites
tracing.RecordSpanError(execSpan, fmt.Errorf("tool execution failed"), "tool execution failed")

// After — intent is self-documenting; internal err goes to caller/logs only
tracing.RecordSpanErrorSafe(execSpan, err, "tool execution failed")

Centralised semconv imports — internal/tracing/semconv.go (new)

Single file that imports go.opentelemetry.io/otel/semconv/v1.41.0 and re-exports all attribute keys and helper functions used across the package. All other non-test files (including proxy/handler.go) now reference these re-exports — a future version bump touches one file instead of five.

BatchSpanProcessor tuning — internal/tracing/provider.go

Explicit MaxExportBatchSize=256 + BatchTimeout=2s (SDK defaults are 512 / 5 s) to reduce export tail latency under gateway throughput patterns.

fanoutExporter.Shutdown context propagation — internal/tracing/fanout.go

Each exporter goroutine now receives a fresh context.WithTimeout derived from the parent's remaining deadline at the moment Shutdown is called. This prevents a slow exporter from consuming the entire budget and leaving subsequent exporters with an already-expired context.

[Copilot]

Pull request overview

This PR improves the gateway’s OpenTelemetry tracing implementation by reducing allocations when tracing is disabled, centralizing semantic convention usage, making trace error recording explicitly “safe” for security-sensitive paths, tuning batch export behavior, and adjusting fanout shutdown behavior.

Changes:

• Added span.IsRecording() guards around AddEvent and attribute slice construction to avoid allocations in noop tracing.
• Introduced RecordSpanErrorSafe and updated an error path to record only a scrubbed/public message in traces.
• Centralized semconv/v1.41.0 usage behind internal/tracing/semconv.go re-exports and tuned BatchSpanProcessor parameters; updated fanout exporter shutdown behavior.

Show a summary per file

File	Description
internal/tracing/span_helpers.go	Adds RecordSpanErrorSafe and migrates span helpers to semconv re-exports.
internal/tracing/span_helpers_test.go	Updates semconv usage and adds tests for the new safe error-recording helper.
internal/tracing/semconv.go	New central semconv re-export module for tracing package callers.
internal/tracing/provider.go	Uses semconv re-exports for resource attrs and tunes SDK batch span processor options.
internal/tracing/http.go	Switches HTTP span attributes to semconv re-exports.
internal/tracing/genai_attrs.go	Clarifies split between semconv-derived GenAI keys (in semconv.go) and custom keys.
internal/tracing/fanout.go	Changes shutdown to use per-exporter timeout contexts based on remaining deadline.
internal/server/unified.go	Adds IsRecording() guards and uses RecordSpanErrorSafe for transport/tool execution errors.
internal/proxy/handler.go	Adds IsRecording() guards and switches HTTP status key usage to semconv re-exports.

Review details

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

• Files reviewed: 9/9 changed files
• Comments generated: 3
• Review effort level: Low

[lpcox]

@copilot fix this failing ci check https://github.com/github/gh-aw-mcpg/actions/runs/28868492598/job/85624699407?pr=8841

[Copilot]

@copilot fix this failing ci check https://github.com/github/gh-aw-mcpg/actions/runs/28868492598/job/85624699407?pr=8841

Fixed in the latest commit. The CI failure was caused by a missing closing brace } at the end of TestRecordSpanErrorSafe_RecordsPublicMsgNotInternalError in internal/tracing/span_helpers_test.go, which resulted in a syntax error (expected '(', found TestRecordSpanErrorSafe_SetsErrorTypeFromPublicErr).